Comparing machine learning and logistic regression for acute kidney injury prediction in trauma patients: A retrospective observational study at a single tertiary medical center

Purpose: The number of elderly patients with trauma is increasing; therefore, precise models arenecessary to estimate the mortality risk of elderly patients with trauma for informed clinical decision-making. This study aimed to develop machine learning based predictive models that predict30-day mortality in severely injured elderly patients with trauma and to compare the predictive performanceof various machine learning models.Methods: This study targeted patients aged ≥65 years with an Injury Severity Score of ≥15 who visitedthe regional trauma center at Chungbuk National University Hospital between 2016 and 2022.Four machine learning models—logistic regression, decision tree, random forest, and eXtreme GradientBoosting (XGBoost)—were developed to predict 30-day mortality. The models’ performancewas compared using metrics such as area under the receiver operating characteristic curve (AUC),accuracy, precision, recall, specificity, F1 score, as well as Shapley Additive Explanations (SHAP) valuesand learning curves.Results: The performance evaluation of the machine learning models for predicting mortality in severelyinjured elderly patients with trauma showed AUC values for logistic regression, decision tree,random forest, and XGBoost of 0.938, 0.863, 0.919, and 0.934, respectively. Among the four models,XGBoost demonstrated superior accuracy, precision, recall, specificity, and F1 score of 0.91, 0.72,0.86, 0.92, and 0.78, respectively. Analysis of important features of XGBoost using SHAP revealedassociations such as a high Glasgow Coma Scale negatively impacting mortality probability, whilehigher counts of transfused red blood cells were positively correlated with mortality probability. Thelearning curves indicated increased generalization and robustness as training examples increased.Conclusions: We showed that machine learning models, especially XGBoost, can be used to predict30-day mortality in severely injured elderly patients with trauma.Prognostic tools utilizing these models are helpful for physicians to evaluate the risk of mortality inelderly patients with severe trauma.

show abstract

Prediction model of intradialytic hypertension in hemodialysis patients based on machine learning

Zhou,

Guo,

wang

et al. 2024

Preprint

View full text Add to dashboard Cite

Objective: The global prevalence of chronic kidney disease (CKD) is escalating, particularly for end-stage renal disease (ESRD), which has led to greater dependence on hemodialysis. This upswing exerts substantial strains on patient families and healthcare systems. A critical concern during hemodialysis is the emergence of Intradialytic Hypertension (IDH), which carries significant health risks. Delayed management of IDH can lead to severe cardiovascular and cerebrovascular complications. The aim of our study was to harness machine learning methodologies to develop a predictive algorithm for IDH, utilizing patient demographic data and dialysis records. Our model equips medical professionals with a robust predictive tool that enhances the detection of patients more susceptible to hypertension during dialysis, thereby advancing the pre-screening for individuals considered at increased risk. Methods: This study developed two predictive models for IDH, named IDH-1 and IDH-2, by employing a suite of machine learning algorithms, namely the Light Gradient Boosting Machine (LGBM), Support Vector Machine (SVM), and TabNet. IDH-1 is specifically engineered to provide immediate predictions of IDH risk prior to a hemodialysis session, utilizing records from the imminent pre-dialysis period combined with historical average dialysis data, whereas IDH-2 employs records from the current dialysis session along with historical average data to forecast the risk of IDH for the next hemodialysis session. The performance evaluation of the models utilized key metrics, including Area Under the Curve (AUC), recall, accuracy, and F1 score, which are crucial in determining the models' precision and reliability. Results: This research analyzed data from 1,405 patients at Shenzhen People's Hospital over 185,125 dialysis sessions and 416 patients at Fuding City Hospital across 71,427 sessions. Data from Shenzhen served as the training set, while Fuding data comprised the test set, supporting the model development and validation process. In the IDH-1 models, the LGBM outperformed SVM and TabNet with an AUC of 0.87. LGBM achieved a recall of 0.73, an accuracy of 0.243, and an F1 score of 0.36. For IDH-2 models, LGBM maintained superior performance, with an AUC of 0.75, a recall of 0.56, an accuracy of 0.17, and an F1 score of 0.26. Predictor importance analysis for the LGBM algorithm identified pre-dialysis diastolic pressures, historical mean arterial pressure, and historical average IDH episodes as significant for the IDH-1 model. For the IDH-2 model, historical average IDH episodes and post-dialysis systolic pressures were most predictive. Conclusions: This study's results highlight the significant potential of machine learning techniques in leveraging demographic and dialysis data to predict IDH in patients undergoing hemodialysis.

show abstract

Comparing machine learning and logistic regression for acute kidney injury prediction in trauma patients: A retrospective observational study at a single tertiary medical center

Cited by 3 publications

References 28 publications

Development and validation of an artificial intelligence mobile application for predicting 30-day mortality in critically ill patients with orthopaedic trauma

Development and validation of an artificial intelligence mobile application for predicting 30-day mortality in critically ill patients with orthopaedic trauma

Predicting 30-day mortality in severely injured elderly patients with trauma in Korea using machine learning algorithms: a retrospective study

Prediction model of intradialytic hypertension in hemodialysis patients based on machine learning

Contact Info

Product

Resources

About