Application of an artificial neural network model to predict delayed decrease of serum creatinine in pediatric...

Area of Science:

• Nephrology
• Computer Science
• Biomedical Engineering

Background:

• Artificial neural networks (ANNs) have shown promise in predicting outcomes for adult kidney recipients.
• Predicting post-transplant serum creatinine levels is crucial for monitoring kidney function in pediatric recipients.
• Delayed decrease in serum creatinine can indicate potential complications or suboptimal graft function.

Purpose of the Study:

• To evaluate the effectiveness of an ANN model in predicting a delayed decrease in serum creatinine among pediatric kidney recipients.
• To identify key input variables that contribute to predicting this outcome.
• To compare the predictive performance of the ANN model against traditional logistic regression analysis.

Main Methods:

• An ANN was constructed using a training set of 107 pediatric kidney recipients.
• Input variables included pre-transplant and early post-transplant clinical data.
• The model's predictive accuracy was validated on an independent set of 41 patients.

Main Results:

• The ANN model achieved an overall accuracy of 87.14% on the entire patient cohort.
• Key predictive variables included early post-transplant serum creatinine, urine volume, diagnostic category, and patient age.
• The ANN model demonstrated superior sensitivity (0.875) and specificity (0.87) compared to logistic regression (0.37 and 0.94, respectively).

Conclusions:

• An artificial neural network model can effectively predict delayed serum creatinine decrease in pediatric kidney recipients.
• The developed ANN model shows higher accuracy and better sensitivity/specificity than logistic regression.
• The availability of the source code facilitates further validation and prospective studies.