Early Detection of Rhabdomyolysis-Induced Acute Kidney Injury through Machine Learning Approaches

Introduction: Rhabdomyolysis-induced acute kidney injury (AKI) is one of the most common complications of catastrophic incidents, especially earthquakes. Early detection of AKI can reduce the burden of the disease. In this paper, data collected from the Bam earthquake was used to find a suitable model that can be used in prediction of AKI in the early stages of the disaster. Methods: Models used in this paper utilized many inputs, which were extracted from the previously published dataset, but depending on the employed method, other inputs have also been considered. This work has been done in two parts. In the first part, the models were constructed from a smaller set of records, which included all of the required fields and in the second part; the main purpose was to find a way to replace the missing data, as data are mostly incomplete in catastrophic events.  The data used belonged to the victims of the Bam earthquake, who were admitted to different hospitals. These data were collected on the first day of the incident via questionnaires that were provided by the Iranian Society of Nephrology, in collaboration with the International Society of Nephrology (ISN). Results: Overall, neural networks have more robust results and given that they can be trained on more data to gain better accuracy, and gain more generalization, they show promising results. Overall, the best specificity that was achieved on testing almost all of the records was 99.24% and the best sensitivity that was achieved in testing almost all of the records was 94.44%. Conclusion: We introduced several machine learning-based methods for predicting rhabdomyolysis-induced AKI on the third day after a catastrophic incident. The introduced models show higher accuracy compared to previous works performed on the Bam earthquake dataset.