ASSESSING THE IMPACT OF MISSINGNESS IN TREATMENT EFFECT MODIFIERS ON BIAS IN ANCHORED AND UNANCHORED MAIC AND STC
Author(s)
Victor Laliman-Khara, MSc1, Ahmad Sofi-Mahmudi, D.D.S, MSc2, Mohammad Sazzad Hasan, PHd3, Mingxin Liu, MSc4;
1Cytel, EVA, Thompson, MB, Canada, 2Cytel, EVA, Toronto, ON, Canada, 3Cytel, EVA, Ottawa, ON, Canada, 4Cytel, EVA, Montreal, QC, Canada
1Cytel, EVA, Thompson, MB, Canada, 2Cytel, EVA, Toronto, ON, Canada, 3Cytel, EVA, Ottawa, ON, Canada, 4Cytel, EVA, Montreal, QC, Canada
OBJECTIVES: We examined how missing baseline characteristics data, outcomes and different missingness mechanisms affect the performance of population-adjusted indirect comparisons (PAIC), specifically Simulated Treatment Comparison (STC) and Matching-Adjusted Indirect Comparison (MAIC).
METHODS: We simulated two studies with patient-level baseline characteristics and outcomes, introduced three missingness mechanisms (MCAR, MAR, MNAR) at 5%, 15%, and 35% missingness, and applied anchored and unanchored STC and MAIC, alternating missingness on either baseline characteristics and outcomes. Bias was compared against analyses without missing data as benchmark. We used two types of outcomes: binary and survival outcomes. We used both absolute error vs actual estimate and 95% confidence interval coverage to summarize the findings.
RESULTS: One of the first findings is that any type of missingness is associated with bias. Higher missingness levels systematically increased bias regardless of mechanism. MAIC consistently reduced bias more effectively than STC across survival and binary outcomes. STC showed no significant advantage over MAIC in reducing the bias or improving estimation.
CONCLUSIONS: Missing data substantially impacts PAIC performance and needs to be carefully addressed to avoid introducing bias in PAIC. MAIC appears more robust than STC under varying missingness conditions, but remains consistently associated with bias in all missingness cases.
METHODS: We simulated two studies with patient-level baseline characteristics and outcomes, introduced three missingness mechanisms (MCAR, MAR, MNAR) at 5%, 15%, and 35% missingness, and applied anchored and unanchored STC and MAIC, alternating missingness on either baseline characteristics and outcomes. Bias was compared against analyses without missing data as benchmark. We used two types of outcomes: binary and survival outcomes. We used both absolute error vs actual estimate and 95% confidence interval coverage to summarize the findings.
RESULTS: One of the first findings is that any type of missingness is associated with bias. Higher missingness levels systematically increased bias regardless of mechanism. MAIC consistently reduced bias more effectively than STC across survival and binary outcomes. STC showed no significant advantage over MAIC in reducing the bias or improving estimation.
CONCLUSIONS: Missing data substantially impacts PAIC performance and needs to be carefully addressed to avoid introducing bias in PAIC. MAIC appears more robust than STC under varying missingness conditions, but remains consistently associated with bias in all missingness cases.
Conference/Value in Health Info
2026-05, ISPOR 2026, Philadelphia, PA, USA
Value in Health, Volume 29, Issue S6
Code
MSR249
Topic
Methodological & Statistical Research
Topic Subcategory
Confounding, Selection Bias Correction, Causal Inference, Missing Data
Disease
No Additional Disease & Conditions/Specialized Treatment Areas