5.1 Etiologic research
Hypotheses are nets: only he who casts will catch.
Novalis
Hypothesis statement
The goal of analytic epidemiology is to clarify causal relations between various determinants (“exposures”) and health outcomes. Advancement in knowledge is an ongoing process that progresses, often gradually, in overlapping stages. An epidemiologic issue may be brought to bear by an intriguing case report. This may be followed by descriptive epidemiologic studies. As hypotheses are generated and refined, more detailed studies follow, perhaps culminating in an epidemiologic experiment. In all instances, the epidemiologist pursues methods that are most advantageous for revealing causal relations.
Not to be overlooked in this process is the need to have sharply focused research questions and hypotheses. The study must then be designed in a manner that will connect the data to these questions and hypotheses. Hypotheses must therefore clearly define:
Let us consider the above list of study design elements (1–8) through a hypothetical example. Use of combination oral contraceptives (COCs) is a known risk factor for venous thromboembolism (VTE). COCs are composed on an estrogen component and progestin component. Let us test the hypothesis that the estrogen dose in COCs is related to VTE risk. A study to address these factors could be set up as follows:
As data emerge from a study, and understandings increase from other sources, foreseeable facets of the study are adjusted. As new questions arise, additional hypotheses are articulated to guide further study.
Variables
Our goal is to make biobehavioral and contextual inferences about the effects of health determinants (“exposures”) on health outcomes (“diseases”). In generic terms we ask: “Is there a causal relation between the explanatory exposure E and disease outcome D?”
Exposure E in the study may represent any explanatory physiological factor, personal attribute, environmental exposure, social determinant, or medical intervention thought to influence health. This is sometimes referred to as the independent variable in a statistical analysis.
Disease D may represent any disease, illness, injury, response, or study endpoint. This is often referred to as the dependent variable in statistical analyses.
In addition to exposure E and health-outcome D, the study addresses other factors associated with exposure E and health outcome D. These additional factors are referred to as potential confounders, control variables, extraneous variables, or cofactors. Let us refer to co-variables as C1, C2, and so on.
The objective of the analytic epidemiologic research is to determine the effects E on D while accounting for the contributions of C1, C2, … Ck.
For the hypothetical study addressed in Illustrative Example 5.1, these variables correspond to:
Data
The data that constitute variables E, D, C1, C2, …, Ck are derived from a variety of sources. Examples of data sources include interviews with study subjects, self-administered questionnaires, employment records, environmental records, health care records, social services records, physical examination, examination of biological specimens, and various types of diagnostic tests. The type of data used in a particular study will depend on the research question being addressed, the cost of obtaining data, the need for confidentiality, the type population being studied, and the available technologies.
Information in medical records is abstracted and coded before analysis. Careful training of record abstractors is essential in order to obtain accurate and uniform information. It is advisable to blind interviewers and medical abstractors to the study hypothesis before data are collected. It is also necessary to obscure or remove sensitive information from medical records in order to blind the abstractors to information that may influence objectivity when assigning codes and to protect the privacy of study subjects. This will prevent conscious and unconscious biases from entering into the abstraction process. When more than one medical record abstractor is involved in a study, it is wise to produce separate analyses for their data to check for consistency of results. Lack of consistency, such as a positive association derived by data from one reviewer but not the other, is cause for concern.
Data collection forms used by interviewers should be brief and simple. The art of asking clearly worded, non-ambiguous, and non-presumptuous questions takes extra thought and planning.a Data collection instruments must be piloted and tested to remove ambiguities and redundancies before being used in the actual study. Completed data forms should be reviewed by a study coordinator before being entered and validated in creating the data files that comprise the database for the study.
5.2 Ethical conduct of studies involving human subjects
Table 5.1 lists three ethical principles for conducting research using human subjects as specified in The Belmont Report (1979). These include respect for persons, beneficence, and justice. As part of the principle of respect for persons, study subjects must freely give their informed consent before participating in a study. This implies that the subjects are given a chance to ask questions, are not coerced, are under no obligation to participate in the study, and may withdraw from the study at any time. A signed statement of consent is required.
Ethical principle | Application |
Respect for persons | Informed consent given freely. Implies ability to comprehend consequences; confidentiality is maintained for private information. |
Beneficence | Risk and benefits are assessed. Benefits can be direct, indirect, collateral, or aspirational. Harms may involve physiological, psychological, or socioeconomic consequences. |
Justice | Selection of study groups should be inclusive, equitable, and avoid exploitation. |
Source: The Belmont Report (1979).
Ethical guidelines are safeguarded by human subjects committees known as institutional review boards (IRBs). IRBs are committees composed of researchers, clinicians, administrators, and laypeople who review the study protocol before the study is begun. Their primary objective is to ensure the ethical treatment of human subjects and to oversee informed consent procedures.
Because experiments involve treatments and interventions, additional ethical constraints are required. To ethically assign treatments, none of the treatments can be known to be superior to any other. Treatments that present special hazards cannot be ethically assigned and, just as importantly, treatments that are believed to be beneficial cannot be ethically withheld. Therefore, a true state of uncertainty or “balanced doubt” about the pros and cons of the intervention must exist before it can be submitted to a trial. This balanced void of knowledge is referred to as equipoise.
Separate from the IRB, studies involving interventions require a Data and Safety Monitoring Board (DSMB). The DSMB is an independent group of outside experts that periodically reviews and evaluates accumulated evidence from the study to monitor its safety and progress. The job of the DSMB is to make recommendations concerning the continuation, modification, or termination of the study.
5.3 Selected study design elements
Let us now address five important elements of epidemiologic study design. These elements are: (a) necessity of including a referent (“control”) group, (b) the distinction between experimental and non-experimental (observational) studies, (c) the unit of observation, (d) the difference between cross-sectional and longitudinal observations, and (e) cohort and case–control samples.
Necessity of a referent (“control”) group
Under all but exceptional circumstances, the only way to demonstrate whether a given exposure is the cause of an outcome is to compare groups to see if different combinations of exposures explain variations in the outcome under a variety of circumstances. Thus, etiologic studies require at least two groups. One group, the index group, is exposed to the factor thought to influence occurrence of the study outcome. The other group, the referent or control group, remains unexposed to provide a reference for comparison.
The effects of the exposure cannot be judged without the benefit of the referent group. Consider an experiment to determine the effectiveness of a treatment. If a certain number of patients recover following treatment, how would we know whether recoveries were due to the treatment or whether recoveries merely represented the spontaneous recovery rate or perhaps some unmeasured factor? Might the treatment have delayed recovery? On the other hand, by administering the treatment to one group of patients while leaving a similar group of patients untreated, observed difference in recovery could then be ascribed to either the treatment, some unnoticed difference in the groups, or to chance. Without the baseline of observation provided by the referent group, it is impossible to determine whether the exposure had a positive effect, a negative effect, or no effect at all.
This same line of reasoning applies to studies that address natural exposures. If a person develops a brain tumor and is a frequent cell phone user, or lives near a toxic waste dump, or is exposed to whatever the media has identified as the hazard of the moment, we might be tempted to attribute their condition to any one of these exposures. Nonetheless, the outcome, however unfortunate, cannot be attributed to any particular exposure without further scrutiny. The question of course is not whether there is an association in the minds of any particular individual. The question is whether the specific exposure contributed to the causal mechanism behind the brain tumor. If the exposure is causal, we would expect an increased occurrence of the outcome in those exposed to the causal factor relative to those who are not exposed, all other things being equal.
Experimental versus observational study designs
The primary way to classify comparative studies in epidemiology is as either experimental or non-experimental (observational). In experimental studies (“trials”), the investigator introduces or withholds an exposure in order to observe its effects. The experimental allocation of the study exposure can be based on chance mechanisms (randomized trials) or on other mechanisms built into the study’s protocol (nonrandomized trials). Randomized designs are superior to nonrandomized trials for reasons that will soon become evident.
In a simple randomized controlled trial, individuals are randomly assigned to either a treatment group or a control group. The treatment group receives the experimental intervention. The control group receives either an inert intervention (placebo) or an alterative active intervention. Study subjects are then followed over time to assess study outcomes.
Because of practical and ethical concerns, however, opportunities for experiments using human subjects are often limited. Thus, most epidemiologic studies are non-experimental. Non-experimental epidemiologic studies are often referred to as observational studies. In contrast to experimental studies, observational studies do not assign treatments to study participants. Instead, subjects are studied under natural circumstances that are thought to be revealing. In a simple observational cohort design, subjects are classified as either “exposed” or “nonexposed” to the study factor of interest and are then followed and assessed for the study outcome. Incidences in the two groups are then compared.b
The Women’s Health Initiative (WHI) was a major 15-year research initiative sponsored by the National Heart, Lung, and Blood Institute. The objective of this program was to address the common causes of death, disability, and poor quality of life in postmenopausal women, with special emphasis on cardiovascular disease, cancer, and osteoporosis. This program included both experimental and observational studies.
Experimental elements of the WHI project were designed to test the effects of postmenopausal hormone therapy, diet modification, and calcium and vitamin D supplements on heart disease, fractures, and breast and colorectal cancer. In the hormone trial, for example, participants were randomly assigned to groups that received either a pill containing estrogen plus progesterone or an identical-looking pill that contained no active ingredients. Incidences of various health outcomes (e.g., coronary disease) were monitored over time in the study subjects.
Observational studies in the WHI program complemented the experimental studies by providing estimates of the extent to which various risk factors predicted heart disease, cancers, fractures, and other adverse health outcomes. Observational studies in the WHI tracked the experience of 93 676 postmenopausal women between the ages of 50 and 79. Women who joined the observational study were not required to take any medication or change their health habits while being monitored.