A prevalence association model for risk factors associated with pressure injuries in older hospitalized patients in China: Development and generalizability assessment.

Purpose

The purpose of this study is to provide a decision-making basis for skin care of older hospitalized patients through a penalized statistical prevalence association model.

Design

We reviewed nine eligible institutions as multi-center research centers in China. SUBJECTS AND

Setting

We conducted a multicenter cross-sectional study on July 17, 2023, from which 4851 valid questionnaires were collected.

Methods

LASSO regression with 10‑fold cross‑validation was first used to select candidate variables. Variables with a selection frequency >0.8 after 1000 bootstrap resampling steps were retained as stable identify. To address rare‑events bias, Firth's penalized logistic regression was then applied. Model discrimination was assessed using the ROC curve, AUC, and C‑index with 95% CI via bootstrap. Calibration was evaluated by calibration plots, slope, intercept, and the Brier score. A leave-one-center-out generalizability assessment was performed to evaluate model applicability across centers. Clinical utility was examined using decision curve analysis (DCA).

Results

Seven variables (age, albumin, hemoglobin, diabetes, painkillers, antipsychotics, Braden score were identified. The model demonstrated excellent discrimination (AUC = 0.916, 95% CI [0.892,0.916]; C-index = 0.916 [0.890,0.939]; sensitivity 0.91, specificity 0.80) and good calibration (slope 1.035, intercept -0.001, Brier score 0.018). In the leave-one-center-out generalizability assessment, the model demonstrated a mean cross-center AUC of 0.905 [0.835-0.967]. Decision curve analysis confirmed clinical utility within the 0-15% threshold probability range.

Conclusions

This large-sample multi-center prevalence association model based on penalization and resampling techniques provides an accurate, stable, and clinically applicable tool for individualized pressure injury risk assessment in hospitalized older patients.