Prescription time-distribution matching and immortal time bias: promise, failure modes, and time-aware alternatives

Purpose of review Immortal time bias remains common in observational studies of treatments started after cohort entry, yet its source is often mischaracterized as the immortal interval rather than the design choices that misalign eligibility, assignment, and follow-up. This review is timely because prescription time-distribution matching (PTDM) is still used as a correction despite recent mathematical, simulation, and applied critiques. Recent findings Immortal time bias arises through misclassification or selection. PTDM can remove visible misclassified person-time, but its exclusion rule can drop controls whose pseudo-treatment dates exceed follow-up, producing left truncation, loss of representativeness, and even reversal of estimated effects. Iterative time-distribution matching reduces this exclusion but can retain residual left-truncation artifacts. Time-dependent Cox models, sequential designs, clone-censor-weight emulation, nested case–control sampling, and competing-risk estimators provide more transparent, time-aware alternatives. Summary PTDM should not be a default primary analysis. Studies should declare time zero, report grace periods and excluded person-time, and, where longitudinal data permit, benchmark PTDM-type analyses against time-varying or target-trial approaches. No single method dominates: relative performance depends on the data structure, the longitudinal information available, the target estimand, and each method's assumptions.