Projects | In Progress

Advancing Diagnostic Excellence for Older Adults through Collective Intelligence and Imitation Learning (INTERLACE)

Research Areas

Principal Investigator

MD, MSHP

Gary Weissman

PAIR Center Research Team

MD, MSc

Matthew Press

PhD

Lyle Ungar

Nicholas Bishop

Alyssa Sliwa

BS, MS

Benjamin Schmid

MD, MPH

Anuradha Vyas

Topics

Overview

About 5% of adult outpatients in the US are negatively impacted by diagnostic errors, or diagnoses that are either inaccurate or untimely, resulting in increased harm to patients. Adults over 65 are at higher risk for diagnostic errors due to medical complexity, high comorbidity burden, and frailty. Diagnostic excellence seeks to improve the diagnostic process among clinicians and individuals, encourage collaboration in the diagnostic process, and reduce the potential for diagnostic errors. There is an opportunity for artificial intelligence (AI) to support diagnostic excellence for older adults in the primary care setting.

First, our team will train and test an AI model to make recommendations for diagnoses and tests based on a patient’s medical history and current symptoms. This model will use collective intelligence and imitation learning—it will recommend diagnoses and tests that are most commonly suggested by Penn clinicians for similar patients. Then, our team will conduct a clinical validation study to test the safety and appropriateness of the AI model’s recommendations. Primary care clinicians will complete retrospective chart reviews and assess a list of recommended diagnoses and tests either from the AI model or human clinician using a randomized, blinded design.

Second, our team will conduct a pilot feasibility study, embedding this AI tool into a real-world clinical workflow. Our team will integrate the AI model into an interactive interface, which will be used during primary care visits of older adult patients. We will collect both qualitative and quantitative data to assess the acceptability, feasibility, and usability of this AI model and interface in the real world and for further evaluation in a future clinical trial.

Results & Impact

Our team has trained an AI model using Penn Medicine data and completed the clinical validation study. The model has also been integrated into an interactive interface in preparation for the pilot study.

Results from the clinical validation study are currently being written up and will be updated when available.