Sage Journals: Discover world-class research

Abstract

Objective: To provide an introduction to the concepts and methods of psychiatric epidemiology for non-specialist readers.

Methods: An overview of concepts and research procedures based on the published literature.

Conclusion: Psychiatric epidemiology is a key mental health discipline that has adopted and developed a variety of methods and procedures for the study of complex mental disorders. In its future development, psychiatric epidemiology is likely to seek an integration of the conceptual and methodological advances of molecular biology and neuroscience within the population-based and social framework of epidemiology as a basic science of public health.

Keywords

epidemiology methods psychiatry

Epidemiology has been defined as ‘the study of the distributions and determinants of states of health in human populations’[1]. The beginnings of modern epidemiology can be traced to the confluence of methods used in late 19th century in the investigation of contagious diseases and the body of knowledge known as ‘sanitary statistics’. The answer to the recurrent question of whether epidemiology is a ‘toolbox’ of methods that can be applied to any inquiry into human disease [2], or a substantive discipline possessing a special knowledge base [3], is that it combines both. As a set of methods and techniques, epidemiology is a potent tool in the study of communicable and non-communicable diseases (with specialized ‘branch’ epidemiologies, e.g. of HIV, malnutrition, cancer, cardiovascular disease, mental disorders). As a body of substantive knowledge about the general laws governing morbidity, mortality and the amelioration of health in human population, it is a basic science of public health.

Modern epidemiology has acquired a highly sophisticated statistical apparatus but its basic theoretical principles and rules of inference were laid down over a century ago in social science [4] and in medicine, which embraced a model of disease causation derived from bacteriology [5]. Thus, the essential elements of any epidemiological inquiry still involve, in one form or another, interactions between ‘hosts’ (individual persons), pathogens (e.g. infectious or toxic agents, mutant genes, stressors) and environments (e.g. urban overcrowding, protective social networks) –all this in the context of a defined population. Two more recent concepts assessing such interactions are exposure (to risk factors) and outcome (in terms of state of health). The applications of epidemiology are commonly divided into descriptive (establishing the frequency and distribution of health or disease states), analytical (search for causes) and experimental (interventions to prevent or treat disease).

Psychiatric epidemiology

The earliest application of an essentially epidemiological method to the study of psychiatric morbidity took place in 1895, when a young female physician, Jenny Koller [7], undertook in Zurich an investigation of the frequency of mental disorders among the relatives of hospitalized psychiatric patients and compared them to the frequency of mental disorders among the relatives of surgical patients (today this would be recognized as a case-control study). The peak achievement of that early period was the work of Joseph Goldberger [8] who started with observations on the population frequency of pellagra psychosis in different geographical areas in the USA and Italy; examined its association with ecological factors (including diet); formulated a hypothesis of the nutritional origin of the disorder; and proceeded with laboratory investigations which ultimately demonstrated the aetiological role of nicotinic acid deficiency. The concluding stage was a public health intervention (dietary supplementation) on a mass scale that resulted in the virtual elimination of a common physical disease associated with significant psychiatric morbidity. The subsequent period in the development of psychiatric epidemiology (from the 1930s to the 1960s) was characterized by great refinements of the field survey method (a number of door-to-door surveys of geographically defined populations in Germany, Scandinavian countries and the USA); longitudinal repeat surveys in which the same individuals were traced in the community and reinterviewed at later time points to establish the ‘natural history’ of disease [9]; and birth cohort studies [10]. A ‘third generation’ [11] of studies was heralded by the introduction of explicit diagnostic criteria such as Research Diagnostic Criteria (RDC), DSM-III/IIIR, and involved the application of standardized, presumably more reliable, diagnostic instruments to representative population samples [12–15].

The current, ‘fourth generation’ of psychiatric epidemiological research is characterized by a strong emphasis on the search for specific risk factors, both biological and psychosocial. Just around the corner, there is an emerging ‘fifth generation’ of studies, which are aiming to integrate recent advances in the understanding of the human genome into the search for causes of psychiatric disorders on a population basis (molecular epidemiology or human genome epidemiology). Current psychiatric epidemiological research is increasingly orientated towards so-called ‘strategic populations’ [16] that may be more informative with regard to teasing out causative pathways in psychiatric illness –such as genetic isolates, samples of specifically configured pedigrees and groups known to be at an increased risk of morbidity.

The questions that psychiatric epidemiology aims to answer are not generically different from those that any branch of disease epidemiology attempts to tackle –these are questions pertaining to the incidence and prevalence of disorders, groups at high or low risk of disease, associations with characteristics of the host and the environment, determinants of disease outcome, and response to specific interventions. Ultimately, establishing the epidemiological ‘signature’ of a disease should provide guidance for laboratory research into its molecular mechanisms. However, psychiatric epidemiology faces challenges that are rarely encountered in other epidemiological investigations. With a few exceptions, the unit of analysis in psychiatric epidemiology –the case –is not flagged by a ‘pathognomonic lesion’ [17] or by a disease marker, such as high blood pressure or tumour cytology, that could reliably identify ‘caseness’. Instead, the psychiatric epidemiologist has to make sense of subjectively reported symptoms or observed behaviour to infer a diagnostic classification of cases. This puts researchers into a sort of ‘double bind’ –on one hand, epidemiological leads are important signposts for aetiological or pathogenetic research in the laboratory; on the other hand, the value of epidemiological data would be greatly enhanced if valid disease markers for use in the field were generated by laboratory research. While validated markers to complement the clinical criteria will eventually be forthcoming, a way out of the present conundrum is to put all available knowledge about psychiatric disease phenotypes to the best possible use by employing standardized, operational definitions and by constantly updating that knowledge through a close, twoway interaction with clinical and laboratory scientists.

Defining the unit of observation

The reliable measurement of the prevalence, incidence and disease expectancy of mental disorders as a prerequisite for research into risk factors and causes depends critically on (i) the capacity to identify within a given population all affected persons (or the great majority of them); and (ii) the availability of a diagnostic assessment system that selects ‘true’ cases corresponding to established clinical concepts. The first requirement refers to the sensitivity of the case finding method (which should minimize the false negative exclusions), while the second requirement has to do with the specificity of diagnostic allocation that is needed to minimize the occurrence of false positive cases.

Case definition and case identification

Cases in treatment contact

Psychiatric hospital or outpatient populations provide relatively easy access to cases for epidemiological investigation. However, the probability of being in treatment depends on a host of factors, such as the availability of services, their location and accessibility, and the rate of their utilization by various population groups –none of which is strongly correlated with the actual burden of morbidity. For example, patients admitted to hospital are unlikely to be a representative sample of all the individuals with a given disorder. The distributions by age and sex, marital state, socioeconomic status, ethnicity, and severity of illness in hospital samples tend to differ from those describing the larger pool of people in the community who exhibit the disorder of interest. The extent of selection bias characterizing clinical populations may vary widely across treatment settings and between different points in time. It has been proposed [18] that, under conditions of adequate service provision, the great majority of people with a severe mental disorder such as schizophrenia eventually become admitted to hospital. However, the general trend in mental health care at present is away from hospital treatment. It has been shown, for example, that as many as 50% of people with first-episode schizophrenia are not admitted to hospital within the three months of their first contact with a primary care facility [19]. As regards non-psychotic disorders, a classic study by Shepherd et al. [20], as well as many subsequent investigations, demonstrated that fewer than 1 in 20 of people with neurotic illnesses seen by general practitioners are ever referred for specialist care. Moreover, recent epidemiological surveys [12–15] have shown that a substantial proportion of community residents with nonpsychotic illnesses diagnosed at interview never sought treatment. Therefore, restricting epidemiological case finding to specialist mental health care settings, or even to cases in any kind of treatment, may miss an important segment of mental morbidity in the community.

Case registers

Psychiatric case registers (or mental health information systems, as they are often referred as at present) are useful tools of epidemiological research. Registers are systematic, cumulative databases which usually cover a wide spectrum of psychiatric morbidity by collating information from multiple sources, including outpatient services. Although registers provide no access to cases that never come up for assessment or treatment, their advantage as an epidemiological tool derives from the coverage of a defined population (providing a reliable denominator for estimates of prevalence and incidence); the cumulative case registration over long periods of time; and the capacity for linking individual case records to other population databases. Notwithstanding their limitation with regard to coverage of cases that have never been in treatment, registers are highly effective research instruments for the study of long-term trends in psychiatric morbidity, for targeted sampling of clinical populations with defined characteristics, and for investigating correlates and associations of mental morbidity with demographic and environmental factors.

Case finding in the community

The identification of the ‘true’ population prevalence and incidence of psychiatric disorders (in contrast to the treated, or ‘administrative’, prevalence and incidence) is an important goal of psychiatric epidemiology. Among other things, knowledge about the proportion of cases not in treatment, their characteristics and needs, can be instrumental in service planning and evaluation. The principal method for achieving this goal is the carefully designed community survey, in which all residents within a demographic unit, or a representative sample of them, are ascertained in a systematic way to detect or rule out ‘caseness’. The ascertainment method can take many forms, ranging from a face-to-face interview to a mail-out of self-administered questionnaires. Although the personal interview may, with some justification, be regarded as the ‘gold standard’ in survey research, alternative techniques, including telephone interviews, questionnaires and rating scales may be as effective, depending on the specific aims of the study. A number of such tools with tested properties (reliability, validity, predictive value, etc.) are available; a selected few are described in Table 1.

Table 1.

Personality traits in Eysenck's and Cloninger's personality models

Generally, a survey can be designed as a single-phase data collection, in which complete diagnostic or other information is obtained through a single contact with every respondent; or as a two-phase survey, in which a brief screening instrument is administered first, to be followed by a more detailed examination of only those respondents who score as ‘screen-positive’, that is, are likely to have the disorder of interest.

Screening for psychiatric disorders in the community

Screening has been defined as ‘the presumptive identification of unrecognized disease or defect by the application of tests, examinations or other procedures which can be applied rapidly. Screening tests sort out apparently well persons who probably have a disease from those who probably do not’ [21]. The important point is that screening is not a diagnostic procedure but a ‘prediagnostic’ filtering process which selects individuals with highest probability of having the disorder for subsequent specific diagnostic evaluation [22]. Screening tests in psychiatric epidemiology are typically self-administered questionnaires or scales, brief face-to-face or telephone interviews, or checklists. Apart from the requirement of reliability (i.e. consistency of performance), the main attribute of a useful screening tool is its validity, that is, its capacity to correctly identify individuals with the disorder of interest (sensitivity) and correctly eliminate those without the disorder (specificity). A schematic representation of these attributes is given in Appendix 2.

Several dozen screening instruments have been used in psychiatric epidemiological research; three of them, including the General Health Questionnaire, (GHQ) [23], which is among the most popular ones for identifying non-psychotic mental morbidity, are briefly described in Table 1.

The use of screening tools in a survey which aims to establish the extent of psychiatric morbidity in the community (see Table 1) should be distinguished from community screening programmes designed to detect cases that need to be offered treatment. According to guidelines published by World Health Organization (WHO) [24], such screening programmes should meet 7 criteria: (i) the condition should represent an important health problem; (ii) screening programmes must be costeffective; (iii) effective treatments must be available for the disorder; (iv) the screening tests should be reliable and valid; (v) the tests should have high cost-benefit to the patient; (vi) early detection by screening should significantly reduce mortality or morbidity; and (vii) treatment administered during the early or presymptomatic phase should be of greater efficacy than treatment administered during the symptomatic phase. It is clear that few screening methods for psychiatric disorders would satisfy at present the majority of these criteria. Currently, screening in psychiatric epidemiology is mainly confined to case finding. It is likely, however, that screening programmes, linked to intervention programmes, will be increasingly feasible for common treatable conditions, such as anxiety disorders and depression.

Diagnosis and classification

Diagnostic concepts, classifications and assessment instruments play a critical role in epidemiological research since a considerable proportion of the variation in the results of various studies is due to variation in diagnostic concepts. Until the late 1960s, the diagnostic rules used in epidemiological research were seldom explicitly stated and the description of assessment techniques often lacked sufficient detail. As demonstrated by the USAUK diagnostic study [25], concepts of schizophrenia used by psychiatrists trained in different medical cultures could differ to an extent that might invalidate epidemiological comparisons. Since the 1980s, the comparability of clinical diagnostic data has been substantially improved by the introduction of operational diagnostic criteria such as RDC, DSM-III/IIIR/IV and ICD-10. However, there is no conclusive evidence that the increased reliability of diagnoses for clinical and biological research has been paralleled by an equally improved validity of diagnostic classification for epidemiological research. Restrictive diagnostic criteria, such as DSM-III and DSM-IIIR, select more homogeneous patient groups and reduce the rate of false positive diagnoses. However, their application as inclusion/ exclusion categories at the point of case identification in epidemiological surveys may result in falsely rejecting potentially eligible cases who, at the time of the initial assessment, fail to satisfy the full set of criteria. Depending on the study objectives and design, it might be preferable to apply such criteria at a later stage, when more extensive clinical assessment has been accomplished, while applying less restrictive criteria at the case finding stage in two-phase surveys. As a general rule, erring on the over-inclusive side is preferable to overrestrictiveness in such designs since false positive cases could easily be eliminated from the final analysis while falsely rejected cases are unlikely to be retrieved.

Diagnostic instruments

Instruments used in epidemiological research in psychiatric disorders differ with regard to purpose and scope, sources of data, setting of application, output format, and type of user. The most widely used current diagnostic instruments fall into two categories. The first comprises fully structured instruments such as the National Institute of Mental Health (NIMH) Diagnostic Interview Schedule, (DIS) [26] and the related World Health Organization–Alcohol, Drug Abuse, and Mental Health Administration (WHO-ADAMHA) Composite International Diagnostic Interview, (CIDI) [27] which have been written to match specifically and exclusively the diagnostic criteria of DSM-IIIR and ICD-10. Such instruments are designed for use by non-psychiatric (lay) interviewers. Clinical judgement plays no part in their administration and scoring. The second category includes semi-structured interview schedules such as the Present State Examination, (PSE) [28] and the Schedules for Clinical Assessment in Neuropsychiatry, (SCAN) [29, 30] which were developed to cover a broad range of psychopathology using professional clinical judgement, and to elicit primary data that can be processed by alternative diagnostic algorithms (including ICD-10 and DSM-IIIR). Each type of instrument has its advantages and disadvantages. Fully structured interviews can be used by lay interviewers with brief (usually two-week) training and achieve a high level of inter-rater reliability. They are capable of generating standard diagnoses in a single-phase survey design but the range of psychopathology covered is restricted to the diagnostic system with which it is interlocked. Both the DIS and CIDI have performed well in surveys of the common, non-psychotic illnesses but their sensitivity and specificity in diagnosing psychotic disorders has been shown to be problematic [31]. On the other hand, the PSE or SCAN type of interview allows the collection of a great amount of clinical information which can be processed and interpreted in alternative ways. Both the reliability and the validity of the PSE/ SCAN are to a large extent a function of the adequate clinical training and skills of the interviewer. This tends to limit their applicability as single-phase epidemiological tools, though they can provide highly informative diagnostic assessment in two-phase designs where potential cases are identified by a screening instrument.

Types of study design

A simple typology of the research designs available to the epidemiologist –depending on the aims and the questions asked –is as follows [32].

– Descriptive studies Cross-sectional (prevalence surveys) Prospective (incidence studies, including repeat cross-sectional surveys)

– Analytical studies Case-control studies (testing specific hypotheses, usually involving risk factors) Cohort studies (enabling longitudinal evaluation of correlates of morbidity) Ecological (examining spatial distributions or time trends in morbidity)

– Experimental studies Intervention studies (evaluating the impact of a new service or prevention programme) Clinical trials (essentially epidemiological in design, though not usually regarded as within the domain of epidemiology)

Cross-sectional surveys

The essential feature of cross-sectional designs is that measures of outcome (e.g. presence or absence of adult depression) and of exposure to a risk factor (e.g. a history of dysfunctional child rearing) are obtained at the same point in time –by face to-face interviews, postal questionnaires, telephone inquiries, or the Internet. There are two principal restrictions on the method: (i) the investigator has limited control over the reliability and validity of retrospective data on exposure –especially in data gathering designs which do not involve direct interactive contact with the respondent; and (ii) cause-and-effect inferences may be problematic with cross-sectional data. Techniques are, however, available to estimate and reduce bias resulting from potentially invalid retrospective data, as well as to enable tentative conclusions about causal relationships. The field, or door-to-door survey (typically involving face-to-face contact), is the method which, to date, has produced the most comprehensive data on the broad range of mental morbidity in various populations. Surveys may involve a census (enumeration) of an entire population or community, or a statistically representative sample which is interviewed to establish current presence (point prevalence) of ‘active’ disorders, or of disorders that have occurred at any time in the past (lifetime prevalence). Examples of sample surveys are the NIMH Epidemiological Catchment Area (ECA) study in which some 20 000 persons were interviewed at five research sites in the United States [12]; the National Comorbidity Survey [13]; the UK National Survey of Psychiatric Morbidity [14]; and the recent Australian National Survey of Mental Health and Wellbeing [15].

Prospective surveys of incidence

Establishing the incidence of disorders requires periodic re-examination or continuous monitoring of a population in order to detect emerging new cases. Repeat cross-sectional surveys of the same population will identify new cases that have become manifest in the interval between the two surveys. Another method of detecting new cases as they arise is to set up and maintain an active case finding network in the community over a period of time. This approach is particularly appropriate for the estimation of incidence of disorders with a long prodromal period, such as schizophrenia, especially when the aim is to identify first-onset cases as early as possible. Since the ‘true’ onset of schizophrenia is usually impossible to pinpoint in the majority of cases (due to the insidious transition from prodrome to overt psychotic illness), consistency in estimating incidence can be ensured by defining onset as the earliest time when the symptom presentation meets agreed diagnostic criteria. This implies screening for likely ‘caseness’ all individuals initiating contact with any relevant ‘helping agency’ (psychiatric or general medical services, youth counselling services, schools, police, alternative medicine practitioners) and inviting for a diagnostic assessment those who score positive on the screen. Continuous monitoring of firstcontact points over two years was employed in the WHO 10-country study of schizophrenia which generated comparative incidence data using identical case finding protocols in 12 geographical areas in developed and developing countries [44].

Longitudinal studies

The cohort study is the paradigm of longitudinal epidemiological research in which exposure and outcome data are mapped onto a ‘real time’ dimension. In cohort studies, subjects are selected by a defined characteristic, such as birth in a specified year or period, or membership in a group with a particular exposure experience (e.g. war veterans). The actual investigation of the cohort may be prospective (through periodic re-examinations), retrospective (with data gathering at predetermined points, e.g. when its members have reached a specified age), or a combination of both. The birth cohort study provides yet another method for determining incidence and morbid risk. The cohort study method was first applied by Klemperer [33] who drew a random sample of 1000 individuals born in Germany from 1881–1890 and attempted to trace them as adults in their fourth decade of life. However, he failed to account for more than 44% of the cohort, which illustrates the main methodological hazard of the method –cohort attrition which may be aggravated in populations with high geographical mobility. Nevertheless, there are examples of remarkable success when the method is applied to stable ‘captive’ populations. Thus, Fremming [34] in Denmark; Helgason [35] and Helgason & Magnusson [10] in Iceland were able to trace 92% and 99.4% of the members of birth cohorts and to collect data for the estimation of lifetime morbidity risk.

Case-control studies

In case-control studies, a group of individuals, selected for presence of a given disease or risk factor and a group of control subjects without that disease or risk factor, are compared in terms of variables that may be associated with, or causally related to, the outcome of interest. Case-control studies are crosssectional, gathering retrospective data on exposures, or prospective, involving a follow-up of the effects of a risk factor over time. Critical issues in case-control designs are: (i) the definition of inclusion/exclusion criteria of ‘caseness’ and (ii) the characteristics and source of recruitment of the control subjects. Since, ideally, cases and controls should be closely similar on all characteristics except one –the disease or risk factor defining ‘caseness’ –the inclusion/exclusion criteria should aim to ensure within-group homogeneity with regard to that characteristic, rather than representativeness vis-à-vis the population. In practical terms, this means that cases should be selected by criteria that maximize their similarity in terms of clinical syndrome, length and stage of illness (or any other relevant variable), and minimize the accretion of factors that may confound the testing of the main hypothesis, such as comorbid conditions or past exposure to very different treatment modalities.

The selection of controls is often of paramount importance for the outcome of any comparisons; errors in this respect may lead to seriously flawed conclusions. To reduce ‘ecological’ bias, controls should be preferably selected from socioeconomic backgrounds or geographical neighbourhoods similar to those of the cases; depending on the research question, similarity of ethnic composition may also be sought. To ensure similarity, matching procedures –individual (pairwise) or group (frequency) matching –may be employed. However, a major problem in psychiatric research stems from the fact that non-psychotic mental disorders, such as episodes of depression and anxiety, or substance use –for example factors that may interact with the variable of interest –are extremely common in the general population [13, 15]. Screening out potential controls for such exposures may result in a sample of ‘super healthy’ controls that would seriously inflate the case-control differences in the analysis.

Another confounding influence may result from control self-selection as a response to recruitment strategies employing advertisements or similar appeals to volunteers (motivational factors related to personality traits or lifestyle are likely to play a role). Such bias can be reduced by using a random or systematic sampling protocol and/or by applying judiciously defined screening criteria. With all these qualifications and cautions, case-control studies are at present among the most costeffective designs in psychiatric epidemiology.

Ecological studies

In ecological studies, rates of morbidity, or their change over time, are statistically related to aggregate measures of exposure within a defined population but no individual exposure data are available (or sought). Examples include: studies finding an excess in winter births among individuals who develop schizophrenia [36]; studies relating variations in the incidence of adult schizophrenia in members of birth cohorts to influenza epidemics that had coincided in time with mothers’ pregnancies [37]; studies reporting associations between adult schizophrenia and urban birth [38]; or studies finding time series correlations between suicide rates and economic cycles [39]. Although ecological studies can be of considerable value in generating hypotheses (to be tested using other designs), their findings qualify, at best, as weak inferences because of their inherent vulnerability to what has been termed ‘ecological fallacy’ [40] –the erroneous imputation of association at the individual level based on an observed association at the group or community level. An example of ecological fallacy is the conclusion that, since a higher than expected proportion of people who develop schizophrenia are born in winter months, winter birth is a risk factor for schizophrenia.

Intervention studies

Epidemiological strategies can be applied to evaluate primary or secondary prevention measures and interventions at population level. Classic examples include the prevention of pellagra and its associated psychosis [8] and, more recently, the prevention of neural tube defects by folate supplementation [42]. However, the prevention of common, multifactorial mental disorders, such as depression, anxiety disorders and addictions, as well as low-prevalence debilitating disorders such as schizophrenia, bipolar affective disorder, or dementia, is still in its infancy and lacks an adequate knowledge base. A comprehensive review of current preventive programmes targeting problems ranging from child and adolescent maladjustment to adult depression and cognitive deterioration in the elderly concluded that ‘as yet, there is no evidence that preventive interventions reduce the incidence of mental disorders’ [43].

Commonly used measures of morbidity and risk

(see also Appendix I.)

Prevalence

Prevalence estimates the number of cases (per 1000 persons at risk) present in a population at a given point in time or over a defined period. Point prevalence refers to ‘active’ (i.e. symptomatic) cases at the point of survey. Since asymptomatic cases in remission are likely to be missed in a point prevalence survey, the assessment of the present mental state needs to be supplemented with appropriate inquiry about past episodes of the disorder. This will yield a lifetime prevalence index, also called proportion of survivors affected (PSA). In disorders with a typically chronic course, such as schizophrenia, the values of point and lifetime prevalence will be closer to one another than in remitting illnesses such as recurrent affective disorders, anxiety disorders and many of the substance use disorders. Another measure is period prevalence, such as the number of cases per 1000 population which are active during a specified period (usually six months or a year).

Incidence

Whereas prevalence is a proportion, incidence is a rate, (estimating the number of disease onsets per unit of population per unit of time). Numerically, the incidence rate equals the average rate of onsets of ‘caseness’ over a specified time interval (e.g. if a population of 50 000 persons at risk is monitored for first onsets of psychosis over a 5-year period, and 75 such cases are actually observed, the numerator would be 75 and the denominator 50 000 · 5 years = 250 000 person-years. The reported incidence rate would then be 0.3 per 1000 person-years). It should be noted that the population at risk equals the total population less: (a) those who already have developed the disease and (b) those whose current age is below or above the age interval within which onsets of the disease normally occur. In many respects, the incidence rate is of greater interest than the prevalence of a disorder since it represents the primary morbidity with less distortion resulting from mortality, migration and accrued comorbidity.

Being temporally closer to the point of action of various risk or precipitating factors, incidence also allows a better ‘take’ on disease associations that may be aetiologically significant. However, the estimation of incidence depends critically on the capacity to accurately pinpoint disease onset or inception, which may not always be the case in disorders with a long prodromal period and a fuzzy boundary between premorbid state and onset of symptoms, such as schizophrenia. Since the true onset of the cerebral dysfunction or biochemical lesion that underlies schizophrenia is at present impossible to determine, a strategy which, as a minimum, would ensure consistency is to define onset operationally –either as the point at which the disorder becomes diagnosable according to specified criteria or as the point at which any ‘helping agency’ (a psychiatric or general medical service, or a traditional practitioner) is contacted by a symptomatic individual for the first time. The latter strategy was used in the WHO 10-country study of schizophrenia to generate comparative incidence data for different populations using identical case finding protocols in 12 geographical areas in developed and developing countries [44].

Morbid risk (disease expectancy)

This is the probability that an individual born into a particular population will develop the disease if he/she survives through the entire period of risk for that disease. If the age and sex-specific incidence rates are known, disease expectancy is estimated directly by summing up the rates across the age groups within the period of risk (under the assumption that the age-specific incidence rates are constant over time). Indirectly, disease expectancy can be estimated from the number of prevalent cases (numerator) and an age-corrected population denominator where: (a) individuals not yet in the risk age period and (b) 50% of the asymptomatic individuals within the risk age period are subtracted from the total population number. This corrected denominator is also referred to as the Bezugsziffer (BZ). The formula is known as Weinberg's abridged method [45].

Relative risk

Relative risk (RR) is the ratio of the occurrence of a given outcome (e.g. disease, death) in persons exposed to a risk factor compared with the risk among the unexposed. When two cumulative incidence rates are compared (i.e. incidence among those exposed versus incidence among those unexposed), RR is synonymous with the rate ratio.

Odds ratio

The odds ratio (OR), or relative odds, is similar to RR but easier to derive. It gives a good approximation to RR, especially when rare outcomes are being estimated. It represents the ratio of the odds of a given outcome in exposed persons relative to the odds of that outcome among unexposed persons. Both the OR and RR are probabilistic estimates, and are usually quoted together with their confidence intervals, or CI. The latter are influenced by the size of the sample, as well as by the size of the observed effect. For example, the RR of developing adult schizophrenia for individuals who had experienced birth complications with hypoxia may be 2.1, with CI 0.87–3.65. Although the RR suggests a twofold increase of risk, this indicates, at best, a trend. It should not be interpreted as ‘significant’, since the confidence interval ‘straddles’ unity (the baseline risk of schizophrenia for people unexposed to such birth complications). On the other hand, the same RR of 2.1 will be ‘significant’ if the CI is, say, 1.32–3.12 –which may be the case with a larger sample. Both RR and OR are frequently used measures of outcome in case-control studies.

Attributable risk

This is also known as population attributable risk or aetiological fraction. Attributable risk (AR) estimates the proportion of cases in an exposed population that can be attributed to that exposure. Depending on the population frequencies of the exposure and the outcome, a risk factor associated with low relative risk may account for a substantial proportion of cases in the population, or vice versa. AR can be a valuable indicator of the net population impact of an exposure only weakly associated with morbidity when the disorder is common, or of an exposure to a strong risk factor associated with a relatively rare disorder. AR can therefore be useful in planning and policy decisions since it estimates the amount of reduced morbidity in the population if exposure to a given risk factor could be prevented.

Statistical aspects of study design and analysis

Bias and confounding

The business of epidemiological research can be described as a persistent effort to detect, correct or prevent distortions that may invalidate estimates, comparisons and inferences. Such distortions may arise at every stage of an epidemiological investigation, because of compromised comparability of groups, missing or incorrect information, or unaccounted for extraneous influences on the outcome variables. The 3 typical classes of bias occurring in epidemiological studies are those of selection, information and confounding [32].

Selection bias may occur in case-control studies because of inappropriate selection of controls (see above); or in surveys because of a restriction of case recruitment to unrepresentative settings, such as hospitals (this is known as Berkson's bias).

Information bias most frequently arise out of missing data (non-response) that affect part of the studied sample; recall bias (e.g. mothers of subjects with schizophrenia may over-report pregnancy complications, due to the so-called ‘effort after meaning’); or measurement and diagnostic error. Confounding occurs when a variable that has an independent effect on the outcome is differentially represented in the groups under comparison, for example, the age at onset of schizophrenia is influenced both by sex (earlier in males) and by marital status (later in those living with a partner). Failure to account for the effect of marital status results in an inflated estimate of malefemale differences in age at onset [46]. All three kinds of bias can be avoided or reduced at the stage of designing the study, as well as at the stage of data analysis, by employing techniques such as stratification or multivariate regression analysis.

Sampling

Since complete enumeration and assessment of all members of a population is rarely possible (except in small, circumscribed communities), epidemiological surveys must rely on population samples that are representative of the total, or target, population (such samples are also called probability samples, in the sense that every individual in the target population has an equal chance of being selected). The sampling protocol is therefore a critical aspect of epidemiological study design. There are many sampling methods, varying by complexity, practical feasibility and cost. The sampling frame is any list or register that can provide access to the totality (or great majority) of the target population, (e.g. electoral tract lists, telephone directories and birth or morbidity registers). The ideal (but rarely practicable) method is the simple random sampling. It presupposes the availability of a complete and accurate listing of all the members of a population and entails a selection of individuals according to a set of random numbers. The advantage of the method is that a relatively small number of persons selected in this way would be statistically representative of the target population at the level of gross characteristics, such as diagnoses of common disorders and their distribution by age and sex. However, its disadvantage is that estimates of the frequency of low-prevalence disorders and their descriptive features are unstable and the generalization of survey findings on such rare disorders to the population may be unreliable. Alternative strategies to use with low-prevalence disorders are: (i) stratified random sampling; (ii) systematic sampling; and (iii) cluster sampling. In stratified random sampling, the sampling frame is a priori partitioned into strata of interest, such as inpatients/outpatients; particular age or ethnic groups; or any other attribute, depending on the aim of the survey. Simple random samples are then drawn within each stratum. Systematic sampling involves drawing every n th individual (or unit, such as a household) in accordance with a predefined sampling step or interval, determined in a way to ensure a required sample size. Cluster sampling usually involves two stages: (i) drawing a random sample from a list of aggregate sampling units, such as hospitals, clinics, general practices, etc.; and (ii) applying within each selected aggregate unit either a simple random, or a systematic sampling procedure, until the desired sample size is reached. Whichever method is chosen, its accuracy in the sense of selecting a representative sample needs to be evaluated by statistical modelling with reference to simple random sampling from the same population.

Standardization of prevalence and incidence rates

Direct comparison of crude prevalence or incidence rates across populations, groups, or over time, can be misleading since factors associated with psychiatric morbidity –such as, age, sex, or socioeconomic status may be differently distributed in the demographic structure of the populations to be compared. Appropriate adjustment of the rates is achieved through methods known as direct and indirect standardization. The basic feature of standardization is the introduction of a standard population with a fixed age structure. The rates obtained for any special population (e.g. a study sample) are then adjusted for the discrepancies in age structure between the standard and that special population [47].

Direct standardization

In this procedure, each ageand sex-specific rate of an outcome (e.g. incidence of depression) obtained from a study population is multiplied by the fraction or percentage of that sex- and agegroup within the standard population (which could be the general population of the country or area) and the products are summed up across all age groups to yield a standard population-weighted average of the specific rates. Thus, the result of direct standardization tells us what the crude rate of the outcome would have been in the study population if the latter had the same structure as the general population. To apply this method, the demographic structure of the population must be known from census data.

Indirect standardization

This method is used to compare rates between a special population and a reference population when specific rates for the special population are not available or show large fluctuations (however, the age- and sex-specific rates for the reference population must be known). Essentially, indirect standardization compares the numbers of observed and expected outcomes by extrapolating the rates for the reference population to the special population. For example, if the reference incidence rate of schizophrenia in age group 15–24 is 0.5 per 1000, the expected annual number of new cases in an area where the population in that age group is 17 500 will be 9; if the actually observed number is 13, the standardized morbidity ratio will be 144 (13/9 · 100), indicating an excess morbidity.

Special applications of epidemiological method

Clinical epidemiology

Although different definitions of clinical epidemiology exist, the one that is gaining ground is ‘the application of epidemiological principles and methods to problems encountered in clinical medicine’ [48]. In other words, epidemiological and biometric concepts such as odds ratio, relative risk, positive/negative predictive value, etc., as well as epidemiological databases, are adapted to assist ‘bedside’ clinical investigation and decision making with regard to the individual patient or clinical groupings of cases. In this sense, clinical epidemiology is very much ‘evidence-based medicine’ in practice.

Genetic and molecular epidemiology

Genetic epidemiology has emerged as a cross-border discipline in the 1970s and is currently viewed as ‘the study of the role of genetic factors and their interaction with environmental factors in the occurrence of disease in human populations’ [49]. Since the majority of the psychiatric disorders are likely to be genetically complex (i.e. they segregate in a non-Mendelian fashion; involve multiple predisposing genes of small to moderate effect interacting among themselves as well as with the environment; and often exhibit genetic heterogeneity in the sense of different sets of genes being involved in causing the disease in different pedigrees or different populations), genetic epidemiology in psychiatry has a rich but daunting research agenda [50]. It uses two main types of research strategy: (i) population-based, such as, association studies comparing marker allele or ‘candidate gene’ distributions in samples of unrelated affected individuals and samples of unrelated controls (these are essentially case-control studies); and (ii) family based, in which the transmission of a disease phenotype or trait in a pedigree is examined for a linked transmission of a genetic marker (linkage studies). With major advances in the complete sequencing of the human genome and the recent innovations in high-throughput molecular research technologies, the prospect of detecting and cloning, within the next decade, genes contributing to the major psychiatric disorders appears realistic, though the unravelling of their function and the molecular pathways to disease is likely to take much longer. (An article on genetic methods is planned for later in this series.)

Molecular epidemiology refers to the utilization of genetic markers in large, population-based studies with the aim of investigating associations between DNA polymorphisms and behavioural phenotypes defined by psychiatric diagnoses or quantitative traits such as scores on personality factors or neurocognitive tasks [51, 52]. The term ‘human genome epidemiology’ has recently been proposed with reference to the wider implications of the novel concepts and technologies arising out of the Human Genome Project [58].

Conclusion

This brief overview suggests that psychiatric epidemiology is a vibrant and growing field of inquiry which has adopted, and is adapting a rich armamentarium of research methods to the study of complex mental disorders. Its future success is likely to depend critically on the ability to incorporate the conceptual and methodological advances of molecular biology and neuroscience into the population-based and social framework that is the hallmark of epidemiology as a basic science of public health.

Footnotes

Appendix I

Attributable risk (AR) The proportion of cases of a disorder in a population that can be attributed to the effects of exposure to a specified risk factor.

Incidence The rate of occurrence of new cases of a disorder within a defined population at risk per unit of time (usually reported as number of new cases per 1000 population per year).

Morbidity A general term for measures estimating the spread of a specified disorder (or group of disorders) in a population.

Morbid risk (disease expectancy, lifetime risk) The probability that a randomly selected member of a defined population will develop a specified disorder if he/she survives to a specified age or over the entire period of risk.

Prevalence The proportion (percentage) of ‘active’ (symptomatic) cases of the disorder per 1000 population at risk at a given point in time (point prevalence) or during a specified period (period prevalence). The ‘at risk’ fraction of the population is usually defined in terms of age and gender (e.g. people over the age 60 for the prevalence of dementia, or women of childbearing age for the prevalence of postpartum depression).

Relative risk (RR) The ratio between the rate of occurrence of a disorder in a population group exposed to a specified risk factor and the rate of occurrence of the same disorder in the population unexposed to the risk factor.

Reliability In psychometric theory, reliability is the proportion of variance on a measure (obtained by using a test, a questionnaire or an interview) that is true variance. In practice, reliability is usually estimated by the proxy measures of agreement between independent observers or raters (interrater reliability) and of reproducibility of ratings made by the same observer or rater (test–retest reliability). The kappa coefficient estimates the ratio between actual agreement and the agreement expected to occur by chance.

Screening The procedure for sorting out apparently well persons who probably have a disease from those who probably do not, by the application of tests, examinations or other tools which can be applied rapidly. Screening is not a diagnostic procedure but a prediagnostic filtering process.

Sensitivity The capacity of a screening or diagnostic procedure to identify correctly the cases who have the disorder of interest. The measure of sensitivity is the proportion of true cases missed, i.e. the rate of false negative identifications.

Specificity The capacity of a screening or diagnostic procedure to identify correctly the cases who do not have the disorder of interest. The measure of sensitivity is the proportion of non-cases incorrectly classified as cases, i.e. the rate of false positive identifications.

Survey A general term for a range of epidemiological methods and study designs aiming to identify and enumerate all cases of a specified disorder (or a group of disorders) within a given population or setting.

Validity The extent to which a measuring tool (e.g. a screening or diagnostic procedure) achieves what it is supposed to measure. Validity can be estimated by the extent to which the measuring tool: (a) conforms to agreed concepts about the nature of the object to be measured (content and construct validity); (b) concurs with measurements of the same object made with another, well established tool (concurrent validity); (c) correctly predicts attributes of the object to be measured (predictive validity).

Appendix II

References

Susser

. Causal thinking in the health sciences. Concepts and strategies in epidemiology. Oxford University Press, New York 1973.

Jones

Cannon

The new epidemiology of schizophrenia

Psychiatric Clinics of North America 1998 21: 1–25.

Marmot

M G

Epidemiology and the art of the soluble

Lancet 1986 338: 897–900.

Durkheim

. The rules of sociological method. Free Press, New York 1986.

Koch

Ueber bakteriologische Forschung

Verhandlungen des X. internationalen medicinischen Congresses Berlin, 4–9 August 1890. Hirschwald, Berlin 1891; 35–47.

Morris

J N

. Uses of epidemiology, 3rd edn. Churchill Livingstone, Edinburgh 1975.

Koller

Beitrag zur Erblichkeitsstatistik der Geisteskranken im Canton Zürich; Vergleichung derselben mit der erblichen Belastung gesunder Menschen durch Geistesstörungen u. dergl

Archiv für Psychiatrie 1895 27: 269–294.

Goldberger

Waring

C H

Willets

D G

The prevention of pellagra. A test of diet among institutional inmates

Public Health Reports 1915 30: 3117–3131.

Essen-Möller

Larsson

Uddenberg

C E

White

Individual traits and morbidity in a Swedish rural population

Acta Psychiatrica et Neurologica Scandinavica 1956, Supplement: 100.

10.

Helgason

Magnusson

The first 80 years of life. A psychiatric epidemiological study

Acta Psychiatrica Scandinavica 1989 79 (Supplement 348) 85–94.

11.

Dohrenwend

B P

Psychology, psychologists, and psychiatric epidemiology

Acta Psychiatrica Scandinavica 1994 90 (Supplement 385) 13–20.

12.

Robins

L N

Regier

D A

. Psychiatric disorders in America. The epidemiologic catchment area study, Robins

L N

Regier

D A

. Free Press, New York 1991.

13.

Kessler

R C

McGonagle

K A

Zhao

Lifetime and 12-month prevalence of DSM-IIIR psychiatric disorders in the United States

Archives of General Psychiatry 1994 51: 8–19.

14.

Bebbington

Brugha

Meltzer

Psychiatric disorder and dysfunction in the UK National Survey of Psychiatric Morbidity

Social Psychiatry and Psychiatric Epidemiology 2000 35: 191–197.

15.

Andrews

Henderson

Hall

Prevalence, comorbidity, disability and service utilisation. Overview of the Australian National Mental Health Survey

British Journal of Psychiatry 2001 178: 145–153.

16.

Rosenthal

The offspring of schizophrenic couples

Journal of Psychiatric Research 1966 4: 169–188.

17.

Charlton

B G

Attribution of causation in epidemiology: chain or mosaic?

Journal of Clinical Epidemiology 1996 49: 105–107.

18.

Ødegaard

The incidence of mental diseases as measured by census investigations versus admission statistics

Psychiatric Quarterly 1952 26: 212–218.

19.

Harrison

Cooper

J E

Gancarczyk

Changes in the administrative incidence of schizophrenia

British Journal of Psychiatry 1991 159: 811–816.

20.

Shepherd

Cooper

Brown

A C

Kalton

. Psychiatric illness in general practice, 2nd edn. Oxford University Press, Oxford 1981.

21.

Last

J M

. A dictionary of epidemiology, Last

J M

. Oxford University Press, Oxford 2001.

22.

Derogatis

L R

Dellapietra

Kilroy

Screening for psychiatric disorder in medical populations

Research designs and methods in psychiatry, Fava

Rosenbaum

J F

. Elsevier, Amsterdam 1992; 145–170.

23.

Goldberg

. The detection of psychiatric illness by questionnaire. Oxford University Press, Oxford 1972.

24.

Wilson

J M

Junger

Principles and practices of screening for diseases

Public Health Papers 34. World Health Organization, Geneva 1968.

25.

Cooper

J E

Kendell

R E

Gurland

B J

Sharpe

Copeland

J RM

Simon

. Psychiatric diagnosis in New York and London. Oxford University Press, London 1972.

26.

Robins

L N

Helzer

J E

Croughan

Ratcliff

K S

National Institute of Mental Health Diagnostic Interview Schedule. Its history, characteristics, and validity

Archives of General Psychiatry 1981 38: 381–389.

27.

Robins

L N

Wing

Wittchen

H U

The Composite International Interview: an epidemiologic instrument suitable for use in conjunction with different diagnostic systems and in different cultures

Archives of General Psychiatry 1988 45: 1069–1077.

28.

Wing

J K

Cooper

J E

Sartorius

. The measurement and classification of psychiatric symptoms. Cambridge University Press, Cambridge 1974.

29.

Wing

J K

Babor

Brugha

SCAN. Schedules for clinical assessment in neuropsychiatry

Archives of General Psychiatry 1990 47: 589–593.

30.

Wing

J K

Sartorius

Üstün

T B

. Diagnosis and clinical measurement in psychiatry. A reference manual for SCAN, Wing

J K

Sartorius

Üstün

T B

. Cambridge University Press, Cambridge 1998.

31.

Kendler

K S

Gallagher

T J

Abelson

J M

Kessler

R C

Lifetime prevalence, demographic risk factors, and diagnostic validity of nonaffective psychosis as assessed in a US community sample. The National Comorbidity Survey

Archives of General Psychiatry 1996 1996 53: 1022–1031.

32.

Tsuang

M T

Tohen

Zahner

G EP

. Textbook in psychiatric epidemiology. Wiley, New York 1995.

33.

Klemperer

Zur Belastungsstatistik der Durchschnittsbevölkerung. Psychosehäufigkeit unter 1000 stichprobemässig ausgelesenen Probanden

Zeitschrift für die Gesamte Neurologie und Psychiatrie 1933 146: 277–316.

34.

Fremming

K H

. Sygdomsrisikoen for sindslidelser og andre sjaelige abnormtilstande i den Danske gennemshitbefolkning. Paa grundlag af en katamnestisk underøgelsse af 5500 personer født, 18831887. Munksgaard, Copenhagen 1947, [In Danish].

35.

Helgason

Epidemiology of mental disorders in Iceland

Acta Psychiatrica Scandinavica Supplement 1964 40 (Suppl. 173).

36.

Torrey

E F

Miller

Rawlings

Yolken

R H

Seasonality of births in schizophrenia and bipolar disorder: a review of the literature

Schizophrenia Research 1997 28: 1–38.

37.

Mednick

S A

Machon

R A

Huttunen

M O

Bonett

Adult schizophrenia following prenatal exposure to an influenza epidemic

Archives of General Psychiatry 1988 45: 189–192.

38.

Eaton

W W

Mortensen

P B

Frydenberg

Obstetric factors, urbanization and psychosis

Schizophrenia Research 2000 43: 117–123.

39.

Gunnell

Lopatatzidis

Dorling

Wehner

Southall

Frankel

Suicide and unemployment in young people. Analysis of trends in England and Wales, 1921–95

British Journal of Psychiatry 1999 175: 263–270.

40.

Greenland

Robins

Ecologic studies –biases, misconceptions and counter-examples

American Journal of Epidemiology 1994 139: 747–771.

41.

Johnson

Statistical methods and clinical trials

Research methods in psychiatry. A beginner's guide, Freeman

Tyrer

. Gaskell, London 1992; 24–61.

42.

Medical Research Council

Prevention of neural tube defects: results of the Medical Research Council vitamin study

Lancet 1991 338: 131–137.

43.

Committee on Prevention of Mental Disorders . Reducing risks for mental disorders, Mrazek

P J

Haggerty

R J

. National Academy Press, Washington, DC 1994.

44.

Jablensky

Sartorius

Ernberg

Schizophrenia: manifestations, incidence and course in different cultures. A World Health Organization ten-country study

Psychological Medicine 1992 20: 1–97, Monograph Supplement.

45.

Weinberg

Methoden und Technik der Statistik mit besonderer Berücksichtigung der Sozialbiologie

Handbuch der sozialen hygiene und gesundheitsfürsorge. Springer, Berlin 1925, Band I.

46.

Jablensky

Cole

S W

Is the earlier age at onset of schizophrenia in males a confounded finding? Results from a cross-cultural investigation

British Journal of Psychiatry 1997 170: 234–240.

47.

Armitage

. Statistical methods in medical research. Blackwell, Oxford 1971.

48.

Feinstein

A R

. Clinical epidemiology. The architecture of clinical research. Saunders, Philadelphia 1985.

49.

Khoury

M J

Beaty

T H

Cohen

B H

. Fundamentals of genetic epidemiology. Oxford University Press, New York 1993.

50.

Kendler

K S

Genetic epidemiology in psychiatry

Archives of General Psychiatry 1995 52: 895–899.

51.

Ambrosone

C B

Kadlubar

F F

Toward an integrated approach to molecular epidemiology

American Journal of Epidemiology 1997 146: 912–918.

52.

Henderson

A S

Blackwood

D HR

Molecular genetics in psychiatric epidemiology: the promise and challenge

Psychological Medicine 1999 29: 1265–1271.

53.

Derogatis

L R

. The SCL-90-R. Clinical Psychometric Research, Baltimore 1975.

54.

Derogatis

L R

Melisaratos

The Brief Symptom Inventory: an introductory report

Psychological Medicine 1983 13: 595–605.

55.

Brugha

T S

Nienhuis

Bagchi

Smith

Meltzer

The survey form of SCAN: the feasibility of using experienced lay survey interviewers to administer a semi-structured systematic clinical assessment of psychotic and non-psychotic disorders

Psychological Medicine 1999 29: 703–711.

56.

Jablensky

McGrath

Herrman

. People with psychotic illnesses. National Survey of Mental Health and Wellbeing. Report 3. Commonwealth Department of Health and Aged Care, Canberra 1999.

57.

McGuffin

Farmer

Harvey

A polydiagnostic application of operational criteria in studies of psychotic illness. Development and reliability of the OPCRIT system

Archives of General Psychiatry 1991 48: 764–770.

58.

Khoury

M J

Human genome epidemiology: translating advances in human genetics into population-based data for medicine and public health

Genetics in Medicine 1999 1: 71–73.

59.

Baer

Jacobs

D G

Meszler-Reizes

Development of a brief screening instrument: the HANDS

Psychotherapy Psychosomatics 2000 69: 35–41.

Research Methods in Psychiatric Epidemiology: An Overview ∗

Abstract

Keywords

Psychiatric epidemiology

Defining the unit of observation

Case definition and case identification

Cases in treatment contact

Case registers

Case finding in the community

Screening for psychiatric disorders in the community

Diagnosis and classification

Diagnostic instruments

Types of study design

Cross-sectional surveys

Prospective surveys of incidence

Longitudinal studies

Case-control studies

Ecological studies

Intervention studies

Commonly used measures of morbidity and risk

(see also Appendix I.)

Prevalence

Incidence

Morbid risk (disease expectancy)

Relative risk

Odds ratio

Attributable risk

Statistical aspects of study design and analysis

Bias and confounding

Sampling

Standardization of prevalence and incidence rates

Direct standardization

Indirect standardization

Special applications of epidemiological method

Clinical epidemiology

Genetic and molecular epidemiology

Conclusion

Footnotes

Appendix I

Appendix II

References