Reviewing the Disparate Death Toll Estimates for Maria

A report commissioned by the Puerto Rico government from George Washington University in collaboration with the University of Puerto Rico has narrowed the uncertainty around the catastrophic death toll from Hurricane Maria. The GWU/UPR report, Ascertainment of the Estimated Excess Mortality from Hurricane Maria in Puerto Rico (see PDF), estimates that Puerto Rico experienced 2975 “excess deaths” (the number of deaths beyond what would one expect at a given time of year) from September 2017 through February 2018.

The government of Puerto Rico asked for this report as a way to objectively assess the death toll from Maria, which has been a hugely controversial point for almost a year. Last December, the territorial government had set the official death toll at 64, which was the number also accepted by the U.S. government. Multiple media outlets carried out their own calculations of excess deaths, resulting in death-toll estimates of at least 499 (CNN, November 2017) and 1052 (New York Times, December 2017). This past spring, a study in the New England Journal of Medicine projected a far higher death toll in the thousands based on interviews with households across Puerto Rico (see our post from May 29). In early August, the Puerto Rico government noted in a report to Congress (see PDF) that 1427 excess deaths had been documented in the four months following Maria. At that point, the government said it would wait for the now-released GWU/UPR study before updating the official death toll.

The new study estimates that Puerto Rico’s population has fallen from 3,327,917 in mid-September 2017 to 3,048,173 in mid-February 2018, a loss of more than 280,000 residents. The researchers factored this post-hurricane migration into their mortality estimates. As one might expect, the risk of death was higher for people at lower socioeconomic rungs of the ladder. For men aged 65 or older, death rates continued to be elevated through February 2018.
 

Figure 1. Age-standardized monthly mortality by year (per 10,000 inhabitants), Puerto Rico, 2010-2011 to 2017-2018. The black line indicates the mortality rate per 10,000 people per month based on reported deaths and assuming no migration from the island post-Maria (the “census scenario”). The red line indicates an upward adjustment to the per-capita mortality rate based on massive migration from Puerto Rico since Maria that was not yet reflected in census data (the “displacement scenario”). Age-adjusted mortality rates for Puerto Rico tend to be higher in the winter and early spring, declining in the summer months. Mortality had been slowly declining from 2010 on, but it increased markedly in the period after September 2017, most dramatically under the displacement scenario. Image credit: Ascertainment of the Estimated Excess Mortality from Hurricane Maria in Puerto Rico (see PDF).

Numbers vs. ranges

One reason the death toll numbers from Maria may have seemed so confusingly divergent is because media have tended to focus on the single best estimate from each study. Calculating death tolls is a surprisingly difficult task, for a variety of reasons, and so the most thorough analyses include not only a best estimate but also a confidence interval, which conveys the likelihood that the people or data sampled in the study accurately reflect the true distribution of the variable in question. Here are the numbers from the two most exhaustive studies, the one from NEJM published in May and the one released this week by GWU/UPR.

NEJM best estimate = 4645
After accounting for household-size factors = 5740
95% confidence interval = 1506 > 9989

GWU/UPR best estimate = 2975
95% confidence interval = 2658 > 3290

These statistics tell us that:

—The new GWU/UPR study falls well within the confidence interval of the NEJM study, even though the latter had a much higher central estimate, so the two studies are not in conflict.

—The GWU/UPR study has a much smaller range of uncertainty than the NEJM, study which gives us added confidence that somewhere close to 3000 people died as a direct or indirect result of Maria.

The unique circumstances following Maria may have led to a greater rate of indirect deaths in Puerto Rico than we have seen in other U.S. hurricanes. However, not all hurricanes have been examined using excess-deaths methodology. One 2007 study that drew on death notices in the New Orleans Times-Picayune, the leading local newspaper, estimated that there were 2358 excess deaths in the local area during the months of January – June 2006, in the aftermath of Hurricane Katrina, compared to the years 2002–2004. These would have been on top of all Katrina-related deaths that occurred from September to December 2005. It is unclear how many of these 2006 deaths may be included in a widely accepted U.S. death toll for Katrina of 1833, as reflected in the National Hurricane Center’s report on the storm.

The bottom line: we may never know exactly how deadly Maria was compared to other catastrophic U.S. hurricanes, but we now have more solid statistical evidence to back up the overwhelming signs that Maria was a human tragedy of vast proportions. For more on this topic, see the excellent writeup by Miami broadcast meteorologist John Morales, "Many Are Misreporting Hurricane Maria's Death Toll. Here's the Messy Reality."

We will have a new post later today on multiple areas of concern across the Atlantic and Pacific tropics.