Application of support vector machine modeling for prediction of common diseases: the case of diabetes and pre-diabetes

Abstract

Background: We present a potentially useful alternative approach based on support vector machine (SVM) techniques to classify persons with and without common diseases. We illustrate the method to detect persons with diabetes and pre-diabetes in a cross-sectional representative sample of the U.S. population.

Methods: We used data from the 1999-2004 National Health and Nutrition Examination Survey (NHANES) to develop and validate SVM models for two classification schemes: Classification Scheme I (diagnosed or undiagnosed diabetes vs. pre-diabetes or no diabetes) and Classification Scheme II (undiagnosed diabetes or pre-diabetes vs. no diabetes). The SVM models were used to select sets of variables that would yield the best classification of individuals into these diabetes categories.

Results: For Classification Scheme I, the set of diabetes-related variables with the best classification performance included family history, age, race and ethnicity, weight, height, waist circumference, body mass index (BMI), and hypertension. For Classification Scheme II, two additional variables--sex and physical activity--were included. The discriminative abilities of the SVM models for Classification Schemes I and II, according to the area under the receiver operating characteristic (ROC) curve, were 83.5% and 73.2%, respectively. The web-based tool-Diabetes Classifier was developed to demonstrate a user-friendly application that allows for individual or group assessment with a configurable, user-defined threshold.

Conclusions: Support vector machine modeling is a promising classification approach for detecting persons with common diseases such as diabetes and pre-diabetes in the population. This approach should be further explored in other complex diseases using common variables.

Figure 1

Demonstration of finding a separating hyperplane in high dimensional space vs in low dimensional space.

Figure 2

ROC curves for Classifications Schemes I (a) and II (b) with SVM models and logistic regression models. Note: see Table 1 for the definitions of Classification Schemes I and II.