Challenges of prolonged follow-up and temporal imbalance in pragmatic trials: Analysis of the ENCOURAGE trial

Academic Article


  • PURPOSE Peer support intervention trials are typically conducted in communitybased settings and provide generalizable results. The logistic challenges of community- based trials often result in unplanned temporal imbalances in recruitment and follow-up. When imbalances are present, as in the ENCOURAGE trial, appropriate statistical methods must be used to account for these imbalances. We present the design, conduct, and analysis of the ENCOURAGE trial as a case study of a cluster-randomized, community-based, peer-coaching intervention. METHODS Preliminary data analysis included examination of study data for imbalances in participant characteristics at baseline, the presence of both secular and seasonal trends in outcome measures, and imbalances in time from baseline to follow- up. Additional examination suggested the presence of nonlinear trends in the intervention effect. The final analyses adjusted for all identified imbalances with accounting for community clustering by supplementing linear mixed effect models with generalized additive mixed models (GAMM) to examine nonlinear trends. RESULTS Largely due to the location of participants across a considerable geographic area, temporal imbalances were discovered in recruitment, baseline, and follow-up data collection, along with evidence for both secular and seasonal trends in study outcome measures. Using the standard analytical approach, ENCOURAGE appeared to be a null trial. After incorporating adjustment for these temporal imbalances, linear regression analyses still showed no intervention effect. Upon further analyses using GAMM to consider nonlinear intervention trends, we observed intervention effects that were both significant (P <.05) and nonlinear. DISCUSSION In community-based trials, recruitment and follow-up may not occur as planned, and complex temporal imbalance may greatly influence the analysis. Real-world trials should use careful logistic planning and monitoring to avoid temporal imbalance. If imbalance is unavoidable, sophisticated statistical methods may nevertheless extract useful information, although the potential problem of residual confounding due to other unmeasured imbalances must be considered.
  • Published In

    Digital Object Identifier (doi)

    Author List

  • Richman JS; Andreae S; Safford MM
  • Start Page

  • S66
  • End Page

  • S72
  • Volume

  • 13