Menu Back to Poster-Presentations-Details

P126: Impact of Lost to Follow-Up on Power and Study Results in Randomized Controlled Studies with Time-to-Event Endpoints

Poster Presenter

Jiewei Zeng

Associate Director
Abbvie
United States

Objectives

Lost to follow-up (LTFU) poses significant challenges in randomized controlled studies with time-to-event endpoints. The objective of the research work is to assess the impact of LTFU on power, Type 1 error and study results.

Method

Simulations using a sample size of 200 subjects, with 1:1 ratio to treatment and control arms. Target number of events is 135 with event times generated using exponential distribution with rate parameter of 0.2. Different LTFU patterns (timing and rates between the arms) have been assessed.

Results

Our findings highlight the critical impact of the timing and proportion of LTFU in both arms on the study's power, Type 1 error, and study results. The study’s power can be highly affected by the LTFU patterns (timing and rates) of the two arms. Compared with the power based on a total of 135 events and the underlying treatment effects (hazard ratio), the simulation shows that the power significantly decreases when the treatment arm experiences a higher early LTFU rate compared to the control arm, and increases when the treatment arm exhibits a lower early LTFU rate than the control arm. Among all the scenarios involving different LTFU patterns, the lowest power is observed when all LTFU cases happen early in the treatment arm, while all subjects in the control arm were followed up adequately until cutoff time. Similar results are observed in the estimated hazard ratios. When there is a higher rate of early LTFU in the treatment arm, the hazard ratio is underestimated compared to the underlying hazard ratio. In contrast, when there is a higher rate of early LTFU in the control arm, the hazard ratio is overestimated. Type I error is maintained within 0.05 when all subjects in both arms are followed until the cutoff time. Based on simulation outcomes, Type 1 error is inflated when there is a higher rate of early LTFU in the control arm and lower rate in the treatment arm. The largest Type I error inflation occurs when all subjects in the treatment arm are followed until the cutoff time, while all LTFU cases occur early in the control arm. This observation suggests that proper and proactive management of early LTFU cases is crucial to control Type I error. The impact of LTFU patterns on power, Type 1 error and estimated HR is also evaluated under different enrollment speeds. Consistent impact for each of the LTFU patterns is observed on power, Type I error and estimated HR across different enrollment speeds.

Conclusion

The impact of LTFU on power, Type 1 error and study results in studies with time-to-event endpoints cannot be overstated. Addressing LTFU is of paramount importance to ensure the validity and reliability of study results. By implementing appropriate strategies, such as thoughtful study design, comprehensive staff training, timely data quality control procedures, and effective subject retention mechanisms, LTFU could be minimized to ensure adequate follow up time on every subject for robust analyses of time to event endpoints. This simulation study underscores the importance of proactively managing LTFU in randomized controlled studies with time-to-event endpoints. By addressing LTFU effectively, researchers can bolster the study's power, accurately estimate treatment effects, increase the likelihood of demonstrating the efficacy of treatment arm, and minimize the risk of Type I error inflation. Comprehensive efforts to prevent LTFU, particularly early LTFU, can lead to more robust and impactful clinical study outcomes, thereby advancing evidence-based healthcare practices.