A Comprehensive Evaluation of Population-Adjusted Indirect Comparison Using Advanced Methodologies
0
조회
0
다운로드
0
인용
0
참고문헌
키워드
초록
Objectives: Estimation of comparative effectiveness is critical for health technology assessment (HTA). Although direct head-to-head randomized controlled trials represent the gold standard, such evidence is often unavailable. In these circumstances, population-adjusted indirect comparison (PAIC) methods have emerged as alternative approaches, especially when individual patient data are available for one intervention, but only aggregate data (AgD) exist for the comparator. This study aimed to implement and compare the performance of various conventional and advanced PAIC methods within a simulation framework. The objective was to provide a more comprehensive under- standing of their performance, limitation and implications for HTA decision-making. Methods: Monte Carlo simulations were conducted using the National Institute for Health and Care Excellence reference logistic model to evaluate the performance of conventional and advanced PAIC methods. A total of seven PAIC methods, including Bucher’s indirect treatment comparison (ITC), were used to estimate relative treatment effects while adjusting for effect modifiers. Performance was assessed through extensive Monte Carlo simulations, evaluating metrics such as mean estimate, bias, root mean squared error (RMSE), and coverage probability of the 95% confidence/credible intervals. Re- sults: Under the conditional simulation framework, the Bucher ITC method showed the largest bias (0.715–0.725 range), consistently estimating an effect opposite to the true value (-0.4), with very low coverage (56%–59%). Matching-adjusted indirect comparison (MAIC) methods (conventional, two- stage, max effective sample size [ESS]) provided mean estimates (-0.356 to -0.386 range) with signifi- cantly lower bias than Bucher ITC, approaching the true value with more repetitions. However, MAIC inherently resulted in ESS loss due to reweighting. The conventional simulated treatment comparison (STC) method showed highly accurate mean estimates (-0.370 to -0.392 range) and high coverage (approximately 94%), while preserving the full sample size. Simulation-enhanced STC yielded similar estimates (-0.318 to -0.382 range) but slightly higher RMSE than conventional STC. The multilevel-network meta-regression analysis, in this simulation with a limited network (2 stud- ies), showed substantial bias and failed to accurately estimate the true value, likely due to insufficient network connectivity for proper model fitting. Conclusion: This simulation study demonstrates that PAIC methods like MAIC and STC can provide substantially less biased estimates of relative treat- ment effects compared to simple AgD-based approaches like Bucher ITC. However, the inherent limitations of each PAIC method must also be carefully considered during the analysis. Continued methodological refinement and further evaluation in a variety of simulation settings are required to establish the best practices for applying PAIC in HTA.
Abstract
Objectives: Estimation of comparative effectiveness is critical for health technology assessment (HTA). Although direct head-to-head randomized controlled trials represent the gold standard, such evidence is often unavailable. In these circumstances, population-adjusted indirect comparison (PAIC) methods have emerged as alternative approaches, especially when individual patient data are available for one intervention, but only aggregate data (AgD) exist for the comparator. This study aimed to implement and compare the performance of various conventional and advanced PAIC methods within a simulation framework. The objective was to provide a more comprehensive under- standing of their performance, limitation and implications for HTA decision-making. Methods: Monte Carlo simulations were conducted using the National Institute for Health and Care Excellence reference logistic model to evaluate the performance of conventional and advanced PAIC methods. A total of seven PAIC methods, including Bucher’s indirect treatment comparison (ITC), were used to estimate relative treatment effects while adjusting for effect modifiers. Performance was assessed through extensive Monte Carlo simulations, evaluating metrics such as mean estimate, bias, root mean squared error (RMSE), and coverage probability of the 95% confidence/credible intervals. Re- sults: Under the conditional simulation framework, the Bucher ITC method showed the largest bias (0.715–0.725 range), consistently estimating an effect opposite to the true value (-0.4), with very low coverage (56%–59%). Matching-adjusted indirect comparison (MAIC) methods (conventional, two- stage, max effective sample size [ESS]) provided mean estimates (-0.356 to -0.386 range) with signifi- cantly lower bias than Bucher ITC, approaching the true value with more repetitions. However, MAIC inherently resulted in ESS loss due to reweighting. The conventional simulated treatment comparison (STC) method showed highly accurate mean estimates (-0.370 to -0.392 range) and high coverage (approximately 94%), while preserving the full sample size. Simulation-enhanced STC yielded similar estimates (-0.318 to -0.382 range) but slightly higher RMSE than conventional STC. The multilevel-network meta-regression analysis, in this simulation with a limited network (2 stud- ies), showed substantial bias and failed to accurately estimate the true value, likely due to insufficient network connectivity for proper model fitting. Conclusion: This simulation study demonstrates that PAIC methods like MAIC and STC can provide substantially less biased estimates of relative treat- ment effects compared to simple AgD-based approaches like Bucher ITC. However, the inherent limitations of each PAIC method must also be carefully considered during the analysis. Continued methodological refinement and further evaluation in a variety of simulation settings are required to establish the best practices for applying PAIC in HTA.