«EPIDEMIOLOGY AND BIOSTASTICS: AN INTRODUCTION TO CLINICAL RESEARCH Bryan Kestenbaum, MD MS University of Washington, Seattle TABLE OF CONTENTS. ...»
Clinic-based studies such as these are generally the easiest and least expensive to conduct, because potential study subjects are often readily accessible to the investigators and important data may already be available as part of clinical practice. However, findings from these types of studies tend to be poorly generalizable to other populations. In this case, antibiotic resistance patterns may be specific to the geographic region where the study was conducted, and could be influenced by antibiotic prescription practices of this particular pediatric clinic. Results from this study will apply only to children who live in Minneapolis, and have the same age, socioeconomic background, and practice patterns as the children who attended this particular pediatric clinic.
Moreover, clinic-based study populations tend to be relatively small and therefore highly subject to sampling variation, or random fluctuation in results. Imagine that there were 500 total softtissue infections in this pediatric clinic, and that 50 (10%) were caused by resistant organisms.
Selecting random samples of 30 cases from the total group of 500 would yield a wide range of estimates of the proportion of resistant organisms.
Health network-based study of hip fracture in chronic kidney disease8 Example 2.2.
Study objective: Estimate hip fracture rates in people with and without kidney disease Study findings: Late stage kidney disease associated with 4-fold greater risk of hip fracture Study population: The source population consisted of all male veterans with at least one outpatient primary care or internal medicine subspecialty clinic visit within the Northwest Veterans Integrated Service Network, a collection of eight Veterans Affairs facilities located in Washington State, Idaho, Oregon, and Alaska. Exclusion criteria were prior history of hip fracture, diagnosis of cancer, chronic dialysis or renal transplantation.
Health network-based studies such as these offer an improvement in generalizability compared to clinic-based studies. In this case, the observed association of late stage kidney disease with hip fracture is more broadly applicable to men in multiple geographic locations, and is not limited to the practice patterns of a single clinic. Further, the closed nature of the Veterans Affairs medical system increases the likelihood that hip fractures will be captured in this study population. This more general study population helps to support the hypothesis that late stage kidney disease might play a causal role in the development of hip fracture. It is important however to note that this study population consisted predominantly of older men. Results may not apply to women, who have a considerably higher underlying prevalence of osteoporosis.
Community-based study of lipoprotein a Lp(a) levels and stroke9 Example 2.3.
Study objective: Examine association of Lp(a) with the risk of incident stroke in older adults Study findings: Higher Lp(a) associated with greater stroke risk in men, but not women.
Study population: The Cardiovascular Health Study (CHS) is a community-based study of heart disease and stroke in 5,888 ambulatory adults aged 65 years and older. Participants were recruited from four communities by randomly sampling from age-stratified Medicare eligibility lists in each area.
Subjects were excluded if they were institutionalized, required a proxy to give consent, required a wheelchair, or were receiving treatment for cancer.
Community-based studies such as these are typically the most costly and complex to conduct because they involve leaving the health care system in favor of the community for subject recruitment. The use of a community-based study population yields the most generalizable study findings, because many people in a community never see a doctor, let alone the inside of a hospital. Study findings for Lp(a) are expected to broadly apply to the general population of ambulatory older adults, not just those who are receiving health care.
C. Where to find information about the study population in a clinical research article Description of the study population is usually, but not always, described in the first one or two paragraphs of the methods section of a research article. This paragraph should explicitly define the source population from which study subjects were selected and detail and justify the specific inclusion and exclusion criteria. This information may also be presented in flow chart form.
II. EXPOSURE AND OUTCOMEA. Definition One broad segment of clinical / epidemiological research focuses on the relationship, or association, between an exposure and an outcome of interest. The term ‘exposure’ is carried over from infectious disease epidemiology; however, it is used to describe any factor or characteristic that may explain or predict the presence of an outcome. Examples of exposures include the use of a particular medication, smoking, and blood pressure.
The term outcome refers to the factor that is being predicted. The outcome is often a disease, but can be any clinical characteristic, such as cholesterol level, vaccination status, or medication use.
The distinction between exposure and outcome is highlighted in the following examples.
A study examines whether vaccination against pneumococcal pneumonia is effective at preventing the disease. Investigators review medical charts from 250 patients enrolled in a primary care clinic to determine whether they received the pneumococcal vaccine.
One possible study question might be:
In this example, the exposure of interest is pneumococcal vaccination, and the outcome of interest is pneumococcal pneumonia. The study would estimate the association of pneumococcal vaccination status with the risk of developing pneumococcal pneumonia.
A study explores whether education level plays a role in a person’s decision to use herbal medications. A group of 500 people from a local shopping mall are asked to complete a questionnaire querying their education level and their frequency of herbal medication use.
A specific study question of interest might be:
In this example, the exposure of interest is education level and the outcome of interest is herbal medication use. Note that the outcome that is being predicted in this example is medication use.
Other studies may examine medication use as the exposure, or predictor, of a disease.
A study is conducted to examine whether heart failure influences survival after a first myocardial infarction. A total of 1,000 people who survive a first myocardial infarction undergo a history, physical examination, and echocardiography testing to determine the presence or absence of systolic heart failure. Subjects are followed until they die or drop out of the study.
A specific study question might be:
In this example, the exposure of interest is systolic heart failure and the outcome of interest is survival. Note that the exposure in this case happens to be a disease. Other studies might focus on risk factors for systolic heart failure, and thus examine heart failure as the outcome of interest.
B. Specifying and measuring the exposure and outcome Effective clinical research requires highly focused definitions of exposure and outcome variables.
For the pneumococcal vaccine study example, possible choices of specific exposures might be:
(1) any previous pneumococcal vaccination (yes versus no), (2) recent pneumococcal vaccination within the past year (yes versus no), or (3) the number of years since last pneumococcal vaccination. Similarly, possible choices for the outcome variable, pneumococcal pneumonia, might be: (1) pneumonia, defined by chest x-ray findings, or (2) pneumonia, defined by cough, fever, rales, and evidence of streptococcal DNA in sputum. For the heart failure example, the exposure variable, systolic heart failure, might be defined by a history of heart failure symptoms, such as shortness of breath and lower extremity swelling, plus evidence of a low cardiac ejection fraction measured by echocardiography.
Once exposures and outcomes are specifically defined, they should be measured as carefully as possible. For example, it is possible that medical chart records that document pneumococcal vaccination are less accurate than computerized pharmacy records that indicate actual disbursement of the vaccine. Errors in measuring exposure and outcome data are common; the consequences of measurement error are discussed in chapter 8.
C. Where to find exposure and outcome data in a clinical research article Information regarding ascertainment of exposure and outcome data is usually described beginning after the description of the study population in the methods section. This section should specifically define the exposure and outcome variables and spell out exactly how they were collected and measured, so that the validity of the study findings can be judged, and so that the study could be repeated under similar conditions. If possible, studies should also describe the accuracy of the data collection methods. For the pneumococcal vaccine example, the authors might state in the methods section, “We defined pneumococcal pneumonia by a hospitalization code for pneumonia plus culture evidence of streptococcus in the sputum. This definition correctly classified 88% of pneumococcal pneumonia cases compared to gold-standard PCR testing for pneumococcus in a subsample of cases.”
III. INTERVENTIONAL VERSUS OBSERVATIONAL STUDY DESIGNSEpidemiologic research studies can be broadly categorized as interventional or observational.
The distinction arises in the method by which study subjects are exposed. An interventional study assigns exposure to study subjects, usually at random, while an observational study observes the exposure, which occurs “naturally.” For the previous example of estrogen use and venous thromboembolism, consider first the interventional approach. At considerable effort and expense, investigators could recruit a large group of postmenopausal women who were not already using estrogen. Each recruited subject would then meet with a study pharmacist who would flip a coin- if the coin comes up heads the pharmacist would assign estrogen therapy and if the coin comes up tails the pharmacist would assign an identical appearing pill that did not contain estrogen (placebo). The coin flip would be conducted in secret, such that neither participant nor study investigators were aware of the results.
Following completion of this random exposure assignment process, the baseline characteristics of exposed (estrogen users) and unexposed (placebo) study subjects can be compared in a table of baseline characteristics, which is usually the first table of a clinical research article. Baseline characteristics of subjects assigned to estrogen versus placebo are presented in Table 2.1.
Baseline characteristics from a randomized trial comparing estrogen to placebo.
Notice how the distributions of baseline characteristics are nearly identical between women assigned to estrogen versus those assigned to placebo, due to the random assignment of the exposure between groups. In contrast, one characteristic that is dramatically different between the groups is the use of estrogen at the beginning of the study; this should be 100% in the estrogen group and 0% in the placebo group. Assuming reasonable compliance with the assigned therapy during the study, potential differences in VTE outcomes between the two groups can be ascribed only to differences in estrogen use and not to other characteristics of estrogen users because in every other respect women assigned to estrogen are similar to those assigned to the placebo.
Now consider the observational approach. A population of postmenopausal women would be identified, but this time researchers would observe whether each study subject was using or not using estrogen. Investigators could ascertain estrogen status by querying computerized pharmacy records or by asking study participants to bring in their medication bottles to the study examination. Notice that in the observational study design, no participants receive a placebo.
Baseline characteristics of estrogen users and non-users are compared in Table 2.2.
Characteristics from observational study comparing estrogen use versus no use.
Under the observational approach, exposure groups are more unbalanced with regard to some of their baseline characteristics. There are also a lower number of estrogen users; only 611 women in the observational study population were using estrogen. Some baseline characteristics appear to be unrelated to estrogen use, such as the serum albumin level. Others, such as previous cardiovascular disease, appear to be strongly linked with estrogen use, possibly because clinicians falsely believed that estrogen use reduced cardiovascular disease and may have prescribed estrogen for that purpose.
Like the interventional design, baseline estrogen use differs dramatically between exposure groups in the observational design: 100% versus 0%. However, if the two groups are found to have different VTE rates during follow-up, there will be residual uncertainty as to whether observed differences are due to estrogen use, or due to differences in other characteristics of the women who used estrogen. This phenomenon is known as confounding and will be covered in detail in chapters 9 and 10. Freedom from confounding is the primary advantage of interventional trials over observational study designs.