An Updated Mortality and Community Discharge Prognostic Model for Older Adults Admitted to Skilled Nursing Facilities for Post-Acute Care

OBJECTIVES: We previously developed a multi-outcome prognostic model for older adults admitted to skilled nursing facilities (SNFs) for short-term rehab using Medicare data. However, incorporating predictors from the Minimum Data Set (MDS), a mandated comprehensive assessment, may improve model performance. This study sought to develop an updated model with MDS elements for use on day 7 of SNF admission when clinical trajectories are more established. DESIGN: Retrospective cohort study. SETTING AND PARTICIPANTS: Twenty percent national sample of community-dwelling Medicare Fee-for-Service beneficiaries aged ≥66 admitted to an SNF for at least 7 days following a hospitalization between 2017 and 2019. METHODS: We predicted 2 outcomes: 6-month mortality and successful community discharge (community discharge without rehospitalization or death in the subsequent 30 days). For model development, we started with predictors from our published Medicare-based model (age, sex, Medicaid status, discharge diagnosis, hospital length of stay, admission type, comorbidities, prior hospitalizations), used Least Absolute Shrinkage and Selection Operator (LASSO) on MDS elements for variable selection, and performed logistic regression to determine predictor coefficients. Model performance was assessed by concordance statistics (c-statistics), calibration plots, and decision curve analysis. RESULTS: The cohort included 426,680 individuals [mean age 81.3 years (SD = 8.3), 62.7% female, 7.9% Black]. Overall, 19.9% died within 6 months, and 57.6% experienced a successful community discharge. The updated MDS model, which included Medicare predictors and 6 MDS items (activities of daily living score, cognitive status, urinary incontinence, bowel incontinence, oxygen use, walking balance), showed improvements over the Medicare model in discrimination [bootstrapped optimism-corrected c-statistic of 0.789 (95% CI, 0.787-0.790) vs 0.747 (95% CI, 0.745-0.749) for 6-month mortality and 0.730 (95% CI, 0.728-0.731) vs 0.685 (95% CI, 0.683-0.687) for successful community discharge, respectively], net benefit, and fraction of new information. Models showed good calibration. CONCLUSIONS AND IMPLICATIONS: Incorporating MDS data from the first 7 days of SNF admission improved the accuracy of predictions of 6-month mortality and successful community discharge.