How many people in the United States die or suffer serious harm from acetaminophen overdose?

To try to answer this question, ProPublica examined three sets of primary data, as well as dozens of scientific studies. While the data sets varied on the total number of deaths, all three showed an increase in acetaminophen-linked fatalities over much of the past decade, with only occasional year-to-year declines.

Here is an in-depth look at the numbers.

Data compiled by the U.S. Food and Drug Administration has linked as many as 980 deaths in a year to drugs containing acetaminophen. In addition, FDA reports of death associated with acetaminophen have been increasing faster than those for aspirin, ibuprofen and many other common over-the-counter pain medicines.

Data obtained from the U.S. Centers for Disease Control and Prevention show that more than 300 people die annually as a result of acetaminophen poisoning.

Beginning in 2006, according to the CDC, the number of people who died after accidentally taking too much acetaminophen surpassed the number who died from intentionally overdosing to commit suicide.

The American Association of Poison Control Centers (AAPCC), a nonprofit that receives federal funds, shows about 113 people dying each year as a result of overdosing on medicines with acetaminophen. Since 2006, acetaminophen has accounted for more fatalities than all other over-the-counter pain relievers combined, according to AAPCC data.

Why the large differences? Because each organization’s data has strengths and weaknesses.

The FDA relies primarily on individual case reports, called FDA Adverse Event Reports, submitted by drug makers, consumers and doctors. Drug makers must submit information about any cases they learn about involving side effects linked to their drugs, while reports by consumers and healthcare workers are voluntary. The reports often lack key information. By some estimates, the FDA system captures from 1 percent to 10 percent of adverse events involving drugs.

Most important, the reports do not demonstrate a causal connection between a person’s death and a particular drug, just an association. This is a crucial issue for acetaminophen, because that drug is often combined with other drugs into one medication. For example, the prescription painkiller Vicodin combines acetaminophen with the powerful opioid painkiller hydrocodone. If a person dies after taking Vicodin, the FDA report does not say whether it was caused by hydrocodone or acetaminophen, or even some other drug. So the FDA data only allows one to say a drug is linked to or associated with the death.

The CDC’s primary source of information for acetaminophen poisoning is death certificates submitted by health care professionals, coroners and medical examiners. These go into the agency’s Multiple Cause of Death database. The CDC estimates that it captures the vast majority of death certificates filed in the United States, and its format attempts to determine the primary cause of death for each case. But the quality of the reports depends upon the judgments of the local officials who review medical files, conduct autopsies and ultimately fill out death certificates.

The AAPCC draws its data from the dozens of poison control centers throughout the nation that advise people who fear they have ingested a poisonous substance, as well as healthcare providers who treat such patients. Those calls form the basis of the AAPCC’s National Poison Data System. Its information is quite detailed – more so than the other two data sets.

But it, too, has a limitation: Not everybody who overdoses on a drug calls a poison control center. In particular, the AAPCC data may underestimate fatalities for the simple reason that there is little need to call a poison control center if a person has already died. Poison control center officials have acknowledged that the AAPCC’s overall fatality numbers capture only a fraction of poisoning deaths.

While the AAPCC publishes annual reports with detailed tables, its full data set is expensive. The organization receives substantial taxpayer money from the CDC, yet it charges citizens and even other government agencies for more detailed access to its raw data.

After examining the data sets and checking the numbers with the CDC and AAPCC, ProPublica has generally relied upon the CDC figures as a primary source of information. We judged that the death certificate information, while imperfect, represented the most rigorous collection of data. (And yes, we are aware of the series we did pointing out the flaws of death investigations in America.) However, we have also used information from the FDA and the AAPCC, such as when the organization’s data was used in a regulatory setting or when it was the sole source of information on a particular issue.

Both the CDC and the AAPCC data show that acetaminophen toxicity results in more deaths per year than do drugs such as ibuprofen and naproxen, the so-called NSAIDs (non-steroidal anti-inflammatory drugs). McNeil Consumer Healthcare, the Johnson & Johnson unit that makes Tylenol and ibuprofen under the Motrin brand, argues that the CDC and AAPCC death counts do not contain enough information to compare the relative safety of different medications. The company contends that those databases “cannot be used in isolation” but must be taken in context with the totality of information available on the risks of the drugs.

In addition, McNeil and some epidemiologists argue that these data sets undercount deaths resulting from chronic use of NSAIDs. They point to studies showing the well-established medical fact that even at recommended doses, NSAIDs can cause stomach bleeding, sometimes serious or even fatal.

If all deaths caused by NSAIDs were counted, McNeil has argued, then those drugs could cause as many or more deaths per year than acetaminophen.

The CDC and AAPCC data sets count actual deaths attributed to medications such as acetaminophen and NSAIDs. They are not extrapolations.

To support its argument that NSAIDs may cause the greater number of deaths, McNeil has pointed to various studies – but they don’t count actual deaths attributed to NSAIDs. Instead, they use assumptions and mathematical models to estimate NSAID-related deaths.

Those studies have other limitations. McNeil and other researchers have cited a 1999 study that “conservatively estimated” as many as 16,500 U.S. deaths a year were related to NSAIDS. However, the lead author of that study wrote in an email that the estimate is outdated and “today is probably one-fourth of that.” Also, that estimate specifically excludes over-the-counter NSAIDs, meaning that it applies to prescription NSAIDs only.

That’s an important limitation. Some NSAIDs are only sold by prescription because they have not been deemed safe enough to be sold over-the-counter. Also, NSAIDs such as ibuprofen can have higher recommended daily doses when sold by prescription than when sold over-the-counter. Acetaminophen, however, has the same maximum recommended daily dose, whether obtained by prescription or not. So comparing prescription NSAIDs to acetaminophen, some researchers have argued, is an apples-to-oranges comparison.

An apples-to-apples comparison, these experts say, would look at deaths caused by over-the-counter NSAIDs to those caused by acetaminophen. No study McNeil cited does that.

One study, partly funded by McNeil, does something similar. It makes an estimate, based on mathematical modeling, about the relative risk of death for acetaminophen and over-the-counter NSAIDs. It found that acetaminophen has a lower risk of causing death than does over-the-counter ibuprofen or naproxen. However, the study states that its analysis is based on some findings “which are subject to considerable error,” the authors write. “We also made a number of assumptions for ease of comparison.” One of those assumptions: Everyone takes the medicines as directed and no one overdoses. But overdosing is the main cause of acetaminophen-related deaths.

For data on nonfatal injuries, ProPublica looked to a study that analyzed several databases. The National Hospital Ambulatory Medical Care Survey, which is conducted by the CDC, estimates that an average of 44,000 people per year visited emergency rooms for acetaminophen overdoses between 2000 and 2007. About half of them overdosed by accident; the rest attempted suicide or their overdoses were of undetermined cause.

A Consumer Products Safety Commission survey estimates more than 78,000 patients a year go to emergency rooms for acetaminophen overdoses, with about 30 percent accidentally taking too much, the study found. The study authors speculated that the difference between the number of emergency-room visits arises from how the two research teams identified acetaminophen overdose cases.

According to yet another database, the CDC’s National Hospital Discharge Survey, about 33,000 people are hospitalized annually due to acetaminophen poisoning, with about 75 percent of them having intentionally overdosed, the study found.

A note for data wonks. One fundamental question can be only roughly approximated: What is the fatality and injury ratio for acetaminophen? That is, how many deaths and serious injuries happen for every 100,000 people who take the drug? Various attempts have been made to answer that question. But nobody knows the numerator, the exact number of deaths and injuries. And no one knows the denominator, how many doses of acetaminophen are sold each year or, of those sold, how many are actually ingested by people. Over-the-counter acetaminophen is sold in bottles containing as many as 1,000 pills; how many people take all of those pills?

All we can say is that only a tiny fraction of people who take acetaminophen appear to suffer injuries or fatalities as a result. Adding up the highest estimates of injuries and deaths linked to acetaminophen would result in a total of a little over 110,000 incidents annually. About 27 billion doses were sold in 2009, the most recent year for which figures are publicly available. If all the pills were consumed, that would mean about one injury for every quarter million doses. In actuality, that almost certainly lowballs the rate of injuries, but by how much, nobody knows.