2003 Volume 11 Issue 1 Pages 96-102
MTS was used to select tests and lifestyle data useful for detecting people at high risk of developing diabetes, and to predict changes in blood glucose levels. The subjects were 935 adult males who had received heath checkups repeatedly over a period of at least ten years, and whose fasting plasma glucose (FPG) levels were initially within the normal range. The reference space consisted of 596 individuals who maintained FPG Values of less than 110 mg/dl over the ten-year observation period; new cases of diabetes (subjects who first exceeded 126 mg/dl during the observation period) were taken as the signal. Twenty-eight categories of health checkup data were used, including age and family data, 15 lifestyle attributes, and 12 test results. Data categories were selected by using the difference between the logarithmic FPG of the subjects diagnosed with diabetes and the average logarithmic FPG ofthe subjects in the reference space as the signal value. The second-year FPG was predicted using the selected data categories. Six test results, two dietary attributes, two alcoholic intake attributes,and two exercise attributes were selected for predicting the second-year FPG from the first-year data. For the third- and fourth-year predictions, the selection included fewer effective test results and more lifestyle attributes. A slight improvement in S/N ratio was obtained by using data from more than one year. When the second-year FPG values were predicted by MTS and the predicted values were compared with the actual measured values by multiple regression analysis (the stepwise method), the estimates made by MTS were found to track the measured values closely even for cases with high blood glucose levels. In an attempt to predict the onset of diabetes from existing health checkup data by using MTS, it was found important to employ not only test data but also lifestyle data obtained by verbal questioning. Even estimated values gave substantially reasonable results.
