4)_Chapter_2_-_Defining_&_measuring_mortality_(pgs 33-52).pdf

Deaths, and counting deaths, is a crucialpublic health indicator for many reasons.Death is the final and most definitive healthoutcome of many important public healthproblems. And most important causes ofpoor health in a population, when commonand severe enough, produce an elevation inthe mortality rate. Death is easily defined,making it a health outcome for which astandardized case definition is easilyapplied. Because death results from somany health problems - from chronic dis-eases, infectious diseases and injuries - themortality rate can provide an overall indica-tor of general health status of a population.Mortality rates can also provide informationon nutritional status because widespreadmalnutrition among children or adultsalmost always results in an elevation of themortality rate, especially if the level of com-municable diseases is high. Nonetheless,the mortality rate is a relatively insensitivemeasure of population health statusbecause conditions often must be quitepoor before it is markedly elevated. Themortality rate is also relatively non-specific;there are many causes of elevated mortali-ty, any one of which might lead to an

increase in the mortality rate. As a result, anelevated mortality rate can indicate thatthere is indeed a health problem in a popu-lation, but it cannot indicate the cause.

Mortality rates have been measured in manycountries for hundreds of years. Deaths areoften counted by various authorities, includ-ing religious leaders, civil authorities andpublic health professionals. In stable popula-tions, mortality rates are usually monitoredby registering deaths using a vital statisticssystem, and reporting deaths is mandatoryin most countries. However, in many coun-tries in the developing world, vital statisticssystems are far from complete. Even if func-tioning, such systems usually are disruptedearly in situations of civil conflict or displace-ment. As a result, cross-sectional surveys areoften necessary to determine accurate mor-tality rates.

Often, especially in emergency situations,initial surveys will combine the assessmentof many different health and nutrition out-comes. Recently, nutrition assessmentsurveys have begun routinely includingmortality measurement. When combin-

MEASURING AND INTERPRETING MALNUTRITION AND MORTALITY 33

Defining and Measuring Mortality

2CHAPTER

Key messages Description of the different types of mortality rates, with formulas used Crude mortality rate/crude death rate (CMR) Age-specific mortality rate (ASMR) Age-specific mortality rate for children under 5 years (ASMR-U5) Under-5 mortality rate (U5MR) Cause-specific mortality rate Infant mortality rate (IMR) Maternal mortality rate (MMR) Retrospective cross-sectional mortality surveys The use of a recall period in mortality surveys Determining the causes of mortality

ing nutrition and mortality assessment inthe same survey, it is important to con-sider the sampling strategy so as not tochoose a sample which is biased foreither outcome (See Chapter 3 for more details). Forexample, you cannot include only house-holds with children 6-59 months in thesurvey sample; the mortality rate in suchhouseholds may not reflect the mortalityrate in all households in the population.

COUNTING DEATHSJust counting the number of deaths in apopulation is not sufficient. For example,if you hear that there have been 139deaths in a certain population, this tellsyou nothing about the rate of death ifneither the size of the population fromwhich these deaths were reported nor thetime period during which these deathsoccurred were known. If you subse-quently learn that the deaths came froma population of 5,000 persons, you stillhave insufficient information: If thesedeaths occurred over 10 years, the rate ofdeath in this population is quite low; onthe other hand, if these deaths occurredin the same month, the rate of death isextremely high. This example illustratesthat every mortality rate must have:

The number of deaths (the numeratorof the mortality rate);

The size of the population from whichthe deaths came (the populationdenominator of the mortality rate);and

The time period during which thedeaths occurred.

If 127 deaths occurred in a population of19,546 over 9 months time, the deathrate would be 127 deaths per 19,546 peo-ple per 9 months. However, this rate can-not be compared to the mortality rate in

other populations of different sizes or tonumbers of deaths counted over differenttime periods. To determine if the deathrate is high, low or normal as comparedwith rates in the same population in pre-vious time periods or compared to ratesin other populations, the mortality rate ina given population must be converted toa rate using a standard populationdenominator and time period.

The first step in such a conversion is todecide how the final rate should beexpressed. There are three differentways to express the same mortality rate. a) # deaths/1,000/year: For many vital

statistics systems, which recorddeaths for longer time periods for anentire nation or province, mortalityrates are often expressed as the num-ber of deaths per 1,000 populationper year.

b) # deaths/1,000/month: In some dis-placed populations, when the acuteemergency is over and the health sit-uation stabilized somewhat, mortalityrates are sometimes expressed as thenumber of deaths per 1,000 popula-tion per month.

c) # deaths/10,000/day: During acutehumanitarian emergencies, when thenumber of deaths is totalled each dayor every few days, the mortality rateis often expressed as the number ofdeaths per 10,000 population per day.

These are three different ways to expressthe same mortality rate; the conversionamong them is merely an exercise inmathematics. For example, to convertfrom 34 deaths per 1,000 per year todeaths per 1,000 per month, just dividethe numerator by 12 (1/12th the numberof deaths will occur in one month as inone year): 34 / 12 = 2.8 deaths per 1,000per month.

34

MEASURING AND INTERPRETING MALNUTRITION AND MORTALITY

DEFINING AND MEASURING MORTALITY

35

2CHAPTER

Because the actual number calculatedvaries depending on the populationconstant and the time period, a mortal-ity rate should never be stated withoutthese parameters. For example, you can-not say that the mortality rate is 1.86;you must say that the mortality rate is1.86 per 10,000 per day. The number1.86 would mean very different things ifthe population constant and time periodwere 10,000 per day, 1,000 per year or1,000 per month.

SOURCES OF DATA TO CALCULATE MORTALITY RATESIn stable populations, it is better to col-lect information of death prospectivelywhere, as each death occurs, it is report-ed to public health or governmentauthorities. Such systems allow the cal-culation of recent death rates as fre-quently as is required, and this data canbe collected easily. In vital statisticssystems, death reporting is oftenmandatory for persons charged withburying bodies. If deaths occur pre-dominantly in clinics or hospitals, med-ical personnel may be responsible fordeath reporting. In many societies, reli-

gious leaders record deaths. In humani-tarian emergencies, someone may behired to count deaths and monitor thearea designated as the graveyard or cre-mation site.

All of these systems of counting deathsrequire separate information on the sizeof the population from which the deathsoccurred in order to calculate mortalityrates. Such information for the popula-tion denominator may come from censuscounts; census projections; populationregistration; or the monitoring of births,deaths, immigration and emigration.. Inhumanitarian emergencies and other sit-uations in which data on population sizeis poor, techniques have been developedto estimate population size.

But which population size is used as thedenominator if deaths are counted for atime period during which the popula-tion size fluctuates? One approachwould be to calculate the average popu-lation during the time period (add thepopulation at the beginning of the timeperiod to the population at the end ofthe time period and divide by 2, produc-ing the arithmetic mean population).

Example 2.1 Expression of mortality ratesEven though they may appear to be very different, mortality rates expressed using different population constants or time periods indicate the same rate of death in apopulation. For example:

Take a mortality rate of 9.6/1,000 population/year. This could be expressed as 0.8/1,000 population/month. This could also be expressed as 0.26/10,000 population/day.

The equivalency of these rates can be seen if we calculate from one rate to the other. For example, if the rate of 0.26 deaths per 10,000 per day remained constant for an enti-re year, we would expect that about 95 people would die (0.26 x 365 days in a year) outof each 10,000 people in the population.

A second approach would be to deter-mine the population at the mid-point ofthe time period and use that (this iscalled the mid-interval populationand is the method used most commonlyin vital statistics systems). However, in many situations, the popu-lation denominator for a mortality rateitself is only a rough population esti-mate and no accounting can be madefor changes in the population during thetime period of interest. This is often truein humanitarian emergencies whichhave the additional complication of highrates of in-migration and out-migration.

TYPES OF MORTALITY RATESMany different rates are used to measuremortality: Crude mortality rate/crude death rate

(CMR) Age-specific mortality rate (ASMR) Age-specific mortality rate for children

under 5 years (ASMR-U5) Under-5 mortality rate (U5MR) Cause-specific mortality rate Infant mortality rate (IMR) Maternal mortality rate (MMR)

Crude Mortality Rate (CMR)The crude mortality rate (CMR), alsocalled the crude death rate or CDR, is

defined as the number of people of allages and both sexes who die in a giventime interval divided by the total pop-ulation at the mid-point of that timeinterval. The CMR always includesthe length of the time interval and astandard population size, called thepopulation constant. For example, aCMR may be 8.5 deaths per 1,000 per-sons per year.

It is calculated by the following formulawhere: the numerator of the fraction in

parentheses is the number of deathswhich occurred in a specific popula-tion during a certain time period.Only deaths which occurred duringthis time period should be included inthe numerator of the mortality rate.

the denominator of the fraction is thenumber of people in the population inwhich these deaths occurred. Thispopulation should be well defined,and only persons fitting this definitionshould be included in the denomina-tor. For example, if you are calculat-ing the mortality rate for a certainprovince, only people who lived inthat province during the time periodof interest should be included in thepopulation denominator.

36



37

2CHAPTER

This means that each person has an average likelihood of 0.006393 (or 0.6393% chance)of dying during the 8-month period. Stating that the CMR is 377 deaths per 58,975 per 8 months has little meaning; it can-not be directly and meaningfully compared to other CMRs from previous periods orfrom other populations. However, if this rate is converted to deaths per 1,000 per year,as follows, it can be compared to other rates expressed in the same way:

1. To convert this likelihood to a likelihood for a standard population size, we multiplythis rate by 1,000 to obtain 6.4. This means that during the period of 8 months, 6.4 peo-ple died for each 1,000 people in the population.

2. To convert this likelihood to a likelihood for 1 year: 6.4 is divided by 8/12ths (8 monthsdivided by the 12 months in 1 year) = 9.6 deaths per 1,000 population per year. This ratemeans that, if the death rate for the 8-month period continued for an entire year, forevery 1,000 people in the population, there would be 9.6 deaths.

Using the formula above, the rate of deaths per 1,000 per year would be calculated asfollows:

Example 2.2 Calculation of CMR with standard time unitIn a specific population, 377 deaths occurred during a period of 8 months in a popula-tion with a mid-interval size of 58,975.

Using the information in the formula you get:377 58,975 = 0.006393 or 0.6393%

Vital statistics systems which calculate mor-tality on an annual basis use the size of thepopulation on July 1 to indicate the averagepopulation between January 1 and December31. Such systems count the number of deathsduring a year's time, divide it by the mid-interval population, then multiply by 1,000(the population constant) to get the numberof deaths per 1,000 population per year.

Since the CMR reflects the overall risk ofdeath in the population among all ages andboth sexes, it is the least specific indicator ofmortality. Mortality reflected in the CMRmay result from causes as varied as thosefrom violent deaths from massacres to thosefrom neonatal tetanus. If only one indicator

of mortality can be calculated, CMR is usu-ally the one chosen. Ideally, a newly calcu-lated CMR should be compared with a pre-vious CMR from the same population todetermine whether the mortality rate is ris-ing or falling. Such trend information can beused as an overall evaluation of health,nutrition and other interventions. However,when prior mortality data are unavailable, arough rule-of-thumb can be used: a CMR of less than 1 death per

10,000/day indicates a reasonable healthsituation;

a CMR of more than 1 death per10,000/day reflects elevated mortality; and

a CMR of more than 2 deaths per10,000/day indicates a health emergency.

Calculating CMR using person-time unitsThe denominator of a mortality rate is thenumber of people in the population; howev-er, the denominator can also be seen asbeing based on person-time instead of thenumber of persons. That is, a rate of 10deaths per 10,000 per day may be seen asthe risk of death in 10,000 people during aperiod of one day, or the risk in 5,000 peo-ple during 2 days, or the risk in 1,000 peo-ple during 10 days. In all three examples, thedenominator is 10,000 person-days (thenumber of people in the population multi-plied by the number of days in the time peri-od). This will be important later when dis-cussing the measurement of mortality incross-sectional surveys. The sample sizerequired to achieve a certain precisionaround the estimate of the CMR is the num-ber of person-time units required in thedenominator of the rate.

In general, if births and deaths are distribu-ted evenly throughout the time period, theneach person who was born or who diedduring the time period contributes, on ave-rage, _ a person-time period to the denomi-nator. For example, vital statistics systems

assume that births and deaths occur evenlythroughout the year. Therefore, each personwho was born or who died during the yearcontributed about _ a year to the denomina-tor. Use of the mid-interval population, asdescribed above, captures half of deaths andhalf of births and is one way of adjusting thepopulation denominator for the incompletecontribution of births and deaths.

Age-Specific Mortality Rates (ASMR)Age-specific mortality rates (ASMR) restrictboth the numerator and denominator to per-sons of a certain age. For example, a morta-lity rate for persons 15-49 years of age is thenumber of deaths of persons 15-49 years ofage divided by the mid-interval population ofpersons 15-49 years of age (adjusted for thelength of the time period).

ASMR is often used to determine if the rate ofdeath is substantially different from thatexpected in any specific age group. Forexample, in a survey in Kosovo, the deathrate among young men was much higherthan expected, indicating that some factordisproportionately increased the risk of deathin this age group.

38

Figure 2.1 Age-specific mortality rates, Badghis Province, March 2001 - April 2002.

3.5

3

2.5

2

1.5

1

0.5

0



39

2CHAPTERAge-Specific Mortality Rate for children under 5 (ASMR-U5)The age-specific mortality rate for childrenunder five (ASMR-U5) is calculated bydividing the number of deaths of childrenunder age 5 during a specific time periodby the number of children under age 5.This is often called the under-5 mortalityrate; however, as described below, thisterm is also used for a completely differentmeasure of mortality in children under 5years of age. For this reason, the termASMR-U5 is used here to maintain adistinction between these two differentmeasures.

The ASMR-U5 often is used as a more sen-sitive indicator of the affect of emergencyconditions on mortality. When health andnutritional status in a population deteriora-tes, the ASMR-U5 often changes earlierand to a greater extent in a crisis situationthan the crude mortality rate becauseyoung children are more susceptible tohealth and nutrition insults than olderchildren and adults.

Cause-specific mortality rateThe cause-specific mortality rate mea-sures the rate of death due to a specificcause (i.e., malnutrition-related deaths,violence related deaths, etc.), and itincludes only those deaths attributed tothe cause of interest in the numerator.The population denominator may inclu-de the entire population, or if the causeof death occurs predominantly in a sub-group of the population, it may includeonly that subgroup. For example, if 28people in a population of 78,904 diefrom tuberculosis during the course of ayear, the cause-specific death rate fortuberculosis during this time period

would be 35.5 deaths from tuberculosisper 100,000 population per year.If data on the causes of deaths in apopulation are available, cause-specificmortality rates provide informationabout the most important causes ofdeath. Such information can be used todesign intervention programs addres-sing these causes. However, in manypopulations, such information is lackingand the causes of death must be obtai-ned from other sources (see below).

Because the number of deaths from a sin-gle cause is usually far lower than thenumber of deaths from all causes (mea-sured by the crude mortality rate), thedenominator of cause-specific death ratesis often expressed as per 100,000 popula-tion. This allows the actual rate to be anumber greater than one. In the exampleabove, the cause-specific death rate fortuberculosis could also be expressed as0.355 deaths from tuberculosis per 1,000per year, but that rate is less easily under-stood than the larger number of 35.5deaths per 100,000 per year.

As mentioned above, the denominator ofmortality rates for causes of death whichoccur predominantly or only in a certainpopulation group may include only thosemost susceptible to death from thatcause. For example, death rates for uteri-ne or ovarian cancer are almost alwaysexpressed as the number of deaths per100,000 adult women because deathsfrom these causes are confined towomen. Similarly, death rates for prosta-te cancer are calculated using a denomi-nator consisting of the number of adultmen in the population.

40

MORTALITY RATIOSSome mortality rates are not truly rates,even though they may be called a morta-lity "rate." These include rates whichuse live births as a denominator, such asthe UNICEF under-five mortality rate, theinfant mortality rate and the maternalmortality rate.

The under-five mortality rate, or U5MRas the term is most commonly used, isthe probability of dying before the age offive, usually expressed per 1,000 livebirths. The U5MR cannot be calculated

directly from data on births and deathsby age in a single year because thedeaths, for example, of four-year-oldsoccur to children born four to five yearsbefore the occurrence. U5MR can be cal-culated using life table methods, frombirth history data (the recording of thedate of each birth and the age of deathfor those that have died), or using indi-rect methods that are beyond the scope ofthis manual. This measure of childhoodmortality is used most frequently by UNICEFand others who conduct large healthassessment surveys.

Comparison of the two measures of under-5 mortality rates

Both U5MR and ASMR-U5 measure the mortality risk for children under 5 years of age,but the two indicators express the risk differently. In the case of U5MR, the risk is expressed as the cumulative probability of dying

before age 5 years in a hypothetical group of 1,000 births. In the case of ASMR-U5, the risk is expressed relative to the mid-interval

population as with the crude and other age-specific mortality rates.

Because U5MR expresses risk over 5 years, whereas ASMR-U5 expresses risk peryear, U5MR is often almost five times as large as ASMR-U5.

As each measure (U5MR and ASMR-U5) has its uses and its advocates, there is nooverwhelming reason to recommend one over the other; they are two different waysof expressing much the same data. Since relief agencies tend to be more familiar withage-specific mortality rates, which are derived in the same way as the CMR (countingdeaths during a period of time and dividing by a population denominator), that methodwill usually be the one used in emergency assessments. Moreover, the usual nutritionand mortality assessment survey done in emergency situations does not gather thedata necessary to calculate the U5MR; only the ASMR-U5 can be calculated.

On the other hand, U5MR is used by UNICEF and others when measuring child morta-lity in stable populations. It is presented as the measure of child mortality in manysummary publications, such as the State of the World's Children. To avoid confusion,any reporting on under-five mortality should specifically note whether it is calculatedas an age-specific mortality rate or as the probability of dying by the age of 5 years.For WFP purposes, the ASMR-U5 should be used.

Infant Mortality Rate (IMR)*The infant mortality rate (IMR) is the num-ber of deaths in children under 1 year ofage in a given period divided by the num-ber of live births in the same time period.The infant mortality rate allows assess-ment of the rate of death in the most vul-nerable age group - children less than 1year of age. It often rises earlier and fasterin the face of poor health and nutritionthan other mortality rates.

The infant mortality rate also is describedas a ratio since many of the children whodie and are recorded in the numeratorwere born before the beginning of the timeperiod, and thus their birth is not recordedin the denominator. If the time period is1996, for example, then a six-month-oldchild dying in March 1996 would havebeen born in 1995; her death would add tothe numerator but her birth would not beadded to the denominator. Likewise, achild born late in the year would still be atrisk of dying under the age of one formuch of the following year.

The difference between a ratio such asthe infant mortality rate and the usualmortality rate can also be seen anotherway. The infant mortality rate calculatesthe chances that a live-born infant willdie before his first birthday - it is a cumu-lative incidence. The denominator is thepopulation at the beginning of the timeinterval of interest. A normal mortality

rate gives the average risk of dying dur-ing the time interval for a person in apopulation and represents a true rate. Ituses as the denominator the averagepopulation during the time interval, ormid-interval population.

Maternal Mortality Rate (MMR) **The maternal mortality rate (MMR) usesas the numerator all deaths of pregnantwomen or pregnancy-related deaths with-in 42 days of the end of pregnancy. Thedenominator is live births. Maternaldeaths are often a relatively rare cause ofdeath and, as such, the rates should becalculated only for relatively large popula-tions (more than 1,000,000). Randomvariation in the maternal death rate calcu-lated in small populations with few birthsmay be misinterpreted as significanttrends, when they are not actually so.

The maternal mortality rate is critical todetermine the need for antenatal andobstetric services. Although the actualproportion of all deaths in a populationresulting from pregnancy-related causesis often small, the effects of a woman ofchildbearing age dying are often muchgreater for her family and the society thanthe deaths of others in the population.Therefore, in a certain country, if therewere 34,459 births in 2004 and 78 mater-nal deaths, the maternal mortality ratewould be 226.4 deaths per 100,000 livebirths.



41

2CHAPTER

*

**

CROSS-SECTIONAL SURVEYS FORRETROSPECTIVE MORTALITY RATESAs mentioned above, it is better tocount deaths prospectively so that themortality rates calculated representrecent events. However, sometimes thesystem to count deaths does not exist ordid not exist during an earlier timeperiod in which you want to measuremortality. In such cases, mortality canbe measured using surveys. Just as withnutritional status or other health outco-mes, mortality information can be col-lected from the randomly selected hou-seholds. The people living in these hou-seholds report the number of deathswhich have occurred in that householdduring a specified time period. Duringdata analysis, the information ondeaths for all the households includedin the survey sample is put together.The total number of deaths in all thehouseholds is counted and becomes thenumerator of the mortality rate. Thedenominator of the mortality rate is thetotal number of people in the all thehouseholds included in the survey sam-ple. The time period is the period in therecent past during which deaths areasked about. Hence, you do not need toknow the size of the entire populationsurveyed or a count of the total numberof deaths in that population in order tomeasure the mortality rate.

Determining the recall periodIn order to calculate a mortality ratefrom data obtained by a survey, onlydeaths which occurred in a definedperiod in the past, called the recallperiod, should be included. To improvethe accuracy of mortality estimates incross-sectional surveys, the beginningof the recall period should be a memo-rable date known to everyone in thepopulation. For example, the start of the

recall period may be a major holiday orfestival (Christmas, beginning ofRamadan, etc.), an election, an episodeof catastrophic weather or other remar-kable event. The end of the recallperiod is usually the date the interviewtakes place. You can then calculate thelength of the recall period by countingthe days between the holiday or otherevent marking the start of the recallperiod and the date of data collection.Of course, this is rarely a nice roundnumber of days, like 90 or 180 days.

To detect such deaths, survey workersask respondents living in randomlyselected households to tell them aboutdeaths which occurred during thisrecall period. The denominator of themortality rate, being the total numberof people living in selected households,can also be seen as the number of per-son-time units (i.e., the number of per-sons in the selected households timesthe number of time units in the recallperiod). For example, if a survey selec-ted 569 households in which 3,243 peo-ple lived and the recall period was 7.3months, the denominator would be23,674 person-months. The same num-ber of person-time units could be obtai-ned from a recall period of 3.65 monthsand a survey sample of 6,846 people.The denominator of the mortality ratecan therefore be increased either byincreasing the number of persons in thesurvey sample or by increasing thenumber of time units (that is, the lengthof the recall period).

42

Lengthening the recall period is one wayto minimize the sample size. However,as you lengthen the recall period, you areasking survey respondents to reportdeaths which occurred in the moredistant past; naturally,an individual'smemory may become less reliable overtime. This is especially true if, along withrecalling the death, you also ask respon-dents about the circumstances or causesof the death. Moreover, by lengtheningthe recall period, you will produce anestimate of the mortality rate for a longerperiod in the past, which may be lessrelevant to current needs than a morerecent mortality rate.

Thus the sample size for a mortality rateis the number of person-time units nee-ded to obtain the desired precisionaround the estimate of the mortality rate.In acute emergencies, the person-timeunit is usually person-days to express themortality rate in terms of the number ofdeaths per 10,000 population per day. Instable populations, the mortality ratemay be expressed as the number ofdeaths per 1,000 population per monthor per year. The procedure for calculatingthe minimum sample size to achieve a

certain precision for the mortality rate isdescribed in the chapter on surveys.

If the mortality rate calculated from sur-vey data includes all reported deathsduring the recall period, that rate is anaverage for the entire recall period. Italso may also be possible to recorddeaths as having occurred in specificparts of the recall period. For example, ifthe recall period is 9.5 months, one couldrecord the death as occurring in one ofthree intervals: 1-3 months, 4-6 months,or 7-9.5 months prior to the interviewdate. Because this increases the comple-xity of the mortality survey, it shouldonly be included if the additional infor-mation is useful and if the persons inter-viewed can reliably place deaths intothese shorter intervals. Most people tendto recall important or traumatic events ashaving occurred more recently thanactual fact, so care must be taken to besure responses are accurate. In practice,in populations where calendar time isnot closely followed and dates are notwell remembered, recalling exactly whendeaths occurred can be very difficult.Nonetheless, if a traumatic or importantdate can be identified within the recall



43

2CHAPTERAdvantages of a longer recall period

A smaller sample size (i.e., number ofhouseholds) is needed for the sameprecision, potentially saving resour-ces and time.

If deaths are recorded for specificparts of the recall period, it is possibleto look at sub-intervals of time (e.g.,before and after displacement) or toexamine monthly trends (dependingon sample size).

Disadvantages of a longer recall period

Mortality rate may be less relevant tocurrent needs than a mortality ratecalculated for a more recent timeperiod.

Additional information, such as causeof death, becomes increasingly unreliable as the recall period lengthens.

period, separate mortality rates can becalculated for both the period before andthe period after such a date. For exam-ple, some surveys done in displacedpopulations have asked if a specific

death in the household occurred beforeor after the household left its home villa-ge. Separate mortality rates could thenbe calculated for the pre-displacementand post-displacement periods.

44

How long should the recall period be?

There is no absolutely correct length for a recall period for surveys measuringmortality rates. The recall period should be based on the objectives of the surveyand the following factors:

Accuracy: the recall period should be short enough to allow accurate recall ofinformation about the death. For most purposes, a recall period greater than oneyear probably will result in less accuracy.

Statistical precision: the recall period should be sufficiently long to provide enoughperson-time units to obtain the desired precision around the estimate of the morta-lity rate. For sample sizes used in many surveys, such as 1,000 households, a recallperiod substantially less than 90 days produces relatively poor precision.

Recent changes in mortality rates: if mortality rates are changing rapidly, you maynot be interested in the average rate over the last year, but rather the average rateover the prior few months. The population should also have a relatively constantmortality rate during the recall period. This may have to be assumed if no informa-tion is available.

Seasonality in mortality: if you are trying to measure the impacts on mortality of fac-tors not determined by season, the recall period should be chosen to cover severalseasons so these effects can be mitigated.

Logistic considerations: longer recall periods reduce the number of householdswhich need to be included in the survey sample and therefore the time needed tocomplete the survey.

During the acute phase of an emergency, it may be advisable to use a short recallperiod, such as 1-3 months, because you may be most interested in the mortalityrate for a very recent time period. When measuring mortality in stable populationswith less fluctuation in the mortality rate, much longer recall periods (such as oneor more years) can be used.

Mortality interview To estimate a mortality rate from a sur-vey we need to know (a) the number ofpeople at risk, and (b) the length of timeover which they were at risk. However,the composition of some of the house-holds will have changed during the recallperiod (death, birth, migration into andout of the household). As a result, thenumber of people within each of thehouseholds may not have been constantduring the recall period.

Figure 2.2 demonstrates this concept gra-phically. Time moves from left to right,with the vertical line on the left being thebeginning of the recall period and thevertical line on the right being the end ofthe recall period - usually the date of datacollection at each household. The asses-sment of mortality will count only thosedeaths which occurred during the recallperiod (those deaths occurring betweenthe vertical lines). Each horizontal lineillustrates a household member. The topline shows a household member wholived in the household at the beginningof the recall period and still lived therewhen the survey team visited the house-hold. The other horizontal lines repre-sent other household members whoentered or left the household during the

recall period by various means, includingbirth and death. The dotted horizontallines represent household members whodied during the recall period; these per-sons would be counted in the numeratorof the mortality rate. The householdshown in Figure 2.2 had three membersat the beginning of the recall period andalso had three members at the end of therecall period; however, only one personwas in the household during the entireinterval. At one time in the middle of therecall period, the household had sixmembers.

Two main methods have been used tocount the number of people in a house-hold in order to calculate a denominatorfor mortality rates: the past householdcensus method and the current householdcensus method. For both methods, a hou-sehold census is taken, whereby a list ismade of all the people living in the house-hold. In the past household method, thecensus is done as of the beginning of therecall period. Interviewers might pose aquestion such as, At the time of {name ofholiday or event}, who lived in this hou-sehold? In the current household censusmethod, the census is done as of the timeof the interview; the question often posedis, Who lives in this household now?



45

2CHAPTER

Beginning ofrecall period

Time End of recall period(Usually when survey data collected)

HH member

Moved into HHduring recall period

Birth during recall period

Moved out of HHduring recall period

Death during recall period

Birth and death duringrecall period

Figure 2.2. Household members' experience during the recall period

In this manual we recommend a modifica-tion of the current household censusmethod. Essentially, a census is done atboth the end and the beginning of therecall period. In order to calculate thedenominator you need to: 1. Ask the household respondent to list all

the household members at the time ofthe survey (the end of the recall period).

2. Ask the household respondent if eachof these household members werepresent at the start of the recall period.

3. Add to the household list all the mem-bers of the household who were presentat the start of the recall period but arenot currently present in the household.

4. Ask the household respondent the cur-rent status of each of the members whowere in the household at the beginningof the recall period but are no longer inthe household. Status may include alivebut living elsewhere, dead or unknown.

5. Ask the household respondent if eachperson on the household list is youngeror older than 5 years of age. Thisallows calculation of an age-specificmortality rate for children under 5years of age.

6. Ask the household respondent if anybabies were born during the recall peri-od and where these newborns are now.

The interviewer also can ask for additionalinformation if other indices of mortalityare to be calculated:1. The age of each member. These data

confirm whether an individual is aboveor below 5 years of age and allow ademographic pyramid of the popula-tion to be constructed. In addition,other age-specific mortality rates couldbe calculated, such as those shownabove for Badghis Province.

2. The sex of each member. These dataallow calculation of separate mortalityrates for males and females.

3. The date of each death if mortalityrates are to be calculated for sub-periods within the recall period.

4. The cause of death if cause-specificmortality rates or proportional causesof death are to be calculated.

These data are collected on a form, usinga separate sheet for each household.Figure 2.3 shows an example of the form.

46



47

2CHAPTERFigure 2.3 Example of household mortality data collection form

1ID

1

2

3

4

5

6

7

8

9

10

11

12

13

14

15

2Sex

F

M

M

F

F

F

M

M

F

M

M

F

M

M

M

3Current

age (in years)

23

26

54

48

18

12

8

2 mos

4

1

3

29

33

8

31

Survey district: Ambo Village: Limbo Cluster number: 4

HH number: 23 Date: 12- Aug - 04 Team number: 2

4Present

now

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

NO

NO

NO

NO

5Present at

beginning of recallperiod

Yes

Yes

Yes

Yes

NO

Yes

Yes

NO

Yes

Yes

NO

Yes

Yes

Yes

Yes

6Current status(1=Alive;2=Dead;

3=Unknown)

1

1

1

1

1

1

1

1

1

1

1

2

1

1

3

Tally (if data on household members will not entired into the computer)

a. Number of HH members at end of recall period

b. Number of children < 5 years at end of recall period

c. Number of HH members at beginning of recall period

d. Number of children < 5 years at beginning of recall period

e. Total number of deaths

Numberabove

11

4

12

2

1

Data comefrom:

Column 4

Columns 3 & 4

Column 5

Columns 3 & 5

Column 6

Sometimes survey managers may be tem-pted to ask each household respondentonly the number of people in the house-hold rather than listing each householdmember. Although this may be faster, it isfar less accurate than asking the respon-dent to list all the household members. Wetherefore strongly recommend that thehousehold members are listed on a formsuch as that above.

The numerator of the crude mortality ratein the survey sample is simply the totalnumber of all deaths counted in the surveysample during the recall period - that is,the sum of all the numbers in row e forall the households selected for the surveysample. The population denominator ofthe mortality rate is the average of the totalpopulation in selected households at thebeginning of the recall period (the sum ofrow c for all households) and the totalpopulation in selected households at theend of the recall period (the sum of row

a for all households). This averagepopulation is then multiplied by thenumber of time units in the recall periodto derive the person-time denominatorfor the mortality rate. The time andpopulation constants are then applied toconvert the mortality rate to a standardform which can be compared to othermortality rates.

For under-five mortality, a comparableprocess is followed. The numerator of theunder-five mortality rate is the number ofdeaths in children under 5 years of agewhich occurred in selected households, orthe total of row f for all households. Thedenominator for this mortality rate is theaverage of the number of children under 5years of age at the beginning and end ofthe recall period - that is, the average ofrows d and b. This mid-interval popu-lation of children under 5 years of age ismultiplied by the number of time units inthe recall period, as described above.

48

What to do with people whose status in the household is reported as unknown?

If such persons represent relatively few people who are no longer in the households,they can be excluded from both the numerator and denominator.

If they represent a large proportion of household members who are no longer in the hou-sehold, two mortality rates can be calculated: one assuming that the unknown membersare alive and one assuming that the unknown members are dead. Some probing with mem-bers of the community may also give you an idea whether those whose status is unknownare more likely to be dead or alive but living elsewhere. For example, in a population wheremany young men fled the household to join a fighting group which has not lost many mem-bers in battle, heads of households may report their status as unknown, but they may belikely to be alive. In other situations where attackers take household members away andthere have been mass executions, such household members may be more likely to be dead.

In any situation where the number of persons with unknown status is larger than 10 per-cent of the number of deaths, calculate two mortality rates (one rate excluding unknownpersons from the numerator, and therefore assuming that they are still alive, and thesecond rate including unknown persons in the numerator and therefore assuming thatthey are dead).

CAUSE OF MORTALITY One of the survey objectives may be todetermine the major causes of death dur-ing the recall period. During a cross-sec-tional survey, this is done by asking ques-tions of a surviving household memberwho was present just before and duringthe death of their relative. The questionsare meant to elicit a description of thesigns and symptoms experienced by theperson who died in order to determinewhat illness caused the death. Thisprocess is notoriously difficult even withan extended interview of a closely relatedperson very soon after death. Lists of vali-dated questions exist; however, the inter-view is much too long for most emergencyassessments and requires highly skilledinterviewers. More abbreviated lists ofquestions have been used in some sur-veys, but such lists have not been proper-ly validated.

Nonetheless, some causes of death may bedistinct enough to diagnose with fewerquestions. For example, malnutrition maybe suspected as a cause of death if thepopulation being surveyed has experi-enced famine and food insecurity, if thesurviving relative reports lack of food inthe household and if the deceased personhad rapid weight loss in the few weeksbefore death. In addition, some diseasesare well known by mothers and others inthe society. Many cultures and languageshave specific terms for measles, neonataltetanus and other common illnesses withrelatively unique appearances.

In many situations of conflict, it may be ofinterest to determine if deaths have beencaused by war-related violence. Becausesuch violence is easily identified by laypeople, a short series of questions, such asthose below, can often accurately identifydeaths due to this cause.

Questions to detect war-related violence:

1) Did (the person) die from some sort ofinjury such as being assaulted, shot orstabbed, a car accident, a fall, drown-ing, poisoning, burn, bite or sting?If YES, go to next question. If NO, record death as not related toinjury or violence.

2) Was this injury caused by someonefighting the war such as from a bullet,bomb, mine, machete or assault? If NO, record non-war-related injury orviolence as cause of death. If YES, record war-related injury or vio-lence as cause of death.

War-related violence normally affectsonly particular areas within the wholesurvey area. As a result, if war-relatedviolence was a major cause of death dur-ing the recall period, the sample size mayhave to be substantially larger to meas-ure mortality rates with any precision(see chapter on surveys for more detailedexplanation). Nevertheless, determiningthe contribution of war-related violenceto overall mortality may be important inmany situations.

In general, because of the complexity ofdetermining the causes of deaths in cross-sectional surveys through interviews,alternate sources of data should be usedto determine the causes of death. Suchsources could be disease surveillance,death registration, clinic or hospital logbooks, or others. If other sources of dataon causes of deaths do not exist, suchinformation could be collected duringsurveys if the following conditions exist:

If local terms exist for causes of inter-est, and respondents can reliably iden-tify them.

If the causes of interest consist only ofviolence.



49

2CHAPTER

50

Example 2.3 Calculation of mortality rates from a cross-sectional survey

The data for a cluster sample survey were gathered between June 11 and 17. The sam-ple contained 387 households. At the end of Ramadan, the November 14, before the sur-vey, 2,818 people lived in the households selected for the survey sample, of which 535were children under 5 years of age. At the end of the recall period (i.e., at the time of sur-vey data collection), the population of these selected households was 2,827, of whom 578were children under 5 years of age. During the recall period, 44 people died; 31 of thesedeaths were children under 5 years of age. Seventeen adults had left the householdduring the recall period and were living elsewhere; 4 adults were no longer in the house-hold, and their current status was unknown.

The first step in calculating any mortality rate is to determine the length of the recallperiod. The mid-point of data collection fell on June 14; this date can be used as the endof the recall period. There are 213 days between November 14 and June 14. These 213days are equivalent to 7 months (November 14 - June 14).

Crude mortality rate (in deaths per 10,000 population per day)

The denominator equals the average population size:

(2,818 + 2,827) = 2,822.52

multiplied by the length of the recall period:

2,822.5 persons x 213 days = 601,192.5 person-days

The numerator equals the 44 deaths reported in all ages during the recall period. Therefore,the mortality rate equals:

44 deaths x 10,000 = 0.73 deaths per 10,000 population per day601,192.5 person-days

Crude mortality rate (in deaths per 1,000 population per month)

If the CMR is to be expressed as deaths per 1,000 per month, the recall period would be 7months. Therefore, the denominator would be:

2,822.5 persons x 7 months = 19,757.5 person-months

and the mortality rate would be:

44 deaths x 1,000 = 2.22 deaths per 1,000 population per month19,757.5 person-months



51

2CHAPTERAge-specific mortality rate for children under 5 years of age (in deaths per 10,000 population per day)

The denominator equals the average size of the population of children under 5 years of age:

(535 + 578) = 556.52

multiplied by the length of the recall period:

556.5 persons x 213 days = 118,534.5 person-days

The numerator equals the 31 deaths reported in children under 5 years of age during therecall period. Therefore, the mortality rate equals:

31 deaths x 10,000 = 2.62 deaths per 10,000 population per day118.534.5 person-days

Age-specific mortality rate for children under 5 years of age (in deaths per 1,000 population per month)

If the age-specific mortality rate for children under 5 years of age is to be expressed as deaths per 1,000 per month, the recall period would be 7 months. Therefore, the denominator would be:

556.5 persons x 7 months = 3895.5 person-months

and the mortality rate would be:

31 deaths x 1,000 = 7.96 deaths per 1,000 population per month3895.5 person-months

REFERENCES

Measuring Mortality, Nutritional Status and Food Security in Crisis Situations: SmartMethodology, Version 1, June 2005. UNICEF and USAID, New York and Washington, D.C.2005. Available at: http://www.smartindicators.org/.

Spiegel PB, Sheik M, Woodruff BA, Burnham G. The accuracy of mortality reporting indisplaced persons camps during the post-emergency phase. Disasters 2001;25:172-180.

Myatt M, Taylor A, Robinson WC. A method for measuring mortality rates using pre-vious birth history. Field Exchange 2002;14:13-15. Available at: http://www.ennonline.net/fex/17/index.html

Woodruff BA. A method for measuring mortality rates using previous birth history:a commentary. Field Exchange 2002;14:16. Available at: http://www.ennonline.net/fex/17/index.html

52

Documents

4)_Chapter_2_-_Defining_&_measuring_mortality_(pgs 33-52).pdf