Prognostic Targeting Improves Statistical Power and Efficiency in Randomized Controlled Trials in Upper Extremity Stroke Rehabilitation.

Abstract

Introduction: Randomized Controlled Trials (RCTs) are essential to underpin the superiority of novel interventions affecting upper extremity capacity post-stroke. However, many RCTs are underpowered, due to heterogeneity in recovery. Prognostic targeting may help reduce sample sizes while maintaining sufficient power.

Objective: This study investigates the effects of prognostic targeting on the required sample size to achieve 70% to 90% power in early post-stroke RCTs with upper extremity capacity measured with the Action Research Arm Test (ARAT) as the outcome.

Patients and methods: Serial data from 4 prospective cohort studies (N = 372 stroke patients) were pooled, with assessments from week 1 to 6 months post-stroke. Using this dataset, we generated synthetic 6-month ARAT outcomes and analyzed data cross-sectionally and longitudinally, with and without prognostic targeting based on a pre-existing prognostic model predicting 6-month outcome. We then calculated power for different sample sizes and assessed trial efficiency, determined by the estimated sample size and inclusion rate.

Results: Prognostic targeting within 3 weeks post-stroke theoretically reduced the required sample size by up to 56% and improved trial efficiency by 40 to 45% for detecting a 6-point ARAT difference at 6 months. The targeted trials needed 220, 270, and 360 patients vs. 470, 560, and 820 in non-targeted trials for 70% to 90% power. Benefits persisted in longitudinal analyses.

Conclusion: This study demonstrates the benefits of prognostic targeting for improving power and efficiency in early post-stroke upper extremity trials using ARAT as outcome. We strongly recommend its use in future stroke rehabilitation and recovery studies.