SHM Converge Scientific Abstract ePoster Gallery-Impact of a Machine Learning-Driven Smartalert on Repetitive Inpatient Lab Ordering

Pdf Summary

This study evaluated whether a machine learning–driven “SmartAlert” could reduce repetitive inpatient lab testing, especially complete blood counts (CBCs), which are often unnecessary. The system used electronic health record data and a probabilistic regression model trained on 7 years of inpatient data to estimate whether a patient’s next CBC would fall within a clinician-defined “stable” range. To build the alert, researchers interviewed 20 inpatient clinicians to define stability thresholds and preferred alert design.<br /><br />The CBC-focused SmartAlert was deployed in randomized fashion to general inpatients, excluding ICU and bone marrow transplant patients. From July 2024 to November 2025, the model screened 9,537 patients and generated 311,012 predictions. About 74% of predictions suggested stability. When the model predicted stability, the next CBC within 28 hours was indeed stable 87% of the time, indicating strong positive predictive value.<br /><br />The alert appeared in two formats: a silent notification and an interruptive display. The interruptive alert was shown 213 times to 96 clinicians for 146 patients, while the silent alert triggered 998 times for 489 patients. A subsequent CBC within 28 hours occurred less often after displayed alerts than after silent alerts (49% vs 66%). Overall, displaying the SmartAlert was associated with a 29% reduction in repeat CBC testing within 28 hours.<br /><br />Clinicians generally responded positively to the alert. The most common reason for dismissing it was perceived medical necessity, with other reasons including peri-procedural testing, medications affecting blood counts, symptom monitoring, and concern that labs were unstable. The team plans to refine the interface, expand the tool to BMPs, and continue monitoring performance for safety, accuracy, and fairness.

Asset Subtitle

April S. Liang

Meta Tag

Author List Aakash Acharya, April S. Liang, Christopher Sharp, Conor K. Corbin, David Mui, David Svec, Fatemeh Amrollahi, Grace Kim, John McKeown, Jonathan H. Chen, Lisa Shieh, Soumya Punnathanam, Sreedevi Mony, Stephen P. Ma, Tho D. Pham, Trevor Crowell, Yixing

Category Research

Concept Inpatient Laboratory Testing

Concept SmartAlert

Concept Lab Stability

Concept Blood Count

Concept Chemistry Panel

Distinguished Finalist

Presenter Organization Stanford University School of Medicine

Presenting Author April S. Liang

Track Value in Hospital Medicine

Keywords

machine learning

SmartAlert

inpatient lab testing

complete blood count

CBC

electronic health record

probabilistic regression model

clinical alert

repeat testing reduction

randomized deployment

Inpatient Laboratory Testing

SmartAlert

Lab Stability

Blood Count

Chemistry Panel