Randomized trial of electronic health record implemented AI risk prediction in kidney transplant care

Abstract Artificial intelligence (AI)-based risk prediction is increasingly implemented in clinical care, but randomized evidence on communication and shared decision-making (SDM) outcomes is limited. In the single-center PRIMA-AI trial, 76 kidney transplant recipients with estimated glomerular filtration rate <30 mL/min/1.73 m² were randomized 1:1 to usual care or usual care plus an electronic health record (EHR)-integrated machine-learning model predicting 1-year graft loss risk. The primary outcome was patient-reported conversations about treatment options after graft loss during 12 months. Conversation frequency did not differ between groups (intervention 14/36 39% vs control 16/40 40%; chi-square p = 1.00). No significant between-group differences were observed for secondary clinical, SDM-related, relationship, or distress outcomes. Post-study user feedback suggested low and variable tool uptake with workflow barriers. Passive EHR availability of AI risk estimates did not improve communication or SDM-related outcomes. Future interventions should strengthen workflow integration and directly support SDM. Trial Registration: ClinicalTrials.gov number, NCT0605651, registered 2023-09-21.