Nearly nine months after Hurricane Maria slammed into Puerto Rico, nobody knows how many people died because of it. Estimates range from the government’s official toll of 64—broadly seen as a gross undercount—to Harvard University researchers’ 4,645, an extrapolation that some say is wildly inflated.

Quartz and CPI set out to get individual stories. In December, we asked Puerto Ricans to report Maria-related deaths through an online survey. Hundreds of people wrote back, making for a total 319 collected cases (once combined with earlier counts by CPI and the government). At least 244 deaths have been verified—the largest number of individual deaths yet reported.

These are real people, whose stories complicate the official causes of death. For nearly 40 days after Hurricane Maria, Mariana Marrero Figueroa waited for heart surgery. The operation was never performed due to lack of electricity, and Figueroa ultimately died of renal failure at age 72. “I want my mother to count,” says María Gómez Marrero, who reported the death.

It’s hurricane season once again. As climate change turns extreme weather into a more common occurrence, better data on the long tail of Maria’s tragedy might help save lives around the world.

Why is the official death count so low?

There are at least eight different Maria-related death tallies in Puerto Rico. The local government’s count is the lowest of all.

The Department of Public Safety says it issued guidelines for hospitals and medical examiners to identify hurricane-related deaths. But out of the nearly 30 long-form certificates we received from survey respondents as evidence, not a single one tied the cause of death to Maria-related conditions.

For example, Marielly Fuentes Torres, 44, slid off a road and into a cliff as she was driving to see her mother, says her husband Gabriel Díaz. A clean-up crew that had cleared a Maria-induced landslide earlier that day left loose dirt behind, which turned into mud. Road workers failed to warn drivers of the dangerous conditions or to install protective barriers to replace the pine trees that, before Maria, had served as a buffer between the highway and the cliff. “Directly or indirectly, she died as a result of Maria and the inaction by the Puerto Rican government and FEMA,” said Díaz.

The official cause of his wife’s death, however, is simply  “serious corporal trauma.”

Noel Llabreras’s death was deemed “natural” in his certificate, the result of a stroke precipitated by brain injury and other ailments. There is no mention of the series of unfortunate events that led to them. It started with Maria blowing the zinc roof off the 69-year-old’s house; followed by a bout of Leptospirosis, a bacterial disease transmitted through water contaminated by animal urine; and ended with neglect during two different hospital stays to treat his worsening condition, according to his son, Javier Llabreras.

“People think a hurricane-related death is when a person who’s out taking pictures of the storm is hit by a flying piece of roof,” he said. “The real Maria deaths are those of people who lived in impassable highways, who had to cross rivers to look for food and died. All of those are direct causes of the hurricane.”

Some survey respondents were unable to obtain death certificates altogether. Some provided other kinds of documents, like death certificate abstracts (which don’t always include cause of death,) cremation certificates, and funerary cards.

A slew of independent counts

Omitting Maria from death certificates appears to have been a widespread practice. The data void it generated and the dozens of bodies overfilling morgues spurred the slew of independent death counts.

About a week after the hurricane, CPI questioned the government’s official count of 16 deaths, saying there could be dozens or hundreds based on reports from hospitals and morgues. Two weeks later, news outlet Vox found 81 cases tied to the storm through a basic news clip search—nearly twice what the government was reporting at that time.

CNN, meanwhile, surveyed half of the island’s funeral homes to inquire about Maria deaths: 499, it counted. Buzzfeed went to crematoriums, and discovered the number of bodies they were processing was much higher than the officially recorded hurricane deaths for those places.

CPI, the New York Times, and a researcher from Pennsylvania State University examined all death records for the first few weeks after Maria and compared them to the number of deaths during that period in previous years. They all estimated around 1,000 extra deaths. 

The government’s number abruptly stopped at 64 in December. By that month, CPI had already documented 66 additional deaths through on-the-ground reporting. In early 2018, Harvard researchers started surveying some 3,000 families across the island. Their estimate is based on 38 post-Maria deaths they discovered through their poll.

Quartz’s own list of verified deaths is just a sample. The Quartz/CPI online survey of more than 200 excluded thousands of people who didn’t have electricity or internet access for months.  The professions of the deceased suggest that our survey was answered by wealthier-than-average Puerto Ricans.

Help us count Maria’s true death toll

Here’s how the conflicting death toll evolved over time:

Fuzzy data

Inconsistency in disaster statistics is common. The death toll from Mexico City’s 1985 earthquake ranges from 5,000 to 45,000; for Haiti’s earthquake, in 2010, it varies from 46,000 to 316,000. The figure most often quoted for Hurricane Katrina, in 2005, is around 1,833, but it’s impossible to know for sure, says John Mutter, a Columbia University professor who carried out his own count.

There are no universal standards for classifying disaster-related death. It’s hard to weed out fraudulent claims and to prevent governments from manipulating data to their advantage. Rich countries tend to undercount deaths to save face, while poor ones lean towards over-inflating them to receive more aid, says Mutter.

“You can’t believe practically any numbers,” he adds. “Everybody counts differently; everybody has motivations.”

In the US, recording death is a state matter, and how it’s handled varies wildly. The Centers for Disease Control and Prevention has some guidelines of how to identify disaster-related deaths during an emergency; in fact, it issued a guide in October of 2017, as the controversy over Puerto Rico’s count was starting to boil over. A note should be made in the death certificate of anyone who died as a result of a catastrophe, whether directly or indirectly, it says.

That includes people who drowned from a flood, for example—that would be a direct death—and people who died because of disruptions caused by that flood, such as closed hospitals. Other situations that qualify as an indirect death, according to the CDC’s guide, read as a checklist of the conditions that led to deaths after Maria as described by the survey respondents: Power outages, blocked-off roads, displacement, property damage, personal loss, stress.

One way of  knowing is asking this question, per the CDC: “But for the disaster (hostile environment), would the person have died when he/she did?” If the answer is no, the death certificate should name the disaster and link it to the death.

“Inconsistencies in reporting a death as disaster-related on the death certificate make it difficult to generate reliable and accurate mortality statistics and to identify the most frequent causes of death associated with disaster events,” it wrote. “For public health and planning and preparedness purposes, recording sufficient information about the disaster-related circumstances on the death certificate is key, whether deaths are directly or indirectly related.”

Puerto Rico’s Public Safety secretary Héctor Pesquera said the cases in the official toll only include those backed by medial and forensic evidence that linked the deaths to the hurricane. To address the discrepancy between the official count and others, in February Puerto Rican governor Ricardo Roselló commissioned George Washington University to look into Maria’s casualties in “an independent and objective manner.” At that time, officials said an approximate death toll would be ready sometime in May; Pesquera said on Monday the government is still waiting for results from that study.

 A long-term emergency

Collecting raw numbers is simply not enough to grasp the scale of death and its causes. To truly understand a disaster like Maria, you have to look at the personal stories behind those numbers, says Rebekah Yore, project director at Rescue Global, a UK-based non-profit that helps governments prepare their disaster response.

Three years after Typhoon Haiyan ravaged the Philippines, Yore followed up with some hard-hit low-income communities. Talking to survivors, she learned that men, rather than women and children, were often in more danger because they chose to stay in place despite government warnings to protect the little they had from looters. “If we know that there is an issue there, we can try to mitigate these vulnerabilities by actually working with those people,” she says.

Our survey responses offer some clues of where to start. We know, from our sample, that the lack of electricity worsened the health of people who were already vulnerable, such as the elderly. We know, too, exactly in which ways the blackouts killed them: Some diabetics couldn’t refrigerate their insulin; others couldn’t get dialysis because treatment centers were closed. Without their adjustable beds and air conditioning, bed-ridden patients got ulcers and bacterial infections. The darkness also made them more likely to take a dangerous spill.

Overall, the cases reveal a fragmented and disorganized health system that was unable to provide medical attention throughout the island, says Luis A. Avilés, a University of Puerto Rico public health professor who collaborated with Quartz and CPI on the survey. “As long as our system doesn’t change, there will always be obstacles to providing services to the whole population that will be exacerbated during a crisis like this one,” he said.

In order to create policies that could save people in the same situation the next time around, Puerto Rican officials would have to carefully study the circumstances behind the death toll, and commit to keep collecting data. Tragically, some Maria-related deaths may not even have occurred yet.

Most communities don’t end up doing long-term studies of past disasters, says Robert Jensen, who’s been a regular at catastrophe sites since the Oklahoma City bombing in 1995. Usually, the instinct is to quickly move on from the tragedy. “We don’t like to think about mass fatality; it reminds us of our own mortality,” says the chief executive of Kenyon International Emergency Services, which provides disaster-related services around the world.

Global disaster intelligence

Facing up to Maria’s deaths in Puerto Rico would ultimately help other areas at risk. Disasters unfold differently from place to place, but their death toll is always related to how humans have chosen to live in their environment.

“The problem is that we continue to look at [disasters] as isolated, separate events, rather than as a continuing pattern,” says Ilan Kelman, a scientist at University College London who has looked at death tolls by disaster type and across geographies. “We lose the fact that it’s all about systemic vulnerabilities.”

Each time a community is wiped out by a flood, an earthquake, or a volcano, locals glean valuable knowledge about how to better protect the vulnerable during an emergency. After a 2003 heat wave killed thousands of elderly people in France, the government put together a national plan that included creating emergency procedures for retirement homes and adding air conditioning; a census of isolated and vulnerable people; and planned visits to them during especially hot days. Those measures reduced deaths during a later heat wave, research suggests.

Lives can be saved if the world’s disaster responders accurately record death tolls and share what they learn from them. Governments are starting to grasp the importance of this disaster intelligence sharing as the predicted effects of climate change become more real. There’s a United Nations-led push to collect data in a systematic way so that it can be compared across regions. “This allows us to know the history of the impact on a place and identify the trends,” says Nayibe Jiménez, a researcher at Corporación OSSO, a Colombian non-profit that helped develop the data-collecting tool and methodology called Desinventar that many of the participating countries are using.

Local lessons could also be spread more widely with technology, she says. Artificial intelligence and machine learning are being used to process large amounts of information in the midst of an emergency, for example, by quickly assessing damage from the analysis of satellite images to determine where help is needed. They could also be deployed to crunch data from around the world and distill it into best practices to prevent death.

Ultimately, though, those models would have to be fed with the contents of the low-tech death certificate. It already has a lot riding on it, from the closure it offers a distraught relative to the historical record it leaves behind for researchers hundreds of years from now. Maria is yet another reminder of the importance of accurately filling it.

“Death certificates aren’t just a piece of paper,” says Jensen, the CEO. “They’re hugely important documents.”