Economic Evaluation of Ambient AI Documentation Across Provider Specialties

• In Event: Regulation, Access, and Efficiency in Health Care Markets

Abstract

Background: Ambient AI documentation tools like DAX Copilot are rapidly adopted to reduce clinician burden and burnout. Published studies show promising signals but rely on designs that cannot support causal inference. No causal economic evaluation exists.

Setting and Design: Henry Ford Health (HFH) deployed DAX Copilot across ambulatory practices beginning February 2024. By August 2025, 410 providers across 30 specialties had adopted at different times. This staggered rollout enables causal identification using a difference-in-differences design with heterogeneity-robust estimation.

Study Aims: The study estimates whether DAX causally reduces documentation time and whether effects vary by specialty and provider type. It examines the dose-response relationship between utilization intensity and efficiency gains, evaluates whether DAX shifts evaluation and management coding and whether that shift reflects documentation quality, and estimates financial return across HFH’s payer mix and value-based care contracts.

Key Contributions: Preliminary data show large specialty variation that is operationally important and causally unstudied. An after-hours time decomposition separates note-writing burden from InBasket load. An encounter-level coding analysis with clinical audit validation provides the first rigorous test of whether coding shifts reflect documentation quality or inflation.

Significance: Findings give health system leaders, workforce planners, and payers the evidence needed to deploy ambient AI based on specialty-specific returns and evaluate financial impact with appropriate rigor.

The study tests whether ambient AI documentation benefits are distributed equitably across provider types and patient populations. Advanced practice providers and physicians may experience different efficiency gains, with direct implications for workforce equity. On the patient side, Henry Ford Health serves a racially and socioeconomically diverse population including substantial proportions of Medicaid-insured, African American, Arab American, and limited-English-proficient patients. The analysis tests whether DAX’s effects on documentation quality and coding differ across these groups, surfacing potential inequities in AI performance that health systems and policymakers need to act on.

Author

• Arjun Teotia, Michigan State University/Henry Ford health
  Presenting Author