Development Of Hypoglycemic Risk Assessing Model For Diabetes Patients: Based On A Systematic Review

Objective: we aim to summarize relative risk factors of hypoglycemia in patients with diabetes mellitus;develop and evaluate the models of assessing risk of hypoglycemia in diabetes.Methods: this study consists four parts.Firstly,systematic review and meta-analysis of “risk factors of hypoglycemia in diabetes” is conducted,aiming at summarizing risk factors of primary questionnaire.Secondly,qualitative interview is conducted in clinical workers and diabetics to find potential risk factors of hypoglycemia as additional items for hypoglycemic risk factor questionnaire.Thirdly,a two-round Delphi method is conducted to add or delete or modify the items in risk factors questionnaire and the results are discussed in research group.Finally,a cohort study is performed,and diabetes patients of a third-class special hospital in Tianjin was selected by cluster sampling between April and September in 2019;basal information outcomes in follow-up period are collected in patients who are included;multivariate logistic regression is conducted to verify the relative factors of hypoglycemia;hypoglycemic risk assessing models are developed and validated.Results: 43 studies with 53 references are included and eight relative factors are included in meta-analysis which are age,gender,having a history of hypoglycemia,using sulfonylurea or insulin,in the status of kidney disease,cardiovascular disease or peripheral nervous disease.Other risk factors are described qualitatively.The risk factors mentioned in systematic review are included as items of the primary questionnaire.We also get 12 potential risk factors for hypoglycemia in qualitative interview and we form the 25-item questionnaire for Delphi method by combining the results of SR and qualitative interview.In Delphi method,3 items are deleted,4 items added,3 items modified and 4 contents of 1 item combined.The final risk factor questionnaire is settled.In 3-month follow-up period,29.19% patients experienced hypoglycemic events,37.80% in 6-month follow-up period,and two multivariate logistic regression models are developed.Model of 3 months includes insulin using,TZD using,gastrointestinal disease,changing hypoglycemic agents in three months,BMI,depression,aged 65 or more;the area under ROC(AUC)is 0.703,and Youden index is 0.310 while sensitivity is 0.629 and specificity is 0.649;bootstrap result estimates c-index(0.702),indicating proper internal validation of the model.Model of 6 months consists seven risk factors which are history of hypoglycemia,insulin using,liver disease,changing hypoglycemic agents in three months,depression,aged 65 or more;the area under ROC(AUC)is 0.698,Youden index is 0.315 while sensitivity is 0.781 and specificity is 0.516.H-L test shows that two models are predicting similar to the reality.Decision curves imply that two Models are good in net benefits of clinical.Conclusion:two models are developed and validated and their internal validation and calibration are acceptable.