Clinical epidemiology : What, why and how?

Rajiv Nath; RC Ahuja; Sandeep Kumar

OPHTHALMOLOGY PRACTICE

Year : 1995 | Volume : 43 | Issue : 2 | Page : 83-87

Clinical epidemiology : What, why and how?

Rajiv Nath¹, RC Ahuja², Sandeep Kumar³
¹ Departments of Ophthalmology, King George's Medical College, Lucknow 226 003, India
² Department of Medicine, King George's Medical College, Lucknow 226 003, India
³ Department of Surgery, King George's Medical College, Lucknow 226 003, India

Correspondence Address:
Rajiv Nath
Departments of Ophthalmology, King George's Medical College, Lucknow 226 003
India

Source of Support: None, Conflict of Interest: None

Check

PMID: 8818319

How to cite this article:
Nath R, Ahuja R C, Kumar S. Clinical epidemiology : What, why and how?. Indian J Ophthalmol 1995;43:83-7

How to cite this URL:
Nath R, Ahuja R C, Kumar S. Clinical epidemiology : What, why and how?. Indian J Ophthalmol [serial online] 1995 [cited 2023 Sep 21];43:83-7. Available from: https://journals.lww.com/ijo/pages/default.aspx/text.asp?1995/43/2/83/25266

Introduction

The term, "clinical epidemiology" is often misunderstood by clinicians and ophthalmologists as a science that deals with field research in a public health setup. In fact, epidemiology is not an independent science. It is a method or tool to investigate distribution and determinants of diseases in the human population,[1],[2] and for applying this knowledge in their prevention. In other words, the term, "clinical epidemiology" simply means "application of epidemiological methods in clinical sciences." Clinical epidemiology helps clinicians in finding answers to the various problems and limitations experienced in their profession through scientifically designed research. Clinical epidemiology is, therefore, just another term for clinical research methodology.

What is Research?

Research can be defined as a careful study and an investigation in order to discover new facts or information. Operationally, research is a "goal-directed process" of finding a "specific answer to a specific question" in an "organized and objective manner" through "unbiased, reliable, and valid measurements."

Why Learn Research Methodology?

With ever-expanding medical literature, most of the present knowledge and principles of management become obsolete. In such a situation, the principles of research methodology help clinicians in selecting only scientifically proven research results for application in clinical practice and, in the process help to improve patient care and health.

Composition of Clinical Epidemiology

Broadly, clinical epidemiology comprises the skills of epidemiology, biostatistics, health social science, and clinical economics. Epidemiology deals with the various kinds of research strategies, their advantages and disadvantages. Biostatistics deals with the application of statistical methods in solving biological (or clinical) problems by describing the characteristics of the study sample (descriptive statistics), and deriving inferences (inferential statistics). Health social science deals with the questionnaire designing, and also helps to discover the social and cultural determinants of a disease (e.g., genetic abnormalities due to consanguinity), and behaviours (e.g., non-compliance to topical drugs in the treatment of glaucoma). Clinical economics helps in taking a rationale and an economically sound decision for the patients (e.g., costs vs benefits of undergoing cataract extraction with or without intraocular lens implantation).

Some Important Terms in Medical Research

Precision in Measurement: The term, "precision" refers to the degree of stability exhibited when a measurement is repeated under identical conditions. In nature, some amount of variability, i.e., lack of precision is inevitable due to biological variations. For example, intraocular pressure can normally fluctuate, though only to a small extent, even at the same time on successive days. This is called "chance variation" and is not given much importance. However, when observer or instrument variation occurs, the variation in measurement is a cause for concern. Unless the conditions are rigidly standardized, different observers may interpret the same event or measurement in different ways. For example, two observers may not grade the optic cup/optic disc ratio equally; or the same observer may grade the ratio differently at successive examinations. Such sources of error increase the spread of the observed values and lead to bias in the results.

Validity in Measurentent: The term, "validity" expresses the closeness of a measurement to the truth. It is defined as "an expression of the degree to which a measurement measures what it purports to measure." The measurement of any variable in medical science may or may not give a truly correct observation. For example, a new tonometer may or may not record the intraocular pressure correctly; it may persistently underestimate or overestimate the pressure.

Random and Randomization: The term, "random" refers to a happening or an event due to chance alone, and is not determined by any other factor. A truly random error due to chance is excusable in research because its total avoidance and detection are not possible, and if large, it may lead to wrong inferences. "Randomization" is the process by which all the subjects in a study are allocated to different groups by chance alone, i.e., all the subjects having equal probability of being allocated into any one of the groups. Generally, the allocation of the subjects to the various groups is not done by the researcher because the researcher may consciously or unconsciously make a systematic error, which would bias the results. For example, while evaluating a new method of surgery, the researcher may consciously or unconsciously tend to include only those cases having a better prognosis in the new surgical method and, in the process may allocate those cases having a poorer prognosis to the current method of treatment. This would make the two groups incomparable, and hence, would bias the results.

Exposure, Predictors, and Outcome Variables: The variable, "exposure" is hypothesized by the researcher as the factor causing or not causing the event of interest, the event of interest being the "outcome." For example, (i) in a study investigating whether or not ultraviolet (UV) light leads to formation of cataract, ultraviolet light is the exposure and cataract is the outcome; (ii) if a new surgical modality is being tried for the control of glaucoma, the new surgical modality is the exposure and the control of glaucoma is the outcome. The term, "predictors" refers collectively to all the variables including the exposure, which can explain the occurrence of the out-come.

Confounding Variables: Any predictor variable associated with the exposure is called a "confounder." In the above example of UV light and cataract, the elderly are more likely to have a higher cumulative exposure to UV light as compared to young individuals, despite ageing being an independent risk factor for cataract. Thus, age can affect the exposure as well as the outcome, and hence, is measured as a confounder.

Blinding (or Masking): Many studies involve two sets of observers; one set of observers for the exposure and the other for the outcome. Thereby, the two sets of observers are kept unaware of the each other's observations. This is so done because knowledge of the exposure status of the subjects can affect the score assigned in the measurement of the outcome, and vice-versa. This process of keeping both sets of observers unaware of each other's findings is called "double-blinding" or "double-masking," and it improves the quality of research.

P Value: The probability of a research result having been obtained by chance alone is assigned a numerical score in biostatistical terms as "p value." The p value should normally be less than 5% (p < 0.05), for a research result to be accepted.[3]

Internal and External Validities: The "internal validity" evaluates the merits and demerits of the study and is therefore an expression of the truthfulness of the results obtained in the study sample. The "external validity" refers to the generalizability of the research conclusions to the reference population taking into account the representativeness of the study sample. The external validity is considered only when the internal validity of a study is strong.

Research Questions and Hypothesis

The research questions may pertain to disease aetiology, distribution, presentation, diagnosis, prognosis, or treatment. The formulation of research questions involves transcription of one's thoughts into a succinct statement of what one intends to do and why. Research questions and hypothesis are closely related but are not quite the same, both in form and purpose. A hypothesis is a statement of belief that can be refuted or confirmed by one's observations. It is made at a much higher and more abstract level, and is actually a prior statement of research inferences which may be accepted or rejected after data verification. Hypotheses are often stated in the null form so that they can be refuted. This is because scientific generalizations while rarely verifiable, are generally falsifiable.

Types of Research Designs

There are two broad groups of research strategies: observational and experimental [Figure - 1]. The observational studies may be descriptive or analytical. The descriptive studies may be case-series or cross-sectional surveys, while the analytic studies may be either longitudinal (cohort) or case-control in type. The experimental studies may be clinical trials, field trials, or community intervention trials. The following is a short account of these research strategies, their advantages, and disadvantages.

Case Series and Case Reports

These studies describe the distribution of symptoms and signs, or investigate the effects of an exposure in a series of patients who already have an outcome, but however, do not involve any comparison group of outcome-free subjects. These studies are, therefore, not considered as valid research. They are instead useful as hypothesis generating exercises, but cannot be used as hypothesis testing exercises. These studies are especially valuable for rare diseases (e.g., ocular manifestations of carotid artery insufficiency[4]), or for newly described illnesses (e.g., ocular manifestations in paediatric AIDS[5]). However, the case series should not be used for investigating the effects of an intervention.

Cross-sectional Studies

These studies are also known as surveys or prevalence studies. In a cross-sectional study, all the subjects in a given population (or a representative sample) are investigated for the presence of outcome and/or exposure of interest [Figure - 2]. For example, determination of the prevalence such as blindness survey or cataract survey. The distribution of the disease or characteristics are studied according to the time, place, and subjects. The important aspect of a cross-sectional study is that data from the subjects are collected only once and the subjects are not followed up. For extrapolating the observations of the study sample to its source population, the study sample must be a fully unbiased representative of the source population.

Strengths of Cross-sectional Studies: The cross-sectional studies are of short duration, less expensive, yield prevalence estimates, and allow several variables to be studied at the same time. They can serve as a first step towards a longitudinal study.

Weaknesses of Cross-sectional Studies: These studies cannot establish the temporality of events. Since the exposure and the outcome are measured at the same time, a logical sequence of the exposure leading to the outcome cannot be proven. Furthermore, the tendency to include the easily available subjects leads to a selection bias and limits the generalizability of the research conclusions. Since the history of exposure is taken retrospectively, there always exists a strong possibility of recall bias (e.g., past history of dehydrational crises is difficult to ascertained[6]). In recall bias, the diseased subjects recall the exposure more easily as compared to the non-diseased subjects. Lastly, the cross-sectional studies are not feasible for rare diseases, because these studies require a very large population to include an adequate number of diseased subjects.

Case-control Studies

The case-control studies are also called as retrospective studies because the data is collected after development of outcome. However, it is best to avoid the terms, "retrospective" and "prospective" because of ambiguity. The case-control design has 3 distinctive features:-(a) both the exposure as well as the outcome would have occurred before data collection; (b) the study proceeds backwards from effect to cause; and (c) it uses a outcome-free control group (comparison group) to support or refute a hypothesis. In case-control studies a group of diseased subjects (cases) and a group of non-diseased subjects (controls) are selected, followed by data collection and data analysis [Figure - 3]. The selection of proper cases and controls are crucial to this study design. The cases should be representative of the disease spectrum and the controls should be representative of the source population. The cases and the controls should be as similar as possible, except for the absence of the outcome but should not be included or excluded on the basis of their exposure status (e.g., if the association between primary open-angle glaucoma and anterior chamber deptah is investigated, the hypermetropes who have shallow anterior chamber depths should not be excluded from the control group[7]). The controls should undergo diagnostic tests to rule out presence of the disease (outcome). The measure of association between the outcome and the exposure in case-control studies is expressed as "odds ratio."

Strengths of Case-control Studies: Case-control studies are easy to conduct, relatively less expensive, and require fewer subjects. Therefore, they are best suited for rare diseases. Since these studies are based on observation, the subjects are exposed to less risks, if any, and moreover, ethical problems are usually not encountered unless the diagnostic tests are invasive. Another advantage is that several aetiological factors can be measured as exposures. Since these studies do not require patient follow-up, the problem of attrition (loss to follow-up) does not arise, which is often a major drawback of longitudinal studies. The odds ratio obtained in case-control design is same as the relative risks obtained in longitudinal design provided the controls represent the source population and the outcome is rare.

Weaknesses of case-control Studies: Like the cross-sectional studies, these studies cannot establish the temporality of events and it also cannot yield incidence rates. A major problem with this design are biases, which have to be carefully prevented. Since the exposure is measured retrospectively, recall bias can be an important bias and may lead to uncertainty in the measurement. This design calls for proper selection of the controls because the odds ratio thus obtained would otherwise be misleading. For example, if easily available subjects are selected to serve as controls in the study (i.e., if the hospital patients suffering from diseases other than the one under study are selected as controls for the sake of easy availability), the odds ratio thus obtained would be misleading as these patients may not be fully representative of the source population. Therefore, due to improper methodologies in the case-control study design, this design has been a target of heavy criticism,[8] and has led to conflicting reports.[9],[10]

Longitudinal Studies

Longitudinal studies are also known as cohort studies. In this study design, the measurement of the exposure is done on outcome-free subjects; some of these would include subjects found to be exposed and the rest would include those found to be unexposed or less exposed. These subjects are followed for a reasonable length of time and are examined and investigated for development of the outcome [Figure - 4]. Thus, the incidence rate can be estimated for the exposed and unexposed groups, which is defined as the number of new cases of the disease identified in per 100 (or 1000) subjects per unit of time (say, per year). The ratio of the incidence rates in exposed and unexposed subjects is called as the "relative risk," which is an index of association between the exposure and the outcome. It is important that the eligibility criteria for inclusion of the subjects be laid out in an objective manner at the beginning of the study. The exposed and the unexposed subjects should be comparable in respect of the confounders as much as possible, or the confounders be adjusted for, in the analysis of the data.

Strengths of Longitudinal Studies: In these studies, the incidence rates and relative risks are important indices of disease load and its association with a given exposure.

These studies can demonstrate temporality between the exposure and the outcome because the outcome follows the exposure in a prospective manner. Since the exposure is measured before development of the outcome, the exposure can be measured without recall bias. Furthermore, a dose-response relationship (if present), can often be established between the exposure and the outcome in an unbiased manner. Also, several exposures and outcomes can be measured in the same study (e.g., Framingham Study[11]).

Weaknesses of Longitudinal Studies: These studies are generally not suitable for rare diseases because a large sample population has to be followed up for development of the outcome in a small number of subjects. Longitudinal studies are also not suitable for chronic disease because the long latent period between the exposure and the outcome necessitates a longer follow-up which, in turn, may result in attrition (loss of patients to follow-up[12]), and often, the research loses its relevance by the end of the study.

Randomized Controlled Trial

This study design is considered as the most superior design and is often used to verify the effectiveness of new methods of treatment. In this design, individuals selected from the same source population are randomly allocated to either of the two groups. One of these groups is given intervention (a new drug or a new surgical procedure), while the other group is given either placebo or the current mode of treatment. The participants of both the groups are observed for occurrence of the event of interest, i.e., the outcome, which may be a cure or regression of disease marker, or improvement in any predefined manner [Figure:5]. For example, in a study comparing peribulbar and retrobulbar anaesthesia for cataract surgery, the patients would be randomly allocated to the two groups after their consent and would receive respective interventions, and then, the data pertinent to the anaesthetic effects would be collected and compared. In order to reduce the patient and observer biases, a randomized controlled trial should be double-masked, as discussed earlier in the article.[13] This research design can also be used in experimental animals for investigating associations between various aetiological factors and diseases; however, this cannot be done in human beings due to ethical reasons.

Strengths of Randomized Controlled Trials: Due to the process of randomization, both the known and unknown confounders get distributed equally between the two groups. As a result, the two groups become similar in baseline characteristics. The eligibility criteria, the intervention, and the outcome assessments can be standardized. This study design allows use of statistical methods which have few inbuilt assumptions. Therefore, this design is considered as the strongest design.

Weaknesses of Randomized Controlled Trials: These studies may be expensive to conduct in terms of time, finances and personnel, and are often found not suitable due to ethical reasons. These studies sometimes tend to be artificial to some extent. For example, the intervention procedure or the patient compliance may be more rigid in the trial than in common practice, and therefore, the efficacy of the intervention may' get overestimated.

Conclusion

A research design should be selected after considering the merits and demerits of the study. The variables should be defined and measured in an objective, reliable and valid manner. Apart from the exposure and outcome variables, other predictors including confounders should be identified and utmost caution should be taken in their measurement. Biostatistics should be used appropriately and the hypothesis should be verified by efficient methods. By using the skills of clinical epidemiology, the research results and conclusions thus obtained with high internal and external validities will earn recognition.

References

1.	Christie D, Gordon I, Heller RF. Epidemiology: An Introductory Text for Medical and Other Health Science Students. Newcastle, New South Wales University Press, 1987.
2.	Sommer A. Epidemiology and Statistics for the Ophthalmologist. New York, Oxford University Press, 1980.
3.	Bourke GJ, Daly LE, McGilvery J. Hypothesis testing: Introduction to statistical tests of significance. In: Interpretation and Uses of Medical Statistics, 3rd Ed. Oxford, Blackwell Scientific Publications, 1989, pp. 64-72.
4.	Nath R, Mehra MK, Agrawal J, et al. Ocular manifestations of internal carotid artery insufficiencies. Indian J Ophthalmol 35:4-6, 1987.
5.	Dennehy PJ, Warman R, Flynn JT, et al. Ocular manifestations in paediatric patients with acquired immunodeficiency syndrome. Arch Ophthalmol 107:978-982, 1989.
6.	Minnasian DC, Mehra V, Verrey JD. Dehydrational Crises: A major risk factor in blinding cataract. Br J Ophthalmol 73:100-105, 1989.
7.	Saxena S, Agrawal PK, Pratap VB, et al. Anterior chamber depth and lens thickness in primary angle-closure glaucoma: A case-control study. Indian J Ophthalmol 41:71-73, 1993.
8.	Sackett DL. Bias in analytic research. Jr of Chr Disease. 32:51-63, 1979.
9.	West SK, Munoz BE, Newland HS, et al. Lack of evidence for aspirin use and prevention of cataract. Arch Ophthalmol 105:1229-1231, 1987.
10.	Van Heyningen R, Harding JJ. Do aspirin-like analgesics protect against cataract? Lancet 1:1111-1113, 1986.
11.	Kahn HA, Leibowitz HM, Ganley JP, et al. The Framingham Eye Study: II. Associations of ophthalmic pathology with single variables previously measured in the Framingham Heart Study. Am J Epidemiol 106:33-41, 1977.
12.	Waring GO III, Lynn MJ, McDonnell PJ. (PERK Study Group). Results of the prospective evaluation of radial keratotomy (PERK) study 10 years after surgery. Arch Ophthalmol 112:1298-1308, 1994.
13.	Agrawal K, Saxena RC, Nath R, et al. Local anesthesia by peribulbar block for cataract extraction in an eye relief camp: A double-masked randomized controlled trial. The Online J of Current Clinical Trials (USA) 40:930319, 1993.