Does gabapentin provide benefit for patients with knee OA? A benefit-harm and cost-effectiveness analysis

Abstract

Objective:

Gabapentin can treat neuropathic pain syndromes and has increasingly been prescribed to treat nociplastic pain. Some patients with knee osteoarthritis (OA) suffer from both nociceptive and nociplastic pain. We examined the cost-effectiveness of adding gabapentin to knee OA care.

Method:

We used the Osteoarthritis Policy Model, a validated Monte Carlo simulation of knee OA, to examine the value of gabapentin in treating knee OA by comparing three strategies: 1) usual care, gabapentin sparing (UC-GS); 2) targeted gabapentin (TG), which provides gabapentin plus usual care for those who screen positive for nociplastic pain on the modified PainDETECT questionnaire (mPD-Q) and usual care only for those who screen negative; and 3) universal gabapentin plus usual care (UG). Outcomes included cumulative quality-adjusted life years (QALYs), lifetime direct medical costs, and incremental cost-effectiveness ratios (ICERs), discounted at 3% annually. We derived model inputs from published literature and national databases and varied key input parameters in sensitivity analyses.

Results:

UC-GS dominated both gabapentin-containing strategies, as it led to lower costs and more QALYs. TG resulted in a cost increase of $689 and a cumulative QALY reduction of 0.012 QALYs. UG resulted in a further $1,868 cost increase and 0.036 QALY decrease. The results were robust to plausible changes in input parameters. The lowest TG strategy ICER of $53,000/QALY was reported when mPD-Q specificity was increased to 100% and AE rate was reduced to 0%.

Conclusion:

Incorporating gabapentin into care for patients with knee OA does not appear to offer good value.

Introduction

Symptomatic knee osteoarthritis (OA) is a chronic, costly, and disabling condition that affects over 14 million Americans^{1, 2}. Pain is the central concern of patients with OA as it limits function, reduces quality of life, and drives individuals to seek treatment³. With no disease-modifying treatments available, pharmacologic OA-pain management agents include acetaminophen, non-steroidal anti-inflammatory drugs (NSAIDs), injections of corticosteroids or hyaluronate products, and/or opioids⁴. These medications offer only modest efficacy and are not tailored to specific pain mechanisms⁵.

Knee OA pain has long been considered nociceptive, arising from abnormal loading of the damaged joint⁶. However, roughly one in three patients with knee OA report neuropathic-like pain symptoms – such as allodynia and hyperalgesia –suggesting that increased sensitization and additional pain mechanisms are present^{5, 7–11}. Mechanistically, neuropathic pain follows from nervous system lesions – which have not been observed in knee OA¹². The International Association for the Study of Pain (IASP) defines nociplastic pain as altered nociception without clear evidence of tissue damage or disease of the somatosensory system causing the pain^{12, 13}. Consequently, we grouped various non-nociceptive components of some patients’ pain profiles under the umbrella term: nociplastic pain¹⁴. Patients with nociplastic pain may experience less relief from OA treatments, including NSAIDs, corticosteroid injections and total knee replacement^{7, 12, 15–17}.

Gabapentinoids are centrally-acting, brain-stabilizing drugs with analgesic efficacy in various neuropathic and nociplastic pain conditions^{18, 19}. Therefore, gabapentinoids may present an opportunity to better treat those with knee OA and nociplastic pain. However, gabapentinoids carry risks of adverse events (AEs), including a 5% chance of drug abuse/misuse and 3% chance of overdose^19–22. Screening for nociplastic pain could help identify patients most likely to benefit while sparing others of gabapentinoids’ serious AEs^{8, 9, 11, 23}. Our analysis focuses on gabapentin, which has shown no difference in safety nor efficacy from pregabalin, the other commonly prescribed gabapentinoid²⁴.

Use of gabapentinoids has skyrocketed. In the US, the rate of ambulatory visits involving gabapentinoids rose from ~1% in 2003 to ~3.5% in 2016²⁵. In the UK, prescription rates for patients with knee OA has increased over 17-fold since 2000²⁶. Yet, to our knowledge, there are no analyses that weigh the risks of AEs with the pain benefits from gabapentin for knee OA. In this analysis, we examine the clinical and economic impact of gabapentin in addition to usual care for patients with knee OA.

Materials and Methods

Analytic Overview

The Osteoarthritis Policy Model (OAPol) is a validated microsimulation model of the natural history and management of knee OA²⁷. We used OAPol to examine if adding gabapentin to usual care (UC) for patients with knee OA and inadequate pain relief from NSAIDs could be cost-effective.

The key outcomes were cumulative quality-adjusted life years (QALYs), lifetime medical costs, and incremental cost-effectiveness ratios (ICERs). We calculated ICERs as the difference in lifetime direct medical costs divided by the difference in QALYs between two strategies. Each year lived is quality-adjusted by a qualify of life (QoL) utility value (1 for perfect health), stratified by BMI, knee pain, age, number of comorbidities, and treatment-related complications. All costs are reported in 2020 USD. Analyses were conducted from a healthcare sector perspective, which includes direct medical costs paid by all payers, including public, private, and individuals²⁸. Costs and QALYs were discounted 3% annually²⁸. Traditionally, cost-effectiveness analyses are anchored in willingness-to-pay (WTP) thresholds, with $50,000 and $100,000/QALY serving as well-established benchmarks²⁹. Here, we report ICERs alone, allowing stakeholders to determine cost-effectiveness. Any strategy or combination of strategies that increased cost and decreased QALYs was labeled “dominated.” To address uncertainty in our inputs, we drew 1000 sets of values for nociplastic pain prevalence, mPD-Q sensitivity and specificity, gabapentin efficacy, gabapentin discontinuation rate, AE rates and QoL decrements associated with AEs from plausible probability distributions specific to each parameter (TA Table 8). We constructed a cost-effectiveness acceptability curve to depict the proportion of model runs where each strategy was preferred at a specific WTP threshold. A strategy was preferred if it produced the greatest number of QALYs while maintaining an ICER below the given WTP threshold. Whenever possible, this cost-effectiveness analysis adhered to the 2022 Consolidated Health Economic Evaluation Reporting Standards statement. We also report results from a benefit-harm analysis, a method for quantitative assessment of the overall value of a treatment based on consideration of its benefits and side-effects. This framework provides an objective and transparent mechanism for combining the various outcomes of treatment into a single ‘net benefit’ metric that can inform clinical decision-making.³⁰

The OAPol Model

OAPol uses prespecified distributions of demographic and clinical characteristics (e.g., age and BMI), comorbidities (e.g., cardiovascular disease and diabetes), and structural and symptomatic severity of knee OA to simulate different patient populations. Using Monte Carlo simulation, in each annual model cycle, subjects may transition to different health states, such as more severe knee OA, according to probabilities determined by their current health state. Subjects accrue QALYs and costs each cycle based on their clinical and/or demographic characteristics. Simulated subjects are followed until death, accounting for all potential benefits and drawbacks of treatment over their remaining lifespan. The average cumulative QALYs and costs for the entire simulated population are used in calculating ICERs.

Knee OA symptom severity is defined by pain ratings on the 100-point Western Ontario and McMaster Universities Osteoarthritis Index (WOMAC) scale, and ratings are split into five pain groups (<1, 1–15, 16–40, 41–70, and 71–100). Additional details about OAPol are published elsewhere and available in the Technical Appendix (TA)^{27, 31}.

Treatments can reduce patient pain in the model. Initial pain reductions are drawn from probability distributions, derived from treatment-specific published literature and secondary analysis of relevant cohort data, to account for variation in pain relief experienced by different individuals. Thus, some individuals receive little-to-no pain relief, while others experience extensive relief from the same treatment. Each cycle following an initial pain reduction, transition probabilities determine whether a subject’s pain relief is sustained. If pain relief is not sustained, a “regimen failure” occurs and the subject is evaluated for the next regimen (Figure 1). Voluntary discontinuation and/or adverse events can also lead to regimen discontinuation. Regimen costs include medication costs as well as dispensation, procedures, physician visits, and AE-associated costs.

Figure 1.

Panel A. The usual care, gabapentin sparing (UC-GS) treatment sequence for knee OA in the OAPol model. Subjects are initialized based on specific cohort characteristics and progress through the regimens outlined, including an NSAID plus corticosteroid injections, tramadol, opioids (for some subjects, others skip the tramadol and opioid regimens), total knee replacement, and revision total knee replacement. Subjects remain on each treatment until it is no longer effective. Death can occur at any point in the sequence.

Panel B. The targeted gabapentin (TG) treatment sequence for knee OA in the OAPol model. Subjects are initialized based on specific cohort characteristics and are screened for nociplastic pain using the modified PainDETECT questionnaire (mPD-Q) at initialization. If they screen positive for nociplastic pain, they receive gabapentin in addition to an NSAID plus corticosteroid injections before moving on to the rest of the UC treatment sequence of tramadol, opioids (for some subjects, others skip the tramadol and opioid regimens), total knee replacement, and revision total knee replacement. If they screen negative for nociplastic pain, they proceed normally through the UC sequence. Subjects remain on each treatment until it is no longer effective. Death can occur at any point in the sequence.

Panel C. The universal gabapentin treatment sequence for knee OA in the OAPol model. Subjects are initialized based on specific cohort characteristics and receive gabapentin in addition to an NSAID plus corticosteroid injections before moving on to the rest of the UC treatment sequence of tramadol, opioids (for some subjects, others skip the tramadol and opioid regimens), total knee replacement, and revision total knee replacement. Subjects remain on each treatment until it is no longer effective. Death can occur at any point in the sequence.

Cohort Characteristics

We modeled a population with demographic and clinical characteristics from the Glucosamine/chondroitin Arthritis Intervention Trial (GAIT) cohort³². We chose these data as they represent a population of patients with knee OA in whom NSAIDs did not lead to satisfactory pain relief. The baseline characteristics of our cohort included a starting age of 58.6 years (SD: 10.3), BMI of 31.7 kg/m² (SD: 7.1) and average WOMAC pain of 47.2 (SD: 14.7).

Our cohort was divided into two sub-cohorts with distinct pain phenotypes: those with nociceptive pain only and those with nociceptive and nociplastic pain. Based on estimates of nociplastic pain prevalence in knee OA, the nociceptive plus nociplastic cohort prevalence was set to 30%¹⁰. Within this group, we assumed that 50% of a subject’s pain was nociceptive and the other 50% nociplastic. In the nociceptive only cohort, all pain was nociceptive (TA Figure 1). We assigned a baseline pain score of 41 to the nociceptive only cohort and 61 to the nociceptive plus nociplastic cohort. With the nociceptive plus nociplastic cohort prevalence at 30% (and thus the nociceptive-only at 70%), this ensured that the overall (combined knee OA) cohort had an average starting pain of 47, equivalent to the GAIT cohort³². These pain values are consistent with studies demonstrating that individuals with nociplastic and nociceptive pain have greater overall pain levels compared to those with nociceptive pain only^{5, 16}.

Prevalence and incidence of comorbidities were derived from the 2017–2018 National Health and Nutrition Examination Survey (NHANES) and were stratified by age, race, and sex³³. Background medical costs were derived from the Medicare Current Beneficiary Survey (MCBS)³⁴, and we used a risk-adjustment model developed by Pope et al. to stratify costs by age and comorbidities³⁵. Annual QoL values were derived from The Osteoarthritis Initiative (OAI) and were stratified by age, BMI, pain severity and comorbidities³⁶.

Treatment Strategies

We examined three treatment strategies (Figure 1):

1. Usual care, gabapentin sparing (UC-GS) – participants are treated with a sequence of a second NSAID plus corticosteroid injections, tramadol, opioids (oxycodone), TKR, and revision TKR.

2. Targeted gabapentin (TG) – participants are screened for nociplastic pain using the modified PainDETECT Questionnaire (mPD-Q)⁸ and receive gabapentin if positive for nociplastic pain, in addition to a second NSAID; if negative, participants receive only NSAIDs. All receive the same UC treatment thereafter.

3. Universal gabapentin (UG) – all participants receive gabapentin in addition to a second NSAID before moving on to the rest of the UC sequence.

QALYs and lifetime medical cost estimates for each of the three strategies were calculated by combining weighted averages of the outputs from the following four OAPol runs, based on nociplastic prevalence as well as the sensitivity and specificity of the mPD-Q:

1. Nociceptive only cohort – treated with UC

2. Nociceptive only cohort – UC and gabapentin

3. Nociceptive plus nociplastic cohort – UC

4. Nociceptive plus nociplastic cohort – UC and gabapentin

For example, the base case UC-GS strategy is a weighted average of run 1 (70%) and run 3 (30%). The UG strategy is a weighted average of run 2 (70%) and 4 (30%). The weights for the TG strategy depend on the sensitivity and specificity of the mPD-Q.

We assumed that NSAIDs only treat nociceptive pain and gabapentin only treats nociplastic pain. Thus, in the base case, the nociceptive only cohort received pain relief strictly from NSAIDs, while the nociceptive plus nociplastic cohort received pain relief from both NSAIDs and gabapentin (TA). Gabapentin offered pain relief to only the nociceptive plus nociplastic cohort, but costs and risks of AEs were incurred by both cohorts.

Efficacy, utilization rates and AE profiles of the UC regimens have been previously published (TA)^{27, 31}. Informed by MCBS³⁴, we set 5% of subjects to receive opioids. Total knee replacement efficacy for subjects’ pain was decreased for subjects receiving opioids (TA)³¹.

Impact of Nociplastic Pain on UC Regimen Efficacies

Studies investigating corticosteroid injections and total knee replacements report worse outcomes in persons with nociplastic pain^{7, 16, 17}. Patients with nociceptive pain only and those with nociceptive and nociplastic pain receive similar absolute reductions in pain due to the given intervention – the worse outcomes reported reflect a variation in initial pain. Those with nociplastic pain have a higher baseline pain, resulting in higher outcome pain, even though the raw benefit received is the same. The equivalent absolute pain decrement represents nociceptive pain relief. Thus, we assume that corticosteroid injections, TKR, and revision TKR strictly treat nociceptive pain (TA). For the tramadol and opioid regimens, we assume equivalent efficacy for nociceptive and nociplastic pain.

Nociplastic Pain Screening Tool

We used the modified 9-item PainDETECT Questionnaire (mPD-Q) to screen for nociplastic pain as it can be completed and scored quickly in clinical settings^{8, 11}. PainDETECT is a validated, self-administered questionnaire that rates seven pathological pain qualities for a total score ranging from −1–38 to aid in identifying persons with a neuropathic pain component^{8, 37}. Scores between 13 and 18 suggest a possible neuropathic component, while scores of 19 and higher are indicative of neuropathic pain¹¹. The mPD-Q was modified for patients with knee OA by adjusting question phrasing to specifically address knee pain as opposed to overall pain⁸; the modified questionnaire was validated against quantitative sensory testing (QST) and reports a sensitivity of 58% and specificity of 71% when a score of 12 is used as the cut-off point for “neuropathic-like” pain¹¹. Given that neuropathy is not recognized to occur in knee OA, we adopted the current IASP terminology of “nociplastic pain” to characterize the neuropathic-like pain symptoms captured with the same mPD-Q cutoff¹⁴.

Gabapentin Characteristics

Efficacy

We derived gabapentin’s pain efficacy (40.6% reduction of nociplastic pain) from a Cochrane systematic review¹⁸ including 7 studies and 1439 participants that compared gabapentin to placebo in treating painful diabetic neuropathy (TA). We assumed the sustainability of this pain benefit was like that reported in NSAID trials³⁸. With no data available for gabapentin’s efficacy in treating knee OA nociplastic pain specifically, we used diabetic neuropathy as a proxy. To address uncertainty in using a proxy, we performed extensive sensitivity analyses for this parameter.

Adverse Events (AEs)

We derived AEs that resulted in immediate discontinuation of gabapentin (drug abuse/misuse, overdose, suicidal behavior, and death) and those that did not warrant discontinuation (dizziness, somnolence, peripheral edema, nausea) from published literature^19–22 (TA). Only AEs that occurred at rates significantly different from placebo were included. Ten percent of the cohort experienced AEs resulting in immediate discontinuation, while 35% experienced other AEs^19–22. Overall, 16% of users discontinued gabapentin by choice with 7% due to AEs that did not necessitate discontinuation²¹.

Cost

We derived the annual cost of a 600 mg dose taken 3x/day of generic gabapentin from Red Book Online³⁹. We used the average wholesale acquisition cost for 100-unit packages, which was $320/year. In addition to the pharmaceutical cost, we included a monthly dispensing fee⁴⁰, bringing the yearly price of gabapentin to $463.

Sensitivity Analyses

We varied input parameters in one-way, two-way, and scenario-based deterministic sensitivity analyses (DSAs) to understand the breadth of possible outcomes. Ranges of values for these analyses were informed by base-case results. We also conducted a probabilistic sensitivity analysis (PSA) to understand the likelihood of possible outcomes when parameter values were simultaneously selected from plausible probability distributions.

One-Way Sensitivity Analyses

We varied the following parameters: initial pain distribution of nociplastic vs. nociceptive pain in the nociceptive plus nociplastic cohort (50% nociplastic to 95% nociplastic); screening test sensitivity (58–100%) and specificity (71–100%); nociplastic prevalence (20–40%)¹⁰; gabapentin nociplastic pain efficacy (20–60% reduction); gabapentin nociceptive pain efficacy (4–20% reduction); and, adverse events (AEs necessitating discontinuation varied from 10% to 0%, AEs not necessitating discontinuation varied from 17.5% to 0%). These ranges are summarized in TA Table 7.

Two-Way Sensitivity Analyses

We conducted additional sensitivity analyses by simultaneously varying two parameters using the same ranges as in the one-way analyses (TA Table 7).

Scenario-Based Deterministic Sensitivity Analyses

To approximate an idealized version of more reliable screening tools like QST, we considered a “perfect” test where both specificity and sensitivity of the mPD-Q were set to 100%. Additionally, we considered an equal pain case where the initial WOMAC pain score for both the nociceptive only and nociceptive plus nociplastic cohorts were set to 47 (SD: 14.9) (TA Figure 2).

Results

Base Case

The UC-GS strategy led to 11.44 QALYs (95% credible interval (CI): 11.42–11.46) and a cumulative lifetime medical cost estimate of $221,301 (95% CI: $221,035-$221,544). The TG and UG strategies led to 11.43 QALYs (95% CI: 11.42–11.46) and 11.41 QALYs (95% CI: 11.39–11.45), and lifetime medical costs of $221,990 (95% CI: $221,484-$222,264) and $223,000 (95% CI: $222,319-$223,248), respectively. Thus, the UC-GS strategy dominated both the TG and UG strategies as both gabapentin-containing strategies increased costs and decreased QALYs. All strategies resulted in comparable opioid and TKR utilization rates (Table 2). Non-quality-adjusted life expectancies were 23.86 (95% CI: 23.85–23.87), 23.85 (95% CI: 23.85–23.87) and 23.84 (95% CI: 23.83–23.88) years for the UC-GS, TG and UG strategies respectively.

Table 2.

Cost-effectiveness analysis results for adding gabapentin to the usual care knee OA treatment sequence in three scenarios.

Base Case
Strategy †	QALY	Cost	ICER	% Using Opioids	% Receiving TKR	TKR Age
UC-GS	11.44	$221,301	--	5.9%	53.9%	64.8
TG	11.43	$221,990	Dominated#	5.9%	54.0%	64.8
UG	11.41	$223,169	Dominated#	5.6%	54.3%	64.7
Perfect Screening Test *
Strategy †	QALY	Cost	ICER	% Using Opioids	% Receiving TKR	TKR Age
UC-GS	11.44	$221,301	--	5.9%	53.9%	64.8
TG	11.44	$221,810	Dominated#	5.9%	53.9%	64.8
UG	11.41	$223,169	Dominated#	5.6%	54.3%	64.7
Equal Pain +
Strategy †	QALY	Cost	ICER	% Using Opioids	% Receiving TKR	TKR Age
UC-GS	11.45	$221,428	--	5.8%	54.5%	64.7
TG	11.44	$222,151	Dominated#	5.7%	54.6%	64.7
UG	11.42	$223,352	Dominated#	5.4%	55.0%	64.6

^*Perfect screening test refers to the scenario where both mPD-Q sensitivity and specificity were set to 100%

⁺Equal pain refers to the scenario where both the nociceptive only and nociceptive plus nociplastic cohorts initialized with an equivalent WOMAC pain score of 47

^†UC-GS is the usual care, gabapentin sparing strategy; TG is the targeted gabapentin strategy; UG is the universal gabapentin strategy

^#Any strategy or combination of strategies that increased cost and decreased QALY was labeled, “Dominated.”

Sensitivity Analyses

One-way Deterministic Sensitivity Analyses

Across all one-way analyses, QALYs ranged from 11.38–11.47, 11.38–11.46 and 11.36–11.45 for UC-GS, TG, and UG, respectively. Lifetime costs remained at $221,301, $221,990, and $223,169 for UC-GS, TG, and UG, respectively. The UC-GS strategy dominated the other strategies in nearly all sensitivity analyses (TA Table 9). When the AE rate leading to immediate withdrawal was decreased from 10% to 0%, the TG strategy had an ICER of $161,000/QALY, the most favorable ICER observed in these analyses.

Two-way Deterministic Sensitivity Analyses

The results were not sensitive to simultaneous change in nociplastic pain prevalence and mPD-Q accuracy, nor gabapentin efficacy and mPD-Q accuracy. Across all ranges considered, the UC-GS strategy dominated. When the nociceptive plus nociplastic cohort’s pain distribution was varied alongside mPD-Q sensitivity, the UC-GS strategy dominated across all ranges. However, when the pain distribution was set to 95% nociplastic and 5% nociceptive, and mPD-Q specificity was set to 100%, the TG strategy had an ICER of $119,000/QALY. The UG strategy remained dominated. For these analyses, QALYs ranged from 11.38–11.44 and 11.38–11.45 for UC-GS and TG, respectively. Lifetime costs ranged from $221,185-$221,811 and $221,460-$222,707 for UC-GS and TG, respectively (TA Table 10).

Results from the two-way sensitivity analyses are displayed in Figure 2 and TA Table 10. In every scenario and range considered, either the UC-GS strategy dominated both gabapentin strategies, or the TG strategy had a lower ICER than the UG strategy. Consequently, we show only the TG ICERs. With mPD-Q sensitivity at 100% and AE rate at 0%, TG had an ICER of $110,000/QALY (Figure 2A). With mPD-Q specificity at 100% and AE rate at 1%, 0.5%, and 0%, the TG ICERs were $79,000, $58,000, and $53,000/QALY, respectively (Figure 2B). When gabapentin efficacy and AE rate (type that results in immediate discontinuation) were varied simultaneously, the lowest TG ICER of $108,000/QALY resulted when gabapentin’s nociplastic pain reduction was increased from 40.6% (base case) to 60%, and its AE rate was decreased from 10% (base case) to 0%.

Figure 2.

Panel A. Bivariate heat map of the modified PainDETECT questionnaire (mPD-Q) sensitivity and adverse event (AE) rate for the targeted gabapentin (TG) strategy. The AE rate refers to the likelihood of experiencing one of the three AEs (drug abuse/misuse, overdose, suicidal behavior, and deaths) that lead to immediate discontinuation of gabapentin. Incremental cost-effectiveness ratios (ICERs) for the TG strategy were calculated for increasing mPD-Q sensitivity values and decreasing AE rates. The AE rate was decreased from the base case value of 10% to 0% and the mPD-Q sensitivity was increased from the base case value of 58% to 100%. All other parameters were held at base case values. ICERs were categorized into buckets and assigned an orange color as specified in the key.

Panel B. Bivariate heat map of the modified PainDETECT questionnaire (mPD-Q) specificity and adverse event rate for the targeted gabapentin (TG) strategy. The AE rate refers to the likelihood of experiencing one of the three AEs (drug abuse/misuse, overdose, suicidal behavior, and deaths) that lead to immediate discontinuation of gabapentin. Incremental cost-effectiveness ratios (ICERs) for the TG strategy were calculated for increasing mPD-Q specificity values and decreasing AE rates. The AE rate was decreased from the base case value of 10% to 0% and the mPD-Q specificity was increased from the base case value of 71% to 100%. All other parameters were held at base case values. ICERs were categorized into buckets and assigned an orange color as specified in the key.

Panel C. Bivariate heat map of gabapentin efficacy and adverse event rate for the targeted gabapentin strategy. The AE rate refers to the likelihood of experiencing one of the three AEs (drug abuse/misuse, overdose, suicidal behavior, and deaths) that lead to immediate discontinuation of gabapentin. Incremental cost-effectiveness ratios (ICERs) for the TG strategy were calculated for increasing gabapentin efficacy values and decreasing AE rates. The AE rate was decreased from the base case value of 10% to 0% and gabapentin efficacy was increased from the base case value of 40.6% to 60% nociplastic pain reduction.

Scenario-based Sensitivity Analyses

In both the “perfect” test and equal pain scenarios, the UC-GS strategy dominated the TG and UG strategies (Table 2).

Probabilistic Sensitivity Analyses

The distribution of model outcomes in the cost-effectiveness plane resulting from the PSA is illustrated in Figure 3. Using these data, 95% credible intervals were constructed for lifetime medical costs and QALYs gained for each strategy (TA Table 11). Figure 4 depicts the results of the PSA in a cost-effectiveness acceptability curve. This shows the proportion of times each strategy was the cost-effective option at willingness-to-pay thresholds of $0 to $200,000/QALY. At a WTP threshold of $50,000/QALY, UC-GS was the cost-effective strategy in 100% of cases. At the $200,000/QALY WTP threshold, the TG strategy was cost-effective in 5.7% of cases and, at the $100,000/QALY WTP threshold, the TG strategy was cost-effective in 0.8% of cases. UG was not cost-effective at any WTP threshold considered.

Figure 3.

The cost-effectiveness points distribution corresponding to the probabilistic sensitivity analyses that were conducted varying gabapentin efficacy, nociplastic prevalence, discontinuation rates, mPD-Q sensitivity and specificity, AE rates and AE QoL multipliers. One thousand sets of probabilistically chosen inputs were drawn from plausible distributions. Each point in the plane marks the average lifetime medical cost and average cumulative quality adjusted life years for a single strategy from one of the 1,000 model runs in the probabilistic sensitivity analysis. Lifetime medical costs are reported in 2020 USD and treatment efficacy is reported in quality adjusted life years. The three treatment strategies include usual care (UC), universal gabapentin (UG), and targeted gabapentin (TG).

Figure 4.

Probabilistic sensitivity analyses were conducted varying gabapentin efficacy, nociplastic prevalence, discontinuation rates, mPD-Q sensitivity and specificity, AE rates and AE QoL multipliers. 1,000 iterations of probabilistically chosen inputs were drawn. Incremental cost-effectiveness ratios (ICERs) were calculated for each iteration comparing the usual care, gabapentin sparing (green line), targeted gabapentin (orange line), and universal gabapentin (purple line) treatment strategies. The ICERs for these strategies were compared at a range of willingness-to-pay (WTP) thresholds, and that which produced the greatest number of quality-adjusted life years while remaining below the WTP threshold was termed the preferred strategy. The probability of being the preferred strategy is plotted against WTP thresholds from $0 to $200,000/QALY.

Discussion

We used a widely-published, validated computer simulation model to evaluate the cost-effectiveness of adding gabapentin to usual care in treating knee OA. Results from our analysis suggest that adding gabapentin to usual care for patients with knee OA leads to greater costs without additional benefits measured by cumulative QALYs. The UC-GS strategy dominated both TG and UG as it led to lower lifetime medical costs and more QALYs. Sensitivity analyses showed that AE rates were the driving factors of gabapentin’s value. The TG strategy led to ICERs below $175,000/QALY only when the likelihood of experiencing an AE that resulted in immediate discontinuation dropped from 10% (base case) to 0.5%.

The finding that gabapentin does not provide good value is driven primarily by high gabapentin AE rates and secondarily by modest gabapentin efficacy. The TG strategy should help ensure that only those who would benefit from gabapentin would also take on the AE risks. However, our results show that this is not enough to increase gabapentin’s value. Even in the “perfect” test scenario, UC-GS dominated. Furthermore, gabapentin use did not delay TKR treatment and the age of TKR was consistent across strategies (65 years). AEs associated with gabapentin treatment led to higher discontinuation rates such that simulated individuals spent comparable amounts of time on the NSAIDs plus gabapentin and the NSAIDs only treatment regimens, despite gabapentin’s additional pain efficacy. In the one-way sensitivity analysis that varied AE rates, the TG strategy had an ICER of $161,000/QALY when the likelihood of experiencing any of the AEs that necessitated immediate discontinuation was set to 0%. The ICER only dropped to $103,000/QALY when the mPD-Q specificity was simultaneously increased to 85%. With the specificity at 100%, the screening strategy ICER reduced to $53,000/QALY. All other analyses led to TG ICERs above the $100,000/QALY WTP threshold. These analyses show that gabapentin-related AEs had a decisive impact on our results. Unless there is some combination of a more accurate nociplastic screening tool, larger pain reduction, and no AEs, the gabapentin-based strategies do not offer sufficient value.

To the best of our knowledge, this is the first study to examine the cost-effectiveness of gabapentin for treating knee OA pain. For a drug with such high prescription rates^{25, 26}, the literature on its cost-effectiveness remains surprisingly sparse. A 2019 knee OA randomized controlled trial⁴¹ reported greater WOMAC pain reduction in the gabapentin group than the acetaminophen group; however, they did not separate the overall cohort by pain status and did not perform a cost-effectiveness analysis (CEA). Gabapentin’s pain benefits must be contextualized by their significant AE profile, which includes overdose and suicidal behavior.

CEAs have shown several pharmacologic treatments to be cost-effective for knee OA including hyaluronate and platelet-rich plasma injections.⁴² However these non-centrally acting treatments may not be as effective among patients with nociplastic knee OA pain.¹² CEAs conducted for both gabapentin and pregabalin in treating neuropathy showed that the value of one over the other relies on just 6–9 pain-free days^{43, 44}. When considered over a lifetime, these differences are negligible and confirm our decision to use gabapentin as a proxy for the gabapentinoids family. O’Connor et al.⁴⁵ and Bellows et al. ⁴⁶ found that both duloxetine and desipramine dominated pregabalin and gabapentin in treating painful diabetic neuropathy. In Bellows et al.’s PSA results, gabapentin was the most cost-effective treatment just 14% of the time at a WTP threshold of $50,000/QALY⁴⁶. Similarly, our results show that incorporating gabapentin into knee OA care is not cost-effective compared to usual care. Furthermore, duloxetine was found to be cost-effective compared to usual care in treating knee OA⁴⁷.

Our results should be viewed considering several limitations. First, we limited our analyses to OA treatments covered by major payers in the US. Due to a dearth of published trials evaluating gabapentin use in pain-stratified knee OA cohorts, studies used to derive gabapentin efficacy only included patients with diabetic neuropathy. While nociplastic pain involves similar symptoms, assuming gabapentin would be equally efficacious in treating nociplastic and neuropathic pain is perhaps an overestimate. We address this limitation with sensitivity analyses that vary gabapentin efficacy. To model the distinct impact of gabapentin versus NSAIDs on pain, we assumed that total pain could be split into two discrete categories. We are unaware of measures that accurately quantify nociplastic vs. nociceptive pain. Sensitivity analyses varying the relative proportions of nociceptive and nociplastic pain showed that our base case conclusions were robust across all scenarios considered. We acknowledge evidence that suggests some usual care treatments for OA, such as injections and TKRs, may affect nociplastic pain⁴⁸, and that opioid treatments may not affect those with and without nociplastic pain equally⁴⁹. However, these assumptions affect all strategies similarly and likely do not confound our comparative results. Moreover, we assume that gabapentin adverse event occurrence is independent of all patient characteristics. While we do perform sensitivity analyses concerning the AE rates for gabapentin, we did not perform a specific subgroup analysis of those experiencing serious gabapentin AEs because we are not aware of any prognostic tools to determine if one individual is more likely to experience a severe AE. Finally, we assumed that gabapentin is only efficacious for nociplastic pain but found this to not affect results in sensitivity analyses.

Nociplastic pain is prevalent in the knee OA community; identifying individuals with non-nociceptive pain components could offer opportunities for personalized treatment strategies^{5, 9, 10}. Those affected by nociplastic pain are at a disadvantage as they are likely to receive less benefit from NSAIDs, corticosteroid injections, and total knee replacements. Gabapentin prescription rates over the past decade have increased in hopes of providing more targeted pain relief for those with nociplastic pain^{25, 26}. However, our analysis suggests that gabapentin is not cost-effective in treating patients with knee OA, even if we could accurately identify those with nociplastic pain 100% of the time, due to its serious AE profile. Further research is needed to determine what interventions could offer value to this subset of patients with knee OA.

Table 1.

Key model inputs (costs in 2020 USD).

Overall Cohort Characteristics
Parameter	Mean (SD)		Source
Age (years)	58.6 (10.3)		Derived from Clegg et al.32
Female	64.1%
Race/Ethnicity
White non-Hispanic	80.8%
White Hispanic	4.0%
Black non-Hispanic	14.4%
Black Hispanic	0.7%
BMI	31.7 (7.0)
Kellgren Lawrence (KL) grade
2	55.3%
3	44.7%
Starting WOMAC knee pain	47 (14.7)
Nociceptive Only Cohort Characteristics
Parameter	Mean (SD)		Source
Prevalence	70%		Derived from Zoli et al.10, Wiffen et al.18
Starting WOMAC knee pain	41 (14.9)		Derived from Clegg et al.32
Nociceptive + Nociplastic Cohort Characteristics
Parameter	Mean (SD)		Source
Prevalence	30%		Derived from Zolio et al.10, Wiffen et al.18
Starting WOMAC knee pain	61 (14.9)		Derived from Clegg et al.32
Percent nociceptive pain	50%		Assumptiona
Percent nociplastic pain	50%
Gabapentin Intervention Parameters
Parameter	Mean (SD)		Source
Annual cost	$463		Derived from Red Book39, CCPA40 (TA)
Percent reduction in nociceptive WOMAC pain	0%		Assumptionb
Percent reduction in nociplastic WOMAC pain	40.6%		Derived from Wiffen et al.18
Annual probability treatment stops providing pain relief*	24% - 75%		Derived from Scott et al.38
Discontinuation rate**	16%		FDA21
Adverse Events	Probability †		Source
Dizziness	28%		FDA21
Somnolence	21%
Peripheral edema	8%
Nausea	13%
Suicidal behavior & death	0.9%		Molero et al.20
Overdose	3.5%
Drug abuse/misuse	5.7%		Evoy et al.22
Screening Test Parameters
Parameter	Value		Source
Sensitivity	58%		Hochman et al.11
Specificity	71%
Cost	$0		Assumptionc
Other Intervention Costs
Intervention	Cost (year 1)	Cost (years 2+)	Source
NSAIDS/PT/Braces	$903	$722	Derived from Losina et al.1 (TA)
Corticosteroid injections	$574	$574	Derived from Losina et al.1 (TA)
Tramadol	$580	$716	Derived from Huizinga et al.31 (TA)
Opioids (oxycodone)	$1,076	$1,328	Derived from Huizinga et al.31 (TA)
Primary TKR	$20,120	$119	Derived from Stanley et al.50 (TA)
Revision TKR	$28,111	$119

^aAssumption = 50% of the cohort’s pain is nociceptive, 50% is nociplastic

^bAssumption = gabapentin is only efficacious in treating nociplastic pain

^cAssumption = the mPD-Q questionnaire costs $0

^*The range spans all pain groups

^**This value does not include the 100% discontinuation rate of the three more severe adverse events

^†Rates of adverse events are halved in years 2+

^#This range refers to pain decrements for pain group 3 to pain group 5. To be in pain group 3, a person must have a WOMAC score of at least 15

^§These values were diminished if a person had previously taken opioids