Abstract Background Acute Kidney Injury (AKI) remains a significant global health challenge, especially in resource‐limited settings. Most existing predictive models rely heavily on serum creatinine (SCr) levels and standardized electronic medical records (EMRs). However, in many low‐resource environments, SCr testing is infrequent, and EMR systems often lack standardization in data structure, terminology, and recording practices (a.k.a., non‐standard EMRs). These limitations hinder the consistent extraction of features needed for accurate AKI prediction and highlight the urgent need for adaptive frameworks tailored to diverse and resource‐limited healthcare environments. Purpose This study aimed to develop and validate a machine learning model using non‐standardized EMRs for predicting AKI, even without SCr data. Methods This multicenter observational study, conducted from 2010 to 2016 across 15 hospitals in China, employed the Light Gradient Boosting Machine (LightGBM) to create predictive models. The model's performance was assessed using area under the curve (AUC), precision, recall, specificity, and accuracy. Results A total of 561 137 hospitalized patients were eligible for the analyses, of whom 45 610 were diagnosed with AKI. The LightGBM model demonstrated high accuracy in predicting AKI, with AUC values ranging from 0.860 to 0.986. The study showed that non‐standard EMRs could effectively predict AKI. Importantly, the model maintained strong predictive performance even without SCr data, indicating that AKI can be accurately predicted without this traditional biomarker. Conclusion Non‐standard EMRs are valuable for predicting AKI, even in the absence of SCr data. This approach is particularly useful in resource‐limited settings, where traditional biomarkers are often unavailable, demonstrating the potential of other clinical features to compensate for missing SCr data in AKI prediction.
Guo et al. (2025) studied this question.