Application of ANN on prediction of abnormal DM/IGT

  Objective   To discuss the application potential of ANN on epidemologic classification of disease.

  Methods   LVQNN and discriminant analysis were applied in terms of the data of epidemiological survey and the case-control study in a general hospital.

  Results   The structure of LVQNN was 25→13→3, the sensitivity of model prediction to DM and IGT was 70.45%, 64.79% respectively, the specificity was 100.00%, the coincident rate was 96.98%, and the veracity to abnormal blood glucose individuals was 92.45%.Through stepwise discriminant analysis, the discriminant equations were established including 11 variables, the sensitivity of equaions to DM and IGT was 67.05%, 60.56% respectively, the specificity was 89.75%, the coincident rate was 87.34%, and the veracity to abnormal blood glucose individuals was 85.53%.To the data from a general hospital, all the collators and 92.37% of DM patients were recognized by LVQ with a veracity of 96.19%, excelling to discriminant analysis in predicting the status of disease. Further analysis on 30 cases with missing values showed that the disagreement ratio of LVQ was 1/30, lower than that of discriminant analysis of 7/30.

  Conclusion   Compared with the conventional statistic method, LVQ not only had better prediction precision, but also can treat the data with missing values satisfactorily, and had no limit to the type and distribution of data. It would be a new powerful method to epidemiologic prediction.