Accuracy and Equity of the End-of-Life Care Index in Predicting 1-Year Mortality

Importance: The End-of-Life Care Index (EOLCI) was developed to predict 1-year mortality and improve communication about serious illness at the point of care. The EOLCI is widely available in US hospitals, yet external validation studies have been limited. Objective: To evaluate EOLCI model performance overall and across key patient subgroups. Design, Setting, and Participants: This prognostic study was performed from January 1, 2022, through December 31, 2023, at 39 hospitals in the Trinity Health and Kaiser Permanente Southern California (KPSC) health systems. Participants included adults (≥18 years of age) hospitalized for 36 hours or longer. Data were analyzed from September 1, 2024, through July 13, 2026. Main Outcomes and Measures: The main outcome was 1-year mortality risk as predicted by the EOLCI using a logistic regression model with inputs including age, sex, insurance, laboratory values, comorbidities, and medications. Model performance was evaluated at a risk threshold of 70% and stratified by health system, based on a use case in a large clinical trial. The evaluation used Scaled Brier Scores (SBS; range -1 to 1) for composite measures of calibration and discrimination, calibration plots, and C statistics. Patient subgroups were defined by age, sex, race and ethnicity, ethnicity alone, Medicaid status, and diagnoses. Results: Among 116 749 Trinity Health patients with 154 063 encounters (median age, 69 [IQR, 56-79] years; 79 448 [51.6%] female across all encounters), 12 054 patients (10.3%) died within 1 year. Among 94 489 KPSC patients with 133 043 encounters (median age, 70 [IQR, 57-80] years; 67 389 [50.7%] female across all encounters), 16 872 patients (17.9%) died within 1 year. The EOLCI SBS across encounters was -0.01 (95% CI, -0.03 to 0.01) at Trinity Health and 0.18 (95% CI, 0.16-0.20) at KPSC, consistent with plots indicating poor calibration in both cohorts. Model discrimination, measured by the C statistic, was 0.76 (95% CI, 0.76-0.77) at Trinity Health and 0.81 (95% CI, 0.81-0.81) at KPSC. Model performance was similar across most subgroups, but worse in the oldest subgroup and some diagnostic subgroups, depending on the health system. Conclusions and Relevance: In this prognostic study of hospitalized adults in 2 large US health systems, the EOLCI showed moderate to high discrimination but poor calibration, with reasonably equitable performance across many sociodemographic characteristics and diagnoses. These findings may inform potential opportunities and limitations for using the EOLCI in clinical settings and serve as benchmarks for health systems developing local mortality risk models.