Can Future Epidemics Be Predicted? – Forbes


Medical staff push a patient on a gurney to a waiting medical helicopter at the Emile Muller … [+] hospital in Mulhouse, eastern France, to be evacuated on another hospital on March 17, 2020, amid the outbreak of COVID-19.
I’ve written before about the need for infectious disease intelligence and whether or not we can insure against damages from future outbreaks. Both ideas assume that epidemics can, to some extent, be predicted. But can they?
The lack of control and idiosyncrasies of the current Covid-19 pandemic (like multiple “waves” of transmission), might cause some to wonder if predicting epidemics is a lost cause. A comparison of the performance of 27 individual models made by many of the best academic scientists showed high high variability in forecast skill across time, geospatial units, and forecast horizons. Only just more than half of the models evaluated showed better accuracy than simply assuming the number of cases per week will stay the same for the next 4 weeks.
Nonetheless, I’m optimistic future epidemics can be predicted. However, we need to change our expectations. Like other catastrophic events (forest fires, hurricanes, tornadoes), unique events are unlikely to be predictable. No one can say exactly when the next tropical storm will appear on the horizon or where the next lightning strike will start a blaze. But these events are predictable in aggregate. We need to adopt a statistical concept of disease prediction.
While we cannot predict in advance exactly how many hurricanes will occur this year or how bad they will be, we know with great confidence that climate change is a risk factor increasing the frequency and severity of hurricanes. Our knowledge of this and all the other risk factors for hurricanes allows us to make a statistical prediction for the coming season.
Similarly, we have known for decades that there are identifiable risk factors associated with disease emergence. For instance, microorganisms that are known to be zoonotic (i.e. transmitted from animals to people) are more likely to emerge as human pathogens than other infectious agents. In fact, around 75% of emerging diseases originate in animals. Viruses are four times more likely to emerge in human populations than the average parasite (a relative risk of 400%) whereas helminths (parasitic worms) are only one fourth as likely to cause novel diseases in humans (a relative risk of only 24%). 
A recent study I contributed to sought to statistically characterize the one hundred largest outbreaks of zoonotic pathogens. Our idea was that if there is a profile for really bad epidemics, then maybe we can intervene to keep those conditions from materializing. Overall, we looked at forty-eight different factors that are known to contribute to outbreak conditions, including things like transmission route, location, and known association with wild or raised animals. We found that the characteristics of large zoonotic outbreaks with thousands to millions of cases differed consistently from those of smaller outbreaks. Specifically, large outbreaks tended to be multi-causal, with numerous contributing factors at play. They also tended to be related to large-scale environmental and demographic factors such as changes in vector abundance, human population density, unusual weather conditions and water contamination.
The second expectation we need to change is how we make infectious disease predictions. I have written before that scientific models are instruments, not oracles. As with any major task that has multiple components, say building a house, more than one tool will be needed. The statistical models I described above may be useful for the task of anticipating where and under what conditions future infectious diseases emerge.
Predicting where and how fast they will spread is another task. Anticipating public policies and individual behavioral responses is yet another. Predicting the rise of new variants is still another. For each task we must build a better toolkit.
Prompted by the first major zoonotic epidemics of the twenty-first century — West Nile virus in the Americas (beginning in 1999) and the first global SARS outbreak (in 2003) — scientists have made steady progress on the models and statistical techniques used for disease forecasting and outbreak analytics. But, as is often the case with preventative maintenance, the work has been underfunded in comparison with the need. Back in 2016, prior to the Covid-19 pandemic, the National Academy of Sciences reported that “the annualized expected loss from potential pandemics is more than $60 billion.  Against this, we propose incremental spending of about $4.5 billion per year — a fraction of what we spend on other risks to humankind.”  Although it is politically difficult to advance massive research investments when the costs appear only sporadically and in hindsight, the expense is nonetheless very well justified. 
Finally, the game changer for disease prediction will be improved data. Early warning systems for earthquakes rely on sophisticated and widely distributed instruments like the USGS’s ShakeAlert public alert program. Severe weather forecasts rely upon a network of ground based doppler radars, polar and geostationary satellites, radiosondes, automated surface-observing systems, supercomputers, and an Advanced Weather Information Processing System (AWIPS). There is nothing currently analogous for predicting the course of epidemics. The data we rely on — case counts, hospitalizations, deaths, vaccines administered — are almost exclusively compiled by hand, are error-prone, and days to weeks out of date when they are finally made available. Moreover, those data are generally not available at the high spatial resolution that would really boost predictive performance.
We have made progress in some specific areas, such as digital contact tracing and molecular epidemiology, which provide data with higher fidelity, higher resolution, and lower latency. But, a revolution in infectious disease forecasting will require overhauling our approach to data collection and management. What we really need is a strategic data stockpile along with real time systems for data collection and dissemination. 
Statistical, pluralistic, data-driven prediction of future epidemics is indeed possible. Actually, it is not only possible, it is imperative.
Full coverage and live updates on the Coronavirus



Demon Slayer season 2 blows the budget on another masterful episode – HITC – Football, Gaming, Movies, TV, Music

Previous article

Intensity and frequency of extreme novel epidemics – pnas.org

Next article

You may also like



More in 今日NEWS