Projecting effectiveness after ending a randomized controlled trial: a two-state Markov microsimulation model
Journal Articles
Overview
Research
Identity
Additional Document Info
View All
Overview
abstract
ObjectiveTo investigate the behavior of restricted mean survival time (RMST) and designs of a two-state Markov microsimulation model through a 2 × 4 × 2 full factorial experiment.MethodBy projecting patient-wise 15-year-post-trial survival, we estimated life-year-gained between an intervention and a control group using data from the Cardiovascular Outcomes for People Using Anticoagulation Strategies Study (COMPASS). Projections considered either in-trial events or post-trial medications. They were compared based on three factors: (i) choice of probability of death, (ii) lengths of cycle, and (iii) usage of half-a-cycle age correction. Three-way analysis of variance and post-hoc Tukey's Honest Significant Difference test compared means among factors.ResultsWhen both in-trial events and post-trial study medications were considered, monthly, quarterly, or semiannually were not different from one other in projected life-year-gained. However, the annual one was different from the others: mean and 95 percent confidence interval 252.2 (190.5–313.9) days monthly, 251.8 (192.0–311.6) quarterly, 249.1 (189.7–308.5) semiannually, and 240.8 (178.5–303.1) annually. The other two factors also impacted life-year-gained: background probability (269.1 [260.3–277.9] days projected with REACH-based-probabilities, 227.7 [212.6–242.8] with a USA life table); half-a-cycle age correction (245.5 [199.0–292] with correction and 251.4 [209.1–293.7] without correction). When not considering post-trial medications, only the choice of probability of death appeared to impact life-year-gained.ConclusionFor a large trial or cohort, to optimally project life-year-gained, one should consider using (i) annual projections, (ii) life table probabilities, (iii) in-trial events, and (iv) post-trial medication use.
