Abstract / Summary
We developed and externally validated a model to predict surgical site infections (SSIs) in older adults undergoing non-cardiac surgery and used latent class analysis (LCA) to characterize clinical heterogeneity. The prediction model was developed using data from older adult patients who underwent surgery at Chinese PLA General Hospital (Hospital A) and was subsequently externally validated using data from Henan Provincial People's Hospital (Hospital B). Multivariable logistic regression was used to develop the prediction model. LCA was conducted separately in both hospitals to identify descriptive clinical profiles and assess their cross-center reproducibility. Among the participants from the two tertiary hospitals, the SSIs incidence rates were 2.64% and 5.35%, respectively. Ten predictors were retained in the final multivariable prediction model. The area under the receiver operating characteristic curves (AUROCs) were 0.785, 0.777, and 0.756 in the derivation, internal validation and external validation cohorts, respectively. A four-class solution was selected in both hospitals, although classification certainty was lower in Hospital B. The classes represented distinct combinations of patient and perioperative characteristics, with Class 4 showing the highest SSIs incidence in both hospitals. The prediction model maintained moderate discrimination across the two centers, and LCA identified clinically distinct descriptive profiles with partial cross-center reproducibility. These findings characterize heterogeneity among older surgical patients but require prospective validation before clinical implementation.