인공신경망을 이용한 알코올 의존 선별 검사의 유용성

Alcohol Dependence Screening Test Using Artificial Neural Network Analysis：The Sensitivity and Specificity Study

Objectives：Alcohol screening test using direct questionnaire had the risk of high false negative. The aim of this study is the development of a reliable and feasible alcohol dependence screening test using Multiphasic Minnesota Personality Inventory (MMPI) and artificial neural network analysis (NN). Methods：The subjects, 191 inpatients with alcohol dependence and 148 normal controls, were randomly assigned into learning group and testing group. Both neural network and logistic regression formula for screening alcohol dependence were acquired using both the conventional MMPI 13 scales and MMPI 73 scales including additional 60 scales in learning group (56 alcoholics and 44 normal controls). The MMPI scales of remained testing group, 140 alcoholics and 104 normal controls, were tested with two LR formula (13 and 73 MMPI scales) and two learned NN (13 and 73 MMPI scales) whether each subject is alcohol dependence or not. The sensitivity and specificity, and area under the curve of 4 analysis methods, LR or NN using 13 or 73 MMPI scales were calculated and compared. Results：When using conventional MMPI 13 scales, the sensitivity, specificity, and area under the curve (AUC) of LR and NN were 71.1% vs. 57.8-75.6%, 62.5% vs. 48.1-60.6%, 0.668± 0.036 vs. 0.577±0.037-0.716±0.034, respectively. But, in the case of using 73 MMPI scales including 13 conventional scales, the sensitivity, specificity, and area under the curve (AUC) of LR and NN were 79.3% vs. 83.0-93.3%, 67.3% vs. 67.5-81.7%, 0.733±0.034 vs. 0.765±0.033-0.855±0.027, respectively. Conclusion：When using 73 MMPI scales, the sensitivity and specificity of NN were higher than that of NN using 13 scales or LR using 13 or 73 scales. The test using NN using 73MMPI scales is a useful alcohol dependence screening tool. It is suggested that the application of NN in screening another substance use disorder or game addiction might be useful.