Mortality in a humanitarian crisis: how it is measured

When a war, a famine or a mass displacement begins, one of the first questions is how many people are dying, and whether that is more than before. It sounds like a statistics question, far from the clinic. It is not. The number of deaths guides where vaccines, food, water and teams go, defines whether a situation is considered an emergency and, later, serves to judge whether the response worked.

The problem is that, precisely when that number is needed most, the systems that would produce it stop working. Registry offices close, hospitals stop recording, families flee. This text explains how mortality is measured under these conditions, what thresholds are used, which biases distort the results and how a clinician can read these numbers with a critical eye.

The unit used and why

In acute humanitarian emergencies, mortality is usually expressed as the number of deaths per 10000 people per day (Prudhon and Spiegel, 2007). The choice of unit is not arbitrary. In a crisis, things change within weeks, and an annual rate would hide exactly the peak that matters. Expressing it per day allows short periods to be compared and the trend to be followed.

Two indicators are most used to measure the severity of a crisis and assess the effectiveness of aid: the crude mortality rate, which includes all ages, and the under-five mortality rate (Working Group for Mortality Estimation in Emergencies, 2007). Young children are the first to feel the lack of food, clean water and vaccines, and so the rate in this age group usually rises earlier and more.

An example helps to give scale. In a camp with 50000 people, a crude rate of 1 death per 10000 per day corresponds to 5 deaths daily, or about 150 in a month. If the rate doubles, that is 10 deaths per day. For those working at the camp health post, this difference shows up in the number of bodies, the number of bereaved families and the strain on the team long before it shows up in a report.

Emergency thresholds serve as a ruler. In the analysis by Ogbu and colleagues, for example, the emergency threshold adopted for under-fives was 2 deaths per 10000 children per day (Ogbu and colleagues, 2022). A study by Medecins Sans Frontieres in eastern Democratic Republic of the Congo describes a territory where a first survey by the organization, in 2013, showed mortality above crisis thresholds and a second, in 2017, showed rates just below them, but still with excess deaths and major health needs (Robinson and colleagues, 2019).

That last point deserves attention. Being below the threshold does not mean being well. The threshold is a practical line for deciding priorities, not a border between health and disease. It is the same care taken with cutoffs in tests: a value just below the cutoff does not guarantee normality, as discussed in the text on sensitivity, specificity and predictive values.

When civil registration disappears

Under normal conditions, mortality comes from civil registration: each death has a certificate, a cause and a place. In crises, prospective registration of vital events is absent or deteriorates in almost all affected populations (Working Group for Mortality Estimation in Emergencies, 2007). For this reason, retrospective household surveys are widely used to estimate and describe mortality patterns (Working Group for Mortality Estimation in Emergencies, 2007).

The working group on mortality estimation in emergencies described the three components of these surveys (Working Group for Mortality Estimation in Emergencies, 2007). The first is selecting a representative sample of households, usually by clusters. The second is administering a questionnaire to a respondent in each household, asking about deaths and other demographic data over a defined recall period. The third is attempting, even if roughly, to determine the cause of each reported death (Working Group for Mortality Estimation in Emergencies, 2007).

Two-stage cluster sampling was originally created to measure vaccination coverage and came to be used for mortality because of the lack of time and resources in emergencies (Working Group for Mortality Estimation in Emergencies, 2007). The authors note that the method is usually followed without considering alternatives, and they call for more research on sampling, household data collection and determination of cause of death (Working Group for Mortality Estimation in Emergencies, 2007).

When there is a minimally functional vital registration system, another approach is possible. Karlinsky and Torrisi used registration data by age and sex to estimate the deaths of the 2020 war in Nagorno-Karabakh, comparing observed deaths with those expected from the 2015 to 2019 trend (Karlinsky and Torrisi, 2023). They estimated almost 6500 excess deaths among people aged 15 to 49, heavily concentrated among adolescent and young adult men (Karlinsky and Torrisi, 2023).

Household surveys in practice

To understand what these surveys can show, it is worth looking at a few concrete examples.

In the Central African Republic, Gang and colleagues conducted surveys in two strata, one in the area under government control and another in the area mostly outside that control, drawing 40 clusters of 10 households in each (Gang and colleagues, 2023). They visited 70 of the 80 clusters, with 699 households and 5070 people interviewed (Gang and colleagues, 2023). The crude mortality rate was 1.57 deaths per 10000 people per day (Gang and colleagues, 2023). Outside the government-controlled area, the rate was 1.85, against 1.31 inside it (Gang and colleagues, 2023).

The data on causes surprises those who think of war only as violence. Families reported malaria or fever and diarrhea as the main causes of death, and violence accounted for 6 out of every 100 deaths (Gang and colleagues, 2023). The article title sums up the tension: the mortality found was much higher than official statistics suggested (Gang and colleagues, 2023).

In the Walikale territory, in eastern Democratic Republic of the Congo, a 2017 survey included 5711 people in 794 households (Robinson and colleagues, 2019). The crude rate was 0.98 and the under-five rate was 1.3 deaths per 10000 per day (Robinson and colleagues, 2019). The most reported causes were fever or malaria, diarrhea and respiratory infections, and the results helped expand integrated community management of these diseases (Robinson and colleagues, 2019).

The two examples show something the clinician recognizes. In protracted crises, most deaths do not come from bullets or bombs, but from common diseases that go untreated because services fall into disarray. Fever in a malaria-endemic region is a problem of geography as much as of clinical care, as the text on how geography changes the probability shows.

Biases and quality of surveys

Every survey measures only those who can be reached, and this creates predictable biases.

The first is absence. In the Central African Republic, about one in five households was absent at the time of the visit, mainly in the safer areas (Gang and colleagues, 2023). Those who are not at home do not answer, and it is not known whether mortality in these households is the same as in those that answered.

The second is that of insecurity. Dianou and colleagues compared a telephone survey with the traditional household survey in Burkina Faso, interviewing 11387 women of reproductive age (Dianou and colleagues, 2026). In regions without direct conflict, under-five mortality was relatively stable, in a pattern similar to that of the household survey (Dianou and colleagues, 2026). In regions affected by armed conflict, however, the telephone survey showed a substantial increase, from 47 to 82 deaths per 1000 live births (Dianou and colleagues, 2026). This increase did not appear in the in-person survey, very likely because the most insecure clusters were excluded from the sample (Dianou and colleagues, 2026).

This finding is a perfect example of selection bias. The places where most people die are precisely the places where interviewers cannot reach. The average result looks reassuring because the worst part was left out. It is the same reasoning as over-triage and under-triage: the error you cannot see tends to be the most dangerous one.

There is a lesson here for anyone reading crisis reports. When a number looks better than expected, the first question should be who was left out. If the sample excluded combat zones, blocked roads or groups that fled, the number says more about who was reached than about the whole population.

The third bias appears when entire families died or scattered. No one is left to tell the story. McGowan and colleagues, studying the war in Tigray, used an online survey with the diaspora, asking about the extended family that remained in the region, and standardized the estimates by rural or urban residence and by wealth level to reduce selection bias (McGowan and colleagues, 2025). It is a creative solution, but one with its own limits, and the authors themselves acknowledge that data from the pre-war period is probably underreported (McGowan and colleagues, 2025).

Even when the survey reaches the population, the quality of execution varies widely. Prudhon and Spiegel analyzed surveys conducted by 33 non-governmental organizations and United Nations agencies between 1993 and 2004: 368 nutritional surveys and 158 crude mortality surveys (Prudhon and Spiegel, 2007). Only 5 of the 158 mortality surveys met the quality criteria, and there was no improvement over that period, unlike what happened with the nutritional surveys (Prudhon and Spiegel, 2007).

The authors identified important errors and imprecision in methodology and reporting, and argued for strengthening initiatives to standardize survey quality (Prudhon and Spiegel, 2007). They recall that adequate sampling and sufficient sample size are essential to ensure representativeness and precision (Prudhon and Spiegel, 2007).

Step	What can go wrong	Likely effect
Sampling	Insecure areas excluded (Dianou and colleagues, 2026)	Underestimating mortality
Household visit	Many households absent (Gang and colleagues, 2023)	Result with uncertain direction
Recall period	Imprecise memory of dates (Working Group for Mortality Estimation in Emergencies, 2007)	Deaths outside the period or forgotten
Cause of death	Lay report, without examination (Working Group for Mortality Estimation in Emergencies, 2007)	Coarse classification of causes
Execution and reporting	Method poorly applied (Prudhon and Spiegel, 2007)	Unreliable numbers

The table reminds us that a mortality number is always the end of a chain of decisions. Anyone who reads the result without asking how it was produced runs the same risk as someone who reads an exam without knowing how the sample was collected.

Excess deaths: what the crisis adds

Knowing how many people died is not enough. The more useful question is how many died beyond what would have happened without the crisis. Checchi describes these excess deaths as those that would not have occurred in a counterfactual scenario without crisis (Checchi, 2023). He argues that humanitarian aid should be assessed by its effect on excess mortality, not on total mortality, since its goal is to mitigate the impacts of the crisis and not the prior conditions (Checchi, 2023).

The hard part is establishing the counterfactual. Checchi lists the problems: there is often no recent estimate of mortality before the crisis, the country average may not reflect the affected region, and in prolonged crises assuming a fixed value is debatable, because without the crisis mortality would probably have kept falling (Checchi, 2023).

Recent studies show the scale that excess can reach. In northwest Syria, Alhiraki and colleagues interviewed 1483 households with 12268 people, comparing the pre-conflict period, from 2000 to 2010, with the conflict period, from 2011 to 2020 (Alhiraki and colleagues, 2022). The crude mortality rate rose 3.55 times between the two periods, and the conflict appears to have caused, directly and indirectly, about 874000 excess deaths (Alhiraki and colleagues, 2022). From 2017 onward, most deaths had non-violent causes (Alhiraki and colleagues, 2022).

In Tigray, of the 310 deaths that occurred in the region during the war and were reported in the survey, 224 were from intentional injuries (McGowan and colleagues, 2025). The standardized mortality rate among adults aged 15 to 49 was 21.3 per 1000 person-years during the war, against 1.0 in the previous period (McGowan and colleagues, 2025). The authors estimate more than 102000 deaths among people aged 15 or older (McGowan and colleagues, 2025).

Two patterns appear in these studies. In some conflicts, direct violence dominates the deaths, as in Tigray (McGowan and colleagues, 2025). In others, after an initial phase, non-violent causes come to predominate, as in Syria (Alhiraki and colleagues, 2022) and the Central African Republic (Gang and colleagues, 2023). For the clinician, this means that untreated chronic diseases, common infections and malnutrition can kill more than trauma over the years.

The Syrian case also draws attention to something that counting deaths does not capture. About one in seven households reported at least one major disability linked to violence since 2011 (Alhiraki and colleagues, 2022). Those who survive with amputation, spinal cord injury or neurological sequelae need care for decades, and this burden does not appear in the mortality rate.

Another detail from Tigray shows how mortality is distributed unevenly. The mortality ratio between men and women was about 2 to 1, both in adults and in the elderly, and mortality among people aged 60 or older also rose sharply during the war (McGowan and colleagues, 2025). Looking only at the average hides who is dying more, and this matters for planning care, just as it matters to protect those who provide care so that the service keeps running.

Surveys do not only count deaths; they help to understand what causes them. Ogbu and colleagues analyzed 1597 surveys from a database of complex emergencies, with information on mortality in children under five, type of population, measles vaccination coverage and global acute malnutrition (Ogbu and colleagues, 2022).

The mean mortality in children under five was 2.07 deaths per 10000 per day among internally displaced children, 0.89 among refugees and 1.20 among resident children affected by the crisis (Ogbu and colleagues, 2022). Higher measles vaccination coverage was associated with a lower chance of exceeding the emergency threshold, and higher prevalence of acute malnutrition with a higher chance (Ogbu and colleagues, 2022). Refugee and resident children had a lower chance of exceeding the threshold than internally displaced children (Ogbu and colleagues, 2022).

These results point to where to act. Vaccinating against measles and treating acute malnutrition are high-impact measures, and the internally displaced child, often with less access to organized aid, is the most vulnerable. In an improvised shelter, the logic of triage when everything arrives at once helps prioritize these children.

Causes of death and forgotten groups

Knowing what people died of is harder than knowing how many died. The verbal autopsy is a structured interview with family members about the signs and symptoms before death, and it can be combined with the social autopsy, which investigates the circumstances and the barriers to care. Thomas and colleagues concluded that the verbal autopsy can determine causes of death in populations affected by crises without another registration system, and that, combined with the social autopsy and community participation, it offers a broad view of the obstacles to seeking and receiving care (Thomas and colleagues, 2018).

The authors recall important ethical precautions, such as protecting participants from harm and avoiding repeated interviews about the death of close relatives (Thomas and colleagues, 2018). They also note that adapting the instruments to the context improves feasibility but limits international standardization (Thomas and colleagues, 2018). This is the same spirit as morbidity and mortality sessions: understanding the circumstances of each death to change what comes next.

Some groups almost disappear from the statistics. Maternal death is one example. Erhardt-Ohren and colleagues observed that there is no reliable way to estimate maternal deaths in refugee and displaced persons camps and settlements (Erhardt-Ohren and colleagues, 2024). In a scoping review of 124 publications, the methods best rated for adaptation to these contexts were those based on community informants and those based on health services, because they require little time, small samples and no statistical training (Erhardt-Ohren and colleagues, 2024). The authors acknowledge that, as in stable contexts, no method is perfect (Erhardt-Ohren and colleagues, 2024).

Why this matters for the clinician

It may seem that measuring mortality is an epidemiologist's task. But the clinician takes part in this chain in several ways. When they record a death carefully, when they note the probable cause in a field record, when they report a disease without waiting for confirmation, as discussed in the text on empirical isolation, they are producing the data that later become a rate.

Reading these numbers also requires the reasoning the clinician uses every day. A rate is an estimate with uncertainty, and a wide confidence interval calls for caution. A sample that excludes the most dangerous areas tends to underestimate the problem. A comparison with the previous period depends on how good the earlier data was. Translating rates into people helps, as the text on natural frequencies shows: 2 deaths per 10000 children per day, in a camp with 20000 children, means 4 children dead every day.

Finally, there is a parallel with prognostic reasoning. Measuring the mortality of a population in crisis is, in a sense, making a prognosis for a group, and that prognosis is only useful if it guides decisions that change the outcome.

Frequently asked questions

Why is mortality expressed per 10000 people per day?

Because, in acute emergencies, the situation changes quickly and this unit allows short periods to be tracked (Prudhon and Spiegel, 2007).

What is the emergency threshold for children?

One threshold used is 2 deaths per 10000 children under five per day (Ogbu and colleagues, 2022).

How is mortality measured without civil registration?

Through retrospective household surveys, with cluster sampling and questions about deaths in a defined period (Working Group for Mortality Estimation in Emergencies, 2007).

What are excess deaths?

They are the deaths that would not have occurred without the crisis, compared with a counterfactual scenario (Checchi, 2023).

Why can surveys underestimate deaths?

Because the most unsafe areas usually stay outside the sample (Dianou et al., 2026).

Are most deaths in war caused by violence?

It depends on the conflict. In several, non-violent causes predominate over time (Alhiraki et al., 2022).

Take-home messages

• In emergencies, mortality is measured as deaths per 10000 people per day (Prudhon and Spiegel, 2007).
• Without civil registration, retrospective household surveys are the main tool (Working Group for Mortality Estimation in Emergencies, 2007).
• Unsafe areas outside the sample lead to underestimating mortality (Dianou et al., 2026).
• Many mortality surveys had insufficient quality (Prudhon and Spiegel, 2007).
• Excess deaths require a counterfactual that is hard to define (Checchi, 2023).
• Measles vaccination and treatment of acute malnutrition reduce the risk of critical mortality (Ogbu et al., 2022).

References

1. Alrashid Alhiraki O, Fahham O, Dubies HA, et al. Conflict-related excess mortality and disability in Northwest Syria. BMJ Glob Health. 2022. PMID 35589154.
2. Checchi F. Inferring the impact of humanitarian responses on population mortality: methodological problems and proposals. Confl Health. 2023. PMID 36998020.
3. Dianou K, Masquelier B, Helleringer S, et al. Under-Five Mortality in Humanitarian Emergencies: A Comparison of Face-to-Face and Mobile Phone Survey Estimates From Burkina Faso. Trop Med Int Health. 2026. PMID 42274292.
4. Erhardt-Ohren B, McCoy SI, Feehan DM, et al. Maternal mortality estimation methodologies: a scoping review and evaluation of suitability for use in humanitarian settings. Confl Health. 2024. PMID 39702374.
5. Gang KBA, O'Keeffe J, Anonymous, et al. Cross-sectional survey in Central African Republic finds mortality 4-times higher than UN statistics: how can we not know the Central African Republic is in such an acute humanitarian crisis? Confl Health. 2023. PMID 37072800.
6. Karlinsky A, Torrisi O. The Casualties of War: An Excess Mortality Estimate of Lives Lost in the 2020 Nagorno-Karabakh Conflict. Popul Res Policy Rev. 2023. PMID 37193053.
7. McGowan CR, Cassidy-Seyoum SA, Ekoriko P, et al. Conflict-attributable mortality in Tigray Region, Ethiopia: Evidence from a survey of the Tigrayan diaspora. Popul Health Metr. 2025. PMID 40399978.
8. Ogbu TJ, Scales SE, de Almeida MM, et al. Predictors of exceeding emergency under-five mortality thresholds using small-scale survey data from humanitarian settings (1999 - 2020): considerations for measles vaccination, malnutrition, and displacement status. Arch Public Health. 2022. PMID 35765108.
9. Prudhon C, Spiegel PB. A review of methodology and analysis of nutrition and mortality surveys conducted in humanitarian emergencies from October 1993 to April 2004. Emerg Themes Epidemiol. 2007. PMID 17543104.
10. Robinson E, Crispino V, Ouabo A, et al. Mortality and health survey, Walikale, Democratic Republic of the Congo, 2017: an example of the use of survey data for humanitarian program planning. Confl Health. 2019. PMID 31832090.
11. Thomas LM, D'Ambruoso L, Balabanova D. Use of verbal autopsy and social autopsy in humanitarian crises. BMJ Glob Health. 2018. PMID 29736275.
12. Working Group for Mortality Estimation in Emergencies. Wanted: studies on mortality estimation methods for humanitarian emergencies, suggestions for future research. Emerg Themes Epidemiol. 2007. PMID 17543103.