Fig. 6

Real-time prediction process of CIAKI in diabetes based on the web calculator platform. a An example of CIAKI prediction in one hospitalized patient. When a patient arrived at the hospital, the doctor obtained the basic information of the patient and made a diagnosis within 0–6 h. At point A, we knew that ACS was 1, stable angina was 0, his previous serum creatinine was 68 µmol/L, and his diabetes history was 5–10 years. According to the CIAKI web predictive platform, other missing values were filled, and we calculated that the risk was 0.196. At point B, we knew that the CHF was 1, and the risk rose to 0.367 and predicted CIAKI occurrence (risk threshold was 0.3338). At point C, the patient’s preoperative DBP was 78 mmHg, glucose was 15.8 mmol/L and the risk rose to 0.434. At point D, the patient took diuretics, and the risk was 0.622. During the period of 6 h-24 h, the patient underwent preoperative examination. At E, hemoglobin was 125 g/L, and the risk was 0.651. At point F, the urine protein level was 2 + , and the risk was 0.689. At point G, LVEF was 35%, and the risk was still 0.689. At point H, uric acid was 435 µmol/L, albumin was 40 g/L, serum creatinine was 71.3 µmol/L, and the risk was 0.717. When the patient arrived 48 h after admission (point I), he underwent CAG and PCI and used contrast volumes of 100 mL, and he had 2 vessels of coronary artery disease; the risk was 0.579. Creatinine was examined at 24 h, 48 h and 72 h after CAG and PCI, and the real occurrence of CIAKI was diagnosed at 48 h after CAG and PCI (point J). b All features of BCPMD were known at point I, and the model output the risk value using the dynamic explainable CIAKI predictor. In this example, our web platform identified patients with possible CIAKI within 1 h of admission