The Ebola epidemic that took off last year in three west African countries—Guinea, Liberia and Sierra Leone —is finally fading. The last three weeks have each seen around 120 new cases, the lowest level since July 2014. In total, around 22,000 people are thought to have been infected, and 9,000 to have died. But predictions last year were much higher, including a worst-case scenario by America’s Centres for Disease Control and Prevention (CDC) of 1.4m cases, reported and unreported, by January 20th. How are such predictions made, and how can they go wrong?
Disease predictions come from mathematical models that group the population at risk of a disease into categories and describe how people move between them. Both the categories and the flows depend on the disease. With Ebola, there is a long lag between infection and the first symptoms, during which it is not contagious. And recovered sufferers are immune. So an Ebola model will group the “susceptible” (who have not yet had the disease), “exposed” (infected but not yet showing signs of the disease), “infectious” (ill and contagious) and “removed” (no longer infectious because they have recovered or died). Estimates for the sizes of these groups and how people move between them—including how many people catch the disease from each infectious person, how long people remain infectious and how likely they are to die—are used to build a computer model showing how the spread of the disease is likely to unfold. The CDC’s Ebola model also sub-divided infectious people into three smaller groups: isolated in hospital, partially isolated at home or elsewhere in the community, and unisolated. And its highest predictions were swollen by the assumption that for every reported case, 1.5 more went unreported.
The results of any mathematical model will only be as good as the numbers that go into it. When, like Ebola, a disease is both novel and raging in developing countries where health-care statistics are close to non-existent, those numbers are quite likely to be far off the mark. The most recent research suggests that each Ebola sufferer infects fewer people than had been thought, for example. It also appears to strike people in clusters, so there are fewer unreported cases than models assumed. But a bigger source of error is that human beings change their behaviour in response to perceived risk—for example by increasing health spending (though aid to the three affected countries took time to materialise, it did eventually pour in) or taking precautions not to get infected (as information about Ebola spread, traditional burial practices that involve close contact with dead bodies became much less common).
The speed and scale of such adaptations are so hard to predict that researchers often leave them out entirely. But that makes models less likely to be accurate over the medium or long term. And there is a deeper dilemma: if an alarmist prediction provokes exactly the response that makes it less likely to come to pass, might it perhaps be justified, or even essential? Some of those studying Ebola say that one aim of the high casualty estimates was to get the world to pay attention. But that is a dangerous line of reasoning. Each new scare that turns out to be unwarranted makes it harder to catch that attention the next time—and more likely that the doom-mongers’ predictions will be fulfilled.
Source: The Economist