Artificial Intelligence empowered recruitment for clinical trials

BACKGROUND: Recruitment of clinical drug trials for Alzheimer's disease (AD) is a lengthy process with an ultimately high screening failure rate, as current research focuses on prodromal stages. We evaluate the option of pre-screening potential participants for AD trials via an automated phone-based assessment using speech analysis.

METHOD: 140 participants were recruited at the memory clinic in Maastricht as part of the MUMC+ study (SCI, MCI, ADD). They underwent a in-person baseline assessment (0M) at the clinic. Cognitive assessments were performed and speech was recorded during each assessment using the Delta platform. For the next assessments (6M), participants were contacted via telephone by a trained research nurse. Cognitive assessments were performed on the phone and speech was recorded during each assessment. Speech from each assessment point was analysed automatically to extract relevant features. Machine learning models to predict disease status, Clinical Dementia Rating Scale (CDR) scores, and Mini-Mental-Status-Examination scores were trained and evaluated on the data.

RESULT: Models based on speech features extracted from the phone assessment were able to predict disease status with an AUC of 0.93±0.06, CDR score with a Mean Absolute Error (MAE) of 1.9±0.8, and MMSE score with an MAE of 2.3±1.1. Adding longitudinal data from baseline assessments increased accuracy across all models.

CONCLUSION: Automated pre-screening through speech analysis could be an effective tool to increase the efficiency and effectiveness of recruitment for AD drug trials.