Impact of Prognostic Notifications on Inpatient Advance Care Planning: A Cluster Randomized Trial

Area of Science:

• Medical Informatics
• Clinical Decision Support
• Palliative Care

Background:

• Advance care planning (ACP) is crucial but often triggered by poor prognosis, which clinicians may inaccurately estimate.
• Physician overestimation of prognosis can delay essential advance care planning conversations.

Purpose of the Study:

• To evaluate if notifying inpatient physicians of high patient mortality risk increases advance care planning (ACP) documentation.
• To assess the impact of machine learning-driven alerts on ACP note completion.

Main Methods:

• A pragmatic cluster randomized trial involving attending physicians on inpatient medicine teams.
• Intervention group physicians received email/page notifications for high-risk patients (30-day/6-month mortality) identified by a machine learning model.
• Primary outcome: ACP note documentation by the randomized physician; Secondary outcomes: any clinician ACP documentation, length of stay, discharge to hospice.

Main Results:

• Physicians in the intervention group were more likely to document an ACP conversation compared to the control group (34.7% vs. 19.6%).
• The machine learning model identified patients at high risk for mortality, prompting targeted interventions.
• No significant difference was observed in ACP documentation by any clinician between groups.

Conclusions:

• Machine learning-based mortality risk notifications effectively prompt physicians to document advance care planning conversations during hospitalization.
• This technology can serve as a valuable tool to enhance end-of-life care discussions.