Analysis of Influencing Factors of Community Management of Diabetes based on 1 Adaptive-Lasso Logistic Regression Model

Background: Influencing factors of community management of diabetes are 9 complex and controversial, and how to select the most effective multiple influencing factors 10 requires in-depth research. This study aims to analyse multiple influencing factors by adaptive- 11 lasso logistic regression model for the effectiveness of diabetes management of the community 12 to improve the efficiency and reduce the burden of diabetes. Methods: A cross-sectional survey 13 (N=1,127) was adopted to establish the adaptive-lasso logistic regression model of influencing 14 factors for community management of diabetic patients based on cluster sampling data of 15 diabetic patients in Chengdu city, China. By comparing with the full-variable logistic model 16 and the ridge logistic model to find the advantages of the adaptive lasso-logistic regression 17 model in community diabetes management. Results: A total of 1,127 diabetic patients were 18 included in the cross-sectional survey. The latest fasting blood glucose was included in the 19 analysis. Among the included population, 90.6% of them had a fasting glucose level higher 20 than 6.1mol/L, and 9.4% of them were below 6.1mol/L. By cross-validation, after folding eight 21 times, the variables involved in the Adaptive lasso-logistic regression model include age, 22 education level, main source of income, marital status, average monthly income, free medical 23 service, basic medical insurance for residents, hospital history, number of follow-up evaluations and 1981, which were lower than the full-variable logistic model (2349, 2023) and the ridge logistic model (2312, 2013). From the perspective of time cost, the adaptive-lasso logistic regression model was better than the other two models. Conclusions : The adaptive-lasso logistic regression model can be used to analyse the influencing factors of community 30 management in patients with diabetes. Community intervention and intensive management 31 measures can significantly improve the blood glucose status of patients with diabetes. 32

Sichuan Province. The inclusion criteria were through community mobilisation and publicity 73 by the community health service centre. To avoid bias in the study design, we conducted a 74 multidisciplinary expert demonstration and formulated reasonable inclusion and exclusion 75 criteria to ensure higher reliability and validity when formulating the questionnaire. Random 76 number tables were used to select registered diabetic patients in three communities in this study.

77
The survey included demographic information (i.e. gender, age, marital status, education, 78 number of children), self-diabetes management status (i.e. self-assessment of glycemic 79 management, hospital history), basic situation of living (i.e. main source of income, average 80 monthly income, medical payment methods), diabetes management in the community in the 81 past six months (i.e. number of follow-up evaluations by the family doctor team, number of 82 health consultations and free consultations, whether to participate in community blood glucose 83 measurement voluntarily) and latest fasting blood glucose level and other information. Family doctor follow-up records, blood glucose records, and health consultations and free 85 consultations were derived from the residents' health management files and the Chengdu 86 regional health information platform. 87 Participants whose home address or contact details were unchanged for more than three years 88 were included in this study. The collection questionnaire was double-entry by two individually 89 researches to avoid bias in the data collection process. During the process of data collection 90 and entry, no data loss occurred.

91
Each participant signed an informed consent form, and they volunteered to withdraw from the 92 questionnaire process at any time without reason. For illiterate and semi-illiterate participants, 93 data were collected by investigators reading informed consent and reading questionnaires. Data 94 collectors were not involved in the data analysis process to avoid bias. The Chronbach 's α 95 coefficient of the knowledge, attitude and behaviour evaluation scale for people with diabetes 96 was 0.757, which can be considered as a good internal consistency.

114
Based on logistic regression,   is defined as its log-likelihood function. When the fasting glucose level of included patients is less than or 116 equal to 6.1mol/L, i y will be defined as 0；when the fasting glucose level is greater than 117 6.1mol/L, i y will be defined as 1, then:

124
Data were entered using Epidata (Version 3.1), data analysis was performed by R (Version 73 of the surveys were considered invalid because they were incomplete.

145
The univariate chi-square test was performed for each categorical variable included in the 146 model and the latest fasting blood glucose (See Table 1). It can be seen that age, education level, 147 main source of income, marital status, average monthly income, medical payment methods

148
(basic medical insurance for residents, basic medical insurance for employee and free medical 149 service), and hospital history were statistically significant (p<0.05). 8

152
Notes: 1) Hospital history means as long as the patient self-reports that they have been hospitalised once, the 153 answer is yes, otherwise the answer is no; 2) Self-assessment of glycemic management is a relatively subjective Due to the inconsistency of the dimensions, the data in this study was standardised by using lar 163 () in R, so that different feature values had the same scale. This study adopted the latest fasting 164 blood glucose measurement (0 or 1) as the dependent variable. The following factors that may 165 affect the management of diabetes were assigned as independent variables (See Table 2). The  Table 3. 172

189
Comparing the three models from Table 3, it was found that the adaptive-lasso logistic model

194
To predict the results of diabetes management, the established full-variable logistic model, 195 adaptive-lasso logistic model and ridge logistic model were used. Table 4 shown the prediction accuracy of the models with different proportions of the training set. The prediction accuracy 197 of the full-variable logistic model and the adaptive-lasso logistic model was slightly higher 198 than that of the ridge logistic model (See Table 4). However, the prediction accuracy of the 199 three models was not significantly different, and the reason for the small difference in accuracy   University of Traditional Chinese Medicine. All participants signed an informed consent form, 271 and they volunteered to withdraw from the questionnaire process at any time without reason.

272
For illiterate and semi-illiterate participants, data were collected by investigators reading 273 informed consent and reading questionnaires. Data collectors were not involved in the data 274 analysis process to avoid bias.

275
Consent for publication 276 All participants agreed that we use the collected data for academic publications.

277
Availability of data and materials 278 All data analysed during this study are included in this manuscript.

280
All authors of the study stated that there were no conflicts of interest.