More than 2.5 billion people—over 40 percent of the world's population—are at risk of infection by the mosquito-borne dengue virus. Most dengue fever victims experience flu-like symptoms of fever, headache, muscle pains, vomiting, and a measles-like skin rash. But an unlucky few—mostly children, the elderly, and those who have had the disease in the past—develop an accumulation of fluid in the chest or abdomen that leads to life-threatening hemorrhage. Worldwide, the virus infects 50 to 100 million people each year; 500,000 will contract severe dengue. In poor, tropical countries, around 5 percent of those with severe dengue will die.
So public health professionals in those nations have a substantial interest in knowing where the disease might next break out. Anna L. Buczak, a researcher with the Applied Physics Laboratory, has been attacking that problem with advanced mathematics. She has developed a statistical model that can predict outbreaks weeks before they occur. Dubbed PRISM—PRedicting Infectious Disease Scalable Method—her method sifts statistical variables such as current dengue incidence rate, temperature, rainfall, population, and percentage of private dwellings with running water. Because these data are publicly available or already among the figures compiled by governments, PRISM is inexpensive—important for low-resource settings.
Take Peru. "Dengue is a big problem in Peru," Buczak says, which made the country a natural fit for a study that began in 2011. (For example, in 2012 the country recorded 21,000 cases of dengue fever and at least 32 fatalities.) Plus APL had previously worked with a U.S. Naval Medical Research Unit in Lima, Peru, to implement a surveillance system for infectious diseases, including dengue. As Buczak developed the PRISM model, she could use this surveillance data to see how accurate her forecasts were. A team of 18 people, including medical doctors, statisticians, epidemiological specialists, and technology experts, worked on the study and found that when PRISM predicted a dengue fever outbreak to happen within four weeks, it occurred 81 percent of the time. "We could even predict six to eight weeks ahead with good accuracy," says Buczak.
After its success in Peru, Buczak's team traveled to the Philippines in 2012 to forecast dengue fever there. "Every country is different in terms of geography, disease, and rainfall, so the model needs to be developed differently for individual countries and areas, explains Buczak. "We develop a new model but use the same procedures, the same software as before. For the Philippines, the team tweaked PRISM, added a variable to account for typhoon weather trends, and succeeded again.
It was a natural leap to turn its attention to other mosquito-borne infections, so next the team tackled malaria in South Korea. The preliminary data suggest predictions there were even better—a positive predictive value of 92 percent—than those from Peru and the Philippines, says Buczak.
Now that the method has been developed for multiple sites and multiple diseases, "we have to create a way for [governments and public health officials] to use PRISM. We're in that process now," says Buczak. With a four-week head start, officials who use PRISM can launch public health interventions to reduce the severity of disease outbreaks by targeted insecticide spraying, mosquito net distribution, and health education campaigns reminding people to wear long sleeves and not leave standing water outside as mosquito breeding grounds.
"Many countries still don't have robust surveillance capabilities, so it's hard to know what the potential impact is," says Sheri Lewis, APL's global disease surveillance program manager. "That's what we're hoping we'll change. APL can arm them not only with the predictive tools they need to have an effective public health response, but with the surveillance tools they need to know whether they're making a difference."
Posted in Health, Science+Technology
Tagged public health, applied physics laboratory, epidemiology, big data, data modeling