64,260 deaths due to traumatic injury in the Gaza Strip between 7 October 2023 and 30 June 2024, with a 95% confidence interval of 55,298 to 78,525, is the only peer-reviewed estimate ever produced against the Gaza Ministry of Health’s registry, published by Jamaluddine and colleagues in The Lancet in 2025. The ministry reported 37,877 decedents for the same window, of whom 28,185 were identified. That study was built out of the ministry’s own hospital records, the ministry’s own online mortality survey and manually scraped social media obituaries, so it measures the completeness of a record rather than counting the dead independently, and it covers violent injury alone. Pushing the other way, the Jerusalem Center for Security and Foreign Affairs published a ceiling of at most 41,000 civilian war-related deaths in 2025, reached by subtracting estimated natural deaths and Israel’s count of 21,000 fighters killed from the ministry total, and stating on the same page that neither input has been independently confirmed.
Key Findings
- The registry runs on two channels, both belonging to the same body: hospital morgue records carrying 22,368 identified decedents, and a rolling online mortality survey the Palestinian Ministry of Health launched on 1 January 2024 carrying 5,817 individual decedents (Jamaluddine and colleagues, The Lancet, 2025).
- The ministry publishes the size of its own unresolved fraction. Its update of 30 June 2024 carried 9,692 unidentified decedents, 26% of its cumulative total at that date, as reported in The Lancet by Jamaluddine and colleagues (2025).
- The gap between the estimate and the reported figure is 26,383 deaths, which is 41% of the 64,260 estimate but 70% of the count it was measured against, so the study’s under-reporting share is not a multiplier on the ministry’s published total.
- 60.8% (n = 59) of the Gaza Strip’s 97 health facilities sustained damage during the first phase of the Israeli military campaign and 35.1% (n = 34) were functionally destroyed on a satellite damage-area threshold, on the analysis Asi and colleagues published in Conflict and Health in 2024.
- The Jerusalem Center for Security and Foreign Affairs published a ceiling of at most 41,000 civilian war-related deaths in 2025, its own arithmetic on the ministry total and Israel’s count of 21,000 fighters killed, and states that neither input has been independently confirmed.
How does a death get onto the ministry’s list?
22,368 identified decedents reached the Palestinian Ministry of Health’s cumulative list through hospital morgue records, and a second channel, a rolling online mortality survey the ministry launched on 1 January 2024, carried 5,817 individual decedents. Both are the records of a health ministry that is a party to the war, and both counts were published in The Lancet (Jamaluddine and colleagues, 2025).
A death enters through one of two doors. The first is the morgue ledger: a body reaching a health facility, identified there, and entered on the cumulative list the ministry publishes. That is the channel behind the 22,368 identified decedents used as one of three input lists in the capture-recapture analysis The Lancet published in 2025, whose authors worked under the London School of Hygiene and Tropical Medicine ethics reference 31101 and published their code and anonymised data on GitHub.
The second door opened on 1 January 2024, almost three months into the war, when the ministry began a rolling online survey through which members of the public could report a death directly. By 30 June 2024 that list held 5,817 individual decedents (Jamaluddine and colleagues, 2025). A survey that begins in January cannot register the first twelve weeks of deaths on the same terms as the weeks that follow, and a channel that depends on a relative with a working phone and a connection is a different instrument from a hospital record.
Both lists belong to the same body, and that body is the health ministry of a party to the war. Neither channel audits the other, and it is the overlap between them that makes them usable as inputs to a statistical test: the analysis The Lancet published averaged alternative models over the three lists’ overlap structure, so the size of the overlap is the arithmetic it runs on. What neither channel reaches is a death that never arrived at a health facility and was never reported on a form.
How much of the ministry’s own record carries no identification?
9,692 unidentified decedents sat in the Palestinian Ministry of Health’s own update of 30 June 2024, 26% of its cumulative total at that date, on the count The Lancet published in 2025 (Jamaluddine and colleagues). The ministry publishes the size of its own unresolved fraction, and the figure is its own count rather than an outside estimate of it.
Those records were excluded from the hospital list used in the analysis, along with 930 people reported missing from the survey list. The reason is mechanical rather than editorial. A capture-recapture test works by matching the same person across lists, and a decedent carried without an identifier cannot be matched to a second record. Entries that cannot be matched are removed before the estimation begins (Jamaluddine and colleagues, 2025).
The consequence is that the estimate was built from a narrower base than the ministry’s own published cumulative total, and the excluded entries did not vanish from the record. They remain on the ministry’s list as deaths it reported and could not resolve. Of the 37,877 decedents the Palestinian Ministry of Health reported for the window, 28,185 were identified and the remainder were not.
A registry that publishes its own unmatched share is disclosing something specific about itself: not that the deaths did not occur, and not that they are duplicates, but that its keeper could not attach a person to them. That is a different property from the completeness question, which asks how many deaths never reached the list at all. The first is visible on the face of the ministry’s own publication. The second is only reachable by comparing the list to something outside it.
What happened to the hospitals the count depends on?
60.8% (n = 59) of the Gaza Strip’s 97 health facilities sustained damage during the first phase of the Israeli military campaign, and 35.1% (n = 34) were functionally destroyed, on the satellite radar analysis of Asi and colleagues, which Conflict and Health published in 2024.
The window is narrow and stated: 7 October to 22 November 2023, forty-six days. The method is multi-temporal coherent change detection run on Sentinel-1A radar imagery and overlaid on facility locations drawn from open-source humanitarian inventories. Functionally destroyed is not a verified loss of service but a threshold, applied where the detected damage covered half or more of a facility’s area (Asi and colleagues, 2024).
Two limits travel with those figures. Health facilities here means 97 clinics and primary-care centres, not the Strip’s 36 hospitals, so 60.8% may not be restated as a share of hospitals. And the detection is satellite-only: the authors state plainly that the absence of ground-truth data prevents them from quantifying the accuracy of radar-based damage assessment (Asi and colleagues, 2024). The denominators come from humanitarian inventories rather than an official facility register, so the percentages move with the inventory.
The relevance to counting is structural and is not a measured link. The first channel of the ministry’s registry is the hospital record, so the condition of health facilities is the condition of one of the two doors through which a death is entered. No published work joins facility damage to registry completeness, and this report sets the two records side by side rather than deriving one from the other. Asi and colleagues also state an allegation of international humanitarian law violation in their own conclusion (Asi and colleagues, 2024); that is their inference from the spatial distribution of damage, distinct from the damage figures themselves.
What happened when the registry was tested against two other lists?
64,260 deaths due to traumatic injury, 95% confidence interval 55,298 to 78,525, is the estimate a three-list capture-recapture analysis produced for the Gaza Strip from 7 October 2023 to 30 June 2024, which The Lancet published in 2025 (Jamaluddine and colleagues). Its three input lists were the ministry’s hospital records, the ministry’s own online survey and manually scraped social media obituaries.
Set against it is the ministry’s own figure for the identical window: the 37,877 decedents the Palestinian Ministry of Health reported as of 30 June 2024, of whom 28,185 were identified. The method, which the paper also calls multiple systems estimation, works from the overlaps. If three lists of the same population share members, the pattern of sharing implies how many members appear on none of them. The authors averaged alternative generalised linear models over the overlap structure, worked under London School of Hygiene and Tropical Medicine ethics reference 31101, published code and anonymised data on GitHub, and record no funding (Jamaluddine and colleagues, 2025).
The provenance of the inputs is what fixes the claim’s size. Two of the three lists are the health ministry’s own, and the third is obituary pages scraped by hand from social media. This is an academic estimate of how complete a party’s record is, built from that party’s record, and it is not an independent enumeration of the dead. The wide interval says the same thing in numbers: the lower bound sits below 56,000 and the upper above 78,000.
One figure often attached to this study is not its estimate. The frequently quoted total above 70,000 comes from the Research in context box of the same paper (Jamaluddine and colleagues, 2025), where the authors applied the same under-reporting rate to the ministry’s October 2024 figure of 41,909 and wrote that the toll probably exceeded 70,000. That is an extrapolation stated as such, not the capture-recapture result.
Does 41% mean the ministry’s count was 41% too low?
26,383 deaths separate the estimate from the reported figure, and that gap is 41% of the estimate but 70% of the count it is measured against. The under-reporting share The Lancet reported is a share of the study’s own estimate, so it is not a multiplier on the ministry’s published total.
The arithmetic is short enough to check. The estimate is 64,260 and the reported figure is 37,877, so the difference is 26,383. Divided by the estimate, that difference is 41%. Divided by the reported figure, the same difference is 70%. Both are true of one gap, and which one a sentence carries decides what the sentence claims (Jamaluddine and colleagues, 2025).
The distinction is not pedantry, because the two readings produce different numbers when a reader acts on them. Multiplying 37,877 by 1.41 yields a total near 53,400, which is neither the point estimate nor anywhere inside the interval the paper published. That product appears nowhere in the paper The Lancet published. The 41% is a statement about how much of the estimated total the registry did not carry, not a correction factor to be applied to what it did carry.
The interval moves the share as well. Against the lower bound of 55,298 the implied unrecorded share is smaller; against the upper bound of 78,525 it is larger. A single percentage quoted without its denominator and without the interval The Lancet published behind it has compressed three published quantities into one, and the compression runs in the direction that makes the claim sound sharper than the paper does.
Which deaths does that estimate leave outside it?
64,260 counts deaths due to traumatic injury and nothing else. Jamaluddine and colleagues exclude deaths from healthcare disruption, food insecurity, water and sanitation failure and disease, and state in The Lancet that their rate ratio holds even when non-injury excess mortality is ignored (Jamaluddine and colleagues, 2025).
That scope decision is deliberate, and The Lancet study records it in its own scope statement. A capture-recapture estimate needs lists that record the same defined event, and a violent death entered in a morgue ledger is such an event. A death from an untreated chronic condition, from contaminated water, or from the collapse of a maternity service is not recorded that way by any of the three lists, so it cannot be estimated from them.
The consequence is that the study answers a narrower question than the public argument usually attaches to it. It measures how many violent deaths a registry of violent deaths failed to carry. It does not measure how many people died because of the war.
The separate category has a named compiler rather than a measurer. Brown University published Crawford’s paper for the Costs of War project at its Watson School in 2025, a university research project report and not peer-reviewed, and it sets direct against indirect death and describes what a health-facility registry does not capture. On the completeness question Crawford relays rather than measures, writing that “Recent analysis by public health experts suggests that the number of fatalities reported by the Gaza Ministry of Health, which faces many obstacles to making a full account of the deaths, may be a significant undercount of the violent deaths” (Crawford, 2025). The paper makes no indirect-death estimate of its own; the ranges printed in it are relayed from a preprint and belong to its authors. So the two challenges pointing in the same direction are different claims with different evidence: one is a modelled estimate for a closed window, the other a statement about a class of death the instrument was never built to record.
How far above pre-war mortality did the tested period run?
39.3 traumatic-injury deaths per 1,000 people per year is the annualised rate The Lancet published for the Gaza Strip across the tested window in 2025 (Jamaluddine and colleagues). That is a rate ratio of 14.0 against all-cause mortality in Gaza in 2022, so the multiple is measured against the territory’s own pre-war experience and not against any other place.
Both figures carry intervals the paper publishes and a citing sentence should carry with them (Jamaluddine and colleagues, 2025). The rate’s 95% confidence interval runs from 35.7 to 49.4, and the rate ratio’s from 12.8 to 17.6. A crude death rate of this kind states deaths per 1,000 people per year, so it can be compared across periods of different length, which a running total cannot.
The choice of baseline is what makes the ratio interpretable. All-cause mortality in Gaza in 2022 is the territory’s own experience in the year before the war, covering every cause of death rather than violent ones alone. The comparison is therefore internal: this population against itself, one year earlier, not this population against a national average elsewhere.
The rate is derived from the estimate rather than from the ministry’s reported total, which means it inherits the estimate’s uncertainty rather than resolving it. It also answers a different question from the one this report opened on. Completeness asks how much of what happened reached the list. Intensity asks how heavily mortality ran in the period, whatever the list holds. A rate ratio of 14.0 is a statement about intensity, and it is one the authors say survives even if non-injury excess deaths are set aside entirely, which is the narrowest possible reading of their own data (Jamaluddine and colleagues, 2025).
Who among the counted could be identified by age and sex?
16,699 of the 28,257 deaths for which age and sex data were available, 59.1%, were women, children and older people, on the figures The Lancet published in 2025 (Jamaluddine and colleagues). The denominator is that identified subset, never the 64,260 estimate.
The Lancet records the categories: women, children under 18, and people aged 65 and over. The subset is defined by what the record carried, namely those deaths for which both an age and a sex were present.
Two nearly identical numbers in the same study measure different things and are not interchangeable. 28,185 is the count of identified decedents among the 37,877 the ministry reported for the window. 28,257 is the count of deaths carrying age and sex data. They differ by 72, they are constructed differently, and a sentence that swaps one for the other has quietly changed its denominator.
The share cannot travel upward either. Applying 59.1% to the 64,260 estimate produces a figure near 38,000 that appears nowhere in the paper, because the estimate covers deaths that by definition were not on any list carrying an age and a sex. And the share is not a civilian proportion: the ministry publishes no combatant-noncombatant field at all, a point recorded in the Costs of War compilation at Brown University (Crawford, 2025), so what the demographic breakdown describes is who the record could categorise, not who was or was not taking part in hostilities. The absence of an attribution field in the ministry’s published list is a property of the instrument rather than a gap in this particular study.
Who argues the published total is too high, and on what arithmetic?
At most 41,000 civilian war-related deaths is the ceiling the Jerusalem Center for Security and Foreign Affairs published in 2025, derived by subtracting estimated natural deaths and Israel’s count of 21,000 Hamas fighters killed from the Gaza Ministry of Health total. The Center states on the same page that neither input has been independently confirmed.
The ceiling is the publisher’s own arithmetic, and the Center describes it as a ceiling rather than a count. Its page records that neither the Hamas claim of 67,000 total deaths nor the Israeli estimate of 21,000 fighters killed has been independently confirmed, and that the former has been challenged. So the subtraction runs on three quantities, two of which its own author marks unverified, and a subtraction cannot be more certain than the numbers it starts from.
The class of input a calculation like this requires is illustrated by the Israel Defense Forces briefing of 29 February 2024, at which the IDF Spokesperson stated over 13,000 people the IDF describes as terrorists eliminated in Gaza since the beginning of the war. That is a party’s own count under a party’s own classification, published by a participant in the events, and the Israel Defense Forces page writes it as a floor rather than an exact total. Its window also closes long before the ministry totals from which the ceiling is computed.
The structural difficulty is that no shared identifier exists. The ministry’s registry records no combatant status, and no Israeli combatant figure is joined to it entry by entry, so a fighter count produced by one party is being deducted from a total produced by another with nothing linking the two records at the level of a person. The result is arithmetic performed across two books that were never designed to be read against each other.
Does the Palestinian statistical authority measure the toll itself?
70,942 deaths in Gaza by the end of December 2025 is the figure the Palestinian Central Bureau of Statistics reports, and the Bureau attributes it to the Palestinian Ministry of Health rather than measuring it. The Bureau’s own measurement is a different quantity entirely.
What the Bureau measures is population. Its end-of-2025 brief puts Gaza at about 2.13 million, a decline of approximately 254,000 people, which it states is a decrease of 10.6% against its pre-aggression population estimates. Those two figures belong together and reconcile on their own base. The brief separately reports that roughly 2.2 million people were living in Gaza on the eve of the war, that nearly two million of them were displaced, and that roughly 100,000 have left the Strip.
The decline is not a death toll and the Bureau does not present it as one. It attributes the change to war-related deaths, missing persons, reduced births and forced out-migration together, four components folded into one net demographic figure with no itemisation between them (Palestinian Central Bureau of Statistics, 2025). A reader cannot extract a casualty number from it, and a reader who divides one endpoint by another is dividing quantities the brief defines differently.
The Bureau is also the Palestinian Authority’s own statistical office, which makes it a body reporting on its own population, and the brief describes its figures as estimates rather than enumeration. What is notable for the counting question is the relationship it declares: the national statistical authority, the institution that would ordinarily be the independent counter of a population, publishes the death figure as the health ministry’s and reserves its own measurement for a different quantity. Two official Palestinian bodies are on the record, and only one of them says it counted the dead. The wider series behind these totals is set out in the Institute’s report on Israeli and Palestinian population growth.
Methodology
This report examines what the Gaza Ministry of Health’s registry captures, what published work has tested it against, and what each of those records measures. Every figure carries the body that produced it, the window it covers and its source type.
Peer-reviewed research. Jamaluddine, Abukmail, Aly, Campbell and Checchi (2025) in The Lancet supplies the capture-recapture estimate of 64,260 traumatic-injury deaths and its interval, the 37,877 and 28,185 ministry figures for the same window, the 22,368 and 5,817 input-list counts, the 9,692 unidentified decedents, the annualised rate and rate ratio, and the 59.1% demographic share with its denominator of 28,257. Asi, Mills, Greenough, Kunichoff, Khan and colleagues (2024) in Conflict and Health supplies the health-facility damage figures for 7 October to 22 November 2023.
Research and university project reports. Crawford (2025) for the Costs of War project at Brown University’s Watson School is a university research project report and not peer-reviewed; it is cited for its direct-and-indirect scope statement and for what it relays about registry completeness, never for a casualty total. The Jerusalem Center for Security and Foreign Affairs (2025) is a research institute and not peer-reviewed; its at-most-41,000 ceiling is its own arithmetic on inputs supplied by parties to the conflict.
Official and party records. The Israel Defense Forces Spokesperson’s briefing of 29 February 2024 is a self-reported count using the IDF’s own classification of those counted, published by a party to the conflict, and its figure is written as a floor. The Palestinian Central Bureau of Statistics brief for the end of 2025 is an official statistical product of the Palestinian Authority reporting on its own population, and its figures are estimates rather than enumeration.
Limitations. The capture-recapture estimate is built from two of the ministry’s own lists plus scraped obituaries, so it tests a record rather than enumerating the dead, and its confidence interval is wide. The satellite damage study has no ground truth, its authors say so, and its denominator is 97 health facilities rather than the Strip’s 36 hospitals. No published work joins facility damage to registry completeness, and the two are set side by side here rather than linked. The 59.1% carries a denominator of 28,257 and is not applied to any other figure. Nothing here is summed or subtracted across bodies: the ministry, the Bureau, the Center and the Israel Defense Forces count different populations over different windows on different definitions, and each figure is reported only against its own base.
Conclusion
So does the Gaza Ministry of Health undercount or overcount? Two published corrections point in opposite directions, and both are computed from the same registry. Jamaluddine and colleagues estimated 64,260 traumatic-injury deaths against 37,877 reported, a gap of 26,383 that is 41% of their estimate and 70% of the ministry figure. The Jerusalem Center for Security and Foreign Affairs subtracted estimated natural deaths and a party’s 21,000-fighter count from the ministry total to reach at most 41,000 civilians, and wrote that neither input is confirmed.
What can be described exactly is the instrument. It has two channels, one of them opened almost three months into the war. It carried 9,692 entries its own keeper could not identify, 26% of its cumulative total, and it publishes that fraction itself. It records no combatant status, so nothing in it can be netted against a fighter count. The facilities feeding its first channel were damaged in 59 of 97 cases inside the first forty-six days. Its demographic breakdown covers 28,257 deaths and not the 64,260 estimate. Each of those is checkable, dated and attributed, and none of them settles the total.
That is the shape of the record: a registry describable to the entry, and a toll that no published source enumerates independently. The correction that raises the figure runs on the registry’s own two lists. The ceiling that lowers it runs on one party’s estimate subtracted from another party’s total, across books with no shared identifier. Both are measurements of the record rather than of the dead, and both are offered to the public as the latter. What would a reader have to be handed, and by whom, before either one could be called a count?
Frequently Asked Questions
What does capture-recapture actually do?
It estimates how many members of a population appear on none of the available lists, by working from how heavily the lists that exist overlap. The Lancet published such an analysis of three lists in 2025, averaging alternative generalised linear models over the overlap structure and publishing its code and anonymised data (Jamaluddine and colleagues, 2025). The paper also calls the technique multiple systems estimation.
Is 64,260 the same as the figure reported in coverage as more than 70,000?
No. The 64,260 is the capture-recapture point estimate for 7 October 2023 to 30 June 2024. The above-70,000 figure comes from the paper’s Research in context box, where the same authors applied their under-reporting rate to the October 2024 total the Palestinian Ministry of Health reported, 41,909, and wrote that the toll probably exceeded 70,000, an extrapolation they label as such (Jamaluddine and colleagues, 2025).
Why is the confidence interval on the estimate so wide?
The published interval runs from 55,298 to 78,525, a span of more than 23,000 deaths around a point estimate of 64,260. An estimate of this kind is driven by how much the input lists overlap, and where the lists are self-reported or scraped, the overlap is measured with less precision than a matched administrative record would allow. The Lancet published the interval rather than the point estimate alone for that reason.
What does functionally destroyed mean in the satellite analysis?
Conflict and Health published it as a threshold for a facility where detected damage covered half or more of the building area, read off Sentinel-1A radar imagery (Asi and colleagues, 2024). It is a threshold applied to imagery, not a verified loss of service, and the authors state that the absence of ground-truth data prevents them from quantifying the accuracy of radar-based damage assessment.
Why can a study built on the ministry’s own lists still test them?
Because the two ministry lists were compiled independently of one another, one from morgue records and one from public reporting, and a third list of scraped obituaries was assembled outside the ministry altogether. The overlaps between them carry information the individual lists do not. What the design cannot deliver is an independent enumeration, which is why the analysis The Lancet published describes an estimate rather than a count.
Does the population decline the Palestinian Central Bureau of Statistics reports work as a death toll?
No. The Bureau reports a decline of approximately 254,000 people, 10.6% against its pre-aggression estimates, and attributes it to war-related deaths, missing persons, reduced births and forced out-migration together. The four components are not itemised, and the Bureau presents the figure as a net demographic estimate rather than a casualty count.
What does the annualised rate measure that the totals do not?
A rate states deaths per 1,000 people per year, so periods of different length can be compared with one another. The Lancet put traumatic-injury mortality at 39.3 per 1,000 per year for the tested window (Jamaluddine and colleagues, 2025), a rate ratio of 14.0 against all-cause mortality in Gaza in 2022. A running total cannot be compared that way, because it grows with the length of the period alone.
Sources
- Asi, Mills, Greenough, Kunichoff, Khan and colleagues, 2024. “Nowhere and no one is safe”: spatial analysis of damage to critical civilian infrastructure in the Gaza Strip during the first phase of the Israeli military campaign, 7 October to 22 November 2023. Conflict and Health, 18(24). DOI 10.1186/s13031-024-00580-x. Peer-reviewed, open access.
- Crawford, 2025. The Human Toll of the Gaza War: Direct and Indirect Death from 7 October 2023 to 3 October 2025. Costs of War project, Watson School of International and Public Affairs, Brown University. costsofwar.watson.brown.edu/paper/HumanTollGaza. University research project report, not peer-reviewed.
- Israel Defense Forces, 2024. Press Briefing by IDF Spokesperson, Rear Admiral Daniel Hagari, February 29th, 2024. idf.il. Official record (a party to the conflict, self-reported).
- Jamaluddine, Abukmail, Aly, Campbell and Checchi, 2025. Traumatic injury mortality in the Gaza Strip from Oct 7, 2023, to June 30, 2024: a capture-recapture analysis. The Lancet, 405(10477), pp. 469 to 477. DOI 10.1016/S0140-6736(24)02678-3. Peer-reviewed.
- Jerusalem Center for Security and Foreign Affairs, 2025. The Death of Genocide. jcfa.org/the-death-of-genocide/. Research institute, not peer-reviewed.
- Palestinian Central Bureau of Statistics, 2025. A Brief on the Status of the Palestinian People at the End of 2025. Ramallah. pcbs.gov.ps. Official record (national statistical office).