To investigate the influencing factors of hypoglycemia during hemodialysis (HD) in patients with diabetic nephropathy (DN), and establish a predictive model and scoring system for hypoglycemia.

242 DN patients who underwent maintenance hemodialysis (HD) in the hemodialysis center of our hospital from February 2019 to July 2021 were selected. According to whether hypoglycemia occurred during HD, they were divided into non-hypoglycemia group (n= 112) and hypoglycemia group (n= 130). The clinical data of patients were collected through questionnaire survey, and the relevant variables were screened by single factor logistic analysis and Lasso-Logistic regression analysis. The logistic regression model and scoring table were established with hypoglycemia as the dependent variable. The discrimination and accuracy of the model were verified by receiver operating characteristic (ROC) curve and calibration curve.

There were significant differences in age, body mass index (BMI), medication compliance, mean blood glucose (MBG), coefficient of variation of blood glucose (CVBG), daily exercise time, reasonable diet control, depression, and care ability between the two groups (P< 0.05). Univariate logistic regression analysis showed that age, BMI, MBG, CVBG, medication compliance, daily exercise time, reasonable diet control, depression, and care ability were the influencing factors of hypoglycemia in patients with DN during HD (P< 0.05). Lasso-Logistic regression analysis showed that MBG, CVBG, medication compliance, reasonable diet control, depression, and care ability were also the influencing factors of hypoglycemia in maintenance HD patients with DN (P< 0.05). The AUC of ROC curve of the constructed logistic risk prediction model was 0.826, and the 95%CI was 0.785-0.897, and the calibration chart showed that the predicted value of the model was consistent with the actual value observed. According to the prediction score table constructed, the occurrence probability of hypoglycemia corresponding to 0-9 points was 7.3%-100.0%. When the maximum Yoden index was 0.49, the cut-off point of the scoring table was 5 points. Under this score, the sensitivity, specificity, and accuracy of the scoring table were 83.61%, 85.83%, and 84.71%, respectively.

The hypoglycemia score scale developed based on Lasso-Logistic regression model combined with MBG, CVBG, medication compliance, reasonable diet control, depression and, care ability could reliably predict hypoglycemia in patients with DN during HD, which may provide basis for blood glucose control during clinical treatment, and have certain clinical application value.