Advanced Care Planning for Hospitalized Patients Following Clinician Notification of Patient Mortality by a Machine...

Area of Science:

• Health Services Research
• Medical Informatics
• Geriatrics

Background:

• Goal-concordant care is challenging in hospitals.
• Identifying high-risk patients can prompt crucial serious illness conversations and documentation of patient goals.

Purpose of the Study:

• To assess the impact of a machine learning mortality prediction algorithm on goals of care discussions (GOCDs) in a community hospital setting.
• To examine if physician notification of high mortality risk influences the frequency and timing of GOCDs.

Main Methods:

• A cohort study compared inpatients with high 30-day mortality risk identified by a machine learning algorithm.
• The intervention group received physician notifications, while the control group did not.
• Propensity-score matching and difference-in-difference analysis were used to compare documented GOCDs before and after the intervention.

Main Results:

• Patients in the intervention group were 5 times more likely to have documented GOCDs by discharge compared to controls.
• GOCDs occurred significantly earlier in the intervention group (median 4 days vs. 16 days).
• Findings were consistent across racial subgroups.

Conclusions:

• Physician awareness of machine learning-derived high-risk predictions significantly increases documented GOCDs.
• This intervention facilitates earlier and more frequent goals of care discussions for high-risk patients.
• Further external validation is recommended to assess broader applicability.