Postoperative MPA-AUC Prediction for Kidney Transplant Recipients Based on Model Interpretability Technique

Mycophenolic acid (MPA) is a commonly used immunosuppressant for organ transplant recipients. the Area Under Curve (AUC) of 0∼12 hours after taking mycophenolic acid (MPA) drugs is often monitored clinically to evaluate drug exposure and adjust medication dosage for individual treatment, so as to achieve the best immunosuppression state. However, in clinical practice, medical staff often need to collect the data of 9 to 12 blood samples within 0 to 12 hours after the patient takes the medicine, which leads to high test cost, heavy workload of medical staff and great pain to patients. This paper proposes a prediction method based on the Model interpretability technique SHAP. Firstly, SHAP calculates the influence of a single feature added to the input on the overall prediction results of the model and analyzes the weight of each feature. The ranking of the weights was used to dynamically select the best blood sampling time-points for prediction. Then we established a prediction model of for kidney transplantation recipients by artificial neural network. The experimental results show that When 4-point sampling is used, the MAE of prediction model is stable below 15%, and the prediction result is better than that of traditional feature selection method and blood collection time-points selection results proposed in other reference.