Drug therapy for obstructive sleep apnoea in adults

Abstract

Background

The treatment of choice for moderate to severe obstructive sleep apnoea (OSA) is continuous positive airways pressure (CPAP) applied via a mask during sleep. However, this is not tolerated by all individuals and its role in mild OSA is not proven. Drug therapy has been proposed as an alternative to CPAP in some patients with mild to moderate sleep apnoea and could be of value in patients intolerant of CPAP. A number of mechanisms have been proposed by which drugs could reduce the severity of OSA. These include an increase in tone in the upper airway dilator muscles, an increase in ventilatory drive, a reduction in the proportion of rapid eye movement (REM) sleep, an increase in cholinergic tone during sleep, an increase in arousal threshold, a reduction in airway resistance and a reduction in surface tension in the upper airway.

Objectives

To determine the efficacy of drug therapies in the specific treatment of sleep apnoea.

Search methods

We searched the Cochrane Airways Group Specialised Register of trials. Searches were current as of July 2012.

Selection criteria

Randomised, placebo controlled trials involving adult patients with confirmed OSA. We excluded trials if continuous positive airways pressure, mandibular devices or oxygen therapy were used. We excluded studies investigating treatment of associated conditions such as excessive sleepiness, hypertension, gastro‐oesophageal reflux disease and obesity.

Data collection and analysis

We used standard methodological procedures recommended by The Cochrane Collaboration.

Main results

Thirty trials of 25 drugs, involving 516 participants, contributed data to the review. Drugs had several different proposed modes of action and the results were grouped accordingly in the review. Each of the studies stated that the participants had OSA but diagnostic criteria were not always explicit and it was possible that some patients with central apnoeas may have been recruited.

Acetazolamide, eszopiclone, naltrexone, nasal lubricant (phosphocholinamine) and physiostigmine were administered for one to two nights only. Donepezil in patients with and without Alzheimer's disease, fluticasone in patients with allergic rhinitis, combinations of ondansetrone and fluoxetine and paroxetine were trials of one to three months duration, however most of the studies were small and had methodological limitations. The overall quality of the available evidence was low.

The primary outcomes for the systematic review were the apnoea hypopnoea index (AHI) and the level of sleepiness associated with OSA, estimated by the Epworth Sleepiness Scale (ESS). AHI was reported in 25 studies and of these 10 showed statistically significant reductions in AHI.

Fluticasone in patients with allergic rhinitis was well tolerated and reduced the severity of sleep apnoea compared with placebo (AHI 23.3 versus 30.3; P < 0.05) and improved subjective daytime alertness. Excessive sleepiness was reported to be altered in four studies, however the only clinically and statistically significant change in ESS of ‐2.9 (SD 2.9; P = 0.04) along with a small but statistically significant reduction in AHI of ‐9.4 (SD 17.2; P = 0.03) was seen in patients without Alzheimer's disease receiving donepezil for one month. In 23 patients with mild to moderate Alzheimer's disease donepezil led to a significant reduction in AHI (donepezil 20 (SD 15) to 9.9 (SD 11.5) versus placebo 23.2 (SD 26.4) to 22.9 (SD 28.8); P = 0.035) after three months of treatment but no reduction in sleepiness was reported. High dose combined treatment with ondansetron 24 mg and fluoxetine 10 mg showed a 40.5% decrease in AHI from the baseline at treatment day 28. Paroxetine was shown to reduce AHI compared to placebo (‐6.10 events/hour; 95% CI ‐11.00 to ‐1.20) but failed to improve daytime symptoms.

Promising results from the preliminary mirtazapine study failed to be reproduced in the two more recent multicentre trials and, moreover, the use of mirtazapine was associated with significant weight gain and sleepiness. Few data were presented on the long‐term tolerability of any of the compounds used.

Authors' conclusions

There is insufficient evidence to recommend the use of drug therapy in the treatment of OSA. Small studies have reported positive effects of certain agents on short‐term outcomes. Certain agents have been shown to reduce the AHI in largely unselected populations with OSA by between 24% and 45%. For donepezil and fluticasone, studies of longer duration with a larger population and better matching of groups are required to establish whether the change in AHI and impact on daytime symptoms are reproducible. Individual patients had more complete responses to particular drugs. It is possible that better matching of drugs to patients according to the dominant mechanism of their OSA will lead to better results and this also needs further study.

Plain language summary

Drug therapy for obstructive sleep apnoea in adults

Background 
 Obstructive sleep apnoea (OSA) is caused by collapse of the upper airway. The mainstay of medical treatment is continuous positive airways pressure (CPAP), delivered through a mask during sleep, aiming to keep the airway opened. Drug therapy has been proposed for individuals with mild OSA and those intolerant of CPAP.

Review question 
 We wanted to look at the evidence from randomised controlled trials which compared drug therapy with placebo for sleep apnoea.

What we did 
 We searched and reviewed all randomised placebo controlled trials of drugs to treat OSA in adult patients. Most of the trials had methodological flaws or were for only one or two nights, which is hardly long enough to find out whether the treatments are effective or acceptable to the people taking them.

Results 
 Of 25 drugs tested, 10 had some impact on the severity of OSA (as measured by the apnoea hypopnoea index or AHI) and four altered symptoms of sleepiness, although in most people changes were only modest.

Eszopiclone, acetazolamide, naltrexone, physiostigmine and nasal lubricant (phosphocholinamine) were trialled for one to two nights only and the long‐term effects are therefore unknown. A topical nasal steroid was well tolerated and reduced the severity of sleep apnoea and improved subjective daytime alertness in a specific subgroup of people with both OSA and rhinitis. Acetazolamide reduced the number of respiratory events per hour of sleep but did not reduce daytime sleepiness and was poorly tolerated long term. Paroxetine had only a small effect on the severity of OSA and while it was tolerated there was no useful effect on daytime symptoms. In contrast, participants reported a symptomatic benefit from protriptyline but there was no improvement in OSA, suggesting a different mechanism for their improved sense of well‐being. The antidepressant mirtazapine improved the severity of sleep apnoea in older trials, however two more recent, larger trials failed to demonstrated the same results; moreover, participants experienced side effects of sleepiness and weight gain. Donepezil, well established in the treatment of Alzheimer's disease, improved oxygen saturation and the severity of OSA in patients with and without dementia. In people with no dementia it was shown to reduce sleepiness but the study was small and neither the AHI nor ESS were normalised. Further trials are needed to confirm these results. Eszopliclone reduced the severity of OSA in selected patients nonetheless larger scale trials are needed to verify the results.

Bottom line 
 Some of the drugs may be helpful, however their tolerability needs to be considered in long‐term trials.

Background

Description of the condition

Obstructive sleep apnoea (OSA) is partial or complete upper airway occlusion during sleep. This leads to increased respiratory muscle effort against an obstructed upper airway in an attempt to maintain air flow, which may lead to arousal from sleep, sleep fragmentation and excessive daytime sleepiness. Furthermore, OSA is associated with neuropsychological impairment, metabolic and cardiovascular co‐morbidities and increased mortality. The first effective treatment for OSA was tracheostomy and this was widely used before less invasive alternatives became available (He 1988). In 1981 continuous positive airway pressure (CPAP) delivered via a nasal mask was shown to be an effective treatment that patients could tolerate at home (Sullivan 1981). CPAP has become the treatment of choice for patients with moderate to severe OSA and has been shown to have many important health benefits including improvement in ventilatory function (Berthon‐Jones 1987); significant reduction in daytime somnolence (Engleman 1998; Giles 2006); improved quality of life (Stradling 1999; Patel 2003); and reduced blood pressure, cardiovascular events and mortality (Doherty 2005; Marin 2005; Barbe 2010). Unfortunately its use is limited by incomplete compliance (Pepin 1999; Weaver 2008), and among those who do tolerate CPAP a majority would prefer alternative treatment such as an oral appliance (Bennett 1998). The role of CPAP in mild OSA has not been fully evaluated.

Description of the intervention

Alternative treatments for OSA remain of interest including conservative measures such as alcohol avoidance, weight reduction, improved sleep scheduling and altering sleep position (Shneerson 2001; Lam 2007). Oral appliances including mandibular advancement splints and surgery to the upper airway are also offered to patients with OSA (Lim 2006; Gagnadoux 2009). Finally pharmacological intervention has been proposed as an alternative to CPAP in some groups of patients (Hudgel 1995; Smith 2006).

How the intervention might work

The drugs that have been most commonly prescribed in an attempt to treat OSA include progestogens, acetazolamide, theophyllines and antidepressants (Brownell 1982; Espinoza 1987; Stepanski 1988; Whyte 1988; Cook 1989; Mulloy 1992; Kraiczi 1999; Hein 2000; Prasad 2010). Progestogens were considered in the management of OSA because it is less common in women than in men, is more common in women after the menopause than before, and because progestogens have been shown to increase ventilatory drive (Cistulli 1994). Acetazolamide also increases respiratory drive and has been proposed as a treatment for OSA. Theophyllines are primarily used as bronchodilators but may also increase ventilatory drive. Antidepressants including tricyclic drugs reduce the proportion of rapid eye movement (REM) sleep and in patients with OSA predominantly in REM sleep might be expected to reduce the number of apnoeas. Clonidine has a similar action and these drugs have therefore been proposed as a treatment for mild to moderate OSA (Hudgel 1995). More recently greater emphasis has been placed on the role of upper airway tone, independent of respiratory drive. A greater understanding of the role of serotonergic receptors has led to trials of drugs to increase (for example SSRIs) or decrease serotonergic tone (ondansetron). The observation of a decline in the number of cholinergic neurons in patients with degenerative brain disease associated with OSA prompted trial of physostigmine and donepezil, which increase cholinergic tone (Gilman 2003, Moraes 2008, Sukys‐Claudino 2012). Attempts have been made to reduce the frequency of OSAs by changing the characteristics of the upper airway using topical therapy to increase its cross‐sectional area or reduce the surface tension acting on the walls (Jokic 1998; Kiely 2004; Clarenbach 2008). Finally, a more recent study by Eckert 2011 assessed manipulation of the arousal threshold (the ease with which a patient wakes up to respiratory stimuli during the respiratory event) by the non‐benzodiazepine sedative eszopiclone. They hypothesised that an increase in arousal threshold (more difficult to wake up) with eszopiclone would allow appropriate time for the accumulation of physiological stimuli to enable upper airway muscle recruitment to abort an apnoea.

Why it is important to do this review

Guideline committees have recognised the potential benefits of an effective pharmacological therapy for OSA, however the role of drugs in the treatment of sleep apnoea remains unclear (SIGN 2003; Veasey 2008; Randearth 2011). This review examines the available evidence.

Objectives

To determine the clinical effectiveness of pharmacological agents in the treatment of adults with OSA.

Methods

Criteria for considering studies for this review

Types of studies

We included randomised, placebo controlled trials (RCTs) which were double or single blind trials of either parallel group or crossover design.

Types of participants

We included adults with a diagnosis of OSA based on a history of sleepiness and the results of studies performed during sleep. Eligible diagnoses were a desaturation index (DI) of at least five per hour, or an apnoea hypopnoea index (AHI) of at least five per hour. We excluded participants already using CPAP, nocturnal oxygen or mandibular advancement devices. We also excluded participants with OSA complicating a neurological disorder and participants with central sleep apnoea.

Types of interventions

Participants were randomised to use either placebo or active medication aimed at reducing the severity of OSA. The drugs could be administered intravenously, orally or topically to the upper airway. No restriction was placed upon the duration of study.

Types of outcome measures

Primary outcomes

1. Apnoea hypopnoea index (AHI)

2. Epworth Sleepiness Score (ESS)

Secondary outcomes

1. Desaturation index (DI)

2. Apnoea index (AI)

3. Hypopnea index (HI)

4. Respiratory disturbance index (RDI)

5. Minimum arterial saturation during sleep (min SaO₂)

6. Total sleep time (TST)

7. Objective measures of daytime sleepiness (including multiple sleep latency tests or the maintenance of wakefulness tests)

8. Measures of quality of life or health status such as the SF‐36

9. Adverse effects

Search methods for identification of studies

Electronic searches

We identified trials using the Cochrane Airways Group Specialised Register of trials, which is derived from systematic searches of bibliographic databases including the Cochrane Central Register of Controlled Trials (CENTRAL), MEDLINE, EMBASE and CINAHL, and handsearching of respiratory journals and meeting abstracts (see Appendix 1 for further details). We searched all records in the Specialised Register coded as 'sleep apnoea' using the following terms:

(drug* or pharmacological and (treatment*)) or progesterone or progestogen or medroxy* or "tricyclic anti depressant*" or protriptyline or amitriptyline or imipramine or ssri or fluoxetine or clonidine or modafinil or stimulant* or Buspirone or doxapram or dopram or naloxone or narcan or "opiod antagonist*" or nicotin* or "ACE inhibitor*" or ACE‐inhibitor* or cilazapril or captopril or enalapril or fosinopril or imidapril or lisinopril or perindopril or quinapril or ramipril or trandolapril or "anti hypertensive*" or anti‐hypertensive* or antihypertensive* or baclofen or mirtazapine or steroid* or *steroid or fluticasone.

The most recent search was carried out in July 2012.

Searching other resources

We searched bibliographies from included studies, reviews and texts for citations. We also contacted authors of studies to locate other unpublished or in‐progress studies which met the inclusion criteria.

Data collection and analysis

Selection of studies

Initially, two review authors (TL and IS, MM and IS for 2012 update) independently assessed the titles, abstracts and citations to determine potential relevance for the full review.

Subsequently, from the full text, both review authors independently assessed studies for inclusion based on the criteria for the population, intervention, study design and outcomes. Agreement was measured by simple agreement and kappa statistics.

Data extraction and management

Two review authors independently extracted data from published selected studies (IES and TL, MM and IES for 2012 update) and entered the data into Review Manager (RevMan). Data in table or graphic form were used if published and the authors were requested to confirm data extraction and provide clarification and additional information for the review.

Assessment of risk of bias in included studies

Included studies were subjected to quality assessment by both review authors (EJW and MM for the 2012 update) using two methods.

Original version to 2006 update

First, using the Cochrane approach to assessment of allocation concealment, trials were scored and entered using the following principles.

Grade A: adequate concealment. 
 Grade B: unclear concealment. 
 Grade C: obviously not adequate concealment.

Each study was also assessed for validity on a 0 to 5 scale, method of Jadad 1996, as follows.

1. Was the study described as randomised? (1 = yes, 0 = no)

2. Was the study described as double blind?(1 = yes, 0 = no)

3. Was there a description of withdrawals and dropouts? (1 = yes, 0 = no)

4. Was the method of randomisation well described and appropriate? (1 = yes, 0 = no)

5. Was the method of double blinding well described and appropriate? (1 = yes, 0 = no)

6. Deduct one point if methods of randomisation or blinding were inappropriate.

2012 update

We assessed the risk of bias for new trials only according to recommendations outlined in the Cochrane Handbook of Systematic Reviews of Interventions (Higgins 2008) for the following items:

1. allocation sequence generation;

2. concealment of allocation;

3. blinding of participants and investigators;

4. incomplete outcome data;

5. selective outcome reporting.

We graded each potential source of bias as high, low, or unclear risk of bias. We also noted other potential sources of bias.

Measures of treatment effect

For continuous variables, we calculated the results of individual studies as fixed‐effect model mean difference (MD) or standardised mean difference (SMD), with 95% confidence intervals (CI). For similar studies, we calculated the pooled MD or SMD and 95% CI.

Unit of analysis issues

For crossover studies, estimates of effects were entered as generic inverse variance data (GIV), deriving MDs between treatment groups with variance determined by the published 95% CI if available, or published P values from statistical tests for paired data if available (Higgins 2011).

Dealing with missing data

For the 2012 update, we contacted trial authors to request missing outcome data as necessary.

Assessment of heterogeneity

We planned to examine heterogeneity using the I² statistic. We planned to investigate heterogeneity by performing prespecified subgroup analysis if we were able to pool data from sufficient studies.

Assessment of reporting biases

We planned to assess potential reporting biases by visually inspecting funnel plots if we were able to perform a meta‐analysis on 10 or more studies for a single outcome.

Data synthesis

We used a fixed‐effect model to analyse data.

Subgroup analysis and investigation of heterogeneity

We planned to conduct subgroup analysis based on the following:

1. population (male versus female);

2. severity of OSA (AHI ≥ to 20 and < 20 per hour).

Sensitivity analysis

We planned to perform sensitivity analyses using the domains:

1. studies at a high risk of bias, where we felt the identified risks of bias were likely to have an effect on the treatment effect or our confidence in it versus studies without a significant risk of bias;

2. random‐effects versus fixed‐effect modelling.

Results

Description of studies

Results of the search

See Table 1 for reporting of search history results. Thirty studies met the entry criteria for this review.

1. Search history detail.

Issue/Year of CLIB	Search results
Issue 3, 2001 (All years to July 2000)	References identified (including duplicates): 56 Studies retrieved: 46 Studies meeting inclusion criteria: 9 Studies identified from additional source: 0 Studies excluded: 37
All years to July 2004	References identified (including duplicates): 117 Studies identified from additional source: 6 Studies retrieved: 27 Studies meeting inclusion criteria: 11. Studies excluded: 16
All years to July 2005	References identified: 156 Studies retrieved: 3 Excluded: 3
July 2005‐July 2006	References identified: 11 Studies retrieved: 1 Excluded: 1
All years to July 2012	References identified (including duplicates):146 Studies retrieved:47 Studies meeting inclusion criteria: 9 Studies excluded:38

Included studies

For a full description of each study, see Characteristics of included studies. An overview of study characteristics is presented in Table 2. A reference to an abstract in the previous version was recognised as describing the same study as in Carley 2007.

2. Study characteristics.

Drug class	Study	Intervention	Control	N participants	Study duration	AHI reported	ESS reported as an outcome
Drugs proposed to act on airway tone	Brownell 1982	Protryptilline 20 mg each night	Placebo	5	2 weeks	yes	‐
	Stepanski 1988	protryptilline 20 mg each night	Placebo	8	3 weeks	yes	‐
	Whyte 1988	protryptilline 10 to 20 mg per night for 2 weeks, acetazolamide 250 mg twice daily for 1 week	Placebo	10	1 week	yes	‐
	Kraiczi 1999	paroxetine 20 mg each night	Placebo	20	2 x 6 weeks	yes	‐
	Prasad 2010 4 arm parallel	ondansetron 24 mg, fluoxetine/ondansetron 5/12 mg, fluoxetine/ondansetron 10/24 mg.	Placebo	35	4 weeks	yes	Stanford sleepiness scale
	Carley 2007	Mirtazapine 4.5 or 15 mg each night	Placebo	12	1 week	yes	‐
	Marshall 2008 (1) 3‐way crossover dose‐finding	Mirtazapine various doses	Placebo	20	2 weeks	yes	‐
	Marshall 2008 (2) 3 way parallel	Mirtazapine 15mg versus mirtazapine 15mg/CD0012	Placebo	64	4 weeks	yes	‐
	Stradling 2003	Ondansetron 16 mg	Placebo	10	2 x 1 night	yes	‐
	Mendleson 1991	Buspirone 20 mg	Placebo	5	1 night? not specified	yes	‐
	Rasche 1999	Salmeterol	Placebo	20	3 nights	yes	‐
Drugs proposed to act on ventilatory drive	Espinoza 1987	IV aminophylline	Placebo	10	1 night	yes	‐
	Mulloy 1992	Theophylline 800 mg each night	Placebo	12	4 weeks	yes	‐
	Hein 2000	Theophylline to maintain therapeutic serum level ≥ 8mg/L	Placebo	22	2 x 2 weeks	yes	‐
	Cook 1989	Medroxyprogestone 50 mg 3 times daily	Placebo	10	1 week	yes	‐
	Diamond 1982	naloxone 2 mg at half‐hourly intervals during sleep	Placebo	4	unclear	‐	‐
	Ferber 1993	naltrexone 50 mg	Placebo	12	2 nights	yes
	Suratt 1986	Doxapram 0.5 mg/kg bolus, then 1 mg/mL infusion overnight	Placebo	4	2 x 1 night	‐	‐
	Mangin 1983	Almitrine 2 or 3 mg/kg up to 200 mg/day	Placebo	9	2 x 5 nights	‐	‐
Vasoactive drugs	Heitmann 1998	Mibefradil 50 mg each day	Placebo	53	8 nights	yes	‐
	Issa 1992	Clonidine 0.2 mg each night	Placebo	8	2 x 10 nights	yes	‐
Topical drugs for the upper airway	Kiely 2004	Fluticasone propionate 100 mcg	Placebo	23	2 x 4 weeks	yes	‐
	Jokic 1998	Phosphocholinamine 0.4 mL at lights out and then again after 3.5 hours	Placebo	10	2 x 1 night	yes	‐
	Clarenbach 2008	Xylomethazoline 3 x 0.15 mg drops of a 0.1% solution	Placebo	12	1 week	yes	yes
Miscellaneous	Bruckert 2010	Fenofibrate 145 mg	Placebo	34	4 weeks	yes	‐
	Eckert 2011	Eszolpiclone 3 mg	Placebo	17	1 night	yes	‐
	Hedner 1996	sabeluzole 10 mg twice daily	Placebo	12	4 weeks	‐	‐
	Hedner 2003	IV physostigmine 0.12 mcg/kg/min	Placebo	10	2 nights	yes	‐
	Moraes 2008	Donezepil 5 mg for 1 month, then 10 mg for 2 months	Placebo	23	3 months	yes	‐
	Sukys‐Claudino 2012	Donezepil 5 mg for 2 weeks, then 10 mg for 2 weeks	Placebo	22	2 x 2 weeks	yes	yes

Study design

Most studies had a crossover design with the exception of six studies which had parallel groups (Heitmann 1998; Marshall 2008 (2); Moraes 2008; Prasad 2010; Bruckert 2010; Sukys‐Claudino 2012). Two studies (Mangin 1983; Espinoza 1987) were single blind studies, the remaining trials were double blind.

Participants

The 30 studies recruited a total of 516 participants. Eighteen studies had sample sizes of fewer than 15 participants. Two studies recruited more than 50 patients (Heitmann 1998; Marshall 2008 (2). Where data on body mass index (BMI) were recorded, the trial populations were overweight or clinically obese, ranging from 26.8 kg/m² (Hedner 2003) to 36 kg/m² (Cook 1989). Using AHI as an indicator of the OSA severity, the participants were generally moderate‐severe where this was reported. Baseline AHI ranged from 13.8 (Hein 2000) to 98.0 (Prasad 2010). Neither threshold AHI nor baseline AHI were reported in Hedner 1996, however participants were described as suffering from moderate to severe OSA. Prasad 2010 excluded patients with severe OSA (SpO₂ < 75% for more than 5% of total sleep time and patients not able to stop CPAP treatment safely) and Eckert 2011 excluded patients with severe OSA (SpO₂ < 70% and AHI > 60 events/hr) as well as patients with a high arousal threshold (‐25 cm H₂O to ‐63 cm H₂O).

Some studies cited a desaturation index (DI) threshold as the entry criterion for OSA, and four studies diagnosed participants on the basis of history and polysomnography (PSG) and did not report AHI (Diamond 1982; Mangin 1983; Suratt 1986; Hedner 1996). It was impossible to be certain that no patients with central sleep apnoeas were recruited. However, where participants were described as suffering from OSA, and AHIs were reported as outcomes, this indicated that the patient populations in each study would have met entry criteria for OSA specific studies.

Participants with co‐existing disease were recruited in four trials: OSA and allergic rhinitis (Kiely 2004), OSA and chronic nasal congestion (Clarenbach 2008), OSA and arterial hypertension (Heitmann 1998), OSA and Alzheimer's disease (Moraes 2008).

Espinoza 1987 and Hedner 2003 excluded participants who had recent therapy with CPAP, and Stradling 2003 recruited participants who were on a waiting list for CPAP treatment.

Interventions

A total of 25 different study drugs were compared with placebo.

Drugs proposed to act on airway tone during sleep

1. Protriptyline (non‐sedating tricyclic, antidepressant) (Brownell 1982; Stepanski 1988; Whyte 1988)

2. Paroxetine (specific serotonin re‐uptake inhibitor, antidepressant) (Kraiczi 1999)

3. Fluoxetine (specific serotonin re‐uptake inhibitor, antidepressant) (Prasad 2010)

4. Mirtazapine (mixed antagonist and agonist acting on serotonin receptors) (Carley 2007; Marshall 2008 (1+2))

5. Ondansetron (5‐HT₃ antagonist, antiemetic) (Stradling 2003; Prasad 2009; Prasad 2010)

6. Buspirone (azapirone anxiolytic acting on serotonin receptors) (Mendleson 1991)

7. Salmeterol (long‐acting beta‐agonist, proposed to relax the inner pharyngoconstricting muscles) (Rasche 1999)

Drugs proposed to act on ventilatory drive

1. Aminophylline (bronchodilator which may increase ventilatory responses) (Espinoza 1987)

2. Theophylline (bronchodilator which may increase ventilatory responses) (Mulloy 1992; Hein 2000)

3. Acetazolamide (carbonic anhydrase inhibitor that produces a metabolic acidosis, which may increase ventilation) (Whyte 1988)

4. Medroxyprogesterone (synthetic progesterone, which may increase ventilatory responses) (Cook 1989)

5. Naloxone (inhibits endogenous opiates that are thought to depress respiration) (Diamond 1982)

6. Naltrexone (inhibits endogenous opiates that are thought to depress respiration) (Ferber 1993)

7. Doxapram (acts as a respiratory stimulant both peripherally at the carotid body and centrally at the respiratory centre) (Suratt 1986)

8. Almitrine (respiratory stimulant that may increase PaO₂ and decrease PaCO₂) (Mangin 1983)

Vasoactive drugs

1. Clonidine (alpha‐adrenergic agonist that suppresses REM sleep) (Issa 1992)

2. Mibefradil (calcium channel antagonist) (Heitmann 1998)

Topical drugs for the upper airway

1. Fluticasone (corticosteroid, thought to reduce nasal airflow resistance) (Kiely 2004)

2. Topical lubricant administered nasally (phosphocholinamin, thought to reduce surface tension forces in the upper airway) (Jokic 1998)

3. Xylometazoline (imidazoline derivate, nasal decongestant increasing nasal conductance) (Clarenbach 2008)

Miscellaneous

1. Sabeluzole (benzothiazole derivative with actions that are poorly understood) (Hedner 1996)

2. Physostigmine (cholinesterase inhibitor, increasing cholinergic tone in sleep) (Hedner 2003)

3. Donepezil (cholinesterase inhibitor, increasing cholinergic tone in sleep) (Moraes 2008; Sukys‐Claudino 2012)

4. Fenofibrate (peroxisome proliferator‐activated receptor‐alfa ligand with anti‐inflammatory and antioxidant effects, used for management of hypertriglyceridaemia and mixed dyslipidaemia) (Bruckert 2010)

5. Eszopiclone (non‐benzodiazepine sedative, increasing arousal threshold without impairing upper airway dilator muscle responsiveness) (Eckert 2011)

These drugs were administered orally apart from three studies that gave the drug as an intranasal spray (Jokic 1998; Kiely 2004; Clarenbach 2008) and two in which the study drugs were given intravenously (Espinoza 1987; Hedner 2003).

The duration of the studies varied considerably. There were nine single‐night studies in laboratory settings (Suratt 1986; Espinoza 1987; Mendleson 1991; Ferber 1993; Jokic 1998; Rasche 1999; Stradling 2003; Hedner 2003; Eckert 2011), seven studies of two to 10 days (Mangin 1983; Whyte 1988; Cook 1989; Issa 1992; Heitmann 1998; Carley 2007; Clarenbach 2008), four studies of two to three weeks duration (Brownell 1982; Stepanski 1988; Hein 2000; Marshall 2008 (1)), and seven studies of four weeks duration (Mulloy 1992; Hedner 1996; Kiely 2004; Marshall 2008 (2); Prasad 2010; Bruckert 2010; Sukys‐Claudino 2012). Two studies were of six to12 weeks duration (Kraiczi 1999; Moraes 2008). One study was of unclear duration (Diamond 1982).

Outcomes

Thirty studies reported the effects of drug therapy on measurements of sleep disordered breathing (namely AHI; AI; HI; DI and arousals). Eight studies recorded data on symptoms (Brownell 1982; Stepanski 1988; Kraiczi 1999; Hein 2000; Kiely 2004; Clarenbach 2008; Bruckert 2010; Sukys‐Claudino 2012) whilst Hedner 1996 reported patient preference for either active or placebo treatments. Total sleep time was recorded in eight studies (Cook 1989; Mendleson 1991; Mulloy 1992; Ferber 1993; Hedner 1996; Heitmann 1998; Kraiczi 1999; Hedner 2003).

Funding

Eight trials were supported by drug companies, five trials were supported by grants from research councils, and four trials were supported by grants from both research councils and drug companies. We were unclear about the funding of 13 trials.

Excluded studies

See Characteristics of excluded studies.

Risk of bias in included studies

Studies in previous published version of the review

Several of the studies had low Jadad ratings, with eight out of the 26 studies scoring two or less. The most common weaknesses were inadequate descriptions of randomisation, blinding and dropouts. In seven studies (Mulloy 1992; Hedner 1996; Jokic 1998; Kraiczi 1999; Hein 2000; Hedner 2003; Kiely 2004), adequate blinding was confirmed with an identical placebo specified. In one study (Whyte 1988) the placebo and active treatment presentation was clearly different, with different frequencies of administration. Two studies were single blind (Mangin 1983; Espinoza 1987). Jadad ratings are described in the Characteristics of included studies tables.

Studies included in the 2012 update

New included studies were assessed using the Cochrane risk of bias tool. Full details can be found in the Characteristics of included studies and see Figure 1 for an overview of our judgements. The following description of risk of bias relates to studies added in the 2012 update.

1.

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

Allocation

Five studies reported adequate randomisation procedures (Marshall 2008 (1+2); Moraes 2008; Eckert 2011; Sukys‐Claudino 2012) while we were concerned that Mangin 1983 was not described as randomised. Five studies reported adequate allocation concealment (Jokic 1998; Marshall 2008 (1+2); Moraes 2008; Eckert 2011). The remaining studies were not described in sufficient detail to allow judgement so were at unclear risk of bias.

Blinding

Eight studies described adequate blinding of participants and personnel (Jokic 1998; Hein 2000; Clarenbach 2008; Marshall 2008 (1+2); Moraes 2008; Eckert 2011; Sukys‐Claudino 2012) and nine studies reported adequate blinding of outcome assessors (Hein 2000; Carley 2007; Clarenbach 2008 ; Marshall 2008 (1+2); Moraes 2008; Prasad 2010; Eckert 2011; Sukys‐Claudino 2012). The remaining studies did not report enough detail to be sure.

Incomplete outcome data

All randomised participants completed the intervention in seven of the nine new studies included in the 2012 update. The number of dropouts in the other two studies did not cause particular concern (Marshall 2008 (1); Sukys‐Claudino 2012).

Selective reporting

The nine new studies for the 2012 update did not appear to have issues with incomplete reporting of outcomes although we did not have access to study protocols. We were able to enter data for AHI for six of these trials (Carley 2007; Marshall 2008 (1+2); Moraes 2008; Eckert 2011; Sukys‐Claudino 2012), however we were concerned that we were not able to extract appropriate AHI outcome data for four of the new trials nor ESS data for seven of these trials, which represent important patient outcomes.

Other potential sources of bias

In five crossover studies (Stepanski 1988; Whyte 1988; Mulloy 1992; Ferber 1993; Hein 2000) there was no washout period leading to a potential for differential carry‐over effects.

In one study 35% of patients also received hypnotherapy, which may have biased the results (Bruckert 2010).

Effects of interventions

The studies reported a number of diverse outcome measures, some of which were idiosyncratic. The absence of common outcome measures between studies hampered direct comparison. No unpublished data were available on request.

Drugs proposed to act on airway tone during sleep

Protriptyline versus placebo

Three crossover studies with 23 participants compared protriptyline with placebo (Brownell 1982; Stepanski 1988; Whyte 1988). In a single study with 10 participants (Whyte 1988), there was no significant difference in AHI or any other measurements of respiratory disturbance or oxygenation during sleep between active and placebo treatments (Analysis 1.1; Analysis 1.2; Analysis 1.3; Analysis 1.4; Analysis 1.5; Analysis 1.6; Analysis 1.7). In two studies, participants reported an improvement in subjective daytime sleepiness with protriptyline (Peto odds ratio (OR) 17.15; 95% CI 3.61 to 81.43), however this was not quantified using any recognised scale and no objective measurements were offered (Brownell 1982; Stepanski 1988). Subjective daytime sleepiness was measured using visual analogue scores in Whyte 1988 and no significant improvement was recorded.

1.1. Analysis.

Comparison 1 Protriptyline versus placebo, Outcome 1 Apnoea Hypopnea Index.

1.2. Analysis.

Comparison 1 Protriptyline versus placebo, Outcome 2 Apnoea Index.

1.3. Analysis.

Comparison 1 Protriptyline versus placebo, Outcome 3 Hypopnea Index.

1.4. Analysis.

Comparison 1 Protriptyline versus placebo, Outcome 4 Min SaO2.

1.5. Analysis.

Comparison 1 Protriptyline versus placebo, Outcome 5 Total sleep time.

1.6. Analysis.

Comparison 1 Protriptyline versus placebo, Outcome 6 Desaturation Index.

1.7. Analysis.

Comparison 1 Protriptyline versus placebo, Outcome 7 Arousal Index.

Paroxetine versus placebo

One crossover trial with 20 people compared paroxetine versus placebo (Kraiczi 1999). A significant difference was observed in AHI (‐6.10 events/hour; 95% CI ‐11.00 to ‐1.20; Analysis 2.1). Kraiczi 1999 reported no significant differences in symptoms and Comprehensive Psychopathological Rating Scale (CPRS), a measurement of psychopathological symptoms (Analysis 2.4), and there was no significant difference in adverse events (including indicators of headache, concentration, memory and mood; Analysis 2.6).

2.1. Analysis.

Comparison 2 Paroxetine versus placebo, Outcome 1 Apnoea Hypopnoea Index.

2.4. Analysis.

Comparison 2 Paroxetine versus placebo, Outcome 4 CPRS symptoms.

2.6. Analysis.

Comparison 2 Paroxetine versus placebo, Outcome 6 Adverse effects.

Mirtazapine versus placebo

Three studies, reported in two papers, comparing mirtazapine with placebo recruited a total of 96 participants (Carley 2007; Marshall 2008 (1+2)). Carley 2007 was a crossover trial assessing high (15 mg) and low (4.5 mg) doses of mirtazapine versus placebo. Marshall 2008 reported two trials: mirtazapine versus placebo, a three‐way crossover dose finding study of mirtazapine (7.5 mg, 15 mg, 30 mg,r 45 mg) and placebo for two weeks (each patient was exposed to a maximum of three doses) and a three‐arm parallel trial of mirtazapine 15 mg, mirtazapine 15 mg and compound CD0012 (dopaminergic and serotoninergic agent undergoing evaluation for efficacy in OSA), and placebo for four weeks. The latter combination is no longer being recommended as therapy by the company who manufacture it.

Carley 2007 reported a significant reduction in mean AHI (P < 0.004) by daily administration of mirtazapine at doses of 4.5 mg (22.3 ± 4.8 to 13.5 ± 3.7) and 15 mg (22.3 ± 4.8 to 11.4 ± 3.6) for one week. The crossover, dose finding study by Marshall 2008 (1) showed a significant increase in AHI in the mirtazapine 15 mg (P < 0.05) and mirtazapine 30 mg (P ≤ 0.05) groups. In the parallel trial by Marshall 2008 (2) there was no statistically significant reduction in AHI (MD ‐1.30; 95% CI ‐13.57 to 10.97; Analysis 3.1). We pooled data from the two crossover trials although the heterogeneity was high (I² = 90%) and the treatment effects were in opposite directions: overall there was no statistically significant benefit. In both studies by Marshall 2008 (1+2) there was a significant weight gain across all treatment groups.

3.1. Analysis.

Comparison 3 Mirtazapine versus placebo, Outcome 1 Apnoea Hypopnea Index.

Ondansetron versus placebo

Two studies with 45 participants compared ondansetron with placebo, one parallel (Prasad 2010) and one crossover (Stradling 2003), but we were unable to pool the data. Stradling 2003 reported no significant differences in oxygen desaturation index (ODI), AHI (Analysis 5.1), AI (Analysis 5.2) or HI. Prasad 2010 reported a trend toward increased AHI post‐treatment but this was not statistically significant (P = NS). No improvement in daytime sleepiness was observed using the Stanford sleepiness scale (Prasad 2010).

5.1. Analysis.

Comparison 5 Ondansetron versus placebo, Outcome 1 Apnoea Hypopnea Index.

5.2. Analysis.

Comparison 5 Ondansetron versus placebo, Outcome 2 Apnoea Index.

Ondansetrone and fluoxetine versus placebo

One trial involving 35 adults compared ondansetron and fluoxetine with placebo (Prasad 2010). High dose combined treatment with ondansetron 24 mg and fluoxetine 10 mg showed a 40.5% decrease in AHI from baseline at treatment day 28.

Buspirone versus placebo

One study with 10 participants compared buspirone with placebo (Mendleson 1991). The reported difference in AHI of 10 per hour (placebo 30 and buspirone 20) was not statistically significant. There were no differences in AI (Analysis 6.1), overnight oxygenation (Analysis 6.3) or total sleep time (Analysis 6.2).

6.1. Analysis.

Comparison 6 Buspirone versus placebo, Outcome 1 Apnoea Index.

6.3. Analysis.

Comparison 6 Buspirone versus placebo, Outcome 3 MinSaO2.

6.2. Analysis.

Comparison 6 Buspirone versus placebo, Outcome 2 Total sleep time.

Salmeterol versus placebo

One study involving 10 participants compared salmeterol with placebo (Rasche 1999). There were no significant differences in AHI (Analysis 7.1) or minimum saturation (Analysis 7.2).

7.1. Analysis.

Comparison 7 Salmeterol versus placebo, Outcome 1 AHI.

7.2. Analysis.

Comparison 7 Salmeterol versus placebo, Outcome 2 Minimum SaO2.

Drugs proposed to act on ventilatory drive

Aminophylline versus placebo

One single blind crossover trial involving 10 participants reported data from a single‐night treatment period (Espinoza 1987). There was no difference in AI (Analysis 8.1) or in HI (Analysis 8.2). There was no improvement in overnight oxygenation, as measured by the minimum oxygen saturation (Analysis 8.3). In contrast there was a clinically large, significant difference in sleep efficiency which favoured placebo (MD ‐24%; 95% CI ‐34.55 to ‐13.45; Analysis 8.4).

8.1. Analysis.

Comparison 8 Aminophylline versus placebo, Outcome 1 Apnoea Index.

8.2. Analysis.

Comparison 8 Aminophylline versus placebo, Outcome 2 Hypopnoea index.

8.3. Analysis.

Comparison 8 Aminophylline versus placebo, Outcome 3 Min SaO2 in REM.

8.4. Analysis.

Comparison 8 Aminophylline versus placebo, Outcome 4 Sleep efficiency.

Theophylline versus placebo

Two trials with 34 participants compared theophylline with placebo (Mulloy 1992; Hein 2000), although paired data were only reported in 23 participants. There was a small and clinically insignificant difference in the AHI between the treatment and placebo (MD ‐2.76 events/hr; 95% CI ‐8.57 to 3.05; Analysis 9.1). Self‐assessed sleep latency, number of awakenings and subjective sleep quality were not different between theophylline and placebo in Hein 2000 (25 versus 22 minutes, 2.1 versus 1.5, and 1.2 versus 1.4 for these outcomes, respectively). Mulloy 1992 reported a significant difference in the number of desaturations favouring treatment (‐79; 95% CI ‐151.53 to ‐6.47; Analysis 9.4). However, the majority of this apparent improvement was due to a reduction in the total sleep time. Participants slept for longer on placebo than with theophylline as measured by total sleep time (MD ‐76 minutes; 95% CI ‐100.88 to ‐51.12; Analysis 9.3). Re‐calculating the number of desaturations as an index, which takes in to account the shortening of sleep time, showed a DI of 51.5 and 49 per hour of sleep for placebo and treatment, respectively.

9.1. Analysis.

Comparison 9 Theophylline versus placebo, Outcome 1 Apnoea Hypopnea Index.

9.4. Analysis.

Comparison 9 Theophylline versus placebo, Outcome 4 Desaturations >4%.

9.3. Analysis.

Comparison 9 Theophylline versus placebo, Outcome 3 Total sleep time.

Acetazolamide versus placebo

One study involving 10 participants compared acetazolamide with placebo (Whyte 1988). There was a significant fall in AHI in participants taking acetazolamide and placebo (MD ‐24 apnoeas per hour; 95% CI ‐44.33 to ‐3.67; Analysis 10.1). However, there was no difference in the number of arousals from sleep through the night (Analysis 10.3) or in self‐scored data for daytime sleepiness. The three participants who appeared to benefit most were offered open label treatment and long‐term follow‐up. Only one participant continued to take the drug, the others had intolerable side effects.

10.1. Analysis.

Comparison 10 Acetozolamide versus placebo, Outcome 1 Apnoea Hypopnoea Index.

10.3. Analysis.

Comparison 10 Acetozolamide versus placebo, Outcome 3 Arousal Index.

Medroxyprogesterone versus placebo

One study involving 10 participants with OSA compared medroxyprogesterone with placebo (Cook 1989). There was no difference in the AHI (Analysis 11.1) or total sleep time (Analysis 11.2) between drug and placebo.

11.1. Analysis.

Comparison 11 Medroxyprogesterone versus placebo, Outcome 1 Apnoea Hypopnoea Index.

11.2. Analysis.

Comparison 11 Medroxyprogesterone versus placebo, Outcome 2 Total sleep time.

Naloxone versus placebo

One study involving 10 participants compared naloxone with placebo (Diamond 1982). There was no significant difference between naloxone and placebo in the number of apnoeic events (Analysis 12.1), duration of apnoeic events and mean per cent periods with arousal (Analysis 12.2).

12.1. Analysis.

Comparison 12 Naloxone versus placebo, Outcome 1 Number of apnoeic episodes.

12.2. Analysis.

Comparison 12 Naloxone versus placebo, Outcome 2 Percent of periods with arousals.

Naltrexone versus placebo

One crossover trial with 12 participants (Ferber 1993) reported significant reductions in AHI in the naltrexone treatment period compared with placebo (MD ‐8.50 events/hour; 95% CI ‐14.38 to ‐2.62; Analysis 13.1). However, there was a lower total sleep time (P < 0.04) and reduced deep sleep (P < 0.02).

13.1. Analysis.

Comparison 13 Naltrexone versus placebo, Outcome 1 AHI.

Doxapram versus placebo

One crossover trial with four participants compared doxapram with placebo (Suratt 1986). Although some differences between doxapram and placebo were noted on measures of AHI, no statistical tests of significance were carried out. No further assessment could be performed due to the inadequate presentation of average values.

Almitrine versus placebo

One crossover trial with nine participants compared almitrine with placebo (Mangin 1983). There were no significant differences in AI (Analysis 14.1) or HI (Analysis 14.2) or total sleep time (Analysis 14.3).

14.1. Analysis.

Comparison 14 Almitrine versus placebo, Outcome 1 Apnoea Index.

14.2. Analysis.

Comparison 14 Almitrine versus placebo, Outcome 2 Hypopnea Index.

14.3. Analysis.

Comparison 14 Almitrine versus placebo, Outcome 3 Total sleep time.

Vasoactive drugs

Clonidine versus placebo

One crossover trial on eight adult men compared clonidine and placebo (Issa 1992). No significant difference in AHI (Analysis 15.1) or minimum arterial saturation overnight (Analysis 15.2) was demonstrated.

15.1. Analysis.

Comparison 15 Clonidine versus placebo, Outcome 1 Apnoea Hypopnea Index.

15.2. Analysis.

Comparison 15 Clonidine versus placebo, Outcome 2 Min SaO2.

Mibefradil versus placebo

One parallel trial with 53 adults compared mibefradil with placebo (Heitmann 1998). There was no significant difference in AHI (Analysis 16.1), total sleep time (Analysis 16.2) or sleep efficiency.

16.1. Analysis.

Comparison 16 Mibefradil versus placebo, Outcome 1 Apnoea Hypopnea Index.

16.2. Analysis.

Comparison 16 Mibefradil versus placebo, Outcome 2 Total sleep time.

Topical drugs for the upper airway

Intranasal fluticasone propionate versus placebo

One crossover trial with 24 participants compared intranasal fluticasone propionate (FP) with placebo (Kiely 2004). Median values indicated that FP led to significantly lower AHI compared with placebo (23.3 versus 30.3; P < 0.05). No significant differences in subjective sleep quality, total sleep time and nocturnal oxygen saturation were apparent. Participants reported an increase in daytime alertness but no validated scale was used.

Nasal lubricant versus placebo

One crossover trial with 10 participants compared a nasal lubricant administered at the start of sleep and after 3.5 h with placebo (Jokic 1998). Jokic 1998 reported significantly lower AHI and arousals in favour of the lubricant (MD ‐10.00; 95% CI ‐14.02 to ‐5.98; Analysis 17.1) and (MD ‐8.00; 95% CI ‐11.88 to ‐4.12; Analysis 17.2), respectively. Since this was a single‐night study there was no information on symptomatic changes.

17.1. Analysis.

Comparison 17 Nasal lubricant versus placebo, Outcome 1 AHI.

17.2. Analysis.

Comparison 17 Nasal lubricant versus placebo, Outcome 2 Arousal index.

Xylometazoline nasal drops versus placebo

One crossover study on 10 participants compared xylometazoline with placebo (Clarenbach 2008) but no data were available for meta‐analysis. Despite a significant increase in mean nocturnal nasal conductance in the xylometazoline group, there was no significant difference in either AHI (xylometazoline 32.2 (SD 32.8) events/hr versus placebo 29.3 (SD 32.5) events/hr; P = NS) or subjective sleepiness (ESS 11.8 (SD 4.4) versus 10.5 (SD 3.8); P = NS).

Miscellaneous

Sabeluzole versus placebo

One crossover trial with 12 participants compared sabeluzole with placebo (Hedner 1996). There was no significant difference in total sleep time (Analysis 18.1). Difference scores for DI were presented and these were found to depend on the plasma concentration, which varied widely between participants.

18.1. Analysis.

Comparison 18 Sabeluzole versus placebo, Outcome 1 Total sleep time.

Physostigmine versus placebo

One crossover trial with 10 non‐obese participants compared physostigmine with placebo (Hedner 2003). AHI and SaO₂ were reported as a paired t‐test. A significant reduction in AHI was reported (MD ‐13.60; 95% CI ‐25.10 to ‐2.10; Analysis 19.1) and an increase in SaO₂ (MD 8.70; 95% CI ‐0.30 to 17.70; Analysis 19.3). However, only one participant of the 10 that were enrolled experienced a reduction in AHI to below 10 events/hour. Although paired data were not reported, there was a statistically significant difference in AI (MD ‐6.00; 95%CI ‐11.96 to ‐0.04; Analysis 19.2) and there was a lower total sleep time (TST) with physostigmine (74 minutes; 95% CI 33.9 to 114.9; Analysis 19.4).

19.1. Analysis.

Comparison 19 Physostigmine versus placebo, Outcome 1 Apnoea Hypopnea Index.

19.3. Analysis.

Comparison 19 Physostigmine versus placebo, Outcome 3 Min SaO2.

19.2. Analysis.

Comparison 19 Physostigmine versus placebo, Outcome 2 Apnoea Index.

19.4. Analysis.

Comparison 19 Physostigmine versus placebo, Outcome 4 Total sleep time.

Donepezil versus placebo

Two parallel trials with 45 participants compared donezepil with placebo (Moraes 2008; Sukys‐Claudino 2012). Moraes 2008 reported a significant reduction in AHI (donepezil 20 (SD 15) to 9.9 (SD 11.5) versus placebo 23.2 (SD 26.4) to 22.9 (SD 28.8); P = 0.035) and percentage of time spent with SpO₂ < 90% (donepezil 13.4 (SD 17.4) to 3.7 (SD 4.8) versus placebo 8.5 (SD 14.7) to 11 (SD 20.1); P = 0.017) in 23 patients with mild to moderate Alzheimer's disease after three months of treatment. The effect of donezepil on OSA in patients without Alzheimer's disease was examined by Sukys‐Claudino 2012 and reported to be statistically significant but small for most patients. There was a mean difference in AHI of ‐9.4 (SD 17.2; P = 0.03) and a clinically significant change in ESS of ‐2.9 (SD 2.9; P = 0.04) after one month of treatment. There was also a reduction in sleep efficiency with the active treatment.

Fenofibrate versus placebo

One parallel trial with 34 participants compared fenofibrate with placebo (Bruckert 2010). There was no significant change in AHI with fenofibrate (MD 14%; 95% CI ‐47% to 40%), however there was a significant reduction in percentage of time spent with SpO₂ < 90% (P = 0.007). The study had several key limitations including short study duration, concomitant hypnotic treatment (35%) and lack of correction for multiplicity testing. The data were not reported in a way that facilitated meta‐analysis.

Eszopiclone versus placebo

One crossover trial with 17 participants compared eszopiclone and placebo (Eckert 2011). The AHI was lowered in people on eszopiclone (MD ‐7.00; 95% CI ‐13.86 to ‐0.14; Analysis 21.1). Compared with placebo, eszopiclone increased arousal threshold (nadir oesophageal or epiglottic pressure preceding arousal) from ‐14 cm H₂O to ‐18 cm H₂O (P ≤ 0.01). A more pronounced AHI drop (45%) was found in eight patients with low arousal threshold (25 (SD 6) events/hr versus 14 (SD 4) events/hr; P ≤ 0.01). However, reductions in AHI to achieve conventional standards of treatment efficacy were not seen in the majority of patients. The study was limited by short duration and recruited a specific cohort of OSA patients with nadir overnight SaO₂ > 70%.

21.1. Analysis.

Comparison 21 Eszopiclone versus placebo, Outcome 1 AHI.

Tolerability

Few data were presented on the long‐term tolerability of the compounds used.

Protryptilline: several of the participants who were given protriptyline had side effects including a dry mouth, urinary symptoms, impotence and visual disturbance. The dose of protryptiline was reduced in two participants in the study where this was an option (Whyte 1988). One study offered a follow‐up phase on protriptyline. Only three out of five participants agreed to participate and one developed urinary retention when the dose of protryptilline was increased (Brownell 1982).

Acetzolamide: acetazolamide produced paraesthesia in eight out of 10 participants who took it in the short term (Whyte 1988). Two out of three participants offered acetazolamide in the long term could not tolerate it, due to paraesthesia in one and enuresis in the other. One patient suffered heart burn but recovered with antacids. One patient in the placebo group suffered from vertigo, resolving spontaneously. One patient withdrew from the study due to acute vestibulopathy.

Paroxetine: Kraiczi 1999 reported that during active treatment side effects included fatigue (n = 2), ejaculation disturbances (n = 3), decreased libido (n = 2), nervousness (n = 1), constipation (n = 1), headache (n = 1), dizziness (n = 1), somnolence (n = 1), infectious pneumonia (n = 1), Lyme disease (n = 1), sweating (n = 1). During placebo treatment side effects reported were dizziness (n = 1), fatigue and somnolence (n = 1) and dryness of mouth (n = 1).

Ondansetron and fluoxetine: there were no significant differences in the incidence of adverse events in the ondansetron and fluoxetine and placebo groups, with most of the reported events being gastrointestinal, including constipation (Prasad 2010).

Mirtazapine: mirtazapine was well tolerated in Carley 2007 with no reported side effects, however in two RCTs by Marshall 2008 enrolment was stopped midway through the first study due to safety concerns over lethargy affecting driving ability. Fifteen patients did not complete the second trial, seven due to unacceptable lethargy. The drug was also associated with significant weight gain.

Theophylline: theophylline caused side effects in five out of 12 participants given it in the short term, including nausea, dyspepsia, headache and depression (Mulloy 1992). One participant withdrew from the study because of these effects. Hein 2000 reported that participants reported mild insomnia (n = 3), tremor (n = 2) and nausea (n = 1). No long‐term data were presented.

Mibefradil: transient side effects were reported in Heitmann 1998 but the differences were non‐significant.

Fenofibrate: adverse events were reported by three patients in the fenofibrate group (bloating, rash, allergic rhinitis, diabetes mellitus type 2, cystitis, metabolic syndrome) and one participant in the placebo group (lumbar pain), none thought to be related to the study drug (Bruckert 2010).

Donezepil: three patients taking donepezil reported mild, transitory side effects including nausea and headache (Moraes 2008), and the most common side effects reported by Sukys‐Claudino 2012 were dizziness, nausea, vivid dreams and nightmares with no statistical difference between the studied groups. Two patients dropped out from the donepezil group, one with an acute prostatic condition and one with diverticulitis.

Discussion

Summary of main results

We identified 30 studies assessing the effects of 25 drugs in the treatment of sleep apnoea. The majority of these studies involved only a small number of participants and for many the design has been poorly reported. For most of the drugs there have been only single trials, indeed we were only able to pool data in the case of five drugs and only for select outcomes. Ten of the drugs trialled were shown to reduce the AHI compared to placebo (Whyte 1988; Ferber 1993; Jokic 1998; Kraiczi 1999; Kiely 2004; Carley 2007; Moraes 2008; Prasad 2010; Eckert 2011; Sukys‐Claudino 2012). The majority of the trials were only for a single night and symptomatic benefit could not therefore be demonstrated. Four longer studies reported a subjective decrease in daytime somnolence or an increase in alertness but this was not corroborated by objective tests of sleep propensity in any of the studies (Brownell 1982; Stepanski 1988; Kiely 2004; Sukys‐Claudino 2012), and reduction in AHI was not always an indicator of improved sleep quality. Ferber 1993 demonstrated a significantly lower AHI with naltrexone but also significantly lower total sleep. The goal of OSA treatment must be to reduce daytime sleepiness (somnolence) but few trials reported this patient important outcome.

Among the drugs administered systemically, paroxetine (Kraiczi 1999), physostigmine (Hedner 2003) and donepezil (Moraes 2008; Sukys‐Claudino 2012) reduced the AHI to an extent. Paroxetine had no impact on daytime symptoms. Its effect was limited to reducing the AHI during non‐REM sleep, and the overall fall in AHI was small at just 17%. Physostigmine reduced the AHI by 24% compared to placebo but the effect was more marked in REM where the reduction was 44%. In addition, the reduction in AHI was inversely related to BMI. It is not known whether physostigmine would be as effective if given orally, how well it might be tolerated, and whether it can reduce daytime sleepiness. Donepezil in patients with and without dementia (Moraes 2008; Sukys‐Claudino 2012) showed a modest but statistically significant decrease in AHI and improvement in nocturnal SaO₂. Moraes 2008 reports a significant reduction in non‐rapid eye movement (NREM) AHI and an increase in REM sleep duration. The improvement in AHI may be due to the improved co‐ordination between the diaphragm and the upper airway muscles, caused by increased hypoglossal to phrenic interval and genioglossal activity. Sukys‐Claudino 2012 reported a decrease in AHI at the expense of reduced sleep efficiency and further trials are needed. Carley 2007 reported reduction in AHI by around 45% for a 15 mg dose of mirtazapine compared to placebo. Two more recent multicentre trials (Marshall 2008) failed to reproduce these results and moreover use of mirtazapine was associated with significant weight gain and sleepiness. Compared to Carley 2007, Marshall 2008 recruited fewer women and obese patients, which may have altered the drug's effect on OSA. Another striking feature in this later trial was an increase in AHI in patients taking 15 mg and 30 mg of mirtazapine, which may suggest a dose specific effect of mirtazapine on OSA, however worsening of OSA was not replicated in either arm of 15 mg of mirtazapine in the second trial. Protriptyline remains the most frequently studied drug and was not shown to improve any measure of respiratory disturbance during sleep (Brownell 1982; Stepanski 1988; Whyte 1988). However, in two out of three studies participants reported a decrease in daytime somnolence. Since sleep parameters were not improved it may be that this was the result of the non‐specific alerting effect of the drug rather than a direct effect on OSA.

Acetazolamide was shown to reduce the number of respiratory events during sleep but this did not improve sleep quality or reduce the subjective impression of sleepiness in the day (Whyte 1988). Theophylline and aminophylline are related compounds and similar results were seen in the trials using these two drugs. Aminophylline decreased the number of hypopnoeas but also reduced the quality of sleep (Espinoza 1987). Theophylline was shown to reduce the number of arterial oxygen desaturations but reduced the total sleep time by more than an hour (Mulloy 1992; Hein 2000). Clonidine improved overnight oxygenation but it is not clear that this would have any clinical relevance (Issa 1992). No statistically significant differences were found between the outcomes for the other drugs studied and placebo.

Use of eszopiclone (Eckert 2011) resulted in a significant increase in arousal threshold and decrease in AHI by 23% (P < 0.005). The study is too small (N = 17) and was only of one‐night duration, and therefore larger scale clinical trials are needed to evaluate these results further.

From the results available, topical intranasal fluticasone propionate has been shown to reduce apnoea for patients with mild OSA and co‐existing rhinitis, although more information is needed regarding the impact on excessive daytime sleepiness and long‐term benefit (Kiely 2004). If a suitable compound can be developed for long‐term use further investigation of nasal lubricants is justified. At present there is insufficient evidence to recommend any systemic drug therapy for OSA. Nevertheless the available results do suggest that further research is warranted. It may be, for example, that patients with OSA almost exclusively in NREM sleep might get more benefit from paroxetine than an otherwise unselected population (Kraiczi 1999). Donepezil was well tolerated orally and reduced AHI and ESS, but again the study was small. Combination therapies may also be of value to control OSA in both REM and NREM sleep.

Overall completeness and applicability of evidence

Some drugs have been used extensively in clinical practice, however two trials on the most commonly used drug, theophylline (Peter 1987; Di Martino 1999), were excluded from this review as neither were randomised, placebo controlled studies. Medroxyprogesterone and clonidine have also been proposed as possible treatments for OSA but there are few reports of their use. More recent work, in particular with animal models of OSA, has recognised the importance of the tone in the upper airway musculature and the role of serotonergic neurons in its control. Other areas of interest, which have more recently been investigated, are the effects of various vasoactive compounds on OSA, the role of topical therapy, use of oral anticholinergic drugs and manipulation of arousal threshold.

From our review some of the new approaches justify further work, but it is possible that those most likely to respond to drug therapy are restricted to selected subgroups of people with OSA. The response may vary according to the relationship between sleep position, sleep stage and severity of sleep disruption (Jokic 1998).

We have not reviewed the effects of medication used to reduce daytime sleepiness in OSA. Modafanil is one such agent that has been trialled in addition to CPAP primarily (Kingshott 2001; Dinges 2003). Concerns that it might reduce usage of CPAP have been expressed (SIGN 2003).

Quality of the evidence

There was much heterogeneity in the types of drugs used in the trials in this review and many of the trials were on just a few patients, with short trial duration and with indiscernible methodological quality.

Although the inclusion criteria in some studies may have allowed the recruitment of some patients with central sleep apnoea, due to the use of the oxygen desaturation index rather than the apnoea hypopnoea index (AHI), the reported number of AHI events in the studies justified the inclusion of these studies in this review. However should new data become available, sensitivity analyses involving the removal of these studies will be performed to see whether a slight variation in the entry criteria accounts for any differential response.

Potential biases in the review process

We did not contact authors to enquire about further details to allow us to assess the risk of bias. We felt that this wouldn't alter our judgements of how the risk of bias might effect the treatment effects in these studies. We tried to minimise bias by screening references in duplicate, double data extracting, and entering the data into RevMan together as recommended in the Cochrane Handbook for Systematic Reviews of Interventions.

Agreements and disagreements with other studies or reviews

Larger, longer‐term studies would offer a clearer picture.

Authors' conclusions

Implications for practice.

There is currently insufficient evidence to recommend any systemic pharmacological treatment for OSA. From the data available it seems likely that unlike CPAP, which is usually effective irrespective of the cause of OSA, drug therapy will need to be targeted according to features such as presence or absence of obesity and the predominance of OSA in a particular sleep stage. Ultimately, a combination of strategies targeting different pathophysiological pathways may be required.

Implications for research.

Evidence from a small trial on topical fluticasone in patients with co‐existent rhinitis and OSA, and trials on four drugs that reduced AHI, paroxetine, physostigmine, acetazolamide and eszopiclone, may justify further work. Donepezil is the most promising proposition at present.

Future trials of drug therapy for OSA should be of adequate size and duration, and should have valid, patient centred outcomes. Many of the studies used multiple outcome variables and it would be valuable to reach a consensus on mandatory outcome variables to measure and report for future studies. In addition to overnight monitoring to confirm a direct effect on the frequency of respiratory events, they should report changes in daytime sleepiness, measured subjectively (ESS) and ideally also objectively. Diagnostic criteria should be explicitly reported, and if any participants are recruited with suspected central apnoea this should also be reported, ideally with separate data for people with each diagnosis. Furthermore, the long‐term tolerability of any drug used should be investigated.

What's new

Date	Event	Description
5 July 2012	New citation required and conclusions have changed	Nine studies added (Bruckert 2010; Carley 2007; Clarenbach 2008; Eckert 2011; Marshall 2008; Moraes 2008; Prasad 2010; Sukys‐Claudino 2012). Marshall 2008 ‐ 2 randomised controlled trials (RCTs) were reported in this paper, Marshall 2008 (1) and Marshall 2008 (2). Four studies inappropriately included in the previous review were withdrawn from this update as the inclusion criteria were not met (Atkinson 1985; Grote 1994; Grote 1995; Grote 1995a). Background updated. Summary of characteristics of included studies table added. Promising results from the preliminary mirtazapine study failed to be reproduced in the two more recent multicentre trials and, moreover, the use of mirtazapine was associated with significant weight gain and sleepiness
5 July 2012	New search has been performed	New literature search run.

History

Protocol first published: Issue 2, 2000
 Review first published: Issue 3, 2001

Date	Event	Description
23 July 2008	Amended	Converted to new review format.
2 February 2006	New citation required and conclusions have changed	Substantive amendment 17 new studies were identified which met the inclusion criteria.These studies introduced new drug treatments in to the review and as such could not be combined with any previous data. The assessment of a single study on mirtazepine has yielded some significant differences on one of the outcomes considered (AHI). This study was conducted over one week and justifies further work in larger and more long‐term clinical trials.

Appendices

Appendix 1. Sources and search methods for the Cochrane Airways Group Specialised Register (CAGR)

Electronic searches: core databases

Database	Frequency of search
CENTRAL (the Cochrane Library)	Monthly
MEDLINE (Ovid)	Weekly
EMBASE (Ovid)	Weekly
PsycINFO (Ovid)	Monthly
CINAHL (EBSCO)	Monthly
AMED (EBSCO)	Monthly

Handsearches: core respiratory conference abstracts

Conference	Years searched
American Academy of Allergy, Asthma and Immunology (AAAAI)	2001 onwards
American Thoracic Society (ATS)	2001 onwards
Asia Pacific Society of Respirology (APSR)	2004 onwards
British Thoracic Society Winter Meeting (BTS)	2000 onwards
Chest Meeting	2003 onwards
European Respiratory Society (ERS)	1992, 1994, 2000 onwards
International Primary Care Respiratory Group Congress (IPCRG)	2002 onwards
Thoracic Society of Australia and New Zealand (TSANZ)	1999 onwards

Sleep apnoea search

1. exp Sleep Apnea Syndromes/

2. (sleep$ adj3 (apnea$ or apnoea$)).mp.

3. (hypopnea$ or hypopnoea$).mp.

4. OSA.mp.

5. SHS.mp.

6. OSAHS.mp.

7. or/1‐6

Filter to identify RCTs

1. exp "clinical trial [publication type]"/

2. (randomized or randomised).ab,ti.

3. placebo.ab,ti.

4. dt.fs.

5. randomly.ab,ti.

6. trial.ab,ti.

7. groups.ab,ti.

8. or/1‐7

9. Animals/

10. Humans/

11. 9 not (9 and 10)

12. 8 not 11

The MEDLINE strategy and RCT filter are adapted to identify trials in other electronic databases

Data and analyses

Comparison 1. Protriptyline versus placebo.

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Apnoea Hypopnea Index	1		Mean Difference (Fixed, 95% CI)	Totals not selected
2 Apnoea Index	2	13	Mean Difference (Fixed, 95% CI)	‐5.81 [‐21.32, 9.70]
3 Hypopnea Index	1		Mean Difference (Fixed, 95% CI)	Totals not selected
4 Min SaO2	1		Mean Difference (Fixed, 95% CI)	Totals not selected
5 Total sleep time	2	13	Mean Difference (Fixed, 95% CI)	2.90 [‐40.38, 46.18]
6 Desaturation Index	1		Mean Difference (Fixed, 95% CI)	Totals not selected
7 Arousal Index	2	15	Mean Difference (Fixed, 95% CI)	‐9.62 [‐22.03, 2.79]
8 Subjective reduction in daytime sleepiness	2	26	Peto Odds Ratio (Peto, Fixed, 95% CI)	17.15 [3.61, 81.43]

1.8. Analysis.

Comparison 1 Protriptyline versus placebo, Outcome 8 Subjective reduction in daytime sleepiness.

Comparison 2. Paroxetine versus placebo.

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Apnoea Hypopnoea Index	1		Mean Difference (Fixed, 95% CI)	Totals not selected
2 Apnoea Index	1		Mean Difference (Fixed, 95% CI)	Totals not selected
3 Hypopnea Index	1		Mean Difference (Fixed, 95% CI)	Totals not selected
4 CPRS symptoms	1		Mean Difference (Fixed, 95% CI)	Totals not selected
5 Sleepiness	1		Mean Difference (Fixed, 95% CI)	Totals not selected
5.1 Morning sleepiness	1		Mean Difference (Fixed, 95% CI)	0.0 [0.0, 0.0]
5.2 Daytime sleepiness	1		Mean Difference (Fixed, 95% CI)	0.0 [0.0, 0.0]
6 Adverse effects	1		Mean Difference (Fixed, 95% CI)	Totals not selected
6.1 morning headaches	1		Mean Difference (Fixed, 95% CI)	0.0 [0.0, 0.0]
6.2 Difficulty in concentration	1		Mean Difference (Fixed, 95% CI)	0.0 [0.0, 0.0]
6.3 Memory complaints	1		Mean Difference (Fixed, 95% CI)	0.0 [0.0, 0.0]
6.4 Low mood	1		Mean Difference (Fixed, 95% CI)	0.0 [0.0, 0.0]

2.2. Analysis.

Comparison 2 Paroxetine versus placebo, Outcome 2 Apnoea Index.

2.3. Analysis.

Comparison 2 Paroxetine versus placebo, Outcome 3 Hypopnea Index.

2.5. Analysis.

Comparison 2 Paroxetine versus placebo, Outcome 5 Sleepiness.

Comparison 3. Mirtazapine versus placebo.

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Apnoea Hypopnea Index	2		Mean Difference (Fixed, 95% CI)	Subtotals only
1.1 high dose (15 mg) Parallel studies	1	39	Mean Difference (Fixed, 95% CI)	‐1.3 [‐13.57, 10.97]
1.2 high dose (15 mg) Crossovers	2	32	Mean Difference (Fixed, 95% CI)	‐4.56 [‐10.91, 1.79]
1.3 Low dose (4.5 mg) crossover	1	12	Mean Difference (Fixed, 95% CI)	‐8.8 [‐14.68, ‐2.92]
2 ESS	1		Mean Difference (IV, Fixed, 95% CI)	Totals not selected
2.1 High dose (15 mg)	1		Mean Difference (IV, Fixed, 95% CI)	0.0 [0.0, 0.0]
3 Arousals	1	24	Events/hr (Fixed, 95% CI)	‐13.0 [‐23.78, ‐2.22]
3.1 Low dose (4.5 mg)	1	24	Events/hr (Fixed, 95% CI)	‐13.0 [‐23.78, ‐2.22]

3.2. Analysis.

Comparison 3 Mirtazapine versus placebo, Outcome 2 ESS.

3.3. Analysis.

Comparison 3 Mirtazapine versus placebo, Outcome 3 Arousals.

Comparison 4. High dose mirtazapine + CD0012 versus placebo.

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Apnoea hypopnea index	1		Mean Difference (IV, Fixed, 95% CI)	Totals not selected
2 ESS	1		Mean Difference (IV, Fixed, 95% CI)	Totals not selected

4.1. Analysis.

Comparison 4 High dose mirtazapine + CD0012 versus placebo, Outcome 1 Apnoea hypopnea index.

4.2. Analysis.

Comparison 4 High dose mirtazapine + CD0012 versus placebo, Outcome 2 ESS.

Comparison 5. Ondansetron versus placebo.

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Apnoea Hypopnea Index	1		Mean Difference (Fixed, 95% CI)	Totals not selected
2 Apnoea Index	1		Mean Difference (Fixed, 95% CI)	Totals not selected
3 SaO2 (>4% dips)	1		Mean Difference (Fixed, 95% CI)	Totals not selected

5.3. Analysis.

Comparison 5 Ondansetron versus placebo, Outcome 3 SaO2 (>4% dips).

Comparison 6. Buspirone versus placebo.

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Apnoea Index	1		Mean Difference (Fixed, 95% CI)	Totals not selected
2 Total sleep time	1		Mean Difference (Fixed, 95% CI)	Totals not selected
3 MinSaO2	1		Mean Difference (Fixed, 95% CI)	Totals not selected

Comparison 7. Salmeterol versus placebo.

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 AHI	1		Mean Difference (Fixed, 95% CI)	Totals not selected
2 Minimum SaO2	1		Mean Difference (Fixed, 95% CI)	Totals not selected

Comparison 8. Aminophylline versus placebo.

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Apnoea Index	1		Mean Difference (Fixed, 95% CI)	Totals not selected
2 Hypopnoea index	1		Mean Difference (Fixed, 95% CI)	Totals not selected
3 Min SaO2 in REM	1		Mean Difference (Fixed, 95% CI)	Totals not selected
4 Sleep efficiency	1		Mean Difference (Fixed, 95% CI)	Totals not selected

Comparison 9. Theophylline versus placebo.

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Apnoea Hypopnea Index	2	23	Mean Difference (Fixed, 95% CI)	‐2.76 [‐8.57, 3.05]
2 Min SaO2	1		Mean Difference (Fixed, 95% CI)	Totals not selected
3 Total sleep time	1		Mean Difference (Fixed, 95% CI)	Totals not selected
4 Desaturations >4%	1		Mean Difference (Fixed, 95% CI)	Totals not selected
5 Self‐assessed sleep latency	1		Mean Difference (Fixed, 95% CI)	Totals not selected
6 Reported number of awakenings	1		Mean Difference (Fixed, 95% CI)	Totals not selected
7 Subjective sleep quality	1		Mean Difference (Fixed, 95% CI)	Totals not selected

9.2. Analysis.

Comparison 9 Theophylline versus placebo, Outcome 2 Min SaO2.

9.5. Analysis.

Comparison 9 Theophylline versus placebo, Outcome 5 Self‐assessed sleep latency.

9.6. Analysis.

Comparison 9 Theophylline versus placebo, Outcome 6 Reported number of awakenings.

9.7. Analysis.

Comparison 9 Theophylline versus placebo, Outcome 7 Subjective sleep quality.

Comparison 10. Acetozolamide versus placebo.

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Apnoea Hypopnoea Index	1		Mean Difference (Fixed, 95% CI)	Totals not selected
2 Desaturation Index	1		Mean Difference (Fixed, 95% CI)	Totals not selected
3 Arousal Index	1		Mean Difference (Fixed, 95% CI)	Totals not selected

10.2. Analysis.

Comparison 10 Acetozolamide versus placebo, Outcome 2 Desaturation Index.

Comparison 11. Medroxyprogesterone versus placebo.

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Apnoea Hypopnoea Index	1		Mean Difference (Fixed, 95% CI)	Totals not selected
2 Total sleep time	1		Mean Difference (Fixed, 95% CI)	Totals not selected

Comparison 12. Naloxone versus placebo.

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Number of apnoeic episodes	1		Mean Difference (Fixed, 95% CI)	Totals not selected
2 Percent of periods with arousals	1		Mean Difference (Fixed, 95% CI)	Totals not selected

Comparison 13. Naltrexone versus placebo.

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 AHI	1		events/hr (Fixed, 95% CI)	Totals not selected

Comparison 14. Almitrine versus placebo.

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Apnoea Index	1		Mean Difference (Fixed, 95% CI)	Totals not selected
2 Hypopnea Index	1		Mean Difference (Fixed, 95% CI)	Totals not selected
3 Total sleep time	1		Mean Difference (Fixed, 95% CI)	Totals not selected

Comparison 15. Clonidine versus placebo.

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Apnoea Hypopnea Index	1		Mean Difference (Fixed, 95% CI)	Totals not selected
2 Min SaO2	1		Mean Difference (Fixed, 95% CI)	Totals not selected

Comparison 16. Mibefradil versus placebo.

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Apnoea Hypopnea Index	1		Mean Difference (IV, Fixed, 95% CI)	Totals not selected
2 Total sleep time	1		Mean Difference (IV, Fixed, 95% CI)	Totals not selected
3 Mean change in Apnoea Hypopnea Index	1		Mean Difference (IV, Fixed, 95% CI)	Totals not selected

16.3. Analysis.

Comparison 16 Mibefradil versus placebo, Outcome 3 Mean change in Apnoea Hypopnea Index.

Comparison 17. Nasal lubricant versus placebo.

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 AHI	1		Mean Difference (Fixed, 95% CI)	Totals not selected
2 Arousal index	1		Events/hr (Fixed, 95% CI)	Totals not selected

Comparison 18. Sabeluzole versus placebo.

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Total sleep time	1		Minutes (Fixed, 95% CI)	Totals not selected

Comparison 19. Physostigmine versus placebo.

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Apnoea Hypopnea Index	1		Mean Difference (Fixed, 95% CI)	Totals not selected
2 Apnoea Index	1		Mean Difference (Fixed, 95% CI)	Totals not selected
3 Min SaO2	1		Mean Difference (Fixed, 95% CI)	Totals not selected
4 Total sleep time	1		Mean Difference (Fixed, 95% CI)	Totals not selected
5 Time to sleep onset	1		Mean Difference (Fixed, 95% CI)	Totals not selected

19.5. Analysis.

Comparison 19 Physostigmine versus placebo, Outcome 5 Time to sleep onset.

Comparison 20. Donezepil versus placebo.

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 Apnoea hypopnea index	2	44	Mean Difference (IV, Fixed, 95% CI)	‐5.07 [‐17.10, 6.96]
2 ESS	1		Mean Difference (IV, Fixed, 95% CI)	Totals not selected
3 Desaturation index	1		Mean Difference (IV, Fixed, 95% CI)	Totals not selected
4 Min SaO2	2	44	Mean Difference (IV, Fixed, 95% CI)	2.79 [‐1.58, 7.15]
5 Total sleep time	2	44	Mean Difference (IV, Fixed, 95% CI)	‐20.14 [‐62.17, 21.90]

20.1. Analysis.

Comparison 20 Donezepil versus placebo, Outcome 1 Apnoea hypopnea index.

20.2. Analysis.

Comparison 20 Donezepil versus placebo, Outcome 2 ESS.

20.3. Analysis.

Comparison 20 Donezepil versus placebo, Outcome 3 Desaturation index.

20.4. Analysis.

Comparison 20 Donezepil versus placebo, Outcome 4 Min SaO2.

20.5. Analysis.

Comparison 20 Donezepil versus placebo, Outcome 5 Total sleep time.

Comparison 21. Eszopiclone versus placebo.

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1 AHI	1		Mean Difference (Fixed, 95% CI)	Totals not selected

Characteristics of studies

Characteristics of included studies [ordered by study ID]

Brownell 1982.

Methods	Randomised, double blind, crossover study. Method of randomisation not reported. Statistical test: Wilcoxon
Participants	5 adult males. Mean age: 45.8 years; AI: 64.4. Participants referred to clinic due to disabling daytime sleepiness. OSA confirmed by PSG
Interventions	Protriptylline (20 mg nocte (every night)) versus placebo. Study duration: 2 weeks (2 week washout)
Outcomes	AI; subjective response
Notes	Jadad score: 3 Funding source: Merck Sharpe and Dohme supplied drugs and placebo.
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	stated randomised
Allocation concealment (selection bias)	Unclear risk	Information not available
Blinding of participants and personnel (performance bias) All outcomes	Unclear risk	stated double blind
Blinding of outcome assessment (detection bias) All outcomes	Unclear risk	stated double blind

Bruckert 2010.

Methods	Randomised, double blind, parallel study, Statistical test: for normally distributed data ANCOVA and for non‐normally distributed data‐Wilcoxon test
Participants	34 adults (12 female, 22 male); age 55.9 ± 5.4 (F); 55.2 ± 6.5 (placebo), median AHI 21, ESS (F) 8.5; (placebo) 7.5; fasting TAG > 2 < 7.5 mmol/L. OSA confirmed on PSG. Exclusion criteria: need for CPAP, uncontrolled diabetes mellitus and other co‐morbidity, pregnancy, drug abuse) 35% patients had concomitant hypnotic treatment
Interventions	Fenofibrate 145 mg (n=18) versus placebo (n=16); 1‐5 weeks run in period, 4 weeks duration
Outcomes	AHI, percentage of time with SpO2 < 90%, attention/vigilance test, ESS, fasting and postprandial triglycerides
Notes	Only 14 patients ESS > 10, median ESS < 11 (patients not sleepy), no power calculation, small sample size, short duration. Concomitant hypnotic treatment, lack of correction for multiplicity of statistical testing Funding source: "This study was supported by a grant from Laboratories Fournier SA, Daix, France."
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	"Randomized" "Laboratoires Fournier SA, Chenove, France was responsible for the preparation of the randomisation schedule for treatment allocation and for supply of the study medication."
Allocation concealment (selection bias)	Unclear risk	See above, unclear
Blinding of participants and personnel (performance bias) All outcomes	Unclear risk	"double‐blind, placebo‐controlled" Comment: not described
Blinding of outcome assessment (detection bias) All outcomes	Unclear risk	"double‐blind, placebo‐controlled" Comment: not described
Incomplete outcome data (attrition bias) All outcomes	Low risk	"All randomised and treated subjects were included in the analysis". All completed.
Selective reporting (reporting bias)	Low risk	Appeared to report the outcomes recorded
Other bias	Unclear risk	"Secondly, concomitant hypnotic use is likely to represent a source of bias. Three subjects took selective serotonin reuptake inhibitors (two fluoxetine, one escitalopram), which are known to influence airway tone, and have been reported to produce statistically significant (albeit modest) improvements in the AHI in OSA patients. Conversely, other hypnotic treatments (such as desloratadine, hydroxyzine, alprazolam, oxazepam and tramadol) exacerbate sleepiness. Anecdotally, venlafaxine has been shown to produce pulmonary obstruction and exacerbate respiratory symptoms in asthma patients, which may be of relevance in this setting. However, these treatments remained unchanged during the study."

Carley 2007.

Methods	Randomised, double blind, 3‐way crossover trial. Statistical assessment: ANOVA with repeated measures
Participants	12 adults, 7 males (mean age 39 years, AHI 22 ± 11.2 events per hour sleep) 5 females (mean age 43 years, AHI 24 ± 22.8 events per hour sleep). Treatment naive OSA. OSA confirmed on PSG. Exclusion criteria: history of substance abuse, other comorbidity, BP > 150/90 (treated with antihypertensive treatment other than diuretics), pregnancy, shift work
Interventions	mirtazapine at doses of 4.5 mg and 15 mg versus placebo daily for 7 days, no washout period
Outcomes	AHI, arousals
Notes	Funding source: Assume Cypress bioscience, see editor's note below. Editor's note: The reader needs to be aware of the following information. On June 27, 2006, Cypress Bioscience Inc. announced “that the results of recently completed Phase IIa trials do not support continuing a development program evaluating combinations of mirtazapine with another approved drug as potential pharmaceutical treatments for obstructive sleep apnea (OSA). Cypress and Organon, the human healthcare business unit of Akzo Nobel, had each independently conducted Phase IIa trials that served as the basis for today’s announcement. A previous independently conducted small preliminary investigator sponsored pilot trial found that mirtazapine was able to reduce the number of abnormal respiratory events over the course of the night by roughly fifty percent. However, those data were not replicated in the recently completed phase IIa trials.”
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	"randomised" comment: not described
Allocation concealment (selection bias)	Unclear risk	not described
Blinding of participants and personnel (performance bias) All outcomes	Unclear risk	"double blind, placebo‐controlled"
Blinding of outcome assessment (detection bias) All outcomes	Low risk	"we sought to minimize variability of respiratory‐event scoring by using a single polysomnographer blinded as to subject and treatment information. Additionally, after an interval of longer than 6 months, we had 5 randomly chosen records recorded by the initial polysomnographer and an additional polysomnographer."
Incomplete outcome data (attrition bias) All outcomes	Low risk	All subjects completed study
Selective reporting (reporting bias)	Low risk	Appeared to report the outcomes recorded

Clarenbach 2008.

Methods	Randomised, double blind, crossover trial. Statistical test: paired t‐test.
Participants	12 adults (10 males), mean age 49 ± 11years, AHI 32.6 ± 24.5 events per hour sleep, ESS 11.8 ± 4.5, patients with chronic nasal congestion; OSA confirmed on PSG. Exclusion criteria: nasal surgery in the last six months, other sleep disorders, use of other nasal decongestant or steroids, other significant co‐morbidity
Interventions	xylomethazoline 0.1% (3 drops of 0.15 mg) versus placebo (normal saline) for one week and one week washout period
Outcomes	AHI, mean nocturnal oxygen saturations, sleep efficiency, ESS, nasal conductance
Notes	Funding source: "This research is supported by Swiss National Science Foundation, Hartmann‐Műller Stiftung, Zurich, and Lung Ligue of Zurich, Switzerland."
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	"randomised" comment: not described
Allocation concealment (selection bias)	Unclear risk	comment: not described
Blinding of participants and personnel (performance bias) All outcomes	Low risk	double blind, placebo controlled "The code of the medication was broken only after completion of data analysis."
Blinding of outcome assessment (detection bias) All outcomes	Low risk	"The code of the medication was broken only after completion of data analysis."
Incomplete outcome data (attrition bias) All outcomes	Low risk	All subjects completed study
Selective reporting (reporting bias)	Low risk	Appeared to report the outcomes recorded

Cook 1989.

Methods	Randomised, double blind, crossover trial. Method of randomisation not reported. Statistical test: Student's paired t test
Participants	10 adult men. Age range: 31 to 67 years; AHI: 77.3 (taken from control group). BMI: 36 Participants referred to clinic due to excessive daytime sleepiness and suspected OSA. OSA confirmed by PSG. Inclusion criteria: AHI: ≥30/hour
Interventions	medroxyprogesterone 50 mg tds (three times a day) 1 week (3 week washout)
Outcomes	AHI; TST; apnoea duration; levels of medroxyprogesterone and plasma testosterone; disordered breathing time as % sleep
Notes	Jadad score: 2 Funding source: "This study was supported by the American Lung Associations of South Carolina, MUSC institutional grant No. CRO2, GCRC grant No. M01RR01070 ad the Upjohn company"
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	stated randomised
Allocation concealment (selection bias)	Unclear risk	Information not available
Blinding of participants and personnel (performance bias) All outcomes	Unclear risk	stated double blind
Blinding of outcome assessment (detection bias) All outcomes	Unclear risk	stated double blind

Diamond 1982.

Methods	Double blind crossover study. Randomisation and method of statistical analysis used was not reported.
Participants	4 participants with OSA. No baseline data/entry criteria were available.
Interventions	Intravenous naloxone (2 mg) at half‐hourly intervals during prolonged sleep versus placebo Study duration: unclear
Outcomes	No. apnoeic episodes; apnoea duration; average nadir value; % periods with arousals
Notes	Unpublished conference abstract Funding source: Do not have a copy of paper for the update so unknown
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	stated randomised
Allocation concealment (selection bias)	Unclear risk	Information not available
Blinding of participants and personnel (performance bias) All outcomes	Unclear risk	stated double blind
Blinding of outcome assessment (detection bias) All outcomes	Unclear risk	stated double blind

Eckert 2011.

Methods	Randomised, double blind, crossover trial. Statistical test: for normally distributed data ‐ paired Student's t test, for non‐normally distributed data‐Mann‐Whitney U test.
Participants	17 adults (10 males), mean age 45 ± 4 years (19‐62), BMI 33 ± 2 kg/m2, AHI 31 ± 5 events per hour sleep, nadir SaO2 > 70%. OSA was diagnosed on PSG. Exclusion criteria: severe OSA defined as SpO2 < 70% or AHI > 60/hr, high arousal threshold (‐25 cmH2O to ‐63 cmH2O)
Interventions	Eszopiclone 3 mg versus placebo for 1 night.
Outcomes	AHI, total sleep time, sleep onset latency, mean overnight oxygen saturations, arousal threshold
Notes	Funding source: "This study was supported by the National Institutes of Health (NIH) [grant number P01 HL095491‐ 01 A1] and an unrestricted investigator‐initiated research grant from Sepracor Pharmaceuticals. Other support includes the NIH [grant numbers HL73146 R01 HL085188‐01A2, R01 HL090897‐01A2, K24 HL 093218‐01 A1]. D.J.E. is supported by an Overseas Based Biomedical (CJ Martin) Fellowship from the National Health and Medical Research Council of Australia [grant number 510392], and the American Heart Association [grant number 10SDG3510018]. The funders had no role in study design, data collection and analysis, decision to publish or preparation of the manuscript."
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	"randomised" "Service in performing the preparation, randomisation and blinding code for the study intervention. D.J.E. is a consultant for Apnex Medical." To ensure that the arousal threshold was comprised of a random allocation of arousals across the night, arousal selection was performed by allocating each individual arousal a sequential number and using a random number generator to select the 20 arousals to be analysed.
Allocation concealment (selection bias)	Low risk	"randomised" "Service in performing the preparation, randomisation and blinding code for the study intervention. D.J.E. is a consultant for Apnex Medical."
Blinding of participants and personnel (performance bias) All outcomes	Low risk	"each patient received placebo or 3 mg of eszopiclone in random order immediately prior to sleep with only a research pharmacist knowing the treatment assignment"
Blinding of outcome assessment (detection bias) All outcomes	Low risk	"each patient received placebo or 3 mg of eszopiclone in random order immediately prior to sleep with only a research pharmacist knowing the treatment assignment"
Incomplete outcome data (attrition bias) All outcomes	Low risk	All subjects completed study.
Selective reporting (reporting bias)	Low risk	Appeared to report the outcomes recorded

Espinoza 1987.

Methods	Randomised, single blind, crossover trial. Method of randomisation not reported. Wilcoxon's signed rank test for paired data
Participants	10 adult males; mean range: 52.6; AHI 76.4; weight (as % ideal): 133.4 Inclusion criteria: > 15 apnoeas/hr; daytime hypersomnolence Exclusion criteria: Evidence of recent URTI; medication for OSA; use of CPAP within 1 week of study commencement
Interventions	IV aminophylline versus saline. Study duration: single night
Outcomes	AI; HI; sleep efficiency; apnoea duration; heart rate
Notes	Jadad score: 2 Funding source: not stated
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	stated randomised
Allocation concealment (selection bias)	Unclear risk	Information not available
Blinding of participants and personnel (performance bias) All outcomes	Unclear risk	stated double blind
Blinding of outcome assessment (detection bias) All outcomes	Unclear risk	stated double blind

Ferber 1993.

Methods	Randomised, double blind crossover trial. Method of randomisation not reported. Statistical analysis: Wilcoxon
Participants	12 adults with OSA (10 males). Mean age: 60.2; Mean BMI: 31. Mean AHI: 32.9 Inclusion criteria: > 10 AHI. Exclusion criteria: severe pulmonary insufficiency/neuroendocrine abnormality
Interventions	Naltrexone 50 mg versus placebo. Study duration: 2 nights. No washout period described.
Outcomes	AHI; TST; REM sleep; Hypoxia/hypercapnia;
Notes	Jadad score: 3 Funding source: not stated
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	stated randomised
Allocation concealment (selection bias)	Unclear risk	Information not available
Blinding of participants and personnel (performance bias) All outcomes	Unclear risk	stated double blind
Blinding of outcome assessment (detection bias) All outcomes	Unclear risk	stated double blind

Hedner 1996.

Methods	Randomised, double blind, crossover trial. Method of randomisation not reported. Statistical test: Student's t test.
Participants	12 adults (11 males); mean age: 49 years (range: 38 to 60). Inclusion criteria: previously diagnosed moderate to severe OSA
Interventions	Sabeluzole 10 mg BD (twice daily) 4 weeks 2 week washout
Outcomes	TST; patient preference; mood; memory; symptoms (VAS); ODI; minimum SaO2; plasma concentrations
Notes	Jadad score: 3 Funding source: This study was supported by grants from the Swedish Medical Research Council (grant no. 14X‐09892), The Swedish Heart and Lung Foundation and by Jansen Pharmaceuticals. R.G. was supported by an overseas travelling fellowship from the National Heart Foundation of Australia.
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	stated randomised
Allocation concealment (selection bias)	Unclear risk	Information not available
Blinding of participants and personnel (performance bias) All outcomes	Unclear risk	stated double blind
Blinding of outcome assessment (detection bias) All outcomes	Unclear risk	stated double blind

Hedner 2003.

Methods	Randomised, double blind crossover study. Method of randomisation: not reported. Statistical analysis: paired t‐test
Participants	10 adult males; Mean age: 48.3; BMI: 26.8; AHI: 54.4; Min SaO2: 73.5 Inclusion criteria: 25‐65 years; previously diagnosed OSA; ODI: 10 or more Exclusion criteria: CHD; cardiac arrhythmia; MI/stroke within 12m; previous/current psychiatric disease; regular use of benzodiazepines; intolerance to cholinesterase; myasthenia gravis; body weight >120% ideal; liver disease; current alcohol/drug abuse; CPAP use ≤1m prior to study entry
Interventions	Intravenous physostigmine (0.12 mcg/kg/minute) versus placebo. Study duration: 2 nights
Outcomes	AHI; AI; min SaO2; total sleep time; REM; sleep stages; heart rate
Notes	Jadad score: 4 Funding source: Supported by grants from The Swedish Heart and Lung Foundation and Faculty grants from the University of Go¨teborg.
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	stated randomised
Allocation concealment (selection bias)	Unclear risk	Information not available
Blinding of participants and personnel (performance bias) All outcomes	Unclear risk	stated double blind
Blinding of outcome assessment (detection bias) All outcomes	Unclear risk	stated double blind

Hein 2000.

Methods	Randomised, double blind, placebo controlled, crossover trial. Withdrawals: described (N = 8). Jadad score: 4. Statistical technique: ANOVA (NB individual patient scores were reported in the trial report)
Participants	22 participants recruited. Data on age and gender not reported. AHI: 13.8 (SD 4) Inclusion criteria: AHI > 5; one or more symptoms compatible with OSA
Interventions	Theophylline versus placebo (theophylline administered to maintain a therapeutic serum level of ≥8mg/L). Study duration: 2 x 14 day. No washout described
Outcomes	AHI; sleep quality; side effects
Notes	Data on AHI collected from second week of treatment period Funding source: not stated
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	stated randomised
Allocation concealment (selection bias)	Unclear risk	Information not available.
Blinding of participants and personnel (performance bias) All outcomes	Low risk	"identical in shape, colour and taste ensuring that both the patient and the investigator were blinded to the substance taken"
Blinding of outcome assessment (detection bias) All outcomes	Low risk	see above

Heitmann 1998.

Methods	Randomised, double blind parallel group trial. Method of randomisation not reported. Statistical analysis: Student's t‐test
Participants	53 adult males recruited (data on 48 who completed were reported). Mean age: 50.7 years. AHI: 62.43; BMI: 32.25 Inclusion criteria: 23‐69 years; mild‐moderate hypertension; OSA diagnosed by PSG and symptoms Exclusion criteria: Malignant/secondary hypertension; major systemic disease; history of alcohol/drug abuse; concomitant antihypertensive medications interfering with calcium antagonists
Interventions	Mibefradil 50 mg QD versus placebo. Study duration: 8 days
Outcomes	AHI; total sleep time; sleep efficiency; NREM; REM; blood pressure
Notes	Jadad score: 3 Funding source: Supported by a grant from Hoffmann‐La Roche AG, Grenzach‐Whylen
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	stated randomised
Allocation concealment (selection bias)	Unclear risk	Information not available
Blinding of participants and personnel (performance bias) All outcomes	Unclear risk	stated double blind
Blinding of outcome assessment (detection bias) All outcomes	Unclear risk	stated double blind

Issa 1992.

Methods	Randomised, double blind, crossover trial. Method of randomisation not reported.
Participants	8 men; 31 to 59 years. (AHI 27 (SD 4.2)).
Interventions	Clonidine 0.2 mg Taken each night for 10 days, then a week wash‐out period, then the next treatment given for 10 days. Sleep study performed in each patient for a total of six nights
Outcomes	AHI; minimum SaO2
Notes	Jadad score 2 Funding source: Do not have a copy of paper for the update
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	stated randomised
Allocation concealment (selection bias)	Unclear risk	Information not available
Blinding of participants and personnel (performance bias) All outcomes	Unclear risk	stated double blind
Blinding of outcome assessment (detection bias) All outcomes	Unclear risk	stated double blind

Jokic 1998.

Methods	Randomised, double blind, crossover trial. Withdrawals: all participants completed. Jadad score: 4. Statistical analysis: Paired t‐test
Participants	10 male adults (mean age: 49); BMI: 31, mild and moderate OSA (median AHI: 16). All participants had been using CPAP for 1‐13 months previously Inclusion criteria: OSA diagnosed by overnight PSG; AHI >/=10
Interventions	Topical lubricant (phosphocholinamin, 0.4 mL) administered at lights out and 3.5 hours later versus placebo (sealed pipette administered at same time). Study duration: 2 x single night assessments
Outcomes	AHI
Notes	Jadad score: 3 Funding source: Supported by a grant from the Heart and Stroke Foundation of Saskatchewan
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	stated randomised
Allocation concealment (selection bias)	Unclear risk	Stated randomised
Blinding of participants and personnel (performance bias) All outcomes	Low risk	Sealed pipette (no fluid dispensed), otherwise identical in appearance and scent
Blinding of outcome assessment (detection bias) All outcomes	Unclear risk	stated double blind

Kiely 2004.

Methods	Randomised, double blind, crossover study.
Participants	24 adult participants (19 male). Apnoeic snorers: N = 13 (AHI ≥10); Mean age: 47; BMI: 29.8; AHI: 26.5 Inclusion criteria: Clinical suspicion of OSA; symptoms of rhinitis Exclusion criteria: Fixed nasal obstruction; previous nasal fracture; deviated septum; use of anti‐histamines and decongestants
Interventions	Intranasal FP 100 mcg BD versus placebo. Study duration: 2 x 4 week treatment periods
Outcomes	AHI; TST; REM sleep; mean SaO2; min SaO2; nasal congestion; daytime alertness; subjective sleep quality; partner‐reported snoring
Notes	Jadad score: 4 Funding source: Supported by a grant from Glaxo Wellcome plc.
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	stated randomised
Allocation concealment (selection bias)	Unclear risk	Information not available
Blinding of participants and personnel (performance bias) All outcomes	Unclear risk	stated double blind
Blinding of outcome assessment (detection bias) All outcomes	Unclear risk	stated double blind

Kraiczi 1999.

Methods	Randomised, double blind crossover trial.
Participants	20 men, mean age 52.1 years. Three withdrawals. BMI 28.7kg/m2. DI 25.4 (SD 13) Inclusion criteria: ODI ≥ 10/hr; 25‐65 years Exclusion criteria: Known psychiatric disease; regular intake of CNS‐active drugs; symptomatic coronary heart disease; intolerance to paroxetine; alcohol or drug abuse
Interventions	Paroxetine 20 mg/day versus placebo. Study duration: 2 x 6 week treatment arms (4 week washout)
Outcomes	AHI; symptoms; sleep architecture (total sleep time; sleep efficiency; sleep latency; REM sleep; sleep stage); CPRS symptoms
Notes	Jadad score 4 Funding source: supported by grant 9892 from the Swedish Medical Reseacrh Council and grants from the Medical Faculty of Gothenburg University and the Swedish Heart and Lung Foundation
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	stated randomised
Allocation concealment (selection bias)	Unclear risk	Information not available
Blinding of participants and personnel (performance bias) All outcomes	Unclear risk	stated double blind
Blinding of outcome assessment (detection bias) All outcomes	Unclear risk	stated double blind

Mangin 1983.

Methods	Double blind, crossover study. Not reported as randomised.
Participants	9 male adults with OSA (diagnosis confirmed with PSG); Mean age: 55.3; mean PaO2: 74.1; SaO2: 94; PaCO2: 46.4 One participant had no apnoea under placebo and was excluded from the analysis.
Interventions	Almitrine (2 or 3 mg/kg up to 200 mg/d) versus placebo. Study duration: 2 x 5‐day treatment arms. No washout phase described.
Outcomes	Apnoea frequency by sleep stage.
Notes	Jadad score: 2 Funding source: Source of support not translated.
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	High risk	not reported as randomised
Allocation concealment (selection bias)	Unclear risk	Information not available
Blinding of participants and personnel (performance bias) All outcomes	Unclear risk	stated double blind
Blinding of outcome assessment (detection bias) All outcomes	Unclear risk	stated double blind

Marshall 2008.

Methods	Two randomised, double blind crossover and three‐arm parallel trial
Participants	First study: 20 adults (15 males), mean age 47 years (28‐64) and BMI 39.4 kg/m2(4.4), AHI 24.1 events per hour (8), Second study: 64 adults (56 males), mean age 52 years (30‐74), BMI 28.2 kg/m2(2.9), AHI 27.4 events per hour (11.1)
Interventions	mirtazapine versus placebo. First study: 3‐way crossover, dose finding study of mirtazapine 0 mg, 7.5mg, 15 mg, 30 mg and or 45 mg for 2 weeks (each patient exposed to maximum of 3 doses). Second study: 3‐arm parallel study of mirtazapine 15 mg (N = 26), mirtazapine 15 mg and compound CD0012 (N = 25) and placebo (N = 13) for 4 weeks
Outcomes	AHI
Notes	Funding source: this study was funded by Cypress Bioscience. Prof Grunstein is supported by National Health and Medical Research Council of Australia Practitioner Fellowships. Dr Wong was supported by the NH&MRC Centre of Clinical Research Excellence for Respiratory and Sleep Medicine.
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	"The pharmacy at Royal Prince Alfred Hospital (Sydney, Australia) facilitated the over‐encapsulation of all study drug to ensure that proper blinding was maintained. Randomization was also facilitated through the onsite pharmacy and was conducted manually using a randomised lists generated by an independent statistician contracted by Cypress Bioscience, Inc., who was not known to onsite study investigators and who never met any of the patients."
Allocation concealment (selection bias)	Low risk	See above, assumed done
Blinding of participants and personnel (performance bias) All outcomes	Low risk	"No study investigators in either study who were aware of dose allocation met any patient who was being treated, and no patient was aware of the dose they were receiving."
Blinding of outcome assessment (detection bias) All outcomes	Low risk	See above. "study investigator who did not meet any of the patients and who was completely independent of the industry sponsor (NSM)." "All studies were scored by a single registered polysomnographic technician who was blinded to the study medication and dose sequencing."
Incomplete outcome data (attrition bias) All outcomes	Low risk	Two participants dropped out due to adverse events
Selective reporting (reporting bias)	Low risk	Appeared to report the outcomes measured

Mendleson 1991.

Methods	Randomised, double blind, crossover trial. Statistical test: unclear.
Participants	5 adult males. Mean age: 45.4 years; AHI 30.8
Interventions	Buspirone 20 mg versus placebo Study duration: single night? (not specified)
Outcomes	AHI; min SaO2; TST; time to sleep onset
Notes	Jadad score 2. Study reported as letter to a journal. Funding source: not stated
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	stated randomised
Allocation concealment (selection bias)	Unclear risk	Information not available
Blinding of participants and personnel (performance bias) All outcomes	Unclear risk	stated double blind
Blinding of outcome assessment (detection bias) All outcomes	Unclear risk	stated double blind

Moraes 2008.

Methods	Randomised, double blind, parallel trial. Statistical analysis: one and two‐way ANOVA
Participants	23 adults (8 males) with mild/moderate Alzheimer disease, age 76.8 ± 6.2 years (donepezil), 72.6 ± 11 years (placebo), AHI 20 ± 15.9 events per hour (donepezil), 23.2 ± 26.4 events per hour (placebo). OSA confirmed by PSG. Exclusion criteria:other co‐morbidities, severe psychiatric disease and concomitant psychoactive and sleep medications
Interventions	Donezepil (N=11) 5 mg for 1 month and 10 mg for 2 months versus placebo (N=12) for 3 months
Outcomes	AHI, SpO2 < 90%, ADAS‐cog score
Notes	Funding source: this work was supported by Fundação de Amparo à Pesquisa do Estado de São Paulo
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	"Each medication‐containing box was packed by a standard pharmacy service. Boxes were coded as A or B, indicating placebo or donepezil (the signification of the codes was kept in a closed envelope). A random number list with uniform distribution from 0 to 1 was generated using software (Statistica; Statsoft; Tulsa, OK)."
Allocation concealment (selection bias)	Low risk	"Patients were consecutively allocated to two treatment groups according to the random number list"
Blinding of participants and personnel (performance bias) All outcomes	Low risk	"Researchers were blind to patient conditions when recording and scoring parameters."
Blinding of outcome assessment (detection bias) All outcomes	Low risk	"Codes were opened and assigned to each patient when statistical analysis was performed."
Incomplete outcome data (attrition bias) All outcomes	Low risk	All subjects completed study
Selective reporting (reporting bias)	Low risk	Appeared to report the outcomes recorded

Mulloy 1992.

Methods	Randomised, double blind, crossover trial.
Participants	12 adult males (3 withdrawals). Age range: 35 to 64 years. Mean AHI 59 Inclusion criteria: AHI >/=15
Interventions	Theophylline 800 mg nocte versus placebo Study duration: 4 weeks (washout not specified)
Outcomes	AHI; min SaO2; DI; AI; TST
Notes	Jadad score 3 Funding source: not stated
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	stated randomised
Allocation concealment (selection bias)	Unclear risk	Information not available
Blinding of participants and personnel (performance bias) All outcomes	Unclear risk	stated double blind
Blinding of outcome assessment (detection bias) All outcomes	Unclear risk	stated double blind

Prasad 2010.

Methods	Randomised, double blind, parallel trial. Statistical test: ANOVA
Participants	35 adults (23 males), age 21‐65 years, AHI >10 (range 10‐98). Exclusion criteria: SaO2 < 75% for more than 5% of total sleep time, severe OSA that precluded withdrawal of CPAP, significant co‐morbidities, CNS active drugs and serotonergic drugs
Interventions	4 arms: 1. Ondansetron 24 mg (N=9) 2. Fluoxetine 5 mg/Ondansetron 12 mg (N=9) 3. Fluoxetine 10 mg/Ondansetron 24 mg (N=10) 4. Placebo (N=7). 7 day run in, 14 and 28 days evaluation by PSG
Outcomes	AHI
Notes	Funding source: this study was supported by a grant from BTG International, which holds a license from University of Illinois at Chicago to issued and pending patents relevant to the combined use of ondansetron and fluoxetine to treat sleep apnea. This study involved off‐label investigational use of ondansetron and fluoxetine. Investigational drug supply was purchased by the University of Illinois at Chicago Investigational Drug Service from a pharmacy. Investigational ondansetron was manufactured by GlaxoSmithKline and fluoxetine was manufactured by Eli Lily. None of the authors have a financial relationship with the manufacturers of ondansetron or fluoxetine. BTG neither manufactures nor markets ondansetron or fluoxetine. No University of Illinois at Chicago or BTG personnel were involved in the statistical analysis of study results, which was contracted to a third‐party service provider: ProSoft Inc. Dr Logan is a full time Vice President of BTG International, the study sponsor. Dr Carley serves as an unpaid Director for SteadySleep Rx Co., and holds stock in this company. Drs Carley and Radulovacki are inventors on patents and patent applications that disclose the use of serotonin antagonists and reuptake inhibitors to treat sleep apnea. Dr Carley and Dr Radulovacki have received research support from SteadySleep Rx Co. Dr Radulovacki holds stock in SteadySleep Rx Co. Dr Prasad, Dr Olpade, and Dr Herdegen have indicated no conflicts of interest.
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	"randomised"
Allocation concealment (selection bias)	Unclear risk	"randomised"
Blinding of participants and personnel (performance bias) All outcomes	Unclear risk	"Randomized subjects were provided blinded study agent in white capsules on coded blister cards and instructed to take one labelled pill..."
Blinding of outcome assessment (detection bias) All outcomes	Low risk	Statistical analyses were contracted to a third‐party service provider. Assumed blinded.
Incomplete outcome data (attrition bias) All outcomes	Low risk	"The primary efficacy analysis was performed on the per‐protocol (PP) population. The PP population included all randomised subjects who: (1) received at least one dose of study medication (the intent to treat [ITT] population); (2) missed no more than 4 PM doses in treatment period 2 (days 15 to 28); and (3) received both doses of study medication (AM and PM) on study day 28 Safety analyses were performed on the ITT population, comprising all subjects who received at least one dose of study medication." "Eight randomised subjects failed to complete the trial per protocol (attrition from ITT to PP population):1 due to AE in the placebo group, and two in the placebo and Ond 24 + Fl 10 treatment groups each, and 1 each from the other groups due to noncompliance."
Selective reporting (reporting bias)	Low risk	Appeared to report the outcomes recorded

Rasche 1999.

Methods	Randomised, double blind, crossover study. Method of randomisation: not reported. Statistical analysis: Paired t‐test
Participants	20 adults with OSA (4F). Mean age: 53; Mean AHI: 35.6 (SD 25.3); Mean BMI: 28 Inclusion criteria: AHI >5; history of excessive daytime sleepiness; ≥ 18 years of age; adequate inhaler technique. Exclusion criteria: diagnosis of asthma/COPD; BMI ≥ 40; evidence of alcohol/drug misuse; sensitivity to ß2‐agonists; ß‐blocker therapy; pregnancy
Interventions	Inhaled long‐acting beta‐agonists (salmeterol) versus placebo Study duration: 3 nights (no washout)
Outcomes	TST; saturation; A/HI; AHI.
Notes	Jadad score 3 Funding source: This study was supported by grants from Glaxo Wellcome, London, United Kingdom
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	stated randomised
Allocation concealment (selection bias)	Unclear risk	Information not available
Blinding of participants and personnel (performance bias) All outcomes	Unclear risk	stated double blind
Blinding of outcome assessment (detection bias) All outcomes	Unclear risk	stated double blind

Stepanski 1988.

Methods	Randomised, double blind, crossover trial. Method of randomisation not reported. Statistical test: paired t‐test
Participants	8 adult males. Mean age 44.9; AHI 87.3; MSLT: 5.3 minutes Inclusion criteria: AHI ≥ 10 and EDS
Interventions	Protriptylline 10 to 20 mg nocte (every night) Study duration: 3 weeks
Outcomes	AI; HI; min SaO2; TST; subjective sleepiness
Notes	Jadad score 3 Funding source: not stated
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	stated randomised
Allocation concealment (selection bias)	Unclear risk	Information not available
Blinding of participants and personnel (performance bias) All outcomes	Unclear risk	stated double blind
Blinding of outcome assessment (detection bias) All outcomes	Unclear risk	stated double blind

Stradling 2003.

Methods	Randomised, double blind crossover trial. Method of randomisation not reported. Statistical test: Paired t‐test
Participants	10 adults (9 females). Participants described as having moderate OSA. Mean age: 53 years; BMI 33.7 kg/m2; AHI: 19‐62. Inclusion criteria: Symptomatic OSA; < 4% SaO2 dips of 10‐40; lived in/near Oxford; awaiting CPAP treatment
Interventions	Ondansetron (16 mg) QD versus matching placebo Study performed on two single night a week apart
Outcomes	AI; HI; ODI; AHI
Notes	Jadad score 2 Funding source: not stated
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	stated randomised
Allocation concealment (selection bias)	Unclear risk	Information not available
Blinding of participants and personnel (performance bias) All outcomes	Unclear risk	stated double blind
Blinding of outcome assessment (detection bias) All outcomes	Unclear risk	stated double blind

Sukys‐Claudino 2012.

Methods	Randomised, double blind, parallel trial. Statistical test: two‐way ANOVA
Participants	22 male between 35‐65 years, mean AHI in placebo group 26.4 ± 14.3 and in donepezil group 42.2 ± 19.4. Mean BMI in placebo group 31 ± 3.5 kg/m2 and in donepezil group 31.3 ± 3.4 kg/m2.
Interventions	Donezepil 5 mg for 2 weeks followed by 10 mg for two weeks (N=10) versus placebo (N=11) for 1 month
Outcomes	AHI, ESS, DI
Notes	Funding source: not stated
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	"Boxes were labelled with the codes X and Y, corresponding to donepezil or placebo, respectively (the codes were kept in a closed envelope). A random list with a uniform distribution of codes was generated by Statistica software. Patients were allocated to the two treatment groups according to the random code list."
Allocation concealment (selection bias)	Unclear risk	Not described
Blinding of participants and personnel (performance bias) All outcomes	Low risk	"Tablets containing a placebo or donepezil were prepared and packed in the same fashion." "Researchers and patients were blind to treatment condition during data collection and analysis."
Blinding of outcome assessment (detection bias) All outcomes	Low risk	"Researchers and patients were blind to treatment condition during data collection and analysis. Codes were opened and assigned to each patient when the statistical analyses were performed"
Incomplete outcome data (attrition bias) All outcomes	Low risk	2 dropouts from donezepil and one dropout from placebo group
Selective reporting (reporting bias)	Low risk	Appeared to report the outcomes recorded

Suratt 1986.

Methods	Placebo controlled, double blind, crossover study. Not described as randomised.
Participants	4 participants (age range 29‐55); BMI: 25 to 47.2. No data on AHI given. Inclusion criteria: not stipulated
Interventions	Doxapram (0.5 mg/kg lean body mass bolus, followed by 1 mg/mL infusion throughout the night) versus placebo. Study duration: 2 x 1 night treatment periods (washout of 1 day described)
Outcomes	AHI; blood pressure; desaturation index
Notes	Not described as randomised. No statistical stets undertaken as there were only 4 participants Funding source: supported by grant HL 30218 and RR 00847 from the A. H. Robins Company.
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	stated randomised
Allocation concealment (selection bias)	Unclear risk	Information not available
Blinding of participants and personnel (performance bias) All outcomes	Unclear risk	stated double blind
Blinding of outcome assessment (detection bias) All outcomes	Unclear risk	stated double blind

Whyte 1988.

Methods	Randomised, double blind, crossover trial.
Participants	10 participants (8 men 2 women); 34 to 67 years. Inclusion criteria: AHI > 15 with 2 from EDS snoring, unsatisfying sleep or awakenings (AHI 50 (SD 26)).
Interventions	Protriptyline 20 mg nocte 2 weeks Acetazolamide 250 mg bd 1 week and then 250 mg qds 1 week, Placebo 2 tabs at night
Outcomes	AHI; desaturation index, arousal index
Notes	Jadad score 2 Funding source: not stated
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	stated randomised
Allocation concealment (selection bias)	Unclear risk	Information not available
Blinding of participants and personnel (performance bias) All outcomes	Unclear risk	stated double blind
Blinding of outcome assessment (detection bias) All outcomes	Unclear risk	stated double blind

AHI: apnoea hypopnoea index

AI: apnoea Index

BD: twice daily

BMI: body mass index

CHF: congestive heart failure

CPAP: continuous positive airway pressure

CPRS: Comprehensive Psychopathological Rating Scale

EDS: excessive daytime sleepiness

ESS: Epworth Sleepiness Scale

Min SaO2: minimum oxygen saturation

MSLT: multiple sleep latency

nocte: every night

ODI: oxygen desaturation index

OSA: obstructive sleep apnoea

PSG: polysomnography

QD: once daily

SpO2: saturation of peripheral oxygen (an estimation of the oxygen saturation level usually measured with a pulse oximeter device)

TST: total sleep time

VAS: visual analogue scale

Characteristics of excluded studies [ordered by study ID]

Study	Reason for exclusion
Ami‐Hai 1986	No control group
Anonymous 1997	Review article
Atkinson 1985	Treatment of blood pressure not OSA
Bartel 1997	Study comparing interaction between different antihypertensive medications in OSA
Baughman 2009	Treatment of day time somnolence in sarcoidosis
Bittencourt 2008	Patients on CPAP
Black 2010	Not RCT
Block 1981	This randomised placebo controlled trial did not meet the inclusion criteria for the review, on the grounds that the investigators did not differentiate between central and obstructive sleep apnoea
Bortolotti 2006	Treatment of gastro‐oesophageal reflux, not OSA
Buttner 2003	Assessment of theophylline given as an attention enhancing agent
Camacho 1995	Study of temazepam in a sample of insomniac patients. Excluded because as a hypnotic it is not intended to treat the signs and symptoms of sleep apnoea
Catesby‐Ware 1983	Review article
Cistulli 1994	No placebo group
Clark 1979	No control group
Collop 1994	This randomised trial was excluded from the review, on the grounds that the patients were administered alcohol in order to exacerbate their sleep apnoea
Conway 1982	No placebo group
Darwish 2010	Patients on CPAP
Davila 1994	Participants recruited had non‐OSA sleep disordered breathing
Demirhan 2010	Paediatric cohort
Di Martino 1999	This trial was randomised, but compared theophylline with no intervention instead of a placebo. This review focuses on trials which compared an active intervention with placebo, as the primary indicator of efficacy
Dinges 2003	Study looking at the additive effect of modafinil to CPAP
Dinges 2007	Patients on CPAP treatment
Dontenwill 1997	This trial focused on the link between hypertension and sleep apnoea syndrome, and not obstructive sleep apnoea alone
Dorow 1991	No control group
El 2006	Effect of allopurinol on endothelial function in OSA patients, not a treatment for OSA
El Solh 2006	Effect of allopurinol on endothelial function in OSA patients, not a treatment for OSA
Empson 1999	Review article
Epstein 1998	Review article
Erman 2003	Patients treated with CPAP
Esteitie 2011	Paediatric cohort
Fietze 1993	The study examined three therapeutic interventions with no placebo control
Finnimore 1995	Participants had less than AHI 5
George 2010	Safety trial of sodium oxabate
George 2011	Safety trial of sodium oxabate
Gooneratne 2010	Patients treated with CPAP
Greenberg 1991	Before and after study
Grote 1994	Treatment of hypertension in OSA patients
Grote 1995	Treatment of hypertension in OSA patients
Grote 1995a	Treatment of hypertension in OSA patients
Guillemineault 1983	No placebo control
Hackett 1986	The allocation of placebo or acetazolamide was randomised, but the participants did not suffer from obstructive sleep apnoea
Hanzel 1991	The study was a randomised trial, but the design was unblinded and there was no placebo control
Hein 1995	Non‐randomised observational study
Heitmann 2010	Treatment of hypertension in OSA patients
Hirshkowitz 2007	Patients treated with CPAP
Hirshkowitz 2007a	Patients treated with CPAP
Hoijer 1994	We excluded this study was it was conducted to see whether there were interactions between a benzodiazepine and sleep apnoea, rather than being a study looking at therapeutic benefit of this drug on sleep apnoea outcomes
Hudgel 1996	Review article
Hudgel 1998	Review article
Ing 2000	Before and after study
Kantola 1999	Treatment of hypertension in OSA patients
Kauffmann 1991	Review article
Kempf 1991	This trial was not randomised and there was no placebo control
Khalil 2008	Paediatric population
Kheirandish‐Gozal 2008	Paediatric population
Killick	Assessment of testosterone on ventilatory response in patients with OSA, not aimed at treatment of OSA. Abstract form
Kingshott 2001	This randomised placebo controlled trial was excluded because all the subjects were treated with CPAP for the course of the study
Kopelman 1992	The study did not look at obstructive sleep apnoea
Krieger 1998	Review article
Kryger 2007	Safety trial
Krystal 2010	Patients on CPAP
Lee 2009	Not treatment of OSA
Lloyd 1983	Review article
Lombard 1985	Review article
Loube 1999	Review article
Lévy 1996	Review article
Martinez 2005	Study not primarily concerned with assessing sleep outcomes
Mayer 1990	Not RCT
Mayer 1991	Not RCT
Montserrat 1998	Review article
Nikolaou 2008	Not RCT
Nussbaumer‐Ochsner 2012	Treatment of CSA, not OSA
O'Malley 2003	Patients on CPAP
Pack 2001	Not RCT
Pelttari 1994	Not RCT
Pelttari 1998	Not RCT
Peter 1987	Not randomised. Patients entered a simple crossover designed study
Planes 1999	Not RCT
Roest 2012	Not treatment of OSA, OSA diagnosed as nocturnal heart rate variability
Rosenberg 2007a	Safety trial
Sadasivam 2011	Patients on CPAP
Scharf 1994	Not RCT
Schiza 2007	Modafinil effect on alertness in OSA patients
Schmidt 1983	Not a randomised controlled trial
Schwartz 2007	Patients on CPAP
Stewart 1992	No placebo group
Strohl 1986	Review article
Suurna 2008	Treatment of gastro‐oesophageal reflux
Tirosh 1995	The study did not meet the inclusion criteria of the review, on the grounds that it was a CPAP study
Trakada 1999	Observational study
Vgontzas 2004	Excluded because it assessed the use of TNF‐alpha antagonist as treatment for sleepiness associated with OSA
Watanabe 2009	Effect of elmisartan on endothelial function and aortic stiffness in hypertensive OSA patients
Weichler 1991	No placebo control
Wirth 1995	Study assessed nicotine versus theophylline
Zevin 2003	Study assessed high versus low dose of nicotine
Zou 2009	Treatment of hypertension in OSA patients
Zou 2010	Treatment of hypertension in OSA patients
Zou 2010a	Treatment of hypertension in OSA patients

Characteristics of studies awaiting assessment [ordered by study ID]

Hoyos 2010.

Methods	RCT, parallel, double blind, placebo
Participants	67 obese man with OSA, BMI 37, AHI 33
Interventions	1g testosterone 3x im injection(n=33) versus placebo (n=34)
Outcomes
Notes	Abstract form only. Not entirely certain that testosterone was aimed to treat OSA. Authors were contacted, however the raw data were not provided. Publication is awaited. This study is therefore still awaiting classification.

Differences between protocol and review

2012 update

• Added adverse effects outcome

• Applied Cochrane risk of bias tool to new included studies

• We halved the numbers of participants for crossover trials in the forest plots so that the correct total or people contributing data to the meta‐analysis was displayed

Contributions of authors

Earlier versions: IS developed the protocol with guidance from John Wright (JW). Toby Lasserson (TL) and IS assessed literature search results, agreed on inclusion and exclusion of studies, and extracted data. Data were entered by IS and TL. TL offered technical support for the development of the meta‐analysis. IS developed the results, discussion and conclusions with input from JW. IS wrote the abstract and synopsis.

2012 update. EJW extracted numerical data and risk of bias information and entered them into RevMan and assisted with drafting the review update. MM assessed the literature search results, extracted and entered the data into RevMan. MM extracted and entered data, updated the background, results, discussion and conclusion with input from IS. MM and IS agreed on exclusion and inclusion of studies. IS drafted and edited text and is the guarantor for the review.

Sources of support

Internal sources

• NHS Research and Development, UK.

• Emma Welsh, UK.

  St George's University of London, London, UK

External sources

• Garfield Weston Foundation, UK.

• Emma Welsh, UK.

  NIHR Core Funding Grant for Cochrane Review Group

Declarations of interest

None known

New search for studies and content updated (conclusions changed)