Predicting unplanned hospital revisits among community-dwelling older adults: a dynamic cohort study

Unplanned hospital revisits (UHR) among older adults are common and contribute to adverse clinical outcomes, caregiver burden and increased healthcare costs. We aimed to develop and validate a risk prediction model for UHR in older adults to support early identification.

We conducted a retrospective cohort study, following the TRIPOD statement, using a Flemish linked database combining primary care and national health insurance data. Adults aged ≥75 years with an all-cause hospital admission in 2014 were included. The primary outcome was UHR, defined as emergency department visits or unplanned hospital admissions within 6 months post-discharge. We used multivariable logistic regression to identify predictors for UHR and develop a risk prediction model. Model performance was assessed using balanced accuracy. Missing data were handled using multiple imputation by chained equations. The model was validated on a held-out test set and a k-nearest neighbour classifier was used to cross-validate risk categories.

Among 3133 patients, 309 (10%) experienced UHR. The best-performing model had a balanced accuracy of 0.56, with a sensitivity of 58% and a specificity of 54%. Predictors were polypharmacy, male sex, haemoglobin level, number of general practitioner contacts and multimorbidity. Excessive polypharmacy (>9 medications) was associated with a 55% increase in UHR odds. Three UHR risk groups were identified: low-risk (5.1%), medium-risk (8.8%) and high-risk (11.6%).

UHR are common in older adults, with excessive polypharmacy emerging as a key predictor. The pragmatic model described here provides a valuable tool to stratify older adults into distinct risk groups, identifying a high-risk group that may benefit from targeted interventions.