13 Research Designs in Oral Epidemiology
Numerous books have been written about research design and the characteristics of good research. This is not one of them, but we do set out in this chapter to give a bare-bones presentation of the essentials of valid research reports. This coverage is intended to permit the reader to make good sense of epidemiologic studies, especially clinical trials, and of other studies involving human beings.
Epidemiology is the study of health and disease in populations, and of how these states are influenced by heredity, biology, physical environment, social environment, and ways of living. Some definitions extend the meaning to include the application of this study to control health problems.14 Epidemiologic studies can first be classified either as descriptive, meaning that the data only describe the distribution of a condition in a population and no specific hypothesis is tested, or as analytic, meaning that the data collection and analysis are designed to answer a particular question. Descriptive and most analytic studies are observational; that is, they observe outcomes without intervening to affect them. Analytic studies in epidemiology look at people with and without the disease in question (the effect or outcome) and with and without exposure to the putative influences that may increase the risk of disease (the exposure).
The clinical trial, an experimental design to test the efficacy of a preventive-control agent or treatment procedure in humans, is an aspect of analytic epidemiology sometimes classified separately as experimental epidemiology. The clinical trial is also an interventional design; that is, something is intervening in the natural history of a condition in an effort to give a beneficial outcome.
Good research demands careful, sometimes exhaustive planning. Every study, no matter how modest, needs a protocol, which is a written plan encompassing the purpose and the detailed operation of the study. The essential elements of a protocol are listed in Box 13-1. A protocol demands careful thinking through of a project, a process that aids its design and also helps the researchers to anticipate potential problems. It also simplifies the writing of a final report, because the protocol forms the basis of the report.
Causality, meaning that a certain exposure results in a particular outcome, can only be demonstrated unequivocally within the experimental study design of a clinical trial (discussed in a later section). Because clinical trials cannot be conducted on many topics for both practical and ethical reasons, causality in the study of disease usually must be imputed from studies with nonexperimental designs. Criteria by which a conclusion of causality can be reached from nonexperimental epidemiologic analyses, known as the Bradford Hill Criteria after the British statistician who developed them, were suggested in 19653 and have become well accepted over the years. The original Bradford Hill criteria have evolved over the years and are summarized in Box 13-2 as they are usually understood today.
Analytic studies, in contrast to descriptive studies, have the general aim of seeking out cause-and-effect associations. Because analytic observational studies cannot directly address cause and effect, however, they seek to quantify the degree of disease risk in specified circumstances. Risk is the probability that a specified event will occur, for example, that an individual will exhibit a disease or die within a stated period of time or by a particular age.14
The criteria listed in Box 13-2 are not all imperative; in fact the only absolute among them is the time sequence—exposure must precede outcome. In many exposure-outcome relationships there are factors that both researchers and clinicians think play a role in causing a disease but which do not satisfy all these criteria. If researchers had to proceed with just the dichotomous judgment of whether a factor is or is not involved in causing a disease, our knowledge of disease causation and development would be seriously hindered. The concept of a risk factor permits quantification of the degree of importance of a particular factor in the development of a disease; some causal factors are more important than others.
A risk factor is broadly defined as an attribute or exposure that is known, from epidemiologic evidence, to be associated with a health condition considered important to prevent.14 Although the term risk factor is applied loosely in the literature, modern usage ascribes a strong causal role to a risk factor: it is either part of the causal chain or is something that brings a person into contact with the causal chain. (An example of the latter situation is an occupation that requires handling toxic materials. The occupation itself is not a risk factor for toxicity, but because it brings a person into contact with toxic materials, which are the risk factors, it does increase the chance of disease.) We prefer the following more complete definition of risk factor:
An environmental, behavioral, or biologic factor confirmed by temporal sequence, usually in longitudinal studies, which if present directly increases the probability of a disease occurring, and if absent or removed reduces the probability. Risk factors are part of the causal chain, or expose the host to the causal chain. Once disease occurs, removal of the risk factor may not result in a cure.2
Part of the concept of a risk factor is that it can be modified: people can stop smoking, lose weight, change to a healthier diet, improve their oral hygiene. The identification of risk factors for a disease then allows the potential for prevention by removing or modifying the risk factors. As examples, smoking is a risk factor for lung cancer; poor oral hygiene is a risk factor for gingivitis. In both instances, removing or modifying the risk factor reduces the risk of disease, although neither exposure is a sole cause of the disease.
As stated in this definition, a risk factor for a disease must be demonstrated as such longitudinally. This is because confirming the necessary time sequence—that is, ensuring that exposure to the risk factor occurs before the disease outcome—can nearly always be demonstrated only longitudinally. This time sequence is one of three criteria that must be met before we can suggest that a particular exposure is a risk factor for a particular disease (Box 13-3 lists the criteria for identifying a risk factor. Box 13-2 gives the criteria for imputing a cause.) The ultimate test of a risk factor is that, if exposure to it is reduced, the risk of subsequent disease diminishes. As an example, quitting smoking reduces the risk of a heart attack.
What if a suspected risk factor cannot be confirmed as such because the necessary longitudinal studies are impractical or unethical? The factor may be classed as a risk indicator, defined as a factor shown to be associated with a disease in cross-sectional data and assumed, on theoretical grounds, to play some causal role.15 Research experience has shown that risk indicators which emerge from cross-sectional studies can disappear in a more rigorous longitudinal analysis, so without longitudinal assessment it cannot be known whether a risk indicator is or is not a true risk factor. As would be expected, many more risk indicators for oral diseases than true risk factors have been identified.
A risk marker is an attribute or exposure that is associated with the increased probability of disease although it is not considered part of the causal chain. A risk marker can also be called a risk predictor when included in predictive statistical models. Some immutable characteristics of a person or group—namely, age, gender, and race or ethnicity—can influence disease occurrence, progression, or outcome. These attributes do not fit the concept of a risk factor because they are not modifiable. Although they can be useful in statistical models whose purpose is to predict disease occurrence, they clearly are of no use when considering disease prevention based on control of risk factors. The literature is unfortunately muddled about what to call these attributes. We suggest the term demographic risk factors to refer to these immutable influences. In addition to age, gender, and race or ethnicity, socioeconomic status is usually considered a demographic risk factor. Theoretically it can be modified, but in practice this is hard to do.
The collection of data to be used for descriptive purposes is commonly called a survey (see Chapter 4). Surveys record the prevalence of various conditions, meaning the number or proportion of persons in the population who exhibit a condition at any given time. (Data from national surveys in the United States are reported in Chapters 19–22 Chapter 19Chapter 20Chapter 21Chapter 22.) Although surveys are important in assessing trends in health and disease, the field of oral epidemiology encompasses much more than surveys.
At the most basic level, epidemiologic study designs are cross-sectional or longitudinal. A cross-sectional study is one in which both exposure to risk factors and the health outcomes in a group of people who are, or are assumed to be, a sample of a particular population (a “cross section”) are assessed at the same time. A longitudinal study is one in which the same group of people is studied on two or more occasions so that incidence, the change in a condition over time, can be assessed. A survey collects cross-sectional data. Comparison of trends by examining the results of a sequence of surveys, even if the same study protocols are used in all of them, is s/>