Intuitive Risk Equation for Post-Transplant Bloodstream Infection Prediction: A Symbolic Regression Approach.

Area of Science:

• Transplant Immunology
• Medical Informatics
• Infectious Disease Epidemiology

Background:

• Liver transplant recipients face high infectious risks from surgery and immunosuppression.
• Post-transplant bloodstream infections (BSI) significantly increase morbidity and mortality.
• Existing BSI prediction models lack applicability and interpretability in liver transplant settings.

Purpose of the Study:

• To develop predictive models for post-transplant BSI using preoperative and perioperative data.
• To derive an interpretable risk equation for BSI using symbolic regression.
• To compare the performance and interpretability of symbolic regression with conventional machine learning models.

Main Methods:

• Retrospective observational study of 245 adult liver transplant recipients.
• Extraction of clinical and laboratory variables from electronic medical records.
• Development and comparison of conventional machine learning and symbolic regression models for BSI prediction.

Main Results:

• Post-transplant BSI occurred in 33.4% of patients.
• Symbolic regression achieved comparable discrimination (AUC 0.63) to conventional models (AUC 0.53-0.64) while providing transparent risk equations.
• Symbolic regression identified perioperative factors and viral serologic markers as key predictors, unlike conventional models relying mainly on lab data.

Conclusions:

• Symbolic regression offers an interpretable approach to predicting post-transplant BSI.
• Viral serologic markers may indicate immunologic vulnerability in transplant recipients.
• This interpretability-focused method can inform future risk stratification models using comprehensive perioperative data.