KEEPING OLDER ADULTS AT HOME - Prediction modeling for 30-day outcomes after sepsis discharge - study

Sep 28, 2026

Uploaded by Ben

Older adults often wish to remain at home and within their communities, yet sepsis survivors receiving home healthcare frequently experience readmissions or transitions to other facilities. This study identifies modifiable risk factors that can alert clinicians and support interventions to keep older sepsis patients at home. Medicare beneficiaries hospitalized for sepsis and discharged with HHC (2021–2022) were analyzed using claims and home-care assessment data, with days-not-at-home as a 30-day binary outcome.

LanguageEnglish

CategoryHealthcare

FormatPDF

Pages1

Reading Time3 min

AccessFree to download

Abstract

Study objective and data

Outcome definition and modeling approach

Cohort and key predictors

Implications for interventions

Innovation in Aging, 2025, Vol. 9, No. S2 1133 Abstract citation ID: igaf122.3800 KEEPING OLDER ADULTS AT HOME: PREDICTION MODELING FOR 30-DAY OUTCOMES AFTER SEPSIS DISCHARGE Sang Bin You1, Jiyoun Song2, Miriam Ryvicker3, Yolanda Barron3, and Kathryn Bowles2, 1. University of Pennsylvania, Philadelphia, Pennsylvania, United States, 2. University of Pennsylvania School of Nursing, Philadelphia, Pennsylvania, United States, 3. VNS Health, New York, New York, United States Many older adults prefer to remain in their homes and communities, yet sepsis survivors receiving home healthcare (HHC) often experience readmissions or transition to other facilities. This study aims to identify modifiable risk factors that can alert clinicians and trigger interventions to keep older sepsis survivors at home. We analyzed Medicare fee-for-service beneficiaries (2021-2022) hospitalized for sepsis and discharged with HHC using claims and home care assessment data. The outcome, days-not-at-home, was defined as binary indicator of whether patients spent time in hospitals, skilled nursing, or long-term facilities within 30 days. Logistic regression, random forest, neural network, and XGBoost models were developed, tuned with cross-validated grid search, and evaluated using 10-fold cross-validation. Performance was assessed using the area under the receiver operating characteristic curve (AUC) . The cohort included 19,019 older adults [mean age 78.5 years, standard deviation (SD)=8.1] with an average of 5.5 chronic conditions (SD = 2.0), mean hospital stay of 8.1 days (SD = 7.0), and nearly all (96.9%) used ≥5 medications. Within 30 days, 25.5% experienced days-not-at-home. The neural network performed best (AUC=0.6052), followed by logistic regression (AUC=0.6044), XGBoost (AUC=0 . 6037), and random forest (AUC=0 .5955) . Top predictors included receiving enteral nutrition therapy, having an ostomy, ≥2 hospitalizations in the past 6 months, positive or missing depression screening, and having pressure ulcer(s) . Although overall accuracy was modest, results suggest that addressing depression, enhancing nutrition support and providing close scrutiny of wounds may reduce risk and prevent institutional care among sepsis survivors. Findings may guide risk stratification and tailored interventions to reduce preventable utilization.

FAQs

What was the study’s main goal?

To identify modifiable risk factors that can alert clinicians and trigger interventions to help older sepsis survivors stay at home after discharge.

How was the 30-day outcome measured?

The outcome, days-not-at-home, was defined as whether patients spent time in hospitals, skilled nursing, or long-term facilities within 30 days.

Which models performed best?

The neural network achieved the highest AUC (0.6052), followed closely by logistic regression (0.6044), XGBoost (0.6037), and random forest (0.5955).

What predictors and actionable findings were highlighted?

Top predictors included enteral nutrition therapy, having an ostomy, at least two hospitalizations in the prior 6 months, depression screening being positive or missing, and pressure ulcer(s). Addressing depression, enhancing nutrition support, and closely monitoring wounds may reduce institutional care and preventable utilization.

Sep 28, 2026

Uploaded by Ben