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

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