Contents - Previous - Next


This is the old United Nations University website. Visit the new site at http://unu.edu


5. Measuring impact using clinical, morbidity, and mortality data


Lincoln Chen and Rachel Feilden

This chapter examines the use of clinical, morbidity, and mortality health indicators in the assessment of food and nutrition interventions. It does not attempt a comprehensive review but instead focuses on conceptual and pratical issues related to the evaluation of feeding programmes. The multiple determinants of child nutritional status are outlined, and their effect on clinical, morbidity, and mortality health indicators is discussed. The major section of the chapter then summarizes techniques of clinical, morbidity, and mortality measurement, followed by a discussion of practical considerations in their use.


Framework for analysis
Methods of measurement
Annex A. Field nutrition assessment form
Annex B. Xerophthalmia field survey forms
Annex C. Diarrhoea/growth study illness surveillance form
Annex D. Brief examination of child
Annex E. Birth report form
Annex F. Death report form
Annex G. Maternity history questionnaire
References
Bibliography


Framework for analysis


Many factors influence the nutritional status of a child: genetic endowment and maternal nutritional status, availability of food to the family, distribution of food within the family and individual nutrient intake, seasonal variation and catastrophic weather conditions, social services (e.g. health care) and environmental sanitation (e.g. sewerage, potable water), and nutrient wastage due to infectious disease, parasitic infestations, and other causes of morbidity. The task facing evaluators of nutrition programmes is to assess whether variations in observed indicators can be attributed to the programme or whether they might be caused by independent individual or community factors. The considerations raised in chapters 1 and 2 concerning the design of the evaluation and sampling strategy are centrally important in gathering data on clinical signs, morbidity, and mortality; not only are these indicators affected by exogenous factors (e.g. education of mother, birth order, availability of health services), but there is also a well recognized synergistic interaction between infection and nutritional status. Either alone or, more often, in combination, they can contribute substantially to mortality. Figure 5.1. is a simplified diagram of these interactions.

Measurement of nutritional status can be undertaken using food balance sheets, dietary surveys, anthropometry. and biochemical and clinical assessment. Annex A. includes a sample field nutrition assessment form. Infectious disease morbidity is primarily episodic, characterized epidemiologically by incidence, prevalence, severity, duration, and degree of incapacitation. Mortality constitutes the ultimate negative consequence of ill health.

Figure 5.1. (see FIG. 5.1. Framework for Energy-Protein Malnutrition) underscores several issues crucial to the selection of an indicator and its measurement in evaluation. First, clinical, morbidity, and mortality indicators are only three of many potential indicators (others are discussed in other chapters). Second, some indicators reflect highly specific problems (e.g. xerophthalmia from vitamin A deficiency), while others are much less specific (e.g. mortality). Third, because indicators reflect many aspects of health, it is often difficult to select which indicators to use in an evaluation. Finally, the chosen indicators should relate to the presumed input of the nutrition intervention. Measurement of pretibial oedema for protein deficiency would be of little use if proteins are not a major programme input, or if baseline surveys indicate extremely low prevalence levels of oedema prior to the beginning of the programme.

The interrelation between morbidity and mortality means that single indicators rarely provide a comprehensive picture. The relationship between clinical, morbidity, and mortality indicators is shown in figure 5.2. (see FIG. 5.2. Relationship of Nutritional Status Infections and Mortality ). The prevalence of clinical disease or morbidity in a population at a point in time is the ratio of those with the disease (or symptoms) to the entire population. The problem here is that many-often most-of the diseased are asymptomatic; symptoms are not observed or not felt or not reported. Thus prevalence figures, even if valid for symptomatic illness, greatly underestimate disease prevalence. The use of objective clinical signs (i.e., observed indicators of morbidity) rarely improves this situation, since these indicators are usually detected only if the disease is severe or advanced; moreover, signs may be present in either symptomatic or asymptomatic individuals.

The rate of illness (R1) and the rate of recovery (R2) are incidence figures. So, too, is the number of deaths that occur in the morbid population over a specified time period (R3). These rates underscore the necessity of an understanding, or hypothesis, with regard to the mode of action of specific nutritional interventions. It is possible, for example, for disease prevalence to increase rather than decline with successful intervention if the health/nutritional technology primarily affects the death rate among the ill (R3), or if the intervention itself improves the accuracy of detection. It may be anticipated that preventive measures that reduce disease incidence (R1) will reduce disease prevalence and disease-related mortality. Even in this instance, however, disease-related mortality (R3) could actually increase rather than decrease, because the prevention might be more successful for mild cases and less so for more severe cases among which higher mortality rates would occur.

In light of these considerations, it may not be surprising to find that the prevalence of a particular clinical sign or anthropometric indicator is not changed in the desired direction with apparently successful interventions. Mortality, by and large, is a useful measure, because it does respond to effective interventions. However, using mortality as an indicator requires large samples because death is a rare event and it may be influenced by many causes, only some of which are nutrition-related.

The remainder of this chapter presents specific methods for measuring clinical, morbidity, and mortality indicators, on the assumption that the evaluation design has taken into account identifiable independent variables or exogenous factors, both at the community level (e.g. availability of health services, schools, travel time to regional centre) and at the household level (e.g. sanitary arrangements, economic resources, education). This strategy permits the analyst to reduce the background noise introduced by confounding factors, even though the interactions between nutritional status, morbidity, and mortality cannot be disentangled.

Methods of measurement


Clinical Signs

Among clinical signs of deficiency, the most frequently used for field assessment of nutritional status are bilateral pedal oedema, xerophthalmia, and various signs of anaemia. Many other signs, such as changes of hair, skin, mucous membrane, and even behaviour, also reflect various nutritional deficiencies, but the specificity, validity, and reliability of most of these signs are limited, thus compromising their usefulness.

Protein Deficiency

The most important nutritional problem addressed by virtually all food and nutrition interventions is proteincalorie malnutrition. As this deficiency state is invariably associated with a host of clinical disturbances, many signs are associated with compromised protein-energy intake.

When the predominant deficiency is protein, kwashiorkor may develop, and the major sign of this protein deficiency state is body oedema. Since oedema of nutritional origin is diagnostic of kwashiorkor, an examination for oedema is vital for determining community rates of the disease. The oedema usually begins in the lower extremities, bilaterally, but it may also be present in the arms and face; in supine babies it may be found in the sacral region. The examination for pretibial oedema is simple and straightforward: with the subject child in a sitting position, the examiner applies firm thumb pressure to the lower anterior surface of both legs for three seconds; identation that remains after the pressure is removed indicates oedema. Because of excess fluid, the child with kwashiorkor may have a normal body weight; therefore, anthropometry is not always specific in identifying or gauging the severity of this type of malnutrition.

Not all observed oedema will be of nutritional origin; but, when the oedema is associated with low weight for height, nutritional deficiency is likely. The simplest means of differential diagnosis for individuals who can be followed is a therapeutic trial by feeding. Those children who fail to respond to supplemental feeding should then be screened for oedema caused by cardiac or renal defects. A nutrition programme that includes dietary supplements provides an ideal opportunity to perform screening for oedema of nutritional origin.

Vitamin A Deficiency

Vitamin A deficiency, if severe and chronic, may result in ocular damage, which may range from barely symptomatic night-blindness to bilateral blindness due to keratomalacia. Field survey forms for the assesment of xerophthalmia, field-tested by the World Health Organization, are shown in Annex B, which also includes a clinic-based case-reporting form. Ocular examinations for the diagnosis of xerophthalmia should be conducted by field staff trained both to recognize the condition and to follow standardized procedures (preferably ophthalmologists). When field staff are not qualified or sufficiently trained to perform standardized ocular examinations, questionnaires and household surveys using interview techniques may also be used. These surveys may contain questions such as "Does your child have trouble seeing at dusk?" or they may use colloquial words for night-blindness (e.g. "chicken blindness" in Zaire). In any case, the questions should be phrased in a manner adapted to local language and situations. Because of variation between observers, it is desirable to do studies calibrating the reliability and validity of data gathered using such questionnaires against data from ocular examinations before undertaking large-scale field surveys.

Iron Deficiency

Despite the high prevalence of iron deficiency anaemia and the existence of multiple diagnostic procedures, this anaemia is difficult to detect by clinical examination alone. A host of laboratory and field-adapted techniques have been developed for standardized field assessment of anaemia. However, it may be due to multiple causes, only one of which is iron deficiency. Moreover, observer variability in the simple examination for pallor of the conjunctive, fingernails, and mucous membranes is high. Often, too, the prevalence of anaemia in a population is so high (virtually 100 per cent) that population-based clinical examinations yield little useful information.

 

Morbidity

Morbidity or illness may be either chronic or acute. Chronic conditions such as lameness, blindness, and other physical disabilities may provide clues to earlier disease processes, such as poliomyelitis. Most acute illness among children is of either infectious or traumatic origin, although in some regions chronic disease (malaria, infestation with helminths, schistosomiasis, trypanosomiasis, filariasis, onchocerciasis, etc.), often with acute episodes, is the cause of a large proportion of morbidity. Population-based morbidity surveys are frequently interested mainly in illnesses of infectious origin, because these are more amenable to limited programme interventions (such as immunization) and do not require extensive community efforts (such as eradicating disease vectors).

It is important to underscore the multiple epidemiologic characteristics of infectious morbidity. The concepts of incidence, prevalence, duration, severity, and case-fatality all differ from each other, and nutritional interventions may affect each differently, at times even paradoxically. It is possible, for example, for supplemental feeding to improve the nutritional status of a child and thus reduce the duration and casefatality of diarrhoea but not the incidence. The fact that duration is reduced while incidence remains constant would suggest a corresponding decline of disease prevalence. Many field surveys of morbidity fail to articulate explicitly the precise epidemiologic characteristic of interest and often measure a characteristic unrelated to the presumed impact of the nutrition intervention.

The criteria by which a "case" of infectious morbidity is defined should be clearly delineated. Is morbidity defined by a respondent, a physician's probing or examination, biomedical tests, bacteriological culture, or other supportive evidence, such as X-rays? Whenever possible, the reliability and validity of the criteria employed should be assessed against other diagnostic methods to gauge the relative usefulness of a particular technique. Irrespective of the method employed, it is crucial that before/after or intervention/control comparisons employ the same criteria of delineating a case of illness in both assessments.

Morbidity data can be collected by village workers noting easily-recognized infectious disease categories such as respiratory, gastrointestinal (diarrhoea), cutaneous (pyrodermus), conjunctivitis and the common communicable diseases of childhood (measles, chicken pox and whooping cough). Another system that records only symptoms is simple to interpret.

The location where morbidity is detected is also important. Case reporting from a hospital is notoriously inadequate for depicting community morbidity rates, because both the sample of patients who turn up for treatment and the range of presenting conditions are biased. Visiting households and inquiring about illness or examining household members avoids these sources of bias but introduces others. If longitudinal data are collected, repeated contact with field workers introduces a source of bias into both the household members' answers and the conditions under investigation. For example, a field worker may offer advice on the treatment of a dehydrated child. If cross-sectional data are collected, the respondent may supply incomplete information, and the opportunity to check on inconsistencies is quite limited.

Nevertheless, the commonest method of determining morbidity in programmes without an extensive research component is through single-round field surveys, sometimes supported by laboratory diagnostic facilities. Many of the problems associated with this method of collecting data can be avoided by careful preparation.

Biases that may be introduced by the interviewer can be reduced through training and supervision of field staff. A well-designed questionnaire improves the quality of data collected. The survey instrument should contain direct and specific questions, such as "Does your child have diarrhoea?" rather than open-ended questions like "Has anyone been ill?" which are likely to elicit vague and incomplete responses. Even a specific term can be misleading or confusing, because some societies may perceive certain illnesses as "normal," or a specific illness may be perceived in many forms; diarrhoea is an example of these phenomena. An understanding of local perception of disease and an appropriate language to describe such conditions are fundamental to the formulation of an effective enquiry.

Another major source of bias in field surveys of morbidity is error in reporting. If child morbidity is of interest, the respondent should be a guardian. If so, who? The mother is usually the most suitable respondent, but problems may arise when the mother is absent and another caretaker is present. Memory bias, omissions, and frank errors of type of morbidity are all hazards to be minimized. Reporting errors are related to the time period referred to in the questions. Repeated experience has demonstrated that recall over the preceding month is not reliable; the period is simply too long for the recall to be valid. Reporting daily or every two days is best, but the incidence and prevalence of the condition sample size may be too low for such a brief period of coverage. Reporting weekly or every two weeks is feasible but should be pretested to determine validity and reliability. Finally, morbidity recall may be undertaken in a cross-sectional survey (prevalence) or may involve repeated household visits over time (prevalence plus indicence). In the former case, the time period of recall may cause errors; The latter case reduces such errors, but repeated measurement introduces the problem of changes in respondent recall induced by the survey itself.

One of the most important considerations of morbidity measurements is the determination by the investigator of whether individual data or group data are of primary interest in the evaluation. The tolerance limit of field enquiries involving recall is obviously greater if group averages are of primary interest. Morbidity data obtained through questionnaires usually show high rates of false-positive and false-negative responses when they are treated at the individual level, virtually irrespective of the time period covered.

Annex C contains a morbidity questionnaire employed in Bangladesh for following a longitudinal sample; field workers visited the household every other day. Validity was improved with a physician follow-up reviewing all reported cases of morbidity at least weekly. Annex D contains a form for recording the results of a brief physical exam, from a cross-sectional household questionnaire developed for use by illiterate field workers.

 

Mortality

Projects and programmes that aim to improve health and nutritional status frequently attempt to measure changes in the target population's experience of mortality.

Changes in mortality rates are most likely to be found in age-groups that have had "high" mortality rates-e.g., children under five and women of childbearing age. Reductions in mortality rates during the relatively short (three to five year) timespan of programme interventions have seldom been satisfactorily documented, even for high-risk groups.

Projects that have reported a measurable reduction have all included a broad range of health service interventions (e.g., immunization, oral rehydration therapy), and have contained a considerable research component that increased the amount of effort invested in organizing and delivering services and in collecting and analysing data (see, for example, Alderman et al. (1 ) on the Hanover Project, in Jamaica and Berggren et al. (2) on Deschapelles, Haiti). Apart from famine relief programmes, we have no evidence that nutrition programmes that are implemented without health services or community development (e.g., sanitation and water supply) will lead to reductions in mortality, even if the programmes are closely monitored and tightly targeted.

Nevertheless, there are persuasive reasons for documenting the level of mortality in a population covered by nutrition programmes. Information on the mortality rate can be used by programme directors when they decide whether scare aid resources should be continued, increased, or allocated elsewhere. Data on mortality levels for infants and children of weaning age can help explain anomalous results of anthropometric surveys; for example, the proportion of children classified as severely malnourished may increase because fewer of these children are dying of measles, diarrhoea, respiratory infections, and other causes associated with malnutrition.

The evaluator who wishes to measure mortality must choose from several evaluation designs, analytic techniques, and methods of data collection. These three components are interrelated. First, we briefly describe the feasible alternatives, and then suggest how the evaluator can assess which configuration is most appropriate for a specific application.

Evaluation Design

There are three major categories of evaluation design, distinguished by the extent to which data for evaluation are collected at the beginning of programme activities and whether data collection continues during the course of the programme.

  1. Baseline/follow-up design: In this case, data are collected at the beginning of the programme with the specific aim of documenting levels of mortality in the target population. After several years of intervention, a second set of data is collected and the two cross-sectional views are compared to assess whether mortality rates have changed.
  2. Longitudinal design: Data on all births and deaths in the area covered by the programme are recorded. These registers of vital events can then be used to calculate mortality rates for specified periods of programme activity.
  3. Single cross-section: No baseline data were collected, and reliable records of births and deaths do not exist; the evaluation must be based solely on information collected after the programme has been operating for some time.

Analytic Techniques

The main techniques for estimating mortality rates are referred to as "direct methods" and "indirect methods." Direct methods can be used on a variety of data sources. Indirect methods all use the same types of data but organize them in different configurations to suit the assumptions of the chosen variant on the basic demographic model.

1. Direct Methods: A mortality rate is calculated by dividing the number of deaths (d) during a period by the total number of people (P) exposed during the same period. The resulting proportion is usually multiplied by one thousand. If all deaths are divided by total population, the result is a crude death rate; age-specific death rates, especially for high-risk age and/or sex groups are more useful for analysing the effects of interventions. Large samples are needed to obtain accurate data on death, which is a relatively rare event (see the section "Sample Size", p. 107). The reference period is usually one year; the denominator should be taken as the number of people in the set who were alive at the middle of the reference period. This procedure does not work for children under one year old; instead, the infant mortality rate should be calculated as the number of children under one year old who died during the year, divided by the number of live births during the same period. This figure includes children who have not yet survived their first year, and is thus conceptually different from lifetable figures, which indicate the probality of dying before age I (q0), calculated from cohort data. The infant mortality rate can be converted to q0 using the appropriate infant mortality separation factor (3).

Sources of data for calculating mortality rates directly include registration records on births and deaths, single-round (cross-sectional) surveys, multi-round surveys, and information gathered using more complex techniques such as dual record systems.

2. Indirect Methods: Estimates of the probality of surviving to age 0, 1, etc. (q0, q1,...) are calculated from data on the number of children ever born (CEB) and the number of children surviving (CS) for women in the target population during their childbearing period to date. Nulliparous women are included in the survey since this technique uses information on the total number of women. The observations are categorized according to woman's age, or duration of marriage (time since first union), or parity, depending upon the socio-economic structure of the population and the specific demographic model being used. The raw data on CEB and CS are ambigous with respect to children's age at death, because the data are categorized by mother's age (or exposure to risk of pregnancy, or parity) at the time of the survey. The indirect method successfully removes this ambiguity if the following assumptions are met:

- accurate data on CEB and CS are available;
- fertility and mortality schedules are known and follow a pattern similar to those embedded in the model;
- fertility and mortality levels are stationary and have been constant for the last 15 to 20 years;
- mortality rates for children are homogenous with respect to woman's age (or duration of marriage, or parity.)

Indirect techniques were developed using data from East Africa; historical data from European censuses were used to test the accuracy and precision of the method, whose estimates of mortality rates were close to the rates calculated from registration records. Several variations on the basic model have been tested using data from developing countries; estimates of the probability of surviving to age two and older are quite satisfactory, if the appropriate age cohorts of mothers are used (e.g., 30-40 for estimating q5). However, infant mortality (q1) is not well estimated by indirect techniques; if all mothers are included, the estimates tend to be too low, due to memory errors. If analysis is confined to women in the youngest age groups (15-24), estimates tend to be too high, because of the higher mortality experience of children born to first-time or young mothers. Evaluators of nutrition programmes are typically working in areas where at least one of the model's underlying assumptions is violated:

- there is no reliable registration system for checking whether CEB or CS are subject to systematic underreporting or overreporting;
- the pattern of fertility and mortality schedules is not known (vis. Haiti, with late age of first union and discontinuous unions). Even if national schedules are known, these may not apply to the specific population targeted by the nutrition programme,
- fertility and mortality levels have been falling in developing countries for several years;
- children's risk of death varies according to mother's age. duration of marriage, parity, level of education and socio-economic class (several of these variables are collinear).

In theory, analytic techniques could be developed which converted information about precise dates of births of CEB, and precise dates of deaths of non-survivors into mortality rates. In practice, these techniques have not been pursued because mothers in the typical target populations have never yet supplied this information with the level of accuracy needed to calculate reliable estimates.

Attempts to circumvent both the real limitations of the data and the fundamental incompatibilities between the models and the intended application have led to the development of methodologies which are more suitable for limited surveys in developing countries. These approaches will be described after the following section on data collection.

Methods of Collecting Data

Methods of gathering data can be divided into three general types: continuous recording of specified events when they occur, for the entire population; one-time contacts with respondents; and regularly-scheduled interviews of selected households or individuals. The first is called registration of vital events, and the second and third use survey questionnaires. These methods overlap with evaluation design but are distinct from it.

1. Registration: Vital events (births, deaths, marriages) are recorded when they occur. To be useful for calculating mortality, registration data must be complete and accurate; registration rates must not be biased by sex or socio-economic group or geographic area. Fetal deaths. still births and live births which subsequently died must be uniformly classified and accurately recorded. If the community does not already keep records, then vital registration data will be more unreliable if births are not attended. prenatal care is not available. migration is high, significant dates (e.g., the President's inauguration) rather than actual dates are preferred, and sex preference exists. Vital events are typically not recorded with desired level of quality in the environments in which nutrition programmes will be evaluated. In addition to being incomplete and inaccurate, registration data tend to be biased (e.g., urban, upper classes are overrepresented), so are unsuitable for calculating mortality rates, especially for a special subgroup at which a nutrition programme is directed.

Special forms can be used to collect accurate data on vital events from the population of interest; examples of Birth Report and Death Report forms from ICDDR, B/Matlab are attached in Annexes E and F respectively. However, special efforts to register vital events in a community where such record-keeping is not the norm have indicated that in these circumstances, registration data may be less reliable than survey data, because births and deaths occuring outside the area covered by the programme were not recorded.

2. Single Contacts: Surveys may be designed so that the inteviewer asks the questions only once of each respondent. Single-round surveys have used two distinct approaches to obtaining information. They may refer to a strictly limited reference period (for example, between now and a well-recognized religious holiday 12 months ago); the simplest form of reference period questionnaire takes three to five minutes to administer. Alternatively, the questionnaire may cover each surveyed woman's complete pregnancy and birth history. The simplest interviewing protocol includes the following items, appropriately worded:

- Present age of mother
- Number of live births
- Number of children born alive who subsequently died.

To obtain as complete and accurate information as possible, and to check it for consistency, this basic set of questions can be expanded to include outcomes of all pregnancies, including abortions; dates of birth and sex of all live-born children; date and cause of death; and various cross-check questions such as how many sons live with you, live away from home; how many daughters, etc. Questions of this nature take 30 to 45 minutes to ask. The World Fertility Survey collects data on complete reproductive histories (4); their questions are included in Annex G.

3. Repeated Contacts: The method of data collection may entail returning to the same respondent after the first contact. In "multi-round surveys", respondents are interviewed one or more times after the initial interview, with an interval of 6 to 12 months. Inconsistencies between the information supplied during the first and subsequent contacts must be clarified by the interviewer, using extensive probing to collect information on all pregnancies, births and deaths in each household. This approach appears to rest on a belief that respondents can successfully conceal or withold information only once. The administrative complexities of returning to the same household and the time lag implied by the survey design make multi-round surveys expensive and relatively slow; they have been used to collect demographic data in some developing countries.

Repeated contacts may be scheduled much more frequently (weekly or daily) in an attempt to construct a complete picture of mortality. (Weekly visits to gather data were made as part of the activities of the International Centre for Diarrhoeal Research, Bangladesh. [51). Data on temporary migrants, infants who did not survive long enough to be considered household members, etc. are likely to be captured by this intensive approach. However, it is expensive in terms of personnel time and data processing, and is only suitable for research purposes. Repeated visits are also likely to change household behavior; for example, if the field worker gives advice on care of sick children, the mortality rate may change more than if less intensive fieldwork were practiced.

Note that the data collected by using frequent visits to each household may approach the level of completeness and accuracy of a well-functioning registration system. However, the methodology is quite different; in the former, the interviewer asks for information during a scheduled series of visits, whereas in the latter, the individual comes forward with information on specified events only when they occur.

It is important to recognize that the baseline/follow-up design for an evaluation usually does not involve a multi-round survey. The administrative complexities of identifying the first set of respondents after three to five years of programme activity make it almost impossible to use this design in populations where there is migration of even five per cent per year. Furthermore, the original cohort of preschool children, who are the principal targets of most programmes, would have passed through this age group by the time of the follow-up. Therefore the second survey typically takes a second random sample of households from the same geographic area or community.

Feasible and Effective Methodology

The preceding description of analytical techniques and methods of data collection includes the range of possible approaches. Serious constraints render several of these approaches unsuitable for use in most evaluations in developing countries, where one is interested in finding out about mortality in a specific age group, during the recent past... Registration systems are beyond the financial realm of most evaluations, as are multi-round surveys and repeated house calls on a weekly or fortnightly basis. Data collected using singleround surveys are subject to uncheckable memory errors. and to a common problem faced by data-gatherers in developing countries: people often have an inexact sense of time in that they tend not to count their birthdays and tend not to identify precisely when vital events took place.

One strategy to limit the seriousness of memory, errors is to restrict the respondent's attention to a relatively recent period. A hybrid methodology using data on a limited time period has been developed. and is appropriate for generating estimates of mortality rates when these rates may be changing. This method (6) uses information on the survivorship status of births occuring during a specified reference period (e.g., 12 months), and converts the proportion of children who died into an infant mortality rate using the infant mortality separation factor. (These multipliers are only published for a reference period of 12 months.)

The technique is more sensitive to recent births and infant deaths than a complete maternity history, and is therefore more likely to enable evaluators to identify a change in infant mortality during the course of an intervention.

Choice of Evaluation Design. Analytic Technique, and Method of Data Collection for Estimating Mortality Rates

It is impossible to draw up a single, universal methodology for evaluating nutrition programmes; the evaluator must carry out some preliminary investigations of the specific site, and must know the features of the programme in question, in order to determine which configuration of design, analytic methods, and type of data is likely to generate the most precise estimates of mortality rates. The following sequence is suggested as a means of clarifying some choices:

  1. Will evaluation start at the beginning of the programme?
  2. Are high-quality survey data, collected specifically from the programme's target population, already available?
  3. Is there a well-functioning vital registration system that covers the entire country (to capture migrants' births and deaths), and that has been operating since before the start of project activity?

If these three questions are answered in the negative, then the evaluation will not be able to generate reliable estimates of changes in mortality rates, but only levels of mortality. Figure 5.3. (see FIG. 5.3. Decision Sequence) summarizes the decision sequence.

Sample Size

The size of sample needed to generate estimates of death with an acceptable degree of precision varies inversely with the crude birth rate and the incidence of mortality. Extensive tables have been produced which calculate total sample sizes implied by various choices of level of precision (coefficient of variation), and for different levels of the crude birth rate and infant mortality rate, given that 40 households will be sampled in each primary, sampling unit (7).

If estimates are calculated from data on recent reference periods (e.g., the last 12 months), then larger sample sizes are needed than for estimates based on children ever born and children surviving. Doubling the reference period approximately halves the sample size needed. Table 5.1. shows the range of sample sizes (households) needed for a reference period of one year in a homogenous population (i.e., estimates for different educational or ethnic groups will not be calculated). with a coefficient of variation (CV) of 0.10. The figures are based on an intraclass correlation of 0.05. (Intraclass correlation indicates the degree to which enumeration units [e.g., births during the reference period] within the primary sampling unit are similar with respect to the population measure being estimated.)

TABLE 5.1. Sample Size (Number of Households) Required for Producing Estimates from Reference-Period Data Covering One Year, for Various Crude Birth Rates and Infant Mortality Rates, with Coefficients of Variation of .10 (and .30).

Crude Birth Rate Infant Mortality Rate
75 100 125 150
30 10,278

(1,142)

7,500

(833)

5,833

(648)

4,722

(525)

40 8,325

(925)

6,075

(675)

4,725

(525)

3,825

(425)

50 7,153

(795)

5,220

(580)

4,060

(451)

3,280

(365)

Figures in parentheses indicate sample sizes for a coefficient of variation of .30. It can be seen that if such an unacceptably low level of precision were tolerated. the sample sizes required would be far smaller.

If we assume that intraclass correlation approaches zero, sample size requirements fall and approach the level generated by less sophisticated techniques for estimating sample size, because these simpler techniques implicitly assume that there is no intraclass correlation. Empirical analysis of World Fertility Survey data indicate that this correlation coefficient is between 0.03 and 0.05, so the figures given in table 5.1. provide a more accurate indication of sample size than do simpler procedures.

Summary on Estimating Mortality

The preciseness of data, which is both demanded by most techniques for estimating mortality and implied by the figures they generate, is considerably beyond the level that can be attained in most developing countries. If an evaluation does not have access to accurate, complete longitudinal data on individuals in the area covered by the programme, then evaluators would do well to concentrate on improving the quality of data collected, by putting more research and care into design of questionnaires, colloquial translations, and training field personnel to probe neutrally for complete information, rather than expending their effort and resources on grooming dirty data on the computer at the other end of the evaluation process.


Annex A. field nutrition assessment form


Dietary History Form

A "qualitative" individual dietary history form is presented. A family-based form would be quite similar: enquire about foods prepared for the family (as opposed to merely those consumed by the child) and omit questions about breast-feeding.

Major categories of food, but only a few specific items, are indicated. The final choice of food items to be listed depends upon local circumstances. For example, wheat is a potentially important vehicle for vitamin A fortification in Indonesia: none is grown locally and all imported wheat is processed in three factories. An extensive list of wheat-based foodstuffs was therefore included in the Indonesian study. This would not be appropriate, however, where wheat is widely grown and processed at a myriad of village mills.

Qualitative Dietary History Form

Sample site       (1) (2) (3)
Head of family: Name   Family number (4) (5) (6)
Individual: Name   Number (7) (8)
Classification: abnormal   1  
  control   2  
  random subsample   4  
  Total     (9)
Items consumed by the child during the past two months
Left-hand column: The frequency with which items were consumed:
1 - several times a day. nearly every day
2 - once a day, nearly every day
3 - less than every day hut at least once a week
4 - loss than once a week, but at least once a month
5 - less than once e month
0 - never
Right-hand column: Source of items consumed:
1 - harvested by the family
2 - bought
3 = harvested and bought
0 - innapplicable (item not consumed)
Staples     Frequency Source
Rice     ———— ————
Casaava     ———— ————
etc.        

FIG 5.A.a. Qualitative Dietary History Form

      Frequency Source
Sources of retinol
Liver     ———— ————
Meal     ———— ————
Eggs     ———— ————
Fish     ———— ————
Fish liver oil     ———— ————
etc.     ———— ————
Sources of -carotenes
Amaranth     ———— ————
Cassava leaves     ———— ————
Drumstick leaves     ———— ————
Mango     ———— ————
Papaya     ———— ————
etc.     ———— ————
Potentially fortifiable items
Salt     ———— ————
Refined sugar     ———— ————
Monosodium glutamate     ———— ————
Cooking oils     ———— ————
Soy sauce     ———— ————
Powdered milk     ———— ————
etc.     ———— ————
If not consumed, reasons why:
1 - unavailable        
2 - too expensive        
3 - child doesn't like it        
4 - child too young        
6 - bad for the child        
6 - other        
Sources of retinol        
It considered "bad for the child", reason why:
If "other" explanation:
Sources of -carotene
If considered "bad for the child", reason why:
If "others" explanation:
Frequency breast-fed per day
1 = once 4 = 4 times      
2 = twice 6 = 5 or more times      
3 = 3 times 0 = never or no longer breast-fed      
Age of child when breast-feeding ceased:
0 = never bread-sod        
1 = loss than 1 month   5 = 1-2 years    
2 = 1-3 months   6 = more than 2 years    
3 = 3 - 6 months   9 = not applicable    
4 = 6-12 months   (still breast-led)    

FIG. 5.A.a. (cont).

A quantitative (24-hour recall) dietary history follows the same format as the qualitative forms, but it is the total amount of each food item consumed by the child during the past 24 hours (coded in appropriately graduated amounts), rather than the frequency, that is investigated.

As in the case of the clinical examination form, a full-scale nutritional survey would require more detailed information covering a more extensive list of items.


Contents - Previous - Next