Background to the characterization of descriptive studies

We reviewed academic textbooks on epidemiology in the international and national literature; the main criterion was to have them available for in-depth review of chapters on descriptive epidemiology and types of studies. In all, the material that underpins this proposal corresponds to 25 textbooks, of which 19 are foreign and six are Brazilian. The material in question was produced by 27 authors or groups of authors. Two of the books contain two chapters by different authors addressing the theme.

The question of the nomenclature of descriptive epidemiological studies begins with the very recognition of a category that includes such studies. Many authors simply ignore or do not mention any type of epidemiological study other than the analytical type, i.e. a type of study that does not have groups for comparison and which does not test hypotheses.

Figure 1, shows eight foreign textbooks, one of which contains two chapters by different authors, in which there is no detailing of descriptive studies, because absolute priority is given to analytical studies. The books begin with one of the first academic works in the area, namely the text by MacMahon & Plugh (1970),² and end with one of the most complete works in terms of its theoretical reflection and in-depth methodological approach, namely the text by Rothman et al. (2008).³ The texts mentioned only present descriptive epidemiology and differentiate between measures of incidence and prevalence.

In Figure 2, twelve textbooks are presented in which there is some degree of detail on certain descriptive epidemiological studies presented in sections or specific chapters. The titles of these chapters often do not mention descriptive epidemiology or even types of descriptive studies. For example, Gordis (2014)⁴ mentions descriptive studies in the chapters entitled “Natural history of disease: ways of expressing prognosis” and “Assessing preventive and therapeutic measures”. In general, the authors recognize the difference between descriptive and analytical studies, placing greater relevance on the second kind. In applied field research, despite the value placed on descriptive epidemiology, greater relevance is attributed to analytical studies, in the sense of their potential for enlightening the etiology of outbreaks.^5,6 In the majority of cases in which specific study types are referred to, mention is made of prevalence studies referred to as population-based surveys or encuestas or, otherwise, mention is made of case report studies and clinical case series studies. There is frequently inconsistency in the denominations used.

The majority of the Brazilian textbooks on epidemiology, shown in Figure 3, include chapters containing descriptive analysis, always giving emphasis to measurements of disease incidence and prevalence in a given population, according to variations relating to people, place and time. Emphasis is also placed on the form of analysis and data presentation. However, similarly to the books published internationally, their nomenclature does not specifically include descriptive studies. The six authors shown in Figure 3 place emphasis on analytical studies. For example, in the same book, the chapter that deals with study types does not mention descriptive studies,⁷ while a chapter specifically about ecological studies separates “exploratory’ strategies from “analytical” strategies. Those exploratory strategies could be called descriptive strategies.⁸

Figure 4 shows two Brazilian textbooks which make more detailed mention of some descriptive studies. Pereira (1995, p. 278)⁹ characterizes descriptive studies as being those which do not have control groups, thus differentiating them from analytical studies. The author lists nine types of epidemiological studies, the first four of which are descriptive: case study, case series, incidence study and prevalence study (descriptive cross-sectional).

Still in relation to the Brazilian publications, Zanetta (2004)¹⁰ recognizes the existence of descriptive studies in contrast to analytical studies. According to this author, descriptive studies “report prevalence rates or describe a situation that appears to be abnormal”, and they are useful for health service planning and for generating hypotheses “in the experimental field”. When listing the “designs of studies in medicine”, she includes the case report, case series and cross-sectional or prevalence studies which she cites as descriptive studies. The remainder are analytical studies.

Classifying descriptive studies: does it make sense?

When reading textbook chapters, it can be perceived that there are different approaches, ranging from complete omission of descriptive studies to their inclusion. When descriptive studies are included, there are different ways of conceptualizing and classifying them. An initial difficulty is the tendency to call descriptive studies only those studies conducted using secondary macrodemographic data or non-clinical data on open populations.

It is also relevant to differentiate between data collection technique and study type. With this in mind it is important to recall the proposal made by Laurenti et al.¹¹ who consider that “gathering statistical data” is the set of operations that allow collection of data which enable an event to be characterized. From this perspective, collection can be done by using existing recorded data (which the author refers to as data gathering per se); or by using existing but unrecorded data by means of interviews or laboratory tests (which the author refers to as surveys); or with data that does not exist but which can be generated by an intervention, such as adverse vaccination events (which the author refers to as experimental data). We must highlight, however, that types of data gathering or collection do not determine the type of epidemiological study.

In the face of this multiplicity of points of view, the domain in which a study is conducted is relevant for its classification. Initially there is a population or community-based domain, so that the need arises to understand that any classification must take two situations in this domain into account: (i) there are studies conducted based on macro-statistical data relating to large population clusters, commonly secondary data, prepared using individualized numerators using case or death notifications and denominators based on population estimates; and (ii) studies conducted using primary data in various community domains (workplaces, education facilities, leisure facilities, households, crèches, public libraries, trade unions and community organizations, religious institutions, clubs, among others), in which both numerators and denominators have been addressed or recognized, and computed as individuals.

Regardless of this population or community domain, health care institutions must be included as a specific domain. In principle, it is in this domain that “cases” are concentrated and that care delivery, interventions, procedures and outcomes are registered. It must be taken into consideration that many of these institutions do not carry out activities only with patients or people who already have a disease or health condition. For example, several types of consultations (prenatal, control of child growth and development, child care, food intake and nutrition, topical dental fluoride application, adult check-ups), in principle involve working with people from “the community”, who have been directed or led to attending a care facility without, however, being classified a priori as “clinical cases”. Other similar environments can be comprised of blood banks or places where other fluids, blood products and organs are donated. It is also relevant to highlight that studies in which the study population is comprised of people assessed in primary care are very close to the household environment. As such, a specific case could be that of studies conducted in Home Hospitalization Services which, in this sense, have clinical characteristics analogous to hospital inpatient care. It must be pointed out that these clinical domain studies are conducted using individualized finite primary data, used as clearly defined numerators and denominators.

We draw attention to the fact that, both in clinical domain studies and population/community-based studies, external validity, understood as the ability to generalize data from a set of subjects that effectively comprise a given study, is not a determinant of classification.

Proposal for classifying descriptive studies

Considering the need for a satisfactory classification, we decided to propose a taxonomy that makes clear potentialities, limitations and consistency among the objectives proposed and the types of studies conducted.

There are several classification possibilities. If the criterion is observation of a reality or measurement of the effects of an intervention that has been conducted by the people in charge of the research, studies can be either observational studies or intervention studies. If the criterion is presence or absence of follow-up of people or groups, studies will be longitudinal (also called follow-up studies) or cross-sectional studies (at a moment in time). With regard to the unit of analysis corresponding to people or clusters, studies can be on an individual basis or an ecological basis. According to the domain in which the study is conducted, studies can be population/community-based or clinical. This classification is not mutually exclusive, i.e. a study can have individual observation units and be observational, or it can be individual and an intervention study, or longitudinal and observational, for instance. The classification proposal is shown in Figure 5.

1.1 Case report

In the clinical domain, an initial objective of researchers may be to report the existence of a case or a small number of cases. The research question could be the existence or not of a given health condition or disease in a country or community, the characteristics of a limited number of clinical cases, or the way in which the clinical history of the case occurred, is suspected, diagnostic impression and confirmation. It is indeed in this way that for centuries accumulated knowledge about nosological entities, their characteristics and responses to treatment has been formed. In this case, a study can be limited to describing in detail the manifestations of the disease, data on the main complaint and anamnesis, symptoms reported, clinical signs detected, laboratory test and imaging results and the diagnostic conclusion.

From the epidemiological point of view, its use may be very restricted for measuring frequency in the population or even for characterizing frequency of manifestations or findings. Nevertheless, the usefulness of this type of study in the clinical domain is that it can alert health professionals as to the existence of an event in their midst, for the purposes of diagnosis and differentiation. Reports of this nature serve later for documenting distribution in relation to events that may be cosmopolitan or exist in some regions, even though their occurrence may be rare. A rare event may, however, be the first sign of an epidemic or an emerging disease. There are precedents of descriptions of divergent patterns of development or distribution that have led to the conclusion being reached as to the existence of new events or unprecedented behaviors of existing diseases. As such, these studies are relevant for epidemiological surveillance, since they can reveal preliminary findings of emerging or reemerging diseases that are spreading in new epidemiological scenarios.

Examples of this kind of epidemiological study are case reports on Kaposi’s sarcoma and pneumonia due to Pneumocystis jirovecii, previously known as P. Carinii, which have served as a sentinel event for the existence of AIDS-related imunossupression.¹⁴ Another example can be the way in which new events unknown before this decade behave clinically, such as Chikungunya and Zika in Latin American countries.^15,16 It is important to point out that the most appropriate term is “case report” rather than “case study”, since the latter can correspond to other research situations (both among individuals and among groups of diverse sizes), in the domain of nursing, psychology, social service or sociology.

How many individuals can be included in a case report? There may not be an exact answer. Fletcher et al.¹⁷ considered that there could be ten cases at the most. Above this number, it would be considered to be a case series or a clinical cohort. There really is no clinical basis for saying that above ten cases (e.g. a case report on eleven individuals), randomness reduces sufficiently for one to get a notion of the true frequency of a given symptom, clinical sign, laboratory or imaging finding.

1.2 Case series

Still within the clinical domain, the research question can intend to gain knowledge of the behavior of a nosological entity or disease, its natural history and manifestations, distribution of individuals by sex, age, race/skin color, by etiological agent subtype, times and places of greatest frequency, among other possibilities. This type of study describes case “profile” and can be called a “case series” study. It goes further than straightforward “case reports”, the importance of which has already been highlighted. In this situation, however, a larger quantity of observations is needed, and it can inform the proportion of individuals who present a given symptom, sign or laboratory or imaging characteristic. This type of study is not population-based and is usually carried out in health care services, most frequently in hospitals.

However, a macro dimension of the case series study is represented by the set of “case” notifications held on information systems for compulsorily notifiable diseases or events (e.g. the Notifiable Health Conditions Information System - Sinan). In this case, duly taking sensitivity and underreporting into consideration, one would be using the “census” of detected cases, which would increase external validity. On the other hand, it would be subject to limitations arising from health care biases, usually present in health care facilities, such as access problems and problems with referrals between levels of health care complexity.

From the point of view of epidemiological analysis, data obtained on frequency of the distribution of characteristics among ill people (cases) are specific for this population of “cases”. The measurement used is the proportion of cases, since it is not a question of prevalence in an open community or population, but rather a very specific situation in which the numerator corresponds to cases that have a given characteristic (e.g. cases reporting fever, male cases or cases with a given laboratory finding) and the denominator corresponds to the total number of patients, assuming that data on the clinical evolution of ill people or cases are not being included. An example is the characterization of 87 cases of Zika virus disease in Pernambuco, highlighting the clinical and imaging characteristics of neurological damage.¹⁸

1.3 Descriptive clinical cohort

Another research question situation relates to the clinical evolution of cases. In this situation, the presence of “new events” is documented as metaphenomena beyond the disease itself, such as complications, appearance of side effects of treatment interventions, cure, sequelae, or death. These studies can be called descriptive clinical cohorts or descriptive prognosis studies. It can be seen that these studies can be documenting both the evolution of the “natural history” of a disease, and also the effects of an intervention, i.e. in certain clinical studies conducted with a group of individuals, the latter may be having standardized treatment and their follow-up checks for clinical outcomes, such as cure or adverse events. This does not mean that this type of study can be used to measure the efficacy of the intervention, since comparison groups would be needed for this, as in clinical trials which are a kind of analytical study.

From the epidemiological analysis point of view, the frequency with which these new events appear is calculated, behaving as incidence in relation to the total number of people. These incident events can be positive, such as cure or remission. In cases in which death is the incident event, fatality can be calculated.

In these studies, as in other strategies for following up on people, the calculations described in the preceding paragraph correspond to the closed or fixed cohort strategy. The dynamic or open cohort is an alternative analysis strategy, in which it is possible to calculate incidence density. In this case, the sum of people-time (e.g. people-year) is computed, according to individual contribution to follow-up in the cohort. The numerator continues to be the number of cases or deaths.

It is important to remember that the “new” events documented in this type of study are part of the evolution of the clinical picture, even when under treatment intervention. The research question refers to case evolution. Data on incidence of adverse effects systematized by clinical researchers have been fundamental for establishing the expected frequency of these events. An example is the frequency of COVID-19 events (clinical outcomes) when, besides symptoms and signs, authors documented the incidence of complications that led to hospitalization in intensive care units (ICUs), use of ventilators or death, i.e. fatality, among 1099 patients in China.¹⁹

2.1 Descriptive observational prevalence studies

These are observational studies the design of which answers the research question about the existence of a given characteristic at the time the study is conducted or the participants are approached. They correspond to cross-sectional studies, also known in the literature as surveys, which document events existing at a given time, such as cases of a disease and risk or protective factors. These studies include those which determine case frequencies in the population, both existing cases and new cases, according to the characteristics of people or contextual variables traditionally attributed to individuals (age, sex, ethnic origin, socioeconomic status, occupation, marital status, sexual orientation, habits); characteristics of the places of occurrence (streets, neighborhoods, administrative regions, census tracts, urban or rural areas, municipalities, states, countries); and time of occurrence (hour, day, month, year).

The research question refers to the frequency of a disease at a given time, the frequency of a risk factor or a specific characteristic of this population or community segment. It is assumed that there is a momentary snapshot in time (what social sciences call synchrony). The name we are suggesting, descriptive prevalence study, describes what was done without the need for other explanations. Another possible name is survey (inquérito in Portuguese), despite the possibility of confusing the study type with a technique for acquiring data by asking questions (the act of inquiring). The term is similar to the word encuesta, used in Spanish (encuesta epidemiológica), or enquête in French. These terms are used with the same meaning and with the same possibility of confusion, despite their being widely used for specific purposes (inquérito epidemiológico [epidemiological survey], inquérito nutricional [nutritional survey], inquérito alimentar [food intake survey], inquérito sorológico [serological survey]). Indeed, in some academic domains, the French term has been translated into Portuguese as enquete. Notwithstanding, in French and French Canadian epidemiological literature, this term can also be used as a synonym for “study”, such as enquête analytique and enquête descriptive, for instance.²⁰

In English, as mentioned above, the term survey is used, originally having the connotation of probing or prospection, and the term opinion survey was used a great deal by Paul Lazarsfeld to refer to studies that became very popular in the United States after the Second World War. Other expressions such as “descriptive cross-sectional study” and “descriptive cross-sectional cohort study” can be synonyms. If their descriptive nature is not specified, this can cause confusion, since in the literature they are widely used to designate the corresponding analytical prevalence studies.

From the analysis point of view, descriptive prevalence studies use calculation of prevalence as the measurement of frequency. Their external validity depends on the sampling strategy.

For the purposes of epidemiological, food and nutrition surveillance, in recent decades the Brazilian State has promoted large-scale prevalence studies which are intended to be repeated periodically, such as the Non-Communicable Disease Risk and Protective Factors Surveillance Telephone Survey (Vigitel), the National School Student Health Survey (PeNSE), the National Survey of Access, Use and Promotion of the Rational Use of Medication in Brazil (PNAUM), the National Health Survey (PNS), among others. These studies are conducted based on complex samples and contact with participants by telephone, household interviews or interviews at school. In principle, through their published reports, these investigations provide the possibility of estimating overall prevalence rates and their distribution and, in this sense, they are descriptive, as shown by the periodical publication reporting on the Vigitel survey findings.²¹ They can be used to carry out analytical studies to answer another type of research question which could be solved by testing hypotheses.

2.2 Descriptive observational incidence studies or descriptive cohort studies

These are studies that involve follow-up of population groups in order to investigate the appearance of new outcomes (cases, relapses, deaths or other events), establishing a diachronic dimension analogous to the clinical cohort study.

When the research question refers to the frequency of new events in the “healthy” population that is having follow-up, the term descriptive cohort study is used. In this case, follow-up will determine the frequency of cases of the disease, deaths or other incident events, such as sexual initiation, seroconversion, relapses, among others. Follow-up is done with a population that has defined characteristics.

From the epidemiological analysis point of view, in descriptive cohorts the frequency of the appearance of these new events is calculated, such as incidence in relation to the total number of people in the community effectively having follow-up (cumulative incidence). In cases of death as an incident event, mortality is calculated. In both cases, the denominators express the population “at risk” of the numerator event happening. For both mortality and morbidity, it is possible to measure incidence density, by creating the person-time artifice in the denominator.

An example is a cohort of elementary school students in Thailand, whose experience is followed up on in order to measure dengue incidence.²²

2.3 Descriptive ecological studies

Finally, it is also necessary to consider studies based on aggregated data, rates or proportions calculated for a given population group, in which the numerators correspond to the number of events notified or recorded (deaths, cases of notifiable diseases, other events, accidents, violence), and the denominators are intercensal estimates of the population. In this case the numerators are finite, but the denominators correspond to estimates, so that any comparison becomes difficult, both because of differences existing in the population base - which requires rates to be standardized, by having recourse to other data (age distribution, for instance) - and also due to the difficulty of establishing people’s exposure status, because they are aggregated when they are assessed. The research question would be: what is the frequency (incidence, mortality) of the event in a given population? How did it evolve over the years? As such, aggregation leads us to an ecological approach, frequently based on secondary data. They could also be used to examine, in an ecological manner, the effects of vaccination on the population level, comparing the incidence rate before and after an intervention.²³