Maximizing Interpretability and Cost-Effectiveness of Surgical Site Infection (SSI) Predictive Models Using...

Area of Science:

• Medical Informatics
• Computational Biology
• Predictive Analytics

Background:

• Surgical site infections (SSI) pose a significant challenge in healthcare, impacting patient outcomes and increasing costs.
• Accurate prediction of SSI risk is crucial for effective prevention strategies.
• Traditional methods often rely on manual feature engineering, which can be complex and time-consuming, especially with temporal data.

Purpose of the Study:

• To develop and evaluate a novel, interpretable, and cost-effective approach for classifying surgical site infection (SSI) risk.
• To leverage temporal patient blood test data for enhanced SSI prediction.
• To integrate prior knowledge, such as blood test costs, into the predictive model.

Main Methods:

• Abstraction of temporal data into three distinct temporal windows.
• Application of penalized logistic regression with maximum likelihood L1-norm (lasso) regularization.
• Integration of prior knowledge through penalty factors based on blood test prices and an early stopping parameter to limit feature selection.

Main Results:

• The baseline C-reactive protein (CRP) classifier achieved a mean Area Under the Curve (AUC) of 0.801.
• The best full lasso model demonstrated a mean AUC of 0.956.
• The optimal lasso model, limited to 20 features, achieved a testing AUC of 0.967, outperforming the baseline significantly.

Conclusions:

• The proposed novel approach offers a highly accurate and interpretable method for SSI prediction.
• The models have the potential to aid domain experts in decision-making and improve preoperative SSI surveillance.
• This cost-effective predictive modeling could inform new guidelines for SSI prevention.