Statistical Models Forecast Locations Likely to See COVID-19
Statistical models forecast the spread of COVID-19 based on pre-pandemic travel from Wuhan, China.
Statistical models incorporating data of international air travel from Wuhan, China—the epicenter of the COVID-19 outbreak—have been applied to identify locations likely to import undetected cases.
Pablo Martinez De Salazar, MD, PhD, Harvard T.H. Chan School of Public Health, and colleagues describe a predictive algorithm that could be used to alert vulnerable locales, in the US Centers for Disease Control and Prevention’s (CDC) journal, Emerging Infectious Diseases.
"Early detection of cases and (implementing) appropriate control measures can reduce the risk for self-sustained transmission in all locations," Salazar and colleagues indicate.
Kamran Khan, MD, founder and chief executive officer of BlueDot, and previously a hospital infectious disease specialist in Toronto described in an interview with WIRED Magazine that he determined to find a better way to track diseases after his experience with the SARS epidemic of 2003.
"In 2003, I watched the virus overwhelm the city and cripple the hospital," Khan said. "There was an enormous amount of mental and physical fatigue, and I thought, 'Let's not do this again'."
Both the models from Martinez de Salazar's group and from BlueDot utilize air travel volume data from the Wuhan region prior to the quarantine measures, to project directions and rates of transmission in, and from destination cities.
Martinez de Salazar and colleagues incorporated data on imported and reported cases of SARS-CoV-2 infection, on daily air travel volume, and on surveillance capacity of the locations. They characterize their model as a simple generalized linear model, requiring only 1 regression coefficient and no extra parameters.
"Our model can be adjusted to account for exportation of cases from locations other than Wuhan as the outbreak develops and more information on importations and self-sustained transmission becomes available," they indicate. "One key advantage of this model is that it does not rely on estimates of incidence or prevalence in the epicenter of the outbreak."
The Imported and reported cases were obtained from the World Health Organization (WHO), which had aggregated them by destination. Air travel volume data were obtained from a separate network-based modeling study that reported volume estimates for 27 locations outside mainland China that were most connected to Wuhan. Travel data were cut off at February 4—the date that exported cases from Hubei Province was significantly reduced—following imposition of travel restrictions for the province on January 23. High surveillance locations were defined as those with a Global Health Security Index (GHS) for category 2, Detection and Reporting, above the 75th quantile.
The investigators assumed that the cumulative imported and reported case counts across 49 high surveillance locations followed a Poisson distribution from the beginning of the epidemic until February 4. They then compared predictions from the model with imported and reported cases across 194 locations from the GHS Index, excluding China, as the epicenter of the outbreak.
Martinez de Salazar and colleagues found that daily air travel volume positively correlates with imported and reported case counts of SARS-CoV-2 infection among high surveillance locations. An increase of flight volume by 31 passengers/day is associated with 1 additional expected imported and reported case.
"To prevent other cities and countries from becoming epicenters of the SARS-CoV-2 epidemic," Martinez De Salazar and colleagues warn, "substantial targeted public health interventions are required to detect cases and control local spread of the virus,".
As the epidemic turned into a pandemic before their warning could appear in the CDC journal's early release section; however, it appears that the best predictive tools are of marginal use if the information isn't quickly assimilated into policies and implemented into practices of early containment.