Cross-sectional assessment of retrospective mortality in humanitarian emergencies
Summary of published research1
A scene from Port au Prince, Haiti, following the earthquake 6 days earlier.
The rates and causes of mortality are critical indicators of the overall health of a population. Because of the importance of such data, it is essential to assess mortality rates in settings without complete vital statistics reporting or death surveillance. Demographers have done extensive work on methods to estimate mortality rates in such settings. Since humanitarian emergencies may differ so substantially from the relatively stable settings for which many retrospective approaches to assessing mortality were developed, the extent to which these approaches can be applied in humanitarian emergencies may be limited. A recent paper reviews basic methods that may be employed in humanitarian emergencies to determine crude, under-five and maternal mortality rates in crosssectional surveys and provides references to more detailed descriptions of these methods.
Basic approaches to measure mortality rates
There are four basic types of approach to measuring mortality rates in cross-sectional surveys. The most basic method consists of 'single-round' survey where surviving members of households, interviewed at a single point of time, report the aggregate number of deaths among household members that occurred during some period in the past. However, this method has severe limitations, e.g. underestimation of rates with greater omission of deaths in certain age groups and differential reporting by gender. The cumulative effect of these errors may result in underestimation of rates by as much as 30-40%. These shortcomings led in the 1970s and 1980s to the development of more complex approaches.
The second method, the 'multi-round' approach, uses consecutive interviews of the same individuals or individuals in the same households conducted at regular intervals to detect changes in household composition over time. Because it relies to a greater extent on the enumerator's observations and less on the recall of respondents, this approach is more robust. However, the approaches are logistically complicated and expensive and may be unsuited to humanitarian emergencies because of the mobility of populations, dissolution of households and evolving security concerns. Moreover, survey results are needed urgently to address emergency health conditions and cannot wait for multiple survey rounds.
The third approach involves the use of survivorship methods. Here, simple questions are asked of surviving relatives. For example, the number of childhood deaths may be estimated by asking women for the total number of live-born children they have ever had, and the number of these still alive. Transformation of the proportion of children still alive into survivorship estimates depends on the application of a set of multipliers to the proportion of children surviving for each five year age group of women. One child survivorship method has been adapted for use in humanitarian emergencies. However, may of these methods have limited use in humanitarian emergencies due to population movements and the fact that rates tend to apply to a period preceding the present by many years.
A fourth approach is the adaptation of the basic single-round method where interviewers ask household respondents to list each member of their households. Both the numerator and denominator of the estimated mortality rate are then based on an enumeration of individuals resident in a household, a process frequently referred to as a household census. This has emerged recently as the approach most frequently employed in humanitarian emergencies. The past period of interest is much shorter than with the survivorship method, thus providing an estimate of recent mortality rate. This approach encompasses several methods.
Retrospective assessment of cause of death is often based on the responses of surviving family members to questions contained in verbal autopsy algorithms.
Data collection methods
Crude mortality rate
The crude mortality rate (CMR) is the rate of death among everyone in a specific population. Regardless of the method of data collection, CMRs are calculated from a numerator (the number of deaths), a denominator (the size of the population within which these deaths occurred), and a time element. Measuring mortality rates retrospectively requires the precise definition of a past time period, called the recall period. When measuring mortality rates in a crosssectional survey, the number of persons needed in the survey sample to achieve a given level of precision around the point estimate of the mortality rate depends on, among other factors, the length of the recall period. The longer the recall period, the more person-time units are included in the denominator with the same survey sample size. However, in humanitarian emergencies, where mortality rates may be changing rapidly and public health professionals require a relatively recent estimate of mortality rates, a shorter recall period, and hence a larger sample size of households, may be preferable.
In the absence of an estimate of the baseline CMR, a CMR of 1/10,000/day or above can be used as the definition of the acute phase of an emergency. This threshold reflects experience in high-fertility countries in Sub-Saharan Africa but predates the HIV/AIDS pandemic. It should also be borne in mind that the CMR is influenced by the underlying age structure of the population. This means a younger population may have a lower CMR because, in general, children (especially those older than five years of age) have a lower rate of mortality than that of elderly persons.
In emergency settings, several methods have been used to gather necessary data for calculation of CMR.
- Some survey methods have focused on only the total number of people who live in the household and the total number of deaths since the beginning of the recall period. This method requires only a few questions and only a few minutes at each household. However, validation studies in stable settings indicate that surveys generally underestimate mortality rates. Several methods have been employed to address this tendency to underestimate number of deaths. One is the 'past household census method' where respondents list each person who lived in the household at the beginning of the recall period and then discuss their current status. The method has to assume that household departures and entrances were evenly spread throughout the recall period.
- Another method is called the 'current house hold census method' where respondents are asked to list each current household member, persons who died or left the house hold during the recall period and to identify those currently members who entered the household during the recall period. The basic population denominator of current household members is then adjusted by subtracting one-half of a recall period for persons who entered the household and adding one-half of a recall period for persons who left the household during the recall period.
- A third method called the mid-interval population involves a complete household census at the time of the survey and at the beginning of the recall period. The population denominator is then the average of the population at the beginning and end of the recall period. Such double enumeration, although taking more time, provides inter nal validation of the data from each household. However, interviewers must take care not to miss deaths among newborns and infants. None of these methods has been validated against a more accurate process for counting deaths, such as a death registration system with good reporting.
If the age and sex of deceased and surviving household members are determined with reasonable accuracy, age or sex-specific mortality rates can be calculated. If some estimate of the cause of each death is determined, causespecific mortality rates can also be ascertained. Sample size will, however, determine precision. If the date of each death can be gauged accurately, death rates for sub-intervals of the recall period can be calculated and used to monitor trends over time.
The household census methods suffer from several potential biases, including those associated with sampling which are inherent in all surveys. Biases in emergencies that are particularly relevant are non-representative samples, possibly due to insecurity and survivor bias when all household member are dead or surviving members cannot be interviewed. Non-sampling biases may include intentional distortion by respondents. This can work both ways, e.g. respondents wish to exaggerate their plight or conversely not report deaths to ensure that food aid is not reduced. Recall bias may also occur as traumatised populations may not remember deaths during the acute emergency or may be confused about timings. To some extent, the potential for these latter biases can be minimised by the employment of a double enumeration, the use of an appropriate household respondent and careful selection, training and supervision of field workers.
Under-five mortality rate
Under-five mortality rates have been of particular interest to public health workers as young children are frequently the target of specific interventions. Furthermore, these rates can be a particularly helpful index of mortality in humanitarian emergencies because young children are more vulnerable to death than older children or adolescents and therefore have higher rates. These rates may also rise before CMR in some emergencies making this an earlier indicator of worsening health.
Under-five mortality rates can be calculated using the household census approach, although sample sizes may frequently be too small to permit precise estimates.
Another family of approaches is based on the 'indirect' method developed in the 1960s in which women in defined age groups respond to pre-set questions on the number of children ever born alive and the number surviving. No information is collected on dates of birth or death. Instead, models of fertility and mortality are used to distribute these events in time and to estimate the probabilities of surviving to certain ages. Such methods are not useful in emergencies because they depend on the simplifying assumptions (unlikely to be true in such contexts) that mortality rates have been stable over time and that survival of children is independent of that of their mothers.
Another approach that has been employed for many years in development contexts is previous birth history (PBH). This method collects the dates of births and deaths of children from a sample of women of reproductive age and uses the data to construct life-tables for the most recent five year period. However, this method still gives estimates of mortality rates centred on a date preceding the survey by approximately two and a half years. PBH has now been amended to give more recent estimates in humanitarian emergencies. The PBH approach minimises underestimate and manipulation by not asking directly about deaths. In addition, respondents have a single relationship to the decedents, unlike other methods in which any available adult household member may report on any household deaths. However, as with household census methods, survivor bias may lead to underestimating the rate. If the mother is missing or dead, no deaths can be reported. This will be particularly important in situations of high maternal and under-five mortality. There have been few attempts to validate the completeness of early childhood deaths collected via the PBH approach. However, this method is the standard one currently used in most large population-based surveys that measure infant and child mortality around the world.
Maternal mortality rate
Measuring maternal mortality is even more challenging than measuring crude and early childhood mortality in humanitarian emergencies and post-conflict settings. Even where measures of maternal mortality are high, maternal deaths remain rare events.
Three basic approaches exist. The first relies on incorporating questions into a comprehensive national population census but this is not realistic in most humanitarian situations. A second approach uses various survivorship methods; by far the best known of these are the direct and indirect sisterhood methods. These methods are also less appropriate for emergencies as they produce an average for many years in the past that may be out of date. They also require sophisticated data analysis techniques with which many personnel in emergencies may not be familiar. The third approach is the reproductive age mortality survey (RAMOS) which gathers data only on women of reproductive age who are included in the survey sample. The RAMOS is conducted in two stages. Identification of all deaths among women of reproductive age through for example, review of burial sites, health records, census data or household surveys, followed by further investigation to determine the cause of each death using, for instance, information from health facilities, death certificates or verbal autopsy interviews with families of deceased women. Because of the relative rarity of maternal deaths, data must be gathered from a very large number of household to obtain a reasonably precise estimate of maternal mortality. This requires many resources and can be logistically difficult, particularly in humanitarian emergencies. Data collected directly in a RAMOS allow direct estimation of proportional mortality among women of reproductive age due to reproductive causes. With information on births and population size, the maternal mortality ratio, maternal mortality rate and lifetime risk of maternal death may also be calculated. Additional methods and tools for maternal mortality have been developed in the past several years, which will merit assessing their feasibility as to usefulness in emergency settings.
In order to produce an accurate point estimate of mortality, prospective death reporting necessitates relatively complete reporting of deaths and a good estimate of the population denominator. These requirements can often not be met in acute humanitarian emergencies. A variety of methods have been developed to measure mortality retrospectively in settings without prospective death surveillance. However, many of the methods require underlying assumptions or produce an estimated mortality rate for time periods in the relatively distant past, making them less suitable for use in humanitarian emergencies. Careful fieldwork, appropriate selection of respondents, and awareness of the limitations and potential sources of bias in those methods best suited for humanitarian emergencies will maximise the accuracy, and hence the usefulness, of the results form such surveys. In addition, it is important to remember that the validity of the results of mortality surveys also depends on all of the factors that must be considered when carrying out any survey, such as the sampling scheme used, careful training and supervision. Finally, studies are needed to validate the methods described against more accurate methods of counting deaths and to compare them to one another.
1Cairns. K et al (2009). Cross-sectional survey methods to assess retrospectively mortality in humanitarian emergencies. Disasters, volume 33 (4), pp 503-521
More like this
FEX: Postscript on measuring mortality rates in cross-sectional surveys: a commentary
Bradley A.Woodruff, MD MPH, Medical Epidemiologist International Emergency and Refugee Health Branch U.S. Centres for Disease Control and Prevention Congratulations to Mark...
FEX: A method for estimating mortality rates using previous birth history
By Mark Myatt, Anna Taylor and W. Courtland Robinson Mark Myatt is a consultant epidemiologist and senior fellow at University College London. His areas of expertise include...
en-net: Confusion over CDR and U5DR sample size calculation
Hi, I recall reading in SMART manual that if U5DR is one of the objectives of the surveys apart from CDR a separate sample size calculation should be done for U5DR. However,...
en-net: MUAC and Mortality relationship
I have found a nutritional survey report conducted in Sudan with GAM of 27.0% and SAM of 7.5%, while the U5MR IS 1.19/10 000/day using ENA for SMART software. Does this result...
FEX: High levels of mortality, malnutrition and measles amongst displaced Somali refugees in Dadaab, Kenya
Summary of published research1 Location: Dadaab, Kenya What we know already: Routine vaccination of children aged 6 months to 15 years, supplemented by mass vaccination...
en-net: Length of recall period during SMART survey
What determines the choice of the 'length of the recall period' in measuring mortality when undertaking SMART assessments . Rogers The recall period in Mortality surveys...
en-net: How to interpret expected death count?
Hi expert in the house I am confused on how to use the crude death per 10,000 per day or per year unit when computing expected death count. eg: I have a crude death rate of...
en-net: Relationship between Mortality and Acute Malnutrition
In most surveys I have seen, there is a positive linear relationship between CMR/U5MR and acute malnutrition. Also, the report...
FEX: Improved surveillance prevents excess mortality: the Gode experience
Summary of published paper1 A study on the epidemiology of the famine in Gode district of Ethiopia has just been published. The primary objectives of the study (which was...
en-net: How to convert DHS under five mortality rate to SMART under 5 mortality estimate?
Is there anyway of converting DHS reported under 5 mortality estimates to SMART survey under 5 mortality estimates? For instance a DHS report stated that the under 5 mortality...
en-net: Negative Confidence Interval
What does negative confidence interval implies? In some assessment findings, I noticed a -ve CI. For e.g., SAM 0.2 [-0.2-0.6 95%CI]. Also, how wider the CI that we allowed to...
FEX: New Method to Estimate Mortality in Crisis-Affected Populations
The Food and Nutrition Technical Assistance II Project (FANTA-2) has prod- uced the report A New Method to Estimate Mortality in Crisis-Affected Populations: Validation and...
FEX: Mortality Rates in Conflict Affected Darfur
Published paper1 Serverly wasted adult male A recent article in the Lancet reported on the findings of a mortality survey in West Darfur, Sudan for the period 2003-4...
en-net: Effect of Recall Period on CMR and U5MR: SMART +Mortality Survey
I am planning to implement a SMART+Mortality survey in an area that has had access issues owing to different administrative structures. Since I can't use the same...
FEX: Patterns of mortality rates in Darfur
Summary of published research1 Several mortality estimates for the Darfur conflict have been reported since 2004, but few accounted for conflict dynamics such as changing...
en-net: When to consider population movement within recall period for mortality survey using SAMRT methodology
when we do mortality survey, if we agreed the recall period is three months then if there someone moved from the HH within this period but will be back agin, shall we count...
FEX: Mortality in Somalia during food insecurity and famine (2010-2012)
Women and children wait for assistance from the drought/famine in Dolo, southern Somali The report summarised below on mortality rates during the Somali famine of 2010-12...
FEX: Mortality in the DRC
Summary of published research1 IRC mortality survey team in transit Since 2000, the International Rescue Committee (IRC) has documented the humanitarian impact of war and...
en-net: A New Method to Estimate Mortality in Crisis-Affected Populations: Validation and Feasibility Study.
In emergency settings, data on mortality rates and the causes and circumstances of death are crucial to guide health interventions and monitor their effectiveness. Current...
FEX: How many lives do our CMAM programmes save? A sampling-based approach to estimating the number of deaths averted by the Nigerian CMAM programme
By Assaye Bulti (UNICEF), Stanley Chitekwe (UNICEF), Chloe Puett (ACF), and Mark Myatt (Brixton Health) Assaye Bulti is responsible for managing routine nutrition programme...
Reference this page
Cross-sectional assessment of retrospective mortality in humanitarian emergencies. Field Exchange 38, April 2010. p11. www.ennonline.net/fex/38/cross