Evaluation of multinomial logistic regression models for predicting causative pathogens of food poisoning cases

The Journal of Veterinary Medical Science

Masashi HYODO¹  Hideya INOUE²  Tomoyuki SUZUKI³ 

[1] Department of Mathematical Sciences, Graduate School of Engineering, Osaka Prefecture University, Sakai, Osaka 599-8531, Japan;Department of Veterinary Science, Graduate School of Life and Environmental Sciences, Osaka Prefecture University, Izumisano, Osaka 598-8531, Japan;Shiga Prefectural Institute of Public Health, 13-45 Gotenhama, Otsu, Shiga 520-0834, Japan

关键词: causative pathogen;    food poisoning;    leave-one-out cross validation;    multinomial logistic regression;   

DOI  :  10.1292/jvms.17-0653

学科分类：兽医学

来源: Japanese Society of Veterinary Science

【 摘 要 】

In cases of food poisoning, it is important for food sanitation inspectors to determine the causative pathogen as early as possible and take necessary measures to minimize outbreaks. Interviews are usually conducted to obtain epidemiological information to aid in the rapid determination of the cause. However, the current method of determining the causative pathogen has the disadvantage of being reliant upon the experience and knowledge of food sanitation inspectors. Here, we analyzed 529 infectious food poisoning incidents reported in five municipalities in the Kinki region to develop a tool for evaluation using a multinomial logistic regression model, which can predict the causative pathogen based on the patients’ epidemiological information. This tool predicts the most probable cause of the incident by generating a list of pathogens with the highest probability. As a result of leave-one-out cross validation, the agreement ratio with the actual pathogen was 86.4%, and this ratio increased to 97.5% when the agreement was judged by including the true pathogen within the top three pathogens with the highest probability. In cases where the difference of probability between the first and second candidate pathogen was ≥50%, the agreement ratio increased to 94.2%. Using this tool, it is possible to accurately estimate the causative pathogen at an early stage based on patient information, and this will further help narrow the target of investigations to identify causative agent, thereby leading to a prompt identification, which can prevent the spread of food poisoning.

【 授权许可】

Unknown