Objective: Patient non-adherence and drop-out significantly extend the duration and cost of clinical trials. A predictive tool identifying patients likely to drop out could enhance clinical trial management through targeted and personalized engagement strategies. We aimed to develop such a tool using predictive modeling on data from two studies on schizophrenia and dry eye. 

Design: In these studies, early termination events were categorized as informed consent withdrawal, non-adherence, adverse events, and lost-to-follow-up, focusing on the first two due to their association with lack of engagement. The timing of these events was also important, as early drop-outs indicate lower engagement levels. We modeled drop-out due to informed consent withdrawal or non-adherence as a survival endpoint (time to drop out) using a multivariate Cox’s model. Baseline predictors of patient engagement, such as study site perception, belief in medicine, and health literacy, were collected via the Compl-AI questionnaire. 

Results: Separate models were constructed for each study and validated on out-of-sample patients from the other trial. The predictive performance was robust, with C-Indexes of 0.73 and 0.87 (p-values < 0.001) demonstrating a strong association between the predicted engagement score and drop-out.  

Conclusion: Informed consent withdrawal and non-adherence can be predicted at baseline, identifying patients at higher risk of dropping out. These patients are prime candidates for targeted engagement strategies to improve retention rates and optimize the efficiency of clinical trials.