“Excess Deaths” and the Toll in Puerto Rico from Hurricane Maria

A bombshell analysis released Tuesday in the New England Journal of Medicine (NEJM, open-access paper) reported that the death toll from Hurricane Maria in Puerto Rico was likely far greater than any estimates to date. The study’s initial estimate of the death toll through December 2017 was 4,645, but when adjusted for household-size factors (see below), the revised best estimate was 5,740, with a 95% confidence range that somewhere between 1,506 and 9,889 people died as a result of Maria in 2017. As shocking as these numbers are, they might still underestimate the true impact of the hurricane, as indirect fatalities no doubt extended beyond the study period into early 2018.

Led by Harvard graduate student and epidemiologist Nishant Kishore, the researchers carried out surveys of 3,299 households in 104 barrios (districts), representing both densely and sparsely populated areas. There were 56 deaths reported in these households for the period from September 20 to December 31, 2017. About a third of those deaths were attributed by household members to medical care being delayed or unavailable. The average household in the survey went 84 days without electricity, 68 days without running water, and 41 days without cellphone service. A full 14% of households had one or more members who could not access medications, and 9% of households could not access 911 emergency help by telephone.

The authors calculated that the 2017 death rate was 62% higher than during the same period in 2016, which extrapolates to the estimated death toll cited above. The wide 95% range corresponds to uncertainty rooted in the relatively small sample size of the households surveyed. It’s worth noting that even the low end of the 95% estimate is far greater than the official death toll.

The NEJM study was preceded by a non-peer-reviewed New York Times analysis of Maria’s impact that also employed the excess-deaths concept. The NYT estimated that Maria caused 1052 deaths from the time of the hurricane through October 31.

The new study adjusted for a variety of potential biases, such as household size (larger households were more likely to have someone at home when the in-person surveyors came by) and “survivor bias” (if the sole person in a household died, there would be nobody left for the surveyors to interview). The effects of post-hurricane migration from Puerto Rico were also taken into account. Most of the authors’ assumptions were on the conservative side, meaning that the death toll is likely higher than calculated. “We have made data available for transparency and to encourage other researchers to conduct more sophisticated analyses,” they wrote.

Figure 1. Aerial photo of flooding in the coastal town of Loiza, on the north shore of Puerto Rico, on September 23, three days after Hurricane Maria tore across the island. Image credit: Dennis M. Rivera Pichardo for The Washington Post via Getty Images.

Calculating the true impact of a disaster

How could the new numbers be so much larger than the most recent official death toll of 64 from the Puerto Rico government? In disasters such as Maria, where everyday life is scrambled for long periods by the hurricane’s damage and disruption, governments are often slow to attribute particular deaths to the disaster, and there may be various pressures not to do so. One well-established tool used by epidemiologists to get around this problem is a measurement called “excess deaths.” The basic idea is easy to grasp: by comparing death rates in one time period (such as September-December) to the same time period in past years, one can calculate the number of additional deaths that occurred. If there are no other major events in play, then the margin of additional fatalities can serve as a raw index of excess deaths—i.e., deaths that would not have occurred had the disaster not happened.

Excess deaths are an especially useful metric in heat waves, which take many lives through air pollution and heat impacts in ways that aren’t as obvious as a direct hurricane or tornado death. The enormity of Europe’s catastrophic 2003 heat wave was not fully recognized until analyses of excess deaths were carried out over the subsequent months. The initial fatality estimate of around 10,000 was sobering enough, but it soon became apparent that more than 70,000 people may have died as a result of the heat wave. (A nation-by-nation calculation carried out in 2015 by Dr. Jeff Masters using the EM-DAT database came up with a total of 71,310.) Many of the victims were elderly, enduring the heat in un-air-conditioned spaces amid the heat-trapping environments of cities such as Paris while relatives were on holiday.

Calculating excess deaths can get more complicated if a disaster affects large numbers of older people, as heat waves often do. In coldly clinical terms, the heat wave might be seen as taking the lives of people in poor health who would have otherwise died within the following months, a phenomenon known as mortality displacement (or even more grimly, "harvesting"). For example, one might see a rise in the death rate in the two months following a major heat wave, followed by a dip afterward due to the harvesting effect. Estimates of the harvesting effect in heat waves vary widely from event to event, from as low as 6% to as high as 71%.

In the case of Maria, the NEJM study found that excess deaths occurred across the age spectrum, suggesting that the mortality displacement effect was mimimal. In short, there is every indication from this study that Hurricane Maria was responsible for the premature deaths of thousands of Puerto Ricans—a stunning outcome largely attributable to a lack of basic services weeks to months after the hurricane struck.

Dr. Jeff Masters contributed to this post.