6.3 Cross-sectional studies
Oxford Textbook of Public Health
J. H. Abramson
Prevalence and incidence
Rapid epidemiological assessment
Uses in community health care
Community education and community involvement
Evaluation of a community’s health care
Uses in clinical practice
Individual and family care
Community-oriented primary care
Studies yielding ‘new knowledge’
Studies of growth and development
Studies of aetiology
This chapter deals with prevalence and other cross-sectional studies, that is, with surveys of the situation existing at a given time (or during a given period) in a group or population or a set of groups or populations. These surveys may be concerned with:
the presence of disorders, such as diseases, disabilities, and symptoms of ill health
dimensions of positive health, such as physical fitness
other attributes relevant to health, such as blood pressure and body measurements
factors associated with health and disease, such as exposure to specific environmental factors, defined social and behavioural attributes (including health practices and attitudes to health and health services), and demographic characteristics; the correlates may be determinants, predictors, or effects of health and disease states.
Such a study may be descriptive, analytical, or both. At a descriptive level it yields information about a single variable (AIDS, haemoglobin concentration, capacity to work, cigarette smoking, and so on) or about each of a number of separate variables, in a total study population, or in specific population groups. At an analytical level, it provides information about the presence and strength of associations between variables, permitting the testing of hypotheses about such associations.
Most cross-sectional studies are individual based, that is, they seek information about the individuals in the group or sample studied. There are also group-based (‘ecological’) surveys, which seek information about groups or populations (Chapter 6.2), and analytical surveys that operate at more than one level; in a Mexican study, for example, an association between dengue infection (the presence of antibodies) and exposure to Aedes aegyptii mosquitoes was not found at an individual level, but it was found when villages were compared (Koopman and Longini 1994).
Cross-sectional studies may be contrasted with incidence and other ‘time-span’ studies that require information relating to two or more points of time. The latter studies, which are discussed in later chapters, measure changes in status (for example, disease onset, growth, changes in blood pressure) or examine associations between variables with a defined temporal relationship, for example, between childhood experiences and health in adulthood, or between treatment and subsequent survival. The difference between cross-sectional surveys and these studies is often likened to the difference between snapshots and motion pictures.
This distinction is not, however, a rigid one. Although the essential feature of cross-sectional surveys is that they collect information relating to a single specified time, they are often extended to include historical information that can be easily collected at the same time. This may lead to the demonstration of statistical associations with past experience, for example, relationships between herpes simplex type 2 infection and the number of previous sexual partners and receptive anal intercourse (van der Laar et al. 1998), and a negative association (in Barbados residents aged under 70) between lens opacities and the regular use of nutritional supplements for at least a year (Leske et al. 1997). Field investigations of epidemics (Chapter 6.4) typically combine a cross-sectional approach (case-finding and the investigation of environmental and other hazards) with the collection of historical information (about possible exposures to infection). Cross-sectional studies in which cases are compared with controls are considered elsewhere (Chapter 6.5).
Current status, and not only historical information, may be indicative of past experiences, permitting consideration of possible aetiological processes. In a study of behavioural problems in school, for example, the lead content of the schoolchildren’s milk teeth was used as an indicator of lead poisoning in early childhood (Needleman et al. 1979). Similarly, assays of iron, arsenic, zinc, and other trace elements in toenail clippings may be used as measures of the prior intake of these elements (Garland et al. 1996). But the temporal relationships of the variables studied in a cross-sectional study (which came first?) is usually uncertain, making causal inferences uncertain.
If cross-sectional studies are repeated, they may be used for the purpose of health surveillance to observe changes in the population’s health status and its determinants. If there are enough surveys extending over a long enough period, they may be used to reconstruct the lifetime experience of birth cohorts. An analysis of successive surveys of smoking habits in Norway, for example, demonstrated the differences between the smoking habits of people born in different periods and the changes that occurred in specific birth cohorts (Ronneberg et al. 1994). Such appraisals are generally based on a comparison of graphs based on a rearrangement of the age-specific findings of successive studies. Ages at the time of the study are converted to birth dates, and birth-cohort graphs are then constructed by bringing together the data for each birth cohort and plotting them against age; also, the data for each age group can be plotted against year of birth or year of death (MacMahon and Pugh 1970). Birth-cohort effects can also be investigated by the median polish procedure (Selvin 1996)—a very simple exploratory data analysis technique that shows whether age and time trends alone can explain the findings—as well as by more elaborate statistical procedures.
The uses of cross-sectional studies can be categorized as follows.
The findings may be used to promote the health of the specific group or population studied; that is, the study can be used as a tool in community health care.
The study may contribute to clinical care.
The study may provide ‘new knowledge’—generalizable inferences that can be applied beyond the specific group or population studied. This knowledge may relate, for example, to the aetiology of a disease or the value of a type of health care.
These uses are not mutually exclusive; a single study may fulfil more than one purpose.
This chapter briefly considers the terms prevalence and incidence, and then reviews the methods used in cross-sectional studies, paying special attention to rapid epidemiological appraisal (cluster surveys and rapid methods of data collection) and to statistical measures, including prevalence measures of various kinds. The next three sections give consideration to the uses listed above. The first of these sections, on uses in community health care, considers community diagnosis, surveillance, community education and community involvement, and the evaluation of a community’s health care. The subsection on community diagnosis deals with studies of health status, determinants of health and disease, associations between variables (including the measurement of impact, risk markers, and community syndromes), and the identification of groups requiring special care. The section on uses in clinical practice briefly describes applications in individual and family care and in community-oriented primary care. The section on studies yielding new knowledge reviews studies of growth and development, studies of aetiology, and programme trials.
Prevalence and incidence
Prevalence refers to the number of individuals who have a given disease or other defined attribute at a specific time, as opposed to incidence, which is based on a count of events. The event may be the onset of a new disease, death, and so on.
The prevalence of a disease in a population at any point in time depends on the prior incidence of new cases and on the average duration of the disease from onset to recovery or death. This relationship is shown diagrammatically in Fig. 1, in which the contents of the container represent prevalence, and the time spent in the container is the duration of the disease.
Fig. 1 Relationship between prevalence and incidence.
If incidence and the average duration have remained constant over a long period (a condition seldom encountered in real life), point prevalence (defined below) is the product of the incidence rate of new cases per time unit t and the average duration of the disease (mean t per case).
For a disease that runs an episodic course, the point prevalence of active disease is (under certain assumptions) the product of the incidence rate, the average duration of an episode, and the average number of episodes per case (Von Korff and Parker 1980). For formulae expressing the change in point prevalence during a specified period, see Rothman and Greenland (1998, pp. 43–5).
Like any other kind of study, a cross-sectional study can yield useful findings only if sound methods are used. At all stages—in the planning phase, during the collection of data, and when the data are processed and interpreted—there is a need for detailed attention to methods (Abramson and Abramson 1999) so as to minimize bias and ensure that the results will be as accurate as practical constraints permit. Simple rapid methods may be called for if resources are limited or speedy results are needed (see section on rapid epidemiologic assessment below).
Findings are obviously determined by the methods used, and different methods may yield very different findings. The prevalence of dementia in a large sample of elderly people in Canada, for example, varied from 3 to 29 per cent, depending on which commonly used set of diagnostic criteria was employed; only 1 per cent had dementia according to all six sets of criteria (Erkinjuntti et al. 1997). Similarly, the prevalence of benign prostatic hyperplasia in a community sample of men ranged, according to different definitions, from 4 to 19 per cent (Bosch et al. 1995). Guidelines for the critical appraisal of a prevalence study have been proposed by Loney et al. (1998); they suggest the following eight questions.
Are the study design and sampling method appropriate for the research question?
Is the sampling frame (from which subjects are selected) appropriate?
Is the sample size adequate?
Are objective, suitable and standard criteria used for measurement of the health outcome?
Is the health outcome measured in an unbiased fashion?
Is the response rate adequate? Are the refusers described?
Are the estimates of prevalence given with confidence intervals and in detail by subgroup, if appropriate?
Are the study subjects and the setting described in detail and similar to those of interest to you?
Cross-sectional studies may be performed in total target groups or populations, or in representative samples. Simple random sampling or systematic, stratified, or cluster sampling may be used.
Methods of collecting information can be broadly classified as follows.
Clinical examinations, special tests, and other observations.
Interviews and questionnaires. The subjects themselves may be questioned, or proxy respondents, for example, household informants may be used.
Clinical records and other documentary sources. Sources of information on the prevalence of diseases include hospital and other medical records, disease registers, records of routine examinations (in schools, prenatal clinics, army induction centres, health insurance schemes, and so on), and published statistics based on these or other records.
Disease prevalence may be studied in two stages, by using a screening test to identify people who are likely to have a given disease, and then subjecting them to more elaborate and specific tests.
Each method of data collection has its own advantages and limitations and carries its own possible biases. If information on the prevalence of a disease is obtained from hospitals, for example, people with mild disease are likely to be under-represented. The degree of bias may vary for different subgroups of the population as a result of variation in the accessibility or use of health services, or of differences between clinical services in their diagnostic and recording procedures. In a rural region in the United States, Anderson et al. (1988) found that 42 per cent of the cases of Parkinson’s disease found in a survey based on screening questions and subsequent neurological examinations had not been diagnosed previously, and would have been missed by a survey based on medical records. Had people in institutions been omitted from the survey, a quarter of the cases would have been overlooked. The associations observed in a study of hospital or clinic patients may differ from those in the general population, if admission rates to the study group are connected with the variables whose associations are studied (Berksonian bias).
Since any method of case-finding may miss cases, it is often recommended that prevalence surveys should use more than one method, and combine the findings. The cases can be cumulated, or the total (including cases not found by any method) can be estimated by the ‘capture–recapture’ and related techniques (McCarty et al. 1993). These are methods originally used in estimating animal populations, where they are based on marking and releasing a batch of captured animals, and then seeing how many are recaptured in the next batch of animals caught.
Applied to two independent case-finding procedures that identify A and B cases respectively, with C cases common to both procedures, one formula for the estimated ‘ascertainment-corrected’ total number of cases is as follows:
[(A + 1)(B + 1)/(C + 1)] – 1.
In a study of childhood diabetes in Madrid, 451 cases were identified—432 by one procedure and 138 by another, with 119 common to both. The estimated total by the above formula was 501, with a 95 per cent confidence interval of 451 to 552. In a town in Japan, where the prevalence of type 2 diabetes mellitus among people aged 50 to 69 was 8.8 per cent according to a cross-sectional survey and 7.1 per cent according to a diabetes registry, the prevalence using the capture–recapture technique was 13.1 per cent (Sekikawa et al. 1999).
The capture–recapture approach is based on assumptions that are not always met. Its limitations are listed by Papoz et al. (1996). In particular, case-finding procedures are usually not independent. If cases ascertained by one method are especially likely to be identified by another method also, the calculated total will be an underestimate; and if cases ascertained by one method are likely to be missed by another, the total will be an overestimate.
This bias can be largely controlled by log-linear modelling (the multiple recapture census). This is a more elaborate technique that takes account of the observed dependencies between procedures (Fienberg 1972; Bishop et al. 1975), and can be used if there are three or more case-finding methods. Frischer et al. (1991) used this technique to estimate the number of injecting drug users in Glasgow, where 2006 cases were ascertained from three sources. The computed ‘ascertainment-corrected’ total was 13 050, a number that the authors reduced to 9424 to compensate for possible false-positive reports (95 per cent confidence interval, 6964–11 884). No method of estimation provides a fully satisfactory answer to the problem of incomplete ascertainment (Fienberg 1972; Armstrong and Hayes 1992; Kiemeney et al. 1994). If, for example, a certain type of case is ‘uncatchable’—that is systematically missed by all procedures—no manipulation can estimate their number. If all cases have in fact been found, the computed total will be an overestimate.
In cross-sectional studies the common sources of selection bias (where the individuals for whom data are available are not representative of the target population) are failure to choose a representative sample and incomplete coverage of the sample or study group. Information bias (due to shortcomings in the gathering or handling of information) is often attributable to the lack of clear diagnostic criteria or other operational definitions, and to inconsistency in their application. In studies that set out to examine causal associations, bias is commonly caused by the respondents’ or investigators’ knowledge that there has been exposure to the putative cause, or that the putative effect is present. In a study of the association between coffee drinking and digestive symptoms, for example, a respondent who drinks much coffee and believes that this beverage causes digestive upsets may be more likely to recall and report symptoms, or an interviewer may tend to be more persistent when asking such a respondent about symptoms. Also, subjects who have symptoms and believe that these are caused by coffee may tend to provide a fuller (or exaggerated) account of their consumption of the beverage, and interviewers may tend to be especially persistent when questioning subjects whom they know to have symptoms. The subjects’ awareness of their symptoms may also have led them to avoid coffee. These and other kinds of bias can be minimized by suitable survey procedures.
Some biases can be avoided, measured, or corrected during the analysis. Others cannot, but must still be taken into account when inferences are drawn from the findings.
As in other epidemiological studies, causal effects can be inferred from the findings of cross-sectional studies only after careful consideration has been given to the possibility of fortuitous and artefactual associations and confounding effects (Susser 1973; Rothman 1988; Rothman and Greenland 1998; Abramson and Abramson 2001). For example, associations between moderate or vigorous sports activities and low levels of blood pressure and other cardiovascular risk factors, observed in three cross-sectional surveys in Germany, could be taken as evidence for the preventive role of physical activity only after the effects of age, social class, body mass index, overall health, treatment for hypertension, activity at work, and other variables had been taken into account in the analysis (Helmert et al. 1994).
Rapid epidemiological assessment
Simple, undemanding, and inexpensive methods can often supply information that will adequately meet a study’s purpose. Such methods are particularly useful if financial, human, and other resources are limited, as in developing countries and in many studies (in both developed and developing countries) carried out by practitioners in the context of community health services. They have special relevance in situations where it is important to obtain real-time results as a basis for programme decisions, and may be essential in mass emergencies, where health needs should be appraised within a day or two (Guha-Sapir 1991). Extensive surveys have been performed in 10 days, including the publication of lengthy reports (Frerichs and Tar 1989; Materia et al. 1995).
The evolution and use of rapid assessment methods are reviewed by Smith (1989), who points out that the introduction of these methods was a major factor responsible for the worldwide eradication of smallpox.
Simple methods will generally provide information that is less detailed and less accurate than would be provided by more elaborate methods. This need not matter, provided that the information meets the study’s purposes, particularly if the alternative is no information at all. Care should be taken to avoid unnecessary inaccuracy, for example by giving due attention to the training of data collectors and by checking completed questionnaires and the coding of data.
Aspects of rapid survey methodology that are relevant to cross-sectional studies—cluster surveys and simplified methods of data collection—are discussed below. In addition, the appropriate use of computers can greatly speed up the performance and analysis of surveys, both in developed countries, where random-digit dialling and computer-assisted telephone interviewing may be useful techniques, and in developing countries, where changes in the cost and portability of computers have made them increasingly useful.
Uses of microcomputers are described in Chapter 6.15.
Two-stage cluster sampling has been advocated by the World Health Organization (WHO) Expanded Programme on Immunization since 1978 as a rapid, cheap, and accurate basis for surveys of immunization coverage (Lemeshow and Robinson 1985). It is based on the selection of a simple random sample not of individual subjects, but of groups or clusters of individuals. The method has been used for studies of specific diseases, service coverage, health service needs, and other topics. It can provide reasonably representative data, and has the advantages that it does not require a detailed sampling frame, it uses clusters of subjects who live close to one another, and it is so simple that subjects can be easily selected in the field with minimal technical support. Since clusters may contain people with similar characteristics, a representative sample requires a reasonably large number of clusters.
Cluster surveys following the Expanded Programme on Immunization pattern are based on the random selection of 30 or more villages, towns, sectors of cities, and so on, with the probability of inclusion being proportional to the cluster’s size (it is possible for a large cluster to be selected more than once). The selection is easily done (Bennett et al. 1991; Abramson and Abramson 1999) if reasonably accurate population estimates are available. Strictly speaking, the subjects in each cluster should be chosen randomly, using a census. But, for simplicity, Expanded Programme on Immunization surveys use a modified method. One household in each cluster is randomly selected, preferably using a list or map; alternatively, the investigator can start at a central point, count the number of households (H) in a randomly-chosen direction (a spinning top or pencil can be used) from the central point to the border of the area, and then randomly choose a number from 1 to H to identify a household for selection. A preset number of subjects (for example, children of a defined age) is then studied in each cluster, starting with the selected household and moving on to the next one (the residence whose front door is closest) or perhaps the fifth-nearest household, until the quota is filled.
The number required in each cluster is decided by applying the usual method of estimating sample size for simple random sampling and then multiplying the result by the ‘design effect’ and dividing it by the chosen number of clusters. The design effect, which expresses the difference in precision between a cluster sample and a simple random sample of the same size, may be estimated from previous surveys (Bennett et al. 1991). A typical Expanded Programme on Immunization cluster survey that aims to provide an immunization coverage rate that is within 10 percentage points of the true rate, with 95 per cent confidence, might use seven non-randomly chosen subjects in each of 30 clusters. Computer simulation has shown that although this method yields overall results that are more biased and variable than those based on simple random sampling, it meets the above requirement in over 95 per cent of replications (Henderson and Sundaresan 1982; Lemeshow and Robinson 1985; Lemeshow et al. 1985). The results in specific clusters or in subsets of clusters cannot be relied on; analyses of a particular cluster are warranted only if the sample is large and not based on non-random selection. Comparisons of subsets of clusters may be feasible, as in an Indian survey that showed a higher level of immunization coverage in urban than in periurban areas and an especially low level in rural areas (Balraj et al. 1993). In a study of disease prevalence, the extent to which the disease is clustered may be of interest; a survey in Tanzania, for example, revealed clustering of trachoma within neighbourhoods in villages, not explicable by known risk factors (West et al. 1991). Simple methods of analysis are described by Bennett et al. (1991) and Frerichs (1989).
Computer simulations suggest that, in view of its ease and cheapness, cluster sampling with non-random selection in the second stage may be a reasonable choice even in analytical studies that use the ratio of two prevalence rates (for example, in people exposed and unexposed to a risk factor) to measure the strength of an association. But if the rates are large and their ratio is high, simple random sampling is much more successful in yielding a ratio within 0.1 of the true value and a confidence interval that includes the true value (Harris and Lemeshow 1991).
Approximate incidence rates can sometimes be derived from the prevalence findings. For example, lame children can be identified in a cluster survey, and the prevalence of lameness attributable to poliomyelitis can then be calculated, using criteria such as the presence of flaccid paralysis and intact sensation, with a history of an acute onset. Correction factors can then be applied to allow for the omission of upper limb paralysis, lethal cases, and complete recovery, and approximate poliomyelitis incidence rates can be computed (LaForce et al. 1980). Such a survey in Nigeria led to the estimate of at least 33 300 paralytic poliomyelitis cases annually in the period before an immunization programme was introduced (Babaniyi and Parakoyi 1991). In Burkina Faso, a cluster survey of this sort indicated a significant decrease in incidence, possibly because of immunization (Schwoebel et al. 1992). Incidence rates may also be estimated by obtaining a history of diseases or fatal conditions (for example, neonatal tetanus) that occurred in the households included in the clusters (Rothenberg 1985).
The main deficiency of the Expanded Programme on Immunization design arises from its non-use of random methods when selecting subjects, and its use of quota sampling for this purpose; results may be misleading if subjects with similar characteristics tend to live close to one another. Ways of mitigating the deficiencies include the use of the fifth-nearest instead of the nearest household, and splitting the community into quadrants and selecting a quarter of each cluster from each quadrant, starting at the centre point of the quadrant. A suggested ‘not quite as quick but much cleaner alternative’ to the Expanded Programme on Immunization design, aimed at retaining the advantages of ease and cheapness but making the procedure more rigorous and appropriate even for surveys that make multiple measurements, is based on the preselection of a ‘target segment size’—the number of households to be surveyed for each cluster—instead of fixing the number of subjects needed in each cluster (Turner et al. 1996). Communities are selected randomly (as in the basic Expanded Programme on Immunization method) and then divided into equal segments, each containing approximately the required number of households; this requires a rough sketch map showing the households in the community. A segment of the community is then chosen at random, and all eligible individuals in all the households in the chosen segment are included in the sample. This method ensures that all households in the study population have approximately the same probability of being selected.
Simplified methods of data collection
Data collection may be simplified in various ways. The most obvious is to restrict the variables to those that are essential to meet the study’s purposes (resulting in very short questionnaires or examinations) and to choose sources that are easily accessible—available records may, despite their deficiencies, contain enough information to obviate the need for a more demanding survey. Purposive sampling will often satisfy a study’s needs. Patients attending a health facility, for example, may be deemed sufficiently representative of the total community to warrant their use (with reservations) as a study sample, and tests and interviews of antenatal clinic attenders and men attending outpatient clinics can provide useful information on the prevalence of sexually transmitted diseases in a refugee camp (Mayaud et al. 1992).
If available, simpler procedures can be chosen rather than more accurate but elaborate ones: for example, a rapid urine test for use in surveys of iodine deficiency (Rendl et al. 1998) or a simple test card for identifying people with low vision (Keeffe et al. 1996). Household food inventories (Patterson et al. 1997) or brief dietary questionnaires may supply enough information on dietary practices, rendering detailed dietary interviews unnecessary.
It may be decided to use simple proxy measures: for example, arm circumference or weight-for-height as easy and cheap indices of malnutrition in children (Velzeboer et al. 1983a, b), night blindness as a relatively easily measured surrogate for vitamin A deficiency (Sommer et al. 1980), or a characteristic depigmentation pattern (‘leopard skin’) as an index of the endemicity of onchocerciasis (Edungbola et al. 1987); or to appraise the burden of lymphatic filariasis by measuring the rate of infection of insect vectors (Pani et al. 1997) or by examining a small sample of men for hydroceles (Gyapong et al. 1996).
If the measures are simple, it is easy to train health workers or others in their use. Schoolteachers can measure weight and height with adequate precision, assistants can be taught simple cataract recognition (Venkataswamy et al. 1989), and traditional midwives have been taught to identify low birth weight babies by using a hand-held scale that shows a coloured signal if the weight is below 2.5 kg (Ritenbaugh et al. 1989). In Tanzania, a simple questionnaire on diseases and symptoms was administered by teachers to children in 245 schools; a comparison with urine tests showed that reports of haematuria or schistosomiasis had a high validity (Lengeler et al. 1991).
Cross-sectional methods—for example, for appraising child growth in a community—are obviously faster than longitudinal ones. Simple data on current infant feeding practices can, if appropriately analysed, rapidly provide a picture of the average duration of breast feeding and the age at introduction of supplements (Ferreira et al. 1996).
Where appropriate, rapid qualitative methods may be used. Qualitative (as opposed to quantitative) methods provide findings that are described in words rather than numbers. They are especially useful in the investigation of knowledge, attitudes, and practices—’beliefs and perceptions regarding health, the prevention and treatment of illness, and the utilization of traditional and biomedical health resources’ (Scrimshaw and Hurtado 1987). Qualitative methods can provide ‘culture specific maps [that] can help to improve the “fit” of programmes to people’. These maps show the presence of beliefs and behaviours, but not their numerical prevalence in the population (Scrimshaw and Hurtado 1987). A survey of patients who had a heart attack, for example, pinpointed the misconceptions (about heart attack symptoms) that contributed to delay in calling for medical help (Ruston et al. 1998), and a study of mothers in a population with a low breast feeding rate highlighted the role of seeing successful breast feeding by a relative or friend, rather than advice, as a determinant of the decision to breast feed (Hoddinott and Pill 1999). These methods have been termed ‘rapid ethnographic assessment’ (Smith 1989), ‘rapid assessment procedures’ (Scrimshaw and Hurtado 1987), and ‘social research methods’ (Smith and Morrow 1991).
Qualitative studies may be based on interviews and conversations with key informants and other members of the community, in which people can express their attitudes, perceptions, motivations, feelings, and behaviour; on observations in health-care facilities and the community at large; and on other methods (some of which require special training), such as focus group discussions, in which a small group of informants talk freely and spontaneously about themes considered important to the investigator (Khan et al. 1991), and the nominal group technique. Rapid methods of evaluating health care include surveys of people attending for care (clinic exit interviews) and checks on clinic facilities and supplies (Anker et al. 1993; WHO 1993). Guidelists for the collection of data on topics related to health and health care are provided by Scrimshaw and Hurtado (1987).
The nominal group technique, which was developed by Van de Ven and Delbecq (1972), is a useful method of obtaining a semi-quantitative picture of the views of a group of people: for example, professionals or laypeople who are invited to provide information on the problems of a specific community, or to suggest possible solutions (Abramson and Abramson 1999). The technique is so called because although the participants sit together, direct interaction is permitted only during specified phases of the process; hence during most phases this is a group ‘in name only’. The Delphi procedure (Linstone and Turoff 1975) is a much more elaborate method.
Methods of validating the results of qualitative studies include ‘triangulation’, that is, the use of more than one qualitative method, to ascertain their common conclusions. Three suggested questions for use in appraising a qualitative study of beliefs and practices (Mays and Pope 1996) are as follows.
How well do the conclusions explain why people behave in the way they do?
How comprehensible would this explanation be to a thoughtful participant in the setting?
How well does it cohere with what we already know?
Qualitative and quantitative approaches may be regarded as complementary (Kroeger 1983). In a study of the reasons for incomplete childhood immunization in Haiti, for example, ethnographic methods were used to identify barriers to the use of preventive services, and these were then measured in a quantitative survey (Coreil et al. 1989). Qualitative methods can also be used as a follow-up to a quantitative study, to explain and expand the findings.
The statistical measures used to summarize the findings of descriptive cross-sectional studies include means and standard deviations, medians, percentiles and other quantiles, measures of prevalence (see below), and other proportions. Ratios other than proportions are occasionally used: for example, the sex ratio (usually the male to female ratio) of people with a specific disease. Separate statistical measures may be provided for specific sex and age categories, ethnic groups, social classes, regions, and so on. Measures of association commonly used in analytical cross-sectional studies are described below.
If the study is based on a random sample, confidence intervals should be calculated in order to obtain an interval that has a high probability of containing the true value of the measure in the total target population. Confidence intervals are also often calculated even in the absence of a ‘chance’ process such as random sampling, to permit generalization to a broad ‘reference’ population, for example ‘the nation’s children’, but in this instance the procedure is open to criticism.
Measures of prevalence
A measure of prevalence expresses the relative frequency of a disease or other qualitative attribute in a group or population; it is a proportion. The convenient and commonly used term ‘prevalence rate’ is disparaged by many experts who prefer to confine the term ‘rate’ to measures of the rapidity of change, and therefore use ‘prevalence’ or ‘prevalence proportion’ rather than ‘prevalence rate’, claiming that the latter is an impossible concept (Elandt-Johnson 1975).
There are a number of prevalence measures. Used without qualification, prevalence usually refers to a point prevalence, that is, the prevalence at a specified point of time. The point prevalence of a disease per 1000 population, for example, is calculated by the formula
The numerator is the total number of people who have the disease at the stated time, irrespective of when the disease commenced. The denominator is the total population (actual or estimated) at that time, including affected and unaffected people. A multiplier of 1000 or any other convenient or conventional multiple of 10 is used in order to eliminate awkward decimals—6.5 per 100 000 is easier to comprehend than the equivalent 0.000065.
The point of time to which prevalence refers need not be a fixed calendar time. The reference may be to a fixed point in the experience of each individual, for example, birth, entry to a job or army service, immigration, death, date of diagnosis, or (in a prevalence survey where interviews or examinations are staggered over a period) the date of examination. In such instances the formula is
A measure expressing the frequency of a finding in an autopsy study is a point prevalence that refers to the time of death. This may be an indication of prevalence in the living, if the disorder does not affect the risk of dying or the probability of an autopsy, as in autopsy studies that showed that 45 per cent of young American soldiers killed in battle had coronary atherosclerotic lesions (MacNamara et al. 1971).
Like other measures, point prevalence may express the findings in a specific subgroup of the population; when so used, the numerator and denominator must both refer to the same population category. As an example, the sex- and age-specific point prevalence of a disease per 1000 men aged 45 to 64 years is calculated by the formula
Paradoxically, there are some point prevalences that can be accurately measured only by a longitudinal study. An example is the prevalence of congenital anomalies per 100 live births, which may be regarded as a point prevalence referring to the moment of birth. Many anomalies become manifest only weeks, months, or years after birth, so that reasonably full case-finding requires long-term follow-up.
Period prevalence refers to prevalence not at a single point in time but during a defined period (usually a specific year). Period prevalence (persons) represents the proportion of the population manifesting the disease at any time during the period. The formula is
The numerator is the number of people with the illness during the specified period, including those whose illness started earlier. The denominator is the average size of the total population during the specified period. It is often estimated by using the population at the middle of the period, or by averaging the size of the population at the beginning and end of the period. Other methods may be needed if the change in population size during the period was large and did not occur at an approximately even pace. It is usually more helpful to know the point prevalence at the beginning of the period and the incidence rate of new cases during the period, rather than the period prevalence.
There is also a type of period prevalence that refers not to a defined calendar period but to a defined period of the individual’s life, for example, pregnancy or childhood. A study of a sample of healthy pregnant women in an American city, for example, revealed that the prevalence of reported physical battering during the current pregnancy was 8 per cent (Helton et al. 1987) and a cross-sectional study of men in England indicated that 5.3 per cent had experienced sexual abuse during childhood (Coxell et al. 1999).
The numerator of the little-used period prevalence (spells) is the number of spells (episodes) of an illness observed during the specified period (including episodes that commenced before the start of the period); the same person may be ill more than once. The denominator is the population at risk. For a short-term disease, this measure is usually similar to the incidence rate (spells).
Lifetime prevalence is a period prevalence referring to the whole of the subject’s prior life. It differs from point prevalence only if the disorder is one that does not always persist. It refers to the presence of the disorder or of a scar, antibodies, historical or other evidence that the disorder was present in the past. The formula is
This measure is usually useful only if it refers to a specific age, and if valid information on prior occurrence is available. As an example, a cross-sectional study in Jerusalem revealed that the point prevalence of inguinal hernia among men aged 65 to 74 years was 30 per cent, whereas the lifetime prevalence (including men with scars of hernia repair operations) was 40 per cent (Abramson et al. 1978). It could be inferred that in this cohort the risk of developing a hernia, among men surviving to the age of 65 to 74 years, was 40 per cent. Such information would be of little value if the disorder were one with an important impact on mortality, such as cancer.
The lifetime prevalence of a disorder among the blood relatives of an index case may be used as a measure of familial risk, especially in genetic studies.
Measures of association
In analytical cross-sectional studies, the most commonly used measures of the association between two variables are odds ratios, rate ratios (for example, ratios of disease prevalences or of prevalences of exposure to a supposed causal factor) and rate differences (for example, differences between disease prevalences or prevalences of exposure). These measures are defined in Table 1, illustrated by fictional data on the association between exposure to fumes and headaches, based on a cross-sectional study in Denmark that found that reported exposure to fumes or chemicals at work was associated with the prevalence of reported headaches during a 1-year period. Table 1 also includes measures of the impact of exposure on the prevalence of headaches (assuming a causal association). Associations may also be measured by correlation and regression coefficients, differences between means, and other statistics.
Table 1 Measures of association and impact (study of a population) using fictional data on headaches and exposure to fumes
A prevalence ratio of 1.8 means that the prevalence of the disease in exposed people is 1.8 times as high as in unexposed people. An exposure ratio of 1.74 means that exposure is 1.74 times as prevalent among people with the disease as it is among those free of it.
An odds ratio is the ratio of one odds to another. An odds is the ratio of the probability that something is so or will occur, to the probability that it is not so or will not occur; in Table 1, a/b is an odds in favour of the presence of headaches. The odds ratio (ad/bc) of 1.89 in Table 1 can be regarded as the disease odds ratio—the odds in favour of headaches are 1.89 times as high among people exposed to fumes (odds = a/b) as among those not exposed (odds = c/d). It can also be regarded as the exposure odds ratio—the odds in favour of exposure are 1.89 times as high among people with headaches (a/c) as among people free of headaches (b/d).
Odds ratios may be difficult to understand, and are easy to misinterpret. A survey of treatment decisions made by doctors presented with feigned patients with identical clinical pictures revealed that the odds ratio for referral for cardiac catheterization was 0.6 for black people, in comparison with white people (Schulman et al. 1999), and this was widely reported in the media in such terms as ‘Doctors are only 60 per cent as likely to order cardiac catheterization for blacks as for whites’. The referral rates were 84.7 per cent and 90.6 per cent respectively, and the rate ratio corresponding to the odds ratio of 0.6 was 0.93 (Schwartz et al. 1999).
Opinions of the utility of odds ratios vary widely (Greenland 1987; Kahn and Sempos 1989; Lee 1994; Osborn and Carraruzza 1995; Selvin 1996; Zocchetti et al. 1997). Their useful features (using the numbers in Table 1 where possible) include the following.
Use of odds ratios facilitates comparisons of results from different kinds of study. Identical odds ratios can be expected in a study of a total population (or representative sample), a comparison of representative samples of people exposed and not exposed to fumes (yielding the disease odds ratio), and a comparison of representative samples of people with and without headaches (yielding the exposure odds ratio). The sampling fractions do not affect the value of the odds ratio; for example, the ratio remains 1.89 if the numbers in the ‘factor absent’ group are reduced to one-tenth (c = 5, d = 85). Under certain conditions, odds ratios from cross-sectional studies can be compared with odds ratios from time-span studies, and odds ratios derived from studies based on analyses of 2 × 2 tables can be compared with those derived from logistical regression analysis, whose regression coefficients are the natural logs of odds ratios.
The odds ratio for freedom from the disease is the reciprocal of the disease odds ratio. The disease odds ratio in Table 1 is 1.89, and the ratio of the odds in favour of freedom from headaches in exposed people (90/10) to the corresponding odds in unexposed people (850/50) is 0.53, which is 1/1.89. This does not hold true for the prevalence ratio; again comparing exposed and unexposed people, the disease prevalence ratio is 1.8, but the ratio of the freedom-from-headaches prevalences, that is the ratio of 90/100 to 850/900, is 0.95—which gives the impression of very little difference between exposed and unexposed people.
Observations in different population groups or strata are often combined by the Mantel–Haenszel procedure, multiple logistic regression analysis, or other techniques, on the assumption that the association has the same strength in each group. There is no problem with this concept if odds ratios are used to measure the association. But the concept of a common value for the prevalence ratio may be untenable if rates are high (Kahn and Sempos 1989). The prevalence ratio in Table 1, for example, is 1.8, but in a different stratum where the prevalence in the unexposed was 70 per cent, a prevalence ratio of 1.8 would be impossible—the highest possible ratio would be 100 per cent/70 per cent, or 1.4.
In aetiological studies of disease, the measure of interest is the ratio of the incidence in persons exposed and unexposed to a putative causal factor, and the odds ratio can sometimes serve as a proxy for this. The prevalence ratio can serve as an indicator of the risk ratio (cumulative incidence-rate ratio) in cross-sectional studies if the risk factor is no longer active: for example, a study of non-lethal birth defects in relation to some prenatal factor, or other studies of diseases with short and well-defined periods of risk, for example, of an epidemic of diarrhoeal illness after a social gathering (Kleinbaum et al. 1982; Rothman 1986).
If the prevalence ratio is not available—for example, if the study compares samples of people with and without the disease—an odds ratio is a good estimator of the prevalence ratio, provided that prevalence is low, and can therefore be used as an estimator of the risk ratio. Selvin (1996) suggests that ‘low’ here means a rate of under 10 per cent in each of the groups that are compared. The odds ratio and the prevalence ratio in Table 1 are fairly close (1.89 and 1.8); they would be closer (the odds ratio would also be 1.8) if the prevalence of headaches was only 1.8 per cent in exposed and 1 per cent in unexposed people.
The odds ratio can sometimes be interpreted as an incidence ratio even if the disease is not rare (Breslow 1982; Miettinen 1985; Rothman 1986; Pearce 1993). This applies to studies in which cases of a disease with a long risk period (like most chronic diseases) are compared with disease-free controls who are representative of the population from which the cases developed, and who at the time that they are studied can be regarded as possible future cases. In such a cross-sectional study the odds ratio is equivalent to the ratio of person-time incidence rates, provided that exposure can be assumed to precede the onset of the disease and not to affect the duration of the disease, and that the disease does not affect exposure status; this equivalence does not apply within narrow age categories, or if aetiological factors have changed in the course of time. The odds ratio can also be interpreted as the ratio of person-time incidence rates in a study comparing cases that develop during a given time with controls selected at the same times as the cases, and as a risk ratio if controls are sampled from the whole population at the beginning of follow-up.
Odds ratios and prevalence ratios based on samples tend to overestimate the true odds and prevalence ratios in the population sampled. This bias may be marked if the sample is small, and the use of estimators that offset the bias has been suggested. Jewell’s (1986) low-bias estimator of the odds ratio is
ad/[(b + 1) (c + 1)]
bc/[(a + 1) (d + 1)].
For the fairly large numbers in Table 1 the disease odds ratio becomes 1.83, instead of 1.89 by the usual formula. A disadvantage of this method is that the odds ratio for freedom from the disease is no longer the reciprocal of the disease odds ratio (Walter and Cook 1991); it becomes 0.48, which is 1/2.11. Jewell’s low-bias point estimate of the prevalence ratio is
[a/(a + b)]/[(c + 1)/(c + d + 1)]
[a/(a + c)]/[(b + 1)/(b + d + 1)]
(1.77 or 0.51 in this instance).
Studies of causal relationships generally use odds or other ratios, but differences, especially between prevalences, may be preferred to ratios when interest lies in the magnitude of a public health problem; for example, if we wish to estimate how many people in a population have headaches because of exposure to fumes, or to use this information in estimating treatment costs or impact on productivity. A useful measure based on the prevalence difference is the number needed to prevent one case, on the assumption that exposure is causal and can be avoided; this is the reciprocal of the prevalence difference. According to Table 1, 22.5 people are needed in the unexposed group to avoid one case.
In analytical cross-sectional studies that aim to explain as well as to describe associations, a variety of measures and techniques may be used to control confounding factors and determine whether other variables modify the association. The procedures range in complexity from stratification and standardization to sophisticated multivariate techniques that permit the simultaneous consideration of a large number of variables and their relationships. The findings in separate strata (for example, sex and age groups) are frequently combined by the Mantel–Haenszel or similar procedures to obtain odds ratios, rate ratios, or rate differences that control for the effects of the stratifying variable or variables; the summary measures should be used only after appraising the homogeneity of the findings in the strata in order to see whether they can be validly combined (Fleiss 1981; Selvin 1996; Abramson and Gahlinger 2001).
Uses in community health care
Cross-sectional studies can fulfil important functions in the health care of a community. They can contribute to the planning of services, to the effective implementation of care, and to decision-making on the continuation and modification of services. In this discussion, ‘community’ may be taken to refer to any aggregation of people for whose care a doctor, health-care team, agency, or authority is responsible; it may be a nation or region, a local neighbourhood, a list of registered patients, a defined group of schoolchildren or workers, inmates of an institution, and so on.
Separate consideration will be given to the use of cross-sectional studies in community diagnosis, in ongoing surveillance, in community health education and the promotion of community involvement, and in evaluation of the community’s health care.
Cross-sectional studies can provide a major part of the epidemiological foundation for community diagnosis, that is, for determining the health status of a community and the factors that influence it. They can supply information on the nature, extent, and impact of health problems, as a basis for the identification of priorities and the planning of intervention. Such studies may relate to a broad spectrum of health states and their correlates, or may be limited in their scope.
Cross-sectional studies may yield useful information on a variety of dimensions of health and disease, including self-appraised health, mental health status, growth and development, physical fitness, the distribution of blood pressures, and so on. The following remarks refer only to the prevalence of disorders; the cross-sectional method for the study of growth and development is discussed below.
It must be remembered that prevalence, especially point prevalence, may provide an incomplete picture because of the under-representation of conditions with a short duration. These include not only the acute non-fatal diseases that constitute a considerable load for the health services, and acute episodes of long-term or recurrent diseases, but also severe and rapidly fatal conditions, such as fatal strokes and sudden deaths from coronary heart disease. This bias was strikingly illustrated during a famine in Chad in 1985, when a rapid assessment displayed no severe malnutrition in children and it was concluded that serious malnutrition did not exist; in fact many children were affected, but they died too soon to be included in the survey (Guha-Sapir 1991).
The most direct evidence of a need for improved secondary and tertiary prevention, at least for long-term diseases that are not rapidly fatal, is an unduly high prevalence of remediable disease that has not been diagnosed or that is untreated or inadequately treated. Prevalence surveys providing such information may be based on examinations, interviews, clinical files, or other documentary sources, or a combination of these. In Australia, a two-stage prevalence study found that almost a million adults had moderate or severe hearing impairment that would probably be helped by a hearing aid, which well over half were not using (Wilson et al. 1999). In Italy, a survey based on the Registry of the Blind showed that the rate of blindness was much higher in the south of the country than in the north; possible causes for the difference in blindness due to treatable conditions such as cataract and glaucoma included a regional difference in the quality or accessibility of care (Nicolosi et al. 1994).
Needs for primary prevention can be inferred from the presence of preventable disorders, that is, those whose incidence can be reduced by known preventive measures. For this purpose too, the prevalence data should be supplemented by data on incidence and mortality, both because diseases with a high fatality rate will otherwise be under-represented and because prevalent cases may be long-standing ones that do not reflect present preventive needs. A high prevalence of crippling due to poliomyelitis does not necessarily mean that current preventive procedures are ineffective. Information on the recent incidence of new cases is to be preferred for this purpose. If prevalence data are to be used, information should be sought on the duration of the disorder, so that the prevalence of disease of recent onset can be measured. In institutional settings where people who develop a disorder are especially likely to remain in the institution, prevalence data may overestimate the need for primary prevention. In a hospital, for example, patients who develop nosocomial infections are for this reason likely to have a longer hospital stay. A prevalence survey of such infections in a hospital may thus give an exaggerated idea of the need for primary prevention.
The use of highly valid measures of the presence of a disease often presents practical difficulties, and reliance may be placed on a proxy measure that is simple, cheap, and acceptable; a screening test may be used for this purpose.
The confidence interval of the prevalence of the disease can be estimated from the prevalence of the proxy attribute (Rogan and Gladen 1978).
Determinants of health and disease
Information on the prevalence of modifiable factors that are known to affect health is of obvious relevance to the planning of health care. These may be factors with broad effects on health, for example, dietary, infant rearing, smoking, and family planning practices and (presumably) the use of health services, and they may be factors that affect the risk of developing specific disorders. They may be factors relevant to the community at large, such as poverty, unemployment, or air pollution, or they may be relevant to specific subgroups, as in studies of bullying behaviour in schools (Forero et al. 1999). They may be risk factors, which increase the risk of ill health, or protective factors, for example, physical activity or specific immunity (natural or acquired) to a pathogenic agent.
Associations between variables
When associations are investigated in a cross-sectional study in the context of community health care, the focus may be placed on specific diseases, disabilities, or other health characteristics, in order to throw light on their determinants or predictors, or on specific risk factors or protective factors, in order to examine their associations with health and disease. In this context, the aim is usually to determine what causal factors or correlates (of those known to be potentially important) are active in the specific community, and to measure their impact. The primary aim is to obtain information that will be useful in practice, not to generate new knowledge about aetiology, although this may be a secondary gain. A cross-sectional survey of respiratory symptoms and exposure to tobacco smoke in schoolchildren in Hong Kong, for example, was undertaken not to reconfirm known aetiological relationships, but to examine the impact of active and passive smoking in this population, as a basis for policy decisions on tobacco control (Lam et al. 1998).
Attention may also be centred on the determinants of supposed risk or protective factors, as in studies of the determinants of cigarette smoking, the use of a health service, or compliance with medical advice, and on associations among diseases or other dimensions of health, or among determinants of health.
Measurement of impact
In a situation where it is believed that an association of a risk factor with the prevalence (or incidence) of a disease expresses a causal relationship, the factor’s impact may be measured by the attributable (or aetiological) fraction in the population. This is the proportion of the disease in the population that can be attributed to exposure to the factor (Table 1). Among workers aged 20 to 64 years in a community in Jerusalem, for example, the fraction of the prevalence of varicose veins that could be attributed to work involving much standing was 16 per cent in each sex, after controlling for effects connected with age, region of birth, weight, and height (Abramson et al. 1981). Such values must be interpreted with caution, as part or all of the apparent causal effect may be due to other (uncontrolled) factors associated with the apparent causal factor. The attributable fraction among the exposed may also be of interest: 31 per cent of the prevalence of varicose veins in men whose work involved much standing could be attributed to their work posture; for women, this fraction was 32 per cent.
For a protective factor, the corresponding measures (Table 1) are the prevented fraction, which is the proportion of the hypothetical total prevalence that has been prevented by exposure to the factor, and the preventable fraction, which is the proportion of the observed prevalence that would be prevented if everyone was exposed to the factor. These fractions are sometimes termed the ‘efficacy’ of the protective factor, particularly with respect to vaccines.
Attributable, prevented, and preventable fractions are specific to a particular population, since they are influenced by the prevalence of the factor, and the factor’s causal effect may be dependent on the prevalence of other factors. The fractions attributable to different causes can sum up to more than 100 per cent, since causes operate in conjunction, and their attributable fractions overlap (Rothman and Greenland 1998).
Measures of impact are generally more helpful as a basis for intervention and policy decisions than odds and prevalence ratios or other measures of association. It is more useful to know that according to a cross-sectional study, 26 to 43 per cent of various asthma-like symptoms in young women in towns in East Anglia were attributable to the use of gas for cooking, than to know that the odds ratios were about 2 (Brauer and Kennedy 1996; Jarvis et al. 1996). If the proportion exposed to a factor is high, the attributable risk may be much higher than might be guessed from the odds or prevalence ratio; for drinking alcohol (ever) and breast cancer, for example, the population attributable fraction according to a study in New York was 25 per cent and the odds ratio 1.4 (Bowlin et al. 1997).
Interest may not be confined to cause–effect relationships. Any attribute or exposure that is strongly associated with a disease or other disorder, even non-causally, has potential value as a predictor, provided that there is reason to believe that it precedes the appearance of the disorder. Such predictors may be used as risk markers to identify vulnerable individuals or groups. The risk marker may be a factor that itself influences the risk, or a precursor or early manifestation of the disorder, or it may be secondarily associated with the disorder because it is associated with a cause or precursor of the disorder.
Risk markers are best identified by longitudinal studies, but can also be detected by cross-sectional ones. A cross-sectional study in Singapore, for example, showed that over 80 per cent of people with corneal arcus (a grey circle around the cornea) at the age of 30 to 49 years had high serum low-density lipoprotein cholesterol levels (Hughes et al. 1992). In The Netherlands, a cross-sectional study showed that divorced people were less healthy than single, married, or widowed people, controlling for age, sex, whether living alone, and other variables (Joung et al. 1994). In Israel, a national cross-sectional study showed a high prevalence of morbidity among 17-year-olds who were extremely underweight or extremely overweight, suggesting the use of low or excessive weight as markers to trigger intervention at an earlier age (Lusky et al. 1996). The reasons for the associations with corneal arcus, divorcee status, and body weight are of interest and may be of practical importance, but are irrelevant to the decision whether to use these characteristics as risk markers.
The value of a risk marker or combination of risk markers depends on the following considerations.
Is its use practical? Questions of simplicity, acceptability, safety, convenience, cost, and resources must be considered.
Is detection of high risk likely to be beneficial? Are resources and techniques available for reducing the risk? Does the benefit outweigh any harm that intervention may cause? What is the predictive value of the risk marker, that is, what proportion of people with the marker are likely to have or develop the disease?
How prevalent is the risk marker? If more than half of the children in a community fall into a high-risk group, might it not be more efficient and possibly more effective to modify the routine care programme so as to give extra attention to all children?
What is the marker’s sensitivity as a predictor? That is, what proportion of the individuals with the disorder will it identify? If this proportion is small, its value is limited.
The answers to these questions may vary in different contexts, as may the associations between specific factors and diseases. A given risk marker may be useful in one setting but not in another.
The term ‘community health syndrome’ may be used to refer to diseases or other health characteristics found to occur together in a community. Examples described by Kark (1974, 1981), who introduced the community syndrome concept and emphasized its potential importance for the development of community health programmes, are a syndrome of malnutrition, communicable diseases, and mental ill health in a poor rural community undergoing rapid change, and the syndrome of hypertension, coronary heart disease, and diabetes frequently found in affluent communities characterized by nutritional imbalance and excesses, limited physical activity, and a drive for achievement.
The components of a syndrome may occur together because they possess shared or related causes, or because they are themselves causally inter-related. The syndrome points to a nexus of causal processes in the community. Even if this nexus is not completely understood, a health programme directed at the syndrome as a whole may be more effective and efficient than an endeavour to deal separately with the individual components.
Associations between diseases or other health states may be detected at a population level or (more convincingly) at an individual level, that is, by finding a tendency to affect the same persons. As an example of the latter approach, a study of coprevalence in Jerusalem revealed clustering of migraine and other common disorders characterized by complaints (rather than objective signs) (Abramson et al. 1982). These disorders were frequently associated with emotional symptoms and with family disharmony or other stressful situations, and people with one or more components of the syndrome made heavy use of medical services. This syndrome represented a considerable burden of discomfort for many individuals and their families, and there was no organized programme to deal with it.
Identification of groups requiring special care
Community diagnosis may focus not only on the community as a whole but on its component groups. Comparisons may identify groups for whom special care may be needed. This identification may be based on the presence of disorders, on screening tests that point to a high probability of having a disorder, on the presence of modifiable risk factors, and/or on the presence of known risk markers indicative of vulnerability and a need for preventive care.
A differential approach in community diagnosis is of basic importance for the identification of priorities and the allocation of resources. Sometimes simple descriptive findings suffice for these purposes, and in other circumstances the planning of effective care requires an understanding of the reasons for the differences found, requiring the use of analytical epidemiological techniques.
The detection of a high-risk or high-morbidity group does not necessarily mean that special care is indicated; this depends on the likely benefits, on what proportion of the community’s cases or prospective cases is concentrated in the group, and on practical and other considerations.
Ongoing surveillance permits the identification of changes in health status and its determinants in the community, and updating of the community diagnosis. Repeated cross-sectional studies, as well as incidence studies, have a clear role, and are the only practicable method for some purposes, for example to detect changes in a community’s health habits or blood pressure distributions. Surveys of different representative samples may be advisable for this purpose, rather than repeated investigations of the same sample, to avoid the possibility that participation in a survey may affect the subjects’ behaviour, including their participation in health programmes and their responses in a later survey (Kroeger 1985; Puska 1991). In North Karelia, during 3 years of operation of a cardiovascular risk factor programme, men who were followed up longitudinally decreased their smoking by 11 per cent, whereas a comparison of the baseline data with a new representative sample showed a drop of only 7 per cent (Puska 1991). Repeated cross-sectional studies, like those of the WHO MONICA (Multinational Monitoring of Trends and Determinants in Cardiac Disease) project, yielded much information about changes in cardiovascular diseases and their risk factors in various countries (Tunstall-Pedoe et al. 1999).
Surveillance of the prevalence of chronic disorders may be based on repeated prevalence surveys, or on the use of a case register that is updated as new cases are found or old ones recover, die, or leave. Changes in the prevalence of chronic disorders may be important as an indication of changing needs for curative and rehabilitative care facilities.
Changes in the prevalence of a chronic disease cannot, however, be glibly taken to indicate changes in the risk of developing the disorder; for this purpose, incidence data should be used. There are a number of possible reasons for changes in prevalence. As illustrated in Fig. 1, the changes reflect the interplay of incidence, recovery, and fatality rates. They may be caused by changes in the demographic characteristics of the population as a result of ageing or inward or outward migration. Especially in studies of small local communities, prevalence may be influenced by a tendency of affected persons to leave or enter the neighbourhood.
Often, apparent changes in prevalence are artefacts caused by changes in methods of case identification (for example, the introduction of a case-finding programme), in the use of medical services, in diagnostic procedures or definitions, or in recording, notification, or registration practices. They may also be caused by incomplete updating of a case register.
Community education and community involvement
Community surveys can be used as tools for community health education. This may be done not only by communicating the findings and their implications to the community and its leaders, but also by using the educational potential of the survey situation itself, for example, by explaining to participants why the collection of specific information is important. If accurate results are required, such explanations should preferably be given after the information has been collected, to minimize bias in the responses.
An example is provided by the ‘Know Your Body’ programme, which aimed to motivate schoolchildren to adopt a healthier lifestyle (Williams et al. 1977). After the measurement of chronic disease risk factors, each child received a feedback of results in a ‘health passport’, together with explanations of desirable ranges for each test, in order to enhance the effect of the curriculum. Trials indicate that this programme effectively modified health knowledge, and had mixed effects on cardiovascular risk factors (Marcus et al. 1987; Walter et al. 1987; Tamir et al. 1990; Resnicow et al. 1993).
Involvement of key community members in the planning and conduct of a health survey may be a useful way to motivate them to a more active participation in the promotion of their community’s health. A community’s interest and involvement in its own health care may find expression in the performance of community self-surveys, even without the participation of professional health workers. Such surveys are usually simple descriptive ones, and may not collect very accurate or sophisticated information.
Evaluation of a community’s health care
In the context of community health care, the purpose of an evaluative study is to yield a factual basis for decisions about the provision of care to a specific community. This kind of evaluative study, the programme review, can be contrasted with programme trials, which aim to provide generalizable inferences about the value of a given type of health programme. In programme reviews, considerable attention is given to evaluation of the process of care (the performance of activities by providers and recipients of care), as well as to measurements of desirable and undesirable effects, especially more immediate outcomes.
Certain findings that are used in the process of community diagnosis as indicators of needs for health care, such as a high prevalence of preventable or remediable disorders, may by the same token be seen as indicators of the value of past health care. In some instances a prevalence survey may reveal more direct evidence of the quality of previous care; for example, the quality of the dental work in subjects’ mouths may be appraised, or the presence of inguinal hernia recurrences may be recorded (in one study, one in five operated hernias showed evidence of recurrence (Abramson et al. 1978)).
Evaluative judgements that relate to the subjects’ prior health care as a whole, however, may be less helpful than those relating to recent or current health care, and especially to care in the context of a specific health programme or service. Studies might deal, for example, with compliance with medical advice (what proportion of hypertensives are taking the medicines prescribed for them?), with satisfaction with medical care, or with immunization status.
In these as in other studies, separate attention is usually paid to population subgroups. The impact of a health programme often varies with age, sex, social class, and other characteristics.
Evidence of change in health status or practices may be provided by repeated cross-sectional studies as well as by incidence and other longitudinal studies. It is usually assumed that such changes (or their absence) are, at least to some extent, reflections of programme effectiveness and not attributable only to outside influences. At the very least, the findings may indicate whether there is a need for more detailed evaluative study.
More rigorous proof that the change is attributable to the programme requires a comparison with controls. For example, the effectiveness of a breast-feeding promotion programme was confirmed by the finding that the prevalence of breast feeding rose much more than in a neighbouring community with no such programme (Palti et al. 1988).
Uses in clinical practice
Epidemiological studies serve important functions in clinical care. The role of cross-sectional studies are considered in individual and family care and in community-oriented primary care.
Individual and family care
Textbooks of clinical epidemiology emphasize that epidemiology is a basic science for clinicians, and demonstrate its use (Sackett et al. 1991). ‘A great many routine clinical decisions about individual patient care … can only be based upon information from properly designed and executed studies in groups or populations’ (Roberts 1977). The systematic use of properly appraised information is the central feature of what is increasingly becoming known as evidence-based medicine—’an approach that integrates the best external evidence with individual clinical expertise and patient choice’ (Sackett et al. 1997).
Cross-sectional studies play a role here, if a modest one, since they can provide information on the prevalence of diseases and their symptoms and causes, the frequency distributions of biochemical and other measurements in the population and in patients with specific disorders, patterns of child growth, health practices, and so on. This information may be derived from any studies whose results can validly be generalized to the population in which the clinician works. A population study in Edinburgh, for example, that demonstrated the high prevalence, in people without varicose veins, of lower limb symptoms commonly attributed to varicose veins, cannot be ignored by a doctor anywhere who is considering the extirpation of varices in order to ameliorate symptoms (Bradbury et al. 1999).
Epidemiological information of this sort is seldom the fruit of the clinician’s own labours, except in a context of community-oriented primary care (see below). Sometimes, however, even a doctor concerned only with individual patients may conduct what in effect are small-scale (usually cross-sectional) epidemiological surveys—for example, of a patient’s contacts for evidence of a communicable disease, to reduce the patient’s risk of reinfection. A family doctor, responsible for the care of whole families, may need to perform such investigations more often; the discovery that a patient has a condition with a known tendency to ‘run in families’ (for example, rheumatic heart disease, diabetes, amoebiasis, AIDS, or Helicobacter pylori infection) may be seen as a signal that the whole family should be surveyed.
Family diagnosis, the process of appraising a family’s health status and the factors that affect it, is an exercise in small-group epidemiology, conducted by a family practitioner. Its aim is to determine a family’s health needs as a basis for the planning of a family health programme. This involves elucidation of the family members’ health status and appraisal of relevant features of the family life situation: for example, the family’s structure and composition, the role performance and health-relevant behaviour of its members, relationships, material resources and their use, and the family’s social and physical environment. At an analytical level, it involves appraisal of the ways in which family members’ health may be affected by other members and by the family life situation.
Community-oriented primary care
Community-oriented primary care refers to the combination of the care of individuals and the care of the community as a whole, in a single integrated practice (Kark 1981; Abramson 1988; Kark et al. 1994; Kark and Kark 1999). The practitioner or team providing clinical care initiates or participates in specific community programmes that deal in a systematic way with the main health needs of the community and its subgroups. There is growing awareness of the potential of this form of integrated practice for improving health in both developing and developed countries (Connor and Mullan 1983; Nutting 1987; Gillam et al. 1994; Tollman and Friedman 1994; Rhyne et al. 1998).
Epidemiology is an indispensable basis for the planning, development, and evaluation of the community health programmes that characterize community-oriented primary care, and the uses (listed above) of cross-sectional studies both in community health care and in individual and family care are relevant to this form of practice.
Obviously, the information required for the practice of community-oriented primary care is not limited to the community served—evidence concerning the effectiveness of various methods of treatment and various community interventions may come from studies elsewhere, and a decision that a specific program is needed (for example, for the control of hypertension) may be based on findings in other or broader populations. But information about the specific community is always required, and the community-oriented primary care practitioner or team must ensure that this is collected, or a community orientation is likely to remain a well-meaning aspiration rather than a means of effecting demonstrable improvements in health.
Information about the community may have three foci, as follows.
A general picture (‘getting to know the community’), whose purposes include identification of health problems that merit detailed study and possible action (a ‘needs assessment’) (Trompeter 1992), and appraisal of their possible causes and the resources and circumstances that may be relevant to their solution. Use is generally made of ‘rapid’ methods, based on easily available sources and qualitative research procedures.
A more detailed community diagnosis with respect to selected health problems and their determinants, in order to decide whether a programme is warranted (is there a sufficient ‘case for action’?) and, if so, to aid in its planning and implementation, and provide baseline data for the measurement of the changes it is expected to produce.
Monitoring of programme activities and surveillance of change, leading to an evaluation of effectiveness.
Much of this information (but obviously, not all) is derived from cross-sectional studies. These do not necessarily refer to a fixed calendar time; the prevalence of a disease, for example, is usually determined by interviews or examinations that are staggered over a period. An important feature is that much of the information needed for community diagnosis, surveillance, and evaluation is generally obtained in the course of routine clinical care. The collection of data thus serves a double function. When a child is weighed or a question is asked about smoking or a diagnosis is made, the results may be used both in the management of the patient and as data for subsequent analysis at a group level. The data may be derived either from routine clinical procedures or from questions or tests specially added for epidemiological purposes. In a practice where periodic health examinations are conducted, these provide an especially useful opportunity for the collection of data for analysis.
This use of clinical data demands careful attention to methods of obtaining, recording, and retrieving data, to make the information as accurate and complete as possible.
Standardized procedures are required, and definitions and diagnostic criteria should be standardized as rigorously as possible. The collection of data by care providers in a care setting carries advantages and disadvantages. On the one hand, it may be relatively easy to obtain answers to awkward questions, and to achieve a high response rate. On the other hand, there are several sources of possible bias.
Except in groups with very high attendance rates (for instance infants and their mothers, pregnant women, and the elderly), there may be a need for supplementary survey procedures to obtain information about members of the community who have not attended for clinical care, and these people may be invited to attend, visited at home, or asked for information by mail or telephone.
If resources are available, information may also be obtained by special surveys. In a community-oriented primary care setting, these generally aim to provide information that will benefit individual participants as well as providing a basis for decisions at a community level. The WHO has published a practical guide to the conduct of simple epidemiological surveys at a district level (Vaughan and Morrow 1989).
Sampling is seldom appropriate, since one aim is usually to identify individuals who need care.
For chronic diseases, a common technique is the maintenance of a case register. This permits the calculation of prevalence rates or other epidemiological indices and may assist in monitoring the performance of tests and other activities, besides its use as a tool to ensure that patients receive the care they need.
The process of information collection can serve to promote community development and the community’s involvement in its own care, which are usually defined as aims of community-oriented primary care.
Meetings with community leaders, designed to learn their opinions about health problems and their solutions, can stimulate interest and promote community action, as can surveys and feedbacks of survey findings. In some community-oriented primary care practices, focus groups have proved to be a valuable means not only of collecting data, but also of enabling the community to recognize its needs, so that it can mobilize to meet them (Plaut et al. 1993).
Studies yielding ‘new knowledge’
Many cross-sectional studies are performed to expand the horizons of knowledge, rather than solely to promote the health of the specific groups or populations studied. They are ‘research’ studies that aim to yield generalizable inferences that are of broad applicability, not relevant only to a specific local context.
Studies of growth and development, and of aetiology and programme trials, are briefly considered. Other research topics include the natural history of health and disease (usually better investigated by time-span studies) and methodological issues. Cross-sectional studies are also commonly used in comparisons of diagnostic criteria and other operational definitions and of study methods, and to appraise the validity of screening tests and proxy measures.
Studies of growth and development
Growth and development, and age trends in the prevalence of disorders, can be studied cross-sectionally as well as longitudinally. The cross-sectional method compares different age groups observed at one point of time, whereas the longitudinal method makes repeated observations of a single cohort as its age changes. The cross-sectional method is simpler, but has limitations. It can provide information about average changes, but not about intraindividual changes or interindividual differences.
The main limitation of the cross-sectional method is that the age groups that are compared may differ in respects other than age, so that the effects of age and other influences may be confounded. There is always potential confounding of age changes with differences between birth cohorts, as the age groups that are compared must belong to different cohorts. Cohort differences in growth may be negligible, but they may be significant if cohorts were exposed to very different circumstances: for example, different infant-feeding or child-rearing practices, changes in economic prosperity, or war. As a result the cross-sectional method may yield a misleading picture. If there has been a secular increase in height, a cross-sectional study may show a decrease in average height throughout adult life, but young adults will be taller because they belong to a more recently born generation, not because they are younger. If a series of cross-sectional studies has been done, suitable rearrangement of the data may permit examination and comparison of the longitudinal changes in different cohorts.
In studies that include the middle-aged and elderly, selective survival may be important. The mean blood pressure may be lower in the very old, not because blood pressure tends to drop with age, but because hypertensive people are more likely to have died and thus left the study sample. Also, the validity of measures may vary with age. The results of a memory test in the elderly may reflect hearing ability, attentiveness, or depression, rather than memory capacity.
Studies of aetiology
Cross-sectional studies often provide useful guides to aetiological processes, especially with respect to influences on long-term disorders and relatively stable measurements and health habits. They have two features, however, that often restrict their value for the testing of causal hypotheses.
Firstly, any associations they reveal are with the presence, not the appearance, of the disorder or other variable studied. Transient or rapidly fatal cases are inevitably under-represented. The causes that determine the appearance of the disorder are confounded with those that influence its duration, and it may be difficult to draw clear inferences about either set of causes. As an example, a high frequency of the A2 human lymphocyte antigen was found in children with acute lymphocytic leukaemia, suggesting that this was a risk factor, but later studies showed that the antigen lengthened the children’s lifespan, which was why prevalent cases included a high proportion who had the antigen (Rogentine et al. 1972, 1973; cited by Newman et al. 1988). Such confounding is relatively unimportant if the disorder studied is seldom fatal and has high chronicity, or data on lifetime prevalence are used. In such instances the main difficulty in a cross-sectional study is that the causal factors may no longer be apparent because of the time-lag since the initiation of the disease.
P>Secondly, in a strictly cross-sectional study the absence of information on time relationships may render it difficult to separate effects on a dependent variable from effects of the dependent variable. The influence of blood pressure, serum cholesterol, and cigarette smoking on the occurrence of myocardial infarction, for example, may be confounded with changes ensuing from the disease episode.
The demonstration in a cross-sectional study that fat abdomens (based on a comparison of waist with hip size) are associated with hypertension, hypertensive heart disease, and diabetes (controlling for sex, age, and ponderal index) is difficult to interpret without knowing which came first, the fat abdomen or the disease (Gillum 1987). The discovery of an inverse association in San Francisco’s bus drivers between hypertension and reported job-related problems—a relationship that was not explained by confounding factors and was specific to hypertension (gastrointestinal, respiratory, and musculoskeletal problems were positively associated with the self-reported stressors)—was given two competing explanations. On the one hand, ’emotional states or coping mechanisms … may play a role in the pathogenesis of hypertension through the repression of anger and hostility’. On the other, ‘the hemodynamic consequences of elevated blood pressure may lead to a physiologic alteration of perception … elevation in blood pressure reduces reactivity to noxious stimuli’ (Winkleby et al. 1988).
There is no ‘chicken-or-egg’ (time-sequence) problem if the postulated cause is blood type or some other genetically determined characteristic or a long-past exposure or long-lived acquired attribute that can be assumed to precede the onset of the latent period of the disease being studied, or if a causal process in the opposite direction does not make sense. A study of postmenopausal women aged 45 to 61 in England, for example, showed that (controlling for numerous possible confounders) fast walking and climbing stairs were positively associated with bone mineral density measured in the femoral trochanter and in the whole body, and walking frequency was associated with a high bone density in the trochanter and the femoral neck, but only in women who walked fast (Coupland et al. 1999). The findings strongly support the hypothesis that these activities affect bone density, since a reverse causal association is untenable, and numerous possible confounders were controlled in the analysis. But even here, the possibility of confounding by some uncontrolled factor (for example, muscularity) that preceded and affected both bone density and activity patterns cannot be excluded.
The value of a cross-sectional study in the search for causes and precursors is limited whenever there is a possibility that the disease may change the subject’s lifestyle, bodily functions and characteristics, or circumstances. To throw light on time sequences, cross-sectional studies are often extended to include historical information on times of disease onset or of other occurrences. Repeated cross-sectional studies of the same population can sometimes establish the order of events.
Cross-sectional studies may be fruitful sources of causal hypotheses for testing in time-span studies, even if, for one or other of the above reasons or because of possible bias or confounding, they do not themselves convincingly support a causal explanation. Here are three examples. The observation by Gregg (1941) that most of the mothers of a series of infants with congenital cataract had had rubella in pregnancy led to a series of studies that proved the causal relationship between rubella in early pregnancy and congenital anomalies. Secondly, a strong association between dietary diversity and anthropometric measures of nutritional status in Kenyan toddlers (Onyango et al. 1998) might be attributed either to the effect of diet on growth or to the effect of health status on the acceptance of new foods; the former explanation, which seems more likely, is open to testing in a cohort or intervention study. Thirdly, a cross-sectional study of American army veterans of the Persian Gulf War revealed a number of specific associations between symptoms and reported exposures (for example, to pesticides, anti-nerve-gas pills, and chemical and biological warfare agents), which the authors proposed to address by analysing the longitudinal data available for the subjects (Proctor et al. 1998).
Cross-sectional studies frequently demonstrate unexplained associations whose investigation in subsequent epidemiological or other studies might lead to important new knowledge. As an example, the most striking finding in a study of the prevalence of human T-lymphotropic virus type I infection in seven villages in Gabon (where this infection is endemic) was that prevalence was very much higher in the Kota-Obamba ethnic group (Le Hasran et al. 1994). No behavioural or other differences were found that could explain this high prevalence. Similarly, a study of young adults in the Antwerp region revealed large unexplained differences between small areas in the prevalence of respiratory symptoms (Wieringa et al. 1998). Associations of this kind pose aetiological riddles—findings awaiting an explanation.
A cross-sectional study may form the first stage of a time-span study, for which it provides baseline measurements of dependent and independent variables, and sometimes a sampling base. If the study is concerned with the incidence of a long-term disease, the baseline prevalence study identifies affected people, who may be followed up in order to study the natural history of the disease, but must be excluded from the population at risk of subsequently developing the disease. A cluster survey in a region of Tanzania, for example, provided information on the prevalence of HIV-1 infection in adults aged 15 to 34 years (the prevalence was 10 per cent, reaching 24 per cent in an urban zone), and seronegative individuals were re-examined 2 years later in order to determine the rate and correlates of seroconversion (Killewo et al. 1993).
Cross-sectional studies can have a role in studies that aim to provide generalizable conclusions and inferences about the value of a given type of health programme.
A simple way of testing the effectiveness of a programme is to compare the status of people who have and have not been exposed to it. In eight clinics in Lesotho, for example, a children’s growth monitoring and nutrition education programme was evaluated by means of a cross-sectional study in which maternal knowledge about infant feeding was measured, and mothers were classified according to whether or not they had previously attended the clinic. Women who had attended were found to be more knowledgeable about the introduction of animal protein foods, the use of oral rehydration salts, and the method of weaning (Ruel et al. 1992). But an evaluative study based on simple comparison of the findings or changes seen in people who voluntarily participate or do not participate in a programme is generally unconvincing, as the comparison may be confounded by differences between the groups. It is difficult to be sure that the difference in the findings can be attributed to the difference in exposure to the programme. Perhaps women who were more knowledgeable, or more educable, were also more likely to attend the clinics. The possible confounding effects are not easy to control. In this study, the difference in knowledge remained apparent when a few easily measured possible confounders—maternal education, working status, parity, and the child’s age—were controlled in the analysis. The authors correctly limited their conclusion to the statement that previous clinic attendance ‘appeared to be’ beneficial.
A similar approach is used in studies that aim to evaluate preventive and therapeutic procedures or programmes by comparing people who have experienced an unfavourable outcome with controls, to see whether they differ in their prior exposure to the procedure or programme. Such studies offer a relatively simple and rapid approach to evaluation (Smith 1989). But they present problems, particularly the possibility that the cases and controls may have differed in their initial characteristics (including prognostic factors) or eligibility for the procedure or programme, and care is required to control for confounding and other biases (Horwitz and Feinstein 1981).
To obtain convincing evidence of the effectiveness of a health programme that aims to modify the distribution of a characteristic in a population or to reduce the prevalence of a disease, it is necessary to measure the change in the population and to demonstrate that this can be attributed to the programme rather than to other causes. A trend shown by repeated cross-sectional studies adds to the force of the evidence—for example, in a population where the prevalence of anaemia in pregnant women was originally 12 per cent, and the introduction of an intervention programme was followed by a progressive drop to 8.8, then 3.3, then 1.6 per cent (Kark 1981). But the cause-and-effect relationship between a programme and its apparent outcome is always difficult to substantiate without observations of a comparison or reference population not exposed to the programme.
In trials of programmes directed at populations, it is unfortunately seldom possible to randomize. There is often little or no choice as to which population will be exposed to the programme under trial, and a restricted choice concerning a control population. Most programme trials are therefore quasi-experiments, in which the control group or groups are selected so as to be as similar as possible to the intervention group, and it remains necessary to control for possible confounders in the analysis. In a 5-year evaluation of the CHAD programme (Community Syndrome of Hypertension, Atherosclerosis and Diabetes) for the control of cardiovascular risk factors in a Jerusalem neighbourhood, for example, where cross-sectional studies revealed a significantly greater improvement in the exposed population than in a neighbouring control population, the confounders that were controlled included age, sex, education, and region of birth. The prevalence of some risk factors declined to some extent in the control population as well, apparently as a result of changes in awareness and health care—underlining the need for comparison groups in such studies. Over the next 18 years the essential features of the programme were adopted in the clinic serving the control population, and differences in risk factors became less obvious (Abramson et al. 1994). Difficulties in the evaluation of community programmes are discussed by Blackburn (1991), who stresses the need for studies involving more communities (possibly smaller), and preferably randomized.
The use of population samples in surveys designed to evaluate health programmes is discussed by Salonen et al. (1986), who compare the advantages and disadvantages of surveying separate samples on each occasion, compared with repeated surveys of the same samples.
This chapter has reviewed uses, methods, strengths, and weaknesses of cross-sectional studies. The essential feature of such surveys is that they collect information relating to a single specified time, but a time dimension is often introduced by the inclusion of easily collected historical information or by comparing successive studies so as to appraise changes in health status and its determinants, and their sequence.
Like other epidemiological studies, cross-sectional studies (descriptive and analytical) can both contribute to the health care of a specific group or community and serve as a research method for the attainment of generalizable new knowledge. These purposes, especially the first, can sometimes be met by the use of simple, undemanding, and inexpensive methods, such as cluster sampling; ‘rapid epidemiological assessment’ is especially relevant in situations (in both developed and developing countries) where resources are limited or results are required as a basis for programme decisions.
Cross-sectional surveys may be used in community diagnosis, ongoing surveillance, community health education and the promotion of community involvement, and evaluation of the community’s health care, and they can contribute to the planning of services, the effective implementation of care, and decision-making on the continuation and modification of services.
In clinical practice, information derived from prevalence studies of the population in which a clinician works is of special pertinence. The performance of cross-sectional studies is often integrated with clinical care in the provision of community-oriented primary care, where (with other sources of information) they play an essential role in the planning, development, and evaluation of community health programmes.
Cross-sectional studies can contribute new knowledge on growth and development, birth-cohort effects, aetiology, the effectiveness of health programmes, the validity of screening tests, and other topics. Uncertainties concerning time relationships limit the value of cross-sectional studies in aetiological research, unless the causal factors are genetically determined ones, long-past exposures, or long-lived acquired attributes that can be assumed to precede the onset of the latent period of the disease under study.
Cross-sectional studies are a fruitful source of causal hypotheses for subsequent testing in other studies. A cross-sectional study may form the first stage of a longitudinal study, for which it provides baseline measurements of dependent and independent variables, and sometimes a sampling base.
Abramson, J.H. (1988). Community-oriented primary care: strategy, approaches and practice. Public Health Reviews, 16, 35.
Abramson, J.H. and Abramson, Z.H. (1998). Survey methods in community medicine: epidemiological research, programme evaluation, clinical trials (5th edn). Churchill Livingstone, Edinburgh.
Abramson, J.H. and Abramson, Z.H. (2001). Making sense of data: a self-instruction manual on the interpretation of epidemiologic data (3rd edn). Oxford University Press, New York.
Abramson, J.H. and Gahlinger, P.M. (2001). Computer programs for epidemiologists: PEPI 4.0. Sagebrush Press, Salt Lake City, UT.
Abramson, J.H., Gofin, J., Hopp, C., Makler, A., and Epstein, L.M. (1978). The epidemiology of inguinal hernia. Journal of Epidemiology and Community Health, 32, 59.
Abramson, J.H., Hopp, C., and Epstein, L.M. (1981). The epidemiology of varicose veins: a survey in western Jerusalem. Journal of Epidemiology and Community Health, 35, 213.
Abramson, J.H., Gofin, J., Peritz, E., Hopp, C. and Epstein, L.M. (1982). Clustering of chronic disorders—a community study of co-prevalence in Jerusalem. Journal of Chronic Diseases, 35, 221.
Abramson, J.H., Gofin, J., Hopp, C., Schein, M.H., and Naveh, P. (1994). The CHAD program for the control of cardiovascular risk factors in a Jerusalem community: a 24-year retrospect. Israel Journal of Medical Sciences, 30, 108.
Anderson, D.W., Schoenberg, B.S., and Haerer, A.F. (1988). Prevalence surveys of neurologic disorders: methodological implications of the Copiah County study. Journal of Clinical Epidemiology, 41, 339.
Anker, M., Guidotti, R.J., Orzeszyna, S., Sapirie, S.A., and Thuriaux, M.C. (1993). Rapid evaluation methods (REM) of health services performance: methodological observations. Bulletin of the World Health Organization, 711, 15.
Armstrong, J.R.M. and Hayes, R. (1992) Estimating prevalence of injecting drug use in an urban population: limitations of the three-sample estimation procedure. International Journal of Epidemiology, 21, 613.
Babaniyi, O. and Parakoyi, B. (1991). Cluster survey for poliomyelitis and neonatal tetanus in Ilorin, Nigeria. International Journal of Epidemiology, 20, 515.
Balraj, V., Mukundan, S., Samuel, R., and John, T.J. (1993). Factors affecting immunization coverage levels in a district of India. International Journal of Epidemiology, 22, 1146.
Bennett, S., Woods, T., Liyanage, W.M., and Smith, D.L. (1991). A simplified general method for cluster-sample surveys of health in developing countries. World Health Statistics Quarterly, 44, 98.
Bishop, Y., Fienberg, S., and Holland, P. (1975) Discrete multivariate analysis: theory and practice. MIT Press, Cambridge, Massachusetts.
Blackburn, H. (1991). Community programmes in coronary heart disease prevention and health promotion: changing community behaviour. In Coronary heart disease epidemiology: from aetiology to public health (ed. M. Marmot and P. Elliott), p. 495. Oxford University Press.
Bosch, J.L., Hop, W.C., Kirkels, W.J., and Schroder, F.H. (1995). Natural history of benign prostatic hyperplasia: appropriate case definition and estimation of its prevalence in the community. Urology, 46 (3 Supplement A), 34.
Bowlin, S.J., Leske, M.C., Varma, A., Nasca, P., Weinstein, A., and Caplan, L. (1997) Breast cancer risk and alcohol consumption: results from a large case-control study. International Journal of Epidemiology, 26, 915.
Bradbury, A., Evans, C., Allan, P., Lee, A., Ruckley, C.V., and Fowkes, F.G.R. (1999). What are the symptoms of varicose veins? Edinburgh vein study cross-sectional population survey. British Medical Journal, 318, 353.
Brauer, M. and Kennedy, S.M. (1996). Gas stoves and respiratory health. Lancet, 347, 412.
Breslow, N. (1982). Design and analysis of case–control studies. Annual Reviews of Public Health, 3, 29.
Connor, E. and Mullan, F. (ed.) (1983). Community oriented primary care: new directions for health service delivery. National Academy Press, Washington, DC.
Coreil, J., Augustin, A., Holt, E., and Halsey, N.A. (1989). Use of ethnographic research for instrument development in a case-control study of immunization use in Haiti. International Journal of Epidemiology, 18, S33.
Coupland, C.A.C., Cliffe, S.J., Bassey, E.J., Grainge, M.J., Hosking, D.J., and Chilvers, C.E.D. (1999). Habitual physical activity and bone mineral density in postmenopausal women in England. International Journal of Epidemiology, 28, 241.
Coxell, A., Kink, M., Mezey, G., and Gordon, D. (1999). Lifetime prevalence. characteristics, and associated problems of non-consensual sex in men: cross-sectional survey. British Medical Journal, 318, 846.
Edungbola, L.D., Alabi, T.O., Oni, G.A., Asaolu, S.O., Ogunbanjo, B.O., and Parakoyi, B.D. (1987). ‘Leopard skin’ as a rapid diagnostic index for estimating the endemicity of African onchocerciasis. International Journal of Epidemiology, 16, 590.
Elandt-Johnson, R.C. (1975). Definitions of rates: some remarks on their use and misuse. American Journal of Epidemiology, 102, 261.
Erkinjuntti, T., Ostbye, T., Steenhuis, R., and Hachinski, V. (1997). The effect of different diagnostic criteria on the prevalence of dementia. New England Journal of Medicine, 337, 1667.
Ferreira, M.U., Cardoso, M.A., Santos, A.L., Ferreira, C.S., and Szarfarc, S.C. (1996). Rapid epidemiologic assessment of breastfeeding practices: probit analysis of current status data. Journal of Tropical Pediatrics, 42, 50.
Fienberg, S.E. (1972). The multiple recapture census for closed populations and incomplete 2k contingency tables. Biometrics, 59, 591.
Fleiss, J.L. (1981). Statistical methods for rates and proportions (2nd edn). Wiley, New York.
Forero, R., McLellan, L., Rissel, C., and Bauman, A. (1999). Bullying behaviour and psychosocial health among school students in New South Wales, Australia: cross-sectional survey. British Medical Journal, 319, 344.
Frerichs, R.R. (1989). Simple analytic procedures for rapid microcomputer-assisted cluster surveys in developing countries. Public Health Reports, 104, 24.
Frerichs, R.R. and Tar, K.T. (1989). Computer-assisted rapid surveys in developing countries. Public Health Reports, 104, 14.
Frischer, M., Bloor, M., Finlay, A., et al. (1991). A new method of estimating prevalence of injecting drug use in an urban population: results from a Scottish city. International Journal of Epidemiology, 20, 997.
Garland, M., Morris, J.S., Colditz, G.A., et al. (1996). Toenail trace element levels and breast cancer: a prospective study. American Journal of Epidemiology, 144, 653.
Gillam, S., Plamping, D., McClenaham, J., Harries, J., and Epstein, L. (1994). Community-oriented primary care. King’s Fund, London.
Gillum, R.F. (1987). The association of body fat distribution with hypertension, hypertensive heart disease, diabetes and cardiovascular risk factors in men and women aged 18–79 years. Journal of Chronic Diseases, 40, 421.
Greenland, S. (1987). Interpretation and choice of effect measures in epidemiologic analysis. American Journal of Epidemiology, 125, 761.
Gregg, N.M. (1941). Congenital cataract following German measles in the mother. Transactions of the Australian College of Ophthalmology, 3, 35.
Guha-Sapir, D. (1991) Rapid assessment of health needs in mass emergencies: review of current concepts and methods. World Health Statistics Quarterly, 44, 171.
Gyapong, J.O., Adjei, S., Gyapong, M., and Asamoah, G. (1996). Rapid community diagnosis of lymphatic filariasis. Acta Tropica, 61, 65.
Harris, D.R. and Lemeshow, S. (1991). Evaluation of the Expanded Programme on Immunization survey methodology for estimating relative risk. World Health Statistics Quarterly, 44, 107.
Helmert, U., Herman, B., and Shea, S. (1994). Moderate and vigorous leisure-time physical activity and cardiovascular disease risk factors in West Germany, 1984–1991. International Journal of Epidemiology, 23, 285.
Helton, A.S., McFarlane, J., and Anderson, E.T. (1987). Battered and pregnant: a prevalence study. American Journal of Public Health, 77, 1337.
Henderson, R.H. and Sundaresan, T. (1982). Cluster sampling to assess immunization coverage: review of experience with a simplified sampling method. Bulletin of the World Health Organization, 60, 253.
Hoddinott, P. and Pill, R. (1999). Qualitative study of decisions about infant feeding among women in east end of London. British Medical Journal, 318, 30.
Horwitz, R.I. and Feinstein, A.R. (1981). The application of therapeutic-trial principles to improve the design of epidemiologic research: a case-control study suggesting that anticoagulants reduce mortality in patients with myocardial infarction. Journal of Chronic Diseases, 34, 575.
Hughes, K., Lun, K.C., Sothy, S.P, Thai, A.C., Leong, W.P., and Yeo, P.B. (1992). Corneal arcus and cardiovascular risk factors in Asians in Singapore. International Journal of Epidemiology, 21, 473.
Jarvis, D., Chinn, S., Luczynska, C., and Burney, P. (1996). Association of respiratory symptoms and lung function in young adults with use of domestic gas appliances. Lancet, 347, 426.
Jewell, N.P. (1986). On the bias of commonly used measures of association for 2 × 2 tables. Biometrics, 42, 351.
Joung, I.M.A., van de Mheen, H., Stronks, K., van Poppel, F.W.A., and Mackenbach, J.P. (1994). Differences in self-reported morbidity by marital status and by living arrangement. International Journal of Epidemiology, 23, 91.
Kahn, H.A. and Sempos, C.T. (1989). Statistical methods in epidemiology (2nd edn). Oxford University Press, New York.
Kark, S.L. (1974). Epidemiology and community medicine. Appleton-Century-Crofts, New York.
Kark, S.L. (1981). The practice of community-oriented primary health care. Appleton-Century-Crofts, New York.
Kark, S. and Kark, E. (1999). Promoting community health: from Pholela to Jerusalem. Witwatersrand University Press, Johannesburg.
Kark, S.L., Kark, E., Abramson, J.H., and Gofin, J. (ed.) (1994). Atencion Primaria Orientada a la Comunidad. Ediciones Doyma, Barcelona.
Keeffe, J.E., Lovie-Kitchin, J.E., Maclean, H., and Taylor, H.R. (1996). A simplified screening test for identifying people with low vision in developing countries. Bulletin of the World Health Organization, 74, 525.
Khan, M.E., Anker, M., Patel, B.C., Barge, S., Sadhwani, H., and Kohle, R, (1991). The use of focus groups in social and behavioural research: some methodological issues. World Health Statistics Quarterly, 44, 145.
Kiemeny, L.A.L.M., Schouten, L.J., and Straatman, H. (1994). Ascertainment corrected rates. International Journal of Epidemiology, 23, 203.
Killewo, J.Z.J., Sandstrom, A., Raden, U.B., Mhalu, F.S., Biberfeld, G., and Wall, S. (1993). International Journal of Epidemiology, 22, 528.
Kleinbaum, D.G., Kupper, L.L., and Morgenstern, H. (1982). Epidemiologic research: principles and quantitative methods. LifeTime Learning Publications, Belmont, CA.
Koopman, J.S. and Longini, I.M. Jr (1994). The ecological effects of individual exposures and nonlinear disease dynamics in populations. American Journal of Public Health, 84, 836.
Kroeger, A. (1983). Health interview surveys in developing countries: a review of the methods and results. International Journal of Epidemiology, 12, 465.
Kroeger, A. (1985). Response errors and other problems of health interview surveys in developing countries. World Health Statistics Quarterly, 38, 15.
LaForce, F.M., Lichnevski, M.S., Keja, J., and Henderson, R.H. (1980). Clinical survey techniques to estimate prevalence and annual incidence of poliomyelitis in developing countries. Bulletin of the World Health Organization, 58, 609.
Lam, T.H., Chung, S.F., Betson, C.I., Wong, C.M., and Hedley, A.J. (1998). Respiratory symptoms due to active and passive smoking in junior secondary school students in Hong Kong. International Journal of Epidemiology, 27, 41.
Lee, J. (1994). Odds ratio or relative risk for cross-sectional data? International Journal of Epidemiology, 23, 201.
Le Hasran, J.Y., Delaporte, E., Gaudebout, C., et al. (1994). Demographic factors associated with HTLV-I infection in a Gabonese community. International Journal of Epidemiology, 23, 812.
Lemeshow, S. and Robinson, D. (1985). Surveys to measure programme coverage and impact: a review of the methodology used by the Expanded Programme on Immunization. World Health Statistics Quarterly, 38, 65.
Lemeshow, S., Tserkovnyi, A.G., Tulloch, J.L., Dowd, J.E., Lwanga, S.K., and Keja, J. (1985). A computer simulation of the EPI survey strategy. International Journal of Epidemiology, 14, 473.
Lengeler, C., Mshinda, H., de Savigny, D., Kilima, P., Morona, D., and Tanner, M. (1991). The value of questionnaires aimed at key informants, and distributed through an existing administrative system, for rapid and cost-effective health assessment. World Health Statistics Quarterly, 44, 150.
Leske, M.C., Wu, S.-Y., Connell, A.M.S., Human, L., Schachat, A.P., and the Barbados Eye Study Grpup (1987). Lens opacities, demographic factors and nutritional supplements in the Barbados Eye Study. International Journal of Epidemiology, 26, 1314.
Linstone, H.A. and Turoff, M. (ed.) (1975). The Delphi method: techniques and applications. Addison-Wesley, Reading, MA.
Loney, P.L., Chambers, L.W., Bennett, K.J., Roberts, J.G., and Stratford, P.W. (1998). Critical appraisal of the health research literature: prevalence or incidence of a health problem. Chronic Diseases in Canada, 19, 170.
Lusky, A., Barell, V., Lubin, F., et al. (1996). Relationship between morbidity and extreme values of body mass index in adolescents. International Journal of Epidemiology, 25, 829.
McCarty, D.J., Tull, E.S., Moy, C.S., Twoh, C.K., and LaPorte, R.E. (1993). Ascertainment corrected rates: applications of capture-recapture methods. International Journal of Epidemiology, 22, 559.
MacMahon, B. and Pugh, T.F. (1970). Epidemiology: principles and methods. Little, Brown, Boston, MA.
MacNamara, J.J., Molot, M.A., Stremple, J.F., and Cutting, R.T. (1971). Coronary artery disease in combat casualties in Vietnam. Journal of the American Medical Association, 216, 1185.
Marcus, A.C., Wheeler, R.C., Cullen, J.W., and Crane, L.A. (1987). Quasi-experimental evaluation of the Los Angeles Know Your Body program: knowledge, beliefs, and self-reported behaviours. Preventive Medicine, 16, 803.
Materia, E., Imoko, J., Berhe, G., Dawuda, C., Omar, M.A., Pinto, A., and Guerra, R. (1995). Rapid surveys in support of district health information systems: an experience from Uganda. East African Medical Journal, 72, 15.
Mayaud, P., Msuya, W., Todd, J., et al. (1992). STD rapid assessment in Rwandan refugee camps in Tanzania. Genitourinary Medicine, 73, 33.
Mays, N. and Pope, C. (1996). Qualitative research in health care. BMJ Publishing, London.
Miettinen, O.S. (1985), Theoretical epidemiology: principles of occurrence research in medicine. Wiley, New York.
Needleman, H.L., Gunnoe, C., Levison, A., Reed, R., Pereshie, H., Maher, C., and Barrett, N. (1979). Deficits in psychologic and classroom performance of children with elevated dentine lead levels. New England Journal of Medicine, 300, 689.
Newman, T.B., Browner, W.S., Cummings, S.R., and Hulley, S.B. (1988). Designing a new study: II. Cross-sectional and case-control studies. In Designing clinical research (ed. S.B. Hulley and S.R. Cummings). Williams and Wilkins, Baltimore, MD.
Nicolosi, A., Marighi, P.E., Rizzardi, P., Osella, A., and Miglior, S. (1994). Prevalence and causes of visual impairment in Italy. International Journal of Epidemiology, 23, 359.
Nutting, P.A. (ed.) (1987). Community oriented primary care: from principle to practice. Health Resources and Services Administration, Public Health Services, Washington, DC.
Onyango, A., Koski, K., and Tucker, K.I. (1998). Food diversity versus breastfeeding choice in determining anthropometric status in rural Kenyan toddlers. International Journal of Epidemiology, 27, 484.
Osborn, J. and Carraruzza, M.S. (1995). Odds ratio and relative risk for cross-sectional data. International Journal of Epidemiology, 24, 464.
Palti, H., Adler, B., Flug, D., Shamir, Z., and Kark, S.L. (1977). Community diagnosis of psychomotor development in infancy. Israel Annals of Psychiatry, 15, 223.
Palti, H., Valderama, C., Pugrund, R., Jarkoni, J., and Kurtzman, C. (1988). Evaluation of the effectiveness of a structured breastfeeding promotion programme integrated into a maternal child health service in Jerusalem. Israel Journal of Medical Sciences, 24, 432.
Pani, S.P., Srividya, A., Krisknamoorthy, K., Das, P.K., and Dhanda, V. (1997). Rapid assessment procedures (RAP) for lymphatic filariasis. National Medical Journal of India, 10, 19.
Papoz, L., Balkau, B., and Lellouch, J. (1996). Case counting in epidemiology: limitations of methods based on multiple data sources. International Journal of Epidemiology, 25, 474.
Patterson, R.E., Kristl, A.R., Shannon, J., Hunt, J.R., and White, E. (1997). Using a brief household food inventory as an environmental indicator of individual dietary practices. American Journal of Public Health, 87, 272.
Pearce, N. (1993). What does the odds ratio estimate in a case-control study? International Journal of Epidemiology, 22, 1189.
Plaut, T., Landis, S., and Trevor, J. (1993). In Successful focus groups (ed. D.L. Morgan). Sage, Beverly Hills, CA.
Proctor, S.P., Heeren, T., White, R.F., et al. (1998). Health status of Persian Gulf War veterans; self-reported symptoms, environmental exposures and the effect of stress. International Journal of Epidemiology, 27, 1000.
Puska, P. (1991). Intervention and experimental studies. In Oxford textbook of public health (2nd edn) (ed. W.W. Holland, R. Detels, and G. Knox), p. 177. Oxford University Press.
Rendl, J., Bier, D., Groh, T., and Reiners, C. (1998). Rapid urinary iodide test. Journal of Clinical Endocrinology and Metabolism, 83, 1007.
Resnicow, K., Cross, D., and Wynder, E. (1993). The Know Your Body program: a review of evaluation studies. Bulletin of the New York Academy of Medicine, 70, 188.
Rhyne, R., Bogue, R., Kukulka, G., and Fulmer, H. (1998). Community-oriented primary care: health care for the 21st century. American Public Health Association, Washington, DC.
Ritenbaugh, C.K., Said, A.K., Gaslal, O.M., and Harrison, G.G. (1989). Development and evaluation of a colour-coded scale for birthweight surveillance in rural Egypt. International Journal of Epidemiology, 18, S54.
Roberts, C.J. (1977). Epidemiology for clinicians. Pitman Medical, Tunbridge Wells.
Rogan, W.J. and Gladen, B. (1978). Estimating prevalence from the results of a screening test. American Journal of Epidemiology, 107, 71.
Rogentine, G.N., Yankee, R.A, Gart, J.J., Nam, J., and Trapani, R.J. (1972). HL-A antigens and disease: acute lymphatic leukemia. Journal of Clinical Investigation, 51, 2410.
Rogentine, G.N., Trapani, R.J., and Henderson, E.S. (1973). HL-A antigens and acute lymphocytic leukemia: the nature of the HLA2 association. Tissue Antigens, 3, 470.
Ronneberg, A., Lund, K.E., and Hafstad, A. (1994) Lifetime smoking habits among Norwegian men and women born between 1890 and 1974. International Journal of Epidemiology, 23, 267.
Rothenberg, R.B. (1985) Observations on the application of EPI cluster methods for estimating disease incidence. Bulletin of the World Health Organization, 63, 93.
Rothman, K.J. (1986). Modern epidemiology. Little, Brown, Boston, MA.
Rothman, K.J. (ed.) (1988). Causal inference. Epidemiology Resources, Chestnut Hill, MA.
Rothman, K.J. and Greenland, S. (1998). Modern epidemiology (2nd edn). Lippincott–Raven, Philadelphia, PA.
Ruel, M.T., Habicht, J.-P., and Olson, C. (1992). Impact of a clinic-based growth monitoring programme on maternal nutrition knowledge in Lesotho. International Journal of Epidemiology, 21, 59.
Ruston, A., Clayton, J., and Calnan, M. (1998). Patients’ action during their cardiac event: qualitative study exploring differences and modifiable factors. British Medical Journal, 316, 1060.
Sackett, D.L., Haynes, K.B., and Tugwell, P. (1991). Clinical epidemiology: a basic science for clinical medicine, (2nd edn). Williams and Wilkins, Baltimore, MD.
Sackett, D.L., Richardson. W.S., Rosenberg, W., and Haynes, R.B. (1997). Evidence-based medicine: how to practice and teach EBM. Churchill Livingstone, New York.
Salonen, J.T., Kottke, T.E., Jacobs, D.R. Jr, and Hannan, P.J. (1986). Analysis of community-based cardiovascular disease prevention studies—evaluation issues in the North Karelia Project and the Minnesota Heart Health Program. International Journal of Epidemiology, 15, 176.
Schulman, K.A., Berlin, J.A., Harness, W., et al. (1999). The effect of race and sex on physicians’ recommendations for cardiac catheterization. New England Journal of Medicine, 340, 618.
Schwartz, L.M., Woloshin, S., and Welch, H.G. (1999). Misunderstandings about the effects of race and sex on physicians’ referrals for cardiac catheterization. New England Journal of Medicine, 341, 279.
Schwoebel, V., Dauvisis, A.-V., Helynck, B., et al. (1992). Community-based evaluation survey of immunizations in Burkina Faso. Bulletin of the World Health Organization, 70, 583.
Scrimshaw, S.C.M. and Hurtado, E. (1987). Rapid assessment procedures for nutrition and primary health care: anthropological approaches to improving programme effectiveness. UCLA Latin American Center Publications, Los Angeles, CA.
Sekikawa, A., Eguchi, H., Tominaga, M., et al. (1999). Evaluating the reported prevalence of type 2 diabetes mellitus by the Oguni diabetes registry using a two-sample method of capture-recapture. International Journal of Epidemiology, 28, 498.
Selvin, S. (1996). Statistical analysis of epidemiologic data (2nd edn). Oxford University Press.
Smith, G.S. (1989) Development of rapid epidemiologic assessment methods to evaluate health status and delivery of health services. International Journal of Epidemiology, 18 (Supplement 2), S1.
Smith, P.G. and Morrow, R.H. (1991). Methods for field trials of interventions against tropical diseases: a ‘toolbox’. Oxford University Press.
Sommer, A., Hussaini, G., Muhilal, T.I., Susanto, D., and Saroso, J.S. (1980). History of nightblindness: a simple tool for xerophthalmia screening. American Journal of Clinical Nutrition, 33, 887.
Susser, M. (1973). Causal thinking in the health sciences: concepts and strategies in epidemiology. Oxford University Press, New York.
Tamir, D., Feurstein, A., Brunner, S., Halfon, S.-T., Reshef, A., and Palti, H. (1990). Primary prevention of cardiovascular diseases in childhood: changes in serum total cholesterol, high density lipoprotein, and body mass index after 2 years of intervention in Jerusalem schoolchildren age 7–9 years. Preventive Medicine, 19, 22.
Tollman, S. and Friedman, I. (1994). Community-oriented primary health care—South African legacy. South African Medical Journal, 84, 646.
Trompeter, T. (1992). Community responsive primary care: a basic guide to planning and needs assessment for community and migrant health. National Association of Community Health Centers, Washington, DC.
Tunstall-Pedoe, H., Kuulasmaa, K., Mahonen, M., Tolonen, H., Ruokokoski, E., and Amouyel, P. (1999). Contribution of trends in survival and coronary-event rates to changes in coronary heart disease mortality: 10-year results from 37 WHO MONICA project populations. Monitoring trends and determinants in cardiovascular disease. Lancet, 353, 1547.
Turner, A.G., Magnani, R.J., and Shuaib, M. (1996). A not quite as quick but much cleaner alternative to the Expanded Programme on Immunization (EPI) cluster survey design. International Journal of Epidemiology, 25, 198.
Van de Laar, M.J.W., Termorshuizen, F., Slomka, M.J., et al. (1998). Prevalence and correlates of herpes simples vitus type 2 infection: evaluation of behavioural risk factors. International Journal of Epidemiology, 27, 127.
Van de Ven, A.H. and Delbecq, A.L. (1972). The nominal group as a research instrument for exploratory health studies. American Journal of Public Health, 62, 337.
Vaughan, J.P. and Morrow, R.H. (ed.) (1989). Manual of epidemiology for district health management. World Health Organization, Geneva.
Velzeboer, M.I., Selwyn, B.J., Sargent, F., Pollitt, E., and Delgado, H. (1983a). Evaluation of arm circumference as a public health index of protein energy malnutrition in early childhood. Journal of Tropical Pediatrics, 29, 135.
Velzeboer, M.I., Selwyn, B.J., Sargent, F., Pollitt, E., and Delgado, H. (1983b). The use of arm circumference in simplified screening for acute malnutrition by minimally-trained health workers. Journal of Tropical Pediatrics, 29, 159.
Venkataswamy, G., Lepkowski, J.M., Ravilla, T., et al. (1989). Rapid epidemiologic assessment of cataract blindness. International Journal of Epidemiology, 18, S60.
Von Korff, M. and Parker, R.D. (1980). The dynamics of the prevalence of chronic episodic disease. Journal of Chronic Diseases, 33, 79.
Walter, S.D. and Cook R.J. (1991). A comparison of several point estimators of the odds ratio in a single 2 × 2 contingency table. Biometrics, 47, 795.
Walter, H.J., Hofman, A., Barrett, L.T., et al. (1987). Primary prevention of cardiovascular disease among children: three years’ results of a randomized intervention trial. In Cardiovascular risk factors in childhood: epidemiology and prevention (ed. B.S. Hetzel and G.S. Berenson), p. 161. Elsevier, Amsterdam.
West, S.K., Munoz, B., Turner, V.M., Mmbaga, B.B.O., and Taylor, H.R. (1991) The epidemiology of trachoma in Central Tanzania. International Journal of Epidemiology, 20, 1088.
WHO (World Health Organization) (Division of Epidemiological Surveillance and Health Situation and Trend Analysis) (1993). News from the World Health Organization: rapid evaluation methods in health services. International Journal of Epidemiology, 22, 578.
Wieringa, M.H., Weyler, J.J., Nelen, V.J., et al. (1998). Prevalence of respiratory symptoms: marked differences within a small geographical area. International Journal of Epidemiology, 27, 630.
Williams, C.L., Arnold, C.B., and Wynder, E.L. (1977). Primary prevention of chronic disease beginning in childhood. The ‘Know Your Body’ program: design of study. Preventive Medicine, 6, 344.
Wilson, D.H., Walsh, P.G., Sanchez, I., et al. (1999). The epidemiology of hearing impairment in an Australian adult population. International Journal of Epidemiology, 28, 247.
Winkleby, M.A., Ragland, D.R., and Syme, S.L. (1988). Self-reported stressors and hypertension: evidence of an inverse association. American Journal of Epidemiology, 127, 124.
Zocchetti, C., Consonni, D., and Bertazzi, P.A. (1997). Relationship between prevalence rate ratios and odds ratios in cross-sectional studies. International Journal of Epidemiology, 26, 220.