Antidepressants for hip and knee osteoarthritis

Abstract

Background

Although pain is common in osteoarthritis, most people fail to achieve adequate analgesia. Increasing acknowledgement of the contribution of pain sensitisation has resulted in the investigation of medications affecting pain processing with central effects. Antidepressants contribute to pain management in other conditions where pain sensitisation is present.

Objectives

To assess the benefits and harms of antidepressants for the treatment of symptomatic knee and hip osteoarthritis in adults.

Search methods

We used standard, extensive Cochrane search methods. The latest search was January 2021.

Selection criteria

We included randomised controlled trials of adults with osteoarthritis that compared use of antidepressants to placebo or alternative comparator. We included trials that focused on efficacy (pain and function), treatment‐related adverse effects and had documentation regarding discontinuation of participants. We excluded trials of less than six weeks of duration or had participants with concurrent mental health disorders.

Data collection and analysis

We used standard Cochrane methods. Major outcomes were pain; responder rate; physical function; quality of life; and proportion of participants who withdrew due to adverse events, experienced any adverse events or had serious adverse events. Minor outcomes were proportion meeting the OARSI (Osteoarthritis Research Society International) Response Criteria, radiographic joint structure changes and proportion of participants who dropped out of the study for any reason. We used GRADE to assess certainty of evidence.

Main results

Nine trials (2122 participants) met the inclusion criteria. Seven trials examined only knee osteoarthritis. Two also included participants with hip osteoarthritis. All trials compared antidepressants to placebo, with or without non‐steroidal anti‐inflammatory drugs. Trial sizes were 36 to 388 participants. Most participants were female, with mean ages of 54.5 to 65.9 years. Trial durations were 8 to 16 weeks. Six trials examined duloxetine. We combined data from nine trials in meta‐analyses for knee and hip osteoarthritis.

One trial was at low risk of bias in all domains. Five trials were at risk of attrition and reporting bias.

High‐certainty evidence found that antidepressants resulted in a clinically unimportant improvement in pain compared to placebo. Mean reduction in pain (0 to 10 scale, 0 = no pain) was 1.7 points with placebo and 2.3 points with antidepressants (mean difference (MD) −0.59, 95% confidence interval (CI) −0.88 to −0.31; 9 trials, 2122 participants).

Clinical response was defined as achieving a 50% or greater reduction in 24‐hour mean pain. High‐certainty evidence demonstrated that 45% of participants receiving antidepressants had a clinical response compared to 28.6% receiving placebo (RR 1.55, 95% CI 1.32 to 1.82; 6 RCTs, 1904 participants). This corresponded to an absolute improvement in pain of 16% more responders with antidepressants (8.9% more to 26% more) and a number needed to treat for an additional beneficial effect (NNTB) of 6 (95% CI 4 to 11).

High‐certainty evidence showed that the mean improvement in function (on 0 to 100 Western Ontario and McMaster Universities Arthritis Index, 0 = best function) was 10.51 points with placebo and 16.16 points with antidepressants (MD −5.65 points, 95% CI −7.08 to −4.23; 6 RCTs, 1909 participants). This demonstrates a small, clinically unimportant response.

Moderate‐certainty evidence (downgraded for imprecision) showed that quality of life measured using the EuroQol 5‐Dimension scale (−0.11 to 1.0, 1.0 = perfect health) improved by 0.07 points with placebo and 0.11 points with antidepressants (MD 0.04, 95% CI 0.01 to 0.07; 3 RCTs, 815 participants). This is clinically unimportant.

High‐certainty evidence showed that total adverse events increased in the antidepressant group (64%) compared to the placebo group (49%) (RR 1.27, 95% CI 1.15 to 1.41; 9 RCTs, 2102 participants). The number needed to treat for an additional harmful outcome (NNTH) was 7 (95% CI 5 to 11).

Low‐certainty evidence (downgraded twice for imprecision for very low numbers of events) found no evidence of a difference in serious adverse events between groups (RR 0.94, 95% CI 0.46 to 1.94; 9 RCTs, 2101 participants). The NNTH was 1000. 

Moderate‐certainty evidence (downgraded for imprecision) showed that 11% of participants receiving antidepressants withdrew from trials due to an adverse event compared to 5% receiving placebo (RR 2.15, 95% CI 1.56 to 2.97; 6 RCTs, 1977 participants). The NNTH was 17 (95% CI 10 to 35).

Authors' conclusions

There is high‐certainty evidence that use of antidepressants for knee osteoarthritis leads to a non‐clinically important improvement in mean pain and function. However, a small number of people will have a 50% or greater important improvement in pain and function. This finding was consistent across all trials. Pain in osteoarthritis may be due to a variety of causes that differ between individuals. It may be that the cause of pain that responds to this therapy is only present in a small number of people. There is moderate‐certainty evidence that antidepressants have a small positive effect on quality of life with heterogeneity between trials.

High‐certainty evidence indicates antidepressants result in more adverse events and moderate‐certainty evidence indicates more withdrawal due to adverse events. There was little to no difference in serious adverse events (low‐certainty evidence due to low numbers of events). This suggests that if antidepressants were being considered, there needs to be careful patient selection to optimise clinical benefit given the known propensity for adverse events with antidepressant use. Future trials should include alternative antidepressant agents or phenotyping of pain in people with osteoarthritis, or both.

Plain language summary

Antidepressants for osteoarthritis

Background

Osteoarthritis is a disease of the joints characterised by reduced cartilage and narrowing of the space between the joints. It may result in pain, deformity and disability.

Antidepressant medicines are thought to affect pain by modulating nerve pathways in the central nervous system.

We aimed to evaluate the clinical benefits and harms of antidepressants for knee and hip osteoarthritis pain.

Study characteristics

This review is current to January 2021. We included nine clinical trials with 2122 participants comparing antidepressants to placebo (a dummy treatment) and non‐steroidal anti‐inflammatory medicines (which are widely used to relieve pain and reduce inflammation). More participants were female (70%) and the average age was 54.4 to 65.9 years. Seven trials examined only knee osteoarthritis. Two also included people with hip osteoarthritis. All trials compared antidepressants to placebo, with or without non‐steroidal anti‐inflammatory medicines.

Key results

Compared with a placebo, antidepressants resulted in a small benefit.

Pain

Pain reduced by 0.59 points more on a 0‐ to 10‐point scale with antidepressants compared to placebo.

– The placebo group had a pain reduction of 1.73 points.

– The antidepressant group had a pain reduction of 2.32 points.

Response

A clinical response is considered a 50% or greater reduction in pain. About 16% more people taking an antidepressant had a reduction compared to placebo (that is, one in six people).

– 28.65% of people taking placebo had a clinical response.

– 45.2% of people taking antidepressants had a clinical response.

Function

Physical function is measured using walking, stair climbing and domestic duties. It improved 6% more with antidepressants compared to placebo. With antidepressants, the function score reduced by 5.65 on a 0‐ to 100‐point scale (lower score = better function), compared to placebo. This is a small improvement.

– The placebo group had an improvement in function of 10.51 points.

– The antidepressant group had an improvement in function of 16.16 points.

Quality of life

Quality of life improved by 4% more with antidepressants compared to placebo. This was 0.04 points higher on a −0.11 to 1 scale in people taking antidepressants compared to placebo (higher score = better quality of life).

– In the placebo group, quality of life improved by 0.07 points.

– In the antidepressant group, quality of life improved by 0.11 points.

Withdrawals due to side effects

The risk of stopping treatment was 2.15 more in the antidepressant group compared to placebo. This means that 1 in 17 people stopped taking the antidepressant because of side effects.

– 5.1% of the placebo group withdrew because of side effects.

– 10.9% of the antidepressant group withdrew because of side effects.

Total side effects

The risk of having any side effects was 1.27 more in the antidepressant group compared to the placebo group. About 14.8% more people reported a side effect in the antidepressant group. This is equivalent to 1 in every 7 people having a side effect.

– 49.3% of the placebo group had a side effect.

– 64.1% of the antidepressant group had a side effect.

Serious side effects

There was no difference in serious side effects between groups.

– 1.7% of the placebo group had serious side effects.

– 1.6% of the antidepressant group had serious side effects.

Quality of the evidence

In people with osteoarthritis, high‐quality evidence shows that antidepressants have a small positive effect on pain and function and that one in six people have a clinically important response of a 50% or greater reduction in their pain. High‐quality evidence also demonstrates that people taking antidepressants have a higher frequency of side effects than those taking placebo.

Moderate‐quality evidence shows very small, probably unimportant, improvements in quality of life for people taking antidepressants compared to placebo. People receiving antidepressants are more likely to stop them because of side effects than placebo.

Low‐quality evidence shows little difference in serious side effects between antidepressants and placebo.

Summary of findings

Summary of findings 1. Antidepressant compared to placebo for osteoarthritis for Hip and Knee Osteoarthritis.

Antidepressant compared to placebo for osteoarthritis
Patient or population: predominantly knee osteoarthritis Setting: community Intervention: antidepressant  Comparison: placebo
Outcomes	Anticipated absolute effects* (95% CI)		Relative effect (95% CI)	№ of participants (studies)	Certainty of the evidence (GRADE)	Comments
	Risk with placebo	Risk with antidepressant
Mean change in pain from baseline (0‐ to 10‐point scale, 0 is no pain) at 8–16 weeks	The mean reduction in pain was 1.73 pointsa	The mean reduction in pain was 2.32 points	MD −0.59 −0.88 to −0.31)	2038 (9 RCTs)	⊕⊕⊕⊕ High	Small possibly clinically unimportant improvement. Absolute improvement 6% (95% CI 3% to 10%)
Responder rate: ≥ 50% reduction in 24‐hour mean pain	28.6%	45.2% (37.5% to 54.9%)	RR 1.55 (1.32 to 1.82)	1904 (6 RCTs)	⊕⊕⊕⊕ High	Absolute improvement: 16% more were responders with antidepressant (8.9% more to 26% more)b. NNTB for a ≥ 50% response 6 (95% CI 4 to 11).
Mean change in physical function from baseline (0–100 WOMAC Total score, 0 is best function) between 0 and 14 weeks	The mean improvement in function was 10.51 pointsa	The mean improvement in function was 16.16 points	MD −5.65 (−7.08 to −4.23)	1909 (6 RCTs)	⊕⊕⊕⊕ High	Small, possibly clinically unimportant improvement with antidepressants.
Quality of life (EuroQol, scale −0.11 to 1.0 where a score of 1.0 indicates perfect health) between 0 and 14 weeks	The mean improvement was 0.07 pointsc	The mean improvement was 0.11 points	MD 0.04 (0.01 to 0.07)	815 (3 RCTs)	⊕⊕⊕⊝ Moderated	Very small, clinically unimportant change with antidepressants.
Proportion of participants who withdrew due to adverse events between 0 and 14 weeks	5.1%	10.9% (7.9 to 15%)	RR 2.15 (1.56 to 2.97)	1977 (6 RCTs)	⊕⊕⊕⊝ Moderated	6% more people withdrew (95% 3% to 10%) with antidepressants. NNTH 17 (95% CI 10 to 35).
Total adverse events between 0 and 16 weeks	49.3%	64.1% (58.6 to 70%)	RR 1.27 (1.15 to 1.41)	2101 (9 RCTs)	⊕⊕⊕⊕ High	14.8% more people reported an adverse event (95% CI 9% more to 21% more) with antidepressants. NNTH 7 (95% 5 to 11).
Serious adverse events between 0 and 16 weeks	1.7%	1.6% (0.8 to 3.3)	RR 0.94  (0.46 to 1.94)	2101 (9 RCTs)	⊕⊕⊝⊝ Lowd,e	No more people reported serious adverse events with antidepressants (0.1% more, 95% CI 0.9% less to 1.6% more). NNTH 1000.
*The risk in the intervention group (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). CI: confidence interval; MD: mean difference; NNTB: number needed to treat for an additional beneficial outcome; NNTH: number needed to treat for an additional harmful outcome; RCT: randomised controlled trial; RR: risk ratio.
GRADE Working Group grades of evidence High certainty: we are very confident that the true effect lies close to that of the estimate of the effect. Moderate certainty: we are moderately confident in the effect estimate: the true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different. Low certainty: our confidence in the effect estimate is limited: the true effect may be substantially different from the estimate of the effect. Very low certainty: we have very little confidence in the effect estimate: the true effect is likely to be substantially different from the estimate of effect.

^aThemean pain reduction from baseline in control group was 1.73 and the mean improvement in function from baseline was 10.51 (Wang 2017).
^bAbsolute benefit was calculated as the improvement in the intervention group minus the improvement in the control group (MD), in the original units, and expressed as a percentage.
^cAs reported in Uchio 2018.
^dDowngraded for imprecision due to the low event rates or small number of participants.
^eDowngraded one level for imprecision due to very low event rates of 10 with antidepressant and 11 with placebo.

Background

Description of the condition

Osteoarthritis is the most common form of arthritis. Hip and knee osteoarthritis combined is one of the leading causes of global disability, ranked as the 11th highest contributors in the most recent Global Burden of Disease study (Vos 2012). The global prevalence of osteoarthritis is estimated to be 3.8% for knees and 0.85% for hips (Cross 2014). It is a multifactorial condition with the main risk factors being advanced age, sex, obesity and joint injury.

Pathologically, osteoarthritis affects structures of the entire joint. The structural changes include articular cartilage erosion and loss; subchondral bone oedema; microfracture and cyst formation; with development of osteophytes (bony outgrowths) and changes in the periarticular  ligaments, bursae and muscles (Felson 2006). These structural changes are classically identified using radiographs, although magnetic resonance imaging (MRI) has been used recently to detect early structural features of osteoarthritis. Radiographically the main characteristics of osteoarthritis are joint space narrowing, as a proxy for loss of cartilage, and the presence of osteophytes (Heidari 2011).

These structural changes are accompanied by the main clinical characteristics of pain, loss of function and stiffness. People with osteoarthritis consider pain to be the most important symptom they experience (Tallon 2000). However, the causes of pain in osteoarthritis are multifactorial, complex and poorly understood (Thakur 2014). With increasing severity of radiographic osteoarthritis, the presence of pain increases, although the severity of pain is not strongly related to the structural severity of the disease (Felson 2005). Whilst some structural changes within the joint itself affecting bone and the synovium may lead to local tissue damage that stimulates pain fibres (nociceptive pain), cartilage, often considered the target tissue in this condition, is aneural. It is becoming increasingly evident that maladaptive pain‐processing mechanisms also occur, with central pain sensitisation contributing to the pain experience in people with osteoarthritis (Thakur 2014).

There are no approved disease‐modifying treatments for osteoarthritis. Weight loss in overweight or obese people with lower limb osteoarthritis has been shown to decrease self‐reported disability, with variable changes in pain level (Christensen 2007); however, sustained weight loss is difficult to achieve. Current treatment strategies focus on the relief of symptoms and optimising function. This may involve the use of simple analgesia, including paracetamol and non‐steroidal anti‐inflammatory drugs (NSAIDs), of which NSAIDs are more effective (Towheed 2006). Intra‐articular injections have modest benefits compared to placebo, with effect reducing over the six months post‐injection (Jüni 2015). Physical therapy strategies, including exercise (Fransen 2014; Fransen 2015), ultrasound (Rutjes 2010), transcutaneous electrical nerve stimulation (TENS) (Rutjes 2009), and hydrotherapy (Bartels 2007) are also used in symptom management of osteoarthritis. These may provide important decreases in pain and better functional ability, but usually only for a short time (less than 12 months). Opiate medications, such as tramadol, are also used to control severe pain, but may be limited by adverse effects such as nausea, vomiting and constipation (Cepeda 2006), and may be associated with cognitive dysfunction, falls, addiction and death (Benyamin 2008).

However, despite the prevalence of osteoarthritis, the management of pain in osteoarthritis is often ineffective (Black 2013). This may be because pain is multifactorial (Neogi 2013). The structural severity of joint disease is associated with the presence of pain, but not the intensity of pain. This suggests that not all pain in osteoarthritis is due to joint disease. For example, pain sensitisation, which relates to changes in the brain and spinal cord that leads to persistent pain, is present in many people with osteoarthritis (Zolio 2021). Pain with these features is less responsive to oral analgesics, including NSAIDs and opioids (Nijs 2011; Woolf 2011), and treatment modalities (O'Leary 2018). Thus alternative, more effective analgesic strategies for pain in osteoarthritis are required. 

Description of the intervention

Drug therapy is one of the most commonly used therapeutic interventions available for the management of osteoarthritis and the control of pain. The drugs usually used to treat osteoarthritis include paracetamol (Towheed 2006), NSAIDs and opiates (Cepeda 2006), which are used for their nociceptive pain management effect. Antidepressant medications comprise one approach to the management of osteoarthritis, as they target central pain‐producing mechanisms responsible for pain sensitivity and neuropathic pain (Dharmshaktu 2012; Dworkin 2007).

For the purpose of this review, we defined antidepressants according to the groupings suggested by the Cochrane Common Mental Disorders Group (cmd.cochrane.org/).

• Tricyclic antidepressants (TCAs): for example, amitriptyline, imipramine, trimipramine, doxepin, desipramine, protriptyline, nortriptyline, clomipramine, dothiepin, lofepramine.

• Heterocyclic antidepressants: for example, mianserin, trazodone, amoxapine, maprotiline.

• Selective serotonin reuptake inhibitors (SSRIs): for example, fluvoxamine, fluoxetine, paroxetine, sertraline, citalopram, escitalopram.

• Monoamine oxidase inhibitors (MAOIs):

  • irreversible: for example, phenelzine, tranylcypromine, isocarboxazid;

  • reversible: for example, brofaromine, moclobemide, tyrima.

• Other antidepressants:

  • Noradrenaline reuptake inhibitors (NRIs): for example, reboxetine, atomoxetine

  • Noradrenaline dopamine reuptake inhibitors (NDRIs): for example, amineptine, bupropion

  • Serotonin noradrenaline reuptake inhibitors (SNRIs): for example, venlafaxine, milnacipran, duloxetine

  • Noradrenaline and serotonin specific antidepressants (NaSSAs): for example, mirtazapine

  • Serotonin antagonist and reuptake inhibitors (SARIs): for example, trazodone

  • Unclassified: for example, agomelatine, vilazodone

Antidepressant medication use may be limited by numerous adverse effects. TCAs can cause dry mouth, drowsiness, confusion, urinary retention, weight gain, nightmares, cardiac conduction abnormalities and orthostatic hypotension (MIMS 2015). MAOIs can cause orthostatic hypotension, dizziness, reflex tachycardia, weight gain and sedation (MIMS 2015). SSRIs and SNRIs have common adverse effects such as drowsiness, dry mouth, nervousness, anxiety, insomnia, increased appetite, long‐term weight gain and decreased sexual function (MIMS 2015). Antidepressants are not known to be addictive, but people may experience withdrawal effects, especially if the drug is stopped suddenly (Richards 2011).

When TCAs are used to treat depression, they are generally given at a higher maximum dose than when used for the treatment of neuropathic pain (eTG 2015). In other studies, antidepressants used for chronic pain syndromes, such as fibromyalgia, have also been used in lower doses than in people with depression (Häuser 2009).

How the intervention might work

Antidepressants, including TCAs, SNRIs and SSRIs,  target central pain‐producing mechanisms. Although they have been used, they have limited efficacy in non‐specific low back pain (Urquhart 2010), and with good efficacy in neuropathic conditions such as post‐herpetic neuralgia, diabetic neuropathy and trigeminal neuralgia (Saarto 2007). Their effect on pain has been attributed to the inhibition of reuptake of serotonin or noradrenaline or both in the brain and spinal cord (Nijs 2011). The neurotransmitters serotonin and noradrenaline are involved in both the pathological basis of depression (for which SSRIs and SNRIs are predominately prescribed), and also in pain perception (Dharmshaktu 2012). When peripheral nerves receive pain input it is transmitted to the spinal cord for processing by the brain. A homeostatic mechanism to mediate this pain input involves serotonin and noradrenaline in the spinal cord. These neurotransmitters help to decrease the ascending pain messages by enhancing inhibitive signals descending from the brain to the periphery. Disorders of serotonin and noradrenaline may therefore diminish descending pain inhibition, and increase the person's perception of pain (Dharmshaktu 2012).

Antidepressant medications aim to increase the concentration of serotonin and noradrenaline in the brain and spinal cord, thereby decreasing pain perception. Whilst depression and increased pain perception have been linked in the literature (Dharmshaktu 2012), antidepressant medications have been shown to exert their effect on pain independently of their effect on depressive symptoms (Chappell 2009a). This has been shown in both people with osteoarthritis and in fibromyalgia pain (Chappell 2009a). A further endogenous homeostatic mechanism to reduce pain perception is the activation and release of opioids (Nijs 2011); opiate‐based medications are often used for the treatment of severe pain. Activation of opioid receptors in the brain and spinal cord also have an inhibitory effect on pain perception (Nijs 2011). TCAs activate this system, as well as altering serotonin and noradrenaline concentrations. Thus, they may also be effective in pain modulation (Dharmshaktu 2012).

Why it is important to do this review

Most people with osteoarthritis do not achieve satisfactory pain control (Black 2013). Thus, there is an urgent need for more effective pain control strategies. The use of duloxetine, an SNRI, is conditionally recommended by the Osteoarthritis Research Society International (Bannuru 2019). Wielage 2014 undertook a modelling exercise that showed the implementation of duloxetine was cost‐effective (compared to usual management with celecoxib, diclofenac, naproxen, hydromorphone and extended‐release oxycodone) in the management of osteoarthritis. Citrome 2012 conducted a systematic review of duloxetine for the management of osteoarthritic pain. It found duloxetine to be an efficacious and tolerable treatment approach for people in this population. However, studies have reported a reducing level of effect of antidepressants for treatment of pain in osteoarthritis over time (Abou‐Raya 2012; Chappell 2009a; Chappell 2011).

One systematic review and meta‐analysis of the use of antidepressants in back pain and osteoarthritis could not exclude a clinically important effect in osteoarthritis and did not examine whether more people achieved a significant, worthwhile benefit from antidepressants than those who were not thus treated (Moore 2014). There is currently no comprehensive systematic review available to fully examine the effectiveness, safety and tolerability of overall antidepressant use in osteoarthritis. Therefore, it is important to conduct our review to clarify the current evidence on the current therapy for osteoarthritis.

This review was conducted according to the guidelines recommended by the Cochrane Musculoskeletal Group Editorial Board (Ghogomu 2014).

Objectives

To assess the benefits and harms of antidepressants for the treatment of symptomatic knee and hip osteoarthritis in adults.

Methods

Criteria for considering studies for this review

Types of studies

We included randomised controlled trials (RCTs) reported as full‐text, published as abstract only and unpublished data. There was no language restriction. We excluded cluster‐RCTs and cross‐over study designs to avoid any carryover effect.

Types of participants

We included adults, as defined in the trials, with a diagnosis of hip and knee osteoarthritis, where possible, using the American College of Rheumatology clinical and radiographic criteria (for the hip: Altman 1991; for the knee: Altman 1986). We did not exclude studies involving participants with conditions that may otherwise require antidepressant prescription (e.g. depression and fibromyalgia).

Types of interventions

We included trials comparing the use of antidepressant medication combined, and stratified by class according to the groupings suggested by the Cochrane Common Mental Disorders Group (Description of the intervention).

We planned to compare antidepressant versus placebo as our primary comparison, followed by antidepressants versus analgesic interventions (including paracetamol, naproxen, indomethacin or other NSAIDs, or cyclo‐oxygenase inhibitors), with other antidepressant medications, or with any other active intervention (including those that were not pharmacological, e.g. exercise therapy).

We included studies longer than six weeks of therapy to ensure a therapeutic drug concentration was obtained.

We included the following co‐interventions provided they were not part of the randomised treatment, and were given to all treatment arms: analgesics (such as paracetamol, naproxen, indomethacin or other NSAIDs, or cyclo‐oxygenase inhibitors) and conservative therapies (such as exercise, weight loss, ultrasound, acupuncture or others).

Types of outcome measures

Major outcomes

• Pain with a hierarchy of 11 levels (when more than one was reported, we used the highest on the list)

  • Pain overall

  • Pain on walking

  • Western Ontario and McMaster Universities Arthritis Index (WOMAC) Pain subscale

  • Pain on activities other than walking

  • WOMAC Global scale

  • Lequesne Osteoarthritis Index Global score

  • Other algofunctional scale

  • Patient's global assessment

  • Physician's global assessment

  • Other outcome

  • No continuous outcome reported

• Responder rate; proportion of participants who had a 50% or greater reduction in pain.

• Physical function with a hierarchy of eight levels (when more than one was reported, we used the highest on the list):

  • Global disability score

  • Walking disability

  • WOMAC Disability subscore

  • Composite disability scores other than WOMAC

  • Disability other than walking

  • WOMAC Global scale

  • Lequesne Osteoarthritis Index Global score

  • Other algofunctional scale

• Quality of life: 36‐item Short Form (SF‐36); EuroQol 5‐Dimensions (EQ‐5D); Sickness Impact Profile (SIP); Nottingham Health Profile (NHP); other

• Proportion of participants who withdrew due to adverse events

• Proportion of participants who experienced any adverse events (total adverse events)

• Proportion of participants who experience serious adverse events

Minor outcomes

• Proportion meeting the OARSI Response Criteria (Pham 2004a), which uses WOMAC subscales and a patient global assessment (of well‐being, using a visual analogue scale (VAS)) to define those who respond to the intervention for osteoarthritis. A patient was a responder if they met either of the following criteria:

  • a high improvement in pain or in function of 50% or greater and absolute change 20% or greater; or

  • improvement in at least two of the following three:

    • pain 20% or greater and absolute change 10% or greater;

    • function 20% or greater and absolute change 10% or greater;

    • global assessment 20% or greater and absolute change 10% or greater

• Radiographic joint structure changes according to the given hierarchy (the first two outcomes, minimum joint space width and median joint space width, are mostly used)

  • Minimum joint space width

  • Median joint space width

  • Semi‐quantitative measurement

• Proportion of participants who dropped out of the study for any reason, as a measure of intervention acceptability.

Timing of outcome assessment

We aimed to use outcomes measured for at least 12 weeks in total, however this depended on what was reported. We extracted data at the 12‐week time point, and also the longest time point reported, as a surrogate measure of longer‐term effectiveness. If data were not reported at 12 weeks (e.g. if the study duration was less than 12 weeks), we extracted the latest time point data that was available. Where different studies reported different time points of data, we combined data points from 12 weeks or fewer, and reported these as short‐term outcomes. If present, we combined data points from six months onwards to illustrate longer‐term outcomes. If there were multiple time points, for example, 4, 6 and 12 months, we extracted the latest time point.

Search methods for identification of studies

Electronic searches

We searched the Cochrane Central Register of Controlled Trials (CENTRAL), Ovid MEDLINE, Ovid Embase, CINAHL and PsycINFO up to 13 January 2021.

For assessments on adverse effects, we searched the websites of the regulatory agencies US Food and Drug Administration – MedWatch (www.fda.gov/Safety/MedWatch/default.htm), European Medicines Evaluation Agency (www.emea.europa.eu), Australian Adverse Drug Reactions Bulletin (www.tga.gov.au/adr/aadrb.htm), and UK Medicines and Healthcare products Regulatory Agency (MHRA) pharmacovigilance and drug safety updates (www.mhra.gov.uk).

We also conducted a search of ClinicalTrials.gov (www.ClinicalTrials.gov) and the World Health Organization (WHO) International Clinical Trials Registry Platform (ICTRP) (www.who.int/ictrp/en/).

We searched all databases from their inception to 13 January 2021, and we imposed no restriction on language of publication.

See Appendix 1 for the search strategies used.

Searching other resources

We checked reference lists of all relevant primary studies and review articles for additional references. We searched relevant manufacturers' websites for trial information. This included any of the pharmaceutical company trial registers: AstraZeneca (www.astrazenecaclinicaltrials.com/); Eli Lilly and Company (www.lillytrials.com/); GlaxoSmithKline (ctr.gsk.co.uk/medicinelist.asp); Bristol‐Myers Squibb (ctr.bms.com/ctd/registry.do); Novartis (www.novartisclinicaltrials.com/webapp/etrials/home.do); Roche (www.roche-trials.com/registry.html); and Wyeth (www.wyeth.com/ClinicalTrialListings).

We searched for errata or retractions from included studies published in full‐text on PubMed (www.ncbi.nlm.nih.gov/pubmed).

Data collection and analysis

Selection of studies

Two review authors (JL, AL) independently screened titles and abstracts for inclusion of all records identified by the search and coded them as 'retrieve' (eligible or potentially eligible/unclear) or 'do not retrieve'. We retrieved the full‐text reports and pairs of review authors (JL, AL, LC, JS) independently screened the full‐texts and identified studies for inclusion, and identified and recorded reasons for exclusion of the ineligible studies. We resolved any disagreements through discussion or, if required, we consulted a third review author (AW). We identified and excluded duplicates and collated multiple reports of the same study so that each study, rather than each report, was the unit of interest in the review. We recorded the selection process in sufficient detail to complete a PRISMA flow diagram (Moher 2009) and Characteristics of excluded studies table.

Data extraction and management

We used a data collection form for study characteristics, which was piloted on at least one study in the review. One review author (AL) extracted study characteristics from included studies. A second review author (AW) spot‐checked study characteristics for accuracy against the trial report. We extracted the following study characteristics.

• Methods: study design, total duration of the study, details of any 'run‐in' period, number of study centres and location, study setting, withdrawals and year of publication of the study.

• Participants: number, mean age, age range, sex, radiographic disease, severity of pain, diagnostic criteria, inclusion criteria and exclusion criteria.

• Interventions: intervention, comparison, concomitant medications, other co‐interventions and excluded medications.

• Outcomes: primary and secondary outcomes specified and collected, and time points reported.

• Characteristics of the design of the trial as outlined at Assessment of risk of bias in included studies.

• Notes: funding for the trial, and notable declarations of interest of trial authors.

Two review authors (AL, JS) independently extracted outcome data from included studies. We extracted the number of events and participants per treatment group for dichotomous outcomes, and means and standard deviations (SDs) and number of participants per treatment group for continuous outcomes. We noted in the Characteristics of included studies table if outcome data were not reported in a usable way and when data were transformed or estimated from a graph. We resolved disagreements by consensus or by involving a third review author (AW). One review author (AL) transferred data into the Review Manager 5 (Review Manager 2014). We double‐checked that data were entered correctly by comparing the data presented in the systematic review with the study reports.

If we identified a mixture of change‐from‐baseline scores (change scores) with their SDs, and final measurement (final values) with their SDs, we prioritised the reporting of change scores on the basis that these remove some of the contribution of between‐person variability, and they are a more useful outcome if the data presented have a skewed distribution (Deeks 2021). However, if final values were explicitly reported and change scores needed to be imputed, we included final values, considering that change scores and final values could be combined in a meta‐analysis when using the (unstandardised) mean difference method.

If both unadjusted and adjusted values for the same outcome were reported, we extracted the unadjusted values. If a mixture of data were available, we adjusted data as required. If data were analysed based on an intention‐to‐treat (ITT) sample and another sample (e.g. per‐protocol, as‐treated), we extracted data based on the ITT principle. If a trial reported multiple time points, we extracted the longest time point available.

Main planned comparisons

Our main planned comparison was antidepressants stratified by class versus placebo including TCAs versus placebo or active comparator, SSRIs versus placebo or active comparator, SNRIs versus placebo or active comparator, heterocyclic antidepressants versus placebo or active comparator and MAOIs versus placebo or active comparator. We planned to pool studies with hip and knee as at the time of protocol publication, knee and hip osteoarthritis were considered to behave similarly and it is usual practice in studies of osteoarthritis pain (Hawker 2008).

The other antidepressants (as per Description of the intervention) were to be grouped against placebo or other active comparator. The secondary comparison was antidepressants versus alternative analgesics including paracetamol, NSAIDs or other antidepressants. 

Assessment of risk of bias in included studies

Review authors (JL, AL, JS) independently assessed risk of bias using the RoB 1 tool for each study using the criteria outlined in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2017). We resolved any disagreements by discussion or by involving another review author (AW). We assessed the risk of bias according to the following domains.

• Random sequence generation

• Allocation concealment

• Blinding of participants and personnel

• Blinding of outcome assessment

• Incomplete outcome data

• Selective outcome reporting

• Other bias: major imbalances in key baseline characteristics, unplanned interim analysis, unequal application of co‐interventions

We graded each potential source of bias as high, low or unclear risk, and provided a quote from the study report together with a justification for our judgement in the risk of bias table. We summarised the risk of bias judgements across different studies for each of the domains listed. We considered blinding separately for different key outcomes where necessary (e.g. for unblinded outcome assessment, risk of bias for all‐cause mortality may be different from for a participant‐reported pain scale). In addition, we considered the impact of missing data by key outcomes.

Where information on risk of bias related to unpublished data or correspondence with a trialist, we noted this in the risk of bias table.

When considering treatment effects, we considered the risk of bias for the studies that contributed to that outcome.

We presented the figures generated by the risk of bias tool to provide summary assessments of the risk of bias.

Assessment of bias in conducting the systematic review

We conducted the review according to the published protocol and reported any deviations from it in the Differences between protocol and review section of the systematic review (Lyttle 2016).

Measures of treatment effect

We analysed dichotomous data as risk ratios (RR) or Peto odds ratio (OR) when the outcome was a rare event (approximately less than 10%), with 95% confidence intervals (CIs). We analysed continuous data as mean difference (MD) or standardised mean difference (SMD, when studies used different scales to measure an outcome), and 95% CIs. We entered data presented as a scale with a consistent direction of effect across studies.

When we calculated SMDs (e.g. disability) we back‐translated them to a typical scale (e.g. 0 to 10 for pain) by multiplying the SMD by a typical among‐person SD (e.g. the SD of the control group at baseline from the most representative trial) (Higgins 2021).

As some studies included normalised data, we normalised all WOMAC Function data to a 0 to 100 scale.

For dichotomous outcomes, we calculated the number needed to treat for an additional beneficial outcome (NNTB) or the number needed to treat for an additional harmful outcome (NNTH), in the case of adverse events, from the control group event rate and the relative risk using the Visual Rx NNT calculator (Cates 2008). Rather than using minimal important differences (MIDs), we judged the magnitude of the effect based upon between‐mean group differences for continuous measures based upon a similar approach used in the American College of Physicians 2017 guidelines for low back pain (Chou 2017). For pain measured on a 0‐ to 10‐point scale, we considered a difference of 0.5 to 1.0 points as slight to small, a difference greater than 1 to 2 points as moderate and a difference greater than 2 points as large. Similarly, for function and health‐related quality of life measured on a 0‐ to 100‐point scale, we considered a difference of 5 to 10 points as slight to small, greater than 10 to 20 points as moderate and a difference of greater than 20 points as large.

For dichotomous outcomes, we calculated the absolute risk difference using the Risk Difference statistic in Review Manager 5 and expressed the result as a percentage (Review Manager 2014).

Unit of analysis issues

Where a trial reported multiple arms, we included only the relevant arms. If two comparisons (e.g. drug A versus placebo and drug B versus placebo) were combined in the same meta‐analysis, we halved the control group to avoid double‐counting. If more than two interventions groups were present in any study, we noted them in the Characteristics of included studies table.

Dealing with missing data

We contacted investigators or study sponsors in order to verify key study characteristics and obtain missing numerical outcome data where possible (e.g. when a study is identified as abstract only or when data are not available for all participants). When this was not possible, and if the missing data were thought to introduce serious bias, we explored the impact of including such studies in the overall assessment of results by a sensitivity analysis. Any assumptions and imputations to handle missing data were clearly described and the effect of imputation explored by sensitivity analyses.

For dichotomous outcomes (e.g. number of withdrawals due to adverse events), we calculated the withdrawal rate using the number of participants randomised in the group as the denominator.

For continuous outcomes (e.g. mean change in pain score), we calculated the MD or SMD based on the number of participants analysed at that time point. If the number of participants analysed was not presented for each time point, we used the number of randomised participants in each group at baseline.

Where possible, we computed missing SDs from other statistics such as standard errors, CIs or P values, according to the methods recommended in the Cochrane Handbook for Systematic Reviews of Interventions (Deeks 2021). If SDs could not be calculated, we imputed them (e.g. from other studies in the meta‐analysis).

Assessment of heterogeneity

We assessed clinical and methodological diversity in terms of participants, definition of osteoarthritis, interventions, outcomes and study characteristics for the included studies to determine whether a meta‐analysis was appropriate by observing the data from the data extraction tables. We assessed statistical heterogeneity by visual inspection of the forest plot to determine obvious differences in results between the studies, and using the I² and Chi² statistics.

As recommended in the Cochrane Handbook for Systematic Reviews of Interventions (Deeks 2021), the interpretation of an I² value of 0% to 40% might not be important; 30% to 60% may represent moderate heterogeneity; 50% to 90% may represent substantial heterogeneity and 75% to 100% represents considerable heterogeneity. As noted in the Cochrane Handbook for Systematic Reviews of Interventions, we kept in mind that the importance of the I² statistic depends on the magnitude and direction of effects and the strength of evidence for heterogeneity.

We interpreted the Chi² test where a P ≤ 0.10 indicated evidence of statistical heterogeneity.

If we identified substantial heterogeneity, we reported it and investigated possible causes by following the recommendations in Chapter 10 of the Cochrane Handbook for Systematic Reviews of Interventions (Deeks 2021). 

Assessment of reporting biases

We planned to create and examine a funnel plot to explore possible small‐study biases. In interpreting funnel plots, we planned to examine the different possible reasons for funnel plot asymmetry as outlined in Section 10.4 of the Cochrane Handbook for Systematic Reviews of Interventions and related this to the results of the review. If we were able to pool more than 10 trials, we planned to undertake formal statistical tests to investigate funnel plot asymmetry, and follow the recommendations in Section 10.4 of the Cochrane Handbook for Systematic Reviews of Interventions (Page 2021).

To assess outcome reporting bias, we checked trial protocols against published reports. For studies published after 1 July 2005, we screened the WHO ICTRP for the a priori trial protocol (www.who.int/ictrp/en/). We evaluated whether there was selective reporting of outcomes.

Data synthesis

Where present, we stratified data according to antidepressant class of drug used against a common comparison, for example, SSRIs versus placebo, then SNRIs versus placebo and estimated the overall effect. We used a random‐effects meta‐analysis model, as we assumed that some clinical diversity and methodological heterogeneity was likely to exist in the studies. We planned to present results for hip and knee combined. We assessed large joints of the lower limb (hip and knee) in combined, as these are often managed as such.

Subgroup analysis and investigation of heterogeneity

We planned to carry out the following subgroup analyses.

• Pain in hip and knee separately as the prevalence of pain sensitisation and neuropathic pain differs according to joint, and that this is hypothesised to be the potential mechanism by which antidepressants may exert an effect.

• People with other chronic conditions (e.g. fibromyalgia, concurrent depression), as the presence of these conditions may affect the measurement of outcomes.

We used the pain outcome in subgroup analyses.

We used the formal test for subgroup interactions in Review Manager 5 (Review Manager 2014), and used caution in the interpretation of subgroup analyses as advised in Chapter 10 of the Cochrane Handbook for Systematic Reviews of Interventions (Deeks 2021). We compared the magnitude of the effects between the subgroups by assessing the overlap of the CIs of the summary estimated. Non‐overlap of the CIs indicated statistical significance.

Sensitivity analysis

We planned to perform the primary meta‐analysis restricted to studies with low risk of bias across the key domains of: sequence generation; allocation concealment; blinding of assessors, participants and personnel; completeness of outcome data; selectivity of outcome reporting and any other area of bias identified on review of the trial, and a sensitivity analysis that included all studies regardless of risk of bias. This enabled confidence that any plausible bias was unlikely to have affected our results. Given the biases identified in most studies, studies with unclear or high risk of bias were included as otherwise there would be no trials for analysis.

If the analysis of heterogeneity found one or more outlying studies with results that conflicted with the other studies, we performed sensitivity analysis to assess the influence of these studies on the results of the meta‐analysis.

We planned to carry out the following sensitivity analyses, for pain, to:

• determine whether the overall results were the same when studies with an unclear or high risk of just selection and detection were included

• determine whether the overall results were the same when studies with co‐interventions (that are equally applied) were included;

• determine whether, if the analysis of heterogeneity found one or more outlying studies with conflicting results, these results affected the results of the meta‐analysis.

Interpreting results and reaching conclusions

We followed the guidelines in the Chapter 15 of the Cochrane Handbook for Systematic Reviews of Interventions (Schünemann 2021a), for interpreting results and were aware of distinguishing a lack of evidence of effect from a lack of effect. We based our conclusions only on findings from the quantitative or narrative synthesis of included studies for this review. We avoided making recommendations for practice; our implications for research suggested priorities for future research and outlined the remaining uncertainties in the area.

Summary of findings and assessment of the certainty of the evidence

We created a summary of findings table using the following outcomes: pain, responder rate, physical function, quality of life, proportion of participants who withdrew due to adverse events, proportion of participants who experienced any adverse events and the proportion of participants who experienced serious adverse events. The comparison was antidepressants versus placebo, and we included participants with hip and knee osteoarthritis. 

Two review authors (AL, JS) independently assessed the certainty of the evidence. We used the five GRADE considerations (study limitations, consistency of effect, imprecision, indirectness and publication bias) to assess the certainty of a body of evidence as it related to the studies that contributed data to the meta‐analyses for the prespecified outcomes. We used methods and recommendations described in Chapter 14 of the Cochrane Handbook for Systematic Reviews of Interventions (Schünemann 2021b) using GRADEpro GDT software (GRADEpro GDT). We justified all decisions to downgrade the certainty of the evidence using footnotes and we made comments to aid the reader's understanding of the review where necessary.

In the comments column of the summary of findings table, we provided the absolute per cent difference and NNTB for dichotomous outcomes.

Results

Description of studies

For study details, see the Characteristics of included studies; Characteristics of excluded studies; Characteristics of studies awaiting classification; and Characteristics of ongoing studies tables.

Results of the search

The results of the search are illustrated in Figure 1. The search returned 865 results (857 through database searching and eight through other sources). There remained 736 records after removal of duplicates. We removed 684 records after reading the titles and abstracts. We assess the full text of 52 records and excluded 34 records, with reasons. We included nine trials, four studies are awaiting classification and five studies are ongoing. The nine included trials enrolled 2122 participants.

1.

Study flow diagram.

All nine included trials were registered in a trial registry (ClinicalTrials.gov or the WHO ICTRP).

Included studies

Nine trials were included in this analysis (see Characteristics of included studies table for full details).

Trial design and setting

All trials were RCTs. The trials were conducted in 11 countries including Canada (Chappell 2011), China (Wang 2017), Germany (Riesner 2008), Greece (Chappell 2011), Japan (Uchio 2018), New Zealand (Hudson 2021), Puerto Rico (Chappell 2009b; Frakes 2011), Romania (Chappell 2009b), Russia (Chappell 2011), Sweden (Chappell 2011), and the US (Chappell 2009b; Chappell 2011; Frakes 2011; Mathur 2015; Tetreault 2016). Six trials were pharmaceutical industry‐funded, receiving funding from Eli Lilly and Company (Chappell 2009b; Chappell 2011; Frakes 2011; Tetreault 2016; Uchio 2018; Wang 2017).

Participants

The studies included people with osteoarthritis involving the knees (Chappell 2009b; Chappell 2011; Frakes 2011; Hudson 2021; Mathur 2015; Riesner 2008; Tetreault 2016; Uchio 2018; Wang 2017), and hips and knees (Riesner 2008; Wang 2017). No participants had prior joint prostheses or concomitant depression. The trials' durations were 10 to 14 weeks. Within the trials, the mean age was 64.0 years in the antidepressant cohort and 63.0 years in the placebo cohort. About 69.2% of participants were female in the antidepressant group and 70.1% were female in the placebo cohort. The mean duration of disease was 6.4 years in the antidepressant cohort and 6.0 years in the placebo. Osteoarthritis pain severity on a 0 to 10 standardised scale was 5.6 in the antidepressant cohort and 5.7 in the placebo cohort using a 24‐hour mean pain score.

Interventions

Antidepressants included the SNRIs duloxetine (Chappell 2009b; Chappell 2011; Frakes 2011; Tetreault 2016; Uchio 2018; Wang 2017) and milnacipran (Mathur 2015); the SSRI fluvoxamine (Riesner 2008); and the TCA nortriptyline (Hudson 2021). Antidepressant dosages were duloxetine 60 mg to 120 mg, milnacipran up 200 mg, fluvoxamine 150 mg and nortriptyline 25 mg to 100 mg. Five trials had participants on a stable dose of antidepressant (Mathur 2015; Riesner 2008; Tetreault 2016; Uchio 2018; Wang 2017), whereas three trials used a dose escalation of antidepressant if there was little pain improvement (Chappell 2011; Frakes 2011; Hudson 2021). One trial had a random re‐allocation to remain on duloxetine 60 mg versus increase to duloxetine 120 mg at week seven of the intervention (Chappell 2009b). All studies included an induction phase and a tapering phase to minimise potential adverse effects.

Several trials used additional analgesia in both intervention and placebo groups (Chappell 2009b; Chappell 2011; Frakes 2011; Hudson 2021; Riesner 2008; Tetreault 2016), whereas other studies discontinued all potential analgesics at the start of the studies (Mathur 2015; Uchio 2018; Wang 2017). Episodic use of rescue medication (NSAID or paracetamol) was allowed for a duration of up to three continuous days throughout all the studies.

Excluded studies

After reviewing full texts, we excluded 34 studies from this review (see Characteristics of excluded studies table).

Thirty were not RCTs (Alaka 2013; Bellingham 2010; Brown 2013; Brunton 2010; Chen 2019; Citrome 2012a; Citrome 2012b; Driban 2012; Ferreira 2021; Fishbain 2000; Gaynor 2010; Gaynor 2013; Hochberg 2012; Moore 2014; Myers 2012; Osani 2019; Peng 2013; Pergolizzi 2013; Sabourin 2013; Sawynok 2003; Skljarevski 2009; Skljarevski 2011; Sullivan 2009a; Sullivan 2009b; Talati 2010; Teran Estrada 2011; Wang 2015; Weng 2020; Williamson 2014; Wohlreich 2013).

Thorpe 1974 also included people with rheumatoid arthritis, and there was no subgroup analysis, precluding its inclusion.

In Lin 2003, the effects of antidepressants could not be delineated from the co‐intervention of psychotherapy. Two trials had mixed patient populations including other types of arthritis, in which people with osteoarthritis could not be separated for analysis (Thorpe 1974; Wohlreich 2009). Sofat 2017 examined hand osteoarthritis only.

Studies awaiting classification

Four studies are awaiting classification. One controlled trial was excluded from our analysis despite fitting inclusion criteria as other studies by the authors have been retracted and the results are under investigation (Abou‐Raya 2012). Eli Lilly and Company 2009 met inclusion criteria; however, this trial had a significant medication labelling error identified and analysis was discontinued. Glick 1976 has no results available. Schiphof 2015 was prematurely ended. We sought further results but they were unavailable for inclusion in this review.

Ongoing studies

Five studies are currently ongoing and may meet the inclusion criteria once completed (ACTRN12615000301561; ACTRN12619001082190; İlleez 2020; IRCT20170716035126N2; ntr4798; see Characteristics of ongoing studies table).

Risk of bias in included studies

The results from the risk of bias assessment are displayed in Figure 2 and Figure 3.

2.

Risk of bias graph: review authors' judgements about each risk of bias item presented as percentages across all included studies.

3.

Risk of bias summary: review authors' judgements about each risk of bias item for each included study.

Allocation

Most studies used computer‐generated systems to randomly allocate participants and concealed allocation. Five studies were at low risk of selection bias (Chappell 2009b; Chappell 2011; Hudson 2021; Uchio 2018; Wang 2017). One trial was at unclear risk selection bias as whilst it was said to be randomised, there was no information on how this was accomplished (despite attempts made to contact the study authors) (Frakes 2011). Two studies were at unclear risk due to lack of details (Mathur 2015; Tetreault 2016). Riesner 2008 was computer generated (low risk of bias) but unclear bias for allocation concealment as this was not described.

Blinding

Five studies reported that participants and personnel were blinded to group allocation and were at low risk of performance bias (Chappell 2009b; Chappell 2011; Frakes 2011; Uchio 2018; Wang 2017). Four studies were at unclear risk of performance bias as there was no information on blinding (Hudson 2021; Mathur 2015; Riesner 2008; Tetreault 2016).

Seven studies were at low risk of detection bias for self‐reported outcomes due to participants blinding (Chappell 2009b; Chappell 2011; Frakes 2011; Hudson 2021; Tetreault 2016; Uchio 2018; Wang 2017). Two studies were at unclear risk of detection bias for self‐reported outcomes as there were no details provided regarding methods of blinding (Mathur 2015; Riesner 2008).

Eight studies had no separate assessor‐reported outcomes, relying on self‐reported outcomes (Chappell 2009b; Chappell 2011; Frakes 2011; Hudson 2021; Mathur 2015; Riesner 2008; Tetreault 2016; Wang 2017). One study provided assessor‐reported outcomes (radiological assessment of joint x‐rays) where there was unclear risk of bias as there were no details on the method of researcher blinding (Uchio 2018). 

Incomplete outcome data

Two studies were at high risk of attrition bias, one because of high dropout rates (Chappell 2009b), and one because there was a significant difference in dropout rate between groups (29.4% with antidepressants versus 10.3% with placebo) (Mathur 2015). Two studies were at high risk of attrition bias as there were no details regarding dropouts (Riesner 2008; Tetreault 2016); the study protocol for Tetreault 2016 noted that if a participant dropped out then they would be replaced.

Four studies had overall high levels of completion and were at low risk of attrition bias (Chappell 2011; Hudson 2021; Uchio 2018; Wang 2017). Frakes 2011 and Uchio 2018 detailed the reasons for participant withdrawal within their publications.

Selective reporting

Six trials reported all their outcomes either on an online database (ClinicalTrials.gov) or in their published reports and were at low risk of reporting bias (Chappell 2009b; Chappell 2011; Frakes 2011; Hudson 2021; Riesner 2008; Uchio 2018). One trial was at high risk of reporting bias as it has not been formally published (results only available on ClinicalTrials.gov) and hence was not accessible in regular database searches (Mathur 2015).One trial was at high risk of reporting bias as it did not provide important baseline data (Tetreault 2016). One trial was at high risk of reporting bias as the Hospital Anxiety and Depression Scale was not reported whilst a non‐prespecified outcome 'concomitant use of short‐acting analgesics' was reported (Wang 2017).

Other potential sources of bias

Riesner 2008 was at unclear risk for other bias as there were no baseline data available, which ensured that any baseline differences between groups was unable to be considered. Mathur 2015 reported baseline differences between the placebo and intervention groups and, in addition had outcome criteria changes after an interim analysis. It was at high risk of other bias. The remaining studies had no other potential sources of bias.

Effects of interventions

See: Table 1

Antidepressant compared with placebo

Nine studies compared antidepressant versus placebo (Chappell 2009b; Chappell 2011; Frakes 2011; Hudson 2021; Mathur 2015; Riesner 2008; Tetreault 2016; Uchio 2018; Wang 2017). See Table 1.

Major outcomes

Pain at 8 to 16 weeks

Nine studies reported pain at the end of treatment (which was 8 to 16 weeks). Five studies measured pain using the Brief Pain Inventory (BPI), three used the WOMAC, one used a VAS (not otherwise specified) and one used the McGill Pain Questionnaire. Results were presented as MDs, normalised to 0 to 10 for all. We were unable to use SMD as the data presentation from Wang precluded this.

High‐certainty evidence indicates that antidepressants lead to a small, clinically unimportant reduction in pain. The mean reduction in placebo was 1.73 points on a 0‐ to 10‐point numerical rating scale, and was further reduced by 0.59 points with antidepressants (95% CI −0.88 to −0.31; I² = 66%; 9 trials, 2038 participants; Analysis 1.1; Table 1).

1.1. Analysis.

Comparison 1: Antidepressants versus placebo in hip and knee osteoarthritis, Outcome 1: Mean pain at 8–16 weeks

For different classes of antidepressants used, there was no evidence of a difference in response between the subgroups (P = 0.07). The SNRI studies had a  mean reduction in pain of 0.68 points lower (95% CI −0.98 to −0.38; 7 studies, 1773 participants). The SSRI study had a mean pain reduction of 0.24 points (95% CI 0.47 lower to 0.95 higher; 1 study, 205 participants). This indicates potentially no difference between antidepressant and placebo cohorts. The TCA study had a mean reduction in pain of 0.56 points lower (95% CI −1.22 to 0.10; 1 study, 60 participants) (Analysis 1.1).

Responder rate (50% or greater pain reduction)

Six trials included 50% or greater pain reduction as an outcome. There was high‐certainty evidence that a small proportion of participants had a 50% or greater pain reduction (RR 1.55, 95% CI 1.32 to 1.82; I² = 40%; 1904 participants; Analysis 1.2; Table 1). About 28.6% of participants receiving placebo and 45.2% (95% CI 37.5% to 54.9%) of participants receiving antidepressant had a 50% or greater reduction in pain. This gave an absolute improvement of 16% (95% CI 8.9% to 26%) with antidepressants. The NNTB 6 (95% CI 4 to 11).

1.2. Analysis.

Comparison 1: Antidepressants versus placebo in hip and knee osteoarthritis, Outcome 2: Responder rate: ≥ 50% reduction in pain

In the SNRI studies, there was an RR of 1.58 (95% CI 1.31 to  1.92; 5 studies, 1703 participants), and in the TCA study, the RR was 1.43 (95% CI 1.07 to 1.92; 1 study, 201 participants). There was no subgroup difference (Analysis 1.2; P = 0.57).

Physical function

Six trials reported WOMAC Total score for measurement of physical function. There was high‐certainty evidence that treatment with antidepressants leads to a small, clinically unimportant improvement in function. The mean improvement with placebo was 10.51 points and with antidepressants was 16.6 points on a normalised 0 to 100 scale, where 0 was best function (MD −5.65 points, 95% CI −7.08 to −4.23; 1909 participants; Analysis 1.3; Table 1).

1.3. Analysis.

Comparison 1: Antidepressants versus placebo in hip and knee osteoarthritis, Outcome 3: Physical function (WOMAC)

The SNRI studies had an MD of 5.68 lower (95% CI −7.34 to −4.02; 5 studies, 1704 participants), and the TCA study had an MD of 5.20 points lower (95% CI −11.32 lower to 0.92; 1 study, 205 participants). There was no subgroup difference (Analysis 1.3; P = 0.88).

Quality of life

Several studies included quality‐of‐life measurements. We analysed the data presented on EQ‐5D as this provided an overall quality of life score, which was present in three of the SNRI (duloxetine) cohort. Moderate‐certainty evidence (downgraded for small number of participants) demonstrated a very small, clinically unimportant improvement of quality of life with duloxetine. The mean improvement with placebo was 0.07 points and with antidepressants was 0.11 points (MD 0.04, 95% CI 0.01 to 0.07; 3 studies, 815 participants; Analysis 1.4; Table 1), which is clinically unimportant on a −0.11 to 1 scale where 1 indicates perfect health.

1.4. Analysis.

Comparison 1: Antidepressants versus placebo in hip and knee osteoarthritis, Outcome 4: Mean change in quality of life at 10–14 weeks

Study withdrawals due to adverse events

We measured tolerability as the proportion of participants withdrawing from the trials due to adverse events. Moderate‐certainty evidence (downgraded for imprecision) from six trials showed that 11% of participants receiving antidepressants withdrew from trials due to an adverse event compared to 5% of participants receiving placebo (RR 2.15, 95% CI 1.56 to 2.97; 1977 participants; Analysis 1.5; Table 1). The NNTH for withdrawals due to adverse events was 17 (10 to 35). This indicated 6% more people withdrew with an antidepressant (95% CI 3% to 10%) with a relative withdrawal rate of 115% (95% CI 56% to 197%).

1.5. Analysis.

Comparison 1: Antidepressants versus placebo in hip and knee osteoarthritis, Outcome 5: Proportion of study withdrawals due to adverse effects

The SNRI studies had an RR for withdrawal of 2.17 (95% CI 1.57 to 3.01; 5 studies, 1772 participants), and the TCA study had an RR of 1.01 (95% CI 0.06 to 15.93; 205 participants). There was no subgroup difference (Analysis 1.5; P = 0.59).

Total adverse events

High‐certainty evidence showed that total adverse events were higher in the antidepressant group (64%) compared to the placebo group (49%). The relative risk of an antidepressant treatment‐related adverse event was 1.27 (95% CI 1.15 to 1.41; 9 studies, 2101 participants; Analysis 1.6; Table 1), which is 14.8% more (95% CI 9% to 21%) people reporting an adverse event with the antidepressant. The NNTH was 7 (95% CI 5 to 11).

1.6. Analysis.

Comparison 1: Antidepressants versus placebo in hip and knee osteoarthritis, Outcome 6: Total adverse events

The SNRI studies had an RR of 1.30 (95% CI 1.19 to 1.42; 7 studies, 1840 participants), the SSRI studies had an RR of 1.33 (95% CI 1.04 to 1.72; 1 study, 60 participants), and the TCA study had an RR of 1.14 (95% CI 1.05 to 1.23; 1 study, 201 participants). There were no subgroup differences (Analysis 1.6; P = 0.06).

Serious adverse events

Low‐certainty evidence (downgraded twice for imprecision for very low numbers of events) indicated no difference in serious adverse events between antidepressant (1.6%) and placebo (1.7%) (RR 0.94, 95% CI 0.46 to 1.94; 9 trials, 2101 participants; Analysis 1.7). The NNTH was 1000. This means 0.1% less people reported serious adverse events with antidepressants (−0.9% to 1.6%).

1.7. Analysis.

Comparison 1: Antidepressants versus placebo in hip and knee osteoarthritis, Outcome 7: Serious adverse events

The SNRI studies had an RR of 1.03 (95% CI 0.42 to 2.54; 7 studies, 1840 participants), the SSRI study had an RR of 0.25 (95% CI 0.03 to 2.11; 1 study, 60 participants), and the TCA study had an RR of 1.37 (95% CI 0.32 to 5.98; 1 study, 201 participants). There were no subgroup differences (Analysis 1.7; P = 0.41).

Minor outcomes

OARSI responders

Five studies reported on whether participants had a clinically important outcome in line with the OARSI criteria (Methods). High‐certainty evidence demonstrated a response to antidepressant compared to placebo, with an RR of response to medication of 1.72 (95% CI 1.49 to 1.98; 5 studies, 1517 participants; Analysis 1.8).

1.8. Analysis.

Comparison 1: Antidepressants versus placebo in hip and knee osteoarthritis, Outcome 8: OARSI responders

The SNRI studies had an RR of 1.74 (95% CI 1.48 to 2.04; 4 studies, 1329 participants), and the TCA study had an RR of 1.60 (95% CI 1.05 to 2.42). There was no subgroup difference (Analysis 1.8; P = 0.71).

Radiographic joint structure changes

Radiographic changes were not widely reported. Uchio 2018 (354 participants) reported the Kellgren Lawrence radiographic criteria showing the risk of joint progression to be 0.99 (95% CI 0.94 to 1.04), indicating no evidence of an effect, but not on joint space width as specified in our protocol. Hence, no further analysis was possible.

Total treatment withdrawals

High‐certainty evidence demonstrated more withdrawals with antidepressants compared to placebo (RR 1.38, 95% CI 1.09 to 1.77; 7 studies, 2023 participants; Analysis 1.10). There was a 5.9% more (95% CI 2.4% to 10.3%) chance of withdrawal in the antidepressant cohort.

1.10. Analysis.

Comparison 1: Antidepressants versus placebo in hip and knee osteoarthritis, Outcome 10: Total treatment withdrawals

The SNRI studies had an RR of 1.37 (95% CI 1.06 to 1.77; 6 studies, 1818 participants), and the TCA study had an RR of 3.03 (95% CI 0.32 to 28.64; 1 study, 205 participants). There was no subgroup difference (Analysis 1.10; P = 0.49).

Subgroup analyses

Comparison of hip versus knee was unable to be completed as all studies but two (Riesner 2008; Wang 2017), examined only the knee. However, although Riesner 2008 included the hip, they did not specify the number of participants with hip and knee involvement. Although Wang 2017 also included the hip, only 3/407 participants had hip osteoarthritis and their data were not presented separately. We performed no subgroup analyses in this review due to the low number of included studies, and lack of information provided regarding the presence or absence of other chronic conditions (depression and fibromyalgia).

Sensitivity analysis

Risk of bias

We performed sensitivity analysis on level of bias for selection and detection bias. The analysis for mean change in pain was used for reference in the sensitivity analysis.

Selection bias

We removed the results of four studies for the sensitivity analysis as they had unclear selection bias (Frakes 2011; Mathur 2015; Riesner 2008; Tetreault 2016). There was no change in mean pain by removal of these studies (MD −0.68, 95% CI −0.86 to −0.49) compared to the original analysis for change in pain (MD −0.59, 95% CI −0.88 to −0.31).

Detection bias

We removed the results of two for the sensitivity analysis (Mathur 2015; Riesner 2008). There was no significant change in mean pain with their results removed (MD 0.64, 95% CI −0.91 to −0.36) compared to the original analysis for change in pain (MD −0.59, 95% CI −0.88 to −0.31).

Co‐interventions

There were no additional co‐interventions in any of the studies.

Heterogeneity

Pain

There was substantial heterogeneity (I² = 66%). When we removed two studies from the analysis (Riesner 2008; Tetreault 2016), heterogeneity reduced to I² = 0% and the MD for change in pain changed from −0.59 (95% CI −0.88 to −0.31) to −0.74 (95% CI −0.91 to −0.57), which was a very small amount on a 10‐point scale. 

Responder rate

There was moderate heterogeneity (I² = 40%). When we removed one study from the analysis (Frakes 2011), heterogeneity reduced to I² = 0%; however, there was no change in the overall results.

Physical function

There was low heterogeneity (I² = 3%), and consequently, sensitivity analysis was not necessary.

Quality of life

There was moderate heterogeneity (I² = 47%). When we removed one study from the analysis (Chappell 2011), heterogeneity reduced to I² = 0%; however, there was no change in the outcome of the analysis.

Withdrawal due to adverse events

There was no heterogeneity (I² = 0%), and consequently, sensitivity analysis was not necessary.

Total adverse events

There was moderate heterogeneity (I² = 47%), with an  overall RR for adverse events of 1.27 (95% CI 1.15 to 1.41). When we removed the one outlying study from the analysis (Mathur 2015), heterogeneity reduced to I² = 43%; however, there was no change in the result overall (RR 1.26, 95% CI 1.16 to 1.39).

Serious adverse events

There was no heterogeneity (I² = 0%), and consequently, sensitivity analysis was not necessary.

OARSI responder

There was no heterogeneity (I² = 0%), and consequently, sensitivity analysis was not necessary.

Radiographic joint structure changes

No sensitivity analysis was possible on radiographic changes in joints as only one study reported these results (Uchio 2018).

Total withdrawals

There was low heterogeneity (I² = 27%), and consequently, sensitivity analysis was not necessary.

Antidepressants versus analgesic interventions or other antidepressants

No studies compared antidepressants to alternative analgesics including paracetamol, NSAIDs or other antidepressants. Several studies had participants who used NSAIDs; however, were in both the placebo and antidepressant cohorts and no comparison was available.

Discussion

Summary of main results

Antidepressants for hip and knee osteoarthritis

The review found that compared with placebo, antidepressants provided a small but not clinically important improvement in overall pain, based on high‐certainty evidence (9 studies, 2122 participants). Those taking antidepressants had a small improvement in pain from baseline on a 0 to 10 scale (MD −0.59, 95% CI −0.88 to −0.31). The studies reported whether there was a clinically important improvement in pain using two different methods; the percentage achieving a significant or moderate response as per IMMPACT guidelines (Dworkin 2008), and those meeting OARSI working group criteria for a clinically important improvement (Pham 2004a). Using the IMMPACT guidelines, participant response was categorised as to whether pain was improved by 50% or greater (a significant response). Combined, the studies show that participants receiving an antidepressant were 55% more likely to have a clinically important reduction in pain compared to those receiving placebo (31% to 92% more likely to have a clinically important reduction in pain). However, the studies' results showed a moderate degree of variation in their results (I² = 40%). There were similar results when using the OARSI Response Criteria, with participants receiving an antidepressant more likely to achieve a clinically important response compared to those receiving placebo (RR 1.72, 95% CI 1.41 to 1.98). Using the IMMPACT guidelines response, six people would need to be treated for one person to show a 50% or greater improvement in pain. Using the OARSI guidelines, five people would require antidepressant treatment for one person to demonstrate a clinically significant improvement.

All studies measured function using the WOMAC Total score. All studies consistently demonstrated an improvement in function from baseline. However, the combined improvement was small, with an MD in improvement of −5.65 on a scale of 0 to 100 (improvement of as much as −7.08 to −4.23). Given the small response, this may be clinically unimportant.

Studies used various measures to assess the effect of an antidepressant on quality of life. Three studies used the EQ‐5D scale. These showed an improvement in quality of life of 0.04 on a scale ranging from −1 to 11 (as little as 0.01 to as much as 0.07). Although the minimally clinically important difference for quality of life has yet to be defined using this scale, this appears to be a small improvement.

The smaller response in quality‐of‐life measures with antidepressants may be related to the higher number of adverse events in the antidepressant group compared to placebo. The risk of having an adverse effect whilst receiving an antidepressant was 1.27 times that of placebo (as little as 1.15 times to as much as 1.41 more likely to have an adverse effect). Combining the data, the NNTH for any adverse effect was 7. There was no increase in the risk of having a serious adverse event using antidepressant compared to placebo (RR 0.94, 95% CI 0.46 to 1.94). However, this was low‐certainty evidence due to very small numbers of events.

The tolerability of antidepressants relates to the number of participants who discontinued the study due to adverse effects compared to the control group. Antidepressants were less well tolerated than placebo in the meta‐analysis. The risk of discontinuing the study if receiving antidepressants compared to placebo was 2.15 (95% CI 1.56 to 2.97). One in 17 participants discontinued antidepressants due to adverse effects (NNTH). This correlated with a higher likelihood of study withdrawal overall in the antidepressant cohort compared with placebo, with the risk of study withdrawal being 1.38 (95% CI 1.09 to 1.77).

Antidepressants do not appear to have any effect on radiographic joint progression, although only one study reported this outcome (Uchio 2018). They reported the risk of joint progression to be 0.99 (95% CI 0.94 to 1.04).

Overall completeness and applicability of evidence

This review included nine RCTs with 2122 study participants across 11 countries. SNRIs, SSRIs and TCAs were all represented in this review. However, duloxetine (an SNRI) had the most data available. The studies included in this discussion are for large joint (hip, knee) lower limb osteoarthritis.

The included studies compared an antidepressant to placebo. All studies allowed regular paracetamol use in both control and antidepressant groups, and Chappell 2009b, Chappell 2011, Frakes 2011, and Riesner 2008 permitted participants to use NSAIDs throughout the study. There were comparison to active placebos or another antidepressant. All participants in Frakes 2011 received concomitant NSAIDs. Wang 2017 did not allow the use of NSAIDs. Between 22.7% and 56.8% of study participants used NSAIDs for 14 days or greater in the three months prior to the study commencing in Chappell 2009b and Uchio 2018. This may reflect the use of non‐prescription medications commonly used in the community.

The age of participants extended from 53.0 to 80.2 years. This is generally representative of the age range presenting with osteoarthritis in the standard clinical setting. The overall mean ages were 63.0 years in the placebo groups and 64.0 years in the antidepressant groups. The populations in these studies may be representative of a younger cohort than the total population experiencing osteoarthritis given that the prevalence of osteoarthritis increases with age (Heidari 2011).

Of the studies included in this review, Chappell 2009b, Chappell 2011, Mathur 2015, Hudson 2021, Tetreault 2016, and Uchio 2018 focused on osteoarthritis of the knee, whilst Wang 2017 and Riesner 2008 included participants with osteoarthritis of the knee and hip. Thus, we analysed trials in participants with knee and hip osteoarthritis together. We were unable to perform subgroup analyses according to joint, as only three participants in Wang 2017, and an unknown number in Riesner 2008 had hip osteoarthritis. Exclusion of Riesner 2008 is unlikely to change the results of the review significantly. Thus, the results are most generalisable to people with knee osteoarthritis. There is less certainty about whether these findings are applicable to pain related to hip osteoarthritis. As the prevalence of neuropathic‐like pain in knee osteoarthritis is higher than in hip osteoarthritis (Zolio 2021), and the mechanism of action of antidepressants is thought to work by centralised pain mechanisms, the effect may differ according to the joint. Thus, future studies should assess the effect of antidepressant therapy on either hip or knee osteoarthritis. It may be inappropriate to combine these analyses.

To evaluate safety in older participants, Uchio 2018 subdivided older (aged 65 years and above) versus younger (aged less than 65 years) participants. Chappell 2011 presented a post hoc analysis of participants in the studies by Chappell 2009b and Chappell 2011 to evaluate the safety and efficacy of older (aged 65 years and above) compared to younger (aged less than 65 years) participants. Uchio 2018 found no difference in adverse effects, whereas Chappell 2011 demonstrated a greater rate of adverse effects in older participants, with duloxetine associated with adverse effects in 22.3% of older participants versus 7.5% of younger participants, and placebo in 11.8% of older participants versus 4.5% in younger participants (P < 0.001).

Given the extensive exclusion criteria with regard to comorbid illnesses, participants in the study populations may be overall healthier than those seen in standard clinical practice. This is especially relevant when considering psychiatric conditions. Pain and depression are linked, with 29% to 86% of people experiencing chronic pain having concomitant depression, perhaps through interactions of fatigue, disability and emotional stress (Bair 2003; Linton 2011). Consequently, the exclusion of people with depression or major psychiatric problems further limits the applicability of the results to this subset of patients.

Studies used several pain scales to assess the efficacy of the antidepressants. The most commonly used were the BPI – Severity (BPI) and the WOMAC Pain subscale. Other scales that incorporated pain were commonly found in secondary outcomes including the Clinical Global Impression (CGI) scale and the Patient Global Impression of Improvement (PGI‐I). We used the BPI 24‐hour mean pain or a visual analogue scale of pain as these were the most widely reported. The studies were powered towards this outcome. This was in accordance with the hierarchy outlined in our protocol (Lyttle 2016).

Many of the scales used to measure treatment effect also incorporated functional assessment. These scales included the WOMAC Physical Function subscale, SF‐36 and BPI – Interference. We used the WOMAC Physical Function scale for analysis. It is specific for osteoarthritis and was prescribed in the study protocol.

There were two methods to demonstrate whether the response to treatment achieved a clinically important improvement. Many of the included studies reported the proportion of participants achieving a 30% or greater or a 50% or greater reduction in mean pain, which was interpreted as the response to treatment. This is in accordance with the IMMPACT recommendations (Dworkin 2008). Fewer studies used the OARSI Response Criteria. Frakes 2011 and Uchio 2018 presented these data and Hochberg 2012  published a post hoc analysis of Chappell 2009b and Chappell 2011, providing the OARSI response results. Hudson 2021 provided results of OARSI responses on request.

Studies used several scales to examine whether treatment was related to an improvement in quality of life. This analysis focused on quality of life, rather than specifically on anxiety and depression, given the exclusion of people with depression from the included studies. Three studies used the SF‐36 (Chappell 2009b; Chappell 2011; Uchio 2018). As the results of the various components reported varied, this precluded the use of this measure in our analyses. Three studies used the EQ‐5D to evaluate quality of life (Chappell 2009b; Chappell 2011; Uchio 2018). The overall effect on quality of life was difficult to determine because studies used different scales and were furthermore not powered towards quality of life as their primary outcome.

The studies in this review had a duration of 8 to 16 weeks. Due to the variation in study duration and the unavailability of raw data, it was not possible to measure clinical outcomes at the same time point in all studies. A post hoc analysis of trial data by Williamson 2014 demonstrated that whilst some participants experienced an early analgesic effect, it could take up to six weeks for duloxetine to become effective in osteoarthritis, and that some people were unlikely to ever have an analgesic response. Consequently, we considered that the results from six weeks onwards are applicable to analgesia outcomes. The efficacy of antidepressants in lower limb osteoarthritis over the long term has not been studied.

The rate of adverse events in both the placebo and antidepressant groups was high (49.3% with placebo and 64.1% with antidepressants). The RR of having an adverse effect when taking antidepressants compared with placebo was 1.27 (95% CI 1.15 to 1.41). The NNTH for any adverse effect was 7. The most common adverse effects were nausea, dry mouth and constipation. Adverse effects were well reported by all nine trials with tables available for all adverse effects on ClinicalTrials.gov. As these were well reported and were consistent with known adverse effects of antidepressants in the literature, we considered that these results are applicable to people using antidepressants in osteoarthritis. The incidence of adverse effects with differing doses of antidepressants was either not reported or there was no difference between participants taking duloxetine 60 mg and those taking 120 mg in Chappell 2009b. None of the trials measured toxicity beyond the end of the study (maximum 16 weeks).

The incidence of serious adverse effects was very low in the included studies. Due to a lack of events, we were unable to determine whether there were any differences between the groups. Serious adverse effects did not appear to be clearly associated with antidepressant use.

We identified five studies ongoing studies (ACTRN12615000301561; ACTRN12619001082190; İlleez 2020; IRCT20170716035126N2; ntr4798). Two studies are awaiting classification (Abou‐Raya 2012; Glick 1976). Glick 1976 focused on imipramine but had sufficient data for inclusion in the review. Abou‐Raya 2012 is awaiting classification because of concerns regarding the accuracy of the data. A post hoc sensitivity analysis was not possible for pain using results from Abou‐Raya 2012 as change in pain data were not available. These studies may be included in updates of this review.

Quality of the evidence

The certainty of the evidence in this review ranged from high to low. Certainty was high for pain, function, response to treatment and adverse effects. According to the GRADE working group, further research is unlikely to alter effect estimates for duloxetine for pain in an unselected population. Single studies evaluated the TCA nortriptyline (Hudson 2021), and the SSRI fluvoxamine (Riesner 2008). Although their results were similar, the potential mechanisms may differ thus a clinically important effect could not be excluded. 

Quality of life and withdrawal due to adverse effects were assessed to be evidence of moderate certainty. These were downgraded for imprecision due to the low event rates or small number of participants. Whilst we believe that these represent likely true trends, further research into this would likely have an impact on results and the confidence that these results reflect the true clinical situation.

The evidence surrounding serious adverse effects was further downgraded to low certainty given the exceedingly low occurrence rate; 15 serious adverse events in the antidepressant cohort and 18 serious adverse events in the placebo cohort out of 2101 total participants. We postulate this could be secondary to the length of the studies (8 to 16 weeks). Consequently, the true effect may differ from the estimated effect. Given the low occurrences of these events in the clinical setting, further research would be of benefit.

Six included studies examined duloxetine. Five of these were industry sponsored (Chappell 2009b; Chappell 2011; Frakes 2011; Uchio 2018; Wang 2017). Due to the limited number of trials, a funnel plot could not be performed for visual assessment of publication bias.

Given that all nine trials included in our study were double‐blind, placebo RCTs, there was a reduced risk of performance or detection bias. However, generalisability maybe limited as many studies excluding depression and other comorbidities that are often concurrent with osteoarthritis (i.e. cardiovascular disease). We assessed four trials at high risk of attrition bias secondary to high dropout rates or lack of details surrounding study discontinuation, or both, and the consequent impacts this could have upon result interpretation (Chappell 2009b; Mathur 2015; Riesner 2008; Tetreault 2016). Three trials were at high risk for reporting bias with issues concerning lack of publication of data or failure to publish the complete results (Mathur 2015; Tetreault 2016; Wang 2017). However, we note that the results, whilst fully published at ClinialTrials.gov were incompletely reported in the published manuscripts.

There are five ongoing studies examining the effect of other antidepressants on pain in osteoarthritis.

Potential biases in the review process

This analysis was performed after an extensive review of the literature, including many databases, and searching the references of identified review articles to ensure that studies were not missed. The literature was searched in duplicate by two of four separate review authors, each acting independently. An additional review author was available to assist and resolve any disagreements. Two separate, independent review authors checked data input for errors. None of the review authors in this analysis had conflicts of interest. This reduced the risk of both bias and error in our review.

This review was limited by the available data. In view of the low numbers of studies included, we were unable to create funnel plots to examine publication bias. Finally, studies had to have specific data available (i.e. baseline measures) and use appropriate scales (i.e. BPI). Two studies were awaiting classification as data were unavailable, possibly introducing reporting bias (i.e. Glick 1976; Abou‐Raya 2012).

We note that several included studies have the same pharmaceutical industry funding. However, we feel that this did not impact our independent review of antidepressants in osteoarthritis.

Agreements and disagreements with other studies or reviews

There have been multiple systematic reviews focusing on the role of duloxetine as an analgesic agent in osteoarthritis, including  Brown 2013, Chen 2019, Citrome 2012a, Hochberg 2012, Myers 2012, and Wang 2015. One review did not include other antidepressants in their analysis of the effect of antidepressants on osteoarthritis, neither did the authors examine the proportions with clinically important benefits (Ferreira 2021). On analysis, these reviews have focused upon either pain or adverse effect safety profile. There has been less of a focus on the overall tolerability of duloxetine as an analgesic agent, which was only been reported in a systematic review (Chen 2019). The number of included studies in each analysis was temporally related to those available in the literature, and there was intermittent inclusion of the outcomes from Abou‐Raya 2012. Our review is the only one to estimate the tolerability and clinical significance of the effect size, and incidence of clinically important response related to multiple antidepressants in osteoarthritis.

The reviews on analgesia focused on statistically significant reductions in pain across the group and measured change in the BPI pain scores. In addition, these reviews examined whether important improvement (a 30% or 50% or greater reduction in pain scores on a 10‐point scale) was observed in each participant and determined the proportion of responders. In our review, we included both mean reductions in BPI values and the percentage achieving a clinically important response as per the IMMPACT guidelines. Most previously published reviews were very positive regarding the use of antidepressants, mainly duloxetine, in osteoarthritis. For example, the electronically published review by Chen 2019 discussed the statistical reductions in pain score (BPI MD −0.74, 95% CI −0.92 to −0.57), but did not report if this was clinically significant. We attempted to demonstrate that overall, this effect is very small. While Ferreira 2021 concluded that a clinically important effect could not be excluded for osteoarthritis, they examined pain as a continuous variable normalised to a 0 to 100 scale and did not examine the effect of antidepressants on a clinically important improvement in pain. Our overall findings were similar across an unselected population. However, when we examined responsiveness, we found more participants experienced a clinically important improvement in pain with duloxetine compared to placebo. As pain is multifactorial in osteoarthritis and the underlying mechanisms may vary between individuals, it is possible that only individuals with a specific underlying pain mechanisms will be responsive to this therapy. We do note that several of these meta‐analyses and review articles received funding in some form from the pharmaceutical industry (i.e. Alaka 2013; Brown 2013; Brunton 2010; Chappell 2011; Citrome 2012a; Citrome 2012b; Hochberg 2012; Moore 2014; Myers 2012; Skljarevski 2009; Skljarevski 2011; Talati 2010; Teran Estrada 2011; Williamson 2014).

Function has also been reported in the literature as it is often considered a secondary outcome. Wang 2015 described a 4.25‐point reduction in pain scores (95% CI −5.82 to −2.68) on a 96‐point scale. We found a slightly larger difference, but did use a normalised scale to 100 points, given the information available. Overall, although there was a functional improvement, given the tools used, we have classified the functional improvement as small.

Other reviews focused on the safety profiles with duloxetine (e.g. Alaka 2013; Brunton 2010; Gaynor 2010). These studies examined different potential influences on adverse effect profiles including race (Gaynor 2010) and age (Alaka 2013). They found no differences in the rate of adverse effects within these subgroups. Brunton 2010 pooled data from several medical conditions, including chronic fibromyalgia and chronic back pain. They found that there were significantly higher numbers of adverse effects with antidepressants compared placebo (72.4% with antidepressants versus 57.2% with placebo for any adverse effect) irrespective of indication or demographic subgroup. These are higher rates than we demonstrated (64.1% with antidepressants versus 49.3% with placebo). However, the numbers of adverse effects in each group were consistent. The most commonly reported adverse effects were nausea, dry mouth, constipation, dizziness, fatigue and somnolence. Reports of serious adverse effects are very low in the literature. These have only been reported for the duration of the study, for no longer than 16 weeks in trial participants with osteoarthritis. 

Our analysis also included tolerability, which is less widely reported upon. We measured tolerability according to the number of participants who withdrew due to adverse effects. Hochberg 2012 reported that the number needed to discontinue due to adverse effects in the duloxetine cohort was 10 (95% CI 7 to 20); this is lower than our reported 17 (95% CI 10 to 35). Chen 2019 defined tolerability using adverse effects, without relating whether these led to study discontinuation. Wang 2015 commented in their discussion that there were higher levels of discontinuation in the duloxetine group, but conversely stated that the majority of adverse effects were well tolerated. We are more cautious in our analysis given the rate of discontinuation of the antidepressant due to adverse effects (NNTH 17).

Authors' conclusions

Implications for practice.

Based on moderate‐ and high‐certainty evidence, although antidepressant use in lower limb osteoarthritis improves pain, function and quality of life, the degree of improvement is small and not clinically important across the whole population. However, also based on high‐certainty evidence, the proportion of people achieving a clinically meaningful improvement in pain is higher in people taking antidepressants. This suggests the possibility that there may be a subpopulation that is more likely to respond to this therapy.

Antidepressant use is associated with an increased number of adverse events compared with placebo and consequentially had reduced tolerability. Serious adverse events were present in all trials, but with no preponderance towards either the antidepressants or placebo. The adverse events may limit the use of antidepressants in the community and the patients' comorbidities need to be considered.

Implications for research.

Future research on duloxetine as an analgesic agent in an unselected osteoarthritis population is unlikely to bring any significant new information regarding pain and functional outcomes. Further research into the use of other antidepressant agents in osteoarthritis is warranted as they may have differing effects to duloxetine, which was most heavily represented in our review.

Further studies could consider the effect of antidepressants in people selected on the basis of  pain phenotypes, and in a single joint, as factors affecting pain in different anatomical joints may differ. As pain is multifactorial in osteoarthritis and there was high level of evidence for a higher likelihood of a clinically important improvement in treated participants compared to untreated participants, identifying those more likely to benefit would reduce the number of people needed to treat for an additional beneficial outcome. For example, people with pain sensitisation may be more likely to respond to antidepressant therapy, or to a specific antidepressant. The mechanism of effect of antidepressants is hypothesised to be by affecting the serotonergic pain pathways in the central nervous system, reducing central sensitisation. Thus studies targeting the use of antidepressants in individuals with a central‐sensitisation or neuropathic‐like pain phenotype may be more effective than its use in an unselected population.

For serious adverse events, further research will continue to provide information on safety profiles.

History

Protocol first published: Issue 4, 2016

Date	Event	Description
26 August 2016	Amended	Correction of author initials

Appendices

Appendix 1. MEDLINE search strategy

1. exp osteoarthritis/

2. osteoarthr$.tw.

3. (degenerative adj2 arthritis).tw.

4. arthrosis.tw.

5. or/1‐4

6. exp Antidepressive Agents/

7. exp Antidepressive Agents, Second‐Generation/

8. antidepressants.mp.

9. (anti‐depressant$ or antidepressant$).tw.

10. exp Serotonin Uptake Inhibitors/ or exp Citalopram/ or exp Serotonin Antagonists/ or exp Fluoxetine/ or SSRI$.tw.

11. selective serotonin reuptake inhibitors.mp.

12. (Citalopram or Celexa).tw.

13. (Fluoxetine or Prozac).tw.

14. (Paroxetine or Paxil or Seroxat).tw.

15. (Sertraline or Zoloft or Lustral).tw.

16. (Escitalopram or Lexapro or Cipralex).tw.

17. (Fluvoxamine or Luvox).tw.

18. (Desvenlafaxine or Pristiq).tw.

19. (Venlafaxine or Effexor).tw.

20. (Duloxetine or Cymbalta).tw.

21. (Milnacipran or Ixel or Savella).tw.8

22. (Reboxetine or Edronax).tw.

23. (Viloxazine or Vivalan).tw.

24. exp Antidepressive Agents, Tricyclic/

25. tricyclic antidepressants.mp.

26. tricyclic antidepressant$.tw.

27. exp Amitriptyline/

28. (Amitriptyline or Elavil or Endep).tw.

29. (Clomipramine or Anafranil).tw.

30. (Desipramine or Norpramin or Pertofrane).tw.

31. (Dosulepin or Dothiepin or Prothiaden).tw.

32. (Doxepin or Adapin or Sinequan).tw.

33. (Imipramine or Tofranil).tw.

34. (Lofepramine or Feprapax or Gamanil or Lomont).tw.

35. (Nortriptyline or Pamelor).tw.

36. (Protriptyline or Vivactil).tw.

37. (Trimipramine or Surmontil).tw.

38. (Amoxapine or Asendin).tw.

39. (Loxapine or Loxpac or Loxitane).tw.

40. (Maprotiline or Deprilept or Ludiomil or Psymion).tw.

41. (Mazindol or Mazanor or Sanorex).tw.

42. (Mianserin or Bolvidon or Norval or Tolvon).tw.

43. (Mirtazapine or Remeron or Avanza or Zispin).tw.

44. (Setiptiline or Tecipul).tw.

45. exp Monoamine Oxidase Inhibitors/

46. monoamine oxidase inhibitors.mp.

47. Monoamine oxidase Inhibitor$.tw.

48. (Isocarboxazid or Marplan or Moclobemide or Aurorix or Manerix or Phenelzine or Nardil or Selegiline or L‐Deprenyl or Eldepryl or Zelapar or Emsam or Trancypromine or Parnate or Moclobemide or Aurorix or Manerix).tw.

49. exp Phenelzine/

50. exp Moclobemide/

51. exp Selegiline/

52. exp Tranylcypromine/

53. exp Isocarboxazid/

54. serotonin‐norepinephrine reuptake inhibitor$.tw.

55. norepinephrine reuptake inhibitor$.tw.

56. serotonin‐noradrenaline reuptake inhibitor$.tw.

57. noradrenaline reuptake inhibitor$.tw.

58. (Duloxetine or Cymbalta or Milnacipram or Ixel or Savella).tw.

59. (Ruboxetine or Edronax or Viloxazine or Vivalan).tw.

60. exp Tryptophan/

61. L‐tryptophan.mp.

62. or/6‐61

63. randomized controlled trial.pt.

64. controlled clinical trial.pt.

65. randomised.ab.

66. placebo.ab.

67. drug therapy.fs.

68. randomly.ab.

69. trial.ab.

70. groups.ab.

71. or/63‐70

72. exp animals/ not humans.sh.

73. 71 not 72

74. 5 and 62 and 73

Appendix 2. Cochrane Library search strategy

1. MeSH descriptor: [Osteoarthritis] explode all trees

2. osteoarthr*

3. (degenerative near/2 arthritis)

4. arthrosis

5. #1 or #2 or #3 or #4

6. MeSH descriptor: [Antidepressive Agents] explode all trees

7. MeSH descriptor: [Antidepressive Agents, Second‐Generation] explode all trees

8. antidepressants

9. (anti‐depressant* or antidepressant*)

10. MeSH descriptor: [Serotonin Uptake Inhibitors] explode all trees

11. MeSH descriptor: [Citalopram] explode all trees

12. MeSH descriptor: [Serotonin Antagonists] explode all trees

13. MeSH descriptor: [Fluoxetine] explode all trees

14. SSRI*

15. selective serotonin reuptake inhibitors

16. (Citalopram or Celexa)

17. (Fluoxetine or Prozac)

18. (Paroxetine or Paxil or Seroxat)

19. (Sertraline or Zoloft or Lustral)

20. (Escitalopram or Lexapro or Cipralex)

21. (Fluvoxamine or Luvox)

22. (Desvenlafaxine or Pristiq)

23. (Venlafaxine or Effexor)

24. (Duloxetine or Cymbalta)

25. (Milnacipran or Ixel or Savella)

26. (Reboxetine or Edronax)

27. (Viloxazine or Vivalan)

28. MeSH descriptor: [Antidepressive Agents, Tricyclic] explode all trees

29. tricyclic antidepressants

30. tricyclic antidepressant*

31. MeSH descriptor: [Amitriptyline] explode all trees

32. (Amitriptyline or Elavil or Endep)

33. (Clomipramine or Anafranil)

34. (Desipramine or Norpramin or Pertofrane)

35. (Dosulepin or Dothiepin or Prothiaden)

36. (Doxepin or Adapin or Sinequan)

37. (Imipramine or Tofranil)

38. (Lofepramine or Feprapax or Gamanil or Lomont)

39. (Nortriptyline or Pamelor)

40. (Protriptyline or Vivactil)

41. (Trimipramine or Surmontil)

42. (Amoxapine or Asendin)

43. (Loxapine or Loxpac or Loxitane)

44. (Maprotiline or Deprilept or Ludiomil or Psymion)

45. (Mazindol or Mazanor or Sanorex)

46. (Mianserin or Bolvidon or Norval or Tolvon)

47. (Mirtazapine or Remeron or Avanza or Zispin)

48. (Setiptiline or Tecipul)

49. MeSH descriptor: [Monoamine Oxidase Inhibitors] explode all trees

50. Monoamine oxidase Inhibitors

51. monoamine oxidase inhibitor*

52. (Isocarboxazid or Marplan or Moclobemide or Aurorix or Manerix or Phenelzine or Nardil or Selegiline or L‐Deprenyl or Eldepryl or Zelapar or Emsam or Trancypromine or Parnate or Moclobemide or Aurorix or Manerix)

53. MeSH descriptor: [Phenelzine] explode all trees

54. MeSH descriptor: [Moclobemide] explode all trees

55. MeSH descriptor: [Selegiline] explode all trees

56. MeSH descriptor: [Tranylcypromine] explode all trees

57. MeSH descriptor: [Isocarboxazid] explode all trees

58. serotonin‐norepinephrine reuptake inhibitor*

59. norepinephrine reuptake inhibitor*

60. serotonin‐noradrenaline reuptake inhibitor*

61. noradrenaline reuptake inhibitor*

62. (Duloxetine or Cymbalta or Milnacipram or Ixel or Savella)

63. (Ruboxetine or Edronax or Viloxazine or Vivalan)

64. MeSH descriptor: [Tryptophan] explode all trees

65. L‐tryptophan

66. #6 or #7 or #8 or #9 or #10 or #11 or #12 or #13 or #14 #15 or #16 or #17 or #18 or #19 or #20 or #21 or #22 or #23 or #24 or #25 or #26 or #27 or #28 or #29 or #30 or #31 or #32 or #33 or #34 or #35 or #36 or #37 or #38 or #39 or #40 or #41 or #42 or #43 or #44 or #45 or #46 or #47 or #48 or #49 or #50 or #51 or #52 or #53 or #54 or #55 or #56 or #57 or #58 or #59 or #60 or #61 or #62 or #63 or #64 or #65

67. #5 and #66

Appendix 3. Embase search strategy

1. exp osteoarthritis/

2. osteoarthr$.tw.

3. (degenerative adj2 arthritis).tw.

4. arthrosis.tw.

5. or/1‐4

6. exp Antidepressive Agents/

7. antidepressants.mp.

8. (anti‐depressant$ or antidepressant$).tw.

9. exp Serotonin Uptake Inhibitors/ or exp Citalopram/ or exp Serotonin Antagonists/ or exp Fluoxetine/ or SSRI$.tw.

10. selective serotonin reuptake inhibitors.mp.

11. (Citalopram or Celexa).tw.

12. (Fluoxetine or Prozac).tw.

13. (Paroxetine or Paxil or Seroxat).tw.

14. (Sertraline or Zoloft or Lustral).tw.

15. (Escitalopram or Lexapro or Cipralex).tw.

16. (Fluvoxamine or Luvox).tw.

17. (Desvenlafaxine or Pristiq).tw.

18. (Venlafaxine or Effexor).tw.

19. (Duloxetine or Cymbalta).tw.

20. (Milnacipran or Ixel or Savella).tw.

21. (Reboxetine or Edronax).tw.

22. (Viloxazine or Vivalan).tw.

23. exp Antidepressive Agents, Tricyclic/

24. tricyclic antidepressants.mp.

25. tricyclic antidepressant$.tw.

26. exp Amitriptyline/

27. (Amitriptyline or Elavil or Endep).tw.

28. (Clomipramine or Anafranil).tw.

29. (Desipramine or Norpramin or Pertofrane).tw.

30. (Dosulepin or Dothiepin or Prothiaden).tw.

31. (Doxepin or Adapin or Sinequan).tw.

32. (Imipramine or Tofranil).tw.

33. (Lofepramine or Feprapax or Gamanil or Lomont).tw.

34. (Nortriptyline or Pamelor).tw.

35. (Protriptyline or Vivactil).tw.

36. (Trimipramine or Surmontil).tw.

37. (Amoxapine or Asendin).tw.

38. (Loxapine or Loxpac or Loxitane).tw.

39. (Maprotiline or Deprilept or Ludiomil or Psymion).tw.

40. (Mazindol or Mazanor or Sanorex).tw.

41. (Mianserin or Bolvidon or Norval or Tolvon).tw.

42. (Mirtazapine or Remeron or Avanza or Zispin).tw.

43. (Setiptiline or Tecipul).tw.

44. exp Monoamine Oxidase Inhibitors/

45. monoamine oxidase inhibitors.mp.

46. Monoamine oxidase Inhibitor$.tw.

47. (Isocarboxazid or Marplan or Moclobemide or Aurorix or Manerix or Phenelzine or Nardil or Selegiline or L‐Deprenyl or Eldepryl or Zelapar or Emsam or Trancypromine or Parnate or Moclobemide or Aurorix or Manerix).tw.

48. serotonin‐norepinephrine reuptake inhibitor$.tw.

49. norepinephrine reuptake inhibitor$.tw.

50. serotonin‐noradrenaline reuptake inhibitor$.tw.

51. noradrenaline reuptake inhibitor$.tw.

52. (Duloxetine or Cymbalta or Milnacipram or Ixel or Savella).tw.

53. (Ruboxetine or Edronax or Viloxazine or Vivalan).tw.

54. exp Tryptophan/

55. L‐tryptophan.mp.

56. or/6‐55

57. random$.tw.

58. factorial$.tw.

59. crossover$.tw.

60. cross over.tw.

61. cross‐over.tw.

62. placebo$.tw.

63. (doubl$ adj blind$).tw.

64. (singl$ adj blind$).tw.

65. assign$.tw.

66. allocat$.tw.

67. volunteer$.tw.

68. crossover procedure/

69. double blind procedure/

70. randomized controlled trial/

71. single blind procedure/

72. or/57‐71

73. 5 and 56 and 72

Appendix 4. CINAHL search strategy

1.((MH "Clinical Trials+") OR "randomi?ed controlled trial*" OR (clinical W3 trial) OR double‐blind OR single‐blind OR triple‐blind OR (MH "Placebo Effect") OR (MH "Placebos") OR placebo* OR random* OR (MH "Random Sample") OR (MH "Study Design+") OR (latin square) OR (MH "Comparative Studies") OR (MH "Evaluation Research+") OR (MH "Prospective Studies+") OR (follow‐up stud*) OR (followup stud*) OR control* OR prospectiv* OR volunteer*) NOT (MH "Animals" NOT MH "Humans")

2. (MH "Osteoarthritis+") OR "osteoarthr*" OR "degenerative N2 arthritis" OR arthrosis* OR ("degenerative" N2 "arthritis")

3. (MH "Antidepressive Agents+") OR (MH "Antidepressive Agents, Tricyclic+") OR "tricyclic antidepressant*" OR (MH "Antidepressive Agents, Second Generation+") OR "antidepressant*" OR "anti‐depressant*" OR (MH "Serotonin Uptake Inhibitors+") OR (MH "Serotonin Antagonists+") OR (MH "Citalopram") OR (MH "Fluoxetine+") OR (MH "Amitriptyline") OR (MH "Monoamine Oxidase Inhibitors+") OR "monoamine oxidase inhibitor*" OR (MH "Selegiline") OR "Selegiline" OR "Tranylcypromine" OR "Isocarboxazid" OR "serotonin‐norepinephrine reuptake inhibitor" OR "norepinephrine reuptake inhibitor" OR "serotonin‐noradrenaline reuptake inhibitor" OR "noradrenaline reuptake inhibitor" OR "SSRI*" OR "Duloxetine" OR "Cymbalta" OR "Milnacipram" OR "Ixel" OR "Savella" OR "Ruboxetine" OR "Edronax" OR "Viloxazine" OR "Vivalan" OR "Moclobemide" OR "Phenelzine" OR (MH "Tryptophan") OR "Tryptophan" OR "Marplan" OR "Aurorix" OR "Manerix" OR "Nardil" OR "L‐Deprenyl" OR "Eldepry" OR "Zelapar" OR "Emsam" OR "Trancypromine" OR "Parnate" OR "Celexa" OR "Prozac" OR (MH "Paroxetine") OR "Paroxetine" OR "Paxil" OR "Seroxat" OR "Sertraline" OR (MH "Sertraline Hydrochloride") OR "Zoloft" OR "Lustral" OR "Escitalopram" OR "Lexapro" OR "Cipralex" OR (MH "Fluvoxamine Maleate") OR "Fluvoxamine" OR "Luvox" OR "Desvenlafaxine" OR "Pristiq" OR (MH "Desvenlafaxine Succinate") OR (MH "Venlafaxine+") OR "Venlafaxine" OR "Effexor" OR (MH "Duloxetine Hydrochloride") OR "Cymbalta" OR (MH "Milnacipran Hydrochloride") OR "Milnacipran" OR "Ixel" OR "Savella" OR "Reboxetine" OR "Edronax" OR "Viloxazine" OR "Vivalan" OR "Elavil" OR "Endep" OR (MH "Clomipramine") OR "Clomipramine" OR "Anafranil" OR (MH "Desipramine") OR "Desipramine" OR "Norpramin" OR "Pertofrane" OR "Dosulepin" OR "Dothiepin" OR "Prothiaden" OR "Doxepin" OR "Adapin" OR "Sinequan" OR (MH "Imipramine") OR "Imipramine" OR "Tofranil" OR "Lofepramine" OR "Feprapax" OR "Gamanil" OR "Lomont" OR (MH "Nortriptyline") OR "Nortriptyline" OR "Pamelor" OR "Protriptyline" OR "Vivactil" OR "Trimipramine" OR "Surmontil" OR (MH "Amoxapine") OR "Amoxapine" OR "Asendin" OR (MH "Loxapine") OR "Loxapine" OR "Loxpac" OR "Loxitane" OR (MH "Maprotiline") OR "Maprotiline" OR "Deprilept" OR "Ludiomil" OR "Psymion" OR "Mazindol" OR "Mazanor" OR "Sanorex" OR "Mianseri" OR "Bolvidon" OR "Norval" OR "Tolvon" OR (MH "Mirtazapine") OR "Mirtazapine" OR "Remeron" OR "Avanza" OR "Zispin" OR "Setiptiline" OR "Tecipul" OR "L‐tryptophan"

4. 1 AND 2 AND 3

Appendix 5. PsycINFO (Ovid) search strategy

1. Clinical Trials/ or Drug Therapy/ or Placebo/ or (randomi*ed controlled trial or controlled clinical trial).mp.

2. (random* or sham or placebo*).ti,ab,hw.

3. ((singl* or doubl*) adj (blind* or dumm* or mask*)).ti,ab,hw.

4. ((tripl* or trebl*) adj (blind* or dumm* or mask*)).ti,ab,hw.

5. (Randomi*ation or Random Allocation).ti,ab,hw.

6. 1 or 2 or 3 or 4 or 5

7. animals/ not (animals/ and human$.mp.)

8. 6 not 7

9. arthritis/ or (osteoarthr* or arthrosis).mp.

10. degenerative arthritis.mp.

11. 9 or 10

12. exp Antidepressant Drugs/ or antidepressant*.mp. or anti‐depressant*.mp.

13. exp Serotonin Antagonists/ or exp Serotonin Reuptake Inhibitors/ or exp Tricyclic Antidepressant Drugs/ or exp Monoamine Oxidase Inhibitors/ or serotonin noradrenalin reuptake inhibitor*.mp. or noradrenaline reuptake inhibitor*.mp. or ssri*.mp. or tricyclic antidepressant*.mp.

14. exp Fluoxetine/ or exp Paroxetine/ or exp Citalopram/ or exp Amitriptyline/ or exp Imipramine/ or exp Sertraline/ or exp Fluvoxamine/ or exp Venlafaxine/ or exp Amitriptyline/ or exp Imipramine/ or exp Chlorimipramine/ or exp Desipramine/ or exp Doxepin/ or exp Nortriptyline/ or exp Mianserin/ or exp Maprotiline/ or exp Loxapine/ or exp Isocarboxazid/ or exp Phenelzine/ or exp Tranylcypromine/ or exp Moclobemide/ or exp Tryptophan/

15. (fluoxetine or prozac or paroxetine or paxil or seroxat or sertraline or zoloft or lustral or escitalopram or lexapro or cipralex or fluvoxamine or luvox or desvenlafaxine or pristiq or venlafaxine or effex or duloxetine or cymbalta or milnacipran or ixel or savella or reboxetine or edronax or viloxazine or vivalan or amitriptyline or elavil or endep or clomipramine or anafranil or desipramine or norpramin or pertofrane or dosulepin or dothiepin or prothiaden or doxepin or adapin or sinequan or imipramine or tofranil or lofepramine or feprapax or gamanil or lomont or nortriptyline or pamelor or protriptyline or vivactil or trimipramine or surmontil or amoxapine or asendin or loxapine or loxpac or loxitane or maprotiline or deprilept or ludiomil or psymion or mazindol or mazanor or sanorex or mianserin or bolvidon or norval or tolvon or mirtazapine or remeron or avanza or zispin or setiptiline or tecipul or isocarboxazid or marplan or phenelzine or nardil or selegiline or eldepryl or zelapar or emsam or trancypromine or parnate or moclobemide or aurorix or manerix or cymbalta or milnacipram or ixel or savella or ruboxetine or edronax or viloxazine or vivalan or tryptophan or l deprenyl or l‐tryptophan).mp.

16. 12 or 13 or 14 or 15

17. 8 and 11 and 16

Appendix 6. WHO Clinical Trial Registry search strategy

1. Osteoarthrtis AND Antidepressants

2. Osteoarthrtis AND Serotonin uptake inhibitors

3. Osteoarthrtis AND Citalopram

4. Osteoarthrtis AND Fluoxetine

5. Osteoarthrtis AND Paroxetine

6. Osteoarthrtis AND Sertraline

7. Osteoarthrtis AND Escitalopram

8. Osteoarthrtis AND Fluvoxamine

9. Osteoarthrtis AND Desvenlafaxine

10. Osteoarthrtis AND Venlafaxine

11. Osteoarthrtis AND Duloxetine

12. Osteoarthrtis AND Milnacipran

13. Osteoarthrtis AND Reboxetine

14. Osteoarthrtis AND Viloxazine

15. Osteoarthrtis ANDTricyclic antidepressants

16. Osteoarthrtis AND Amitriptypline

17. Osteoarthrtis AND Clomipramine

18. Osteoarthrtis AND Desipramine

19. Osteoarthrtis AND Dosulepin

20. Osteoarthrtis AND Doxepin

21. Osteoarthrtis AND Imipramine

22. Osteoarthrtis AND Lofepramine

23. Osteoarthrtis AND Nortriptyline

24. Osteoarthrtis AND Protriptyline

25. Osteoarthrtis AND Trimipramine

26. Osteoarthrtis AND Amoxapine

27. Osteoarthrtis AND Loxapine

28. Osteoarthrtis AND Maprotiline

29. Osteoarthrtis AND Mazindol

30. Osteoarthrtis AND Mianserin

31. Osteoarthrtis AND Mirtazapine

32. Osteoarthrtis AND Setiptiline

33. Osteoarthrtis AND Monoamine Oxidase Inhibitor

34. Osteoarthrtis AND Isocarboxazid

35. Osteoarthrtis AND Moclobemide

36. Osteoarthrtis AND Phenelzine

37. Osteoarthrtis AND Selegiline

38. Osteoarthrtis AND Tryptophan

Appendix 7. ClinicalTrials.gov search strategy

1. Osteoarthrtis AND Antidepressants

2. Osteoarthrtis AND Serotonin uptake inhibitors

3. Osteoarthrtis AND Citalopram

4. Osteoarthrtis AND Fluoxetine

5. Osteoarthrtis AND Paroxetine

6. Osteoarthrtis AND Sertraline

7. Osteoarthrtis AND Escitalopram

8. Osteoarthrtis AND Fluvoxamine

9. Osteoarthrtis AND Desvenlafaxine

10. Osteoarthrtis AND Venlafaxine

11. Osteoarthrtis AND Duloxetine

12. Osteoarthrtis AND Milnacipran

13. Osteoarthrtis AND Reboxetine

14. Osteoarthrtis AND Viloxazine

15. Osteoarthrtis ANDTricyclic antidepressants

16. Osteoarthrtis AND Amitriptypline

17. Osteoarthrtis AND Clomipramine

18. Osteoarthrtis AND Desipramine

19. Osteoarthrtis AND Dosulepin

20. Osteoarthrtis AND Doxepin

21. Osteoarthrtis AND Imipramine

22. Osteoarthrtis AND Lofepramine

23. Osteoarthrtis AND Nortriptyline

24. Osteoarthrtis AND Protriptyline

25. Osteoarthrtis AND Trimipramine

26. Osteoarthrtis AND Amoxapine

27. Osteoarthrtis AND Loxapine

28. Osteoarthrtis AND Maprotiline

29. Osteoarthrtis AND Mazindol

30. Osteoarthrtis AND Mianserin

31. Osteoarthrtis AND Mirtazapine

32. Osteoarthrtis AND Setiptiline

33. Osteoarthrtis AND Monoamine Oxidase Inhibitor

34. Osteoarthrtis AND Isocarboxazid

35. Osteoarthrtis AND Moclobemide

36. Osteoarthrtis AND Phenelzine

37. Osteoarthrtis AND Selegiline

38. Osteoarthrtis AND Tryptophan

Data and analyses

Comparison 1. Antidepressants versus placebo in hip and knee osteoarthritis.

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1.1 Mean pain at 8–16 weeks	9	2038	Mean Difference (IV, Random, 95% CI)	‐0.59 [‐0.88, ‐0.31]
1.1.1 SNRI	7	1773	Mean Difference (IV, Random, 95% CI)	‐0.68 [‐0.98, ‐0.38]
1.1.2 SSRI	1	60	Mean Difference (IV, Random, 95% CI)	0.24 [‐0.47, 0.95]
1.1.3 TCA	1	205	Mean Difference (IV, Random, 95% CI)	‐0.56 [‐1.22, 0.10]
1.2 Responder rate: ≥ 50% reduction in pain	6	1904	Risk Ratio (M‐H, Random, 95% CI)	1.55 [1.32, 1.82]
1.2.1 SNRI	5	1703	Risk Ratio (M‐H, Random, 95% CI)	1.58 [1.31, 1.92]
1.2.2 TCA	1	201	Risk Ratio (M‐H, Random, 95% CI)	1.43 [1.07, 1.92]
1.3 Physical function (WOMAC)	6	1909	Mean Difference (IV, Random, 95% CI)	‐5.65 [‐7.08, ‐4.23]
1.3.1 SNRI	5	1704	Mean Difference (IV, Random, 95% CI)	‐5.68 [‐7.34, ‐4.02]
1.3.2 TCA	1	205	Mean Difference (IV, Random, 95% CI)	‐5.20 [‐11.32, 0.92]
1.4 Mean change in quality of life at 10–14 weeks	3	815	Mean Difference (IV, Random, 95% CI)	0.04 [0.01, 0.07]
1.4.1 SNRI	3	815	Mean Difference (IV, Random, 95% CI)	0.04 [0.01, 0.07]
1.5 Proportion of study withdrawals due to adverse effects	6	1977	Risk Ratio (IV, Random, 95% CI)	2.15 [1.56, 2.97]
1.5.1 SNRI	5	1772	Risk Ratio (IV, Random, 95% CI)	2.17 [1.57, 3.01]
1.5.2 TCA	1	205	Risk Ratio (IV, Random, 95% CI)	1.01 [0.06, 15.93]
1.6 Total adverse events	9	2101	Risk Ratio (M‐H, Random, 95% CI)	1.27 [1.15, 1.41]
1.6.1 SNRI	7	1840	Risk Ratio (M‐H, Random, 95% CI)	1.30 [1.19, 1.42]
1.6.2 SSRI	1	60	Risk Ratio (M‐H, Random, 95% CI)	1.33 [1.04, 1.72]
1.6.3 TCA	1	201	Risk Ratio (M‐H, Random, 95% CI)	1.14 [1.05, 1.23]
1.7 Serious adverse events	9	2101	Risk Ratio (M‐H, Random, 95% CI)	0.94 [0.46, 1.94]
1.7.1 SNRI	7	1840	Risk Ratio (M‐H, Random, 95% CI)	1.03 [0.42, 2.54]
1.7.2 SSRI	1	60	Risk Ratio (M‐H, Random, 95% CI)	0.25 [0.03, 2.11]
1.7.3 TCA	1	201	Risk Ratio (M‐H, Random, 95% CI)	1.37 [0.32, 5.98]
1.8 OARSI responders	5	1517	Risk Ratio (M‐H, Random, 95% CI)	1.72 [1.49, 1.98]
1.8.1 SNRI	4	1329	Risk Ratio (M‐H, Random, 95% CI)	1.74 [1.48, 2.04]
1.8.2 TCA	1	188	Risk Ratio (M‐H, Random, 95% CI)	1.60 [1.05, 2.42]
1.9 Radiographic joint structure changes	1		Risk Ratio (M‐H, Random, 95% CI)	Totals not selected
1.9.1 SNRI	1		Risk Ratio (M‐H, Random, 95% CI)	Totals not selected
1.10 Total treatment withdrawals	7	2023	Risk Ratio (M‐H, Random, 95% CI)	1.38 [1.09, 1.77]
1.10.1 SNRI	6	1818	Risk Ratio (M‐H, Random, 95% CI)	1.37 [1.06, 1.77]
1.10.2 TCA	1	205	Risk Ratio (M‐H, Random, 95% CI)	3.03 [0.32, 28.64]

1.9. Analysis.

Comparison 1: Antidepressants versus placebo in hip and knee osteoarthritis, Outcome 9: Radiographic joint structure changes

Characteristics of studies

Characteristics of included studies [ordered by study ID]

Chappell 2009b.

Study characteristics
Methods	Study design: multicentre, 3‐arm, double‐blind RCT Study grouping: parallel group Setting: community (USA) Time period: November 2006 to October 2007 Intervention: duloxetine vs placebo Sample size calculation: > 80% power to detect a treatment difference of 0.1 point in mean change of the primary variable and > 85% power to detect a treatment group difference of 25% in response rate based on data from duloxetine studies of diabetic peripheral neuropathic pain. Analysis: conducted on an ITT basis, assuming there was 1 measurement postbaseline. Missing data were imputed using LOCF.
Participants	Number of participants Number screened: 343 Number excluded: 87 ineligible, 25 declined to participate Number randomised: placebo: 120; duloxetine: 111 Number included in analysis Placebo: initial (week 0) 120, at re‐randomisation (week 7) 103, at conclusion (week 13) 96 Duloxetine 60 mg: initial (week 0) 111, at re‐randomisation (week 7) 46, at conclusion (week 13) 39 Duloxetine 60 mg uptitrated to 120 mg: initial(week 0) 0: at re‐randomisation 43, at conclusion 38 Baseline characteristics Antidepressant (duloxetine 60 mg and 120 mg) Age (years): 62.1 (SD 9.6) % female: 63.1% Analgesic use – % NSAID: 52.3% Disease duration (years) mean: 6.9 (SD 8.4) 24‐hour pain severity (mean, scale 0–10): 6.1 (SD 1.3) Function (WOMAC Total score, mean score 0–96): 53.73 (SD 16.51) Quality of life (EQ‐5D score mean): 0.68 (SD 1.18) Placebo Age (years): 62.5 (SD 9.3) % female: 67.5% Analgesic use – % NSAID: 49.2% Disease duration (years) mean: 7.1 (SD 7.2) 24‐hour pain severity (mean, scale 0–10): 6.2 (SD 1.3) Function (WOMAC Total score 0–96): 53.36 (SD 14.73) Quality of life (EQ‐5D score mean): 0.7 (SD 0.16) Inclusion criteria Male and female outpatients aged > 40 years Met ACR radiographic and clinical criteria for the diagnosis of knee OA Pain ≥ 14 days of each month for 3 months before study entry Mean score ≥ 4 on the 24‐hour mean pain score using the mean of daily ratings from visit 1 to visit 2 People with bilateral OA were required to identify an index knee on which to base ratings throughout study Agreed to maintain their usual activity level throughout course of study Exclusion criteria BMI > 40 kg/m² Confounding painful condition that would interfere with assessment of index joint Diagnosis of inflammatory arthritis or an autoimmune disorder Received invasive therapies to knee in past 3 months Knee arthroscopy of the index knee within past year Joint replacement of index knee at anytime Prior synovial fluid analysis indicative of a diagnosis other than OA Non‐ambulatory or required use of crutches or a walker People with psychiatric disorders including major depressive disorder (as identified using the Mini International Neuropsychiatric Interview) Previous exposure to duloxetine Pregnant or breastfeeding History of substance abuse or dependence Positive urine drug screen for any substance of abuse Existence of any serious medical or psychiatric condition that could compromise participation in study History of recurrent seizures Uncontrolled narrow‐angle glaucoma Acute liver injury or severe cirrhosis Known hypersensitivity to duloxetine or any of inactive ingredients Frequent or severe allergic reactions to multiple medications Baseline differences: mean height was greater in the duloxetine compared to placebo group.
Interventions	Intervention characteristics Antidepressant Type of antidepressant: duloxetine (SSRI) Route: oral Dose: 60–120 mg Frequency: daily Length of administration: 13 weeks of duloxetine 60 mg (started on 30 mg for 1 week). At week 7, all duloxetine participants were randomly re‐assigned in a 1:1 ratio to either duloxetine 60 mg or 120 mg, continued for the remainder of 13‐week treatment phase. 2‐week taper at conclusion of study. Placebo Route: oral Dose: placebo Frequency: daily Length of administration: 13 weeks
Outcomes	Outcomes measured at baseline, and 13 weeks with weekly measures Outcomes measured in trial Weekly change from baseline in the 24‐hour mean pain rating using 11‐point numerical Likert scale patient diary (0 no pain to 10 'worst pain imaginable') PGI‐I at 13‐week endpoint WOMAC change from baseline to 13‐week endpoint (outcomes measured weekly) Pain subscale Stiffness subscale Physical Function subscale Total score Change from baseline to 13‐week endpoint in weekly mean of the 24‐hour worst pain score Weekly change from baseline in the 24‐hour worst pain score Change from baseline to 13‐week endpoint in weekly mean of 24‐hour mean pain in the re‐randomised treatment phase Change from baseline to 13‐week endpoint in CGI‐S Change from baseline to 13‐week endpoint in BPI Worst pain score Least pain score Mean pain score Pain right now score General activity Mood Walking ability Normal work Relations with other people Sleep Enjoyment of life Mean interference Response to treatment Number of participants with ≥ 30% reduction of weekly mean in 24‐hour mean pain severity ratings Response to treatment – number of participants with ≥ 30% reduction of weekly mean in 24‐hour mean pain severity ratings in the re‐randomised treatment phase (weeks 7–13) Change from baseline to 13‐week endpoint in SF‐36 Mental Health Component Summary Physical Component Summary Change from baseline to 13‐week endpoint in EQ‐5D: US‐based index score, UK‐based index Change from baseline to 13‐week endpoint in BDI‐II – Total score Change from baseline to 13‐week endpoint in HADS‐A subscale Change from baseline to 13‐week endpoint in laboratory data – chemistry analytes Alkaline phosphatase Uric acid Change from baseline to 13‐week endpoint in vital signs Pulse rate Diastolic blood pressure Systolic blood pressure Weight Proportion of participants who discontinued study Proportion of participants who discontinued study due to adverse effects Adverse effects Outcomes included in review (selected to align with protocol) Major Pain – mean change from baseline overall pain (0–10 Likert 24‐hour mean pain scale) Responder, ≥ 50% reduction in mean pain Function – change from baseline WOMAC Total (0–96 scale) Quality of life: EQ‐5D (−0.11 to 1.0 scale, 1.0 indicated perfect health) Proportion of participants who discontinued study due to adverse effects Proportion of people who experienced any adverse effects Proportion of people who experienced serious adverse events Minor Proportion of participants who withdrew from study for any reason
Identification	Sponsorship source: Eli Lilly and Company Country: USA Trial registration: NCT00408421 Author's name: AS Chappell Institution: Lilly Research Laboratories, Indianapolis, IN, USA Email: aschappell@lilly.com
Notes	Adverse events Total adverse events: placebo: 55/111; duloxetine: 49/120 Number of serious events: placebo: 2/120 (1.7%); duloxetine: 1/111 (0.9%) Nature of events: dehydration, gouty arthritis, myocardial infarction, asthma, bronchiectasis, allergic rhinitis Withdrawals due to adverse effects: placebo: 7/120; duloxetine: 15/111 Analysis Normalised WOMAC Total score 0–96 to 0–100 using equation (score/96) × 100 Requests made for additional information to authors unsuccessful.
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Participants randomly assigned. Quote: "randomly assigned 1:1, as determined by a computer‐generated random sequence using an Interactive Voice Response System (IVRS)."
Allocation concealment (selection bias)	Low risk	Use of interactive voice response system.
Blinding of participants and personnel (performance bias) All outcomes	Low risk	Participants were blinded as to which treatment they received. Quote: "All patients, clinicians and raters were blinded to treatment."
Blinding of outcome assessment (detection bias) Self‐reported outcomes (pain, function, quality of life, adverse events)	Low risk	Participants were blinded to treatment they received and hence had a low risk of this influencing outcome assessment.
Blinding of outcome assessment (detection bias) Assessor‐reported outcomes (radiographic joint structure)	Low risk	No assessor‐reported outcomes included.
Incomplete outcome data (attrition bias) All outcomes	High risk	75% of participants completed study. Withdrawals from study: placebo: 24/120 (20%); duloxetine: 34/111 (30.63%) Greater loss to follow‐up in duloxetine group and greater discontinuation due to adverse event in duloxetine group.
Selective reporting (reporting bias)	Low risk	All outcomes reported as per initial trial protocol (even if no response).
Other bias	Low risk	No other sources of bias identified.

Chappell 2011.

Study characteristics
Methods	Study design: RCT Study grouping: parallel group Setting: 21 clinical sites in USA, Canada, Greece, Russia and Sweden by GP and rheumatologists Time period: February 2007 to May 2008 Intervention: duloxetine vs placebo Sample size calculation: not reported Analysis: LOCF or baseline observation carried forward
Participants	Number of participants Number screened: not reported Number excluded: not reported Number randomised: placebo: 128; duloxetine: 128 Number included in analysis: placebo: 128 at week 0, 111 at week 13; duloxetine: 128 at week 0, 93 at week 13 Baseline characteristics Antidepressant (duloxetine 60 mg and 120 mg) Age, mean (years): 63.2 (SD 8.8) % female: 69.5% Analgesic use – % NSAID: 36.7% Disease duration (years): 6.2 (SD 5.9) BPI mean pain score, mean: 6.1 (SD 1.4) Function: no baseline data reported Quality of life (mean) EQ‐5D, −0.11 to 1.0 scale: 0.68 (SD 0.16) SF‐36 Mental Component, scored 0–100: 55.86 (SD 10.00) Placebo Age, mean (years): 61.9 (SD 9.2) % female: 83.6% Analgesic use – % NSAID: 41.4% Disease duration (years): 5.6 (SD 6.2) BPI mean pain score, mean: 6.1 (SD 1.3) Function: no baseline data available Quality of life, mean EQ‐5D, −0.11 to 1.0 scale: 0.66 (SD 0.16) SF‐36 Mental Component, scored 0–100: 54.47 (SD 10.16) Inclusion criteria Male and female outpatients age > 40 years ACR clinical and radiographic criteria for diagnosis of knee OA Pain > 14 days per month during 3 consecutive months preceding study entry Required to have pain severity ≥ 4 on the 24‐hour mean pain severity scale, using the mean of daily ratings from the week preceding randomisation Had to consent to maintain their usual activity level throughout course of study Exclusion criteria BMI > 40 kg/m² Diagnosis of inflammatory arthritis or an autoimmune disorder History of invasive therapies to the index knee during past 3 months or joint replacement of index knee at anytime Prior synovial fluid analysis indicative of a diagnosis other than OA Non‐ambulatory or crutch or walker dependent Presence of psychiatric disorders, including major depressive disorder (identified using the Mini International Neuropsychiatric Interview) Previous exposure to duloxetine Pregnant or breastfeeding History of substance abuse or dependence Presence of serious medical condition History of recurrent seizures Uncontrolled narrow‐angle glaucoma Acute liver injury or severe cirrhosis Known hypersensitivity to duloxetine or any of the inactive ingredients Frequent or severe allergic reactions to multiple medications Baseline differences: significantly fewer women in duloxetine group compared to placebo (P = 0.012). Fewer participants in duloxetine group used NSAIDs. Other differences: more participants in duloxetine group discontinued from study due to adverse events (P = 0.002).
Interventions	Intervention characteristics Antidepressant Dose: 60 mg or 120 mg Frequency: daily Type of antidepressant: duloxetine (SSRI) Route of administration: oral Length of administration: 13 weeks of 60 mg (started on 30 mg for 1 week). At week 7, duloxetine dose increased to 120 mg daily in participants reporting < 30% pain reduction from baseline using BPI 24‐hour mean pain rating and no tolerability concerns. Placebo Dose: placebo Frequency: daily Route of administration: oral Length of administration: 13 weeks
Outcomes	Outcomes measured in trial Primary outcome Change in BPI 24‐hour mean rating Secondary outcomes Mean values at 13‐week endpoint in PGI‐I Change from baseline to 13‐week endpoint in WOMAC Physical Function subscale Pain subscale Stiffness subscale Change from baseline to 13‐week endpoint in weekly mean of the 24‐hour mean pain and worst pain scores Change from baseline to 13‐week endpoint in CGI‐S Number of participants who responded to treatment at 13‐week endpoint, ≥ 30% and ≥ 50% Mean change from baseline to 13‐week endpoint in SF‐36 Mental Health Component Summary Physical Component Summary Domain scores Change from baseline to 13‐week endpoint in EQ‐5D Change from baseline to 13‐week endpoint in BDI‐II Change from baseline to 13‐week endpoint in HADS‐A Change from baseline to 13‐week endpoint in BPI‐S: Worst pain score Least pain score Mean pain score Pain right now score General activity Mood Outcomes included in review (selected to align with protocol) Major Pain – mean change from baseline overall pain (0–10 Likert 24‐hour mean pain scale) Responder, ≥ 50% reduction in mean pain Function – change from baseline WOMAC Total score (0–96 scale) Quality of life: EQ‐5D (−0.11 to 1.0 scale, 1.0 indicated perfect health) Proportion of participants who discontinued study due to adverse effects Proportion of people who experienced any adverse effects Proportion of people who experienced serious adverse events Minor Proportion of participants who withdrew from study for any reason
Identification	Sponsorship source: Eli Lilly and Company. Mrs Chappell, Skljarevski, Desaiah, Liu‐Seifert and Ms Zhang are employees and stockholders of Eli Lilly and Company. Drs Belenkov and Brown were participating investigators in the conduct of the study and received funding from Eli Lilly and Company. Country: US, Russia and Canada Trial registration: NCT00433290 Author's name: Amy Chappell Institution: Eli Lilly and Company, Lilly Corporate Centre, Indianapolis, IN, US Email: desaiahdu@lilly.com
Notes	Adverse effects Total adverse effects: placebo: 41/128; duloxetine: 64/128 Serious adverse effects: placebo: 2/128 (1.56%); duloxetine: 3/128 (2.34%) Nature of events: atrial fibrillation, acute pyelonephritis, drug intolerance, memory impairment, supraventricular tachycardia Withdrawals due to adverse effects: placebo: 7/128; duloxetine: 24/128 Analysis Normalised WOMAC Total score from 0–96 to 0–100 using equation (score/96) × 100. Note: least squares mean used as only data available Requests made for additional information to authors unsuccessful.
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Computer‐generated random sequence using an interactive voice response system to assure random blinding.
Allocation concealment (selection bias)	Low risk	Use of interactive voice response system.
Blinding of participants and personnel (performance bias) All outcomes	Low risk	Quote: "Double blind."
Blinding of outcome assessment (detection bias) Self‐reported outcomes (pain, function, quality of life, adverse events)	Low risk	Study participants were blinded to their group allocation (not reported).
Blinding of outcome assessment (detection bias) Assessor‐reported outcomes (radiographic joint structure)	Low risk	No assessor‐reported outcomes included in study.
Incomplete outcome data (attrition bias) All outcomes	Low risk	80% of participants completed study. Withdrawal scores: placebo: 17/128 (13.3%); duloxetine: 35/128 (27.3%). Increased withdrawals in duloxetine group related to adverse effects.
Selective reporting (reporting bias)	Low risk	Compared to study protocol (available from ClinicalTrials.gov), primary outcome (BPI 24‐hour mean) and all secondary outcomes reported.
Other bias	Low risk	Judgement comment: disclosure regarding sponsorship by Eli Lilly and Company provided.

Frakes 2011.

Study characteristics
Methods	Study design: RCT Study grouping: parallel group Setting: 42 study sites in US and Puerto Rico between November 2009 and April 2011. Care provided by clinicians from a variety of specialities. Time period: 10 weeks Intervention: duloxetine vs placebo Sample size calculation: 253 participants per group to provide 80% power to detect a treatment effect of 0.25 based on 0.05 level of significance for change from baseline in the weekly mean of the 24‐hour mean pain score at 8 weeks Analysis: modified ITT basis, with all participants remaining in the group to which they were randomised but only participants with baseline and ≥ 1 postbaseline observation were included in the efficacy‐based results.
Participants	Number of participants Number screened: 1128 Number excluded: 604 Number randomised total: placebo: 260; duloxetine: 264 Number included in analysis: placebo: 260 at week 0, 199 at week 10; duloxetine: 164 at week 0, 189 at week 10 Baseline characteristics Antidepressant Age, mean (years): 61.6 (SD 9.2) % female: 57.6% Analgesic use–: 100% Disease duration (years): 9.8 (SD 8.9) Baseline pain, mean: weekly mean pain severity scale 6.27 (SD 1.41) Function: no baseline data available for WOMAC Total score. WOMAC Physical Function at baseline: 37.4 (SD 10.1) Quality of life: no baseline data available, b‐POMS (score 0–20) Placebo Age, mean (years): 60.3 (SD 9.2) % female: 56.5% Analgesic use – % NSAID: 100% Disease duration (years): 9.2 (8.9) Baseline pain, mean: weekly mean pain severity scale 6.36 (SD 1.41) Function: no baseline data available for WOMAC Total score. WOMAC Physical Function at baseline: 37.5 (SD 9.4) Inclusion criteria Men and women aged > 40 years ACR clinical and radiographic criteria for OA of knee Knee pain for > 14 days/month in 3 months preceding study entry and reported use of oral NSAIDs for treatment of knee pain on most days during that time period Exclusion criteria Gout, pseudogout, inflammatory arthritis, end‐stage bone‐on‐bone OA Knee surgery or intra‐articular injection in last 6 months Another chronic painful condition such as fibromyalgia that could interfere with assessment of knee BMI > 40 kg/m² Non‐ambulatory or required equipment other than 1 walking stick Using opioid analgesics > 3 days/week or unwilling to taper and discontinue their use prior to randomisation Other serious or unstable medical conditions Taking warfarin Anaemia at baseline (Hb < 110 for men or < 100 for women) History of peptic ulcer disease History of bleeding disorder or other risk factor for bleeding Baseline differences: no significant demographic differences. Tendency for participants in placebo group to be younger (60.3, SD 9.2 years) than those receiving duloxetine (61.6, SD 9.1 years; P = 0.08)
Interventions	Intervention characteristics Antidepressant Frequency: daily Dose: 30 mg/day for 1 week followed by 60 mg/day for 2 weeks. At week 3 of active treatment, participants who had a mean pain severity rating of ≥ 4 during previous week had a blinded dose escalation to 120 mg/day. Route of administration: oral Type of antidepressant: duloxetine (SSRI) Length of treatment: 8 weeks primary outcome. Sham endpoint at 10 weeks, with dose reduction between weeks 10 and 12. Placebo Frequency: daily Dose: placebo Route of administration: oral Length of treatment: 8 weeks primary outcome. Sham endpoint at 10 weeks, with dose reduction between weeks 10 and 12.
Outcomes	Outcomes measured in trial Primary outcome Change from baseline in weekly mean of the 24‐hour mean pain score This was altered to: Change from baseline in the weekly mean of the 24‐hour mean pain score at 8 weeks Secondary outcomes PGI‐I Change from baseline in weekly mean of 24‐hour night pain and worst pain scores Change from baseline in the WOMAC Pain, Stiffness and Physical Function subscale scores Change from baseline in the BPI‐Severity and Interference scores Change from baseline in the CGI‐S Change from baseline in the PGAI Change from baseline in the b‐POMS Total and subscale scores Weekly use of paracetamol Response status using OARSI Response Criteria Response status based on a ≥ 30% and ≥ 50% reduction in the weekly mean of the 24‐hour mean pain score Response status based on a ≥ 30% and ≥ 50% reduction in the BPI mean pain score Abnormal laboratory values statistically different between treatment groups Vital signs statistically different between treatment groups Discontinuation due to adverse events Outcomes included in review (selected to align with protocol) Major Pain – mean change from baseline, overall pain (0–10 Likert 24‐hour mean pain scale) Responder (≥ 50% reduction in mean pain) Function – change from baseline WOMAC Total (0–96 scale) Proportion of participants who discontinued study due to adverse effects Proportion of people who experienced any adverse effects Proportion of people who experienced serious adverse events Minor Responder (OARSI Response Criteria) Proportion of participants who withdrew from study for any reason
Identification	Sponsorship source: financial support was provided by Eli Lilly and Company, Indianapolis, IN, USA. Employees of Lilly were involved in study design, analysis of data, critical revision of manuscript and decision to submit manuscript for publication. Country: USA Trial registration: NCT01018680 Author's name: Elijah Frakes Institution: Eli Lilly and Company, Indianapolis, IN, US Email: efrakes@lilly.com
Notes	Adverse effects Total adverse effects: placebo: 130/260; duloxetine: 167/264 Serious adverse effects: placebo: 3/260; duloxetine: 5/264 Nature of events: coronary artery disease, fall, metastasis to central nervous system, syncope, urosepsis, breast cancer, cerebrovascular accident, cholecystitis, pancreatitis Withdrawals due to adverse events: placebo: 23/260; duloxetine: 40/264 There have been alterations to outcomes in ClinicalTrials.gov, submitted 2012 (after publishing) Requests made for additional information to authors unsuccessful
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	No details on method of randomisation available. Allocation was "randomised."
Allocation concealment (selection bias)	Unclear risk	No details provided as per allocation concealment, but (quote): "patient and clinicians were blind to the criteria for advancement into the treatment phase" (p 2362).
Blinding of participants and personnel (performance bias) All outcomes	Low risk	Participants blinded to group allocation.
Blinding of outcome assessment (detection bias) Self‐reported outcomes (pain, function, quality of life, adverse events)	Low risk	Participants blinded to their treatment groups.
Blinding of outcome assessment (detection bias) Assessor‐reported outcomes (radiographic joint structure)	Low risk	No assessor‐reported outcomes included in study.
Incomplete outcome data (attrition bias) All outcomes	Low risk	74% of participants completed study. Withdrawals: placebo: 61/260 (23.46%); duloxetine: 75/264 (28.41%).
Selective reporting (reporting bias)	Low risk	All outcomes in protocol reported on online if not in published paper (can be found on ClinicalTrials.gov).
Other bias	Low risk	Declaration of funding provided. No other sources of bias detected.

Hudson 2021.

Study characteristics
Methods	Study design: RCT Study grouping: parallel group Time period: 2012–2018 Intervention: nortriptyline vs placebo Sample size calculation: sample size calculated to give ≥ 90% power at a 2‐sided significance level of 0.05 to detect a difference in treatment effect of 5 points between the nortriptyline and placebo groups. The sample size was calculated using a pooled SD 10 points estimated from previous studies  and conservatively assuming no correlation between baseline and follow‐up scores. Analysis: ITT and per‐protocol analysis
Participants	Number of participants Number screened: 631 Number excluded: 426 Number randomised: total 205; placebo: 103; nortriptyline: 102 Number included in analysis: placebo: 103; nortriptyline: 102 Baseline characteristics Antidepressant Age, mean (years): 64.4 (SD 7.9) % female: 57% Analgesic use – % NSAID: not reported Disease duration (years): 8.5 (SD 7.9) Baseline pain, mean: 60.2 (SD 13.5) WOMAC Pain standardised on a 0–100 scale where 0 is the best possible outcome Function, mean: 62.8 (SD 15) WOMAC Function standardised on a 0–100 scale where 0 is the best possible outcome Quality of life: no overall score available Placebo Age, mean (years): 64.6 (SD 10.3) % female: 58% Analgesic use – % NSAID: not reported Disease duration (years): 6.6 (SD 7.1) Baseline pain, mean: 61.2 (SD 12.5) WOMAC Pain standardised on a 0–100 scale where 0 is the best possible outcome Function, mean: 59.9 (SD 14.8) WOMAC Function standardised on a 0–100 scale where 0 is the best possible outcome Quality of life: no overall score available Inclusion criteria Primary knee OA defined according to ACR classification criteria (knee pain plus 3 of: aged > 50 years, morning stiffness < 30 minutes, crepitus, bony tenderness, no palpable warmth) Pain severity ≥ 20 points on WOMAC Numerical Rating Scale (range 0–50 points) at study knee Stable analgesic regimen for 2 months before entering study Exclusion criteria Prior joint replacement surgery on study knee Intra‐articular steroid injection within previous 3 months Secondary OA (OA due to inflammatory arthritis (e.g. gout, rheumatoid arthritis, juvenile arthritis), septic arthritis or trauma (articular fracture)) Known hypersensitivity to nortriptyline or history of adverse reaction to any TCA Current use of nortriptyline or other antidepressants, amiodarone or domperidone Myocardial infarction within 6 months before study entry Heart block Postural hypotension Pregnancy Hyperthyroidism or phaeochromocytoma under current investigation or treatment History of epilepsy or other seizure History of bipolar disorder or manic episode History of increased intra‐ocular pressure or history of angle‐closure glaucoma Chronic constipation Urinary retention Baseline differences: greater proportion of overweight participants in placebo and greater proportion of obese participants in antidepressant group
Interventions	Intervention characteristics Antidepressant Type of antidepressant: nortriptyline Frequency: daily Dose: 25–100 mg Route of administration: oral Length of treatment: week 0–8 dose adjustment period where participants receive nortriptyline 25–100 mg as titrated by study nurse. Week 8–14 steady dose treatment period Placebo Frequency: daily Dose: placebo Route of administration: oral Length of treatment: 14 weeks
Outcomes	Outcomes were measured at baseline, and 14 weeks with fortnightly measures Outcomes reported in study Primary outcomes Change from baseline to 14‐week endpoint in WOMAC Pain subscale 0–100 standardised scale where 0 is best outcome Secondary outcomes Physical function using the WOMAC Function subscale 0–100 standardised scale where 0 is best outcome Participant‐rated global assessment using a VAS 0–100 standardised scale where 0 is best outcome Difference in proportion of participants reporting a treatment effect, defined according to OARSI set of responder criteria Quality of life using SF‐36 Participant‐recorded NSAID and other analgesic use in final 2 weeks of study Adverse events Serious adverse effects Study withdrawals Outcomes included in review (selected to align with protocol) Major Change from baseline to 14‐week endpoint in WOMAC Pain subscale 0–100 standardised scale where 0 is best outcome Physical function using the WOMAC Function subscale 0–100 standardised scale where 0 is best outcome Adverse events Serious adverse effects Study withdrawals Minor Difference in the proportion of participants reporting a treatment effect, defined according to the OARSI set of responder criteria
Identification	Source: Health Research Council of New Zealand Country: New Zealand Setting: community Trial registration: ACTRN12614000683639 Author's name: Dr Ben Hudson Institution: University of Otago, Christchurch 8140, New Zealand Email: ben.hudson@otago.ac.nz
Notes	Adverse effects Total adverse effects: placebo: 88/102 (87%); nortriptyline: 97/99 (99%) Serious adverse effects: placebo: 3/102 (3%); nortriptyline: 4/99(4%) Nature of events: lower back pain, atrial fibrillation, myocardial infarction, epistaxis, renal calculi, lung infection, hyperglycaemia Withdrawals from trial due to adverse effects: not reported Analysis 0–100 scale for pain altered to 0–10 scale by calculation; result/10
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Computer‐generated randomisation list with blocks of varying size (1–4) prepared by study statistician.
Allocation concealment (selection bias)	Unclear risk	Blinded; however, no details provided.
Blinding of participants and personnel (performance bias) All outcomes	Unclear risk	Blinded; however, no details provided.
Blinding of outcome assessment (detection bias) Self‐reported outcomes (pain, function, quality of life, adverse events)	Low risk	To prevent accidental unblinding there was no contact between study team and participant until after data cleaning, database lockdown and analysis were complete.
Blinding of outcome assessment (detection bias) Assessor‐reported outcomes (radiographic joint structure)	Low risk	Not applicable to this study.
Incomplete outcome data (attrition bias) All outcomes	Low risk	98% of participants completed study.
Selective reporting (reporting bias)	Low risk	OARSI reported as non‐significant. Further results provided upon request.
Other bias	Low risk	No significant baseline imbalance, co‐interventions used across both groups and no unplanned interim analysis performed.

Mathur 2015.

Study characteristics
Methods	Study design: RCT Study grouping: parallel group Setting: not reported Time period: 11 weeks Intervention: milnacipran vs placebo Sample size calculation: not performed. Pilot study Analysis: not reported
Participants	Number of participants Number screened: not reported Number excluded: not reported Number randomised: total: 46: placebo: 17; milnacipran: 29 Number included in analysis: placebo: 12; milnacipran: 26 Baseline characteristics Antidepressant Age, mean (years): 57 (SD 9) % female: 65.4% Analgesic use – % NSAID: not reported Disease duration (years): not reported Baseline pain, mean: 44 (SD 24) VAS where 0 is no pain and 100 is the most pain Function: not reported Quality of life: not reported Placebo Age (years, mean): 53 (7) % female: 50% Analgesic use – % NSAID: not reported Disease duration (years): not reported Baseline pain, mean: 55 (SD 28) VAS where 0 is no pain and 100 is the most pain Function: not reported Quality of life: not reported Inclusion criteria Knee pain and osteophytes on x‐rays Aged ≥ 40 years, morning stiffness lasting ≤ 30 minutes, and crepitus (cracking sound) during motion Experiences chronic pain for ≥ 6 months Mean pain rating of worse knee ≥ 4 on a 0–10 scale Not pregnant or breastfeeding, and not currently attempting to conceive; if of childbearing potential, using highly effective method of contraception (as determined by principal investigator) Able to walk ≥ 0.5 of a city block a day, and agreed to try and slowly increase that over course of study Able to read and speak English and provide informed consent Able to understand and comply with all data collection methodology including electronic diary Agrees to continue their stable drug regimen with no changes during course of study; only use paracetamol 325 mg tablets (maximum 8 tablets a day) for breakthrough pain; not use paracetamol or any other pain medicines 12 hours before testing. Exclusion criteria Already receiving SNRIs (milnacipran, duloxetine, venlafaxine) or other neuroamine reuptake blockers for mood disorders Allergic to SSRIs, SNRIs or milnacipran Severe or untreated psychiatric disturbance (e.g. mania, depression, anxiety, substance dependence) Medicated with triptans, MAOIs, SSRIs, other SNRIs, tricyclic or heterocyclic antidepressants, lithium, adrenaline, noradrenaline, clonidine or digoxin during trial Clinical diagnosis of fibromyalgia Severe ongoing or unaddressed medical conditions (e.g. renal or hepatic disease (creatinine > 1.5 mL/dL; AST or ALT > 3 × normal limit), uncontrolled hypertension, severe cardiac rate or rhythm disorders, rheumatological disease (e.g. polymyalgia rheumatica), narrow‐angle glaucoma, hyponatraemia, clotting disorders, uncontrolled seizure disorder or urinary retention) Cardiac implants Knee replacement Plans to start new pain treatments or therapies during study (e.g. new pain medication, injections, physiotherapy, surgery) Baseline differences: more women in the milnacipran group. Baseline pain scores were higher in the placebo group compared to milnacipran group. Unknown if these differences were significant.
Interventions	Intervention characteristics Antidepressant Frequency: twice daily Dose: target dose 200 mg/day 25 mg daily for 2 days 50 mg daily for 2 days 100 mg daily for 2 days 150 mg daily for 4 days 200 mg daily ongoing If intolerable adverse effects, dose reduced to 100 mg/day. If unable to tolerate then participant would discontinue study Route of administration: oral Type of antidepressant: milnacipran (SNRI) Length of treatment: 11 weeks Placebo Frequency: twice daily Dose: placebo Route of administration: oral Length of treatment: 11 weeks
Outcomes	Outcomes measured in trial Primary outcomes Efficacy of milnacipran, measured by the change of knee pain, in treating chronic knee OA pain – no details provided regarding scales used Secondary outcomes Change of physical functional ability due to change in knee pain – no details provided regarding scales used Outcomes reported in study (altered 2015) Primary outcomes McGill Pain Questionnaire – Short Form baseline and endpoint scores (0‐ to 45‐point scale where 0 is no pain and 45 is worst pain) PamSys Actigraph Data baseline and endpoint scores PASS baseline and endpoint scores PDI baseline and endpoint scores CES‐D‐10 baseline and endpoint scores Pain VAS baseline and endpoint scores Secondary outcomes Daily diary entries with pain, fatigue and functioning scores 3 times a day Outcomes included in review (selected to align with protocol) Major Pain Proportion of people who experienced any adverse effects Proportion of people who experienced serious adverse events Minor None
Identification	Sponsorship source: Forrest Research Institute Country: USA Trial registration: NCT01510457 Author's name: interim analysis by S Mathur. Dr R Norman Harden is the point of contact. Institution: Shirley Ryan Ability Lab Pain Management Center, Chicago, IL, US Email: centerforpainstudies@ric.org
Notes	Adverse effects Total adverse effects: placebo: 0/12 (0%); milnacipran: 9/26 (34.6%) Serious adverse effects: placebo: 0/12 (0%); milnacipran: 0/26 (0%) Nature of events: not applicable Withdrawals from trial due to adverse effects: not reported Additional notes An interim analysis of this trial was published. Final results were available on ClinicalTrials.gov but no formal publication of final results has occurred. There have been alterations to outcomes of study in ClinicalTrials.gov, submitted 2015 Analysis Pain scores normalisation of 0–45 scale to 0–10 scale using (result/45) × 10
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	2:1 random allocation, no further details provided.
Allocation concealment (selection bias)	Unclear risk	Blinded; however no details provided.
Blinding of participants and personnel (performance bias) All outcomes	Unclear risk	Blinded; however no details provided.
Blinding of outcome assessment (detection bias) Self‐reported outcomes (pain, function, quality of life, adverse events)	Unclear risk	Blinded; however no details provided.
Blinding of outcome assessment (detection bias) Assessor‐reported outcomes (radiographic joint structure)	Low risk	Not applicable for this study.
Incomplete outcome data (attrition bias) All outcomes	High risk	82.7% of participants completed the study. Withdrawals: placebo: 5/17 (29.4%); milnacipran 3/29 (10.3%) No details provided regarding reasons for withdrawal. It was unclear whether ITT analysis performed.
Selective reporting (reporting bias)	High risk	Final results of study have never been formally published. They are available on ClinicalTrials.gov.
Other bias	High risk	Baseline imbalance in pain scores between placebo and antidepressants groups present. Interim analysis reported in 2013, with changes to outcome criteria post this.

Riesner 2008.

Study characteristics
Methods	Study design: RCT Study grouping: parallel group Time period: 8 weeks Intervention: fluvoxamine vs placebo Sample size calculation: not described Analysis: ITT
Participants	Number of participants Number screened: not reported Number excluded: not reported Number randomised: total 60; placebo: 30; fluvoxamine: 30 Number included in analysis: placebo: 30; fluvoxamine: 30 Baseline characteristics Antidepressant Age, mean (years): not reported % female: not reported Analgesic use – % NSAID: not reported Disease duration (years): not reported Baseline pain, mean (SD): not reported Function: not reported Quality of life: not reported Placebo Age, mean (years): not reported % female: not reported Analgesic use – % NSAID: not reported Disease duration (years): not reported Baseline pain, mean (SD): not reported Function: not reported Quality of life: not reported Inclusion criteria Advanced gonarthrosis or coxarthrosis (OA) of the hip or knee Chronic pain caused by OA with therapy indication Initial pain on VAS > 50 mm at first examination Aged 30–80 years No MAOI use for ≥ 2 weeks before start of study Taking only prescribed NSAIDs (diclofenac, piroxicam, ibuprofen) Regular weekly telephone calls for evaluation (possible) Exclusion criteria Multimorbidity (cardiovascular, liver, kidney diseases, reduced general condition, endogenous depression) Organic brain psychiatric syndromes and dependencies Pregnant and breastfeeding, or people in custody Speech and communication problems Baseline differences: no baseline details provided on study cohorts to determine any pretreatment differences that may affect the results.
Interventions	Intervention characteristics Antidepressant Frequency: daily Dose: up to 150 mg 50 mg daily for weeks 1–2 50 mg or 100 mg daily for week 3 50 mg, 100 mg or 150 mg daily weeks 4–8 Route of administration: oral Type of antidepressant: fluvoxamine Length of treatment: 8 weeks Placebo Frequency: daily Dose: placebo Route of administration: oral Length of treatment: 8 weeks
Outcomes	Outcomes measured in trial Primary outcome Pain as assessed with VAS Secondary outcomes WOMAC scale Zerssen Mental State scale CGI Griss scale Outcomes included in review (selected to align with protocol) Major Pain Minor Proportion of people who experienced any adverse effects Proportion of people who experienced serious adverse events
Identification	Sponsorship source: Leipzig University, Hessing Foundation Country: Germany Setting: outpatient and telephone Trial registration: unregistered Author's name: HJ Riesner Institution: Leipzig University, Leipzig, Germany Email: hajo.riesner@gmx.de
Notes	Adverse effects Total adverse effects Placebo cohort: 21 Fluvoxamine cohort: 28 Serious adverse effects Placebo cohort: 4 Fluvoxamine cohort: 1 Nature of events: not specified Withdrawals from trial due to adverse effects Placebo cohort: incomplete data Fluvoxamine cohort: incomplete data Additional notes Study required translation from German. Requests made for additional information to authors unsuccessful. SDs for change in pain not provided. SDs taken from baseline pain scores.  Scale changed to a 0–10 scale (result/100)
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Computer‐generated, random allocation.
Allocation concealment (selection bias)	Unclear risk	Blinded; however, no details provided.
Blinding of participants and personnel (performance bias) All outcomes	Unclear risk	Blinded; however, no details provided.
Blinding of outcome assessment (detection bias) Self‐reported outcomes (pain, function, quality of life, adverse events)	Unclear risk	Blinded; however, no details provided.
Blinding of outcome assessment (detection bias) Assessor‐reported outcomes (radiographic joint structure)	Low risk	Not applicable.
Incomplete outcome data (attrition bias) All outcomes	High risk	No details provided regarding numbers of individuals who discontinued study or reasons why.
Selective reporting (reporting bias)	Low risk	Outcomes described reported.
Other bias	Unclear risk	Given lack of baseline data available, unable to commence on baseline differences having impact.

Tetreault 2016.

Study characteristics
Methods	Study design: RCT (study phase II from paper) Study grouping: parallel group Setting: US Time period: 16 weeks Intervention: duloxetine vs placebo Sample size calculation: exploratory study. Size based on other studies regarding grey mater restoration (references not given) Analysis: no ITT. If participants failed to complete study then further participants were recruited
Participants	Number of participants: 39 Number screened: 70 Number excluded: 31 Number randomised total: placebo: 20; duloxetine: 19 Number included in analysis: placebo: 20; duloxetine: 19 Baseline characteristics Antidepressant Age, mean (years): 58.6 (SD 4.2) % female: 52.6% Analgesic use – % NSAID: not reported Disease duration (years): not reported Baseline pain, mean (SD): not reported Function: not reported Quality of life: not reported Placebo Age, mean (years): 58.9 (SD 8.6) % female: 47.6% Analgesic use – % NSAID: not reported Disease duration (years): not reported Baseline pain, mean (SD): not reported Function: not reported Quality of life: not reported Inclusion criteria Aged 45–80 years ACR criteria for OA including Kellgren‐Lawrence radiographic OA grades II–IV VAS pain score > 5/10 within 48 hours of telephone screen and visit 1 (screening) Knee OA for a minimum of 12 months Need for daily pain medication to manage symptoms of OA Exclusion criteria Currently receiving MAOIs or any centrally acting drug for analgesia, depression Narrow‐angle glaucoma Uncontrolled hypertension Co‐existing inflammatory arthritis, fibromyalgia or other chronic pain state Pregnant, trying to become pregnant or breastfeeding Major depressive disorder Substantial alcohol use or history of significant liver disease Use of MAOIs, triptans, serotonin precursors (tryptophan) Use of potent CYP1A2 inhibitors, thioridazine, and antidepressants Diabetes, type 1 or type 2 Condition which Investigator believed would interfere with person's ability to comply with study instructions, or might confound the interpretation of study results or put the person at undue risk MRI safety necessitates the exclusion of people having ≥ 1 of: Metal fragments in the eye or face, or having worked previously in the metal industry Implantation of any electronic devices (e.g. cardiac pacemakers, cardiac defibrillators and cochlear implants or nerve stimulatory) Surgery on blood vessels of brain Claustrophobia Piercings or tattoos > 113 kg in weight Obvious brain abnormalities Baseline differences: higher proportion of participants who received antidepressants were female. Significance of this not reported.
Interventions	Intervention characteristics Antidepressant Frequency: daily Dose: 30 mg for 1 week followed by 60 mg for 14 weeks then 30 mg for 1 week Route of administration: oral Type of antidepressant: duloxetine Length of treatment: 16 weeks Placebo Frequency: daily Dose: placebo Route of administration: oral Length of treatment: 16 weeks
Outcomes	Outcomes measured in trial Primary outcome Change in brain gray matter volume Secondary outcome Change in pain magnitude (WOMAC Pain subscale) Outcomes included in review (selected to align with protocol) Major Pain Proportion of people who experienced any adverse effects Proportion of people who experienced serious adverse event
Identification	Sponsorship source: Eli Lilly and Company; National Institute of Neurological Disorders and Stroke National Center for Complementary and Integrative Health Country: US Trial registration: NCT01558700 Author's name: Pascal Tetreault Institution: Department of Physiology, Northwestern University, Feinberg School of Medicine, Chicago, IL, US Email: a‐apkarian@northwestern.edu
Notes	Adverse effects Total adverse effects: placebo: 6/21 (28.6%); duloxetine: 6/19 (31.6%) Serious adverse effects: placebo: 0; duloxetine: 0 Nature of events: placebo: 3 worsening of pain, 3 dizziness/grogginess; duloxetine: 2 worsening of pain, 4 dizziness/grogginess Withdrawals due to adverse events: not reported Analysis Pain scores normalisation of WOMAC score 0–96 scale to 0–10 scale using (score/96) × 10
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	1:1 randomised allocation performed. Inadequate details provided.
Allocation concealment (selection bias)	Unclear risk	Participants blinded to allocation; no details provided.
Blinding of participants and personnel (performance bias) All outcomes	Unclear risk	Assessors blinded to allocation; no details provided. Inadequately described.
Blinding of outcome assessment (detection bias) Self‐reported outcomes (pain, function, quality of life, adverse events)	Low risk	Participants completed assessments individually.
Blinding of outcome assessment (detection bias) Assessor‐reported outcomes (radiographic joint structure)	Low risk	Assessors (radiologists for MRI) were blinded to participant allocation. Images analysed by computer program.
Incomplete outcome data (attrition bias) All outcomes	High risk	No data regarding discontinuation of study. Protocol stated (quote) "subjects that terminate the trial before completion of the final visit and scan will be replaced."
Selective reporting (reporting bias)	High risk	Baseline data on prior medication use of baseline WOMAC Pain scores not provided.
Other bias	Low risk	No other sources of bias identified.

Uchio 2018.

Study characteristics
Methods	Study design: RCT Study grouping: parallel group Setting: 47 medical institutions in Japan Time period: October 2014 to June 2015 Intervention: duloxetine vs placebo Sample size calculator: in a meta‐analysis of the results from Chappell 2009b and Chappell 2011, the difference in the change in BPI‐Severity mean pain between the duloxetine and placebo groups was estimated to be −0.9, with an SD of 2.0. Because a dose increase from 60 mg to 120 mg was performed in the duloxetine group, this estimated size effect may have included the dose‐increase effect. Therefore, by estimating a 20% smaller effect size in the present study and by ensuring a power of ≥ 90% for the between‐group comparisons using 2‐sample t‐tests at a 2‐sided significance level of 5%, the target sample size was determined to be 340 participants, with 170 participants per group Analysis: ITT with participants who received ≥ 1 dose of study drug and had a BPI‐Severity mean pain recorded at baseline.
Participants	Number of participants Number screened: 395 Number excluded: 41 Number randomised: placebo: 176; duloxetine: 178 Number included in analysis: placebo: 176 at week 0, 162 at week 14; duloxetine: 77 at week 0, 161 at week 14 Baseline characteristics Antidepressant Age, mean (years): 65.5 (SD 8.0) % female: 80.2% Analgesic use – % NSAID: 55.4% Disease duration (years): 4 (SD 4.2) Baseline pain, mean: 5 (SD 1.0) Function: no baseline data reported (WOMAC Total score) Quality of life: no baseline data reported (SF‐36 Mental Health Component/EQ‐5D) Placebo Age, mean (years): 66.4 (SD 8.4) % female: 75.0% Analgesic use – % NSAID: 56.8% Disease duration (years): 4.5 (SD 4.3) Baseline pain, mean: 5.1 (SD 1) Function: no baseline data reported (WOMAC Total score) Quality of life: no baseline data reported (SF‐36 Mental Health Component) Inclusion criteria Men and women aged 40–80 years ACR criteria for idiopathic knee OA (knee pain, bone spurs detected on x‐ray images, ≥ 1 of age > 50 years, morning stiffness resolving within 30 minutes or crepitus) Had pain for ≥ 14 days of each month for 3 months prior to study entry. Participants must have a score of ≥ 4 on the BPI mean pain score before randomisation. Women of child‐bearing potential must test negative on a pregnancy test Exclusion criteria Previous administration of duloxetine Inflammatory arthritis (e.g. rheumatoid arthritis) or autoimmune diseases (except Hashimoto's disease and type 1 diabetes) Invasive treatment (e.g. joint lavage or intra‐articular hyaluronic acid or steroid injections) in either knee within 1 month before visit 1 Arthroscopic surgery of the affected joint within 1 year before visit 1 or history of joint replacement or osteotomy End‐stage OA (e.g. loss of joint space or loss of articular cartilage) or scheduled to undergo surgery of affected joint during study Major depressive disorders based on the Mini International Neuropsychiatric Interview; and suicidal tendencies according to the C‐SSRS
Interventions	Intervention characteristics Antidepressant Frequency: daily Dose weeks 0–1 duloxetine 20 mg weeks 1–2 duloxetine 40 mg weeks 2–14 duloxetine 60 mg weeks 14–15 duloxetine 40 mg weeks 15–16 duloxetine 20 mg Route of administration: oral Type of antidepressant: duloxetine (SSRI) Length of treatment: 14 week (16 weeks' duloxetine exposure) Placebo Frequency: daily Dose: placebo Route of administration: oral Length of treatment: 14 weeks (16 weeks' total exposure)
Outcomes	Outcomes measured in trial Primary outcome Change from baseline to 14 weeks on BPI 24‐hour mean pain score Secondary outcomes PGI‐I Change from baseline to 14 weeks on CGI‐S Change from baseline to 14 weeks on SF‐36 Change from baseline to 14 weeks on the BDI‐II total score Number of participants with fall events from Fall Questionnaire Change from baseline to 14 weeks on WOMAC Total score Change from baseline to 14 weeks on PGAI score Change from baseline to 14 weeks on EQ‐5D Version of the European Quality of Life Instrument Change from baseline to 14 weeks in BPI Pain Severity Item and Interference Item. Interference items (general activity, mood, walking ability, normal work, relationship with others, sleep and life enjoyment). Proportion of participants with a ≥ 30% and ≥ 50% reduction in mean pain score on weekly mean of the 24‐hour mean pain score on the 11‐point Numeric Rating Scale Change from baseline to 14 weeks on weekly mean of the 24‐hour mean pain and worst pain score, least pain and pain right now Proportion of participants with reduction of ≥ 30% and ≥ 50% in BPI mean pain score Change from baseline to 14 weeks on the WOMAC Physical Function Pain subscale Change from baseline to 14 weeks on the WOMAC Stiffness subscale Kellgren Lawrence grade of radiological progression; reported as proportion who were worse, proportion who were the same and proportion who improved at 15 weeks Health‐related quality of life specific to OA assessed using WOMAC OMERACT response rate, calculated using the 24‐hour BPI‐Severity mean pain score, the WOMAC subscale score for "difficulty in performing daily activities," and the PGAI. Outcomes included in review (selected to align with protocol) Major Pain – mean change from baseline overall pain (0–10 Likert 24‐hour mean pain scale) Responder, ≥ 50% reduction in mean pain Function – change from baseline WOMAC Total scale (0–96 scale) Quality of life – EQ‐5D Proportion of participants who discontinued study due to adverse effects Proportion of people who experienced any adverse effects Proportion of people who experienced serious adverse event Minor Responder (OARSI Response Criteria) Radiographic progression of joint changes (based on Kellgren Lawrence scores) Proportion of participants who withdrew from the study for any reason
Identification	Sponsorship source: Eli Lilly and Company Country: Japan Setting: outpatients Trial registration: NCT02248480 Author's name: Yuji Uchio Institution: Shimane University School of Medicine, Izumo, Shimane, Japan Email: orthop@med.shimane‐u.ac.jp
Notes	Adverse effects Total adverse effects: placebo: 98/176; duloxetine: 120/178 Serious adverse effects: placebo: 1/176 (0.57%); duloxetine: 1/178 (0.56%) Nature of events: cancerous peritonitis, cerebellar tumour Withdrawals from trial due to adverse effects: placebo: 2/176; duloxetine: 11/178 Analysis Normalised WOMAC Total score from 0–96 to 0–100 using equation (score/96) × 100 For radiographic changes participants were said to have a response if their Kellgren Lawrence score remained the same at conclusion of treatment or improved at conclusion of treatment.
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Random allocation using a Web response system and stochastic minimisation method using the BPI pain severity mean pain as an allocation factor.
Allocation concealment (selection bias)	Low risk	Interactive web response system used.
Blinding of participants and personnel (performance bias) All outcomes	Low risk	Double‐blind. Quote: "Drugs were identical in appearance, odour and taste."
Blinding of outcome assessment (detection bias) Self‐reported outcomes (pain, function, quality of life, adverse events)	Low risk	Double‐blind. Note: there is no mention of when study personnel were unblinded.
Blinding of outcome assessment (detection bias) Assessor‐reported outcomes (radiographic joint structure)	Unclear risk	Investigators were blinded to participants allocated groups. No specific mention made to review of radiographic joint structures. Quote: "there were no changes to the study design or statistical analyses after starting enrolment and unblinding."
Incomplete outcome data (attrition bias) All outcomes	Low risk	Details of those who discontinued treatment were given including participants found to be ineligible, participant request, adverse events, lack of efficacy or worsening disease and other reason. 91% of participants completed study. Withdrawals: placebo: 14/176 (7.95%); duloxetine: 17/178 (9.55%) Reasons for withdrawals published
Selective reporting (reporting bias)	Low risk	All specified outcomes reported on. Note: reporting of Kellgren Lawrence scores could not be easily interpreted.
Other bias	Low risk	No other sources of bias identified.

Wang 2017.

Study characteristics
Methods	Study design: double‐blind RCT Study grouping: parallel group Setting: 17 study centres in China Time period: 13‐week double‐blind research period and 1‐week taper phase Intervention: duloxetine vs placebo Sample size calculator: sample size was calculated based on 90% power to detect a 0.73‐point difference with 2.2 SD between duloxetine and placebo for BPI – 24‐hour mean pain rating (primary outcome) Analysis: for efficacy variables, analyses included all randomised participants with a baseline and ≥ 1 postbaseline observation. For the primary objective, change from baseline to endpoint on the BPI 24‐hour mean pain rating was analysed using a mixed‐model repeated‐measures approach. In general, when a repeated measures analysis was used, the model included the fixed categorical effects of treatment, pooled investigative site, visit and treatment‐by‐visit interaction, as well as the fixed continuous covariates of the baseline value of the variable being analysed and baseline value of the variable being analysed‐by‐visit interaction. ANCOVA using a LOCF approach for imputation of missing data applied as a supportive analysis. When an ANCOVA model was used to analyse a continuous variable, model contained main effects of treatment and pooled investigative site, and appropriate baseline value as covariates. Change from baseline to endpoint for other efficacy variables was analysed using a similar approach. A Fisher exact test was applied to categorical variables. Analysis of proportions of participants experiencing ≥ 30% and ≥ 50% reductions at endpoint (using LOCF) was performed for BPI 24‐hour mean pain. The proportions of participants reporting endpoint responses of either "much better" or "very much better" on the PGI‐I scale were also calculated.
Participants	Number of participants Number screened: 481 Number excluded: 74 Number randomised: 407; placebo: 202; duloxetine: 205 Number included in analysis: placebo: 202 at week 0, 176 at week 13; duloxetine: 205 at week 0, 166 at week 13 Baseline characteristics Antidepressant Age, mean (years): 61.2 (SD 8.2) % female: 78.0% Analgesic use – % NSAID: not reported Disease duration (years): 2.9 (SD 4.4) Baseline pain, mean: BPI 24‐hour mean 5.49 (SD 1.27) Function (WOMAC Total score): 37.75 (SD 13.97) Placebo Age, mean (years): 59.8 (SD 8.4) % female: 74.8% Analgesic use – % NSAID: not reported Disease duration (years): 2.7 (SD 4.2) Baseline pain, mean: BPI 24‐hour mean 5.41 (SD 1.21) Function (WOMAC Total score): 36.68 (SD 13.95) Inclusion criteria Men and women aged ≥ 40 years who met clinical and radiographic criteria for the diagnosis of OA of the knee or hip. Knee OA criteria included knee pain, osteophytes (with radiographic evidence) and ≥ 1 of: aged > 50 years, morning stiffness < 30 minutes or crepitus. For the hip, OA criteria included hip pain and ≥ 2 of: erythrocyte sedimentation rate < 20 mm/hour, femoral or acetabular osteophytes (with radiographic evidence) pace narrowing (superior, axial, medial or a combination) Pain for > 14 days of each month for 3 months prior to study entry Rating > 4 on the BPI 24‐hour mean pain item during screening, prior to treatment Exclusion criteria Previously completed/withdrawn from this study or any other study investigating duloxetine Previous exposure to duloxetine Diagnosis of psychosis, bipolar disorder or schizoaffective disorder Current (within 1 year of screening) Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition Axis I diagnosis of major depressive disorder, anxiety disorders (excluding phobias), alcohol or eating disorders, as determined by the Mini‐International Neuropsychiatric Interview or a previous diagnosis History of substance abuse or dependence within the past year, excluding nicotine and caffeine Receiving any excluded medications that could not be discontinued at screening Treatment with an MAOI within 14 days of randomisation or the potential need to use an MAOI during study or within 5 days of discontinuation of investigational product Positive urine drug screen for any substance of abuse or excluded medication Serious cardiovascular, hepatic, renal, respiratory or haematological illness, or other medical or psychiatric condition that, in the opinion of the investigator, would compromise participation or be likely to lead to hospitalisation during study History of recurrent seizures other than febrile seizures Judged clinically by the investigator to be at suicidal risk according to the C‐SSRS: a "Yes" answer to either Question 4 (Active Suicidal Ideation with Some Intent to Act, Without Specific Plan) or Question 5 (Active Suicidal Ideation with Specific Plan and Intent) on the "Suicidal Ideation" portion of the C‐SSRS or a "Yes" answer to any of the suicide‐related behaviours (actual attempt, interrupted attempt, aborted attempt, preparatory act or behaviour) on the "Suicidal Behavior" portion of the C‐SSRS, with the ideation/behaviour having occurred within the previous month Uncontrolled narrow‐angle glaucoma Acute liver injury (such as hepatitis) or severe cirrhosis Hypersensitivity to duloxetine or any of the inactive ingredients or had frequent/severe allergic reactions to multiple medications Had frequent falls that could result in hospitalisation or could compromise response to treatment Diagnosis of inflammatory arthritis (i.e. rheumatoid arthritis) or an autoimmune disorder (excluding inactive Hashimoto's thyroiditis) Received intra‐articular hyaluronate/steroids, joint lavage or other invasive therapies to the index joint in the previous 3 months Had arthroscopy of the index joint within previous year or joint replacement of index joint at any time Had surgery of index joint scheduled during trial or anticipated by investigator to require surgery for treatment of OA of the index hip or knee along duration of study Had a prior synovial fluid analysis showing a white blood cell count ≥ 2000 mm³ that is indicative of a diagnosis other than OA Non‐ambulatory or require use of crutches or a walker BMI > 40 kg/m² Anticipated by the investigator to require use of analgesic agents including, but not limited to, NSAIDs, paracetamol and opioids, or other excluded medication for duration of study
Interventions	Intervention characteristics Antidepressant Frequency: daily Dose: 60 mg 1‐week titration phase (duloxetine 30 mg) 12‐week treatment phase (duloxetine 60 mg) 1‐week taper phase (duloxetine 30 mg) Route of administration: oral Type of antidepressant: duloxetine (SSRI) Length of treatment: 13 weeks (at time of analysis prior to 1‐week taper) Placebo Frequency: daily Dose: placebo Route of administration: oral Length of treatment: 14 weeks
Outcomes	Outcomes measured in trial Primary outcome Change from baseline to 13 weeks in BPI 24‐hour mean pain rating Secondary outcomes PGI‐I Change from baseline in WOMAC Total and subscale scores (Pain, Stiffness and Function) Change from baseline in CGI‐S score Change from baseline in BPI‐Severity including worst pain, least pain and pain right now Change from baseline in BPI‐I Change from baseline in HADS‐D or HADS‐A subscale scores Percentage of participants with reduction of ≥ 30% and ≥ 50% in BPI 24‐hour mean pain score Percentage of participants with response to treatment on PGI‐I at endpoint Outcomes included in review (selected to align with protocol) Major Pain – mean change from baseline overall pain (0–10 Likert 24‐hour mean pain scale) Responder, ≥ 50% reduction in mean pain Function – change from baseline WOMAC Total scale (0–96 scale) Proportion of participants who discontinued study due to adverse effects Proportion of people who experienced any adverse effects Proportion of people who experienced serious adverse events Minor Proportion of participants who withdrew from the study for any reason
Identification	Sponsorship source: Eli Lilly and Company Country: China Setting: 12 study centres in China Trial registration: NCT01931475 Author's name: G Wang Institution: China‐Japan Friendship Hospital; Lilly Suzhou Pharmaceutical Co Ltd, Shanghai Branch 21F, Shanghai, China Email: guochunwang@hotmail.com
Notes	Adverse effects Total adverse effects: placebo: 83/198; duloxetine: 121/199 Serious adverse effects: placebo: 3/198 (1.5%); duloxetine: 0/199 (0%) Nature of events: ventricular extrasystoles, intestinal polyp, gastrointestinal haemorrhage Withdrawals from trial due to adverse effects: placebo: 26/202; duloxetine: 39/205 Data analysis For continuous outcomes, used least squares mean (take from ClinicalTrials.gov) – we included these data as presented. Normalised WOMAC Total score from 0–96 to 0–100 using equation (score/96) × 100
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Quote: "randomized in a 1:1 ratio… determined by a computer‐generated random sequence using interactive web‐response system."
Allocation concealment (selection bias)	Low risk	Quote: "IWRS was used to assign investigational product packages to each patient."
Blinding of participants and personnel (performance bias) All outcomes	Low risk	Quote: "Patients, investigators, and all other personnel involved in the conduct of the study were blinded to individual treatment assignments during the double‐blind treatment phase."
Blinding of outcome assessment (detection bias) Self‐reported outcomes (pain, function, quality of life, adverse events)	Low risk	Quotes: "Patients, investigators, and all other personnel involved in the conduct of the study were blinded to individual treatment assignments during the double‐blind treatment phase." "Patients' treatment assignment was kept blinded to patients… until the end of the study."
Blinding of outcome assessment (detection bias) Assessor‐reported outcomes (radiographic joint structure)	Low risk	No assessor‐reported outcomes were measured.
Incomplete outcome data (attrition bias) All outcomes	Low risk	84% of participants randomised completed study. Withdrawals: placebo: 26/202 (12.87%); duloxetine: 39/205 (19.02%)
Selective reporting (reporting bias)	High risk	1 prespecified secondary endpoint omitted (HADS) not included in results and 1 non‐prespecified endpoint (concomitant use of short‐acting analgesics) included in results.
Other bias	Low risk	No other sources of bias identified.

ACR: American College of Rheumatology; ALT: alanine aminotransferase; ANCOVA: analysis of covariance; AST: aspartate transaminase; b‐POMS: Profile of Mood States – Brief Form; BDI‐II: Beck Depression Inventory – II; BMI: body mass index; BPI: Brief Pain Inventory; C‐SSRS: Columbia – Suicide Severity Rating Scale; CES‐D: Center for Epidemiological Studies Depression Scale; CGI‐S: Clinical Global Impression – Severity; EQ‐5D: EuroQol 5‐Dimensions; HADS: Hospital Anxiety and Depression Scale; HADS‐A: Hospital Anxiety and Depression Scale – Anxiety subscale; HADS‐D: Hospital Anxiety and Depression Scale – Depression; Hb: haemoglobin; ITT: intention‐to‐treat; LOCF: last observation carried forward; MAOI: monoamine oxidase inhibitor; MCS: Medical Component Summary; MRI: magnetic resonance imaging; NSAID: non‐steroidal anti‐inflammatory drug; OA: osteoarthritis; OARSI: Osteoarthritis Research Society International; PASS: Pain Anxiety Symptoms Scale; PCS: Physical Component Summary; PDI: Pain Disability Index; PGAI: Patient Global Assessment of Illness; PGI‐I: Patient Global Impression of Improvement; RCT: randomised controlled trial; SD: standard deviation; SF‐36: 36‐item Short Form; SSRI: selective serotonin reuptake inhibitor; TCA: tricyclic antidepressant; VAS: visual analogue scale; WOMAC: Western Ontario and McMaster Universities Arthritis Index.

Characteristics of excluded studies [ordered by study ID]

Study	Reason for exclusion
Alaka 2013	Review – not an RCT
Bellingham 2010	Review – not an RCT
Brown 2013	Review – not an RCT
Brunton 2010	Review – not an RCT
Chen 2019	Review – not an RCT
Citrome 2012a	Review – not an RCT
Citrome 2012b	Review – not an RCT
Driban 2012	Not an RCT
Ferreira 2021	Review – not an RCT
Fishbain 2000	Review – not an RCT
Gaynor 2010	Review – not an RCT
Gaynor 2013	Not an RCT
Hochberg 2012	Not an RCT
Lin 2003	Wrong patient population, participants included had depression and no details provided regarding types of antidepressants. Co‐intervention of psychotherapy not in placebo group.
Moore 2014	Review – not an RCT
Myers 2012	Review – not an RCT
Osani 2019	Review – not an RCT
Peng 2013	Not an RCT
Pergolizzi 2013	Review – not an RCT
Sabourin 2013	Oral presentation – not an RCT
Sawynok 2003	Review – not an RCT
Skljarevski 2009	Review – not an RCT
Skljarevski 2011	Review – not an RCT
Sofat 2017	Participants did not meet inclusion criteria
Sullivan 2009a	Not an RCT
Sullivan 2009b	Not an RCT
Talati 2010	Review – not an RCT
Teran Estrada 2011	Review – not an RCT
Thorpe 1974	Wrong patient population – contained people with rheumatoid arthritis as well and unable to separate groups.
Wang 2015	Review – not an RCT
Weng 2020	Not an RCT
Williamson 2014	Review – not an RCT
Wohlreich 2009	Wrong patient population – did not clarify type of arthritis
Wohlreich 2013	Letter – not an RCT

RCT: randomised controlled trial.

Characteristics of studies awaiting classification [ordered by study ID]

Abou‐Raya 2012.

Methods	Randomised controlled trial Parallel group
Participants	Baseline characteristics Antidepressant Age, mean (years): 68.9 (SD 6.2) % female: 84.0% Analgesic use – % NSAID: 89% Disease duration (years): 5.7 (SD 4.9) Baseline pain, mean: 9.1 (SD 4.6) – WOMAC Pain scale Placebo Age, mean (years): 68.5 (SD 5.8) % female: 83.3% Analgesic use – % NSAID: 90% Disease duration (years): 5.6 (SD 4.5) Baseline pain, mean: 8.9 (SD 5.1) – WOMAC Pain scale Inclusion criteria Clinical ACR clinical and radiographic criteria of primary knee OA with knee pain (> 40 on the 24‐hour mean pain severity scale 0–100 using mean of daily ratings from week preceding randomisation for 14 days/month during 3 consecutive months. Radiological Kellgren‐Lawrence grade I–III tibiofemoral or patellofemoral OA found on weight‐bearing anterioposterior and sunrise view radiographs Exclusion criteria Morbid obesity BMI > 32 kg/m² Inflammatory joint conditions or crystal arthropathies Concomitant disease (neuropsychiatric disease, including cognitive impairment, Alzheimer's disease, Parkinson's disease, cerebrovascular disease, cardiovascular disease, liver and renal disease) Use of antidepressants that could interfere with evaluation of intervention
Interventions	Antidepressant Frequency: daily Dose: 60 mg Route of administration: oral Type of antidepressant: duloxetine (SSRI) Length of treatment: 16 weeks Placebo Frequency: daily Dose: placebo Route of administration: oral Length of treatment: 16 weeks
Outcomes	Primary outcome Clinical response according to OARSI 2004 criteria at the end of 16 weeks (Pham 2004b). Participants were classified as responders if the pain or physical function score decreased by ≥ 20% and ≥ 20 mm on VAS pain Secondary outcomes Improvements in function using the WOMAC scores Use of OA rescue medication (paracetamol and NSAIDs)
Notes	Trial registration: NCT01425827 Registered: 30 August 2011 Primary completion date: June 2011 Results currently under investigation

Eli Lilly and Company 2009.

Methods	Randomised controlled trial Parallel group
Participants	Male or female outpatients Aged > 40 years ACR clinical and radiographic criteria for the diagnosis of OA of knee Pain for ≥ 14 days of each month for 3 months Pain rating ≥ 4 on the BPI mean pain rating scale at both visit 1 and visit 2 ACR clinical and radiographic criteria for classification of idiopathic OA of knee include: knee pain, osteophytes (with radiographic evidence) and ≥ 1 of: age > 50 years, morning stiffness < 30 minutes and crepitus Presence of osteophytes in index knee required documented radiographic evidence from a past imaging study (plain x‐ray, computerised tomography or magnetic resonance imaging), or from a current x‐ray taken during the screening period.
Interventions	Duloxetine 60 mg vs placebo
Outcomes	Primary outcomes BPI Secondary outcomes Discontinuation rates Adverse events Treatment‐emergent adverse events Discontinuations due to adverse events Serious adverse events Changes in laboratory test values Changes in vital signs and weight Columbia – Suicide Severity Rating Scale
Notes	Drug labelling error led to discontinuation of study

Glick 1976.

Methods	Double‐blind crossover trial
Participants	OA
Interventions	Imipramine 25 mg 3 times daily vs placebo
Outcomes	No further details
Notes	Unable to locate any results for this study despite literature search. Given age of study, it is presumed that it has concluded and no results have been published.

Schiphof 2015.

Methods	Randomised controlled trial
Participants	Age >18 years Hip or knee OA based on the clinical ACR criteria Chronic pain (most days of the last 3 months) in hip or knee Have: contraindication for NSAIDs or adverse reactions of NSAIDs or insufficient benefit of NSAIDs.
Interventions	Duloxetine vs placebo
Outcomes	Primary outcome Pain Secondary outcomes Cost‐effectiveness Presence of a neuropathic pain component favourably modifies response to treatment Whether it is only cost‐effective in people with a neuropathic component
Notes	Study ended prematurely. No details regarding outcomes or reason study ended prematurely.

ACR: American College of Rheumatology; BMI: body mass index; BPI: Brief Pain Inventory; NSAID: non‐steroidal anti‐inflammatory drug; OA: osteoarthritis; OARSI: Osteoarthritis Research Society International; SD: standard deviation; SSRI: selective serotonin reuptake inhibitor; VAS: visual analogue scale; WOMAC: Western Ontario and McMaster Universities Arthritis Index.

Characteristics of ongoing studies [ordered by study ID]

ACTRN12615000301561.

Study name	Does amitriptyline reduce pain in knee osteoarthritis?
Methods	Randomised controlled trial of low‐dose amitriptyline versus benztropine over 3 months
Participants	Men and women with symptomatic knee OA for ≥ 3 months, defined by the ACR clinical and radiographic criteria (i.e. pain and radiographic OA) Pain score ≥ 30 mm on 100‐mm VAS Age > 40 to < 75 years
Interventions	Intervention: low‐dose amitriptyline 25 mg orally in capsule form, alternate days for 2 weeks, then daily administration for 3 months Comparator: benztropine 1 mg orally in capsule form, on alternate days for 2 weeks then daily administration for 3 months
Outcomes	WOMAC Pain subscale Change in WOMAC Total score OARSI Response Criteria
Starting date	7 July 2015
Contact information	Ms Molly Bond; jointstudy@monash.edu
Notes

ACTRN12619001082190.

Study name	Serotonin noradrenaline reuptake inhibitor (SNRI) medications for the treatment of osteoarthritis pain (STOP) trial
Methods	Double‐blind, randomised, non‐inferiority trial of venlafaxine vs duloxetine for 9 weeks Participant adherence will be monitored using daily drug diaries, weekly follow‐up telephone calls and the return of any unused medications at end of intervention period.
Participants	Men and women aged ≥ 40 years who have radiographic evidence of knee OA and meet ACR clinical criteria for the diagnosis of knee OA History of knee pain for > 14 days of each month for ≥ 3 months BPI mean pain rating ≥ 4/10 at time of initial screening
Interventions	Intervention: inert gelatine placebo taken orally once a day in 1 opaque microcellulose capsule for 1 week; venlafaxine 75 mg taken orally once a day in 1 opaque microcellulose capsule for 1 week; venlafaxine 150 mg taken orally once a day in 1 opaque microcellulose capsules for 7 weeks (trial endpoint); venlafaxine 75 mg taken orally once a day in 1 opaque microcellulose capsule for 2 weeks (tapering dose). Control: inert gelatine placebo taken orally once a day in 1 opaque microcellulose capsule for 1 week; duloxetine 30 mg taken orally once a day in 1 opaque microcellulose capsule for 1 week; duloxetine 60 mg taken orally once a day in 1 opaque microcellulose capsules for 7 weeks (trial endpoint); duloxetine 30 mg taken orally once a day in 1 opaque microcellulose capsule for 2 weeks (tapering dose).
Outcomes	Primary outcome Change in the BPI mean pain rating Secondary outcomes Change in the BPI worst pain rating Change in the BPI least pain rating Change in the BPI pain right now rating Change in symptoms of anxiety and depression (Hospital Anxiety and Depression Scale) Change in physical function (lower limb tasks questionnaire) Change in quality of life (EQ‐5D) Patient Global Impression of Change with respect to their overall OA condition Number of treatment responders (≥ 50% improvement in BPI mean pain score) Number of treatment responders (≥ 30% improvement in BPI mean pain score)
Starting date	30 September 2019
Contact information	Dr David Rice; david.rice@aut.ac.nz
Notes

İlleez 2020.

Study name	Comparison of the efficacy of duloxetine and pregabalin in patients with knee osteoarthritis with mix type pain
Methods	Randomised parallel controlled trial
Participants	66 participants Inclusion criteria Aged 40–69 years Diagnosed with knee OA according to ACR criteria Grade 2–3 knee OA according to Kellgren‐Lawrence grading system VAS pain score: ≥ 4 Neuropathic Pain Diagnostic Questionnaire score: ≥ 4 BMI: ≤ 40 kg/m² Exclusion criteria History of trauma or surgical operation in the knee region Diabetes mellitus Fibromyalgia Inflammatory rheumatic disease Central or peripheral neurological disorder Severe cardiac, pulmonary or malignant disease Invasive treatment for knee in last 3 months
Interventions	Duloxetine: 33 participants received duloxetine HCl 60 mg/day orally for 12 weeks Pregabalin: 33 participants received pregabalin 300 mg/day orally for 12 weeks
Outcomes	Assessing VAS score change Neuropathic Pain Diagnostic Questionnaire Assessing WOMAC score change At 0, 4 and 12 weeks
Starting date	October 2016
Contact information	Özge Gülsüm İlleez, MD;  Fatih Sultan Mehmet Training and Research Hospital
Notes

IRCT20170716035126N2.

Study name	Evaluating the efficacy of duloxetine and gabapentin in pain reduction in patients with knee osteoarthritis
Methods	Patients randomised into 3 groups using random number blocks and sealed envelopes Measurements taken at 0, 1 and 3 months after intervention
Participants	Inclusion criteria Idiopathic knee OA Age 45–75 years Ability to walk Pain score > 5 on VAS and > 48 using WOMAC
Interventions	Intervention 1: duloxetine 30 mg daily, increased to 60 mg daily Intervention 2: gabapentin 300 mg daily, increased to 600 mg daily Control group: paracetamol 1000 mg, increased to 2000 mg daily
Outcomes	Pain: VAS Function: WOMAC score Quality of life: SF‐36 questionnaire
Starting date	March 2018
Contact information	Afsaneh Enteshari Moghaddam; afsanehenteshary@gmail.com Afshin Habibzadeh; afshin.habibzadeh@gmail.com
Notes	Currently recruiting

ntr4798.

Study name	Duloxetine for chronic osteoarthritis pain; an important alternative?
Methods	Pragmatic open, cluster randomised trial Included people with pain due to hip or knee OA on most days of past 3 months with insufficient benefit from NSAIDs or contraindications or intolerable adverse effects to NSAIDs. General practices randomised to duloxetine plus usual care or usual care only.
Participants	Inclusion criteria Age ≥ 18 years with: hip or knee OA based on the clinical ACR criteria chronic pain (most days of last 3 months) in hip or knee contraindication for NSAIDs; adverse reactions to NSAIDs or insufficient benefit of NSAIDs.
Interventions	Intervention: duloxetine 60  mg once a day. During the first week of treatment, duloxetine 30 mg per day to minimise potential adverse events. If tolerated well, dose increased to 60  mg/day in second week. Therapeutic benefit of duloxetine assessed regularly by treating GP (after 2  weeks; 1, 3, 6, 9, 12 months and when necessary). If participants experienced no effect of duloxetine after 3  months or intolerable adverse effects, duloxetine can gradually be discontinued. Participants also received usual care according to the GP guidelines in the Netherlands Comparator: usual care: treated according to the current Dutch GP guidelines which comprise education, lifestyle advice, dietary therapy, physiotherapy and analgesics. Analgesics prescribed according to opinion and experience of treating GP. Analgesic choice depends on what the GP and participants decide. Paracetamol and NSAIDs have failed in these participants. A GP can decide to prescribe or continue an opioid, or the treatment can remain the same. Depending on the comorbidities of the patients, contraindications and opinion of the GP other interventions are also allowed.
Outcomes	Primary outcome Pain at 3 months measured with the WOMAC Pain subscale Secondary outcome Pain at 1 year (WOMAC Pain subscale), disability (WOMAC Function subscale) Adverse reactions Quality of life with the EQ‐5D Compliance to treatment (Brief Medication Questionnaire) Patients' satisfaction OARSI‐OMERACT responder criteria Costs; including direct medical and patient costs (iMCQ) and productivity costs (iPCQ)
Starting date	1 April 2015
Contact information	Dr D Schiphof; d.schiphof@erasmusmc.nl
Notes

ACR: American College of Rheumatology; BMI: body mass index; BPI: Brief Pain Inventory; EQ‐5D: EuroQol 5‐Dimensions; GP: general practitioner; iMCQ: Institute for Medical Technology Assessment Medical Consumption Questionnaire; iPCQ: Institute for Medical Technology Assessment Productivity Cost Questionnaire; NSAID: non‐steroidal anti‐inflammatory drug; OA: osteoarthritis; OARSI: Osteoarthritis Research Society International; SF‐36: 36‐item Short Form; VAS: visual analogue scale; WOMAC: Western Ontario and McMaster Universities Arthritis Index.

Differences between protocol and review

Title 

Initially this was assumed to examine large joint lower limb osteoarthritis only. This was evident in the outcome measure hierarchy in the protocol. Thus we have amended the title to "Antidepressants for hip and knee osteoarthritis."

Types of participants

In the initial protocol, all joints were considered together. Since the protocol was published, increasing evidence suggests that the prevalence of pain sensitisation differs according to joint. It is more common in knee osteoarthritis than in hip osteoarthritis (Zolio 2021). Pain in hand osteoarthritis behaves differently to pain in knee and hip osteoarthritis (Allen 2009). Thus we considered hip and knee joints in this review.

Types of interventions

In the first instance, we examined the combined effect of the different antidepressant classes on osteoarthritis. However, as the various classes of antidepressant have alternative mechanisms, which may have differing effects on pain sensitisation, we considered the antidepressant classes separately, according to the Cochrane Common Mental Disorders Group.

We included studies of eight weeks or more as osteoarthritis is a chronic pain condition, in which pain levels fluctuate over weeks. Thus, a longer study period was desirable to capture any possible effect.

For example, we included (but did not limit to) the following comparisons: tricyclic antidepressants versus placebo or active comparator, selective serotonin reuptake inhibitors versus placebo or active comparator, serotonin noradrenaline reuptake inhibitors versus placebo or active comparator, heterocyclic antidepressants versus placebo or active comparator, and monoamine oxidase inhibitors versus placebo or active comparator. We grouped the other antidepressants (as per Description of the intervention) and compared against placebo or other active comparator.

Measures of treatment effect

For comparability of studies, data was normalised to a 0–100 scale for physical function (WOMAC), as some studies presented normalised data.

Subgroup analyses

We planned to conduct subgroup analyses based on the presence of other chronic conditions (e.g. fibromyalgia, concurrent depression), as the presence of these conditions may affect the measurement of outcomes. However, as no relevant data were available, this was not performed.

Results were stratified according to drug class, thus no subgroup analyses were performed.

Outcomes investigated

In the protocol, we set out to determine if antidepressants had physical effects on joints that were to be monitored radiographically as a major outcome. Given the lack of research into this area, whilst included, we have altered it to a minor outcome.

We have additionally made responder criteria a major outcome in this review, as it reflects clinical impacts of antidepressant use, and we considered this was of greater importance to patients and clinicians.

We used the WOMAC Global score rather than the functional subscale as it was consistently measured throughout the studies.

Contributions of authors

AL: study extraction, risk of bias assessment, data insertion, data analysis, review of results and discussion.

JS: study extraction, data extraction, risk of bias assessment, summary of findings table creation.

JL: first draft, preparation of protocol and search strategy, review of protocol, initial search, study extraction, risk of bias assessment.

LC: risk of bias assessment.

DU: review of first draft, review of protocol and search strategy, review of protocol and review.

FC: conception, review of protocol and search strategy, review of protocol and review.

AW: conception, supervision of first draft, preparation of protocol, search strategy, review of protocol, data extraction, review of results and discussion.

Sources of support

Internal sources

• St Kilda Road Department of Monash University's Department of Epidemiology and Preventative Medicine, Melbourne, Australia

External sources

• NHMRC, Australia

  • DU is the recipient of a Career Development Fellowship (Clinical Level 1, 1011975). AW is the recipient of a Career Development Fellowship (Clinical Level 2, 1063574)

• NHMRC, Australia

  Does low dose amitriptyline reduce pain in knee osteoarthritis? A double blind, randomised, pragmatic, placebo controlled clinical trial of amitriptyline in addition to usual care (AW, DU, FC – 103664)

• Royal Australasian College of Physicians, Australia

  AEW is the recipient of the Fellows Career Development Award

• NHMRC, Australia

  FMC is the recipient of an Investigator Grant (APP1194829)

Declarations of interest

AL: none.

JL: none.

JS: none.

DU: is a co‐investigator in a current ongoing study regarding amitriptyline in osteoarthritis (ACTRN12615000301561). They did not perform data extraction for this trial.

FC: is a co‐investigator in a current ongoing study regarding amitriptyline in osteoarthritis (ACTRN12615000301561). They did not perform data extraction for this trial.

LC: none.

AW: is the lead researcher in a current ongoing study regarding amitriptyline in osteoarthritis (ACTRN12615000301561). They did not perform data extraction for this trial. AW is currently a lead author on a systematic review and meta‐analysis on the prevalence of neuropathic‐like pain and central sensitisation in hip and knee osteoarthritis.

New