Monitoring in clinical trials: Identifying poor performance at recruitment sites

This Data Study Group (DSG) Challenge aimed to build a human-in-the-loop machine learning model that could assist the Medical Research Council Clinical Trials Unit (MRC CTU) at University College London, in identifying clinical trial sites that are performing poorly and require monitoring intervention. When running clinical trials, patient safety is the greatest priority, with robust operation a secondary but nevertheless key feature in ensuring successful outcome of the trial. Poor performance therefore relates to any trial conduct that could compromise safety or lead to variation in how a trial is conducted between sites, making monitoring trials a large and complex challenge.

The CTU is responsible for ensuring that all clinical trial sites gather quality data, while also protecting patient rights and well-being. Currently risk-based monitoring is used to evaluate whether individual sites are following trial protocol. This approach relies on pre-defined metrics, such as those measuring the correct and timely completion of trial forms, as indicators of site performance and uses a trigger system based on those indicators to alert the CTU when a site visit is considered necessary. However, a recent study found that the intuitively developed pre-defined metrics were not sufficient at discriminating sites requiring a site visit, where relevant features may have been missed.

The purpose of this DSG challenge was therefore to undertake an initial investigation into whether machine learning approaches can improve identification of poorly performing sites.

Citation information

Data Study Group team. (2022, September 13). Data Study Group Final Report: MRC Clinical Trials Unit, UCL. Zenodo. https://doi.org/10.5281/zenodo.7075637

Additional information

PI: Louise Coutts