Kamal Gholipour1,2, Mohammad Asghari-Jafarabadi3,4, Shabnam Iezadi5, Ali Jannati1,2, Sina Keshavarz6. 1. Iranian Center of Excellence in Health Management, School of Management and Medical Informatics, Tabriz University of Medical Sciences, Tabriz, Islamic Republic of Iran. 2. Tabriz Health Services Management Research Center, Tabriz University of Medical Sciences, Tabriz, Islamic Republic of Iran. 3. Road Traffic Injury Research Center, Health Management and Safety Promotion Research Institute, Tabriz University of Medical Sciences, Tabriz, Islamic Republic of Iran. 4. Department of Statistics and Epidemiology, Faculty of Health, Tabriz University of Medical Sciences, Tabriz, Islamic Republic of Iran. 5. Social Determinants of Health Research Center, Health Management and Safety Promotion Research Institute, Tabriz University of Medical Sciences, Tabriz, Islamic Republic of Iran. 6. Public Health and Preventive Medicine, University of Social Welfare and Rehabilitation Sciences, Tehran, Islamic Republic of Iran.
Abstract
BACKGROUND: Type 2 diabetes mellitus (T2DM) is a metabolic disease with complex causes, manifestations, complications and management. Understanding the wide range of risk factors for T2DM can facilitate diagnosis, proper classification and cost-effective management of the disease. AIMS: To compare the power of an artificial neural network (ANN) and logistic regression in identifying T2DM risk factors. METHODS: This descriptive and analytical study was conducted in 2013. The study samples were all residents aged 15-64 years of rural and urban areas in East Azerbaijan, Islamic Republic of Iran, who consented to participate (n = 990). The latest data available were collected from the Noncommunicable Disease Surveillance System of East Azerbaijan Province (2007). Data were analysed using SPSS version 19. RESULTS: Based on multiple logistic regression, age, family history of T2DM and residence were the most important risk factors for T2DM. Based on ANN, age, body mass index and current smoking were most important. To test for generalization, ANN and logistic regression were evaluated using the area under the receiver operating characteristic curve (AUC). The AUC was 0.726 (SE = 0.025) and 0.717 (SE = 0.026) for logistic regression and ANN, respectively (P < 0.001). CONCLUSIONS: The logistic regression model is better than ANN and it is clinically more comprehensible.
BACKGROUND: Type 2 diabetes mellitus (T2DM) is a metabolic disease with complex causes, manifestations, complications and management. Understanding the wide range of risk factors for T2DM can facilitate diagnosis, proper classification and cost-effective management of the disease. AIMS: To compare the power of an artificial neural network (ANN) and logistic regression in identifying T2DM risk factors. METHODS: This descriptive and analytical study was conducted in 2013. The study samples were all residents aged 15-64 years of rural and urban areas in East Azerbaijan, Islamic Republic of Iran, who consented to participate (n = 990). The latest data available were collected from the Noncommunicable Disease Surveillance System of East Azerbaijan Province (2007). Data were analysed using SPSS version 19. RESULTS: Based on multiple logistic regression, age, family history of T2DM and residence were the most important risk factors for T2DM. Based on ANN, age, body mass index and current smoking were most important. To test for generalization, ANN and logistic regression were evaluated using the area under the receiver operating characteristic curve (AUC). The AUC was 0.726 (SE = 0.025) and 0.717 (SE = 0.026) for logistic regression and ANN, respectively (P < 0.001). CONCLUSIONS: The logistic regression model is better than ANN and it is clinically more comprehensible.
Authors: Abdallah Y Naser; Ian C K Wong; Cate Whittlesea; Hassan Alwafi; Amjad Abuirmeileh; Zahra Khalil Alsairafi; Fawaz Mohammad Turkistani; Nedaa Saud Bokhari; Maedeh Y Beykloo; Dalal Al-Taweel; Mai B Almane; Li Wei Journal: PLoS One Date: 2019-10-24 Impact factor: 3.240