| Abstrakt: |
Background. Recent studies suggest machine learning represents a promising predictive option for patients in intensive care units (ICU). However, the machine learning performance regarding its actual predictive value for early detection in acute kidney injury (AKI) patients remains uncertain. Objective. This study represents the inaugural meta-analysis aiming to investigate the predictive value of machine learning for assessing the risk of AKI among ICU patients. Methods. PubMed, Web of Science, Embase, and the Cochrane Library were all thoroughly searched from inception to June 25, 2022. Eligible studies for inclusion were those concentrating on the predictive value and the development, validation, or enhancement of a prediction model for AKI patients in the ICU. Measures of effects, including c-index, sensitivity, specificity, and their corresponding 95% confidence intervals (CIs), were employed for analysis. The risk of bias in the included original studies was assessed using Probst. The meta-analysis in our study was carried out using R version 4.2.0. Results. The systematic search yielded 29 articles describing 13 machine-learning models, including 86 models in the training set and 57 in the validation set. The overall c-index was 0.767 (95% CI [0.746, 0.788]) in the training set and 0.773 (95% CI [0.741, 0.804]) in the validation set. The sensitivity and specificity of included studies are as follows: sensitivity [train: 0.66 (95% CI [0.59, 0.73]), validation: 0.73 (95% CI [0.68, 0.77])]; and specificity [train: 0.83 (95% CI [0.78, 0.87])], validation: 0.75 (95% CI [0.71, 0.79])]. Conclusion. The machine learning-based method for predicting the risk of AKI in hospital ICU patients has excellent predictive value and could potentially serve as a prospective application strategy for early identification. PROSPERO Registration number ID: CRD42022362838. [ABSTRACT FROM AUTHOR] |
