WEDNESDAY, Nov. 6, 2024 (HealthDay News) -- Clinical data and machine learning can help to predict intradialytic hypotension (IDH) for patients undergoing hemodialysis, according to a study published online Sept. 14 in the Journal of Kidney Care.
Shamsul K. Masum, Ph.D., from the University of Portsmouth in the United Kingdom, and colleagues investigated the scope of machine learning techniques in predicting IDH by analyzing various standard clinical variables measured by clinicians. The dataset included 73,323 hemodialysis sessions (3,944 patients) with 36,662 IDH events seen at 10 centers during 2000 to 2020.
The researchers found that systolic blood pressure (SBP) and diastolic BP (DBP) were key predictor variables. Predialysis, patients with IDH had lower SBP, as well as a greater percentage drop in SBP during dialysis. Similarly, people with IDH had lower predialysis DBP and a greater percentage drop in their DBP. IDH probability increased with lower pre-SBP, greater delta-systolic, and increased weight loss. Highest specificity (73.9 percent) and highest overall predictive accuracy (75.5 percent) were seen with a machine learning model with Random Forest. However, a model with Bidirectional Long Short-Term Memory (Bi-LSTM) achieved the highest sensitivity (78.5 percent) in predicting IDH. In a separate validation dataset, the Bi-LSTM model achieved accuracy of 74.09 percent, sensitivity of 74.81 percent, specificity of 73.3 percent, and a receiver operating characteristic-area under the curve (ROC-AUC) of 0.816. Based only on predialysis data, the prediction performance dropped to accuracy of 68.60 percent, specificity of 69.81 percent, sensitivity of 67.40 percent, and an ROC-AUC of 0.757.
"A prediction model using machine learning algorithms offers great promise as a tool in identifying patients at risk of IDH in advance," the authors write.
Abstract/Full Text
