PERFORMANCE OF TRANSPORTABILITY METHODS UNDER EFFECT MEASURE MODIFIER OMISSION: A PLASMODE SIMULATION STUDY IN LUNG CANCER SCREENING

Author(s)

I-Hsuan Su, MS1, Jennifer L. Lund, PhD, MSPH1, Michael Webster-Clark, PhD, PharmD2;
1University of North Carolina at Chapel Hill, Department of Epidemiology, Chapel Hill, NC, USA, 2Wake Forest University School of Medicine, Division of Public Health Sciences, Winston-Salem, NC, USA
OBJECTIVES: Randomized controlled trials (RCTs) provide internally valid treatment effect estimates but may have limited generalizability to real-world populations. Transportability methods, such as inverse odds weighting (IOW) and parametric g‑computation, adjust for differences in effect measure modifiers (EMMs) between trial and target populations but rely on correct model specification. We compared the performance of IOW and parametric g‑computation when models are misspecified due to omission of influential EMMs using plasmode simulations. Unlike prior studies relying solely on fully synthetic data, our plasmode approach leveraged real‑world data to preserve realistic covariate structures.
METHODS: Plasmode simulations were conducted using trial participants from the National Lung Screening Trial and a lung cancer-screening-eligible target population from the U.S. Behavioral Risk Factor Surveillance System. The estimand was the 5‑year lung cancer mortality risk difference comparing annual screening with low‑dose computed tomography to chest radiography. The trial sampling model was specified using multivariable logistic regression, and outcomes were generated using multivariable log‑linear regression. Transported risk differences were estimated using IOW and parametric g‑computation under scenarios omitting key EMMs, including demographics, smoking history, and respiratory conditions. Performance was assessed using bias, empirical variance, 95% confidence interval (CI) coverage, and root mean squared error (RMSE). Additional scenarios varied the magnitude of sex and exposure-sex interaction effects.
RESULTS: When individual EMMs were omitted, absolute bias ranged from 0% to 0.04% for both methods. IOW exhibited slightly higher variance and RMSE than parametric g‑computation, while CI coverage was approximately 95% for both. Increasing the magnitude of sex and exposure-sex interaction effects increased absolute bias up to 0.11% for IOW and 0.13% for g‑computation, with modest variance inflation.
CONCLUSIONS: Neither method was uniformly more robust to omitted variable bias. However, strong unmeasured EMMs may substantially impair accuracy, underscoring the importance of careful EMM identification when transporting RCT findings.

Conference/Value in Health Info

2026-05, ISPOR 2026, Philadelphia, PA, USA

Value in Health, Volume 29, Issue S6

Code

MSR168

Topic

Methodological & Statistical Research

Topic Subcategory

Confounding, Selection Bias Correction, Causal Inference

Disease

SDC: Oncology

Your browser is out-of-date

ISPOR recommends that you update your browser for more security, speed and the best experience on ispor.org. Update my browser now

×