Paper Type
Complete
Abstract
The effects of a communication LLM that automatically drafts clinician responses to patient portal messages on healthcare utilization remain unknown. We analyze a US hospital dataset with 267 hospitals adopting communication LLM and 747 not adopting in 2023 via a Difference-in-Differences (DiD) regression. Additionally, we cross-validate our results with an Instrumental Variables (IV) regression and a battery of robustness checks. Our results show that LLM led to a 12.5% reduction in the number of hospital visits and a 12.2% increase in the median dwell time. Drawing on process virtualization theory, we demonstrate that the LLM acts as a demand filter, effectively absorbing low-complexity interactions and concentrating remaining in-person visits on higher-acuity cases. This provides the first large-scale causal evidence that Electronic Health Record-integrated LLMs measurably restructure hospital utilization, carrying profound implications for capacity planning, staffing models, and reimbursement design.
Paper Number
1508
Recommended Citation
Wu, Connor; Bao, Chenzhang; and Sharda, Ramesh, "Fewer Visits, Longer Stays: How AI-Mediated Communication in Electronic Health Record Systems Transforms Healthcare Utilization" (2026). AMCIS 2026 Proceedings. 11.
https://aisel.aisnet.org/amcis2026/sigadit/sigadit/11
Fewer Visits, Longer Stays: How AI-Mediated Communication in Electronic Health Record Systems Transforms Healthcare Utilization
The effects of a communication LLM that automatically drafts clinician responses to patient portal messages on healthcare utilization remain unknown. We analyze a US hospital dataset with 267 hospitals adopting communication LLM and 747 not adopting in 2023 via a Difference-in-Differences (DiD) regression. Additionally, we cross-validate our results with an Instrumental Variables (IV) regression and a battery of robustness checks. Our results show that LLM led to a 12.5% reduction in the number of hospital visits and a 12.2% increase in the median dwell time. Drawing on process virtualization theory, we demonstrate that the LLM acts as a demand filter, effectively absorbing low-complexity interactions and concentrating remaining in-person visits on higher-acuity cases. This provides the first large-scale causal evidence that Electronic Health Record-integrated LLMs measurably restructure hospital utilization, carrying profound implications for capacity planning, staffing models, and reimbursement design.
Comments
SIG ADIT