Related Experiment Video
Updated: Mar 9, 2026
Using Zebrafish Models of Human Influenza A Virus Infections to Screen Antiviral Drugs and Characterize Host Immune Cell Responses
Published on: January 20, 2017
Development of machine-learning models to diagnose influenza among international travellers based on symptoms and
Yusuke Asai1, Kei Yamamoto2, Hidetoshi Nomoto2
1Japan Institute for Health Security, National Center for Global Health and Medicine, Disease Control and Prevention Center, Tokyo, Japan; Japan Institute for Health Security, Center for Clinical Sciences, Department of Data Science, Tokyo, Japan.
Introduction:
The importation of infectious diseases has increased dramatically in recent years. The diagnosis of such diseases has traditionally been based on symptoms, as well as blood and biochemical tests. In this study, we developed machine-learning models to diagnose influenza cases among international travellers using information on the number of infected individuals in destination countries and the incubation period, which have not been utilised for diagnosis.
Methods:
This study examined cases recorded in Japan Registry for Infectious Diseases from Abroad (https://jrida-jprecor.ncgm.go.jp/en/j-rida/index.html), which included influenza test results and information on travel destination and duration. Multivariable logistic regression and machine-learning methods were used to build diagnostic models, with symptoms and an epidemiological score calculated from information on the number of cases in the destination country and the incubation period serving as explanatory variables.
Results:
Logistic regression analysis revealed that symptoms such as fever (p < 0.001) and cough (p < 0.001), as well as the epidemiological score (p < 0.001), were significant predictors of influenza infection. The machine-learning models were then developed and tested using training and test data, respectively. The constructed models showed good specificity and high accuracy. Among these models, the neural network had the highest accuracy (93%).
Discussion:
Adding epidemiological information to the usual clinical information can improve diagnostic accuracy. Machine-learning models capable of handling a wide variety of data would be particularly beneficial for diagnosis. The knowledge gained should also be used to raise awareness among travellers.
