A novel method to predict white blood cells after kidney transplantation based on machine learning

Area of Science:

• Nephrology
• Transplant Surgery
• Machine Learning in Medicine

Background:

• Abnormal white blood cell counts post-kidney transplant are a significant adverse outcome.
• Low white blood cell counts increase infection risk and decrease transplant success.
• Immunosuppressive agents and other factors contribute to abnormal counts.

Purpose of the Study:

• Develop a machine learning model to predict leukocyte drop to abnormal levels after kidney transplantation.
• Provide a clinical reference for managing post-transplant white blood cell counts.

Main Methods:

• Utilized data from 546 kidney transplant patients.
• Introduced time correlation features for variable analysis.
• Applied Least Absolute Shrinkage and Selection Operator (LASSO) for variable selection, retaining 20 key variables.
• Evaluated eight machine learning algorithms using five-fold cross-validation.

Main Results:

• The multilayer perceptron model achieved 71.34% accuracy, 61.18% sensitivity, 82.28% specificity, and 77.30% AUC.
• Key predictors for leukopenia included time proportion of lymphocytes below normal, blood group AB, gender, and platelet CV.
• The multilayer perceptron model demonstrated strong predictive performance.

Conclusions:

• The multilayer perceptron model shows significant potential for predicting abnormal white blood cell counts after kidney transplantation.
• This predictive model can aid in risk stratification for transplant recipients.
• Further external and prospective validation is recommended.