| Objective: The purpose of this study was to understand the basic characteristics of patients with T2 DM outpatient insulin therapy and their blood glucose control status in a North China region through a multi-center cross-sectional study.Based on this,the demographic indicators and the diagnosis and treatment of diabetes were used to analyze the influencing factors of blood glucose compliance in T2 DM patients treated with insulin,and the predictive analysis of blood glucose compliance was carried out.In addition,this study will explore the application value of the combination of elastic network algorithm and machine learning algorithm in predicting blood glucose control in patients with type 2 diabetes through comparison with traditional Logistic regression algorithms.Methods: This study collected data from outpatient T2 DM patients in 27 hospitals in 6 cities in North China from January 2016 to December 2017.The data of the first five outpatients who met the inclusion and exclusion criteria were collected,including: general characteristics,physical measurement,T2 DM condition,T2 DM related treatment and medication information,and laboratory inspections.Glycosylated Hemoglobin ≤ 7.0% was used to define the glycemic control target,and a univariate analysis was performed between the control group and the non-control group.First,the elastic network algorithm was used to reduce the dimensions of statistically significant indicators in a single factor analysis.The EN-Logistic model was used to perform multi-factor analysis of blood glucose standards.Second,machine learning models were established using indicators before and after the dimension reduction(Random forests,support vector machines,and artificial neural networks)and stepwise logistic models were compared to model the results.Results: This study finally included 2,787 T2 DM outpatients treated with insulin from 27 centers in 6 cities.Among them,there were 1407 male T2 DM patients,accounting for 50.5%.The patients were 56.37±11.41 years for average,almost all married(2745 cases,98.5%),and the average duration of T2 DM was 6.57±5.28 years.(1)Basic information: Among all the subjects,52.2% were overweight and 16.8% were obese;1452 patients were centrally obese,accounting for 52.1%;919(33.0%)patients had a history of smoking,891(32.0%)patients had a history of drinking.859 patients(30.8%)had a family history of type 2 diabetes.Among all patients,717(25.7%)patients had a previous medical history,of which 493 had a history of hypertension;225 had a history of ASCVD.46.9% of patients have the typical onset characteristics of diabetes;44.2% of patients have diabetes-related complications,including 382 cases of diabetic nephropathy(13.7%);566 cases of diabetic retinopathy(20.3%);814 cases of diabetic peripheral neuropathy(19.2%);41 cases of diabetic foot(1.5%);347 cases of diabetic lower extremity vascular disease(12.5%).The patient’s average insulin dose was 17.17 U,and 222(8.0%)patients did not use oral hypoglycemic agents.82.3% of patients had exercise habits,and 89.5% of them took measures to adjust their eating habits after diagnosis of T2 DM.Among laboratory indicators,26.8% of patients had good blood pressure control;26.6% of patients had TC within normal range;36.8% of patients had TG within normal range;33.6% of patients had good HDL-C control,and 37.9% of patients had LDL-C is well controlled.There were 1277 patients with blood glucose control level,and the compliance rate was 45.82%.(2)Univariate analysis: Differential analysis was performed between the blood glucose control group and non-control group.Age,gender,BMI classification,central obesity,smoking history,family history,exercise,diet,previous medical history,course of disease,insulin dose,typical characteristics of diabetes mellitus(polyphagia,polyphagia,polyuria and weight loss),complications,comorbidities,previous hypoglycemia,duration of basal insulin use,oral hypoglycemic drugs,and compliance with doctor’s orders were related to blood glucose control.(3)Multi-factor analysis: The results of the EN-Logistic model show that married,exercise,and adjustment of vegetable oil intake are protective factors for blood glucose compliance,central obesity,family history of T2 DM,long duration,complications,high insulin doses,and A typical morbidity characteristic is a risk factor for blood glucose compliance.(4)Predictive analysis: Compare the machine learning model and stepwise logistic regression model by the sensitivity,specificity,accuracy and AUC of each model.Among the four models,the random forest algorithm has the best prediction performance,and its sensitivity,specificity,accuracy,and AUC are all 0.70,which are 16.67%,6.06%,9.8%,and 4.48% higher than the progressive Logistic,respectively;followed by the support vector machine algorithm,its sensitivity,specificity,accuracy,and AUC are 0.69,0.65,0.67,and 0.They were 67 respectively.The sensitivity and accuracy are higher than the Logistic regression model,AUC is consistent with Logistic regression,and the specificity is slightly lower than Logistic.Among the four models,the performance of neural network is poor.Conclusion: Although the blood glucose control level of T2 DM outpatients in northern China is slightly higher than the blood glucose control rate reported by various domestic research results,this control rate is still not ideal for the control of T2 DM.In the results of this study,married,exercising,and adjusting vegetable oil intake were protective factors for blood glucose compliance.Central obesity,a family history of T2 DM,long duration of T2 DM,complications of T2 DM,high insulin doses,and typical onset characteristics were risk factors for blood glucose compliance.In the daily management of diabetics,we should focus on the high-risk groups,and adopt individualized treatment programs to comprehensively control blood sugar,control blood sugar to a reasonable level,and to reduce the incidence of related complications.In addition,this study found that elastic networks and machine learning algorithms have good application value in predicting blood glucose compliance. |
PDF Full Text Request