Senior officials at the World Health Organization warned on Monday that drug-resistant infections could cause 39 million deaths over the next 25 years. The number did not come from the agency’s own counting. It came from a model published in The Lancet in September 2024.
The model belongs to the Global Research on Antimicrobial Resistance project, known as GRAM, a partnership between the Institute for Health Metrics and Evaluation at the University of Washington and the University of Oxford. Its central estimate is 39.1 million deaths caused directly by resistant bacterial infections from 2025 through 2050, with an uncertainty range of 33.0 million to 46.0 million, according to [the paper](https://www.thelancet.com/journals/lancet/article/PIIS0140-6736(24)01867-1/fulltext). Dr. Jean-Pierre Nyemazi, the acting director of the W.H.O.’s antimicrobial resistance department, told a virtual media forum that resistance “is widespread and threatening the future of modern medicine,” Al Jazeera reported.
Checked against the record, the figure is sturdier than the claim it replaced and softer than the headlines make it sound. Its annual toll for 2050 is about a fifth of the 10-million-deaths-a-year forecast that dominated the subject for a decade, and in the one region where a like-for-like check has been published, the GRAM method produced a lower count than the official European estimate. But much of the projected rise follows the aging of the world’s population, the model’s inputs are thinnest in the countries where it places the most deaths, and epidemiologists who study the problem say resistance cannot be forecast reliably a quarter-century out. The order of magnitude holds up. The precision does not.

What the number counts
GRAM reports two kinds of death, and they answer different questions. Deaths “attributable” to resistance are those that would not have happened had the infection responded to standard drugs. Deaths “associated” with resistance are all deaths in which a resistant infection played a part, measured against a world in which the infection never occurred. The second figure is always several times the first.
For 2021, the last year the study estimated rather than forecast, GRAM put attributable deaths at 1.14 million and associated deaths at 4.71 million. By 2050 it projects 1.91 million attributable deaths a year, a 67.5 percent increase, and 8.22 million associated deaths, according to the institute’s announcement. Added up from 2025 through 2050, associated deaths come to 169 million, within a range of 145 million to 196 million. The 39 million is the smaller, stricter count.
How the number is phrased depends on who repeats it. Reuters wrote that drug-resistant infections “could contribute to more than 39 million deaths worldwide over the next 25 years, if current trends persist.” Al Jazeera’s headline used “could kill.” The W.H.O.’s written release on Monday did not mention the forecast, and the agency’s fact sheet, updated July 16, cites only the 4.7 million associated deaths for 2021. When the paper appeared, The Lancet announced it with the conditional intact.
🆕 Antimicrobial resistance poses an important global health challenge in the 21st century.
A new in-depth global analysis suggests more than 39 million deaths from antibiotic-resistant infections could occur between now & 2050.
Explore the data: #AMR
The number it replaced
In December 2014, a review commissioned by Prime Minister David Cameron of Britain and chaired by the economist Jim O’Neill estimated that if resistance kept rising, 10 million people a year would die of it by 2050 and global economic output would be 2 to 3.5 percent lower. By its own description the figure was rough. It set that against a low estimate of 700,000 deaths a year at the time, and it drew on two modeling exercises, one by RAND Europe and one by the consulting firm KPMG, according to the review’s first paper.
Two years later, Marlieke de Kraker and Stephan Harbarth of Geneva University Hospitals, with Andrew Stewardson, took the figure apart in PLOS Medicine. The headline number, they wrote, appeared to rest on a scenario in which resistance rates jumped 40 percentage points and then held steady until 2050 while infection rates doubled. “To date, there is no empirical data supporting any of these scenarios,” they wrote, and the estimate had been repeated without caveats or confidence intervals.
Predictions always require assumptions, but modeling future scenarios using unreliable contemporary estimates is of questionable utility.
Eric A. Friedman, a global health justice scholar at Georgetown University’s O’Neill Institute, went looking for the source in 2020. The two underlying studies modeled only six diseases and bacteria, and one assumed that within 15 years there would be no drugs left to treat them. “Review the two studies themselves, and you will not find that number,” he wrote.
The 700,000 baseline fared no better with the people who produced it. In a 2022 post for the Center for Global Development, Anthony McDonnell and Katherine Klemperer called it “flawed” and said one of them had done the calculations behind it. The United States, they wrote, was a poor model for the world, with lower resistance than average and generally good health care.
This low estimate was primarily based on extrapolating worldwide deaths from US deaths.
The 10 million figure outlived its critics. In June 2025, nine months after the GRAM forecast appeared, the drugmaker Roche attributed it to the W.H.O. in a post on X, still framed as an annual toll.
Antimicrobial resistance (AMR) is a pressing reality. The WHO projects that AMR could cause up to 10 million deaths annually by 2050 if current trends continue. Let’s take action now to ensure life-saving medicines remain effective for future generations.…
Set beside it, GRAM’s numbers are modest. Even on the broader associated definition, the forecast for 2050 is 8.22 million deaths, below 10 million, and the 1.91 million attributable deaths is roughly a fifth of it. The GRAM paper itself calls the 2014 projection “the subject of scientific criticism,” in its open-access version.
How the model works
GRAM built its historical estimates from 520 million individual records covering 22 pathogens, 84 pairings of bacteria and drugs and 11 types of infection in 204 countries and territories, CIDRAP reported. Seven of those 204 places had no data for any part of the model. The method produces estimates for every location anyway, filling the gaps with statistical prediction.
The forecast does not model resistance from its causes, such as antibiotic use. According to the paper’s methods, the team took the historical share of deaths from each broad cause that was attributable to resistance and projected that share forward with an ensemble of 12 statistical sub-models, weighted by how well each predicted data held out of the fitting. It then applied those shares to the institute’s separate forecasts of deaths from all causes, which rest on projections of income, education and fertility.
The associated count is not forecast on its own. The team took the 2021 ratio of associated to attributable deaths for each cause, age, sex and location, and multiplied. So the 169 million inherits every assumption in the 39 million and adds one more: that the ratio between the two holds for three decades.
The paper also ran two alternative futures. Better care for severe infections and wider access to antibiotics could avert 92.0 million deaths from 2025 to 2050, and a steady supply of new drugs against Gram-negative bacteria could avert 11.1 million, the authors estimated, noting that both results depend on the assumptions behind them. The drug scenario assumed a 15-year scale-up of the pipeline that cut Gram-negative deaths in half. Neither is funded.

Where the rise comes from
The historical line is nearly flat. The study estimated 1.06 million attributable deaths in 1990 and 1.14 million in 2021, according to Smithsonian magazine’s account of the paper, an increase of about 8 percent over 31 years. The forecast has the annual toll rising 67.5 percent over the next 29.
Much of the difference lies in who gets sick. Attributable deaths among young children fell from 488,000 in 1990 to 193,000 in 2021, while those among adults over 70 rose 89 percent, according to CIDRAP. Authia Gray, a researcher at the institute, told CIDRAP that the drop in child deaths came from fewer infections, not from less resistance.
When children get infections today, they’re actually more likely to have a resistant infection compared to the past.
The forecast carries that pattern forward. People 70 and older account for 65.9 percent of the attributable deaths projected for 2050, according to the paper. Deaths rise 69.6 percent from 2022 to 2050. Measured in years of healthy life lost, which counts a child’s death far more heavily than an 80-year-old’s, the burden rises only 9.4 percent, within a range that runs from a 6.9 percent fall to a 29 percent rise.
Resistance itself is climbing in the data the W.H.O. can see. The agency’s surveillance report of Oct. 13, 2025, built on more than 23 million laboratory-confirmed infections reported by 104 countries, found that one in six was resistant in 2023, and that resistance rose in more than 40 percent of the bacteria-drug pairings monitored from 2018 to 2023, by an average of 5 to 15 percent a year in relative terms, according to the agency’s announcement. Dr. Mohsen Naghavi, who led the GRAM analysis at the institute, told CIDRAP in 2024, “This trend is consistent with everything we calculated in 2019.”
The case against precision
The sharpest caveats arrived with the paper. In a comment published alongside it in The Lancet, Dr. Samuel Kariuki of the Kenya Medical Research Institute wrote that scarce data in many low- and middle-income countries was a major drawback, and that where data existed it was often too fragmented for trend analysis. He also noted, CIDRAP reported, that the model could not account for new resistant strains that might emerge.

Dr. de Kraker, the Geneva epidemiologist who dissected the 2014 figure, was asked about the new one. “Predicting antimicrobial resistance trends is very unreliable,” she told New Scientist, as Smithsonian reported, because resistant strains can appear or fade quickly for reasons scientists do not well understand.
Researchers who build on GRAM’s numbers have pointed to the same soft spot. A team from the Global Antibiotic Research and Development Partnership in Geneva, writing in The Lancet Infectious Diseases in April 2025, said that in poorer countries GRAM “relied more heavily on regional trends and methodological assumptions” than in rich ones, which “might have contributed to bias,” even as they called it the most comprehensive analysis available. A pooled analysis of six East African countries, published in December 2025 in Tropical Medicine and Health, found that GRAM’s uncertainty ranges “may not fully reflect the true variability in data-scarce settings,” its authors wrote, “potentially leading to over-confidence in some estimates.”
The bias need not run one way. In a Perspective published in July in Nature Communications, Cherry Lim, Koen Pouwels, Ben Cooper and colleagues at Oxford and the Mahidol Oxford Tropical Medicine Research Unit argued that uncorrected biases in the hospital studies feeding burden estimates can produce underestimates or overestimates, depending on the setting. Letters in The Lancet’s May 31, 2025, issue pressed other points: one asked how Covid-19’s disruption of infection patterns could shift the long-range forecast, and Dr. Manica Balasegaram of the Geneva drug-development partnership argued that most of the world’s population, particularly in sub-Saharan Africa, has no access to effective antibiotics at all.

The associated figure is the softer of the two. Its comparison case, a world where resistant infections vanish rather than being replaced by treatable ones, is one the GRAM team’s [2022 paper](https://www.thelancet.com/journals/lancet/article/PIIS0140-6736(21)02724-0/fulltext) said the evidence could not yet settle.
Weighing the evidence
If the model inflated deaths, it should show where official counts exist. Europe offers the cleanest test. For the 11 bacteria-drug pairings both studies covered, the European Centre for Disease Prevention and Control’s estimate, led by Alessandro Cassini, put deaths in the European Union and European Economic Area at about 30,000 in 2015. GRAM’s 2019 global analysis put deaths from the same pairings at 23,100 for western and central Europe, according to a 2022 GRAM paper on the European region. The years and borders differ, so the comparison is rough. It still points toward the model running low.
The newest American count went up. On Oct. 27, the Centers for Disease Control and Prevention estimated 9.7 million drug-resistant infections and 50,800 deaths in the United States in 2023, against 2.8 million infections and 35,000 deaths in its 2019 report, Healio reported. The agency warned against reading that as a trend because its methods and data sources changed, though Dr. Dawn Sievert of the C.D.C. said real increases were also visible.
We are saying to not compare the data from this ... directly back with numbers from the 2019 threats report.
Taken together, the critics are strongest on precision. The 33.0 million to 46.0 million range describes uncertainty inside the model. It does not capture the question Dr. de Kraker and Dr. Kariuki raise, which is whether the next 25 years of resistance will look like the last 30. A new resistant strain, or a new class of antibiotics, could push the real toll outside that band, and a model that extends past trends has no way to see either coming.
They are weaker on direction. The published critiques of the 2024 forecast argue that it is uncertain, not that it is too high, and the European check points the other way. The aging effect also works in the model’s favor: the size of the over-70 population in 2050 is among the most predictable inputs a forecaster has, since nearly everyone in it has already been born.
What the evidence supports is a scale: somewhere between one million and two million direct deaths a year, rising as populations age, with resistance climbing in the bacteria the W.H.O. tracks. What it does not support is treating 39 million as a count. The main alternative still in circulation, 10 million a year, comes from a 2014 review that described its own number as rough.

What W.H.O. measured this week
The agency’s new material on Monday was a governance survey, not a death count. Of 183 countries that answered its annual self-assessment, 173 reported a national action plan against resistance, up from 79 in 2017, but only 42 said their plans were financed and implemented, according to Reuters. The W.H.O. put that share at 22 percent. “An unfulfilled plan helps no one,” Dr. Tedros Adhanom Ghebreyesus, the agency’s director general, said in the release.
Of 128 countries with national infection-prevention programs, up from 90 in 2018, only 69, or 38 percent, had put them in place nationwide, against a 2030 target of 90 percent. The number monitoring antibiotic use rose to 113, from 65 in 2017, and the number tracking resistance to 109, from 39, but only 21 link the two. Just 66 countries, 36 percent, have a national reference laboratory that can test for resistance in both the bacteria and the fungi the W.H.O. monitors, against a target of 80 percent. Dr. Silvia Bertagnolio, who heads the agency’s surveillance and laboratory unit, said at the briefing what that means for doctors and for anyone trying to measure the problem, CIDRAP reported.
Without testing, we cannot see where resistance is emerging.
Those blind spots are the same ones the GRAM model fills with assumptions. The W.H.O.’s global surveillance system drew data from 104 countries in 2023. In Asia and the Pacific, a regional analysis released Oct. 5 found that 49 percent of 27 national plans carried no costing at all and only 44 percent were fully costed, according to the W.H.O.’s Western Pacific office.
The forecast and the policy are tied together. The 2024 U.N. political declaration committed countries to cut deaths associated with bacterial resistance 10 percent by 2030 from a 2019 baseline of 4.95 million, according to the U.N.’s summary. That baseline is GRAM’s own 2019 estimate. The 2024 paper’s authors concluded that “without additional measures, we will not reach the 10% reduction.”
What comes next
A multistakeholder meeting runs in Rome from Nov. 16 to 18, during World AMR Awareness Week. The Fifth Global High-Level Ministerial Conference on Antimicrobial Resistance then convenes in Abuja, Nigeria, from Dec. 6 to 8, the first time an African country has hosted it, according to its organizers. The meeting was first planned for June, Health Policy Watch reported.
Member states [adopted a Global Action Plan on Antimicrobial Resistance](https://www.who.int/news/item/25-05-2026-the-world-health-assembly-adopts-updated-global-action-plan-on-antimicrobial-resistance-(2026-2036)) for 2026 to 2036 on May 23, under a decision that asks the director general to report progress in 2027, 2029 and 2031, and the U.N. declaration calls for another high-level meeting in 2029. On the mortality side, the Global Burden of Disease project has begun publishing 2023 results one infection type at a time, including a paper on lower respiratory infections in April. The 2024 GRAM paper remains the latest global forecast of resistance deaths.
The first progress report under the new plan goes to the World Health Assembly in 2027, three years before the deadline for the 10 percent cut.

