Predicting hospital-acquired infections by scoring system with simple parameters

Area of Science:

• Medical Informatics
• Epidemiology
• Clinical Prediction Models

Background:

• Hospital-acquired infections (HAI) significantly increase patient morbidity, mortality, and healthcare costs.
• Developing reliable prediction models for HAI is crucial for effective clinical management.
• Existing methods may lack simplicity or broad applicability in diverse healthcare settings.

Purpose of the Study:

• To develop and validate a simple scoring system for predicting HAI risk.
• To utilize Logistic Regression (LR) for model derivation and Artificial Neural Networks (ANN) for validation.
• To identify key clinical parameters associated with HAI development.

Main Methods:

• A retrospective study included 476 HAI patients and 1,376 non-HAI controls, with external validation on 2,500 patients.
• Sixteen variables were extracted from Electronic Health Records (EHR).
• LR and ANN models were employed, with stepwise selection identifying seven significant predictors: Foley catheterization, central venous catheterization, arterial line, nasogastric tube, hemodialysis, stress ulcer prophylaxis, and systemic glucocorticosteroids.

Main Results:

• Both LR and ANN models demonstrated excellent discrimination in internal validation (AUCs 0.969 and 0.964, respectively).
• External validation also showed high performance for both models (AUCs 0.870 and 0.850, respectively).
• The developed scoring system achieved high AUC values in both internal (0.965) and external (0.871) validations.

Conclusions:

• A validated scoring system for predicting HAI using simple, observable parameters was successfully developed.
• This scoring system enables efficient identification of high-risk patients, facilitating targeted interventions.
• The model's reliance on EHR data and medical device observation allows for flexible application across various clinical settings.