Study on clinical classification of Japanese encephalitis by Bayes discriminant analysis

Abstract

Objective: To construct Bayes discriminant function for clinical classification of common and severe Japanese encephalitis (JE) cases, and to identify cases accurately with quantitative indicators. Methods: Samples of confirmed common and severe JE cases reported by the epidemic surveillance system of Gansu Provincial Center for Disease Control and Prevention from 2005 to 2017 were collected. Non-conditional logistic regression analysis and Bayes stepwise discriminant analysis were used to screen meaningful clinical indicators, so as to construct and evaluate Bayes discriminant function. Results: There were 256 common JE cases and 257 severe JE cases. There were no significant differences in sex, age and occupation distributions between the two groups (P>0.05) and there was significant difference in case fatality rate (P<0.05). Non-conditional logistic regression analysis and Bayes stepwise discriminant analysis, combined with using related literature, to screen 11 clinical indicators for the construction of Bayes discriminant function. Interactive validation showed that the sensitivity of discriminant function was 71.48% (95%CI: 65.53%-76.93%) and the specificity was 73.93% (95%CI: 68.11%-79.19%). The area under ROC curve was 0.761 (95%CI: 0.720-0.803) and the total accuracy rate was 72.71%. Conclusion: Bayes discriminant function can be used to identify common and severe JE cases more accurately, which is helpful for the reasonable treatment and good prognosis of JE patients.