Index overestimates one-year mortality risk

The End-of-Life Care Index ranked mortality risk reasonably well but overestimated absolute risk in hospitalized adults.
*Retrospective prognostic validation cohort; Level 2 (OCEBM).

Citation

Kohn R, Courtright KR, Grau-Sepulveda M, et al. Accuracy and Equity of the End-of-Life Care Index in Predicting 1-Year Mortality. JAMA Network Open. 2026;9(9):e2633038. doi:10.1001/jamanetworkopen.2026.33038

Background

Hospitals increasingly use electronic risk tools to identify patients who may benefit from serious illness conversations or palliative care. This study tested whether a widely available commercial tool accurately predicted death within 1 year across two large health systems.

Patients

Adults hospitalized at least 36 hours in 39 United States hospitals during 2022. Exclusions included psychiatry, rehabilitation, hospice, obstetrics, and neonatology admissions.

Intervention

End-of-Life Care Index prediction using electronic health record data at about 36 hours after admission.

Control

Observed vital status in each health system.

Outcome

Predicted versus observed 1-year mortality, including calibration and discrimination.

Follow-up Period

1 year after the 36th hospital hour.

Results

Health system Encounters Observed 1-year deaths Overall performance Discrimination At 70% risk threshold
Trinity Health 154,063 10.3% Scaled Brier score −0.01 (95% CI −0.03 to 0.01); poor calibration C statistic 0.76 (95% CI 0.76 to 0.77) Positive predictive value 0.45; sensitivity 0.13
Kaiser Permanente Southern California 133,043 17.9% Scaled Brier score 0.18 (95% CI 0.16 to 0.20); poor high-risk calibration C statistic 0.81 (95% CI 0.80 to 0.81) Positive predictive value 0.62; sensitivity 0.27

C statistic: ability to rank higher-risk patients above lower-risk patients. Scaled Brier score: combined accuracy measure.

Performance was broadly similar across sex, race and ethnicity, insurance, and diagnoses, but was worse in the oldest patients and some diagnostic groups.

Limitations

Deaths outside the health system may have been missed, making the tool appear to overpredict mortality. Race and ethnicity data may be inaccurate. The model was not recalibrated locally, and some subgroup samples were small.

Funding

Patient-Centered Outcomes Research Institute; no funder role reported.

Clinical Application

Use only after local validation; it may flag higher-risk patients but should not guide prognosis or hospice eligibility alone.