A Novel Artificial Intelligence Platform to Automate Clinical Consultation Notes and Enhance Diagnostic Efficiency in the Outpatient Clinic: A Multi-Center, Multi-Disciplinary, Prospective Randomized Controlled Trial

Presented herein is a proposal for a protocol for a multi-center, multi-disciplinary randomized controlled trial (RCT) to evaluate a novel artificial intelligence (AI)-based technology that automates the construction of the clinical consultation note (CCN) and enhances diagnostic assessments in the outpatient clinic setting. This innovative tech-platform automatically generates the CCN and presents it to the provider in advance of the patient consultation, without any work done by the provider. The constructed CCN is presented either in the native electronic health record (EHR) or in a secure web-based application, in a HIPAA-compliant manner. The proposed prospective prospective trial will compare this novel AI/ML technology (NAMT) versus the current standard-of-care (SOC) in the outpatient setting. Outpatient clinic-days will be randomized to either "SOC clinic-day" or the "NAMT clinic-day" based on whether the SOC or the NAMT was used to construct the CCN for all patients seen on that particular clinic-day. Randomized cross-over of each provider between "SOC clinic-day" and "NAMT clinic-day" will result in each provider serving as her/his own internal control. Objective data will be used to compare study endpoints between the SOC and the NAMT. Co-primary endpoints include a) CCN diagnostic accuracy/quality (based on standardized QNOTE metrics); and b) Work-outside-work (WOW) time required by providers to complete clinic-related documentation tasks outside clinic hours (based on EHR meta-data). Secondary endpoints include a) Provider productivity (based on provider "walk-in, walk-out time from the consultation room); b) Provider satisfaction (based on the standardized AHRQ EHR End User Survey); and c) Patient satisfaction (based on the standardized Press Ganey/CG-CAHPS survey). To assess generalizability across the health-care spectrum, the study will be conducted in four different types of health-care settings (large academic medical center; non-academic hospital; rural hospital; community private practice); in four different disciplines (cardiology; infectious disease; urology; emergency medicine); using four different EHR systems (Cerner; Epic; AllScripts; MediTech/UroChart). We estimate an aggregate RCT sample size of 150 clinic-days (involving 3,000 total patients; 15-30 providers). This will randomize 75 clinic-days (1,500 patients) to the control SOC arm, and 75 clinic-days (1,500 patients) to the intervention NAMT arm. We will use a two-sided Z-test of difference between proportions with 90% power and two-sided 5% significance level. This RCT is the first to evaluate the efficiency and diagnostic accuracy of pre-constructing CCNs in an automated manner using AI/ML technology, deployed at a large-scale, multi-institutional, multi-disciplinary, multi-EHR level. Results from this study will provide definitive level 1 evidence about the desirability and generalizability of AI-generated automatically constructed CCNs, assessing its potential benefits for providers, patients, and healthcare systems.