Hyperglycemia screening based on survey data: an international instrument based on WHO STEPs dataset

Abstract

Abstract Background Hyperglycemia is rising globally and its associated complications impose heavy health and economic burden on the countries. Developing effective survey-based screening tools for hyperglycemia using reliable surveillance data, such as the WHO STEPs surveys, would be of great importance in early detection and/or prevention of hyperglycemia, especially in low or middle-income regions. Methods In this study, data from the nationwide 2016 STEPs study in Iran were used to identify socioeconomic, lifestyle, and metabolic factors associated with hyperglycemia. Furthermore, the ability of five commonly used machine learning algorithms (random forest; gradient boosting; support vector machine; logistic regression; artificial neural network) in the prediction of hyperglycemia on STEPs dataset were compared via tenfold cross validation in terms of specificity, sensitivity, and the area under the receiver operating characteristic curve. Results A total of 17,705 individuals were included in this study, of those 29.624% (n = 5245) had (undiagnosed) hyperglycemia. Multivariate logistic regression analysis showed that older age (for the elderly group: OR = 5.096; for the middle-aged group: OR = 2.784), high BMI status (morbidly obese: OR = 3.465; obese: OR = 1.992), having hypertension (OR = 1.647), consuming fish more than twice per week (OR = 1.496), and abdominal obesity (OR = 1.464) were the five most important risk factors for hyperglycemia. Furthermore, all the five hyperglycemia prediction models achieved AUC around 0.70, and logistic regression (specificity = 70.22%; sensitivity = 70.2%) and random forest (specificity = 70.75%; sensitivity = 69.78%) had the optimal performance. Conclusions This study shows that it is possible to develop survey-based screening tools for early detection of hyperglycemia using data from nationwide surveys, such as WHO STEPs surveys, and machine learning techniques, such as random forest and logistic regression, without using blood tests. Such screening tools can potentially improve hyperglycemia control, especially in low or middle-income countries.

Keywords

• Hyperglycemia
• Logistic regression
• Machine learning
• Prediction models
• Random forest
• Screening