Structured Handoff Using Intelligent Framework for Transitions Trial

The Structured Handoff Using Intelligent Framework for Transitions (SHIFT) Trial is a pragmatic, parallel-group randomized controlled trial designed to evaluate whether an artificial intelligence (AI) tool integrated into the Carelign handoff platform can reduce documentation time and burden among inpatient hospitalists. The AI feature uses a large language model (LLM) tuned for clinical summarization to generate an editable draft of the end-of-rotation handoff. Hospitalists randomized to the intervention arm will have access to a "Draft Handoff" button within Carelign, which automatically generates a structured, editable handoff draft using data from recent notes in the electronic health record. Clinicians are required to review and edit all draft content before finalizing and sharing the handoff with colleagues. Providers in the control arm will continue using standard handoff workflows within Carelign without AI assistance. The study will be conducted on general medicine services at the Hospital of the University of Pennsylvania (HUP) and Penn Presbyterian Medical Center (PPMC). Eligible participants include attending hospitalists scheduled for at least five consecutive days on service. Providers will be randomized 1:1 to the intervention or control arm, stratified by site and team. The study will enroll approximately 90 hospitalists contributing about 120 eligible rotations over a 12-week period. The primary outcome is total time (in minutes) spent drafting the end-of-rotation handoff, measured using automated Carelign audit logs. Secondary outcomes include documentation burden (modified NASA Task Load Index), work exhaustion (Stanford Professional Fulfillment Index), and self-reported drafting time. Exploratory measures include tool usability (Net Promoter Score), receiving provider ratings of handoff quality, and electronic health record (EHR) usage metrics (total and after-hours "pajama time"). The trial is unblinded and will be conducted under real-world conditions to maximize generalizability. Analyses will follow an intention-to-treat approach, using linear mixed-effects models with random intercepts for provider to account for repeated rotations. All data will be stored and analyzed in secure, HIPAA-compliant institutional environments. The study has received approval from the University of Pennsylvania Institutional Review Board and Penn Medicine's AI Governance Council.