COMPARISON OF SIMULATED TREATMENT COMPARISON AND MATCHING-ADJUSTED INDIRECT COMPARISON FOR LONGITUDINAL BINARY OUTCOMES WITH LOW POPULATION OVERLAP
Author(s)
Olga Mironenko, PhD1, Andrei Lazarev, MS1, Natalia Sableva, MA1, Daria Tolkacheva, MS1, Taras Khimich, PhD1, Valentina Batorova, MA1, Kirill Sapozhnikov, MD2.
1Russian Presidential Academy of National Economy and Public Administration, Moscow, Russian Federation, 2Russian Presidential Academy of National Economy and Public Administration, Moscow, Russian Federation, Moscow, Russian Federation.
1Russian Presidential Academy of National Economy and Public Administration, Moscow, Russian Federation, 2Russian Presidential Academy of National Economy and Public Administration, Moscow, Russian Federation, Moscow, Russian Federation.
OBJECTIVES: To compare the performance of simulated treatment comparison (STC) versus matching-adjusted indirect comparison (MAIC) for longitudinal binary outcomes in the presence of low population overlap between clinical trials, using netakimab (NTK) data against registered comparators in radiographic axial spondyloarthritis (r-axSpA).
METHODS: Individual patient data (IPD) from Phase III RCT ASTERA for NTK and aggregate data from 10 comparator trials were used. STCs were applied alongside MAIC to assess relative performance under low population overlap. Covariate correlations (age, sex, disease duration, HLA-B27, baseline BASDAI, ASDAS, CRP) and AIC-driven parametric distributions for continuous covariates (truncated normal/lognormal/gamma) were estimated. Logistic regressions with patient-level random intercepts were fitted to IPD on post-baseline visits in NTK arm for each endpoint (ASAS 40, ASAS 20, BASDAI 50, ASDAS-ID, ASDAS-LDA, ASDAS-MI, ASDAS-CII). Model specifications (natural cubic s`plines for time and continuous covariates, covariate×time interactions) were selected via 5-fold cross-validation. Synthetic populations (n=100,000 per comparator) were generated via NORTA algorithm with biologically constrained rejection sampling. Missing covariates used ASTERA marginal distributions. Marginal response probabilities for NTK in comparator populations were obtained by Monte Carlo integration over 1,000 random-effect draws per patient. Variability was propagated via 2,000 bootstrap iterations.
RESULTS: The NORTA-based STC approach successfully mitigated biases associated with low overlap, while MAIC suffered substantial effective sample size loss (ESS < 30% of original). Risk differences at maximum timepoints demonstrated that NTK exhibited superior efficacy compared to TNFα inhibitors, IL-17 inhibitors, and tofacitinib across key endpoints (ASAS 40, ASAS 20). STC produced more precise risk difference estimates with narrower confidence intervals compared to MAIC, confirming its superiority in settings with low population overlap.
CONCLUSIONS: STC outperforms MAIC with low population overlap, providing more precise and stable estimates without MAIC's effective sample size loss. This offers a practical and robust framework for indirect comparisons when direct head-to-head trials are unavailable or insufficiently overlapping.
METHODS: Individual patient data (IPD) from Phase III RCT ASTERA for NTK and aggregate data from 10 comparator trials were used. STCs were applied alongside MAIC to assess relative performance under low population overlap. Covariate correlations (age, sex, disease duration, HLA-B27, baseline BASDAI, ASDAS, CRP) and AIC-driven parametric distributions for continuous covariates (truncated normal/lognormal/gamma) were estimated. Logistic regressions with patient-level random intercepts were fitted to IPD on post-baseline visits in NTK arm for each endpoint (ASAS 40, ASAS 20, BASDAI 50, ASDAS-ID, ASDAS-LDA, ASDAS-MI, ASDAS-CII). Model specifications (natural cubic s`plines for time and continuous covariates, covariate×time interactions) were selected via 5-fold cross-validation. Synthetic populations (n=100,000 per comparator) were generated via NORTA algorithm with biologically constrained rejection sampling. Missing covariates used ASTERA marginal distributions. Marginal response probabilities for NTK in comparator populations were obtained by Monte Carlo integration over 1,000 random-effect draws per patient. Variability was propagated via 2,000 bootstrap iterations.
RESULTS: The NORTA-based STC approach successfully mitigated biases associated with low overlap, while MAIC suffered substantial effective sample size loss (ESS < 30% of original). Risk differences at maximum timepoints demonstrated that NTK exhibited superior efficacy compared to TNFα inhibitors, IL-17 inhibitors, and tofacitinib across key endpoints (ASAS 40, ASAS 20). STC produced more precise risk difference estimates with narrower confidence intervals compared to MAIC, confirming its superiority in settings with low population overlap.
CONCLUSIONS: STC outperforms MAIC with low population overlap, providing more precise and stable estimates without MAIC's effective sample size loss. This offers a practical and robust framework for indirect comparisons when direct head-to-head trials are unavailable or insufficiently overlapping.
Conference/Value in Health Info
2026-11, ISPOR Europe 2026, Vienna, Austria
Value in Health, Volume 29, Issue 12S
Code
MSR136
Topic
Methodological & Statistical Research, Study Approaches
Disease
Musculoskeletal Disorders (Arthritis, Bone Disorders, Osteoporosis, Other Musculoskeletal)