|
Indian Journal of Dermatology, Venereology and Leprology, Vol. 74, No. 3, May-June, 2008, pp. 288-291 Resident's Page Observational studies: How to go about them? Setia ManinderSingh Department of Epidemiology and Biostatistics, McGill University, Montreal Code Number: dv08123 Introduction As dermatologists, we are often involved in clinical research, either during our residency or later in academics and clinical practice. Clinical research involves a few major steps: identifying the research question, review of the literature, designing a study, ethical conduct of the research, analysis of the collected data, and publication of results. Each and every step is equally important for conduct of good research. A badly designed and conducted study will provide biased estimates, and cannot be salvaged at the analysis stage. Hence, it is imperative to start thinking about all these issues when the research is being conceived and not wait to address them later in the process. The present article builds on previous articles published in the Residents′ page on research methodology and discusses observational studies in clinical research. Clinical studies can be classified into two main categories: experimental and observational. In experimental studies, the investigator intervenes to change the exposure status in the population under study and assesses the change in outcome. The main forms of experimental studies are randomized clinical trials and other large controlled trials, which may not be randomized (for example, field trials). In observational studies, the investigator does not actively intervene to change the exposure status, but merely observes the population under study. The current article focuses on three types of observational studies: cross-sectional studies, case-control studies, and cohort studies. It addresses the methods involved, advantages and potential limitations, and finally briefly mentions the types of analyses used in these methods. Cross-Sectional Studies Design Example of a cross-sectional study Strengths, limitations, and biases However, there are certain limitations of these studies - one usually cannot comment on the temporality of events in cross-sectional studies. In the above example, if we find that hypercholesterolemia is associated with psoriasis, we cannot comment on the sequence of events; whether hypercholesterolemia preceded psoriasis or vice versa. This is a potential limitation of cross-sectional studies. The assessment of exposure in this type of study does not necessarily distinguish between its etiological or prognostic roles. These studies are not very efficient to assess rare diseases. One can often only measure the prevalence of disease or exposure and not the incidence in these types of studies (e.g., HIV ELISA, VDRL, HSV 2 IgG, cholesterol levels). Hence, the term ′incidence′ should be avoided while presenting results from these cross-sectional studies. Further, it is appropriate to report that X and Y are associated rather than X causes Y from a single cross-sectional analysis. In a cross-sectional study, long duration cases will be over represented and short duration cases (mild or very severe leading to mortality) will not be adequately represented. This is called the length-time bias and one has to be aware of this while interpreting results from a cross-sectional study. The measure of association in cross-sectional studies is the odds ratio. Case-Control Studies Design Example of a case-control study Matching in a case-control study Matching can have its disadvantages: it may be expensive or time-consuming to find a good matched control for the case. If there are no appropriate controls for the identified case, then the case is lost and cannot be included in the study. It may be difficult to account for the effects of the matching variable on the outcome of interest. Selection of controls One should be aware of the population attending the hospital while selecting cases and controls. If we are interested in evaluating a rare disease and the hospital is a referral centre for this condition, it may be difficult to select the cases and controls from the same study base. The cases will be referred from all regions to this hospital, whereas the controls will be from the usual catchment area of the hospital. These issues have to be addressed while designing a case-control study in a hospital setting. Other controls Although friends may be easy to access, it is possible that they may have similar exposure distribution (particularly for exposures such as alcohol, smoking, physical activity, socio-economic conditions etc). Hence, this might not be representative of the general population. Population controls from the same catchment area are good for these studies but very difficult to access and practically the most difficult group, particularly in our research settings. Strengths, limitations, and biases As with other studies, there are certain biases one has to be aware of in these studies. Exposure information is collected after the occurrence of disease (in cases). Such individuals with the disease (cases) are more likely to recall past habits and exposures compared with those without the disease (controls). This may introduce a recall bias, leading to overestimation of the association between the exposure and the outcome. There could be an interviewer bias or observer bias in ascertaining the exposure if the outcome is known ( i.e ., the disease state is known). For example, many case-control studies were done to study the role of BCG in leprosy. Researchers identified cases of leprosy and controls without leprosy. They ascertained the exposure by looking at the BCG scar in both. It is known that sensitivity of reading a BCG can vary. Observers who are aware of the study hypothesis and disease status of the study subjects are more likely to report the presence of a scar in nonleprosy patients than in leprosy patients. Thus, the protective effect of BCG in leprosy will be overestimated (a biased estimate). These biases can be minimized by adequately blinding the observers to the outcome status and standardizing the criteria for evaluation of exposure in both cases and controls. The measure of association in a case-control study is the Odds Ratio. The details of these ratios (calculation and interpretation) in these studies will be discussed in a subsequent article. Cohort Studies Design Prospective studies It is very important that we define the exposure and outcome clearly. Often, it is easy to define the exposure; for example, blood group, some drug exposure, teratogen exposure at birth. It may be more complicated in other scenarios, such as smoking status (how much, how frequent, etc), alcoholism status or physical activity. Similarly, the outcome has to be clearly defined at the beginning of the study ( e.g ., is depigmentation the end point or will hypopigmentation be sufficient as the end point?). Retrospective studies Example of a cohort study Strengths, limitations, and biases There could be losses to follow-up in prospective cohort studies; this may lead to bias if the loss to follow-up is due to the exposure. These studies are not very efficient if the outcome is a rare disease. In retrospective cohort studies, the investigator has to rely on the information already collected. There could be missing data or missing records. The measurement of exposure or outcome may not be by the most appropriate method. However, retrospective studies are less expensive and relatively quicker to conduct. In prospective cohort studies, observers who are aware of the study hypothesis and exposure status may be more likely to confirm it in exposed individuals compared with nonexposed ones - the observer′s bias. This may be more relevant for soft health outcomes (intensity of pain, improvement in a lesion, mental condition etc) or if the outcome is not clearly defined. We can calculate the incidence ratio (or incidence rate) in cohort studies as a measure of association. There are multiple variants and exceptional situations for each of these studies; the information on these variants can be obtained from the references at the end of this article. The purpose of this article was to provide an overview of these different types of observational studies. As mentioned earlier, one should choose an appropriate design to conduct research. The choice of this design will depend on the study hypothesis, and on the type of exposure and disease. It also depends on the time and funds available to conduct the study (particularly in case of postgraduate dissertations). Often, in academic settings time and funds are sought after designing the study. The golden rule is to design the study well. All the potential limitations and biases in each study design and the methods to address them should be considered before initiating the study. It will be appropriate to reiterate here that one should not attempt to salvage a poorly designed and conducted study by using statistical methods at the end of the study. Finally, the type of study design, methods used to measure the exposure and outcome, and the limitations in the study should be clearly mentioned in the published report. Acknowledgment MSS is funded by CIHR-IHSPR Fellowship and CIHR-RRSPQ, Public Health Training Programme for his doctoral studies at McGill University.[6]References
Copyright 2008 - Indian Journal of Dermatology, Venereology and Leprology |
|