Senior Biostatistician Komodo Health, United States
Background: A combination product of sodium phenylbutyrate and taurursodiol (SP-T) was FDA-approved for amyotrophic lateral sclerosis (ALS) treatment in September 2022, then later withdrawn in April 2024, following unfavorable Phase 3 trial results. For rare diseases such as ALS, it is particularly difficult to recruit an adequate number of patients to participate in clinical trials. One approach to study the effectiveness of medical interventions with a small sample size is to emulate a sequential target trial, using observational data.
Objectives: To estimate the effect of SP-T initiation on time-to-death by emulating a sequential target trial.
Methods: Using the Komodo Research Dataset, a claims database with mortality data, we implemented the target trial framework to emulate a sequence of hypothetical randomized trials comparing SP-T initiators vs non-initiators (standard of care [riluzole or edaravone]) in each calendar month from October 2022 to March 2024 to estimate the observational analog of the intention-to-treat effect. Initiators in a given emulation were considered ineligible in subsequent emulation months. We used optimal full matching to account for confounding with 44 pre-index demographic and clinical characteristics as covariates in our propensity score model and exact matching on emulation month. Individuals in each emulation were followed from index date to whichever occurred first: death, end of continuous enrollment, or October 31, 2024. Data were pooled across all emulations to estimate survival time (via Kaplan-Meier estimator) and the effect on time-to-death (via Cox proportional hazards model). Confidence intervals (CI) were estimated via bootstrapping since unique individuals could contribute to multiple emulations.
Results: Among 709/10,693 respective non-unique SP-T initiators and non-initiators (3,117 unique patients), there was adequate post-matching balance (median, 62/63 years old; 40%/44% women, 64%/60% white, 60%/54% with commercial insurance). After incorporating matching weights, we observed 2,570 non-unique deaths (119/2,451 [initiator/non-initiator]) over 788 days. The estimated effect of SP-T initiation on time-to-death was 0.70 (hazard ratio [95% CI: 0.50, 0.83]).
Conclusions: We utilized a sequential target trial emulation to improve statistical efficiency in estimating the effect of SP-T initiation on time-to-death. Our results suggest a lower risk for death for ALS patients who initiated SP-T compared with standard of care.
