Cost-effectiveness of GLP-1 receptor agonists for treating obesity in people without diabetes: A systematic review and meta-analysis

Obesity is a chronic medical condition that leads to numerous complications and independently increases the risk of cardiovascular disease and mortality.1 It was estimated that in 2022, approximately 1 billion people worldwide were living with obesity, which has more than doubled since 1990.2 The global burden of obesity and its associated comorbidities has elevated substantial healthcare expenditure, lessened individual quality of life, and caused loss of productivity.3 Effective treatments for obesity include lifestyle interventions, pharmacotherapy, and bariatric surgery.4

Glucagon‐like peptide 1 receptor agonists (GLP‐1RAs) are one effective treatment option recommended by various clinical practice guidelines for treating obesity.5, 6, 7 The current comparative evidence suggests that semaglutide provides substantially larger benefits in weight reduction than other drugs with a similar risk of adverse events.8 However, the costs of GLP‐1RAs are considered relatively high compared with other medications.9

Several economic evaluation (EE) studies showed inconsistent findings; some identified semaglutide as the most cost‐effective option,10, 11 while others reported high incremental cost‐effectiveness ratios (ICERs), suggesting it may not be cost‐effective.10, 11, 12 These mixed results make it unclear whether GLP‐1RAs are cost‐effective for obesity treatment. Crespo et al. proposed that a meta‐analysis of cost‐effectiveness studies could provide a quantitative summary of evidence to guide decision‐making.13 This approach has recently been endorsed by the WHO's Immunization and Vaccine‐related Implementation Research Advisory Committee (IVIR‐AC), especially for countries lacking context‐specific data,14 such as national estimates of direct and indirect healthcare costs or local willingness‐to‐pay thresholds, which are required to conduct tailored economic evaluations. In these settings, pooled cost‐effectiveness estimates from meta‐analyses may provide useful guidance for policy decision‐making. We conducted a systematic review and meta‐analysis of cost‐effectiveness studies and estimated the pooled incremental net benefits (INBs) to assess the cost‐effectiveness of using GLP‐1RAs for treating obesity in people without diabetes.

This study followed the Preferred Reporting Items for Systematic Reviews and Meta‐Analysis Protocols (PRISMA).15 The study protocol was registered in PROSPERO (CRD42024555582).16

The growing burden of obesity underscores the need for accessible treatments. While GLP‐1RAs have demonstrated effectiveness for obesity treatment, their relatively high cost raises concerns about cost‐effectiveness, requiring evidence to support their use. Our study addressed this gap by pooling INBs from EEs of GLP‐1RAs, providing new insights into their cost‐effectiveness for treating obesity in people without diabetes. We found that both semaglutide and liraglutide are not cost‐effective compared to other interventions such as lifestyle modifications, phentermine‐topiramate, or naltrexone‐bupropion. These findings are consistent across studies and are due mainly to the high costs of these medications.

While the magnitude of pooled INBs varied, our findings showed a consistent trend: GLP‐1RAs were not cost‐effective compared with either no treatment or active anti‐obesity medications, including lifestyle interventions, phentermine‐topiramate, and naltrexone‐bupropion. This pattern was observed across the main analysis and sensitivity analyses. The only exception was in a subgroup comparison where GLP‐1RAs appeared cost‐effective relative to no treatment when modelled over extended time horizons (>10 years). These results highlight how cost‐effectiveness estimates can differ depending on the comparator and underscore the importance of distinguishing comparator type and time horizon when interpreting the economic value of GLP‐1RAs.

Our results differ from a previous systematic review of the economic evaluation of anti‐obesity drugs by Xue et al.,34 which found GLP‐1RAs to be cost‐effective for obesity management. This discrepancy may stem from differences in the included populations, as Xue et al. incorporated studies involving patients with T2D,34 where GLP‐1RAs address broader health complications and lead to greater cost offsets. In contrast, our study focused primarily on obesity treatment in people without diabetes. Moreover, the review by Xue et al. included only a small number of studies specifically evaluating GLP‐1RAs, most of which used short intervention durations of less than 5 years. These limited time horizons may underestimate long‐term costs and benefits. In addition, several studies adopted a societal perspective, incorporating indirect benefits such as productivity gains. These methodological differences likely contributed to the more favourable cost‐effectiveness outcomes reported by Xue et al. compared to our analysis including longer‐term, payer‐perspective studies. Including diabetes‐related health states in economic models likely amplifies the perceived benefits of obesity treatment, resulting in more favourable cost‐effectiveness outcomes in the original EE studies included in the previous review.

Several factors likely contributed to the findings that GLP‐1RAs are not cost‐effective for treating obesity in people without diabetes. One major factor is the high cost of these medications, particularly when compared to alternative weight management strategies such as lifestyle modifications, behavioural therapies, or other anti‐obesity medications. These alternatives typically have significantly lower costs and can result in meaningful, sustained weight loss over time.35 Additionally, most studies included in this review assumed continuous use of GLP‐1RAs throughout the study's time horizon, further compounding the overall cost. For obese people without diabetes, especially those at lower risk for obesity‐related complications, the benefits of weight loss may not justify the ongoing expense of long‐term treatment. Moreover, most simulation models may not account for broader benefits such as improved mental health, or productivity, which could lead to an underestimation of the overall value of GLP‐1RAs. Additionally, the indirect economic benefits of weight loss achieved through GLP‐1RAs, such as reductions in hospitalisations, surgeries, or medications for obesity‐related comorbidities, may take years to materialise. In the short term, these benefits are unlikely to offset the high costs of the medication, particularly for individuals who are not at high risk for obesity‐related health issues, or when such complications were not accounted for in the economic models.

The cost of GLP‐1RAs was consistently identified as a key driver of their cost‐effectiveness across included studies. Although most studies did not report whether prices reflected negotiated discounts or other pricing arrangements, and the cost inputs used were generally based on current or recent market prices. As such, the observed unfavourable cost‐effectiveness may partly reflect the high drug prices at the time of analysis. Notably, the majority of studies were conducted in the United States, where medication costs are typically higher than in many other countries We also acknowledge recent reductions in the market prices of GLP‐1RAs,36 may improve their cost‐effectiveness in future analyses. As pricing evolves, updated evaluations will be important to inform health policy and clinical decision‐making.

All included studies used a healthcare system or payer perspective, without considering a societal perspective that includes patient time costs or productivity losses. These broader costs can substantially affect cost‐effectiveness estimates, especially for obesity treatments that may reduce time lost from work or daily activities. Evidence from a previous review37 and CEA registries38 has shown that including such costs often results in more favourable cost‐effectiveness outcomes. Future evaluations should consider societal perspectives to better capture the full economic impact of GLP‐1RAs.

It is also worth exploring the potential cost‐effectiveness of GLP‐1RAs in specific subgroups of the obese population. For instance, individuals with severe obesity (BMI ≥40 or BMI ≥35 with comorbidities) or those with obesity‐related conditions such as hypertension or sleep apnoea may achieve more substantial health improvements from weight loss, potentially making these medications more cost‐effective for them. However, due to the limited number of studies and the lack of reported information, we were unable to perform subgroup analyses to examine this possibility in greater detail.

A specific model condition, as demonstrated in a study by Kim et al.,28 found semaglutide to be significantly cost‐effective compared to no treatment over a time horizon exceeding 10 years. This industry‐sponsored study applied several assumptions or methodologies that may have influenced the results. For example, it evaluated 2 years of semaglutide use over a 30‐year study time horizon that incorporated diabetes‐related health states. The 2‐year treatment duration was based on real‐world data showing most patients discontinue anti‐obesity medications (AOMs) within that period. The model assumed weight rebound after discontinuation until BMI returns to baseline, followed by natural weight gain. Although this reflects a plausible treatment pattern, the favourable cost‐effectiveness observed, particularly in the long‐term comparison of semaglutide versus no treatment, may be more strongly influenced by methodological factors such as time horizon, treatment duration, discontinuation patterns, and model structure, rather than reflecting true subgroup effects or biological differences. While Kim et al.'s study was assessed as low risk of bias for sponsorship using the ECOBIAS checklist, its unique methodological approach may have contributed to heterogeneity and influenced the pooled findings.

Furthermore, many included models assumed long‐term or lifelong adherence to GLP‐1RA therapy, which may not reflect real‐world usage patterns. In practice, discontinuation rates are high due to factors such as cost, side effects, and patient preferences.39, 40 These real‐world adherence issues may affect the actual effectiveness and economic value of GLP‐1RAs, particularly if early discontinuation limits the duration of health benefits.41 Incorporating these considerations is essential for improving the generalisability and accuracy of cost‐effectiveness estimates.

Our study identifies methodological variability as a significant challenge in economic evaluations of GLP‐1RAs. The included studies employed diverse modelling techniques, input parameters, and time horizons, contributing to high heterogeneity in some comparisons. Other potential sources of heterogeneity include variations in target populations (adults versus adolescents), assumptions regarding treatment duration, and the inclusion of diabetes‐related health states in some models but not others. Importantly, findings from sensitivity and subgroup analyses indicated that heterogeneity was reduced when studies with similar characteristics, such as comparable time horizons, age groups, or treatment duration, were analysed within the same subgroup. This suggests that heterogeneity in certain comparisons arose from structural and contextual variations rather than random error. Accordingly, while the pooled INB estimates provide useful summary measures of economic value, their reliability and generalisability depend on the degree to which decision contexts align with the modelling assumptions and structural characteristics of the included studies.

This variability reflects the broader challenge in generalising findings from individual studies to inform health policy and reimbursement decisions. For example, the study by Atlas et al.25 reported a wide 95% confidence interval for the INB in certain comparisons, likely due to differences in model assumptions such as long‐term treatment effects, weight regain, and input parameter values. Despite this, our sensitivity analyses excluding studies lacking comprehensive parameter exploration confirmed the robustness of the pooled findings. By applying a meta‐analytic approach using INB, we synthesised results across studies and provided a more generalisable estimate of cost‐effectiveness that accommodates variation in modelling frameworks and input assumptions. These results highlight the importance of adopting standardised methodologies to improve comparability across studies. Key recommendations include consistent definitions of health states, utility measures, and cost inputs, along with robust sensitivity analyses.

Our findings suggest that while GLP‐1RAs may not be the most cost‐effective option for the general population without diabetes, they could still be valuable in specific clinical scenarios. For example, semaglutide may be a reasonable choice for patients who require long‐term treatment or for whom alternative weight management options are unsuitable or unavailable. Moreover, the American Gastroenterological Association (AGA) guidelines advocate for personalised treatment plans considering comorbidities, patient preferences, and costs.5 In this context, semaglutide and liraglutide could be considered for patients whose clinical profiles align with these criteria, despite their higher economic burden. However, discontinuation of GLP‐1RAs is a significant concern, driven by their high cost and side effects.42 These factors can reduce adherence, compromise effectiveness, and further increase the overall cost burden. Among the included studies, adherence assumptions were generally based on discontinuation rates from clinical trials.

The choice of a uniform WTP threshold of $100 000 per QALY was made to enhance comparability across the US studies. However, we acknowledge that this may not reflect thresholds used in all healthcare systems. We initially planned a subgroup analysis by income level to account for this variability, but all included studies were from high‐income countries, limiting this approach. To test the influence of threshold assumptions, we performed a sensitivity analysis using a higher WTP of $150 000 per QALY, which showed consistent results with the main analysis. Additionally, because GLP‐1RAs were generally not cost‐effective even at the $100 000 threshold, it is reasonable to infer that cost‐effectiveness would be even less favourable at the lower threshold, such as $50 000. This suggests that the findings are robust within the US contexts. However, cost‐effectiveness thresholds vary widely across countries, and may be much lower in resource‐limited settings. Therefore, interpreting these findings requires consideration of local healthcare priorities and WTP benchmarks.

Given that GLP‐1RAs were not found to be cost‐effective in high‐income countries, their economic value is likely to be less favourable in lower‐income settings, where healthcare budgets are more limited and WTP thresholds are considerably lower. In such settings, the high price of GLP‐1RAs could divert limited resources away from more cost‐effective public health interventions.

Despite methodological robustness, our study has some limitations. Most of the included studies focused exclusively on weight‐related outcomes when estimating effectiveness. While weight loss is an important endpoint in obesity treatment, GLP‐1RAs have also shown potential benefits in improving cardiovascular risk factors and metabolic parameters. The exclusion of these broader clinical outcomes may underestimate both the health benefits and the economic value of these medications. Future evaluations should incorporate a wider range of outcomes to provide a more comprehensive assessment of the value of GLP‐1RAs in obesity management.

In addition, none of the included studies adopted a societal perspective. This is an important limitation, as excluding indirect costs, such as productivity losses and patient time costs, may underestimate the broader economic value of GLP‐1RAs. For obesity treatments that may reduce work absenteeism and improve quality of life, inclusion of a societal perspective could lead to more favourable cost‐effectiveness estimates, as demonstrated in a study on diabetes where analyses from a societal perspective yielded more positive economic outcomes compared to those from a payer perspective.43 Future economic evaluations should incorporate this perspective to better reflect the full value of these therapies.

There was also a lack of reported relevant variance in some studies. To address this, we imputed or borrowed variance from similar studies under Scenario 5, as previously described,21 to retain as many relevant studies as possible in the pooled analysis. When studies with variance borrowed under Scenario 5 were excluded in sensitivity analyses, the findings were mixed. Some comparisons showed results consistent with the main analysis, while others exhibited a similar direction of effect but with greater statistical significance when Scenario 5 studies were excluded. This suggests that including studies without reported variance may have influenced the precision and statistical accuracy of the pooled estimates. Nonetheless, the overall direction of findings remained consistent across analyses, supporting the robustness of the main conclusions despite some variability in statistical significance.

Third, although there was considerable heterogeneity among the included studies, our aim was to estimate the pooled INB while acknowledging that true values may vary across settings. We addressed this variation through subgroup and sensitivity analyses and highlighted it in the discussion to support careful interpretation of the results. Lastly, another important methodological consideration in our analysis is the choice of a uniform WTP threshold. Although the original studies included in our review used varying thresholds (e.g., $50 000, $100 000, or $150 000 per QALY), we recalculated all INBs using a consistent threshold of $100 000 per QALY. This standardisation enhances comparability across studies and aligns with commonly cited decision thresholds in high‐income countries. Future studies may consider conducting threshold‐based subgroup analyses when more data are available.

In light of the observed variability and the nature of the economic outcomes analysed, an assessment of the certainty of evidence is important for contextualising our findings. Although there is no standard tool for evaluating the certainty of evidence in meta‐analyses of economic evaluations (MAEE), the GRADE framework, which was primarily developed for clinical outcomes, has limited applicability in this context. For example, the domain of imprecision requires a minimally important difference, which is not defined for economic outcomes such as INB. Based on key considerations, the certainty of our pooled evidence is unlikely to be high, mainly due to imprecision from the wide range of INBs. Indirectness is also present, as most studies were conducted in high‐income countries, particularly the United States. Although heterogeneity was observed, the direction of effect estimates was consistent, and variability was likely influenced by differences in model structures and assumptions rather than conflicting results.

Our study highlights that while GLP‐1RAs are effective for weight reduction, they are generally not cost‐effective for obesity treatment in people without diabetes compared to other interventions. This reflects the high cost of these medications, which limits their economic value despite their proven efficacy in weight reduction. However, GLP‐1RAs may be cost‐effective in certain subgroups, particularly when the benefits are evaluated over longer time horizons such as more than 10 years in comparison with no treatment. These findings offer potentially useful quantitative economic evidence to support decision making in obesity management.

Nathorn Chaiyakunapruk received consulting honoraria from Novo Nordisk. And other authors declare no conflicts of interest.

The research reported in this publication was supported (in part or in full) by the National Center for Advancing Translational Sciences of the National Institutes of Health under Award Number(s) UM1TR004409. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.