.png)
Congo War Death Toll: The 5.4 Million That Was Never Counted (Cold Files 021)
In January 2008 the International Rescue Committee published a number that has outlived every argument made against it. Five point four million people dead in the Democratic Republic of the Congo between August 1998 and April 2007. The organisation's president, George Rupp, said the toll surpassed any conflict since the Second World War. The Washington Post and NBC carried it within hours. Eighteen years later it is still in the lead of Wikipedia's article on the war, still on the IRC's own website, and still cited in a 2026 medical preprint written by the epidemiologist who ran the first survey.
It is not a count of the dead. It is a subtraction. Almost none of the deaths inside it were measured as violent, and the quantity subtracted from it was not a Congolese measurement.
Part of the Cold Files series.
The Number Is a Subtraction
The five surveys behind the figure did not look for bodies. They measured a crude mortality rate: deaths per 1,000 people per month, established by asking households who had died since a given date. Excess deaths are then what is left when you subtract an assumed peacetime rate from the measured one, and multiply the difference by the population and the number of months.
The last and best of the surveys, fielded between May and July 2007, is the clearest example. It visited 14,000 households across 700 clusters in all eleven provinces, and measured a national rate of 2.2 deaths per 1,000 per month, with a 95 percent confidence interval of 2.1 to 2.3. The east came in at 2.4, the west at 2.0. Subtract the assumed baseline of 1.4, apply the remainder to the population over sixteen months, and you get roughly 727,000 excess deaths.
What the surveys recorded as violence was almost nothing. In that final survey, deaths from violence were 0.4 percent of all deaths nationally, weighted: eleven of 1,915 recorded deaths in the east, three of 984 in the west. In the 2004 survey it was under 2 percent, with a single violent death recorded in the entire western half of the country. Across the earlier studies IRC put it at less than 10 percent. The IRC said so itself in the press release: the vast majority "died from non-violent causes such as malaria, diarrhea, pneumonia and malnutrition."
This is not a trick, and it is worth being clear about that before going further. It is standard humanitarian epidemiology, and the causal claim behind it is coherent. Richard Brennan, who co-authored the last two surveys, put the mechanism plainly on Voice of America: "Clinics don't work; vaccination programs don't operate; farmers can't work their land." A war that displaces millions and collapses a health system kills people who never see a soldier. The 2006 Lancet paper found that if the effects of violence were removed, all-cause mortality could fall to almost normal rates.
But a measured rate is not an excess. The excess is the gap between a number that was observed and a number that was chosen.
The Rate Nobody Measured in the Congo
The chosen number was 1.4 deaths per 1,000 per month. It is the average crude mortality rate for sub-Saharan Africa, taken from UNICEF's State of the World's Children report for 2007 and cited at the seventh endnote of the IRC study. For the stretch between May 2004 and December 2005 the report used 1.5 instead. The 2004 survey and the 2003 paper in the CDC's Morbidity and Mortality Weekly Report had both used 1.5, in the latter case citing a 1992 reference on famine-affected, refugee and displaced populations.
So four numbers drift through the series: 1.2, 1.3, 1.4 and 1.5. None of them is a Congolese measurement.
The IRC's defence of this was real and should be printed. It knew about the lower figures and rejected them deliberately. Its own reports name a pre-war Congolese rate of 1.2 from UNICEF and a 1997 pre-war rate of 1.3, and use the higher continental average anyway, on the ground that doing so produced the more conservative estimate. The 2007 report says the same thing in its discussion: "we have erred on the conservative side in our use of external sources."
The reason no Congolese figure was available is stated in the report itself. There had been no national census since 1984. The IRC wrote that "no official census has been conducted for over 20 years." A first one since 1984 began only in 2026. No defensible pre-war Congolese mortality rate exists. Everybody in this argument is choosing one.
The choice does almost all of the work. At 1.4, that sixteen-month window yields about 727,000 excess deaths. At 2.0 it yields close to nothing. The denominators were modelled too: the report applied its rates to the median of ten separate population estimates, 59.9 million, from a range running from 56.8 million to 69.9 million, with earlier years grown backwards at 2.5 percent a year.
Forty-One Months Out of a Hundred and Five
Five point four million covers 105 months. Three of the five surveys were nationwide, covering forty-one of those months; the two strongest covered thirty-two.
The first ran from April to July 2004: 19,500 households, covering more than 90 percent of the population, with a recall period running from January 2003 to April 2004. The second is the 2007 survey already described, recalling January 2006 to April 2007. Both are serious pieces of fieldwork, and the second answered most of the design criticisms made of the first.
The other sixty-four months are a different matter.
May 2004 to December 2005 was not surveyed at all. The IRC says so without hedging: it is "a period not covered by any IRC survey." The roughly 735,000 excess deaths assigned to it were computed by averaging the crude mortality rates from the 2004 and 2007 surveys and subtracting 1.5.
The largest single block is older and weaker. IRC's own 2001 report put August 1998 to April 2001 at 2.5 million, and it rests on two early studies. The first, released in June 2000, was five surveys in parts of the eastern provinces. The second, in 2001, was eleven surveys across seven eastern areas. Neither used random site selection. Michael Spagat, who later did the technical work on the critique, wrote that because of that, these surveys "are not appropriate for extrapolation" to region-wide estimates. The IRC extrapolated from them anyway.
The 2002 survey, which produced the 3.3 million then in circulation, excluded about five million people for insecurity: roughly 35 percent of the eastern population, in the part of the country where the fighting was.
And no confidence interval was ever published on 5.4 million. The report gives intervals on rates, and offers a floor and ceiling for the final window alone, 522,000 and 1.05 million. The headline number was released bare, and it was quoted bare.
The Correction That Corrected Nothing
On 20 January 2010 the Human Security Report Project, based at Simon Fraser University and directed by Andrew Mack, pre-released a chapter on the Congo at United Nations headquarters in New York. The full volume followed from Oxford University Press in 2011.
Its case had four parts. The baseline was too low. The 1998 to 2001 block could not bear extrapolation at all. No real confidence intervals had ever been published. And the IRC's child mortality figures ran roughly double those of the 2007 Demographic and Health Survey, which showed no comparable wartime spike.
For a substitute baseline the project did not reach for another continental average. It used 2.0, the rate the IRC had itself measured in western Congo, the region the war had largely bypassed. On that assumption the IRC data supported about 860,000 excess deaths for May 2001 to April 2007, with a 95 percent confidence interval running from minus 550,000 to 2.4 million. The substitution has its own circularity. Western Congo in wartime was not a country at peace, so 2.0 assumes the war killed nobody there.
An interval that spans zero is not a rival estimate. It is a statement that the instrument cannot answer the question, and Mack said as much. "We do not deny the veracity of the IRC report," he told the Huffington Post, adding that no one knows the correct figure. To the Christian Science Monitor he said the exercise was "a way of trying to demonstrate that relatively small change in assumptions can produce a huge change in outcomes," and that the point was "not to say we're right, you're wrong." He disowned the halved-death-toll headline that the press had already run.
The IRC did not move. Its public position held that 5.4 million remained its best estimate under conservative assumptions. Les Roberts, who led the first three surveys, called it "unbecoming to grab a headline a decade after by tearing down a study with erroneous speculation." Brennan accused the project of cherry-picking data.
Then nothing happened. No institution adopted 860,000. The Human Security Report Project wound down a few years later and its website went offline, taking its published replies to the IRC with it. The IRC's report page still carries 5.4 million in 2026, with no revision note. And popular usage moved the other way entirely: the Council on Foreign Relations conflict tracker and the United Nations regional information centre for Western Europe now print the same sentence, word for word: "Since 1996, conflict in eastern DRC has led to approximately six million deaths." That figure is 600,000 larger, starts two years earlier, and has no survey behind it at all.
What Both Sides Agree
Even the lowest estimate runs to six figures.
The lowest published estimate is a demographic reconstruction by Andre Lambert and Louis Lohle-Tart for ADRASS, which put conflict-attributable deaths from 1998 to 2004 at about 183,000, and which its authors present as a high-end figure built on deliberately generous assumptions about which provinces were at war. That lowest number is still many times what conflict-event datasets record, because those count deaths from fighting and exclude deaths from disease and starvation by definition. An independent reconstruction by Richard Kapend, Jakub Bijak and Andrew Hinde in the Quetelet Journal in 2020 declines to give a point estimate and offers a range of one to 1.9 million. Their stated conclusion is the thing both camps arrived at from opposite directions: when data are scarce, the choice of baseline rate strongly affects the result.
Nor is the error plainly in one direction. The IRC could not reach the districts where the killing was worst, and it said so. Roughly five million people fell outside the 2002 sample and about five million outside the 2004 one, almost all of them in the east. The 2007 report concedes that its results "could have underestimated mortality because of survival bias." The excluded zones were the violent ones.
The figure also has a constituency now. Since 2023 the Congo has marked 2 August, the day the war began, as a national day of commemoration under the name Genocost; campaigners use "more than 6 million lives"; and in 2025 President Felix Tshisekedi asked parliament to adopt a resolution recognising genocide committed on Congolese soil. Auditing the arithmetic of a death toll is not the same act as denying the dead, and this file is not doing the second thing.
In 2025 Les Roberts went back. A survey he co-authored with Jennifer OKeeffe and Augustin Gang Karume covered 899 households in North and South Kivu after the M23 offensive and the collapse of humanitarian funding, and found the crude mortality rate had roughly doubled. It reports about 165,391 excess deaths, with an interval of 63,557 to 267,225. It is a preprint and has not been peer reviewed.
What it did differently matters more than what it found. The baseline it subtracts was measured in the same households, over the six months before the crisis, instead of being lifted from a continental indicator table. Twenty-five years after the first survey, somebody finally measured the thing being taken away. For the war years it can only be reconstructed.
Sources consulted
IRC, Mortality in DR Congo: An Ongoing Crisis, 2008
Coghlan et al., Mortality in DR Congo, Lancet, 2006
Coghlan et al., Update on Mortality, Disaster Medicine, 2009
IRC, Nationwide Survey, April to July 2004
Roberts et al., Elevated Mortality, MMWR, 2003
IRC, Mortality in Eastern DR Congo, 2000
IRC press release on 5.4 million, 2008
Spagat, Estimating the Human Costs of War
Human Security Report Project, Shrinking Costs, 2011
UN DESA, Meeting on Mortality Crises, 2011
Lambert and Lohle-Tart, La surmortalite au Congo, 2009
Kapend, Bijak and Hinde, excess mortality, Quetelet, 2020
Brennan et al., Mortality surveys, Humanitarian Exchange, 2006
Stearns, How many really died, Congo Siasa, 2010
Nienaber, Data Bombshell, Huffington Post, 2010
Christian Science Monitor, on overestimated war deaths, 2010
De Capua, Brennan interview, Voice of America, 2008
Associated Press, Review halves Congo war toll, 2010
OKeeffe, Karume and Roberts, Kivu mortality, medRxiv, 2026
Council on Foreign Relations, Global Conflict Tracker, 2026
Uppsala Conflict Data Program, Our World in Data
More Cold Files: Tigray and the Zaire camps.
If this is the kind of work you want more of, the daily brief and every Cold File go out on Telegram.
Next in Cold Files: Operation Menu, 1969 to 1970. The bombing of Cambodia was logged under the wrong country, and the records that proved it were the ones the United States kept itself.
Comments