Constructing a model of MHD capacity overload and its clinical effectiveness

ObjectiveTo construct an early warning model of maintenance hemodialysis （MHD） capacity overload based upon basic dialysis data, ultrasound and routine blood biochemistry and examine its clinical effectiveness. Methods From January 2023 to December 2023, 603 patients on MHD were retrospectively reviewed. They were assigned into two sets of modeling （n = 402） and validation （n = 201） according to a ratio of 2:1. The relevant clinical were collected. And the influencing factors of MHD capacity overload in modeling set were analyzed by univariate and LASSO-Logistic regression. The risk prediction nomogram model was constructed. And consistency index, calibration curve, receiver operating characteristic （ROC） curve and decision curve were utilized for verifying the prediction model internally and externally. Results Serum albumin, Kt/V, RRF and left ventricular ejection fraction （LVEF） in overload group were lower than those in normal group. NT-proBNP, ultrasonic lung B-line and right atrial diameter were higher than those in normal group （P＜0.05）. LASSO results indicated that serum albumin, Kt/V, RRF, LVEF, NT-proBNP, ultrasonic lung B-line and right atrial internal diameter were Top 7 characteristic variables of volume overload in MHD. Logistic regression results revealed that Kt/V, serum albumin, RRF and LVEF were the protective factors for MHD volume overload. And NT-proBNP, ultrasonic lung B-line and right atrial diameter were the risk factors for MHD volume overload （P＜0.05）. The risk prediction nomogram model of capacity overload was constructed according to the influencing factors of Logistic regression. Consistency index in modeling and validation sets was 0.860 and 0.814, respectively. Calibration curve showed that the incidence of MHD capacity overload predicted by the model in modeling and validation sets was basically consistent with the actual incidence. ROC curve indicated that the area under the curve as predicted by the model in modeling and validation sets was 0.917（95%CI: 0.886-0.948） and 0.916（95%CI: 0.869-0.962）. Decision curve revealed that when the high-risk threshold of modeling set was 0-0.7 and the high risk threshold of validation set 0-0.7. The model had an excellent clinical net benefit. Conclusion In MHD patients, the occurrence of volume overload is affected by serum albumin, Kt/V, RRF, LVEF, NT-proBNP, ultrasonic lung B-line and right atrial diameter. The early warning model built according to the factors related to volume overload has excellent predictive capability and clinical benefit.