This article is more than 5 years old. Information may no longer be current.
Forecasting model predicts subtropical influenza epidemics
Click Here to Manage Email Alerts
An influenza forecasting system may be the first to accurately predict the timing and magnitude of epidemics in subtropical climates, according to data recently published in PLoS Computational Biology.
“These forecasts provide information at lead times that can be valuable for both the public and health officials,” Jeffrey Shaman, PhD, of the Mailman School of Public Health at Columbia University, said in a press release. “Individuals may choose to get a flu vaccine to protect themselves against infection, while officials can anticipate how many vaccines and other supplies are needed, as well as the number of clinicians and nurses needed.”
Using weekly records from a surveillance network of approximately 50 outpatient clinics, researchers from Columbia University and the University of Hong Kong developed a susceptible-infected-recovered (SIR) epidemic model. With this, two forecast systems were built to specifically predict influenza epidemics in subtropical climates, where outbreaks can occur irregularly and at any time of the year. Researchers “trained” the systems using previously reported surveillance data, and, by repeating each run 100 times to account for random effects from system initialization, they evaluated their accuracy by retroactively forecasting H1N1, A strain (H3N1) or B strain influenza epidemics in Hong Kong from 1998 to 2013.
“Hong Kong is a crossroads to Asia and the rest of the world, serving as an entry and exit point for flu outbreaks year-round, and the region of South East Asia with Hong Kong at its center is often referred to as the global epicenter for flu,” Benjamin J. Cowling, PhD, of the University of Hong Kong, said in the press release.
Both forecast systems accurately predicted an ongoing epidemic, with the more successful of the two (SIR particle filter) demonstrating 90% sensitivity and 95% specificity; however, these results were lower for H1N1 and Type B. While each system was capable of predicting peak timing and peak magnitude, the SIR system using a particle filter was again more promising. Neither system was able to predict outbreak onset or duration before its occurrence.
“As our understanding of influenza transmission dynamics in subtropical and tropical regions improves in the future, more mechanistic and detailed models could be used in conjunction with the filters,” the researchers wrote. “These more complicated model-filter systems could further improve the forecast performance.”
Disclosures : Cowling reports receiving research funding for MedImmune and Sanofi Pasteur, and consults for Crucell NV. Shaman reports consulting for JWT and Axon Advisors, and is a partial owner of SK Analytics. All other authors report no relevant financial disclosures.
Click Here to Manage Email Alerts