| Background:Diabetes was one of the most important non-communicable diseases worldwide.China had the largest number of diabetic patients in the world,with the severe situation of prevention and control.Comprehensively reflecting of the prevalence and prevalent trend,dynamically discovering,and monitoring of the risk factors,and scientifically formulating and evaluating of the prevention and control strategies and measures have become urgent tasks.Many studies have been conducted to evaluate the individual risk factors of diabetes,and a broad consensus had been reached.Carrying out identification of the influencing factors at the regional level was also important for health-related decisions,especially those in other areas related to health.In recent years,the content of medical and health services for diabetes had increased considerably.Identification of high-risk patients and potential risks timely had an important role in the secondary prevention of diabetes.At the same time,the specialization in the management of diabetics in primary-level medical and health institutions in China was low,and community health workers did not have a sufficient understanding of diabetic treatment,and offer the management timely and personally.Developing some easy,comprehensive and patient-centric tool for risk evaluation of diabetics based on existed guidelines and research consensus and integrated cross-departmental health records,will be useful to strengthen the pertinence and timeliness of existed clinical treatment and management in health sectors,especially in community health centers,or some other grassroots health institutions.Objective:This study took the occurrence and development risk of diabetes as the evaluation object,and integrated existed monitoring,diagnosis and treatment of diabetes,and other data in the health related fields to analysis the regional risk factors,evaluate the patients management effect of diabetes,and provide a measurable and transferable tool for diabetes risk evaluation and relative decision-making.Furthermore,the field application of the evaluation tool was implemented based on the regional health information platform in one city in China.Materials and Methods:Data used in this study were derived from the 2010 survey of the China Chronic Diseases and Behavioral Risk Factors Surveillance,the Atlas of Soil Environmental Background Values of the People’s Republic of China issued by the China Environmental Monitoring Station,the Tabulation on the 2010 Population Census of the People’s Republic of China By County by the National Bureau of Statistics,2011 China City Statistical Yearbook and 2011 China Regional Economic Statistics Yearbook publish by the National Bureau of Statistics;data of diagnosis and treatment,physical examination and nation basic public health service of type 2 diabetes patients in a city’s regional health information platform from 2015 to 2017.This study included three parts:1.Analysis of regional influencing factors of diabetes prevalenceA comprehensive analysis of the association between the soil trace element content and socioeconomic development and the prevalence of diabetes at the regional level among people aged from 35 to 74 was conduct respectively,by integrating existed disease surveillance and health-related data resources.Principal Component Analysis was used to reduce the dimension of indicators of socioeconomic development.A two-level binary logistic regression model was applied to determine the association,with participants nested within districts/counties.2.Development of a comprehensive evaluation tool for management effect of diabeticsThe effect of diabetes management was built on a two-dimensional structure,by decomposed into two aspects:health status(no-risk,low-risk,high-risk)and health-related behavior(no-risk,risk).Principal Component Analysis was used to reduce the dimension of indicators of health status and health-related behavior.Common factors were extract based on a comprehensive consideration of eigenvalue,contribution rate,common variance,and factor load matrix to extract.A comprehensive risk index was constructed by using the factor score coefficient as the index weight and the matrix contribution factor after the maximum variance rotation as the weight of the common factor.ROC curve was used to determine the critical value of the comprehensive risk index.Health status was evaluated by the indicators included in the comprehensive control goals for type 2 diabetes in China.Health-related behaviors were evaluated by the indicators included in the follow-up records of management of diabetic patients in the National Basic Public Health Service from a city.3.Application of the comprehensive evaluation tool for management effect of diabeticsThe type 2 diabetes patients from a city’s Regional Health Information Platform who received National Basic Public Health Services between 2015 to 2017 were used as the analysis sample,the comprehensive evaluation of management effect was completed with patients’ medical records and follow-up records.A two-level binary logistic regression model with repeated measurement data was applied to determine the association between patients’ management effect risk grades and the prevalence of diabetic complications,with measurement points nested within participants.Based on the full sample cohort and propensity score matching cohort,Kaplan-Meier method and Cox proportional hazard regression model were used to analyze the risk of complications during the follow-up period among uncomplicated patients with different risk assessment grades of the first evaluation from 2015 to 2017.Six traditional machine learning classifiers,K-Nearest Neighbor,Decision Tree,logistic,Naive Bayes,Support Vector Machine and Adaboost,and neural network model,were used to predict the complications of type 2 diabetes.Results:1.Analysis of regional influencing factors of diabetes prevalenceAmong the 9 trace elements,the risk of diabetes with different iron and selenium concentrations in soil was different.After adjusting for all confounding factors of interest,the soil iron concentration of 431.407mg/kg-486.251mg/kg was positively associated with diabetes compared to a concentration of 270.181mg/kg-384.713mg/kg,with the adjusted OR was 1.261(95%CI,1.065-1.493).the soil selenium concentration of 0.247mg/kg～0.338mg/kg was positively associated with diabetes compared to the concentration of 0.186-0.219 mg/kg,with the adjusted OR was 1.203(95%CI,1.018-1.421).The 15 socioeconomic development indicators were reduced to 5 common factors by Principal Component Analysis,which potentially represented social development,industrial development,pollutant emission,urban construction,and population aging.After adjusting for all possible confounding factors of interest,both the high-level group of social development and pollutant emission had adjusted OR of 1.256(95%CI,1.082-1.457)and 1.174(95%CI,1.018-1.355)relative to their own low-level group.2.Development of a comprehensive evaluation tool for management effect of diabeticsThe 10 health status evaluation indexes were reduced to 4 common factors,F1-F4,and the comprehensive risk index f of type 2 diabetes patients’ health status was calculated by using the contribution rate as the factor weight.The value of franged from-0.90 to 1.08,which had a skewed distribution with a median of-0.02 and an interquartile range of-0.27 to 0.25.The evaluation criteria were as follows:f(Ⅰ)no risk,f<-0.09;f(Ⅱ)low risk,0.09≤f<0.21;f(Ⅲ)high risk,f≥0.21.Women,60-69 years old and patients treated with insulin injections were more likely to have health risks.Compared with the reference group,the OR values of low health risk were 1.258(95%CI:1.074-1.475),1.230(95%CI:1.014-1.491)and 1.260(95%CI:1.024-1.551),respectively;the OR values of high health risk were 1.454(95%CI:1.246-1.697),1.204(95%CI:0.999-1.450)and 1.238(95%CI:1.011-1.517),respectively.The 6 health-related behaviors evaluation indexes were reduced to 3 common factors,S1-S3,and the comprehensive risk index s of type 2 diabetes patients’ health status was calculated by using the contribution rate as the factor weight.The value of s ranged from-0.33 to 2.45,which had a skewed distribution with a median of-0.33 and an interquartile range of-0.15 to 0.21.The evaluation criteria were as follows:s(Ⅰ)no risk,s<0.12;s(Ⅱ)risk,s≥0.12.Increased age was a protective factor for the risk of health-related behaviors.Compared with the 35-59 years old group,the OR values of the health-related behaviors risk for the 60-69 years old group and 70-year-old and above group were 0.187(95%CI:0.036-0.975)and 0.105(95%CI:0.020-0.542),respectively.Based on the two-dimensional structure of the risk classification of the comprehensive risk index of the health status f and the health-related behavior s,the final risk classification for diabetes management effect was:0=f(Ⅰ)+s(Ⅰ),1=f(Ⅰ)+ s(Ⅰ)or f(Ⅱ)+s(Ⅰ),2=f(Ⅱ)+(Ⅱ)or f(Ⅲ).3.Application of the comprehensive evaluation tool for management effect of diabeticsFour types of medical and health management practice data from inpatient records,outpatient records,health examination records,and national basic public health services could support the application of comprehensive evaluation tool for management effect of diabetics.From 2015 to 2017,there were 2613 type 2 diabetics who had medical records on a city’s regional health information platform and followed up in the National Basic Public Health Services.21857 evaluation records were integrated after data filtering,cleaning,stitching and deduplication.In the evaluation of 21,857 person-times,the proportions of no-risk,low-risk and high-risk of patient management effect were 59.83%,33.64%and 6.53%,respectively,and the proportions of risk grading decreased and increased were 12.57%and 12.13%respectively.The proportion of patients with risk grading increased in the 35-59 years old group(7.80%)was lower than that in the 60-69 years old(12.17%)and 70 years old and above group(14.83%).A total of 680 of 2613 patients were associated with complications(26.02%),and 483 were associated with multiple complications(18.48%).In the evaluation of 21,857 person-times,there were 2763 person-times with complications(12.64%)and 1791 person-times with multiple complications(8.19%).Patients with high-risk of management effect were more likely to have complications than patients with on-risk,the adjusted OR value was 1.462(95%CI,1.001-2.134).During the three years of follow-up,530 patients and 398 patients of the 2375 patients who had no complications at the start of observation were diagnosed with complications(22.3%)and multiple complications(16.3%),respectively.After matching the propensity scores,compared with patients who were initially evaluated as no-risk,patients as high-risk were more likely to be diagnosed with multiple complications and neurological complications during follow-up,with HR values of 2.054(95%CI:1.097-3.846)and 2.246(95%CI:1.080-4.671).Compared with patients who were initially evaluated as low-risk,patients as high-risk were more likely to be diagnosed with complications with the HR value was 1.679(95%CI:1.003-2.811).The same associations were also observed in the diagnosis of neurologic,peripheral circulatory,and multiple complications,with the HR values of 2.336(95%CI:1.125-4.849),3.251(95%CI:1.212-8.721)and 1.891(95%CI:1.025-3.488).Among the six traditional machine learning classifiers,the K-Nearest Neighbor had the highest forecast accuracy,for complications the accurate rate was 85.94%,and for kidney,eye,neurologic,peripheral circulation,and multiple complications were 93.98%,90.96%,90.67%,92.15%,and 88.59%,respectively.For the feedforward neural network of the six kinds of complications,the optimal hidden layer number of each’s model were 4,4,2,2,4 and 4,and the optimal single layer optimal neuron number were 120,140,120,120,120 and 120.The performance of the optimal model of feedforward neural networks were better than the traditional machine learning classifiers,the forecast accuracy for complications was 88.71%,and for kidney,eye,neurologic,peripheral circulation,and multiple complications were 93.22%,92.43%,93.43%,94.05%,and 90.68%,respectively.Conclusion:1.Analysis of regional influencing factors of diabetes prevalenceAt the regional level,the relative increase in risk was small but of possible public health importance because of the high incidence of diabetes and the prevalence of regional influencing factors exposure,which should be highly valued by the government.In addition to the fields of health and wellness,the results had implications for the formulation of policies and measures in some health-related areas,such as food circulation,environmental protection,transportation,etc.2.Development of a comprehensive evaluation tool for management effect of diabeticsThe comprehensive risk evaluation tool for management effect of type 2 diabetes patients established in this study was a two-dimensional risk grading structure model,including a health risk index f and a health-related behavior risk index s,and the final risk can be classified as:f(Ⅰ)+s(Ⅰ),1=f(Ⅰ)+s(Ⅰ)or f(Ⅱ)+s(Ⅰ),2=.f(Ⅱ)+s(Ⅱ)or f(Ⅲ).The model was a simple evaluation tool developed based on the existed practice data,which could be transformed into suitable technologies that can be applied at the grassroots level to improve the pertinence and timeliness of diabetes diagnosis,treatment and management services.3.Application of the comprehensive evaluation tool for management effect of diabeticsThe tool could implement dynamic evaluation in the city’s regional health information platform,by integrated the real clinical diagnosis and treatment,physical examination,and public health management practice data.Diabetic patients with high management effect risk were more likely to suffer from complications had been observed during a period of 3-year follow-up,by selecting an appropriate analysis model.The evaluation tool can be used to help the health institutions,especially community health and other grass-roots institutions to identify high-risk patients quickly. |
PDF Full Text Request