Local Interpretability of Calibrated Prediction Models: A Case of Type 2 Diabetes Mellitus Screening Test