Finally, confounding factors (or confounders) are essentially alternative possible explanations of the relationship between risk factors of interest and health outcomes. Confounding factors are associated with both risk factors and health outcomes, but they do not come in the causal pathway linking the two. For example, people working in the asbestos mines and exposed to asbestos fibers are prone to develop lung cancer. Thus, occupational exposure to asbestos is an independent risk factor for lung cancer. However, smoking is also a known independent risk factor for lung cancer. If prevalence of smoking is found to be higher among asbestos miners, then, in studying the association between occupational asbestos exposure and lung cancer, cigarette smoking should be considered a confounding factor. Two common methods for controlling the effects of confounding factors in the risk factor-health outcome linkage are matching, and stratification for the levels of the confounding variables (4). Matching is achieved during the study design phase when participants are selected on the basis of known confounding variables (often with respect to age and gender). Stratification is a data analytical technique where the comparable group statistics are studied after dividing the study populations into different levels depending on the confounding variable (4).
From an epidemiological perspective, causality is subjective. While a statistically valid association does not necessarily imply a cause-and-effect relationship, the relationship is commonly evaluated by using multiple criteria. These include strength of association, temporality, dose-response relationships, specificity, biological plausibility, and replicability of the findings (5). Strength of association implies the magnitude of effect size of the risk factor on the outcome. The stronger the effect, the higher is the likelihood of a cause-effect relationship. Temporal precedence indicates that the risk factor must precede the outcome. This is a necessary and logically the most powerful criterion for cause and effect relationship. Dose-reponse relationship indicates that as the amount of the risk factor increases, there occurs a corresponding increment in the magnitude of the effect. Specificity and biological plausibility are softer criteria. Specificity indicates a single risk factor for a single health outcome, and biological plausibility indicates that a sound biological explanation based on existing paradigms should be offered for an observed relationship. While these criteria are definitely useful and important in conjunction with other criteria, their logical values as sufficient criteria for causality are open to question (3).
Since one risk factor can give rise to several different diseases (smoking for instance, or exposure to noise or sound above 90 decibels), and likewise, one disease may have multiple plausible risk factors (hypertension for instance), the specificity clause is a weak criterion. Also, for all probable risk factor-outcome pairs, it is not possible to offer biological explanations under the existing paradigms of science. For example, no reliable animal model of carcinogenesis exists for chronic exposure to high levels of inorganic arsenic through drinking water, yet epidemiological evidence suggests that long term exposure to high concentrations of inorganic arsenic can induce several different forms of cancer including those of skin, urinary bladder, and lung. Finally, replicability indicates that if studies were to be conducted to evaluate the association between a suspected risk factor and a health outcome for different populations and under differing circumstances, the results would be similar or at least close for these different studies conducted in different populations and circumstances using similar methods. The repetitiveness criterion helps to establish the cause and effect relationship by showing that the relationship is indifferent of the population or other characteristics.
Several types of research study designs are used in epidemiology. The purposes are either a) to generate hypothesis, b) to describe relevant risk factors and health outcomes, or c) to study causal linkages between risk factors and diseases. Study designs that help to generate hypotheses include ecological studies, case studies, and case series. Ecological studies are epidemiological studies where risk factors and health outcomes are studied at population levels and hence the data is available in aggregates. For example, in studying the effects of air pollution on respiratory health, air pollutants are measured on a daily basis in a city. At the same time, hospital admissions due to respiratory diseases are measured from hospital records in select areas. Finally, information on the levels of air pollutants and the number of hospital admissions are statistically analyzed to derive estimations about the relationship between air pollution and respiratory morbidity. The biggest problem with this type of study is lack of inference regarding association at an individual level. Any conclusion arrived at about the cause-effect relationships on the basis of an ecological study is therefore open to ecological fallacy that conclusion about an individual cannot be derived from population level aggregated data (6).
However, these types of studies are important for generation of hypotheses. Other hypothesis generating studies include case studies, case series, and cross-sectional surveys. Case studies are descriptions of individual cases of diseases and case series are descriptions of a series of similar cases. Cross sectional surveys are “snapshots” of risk factors, health outcomes, and other relevant factors in a population. The data are collected at the level of individuals with respect to health outcomes, and risk factors. This is typically done using a survey questionnaire or some other means of measurement, where a community of individuals is approached and information taken at a point in time. The problem with these studies from the perspective of causal linkages is lack of comparison groups. While case studies and case series provide information about possible linkages, they cannot control for the effects of alternative explanations or confounding variables. However, cross sectional surveys provide limited information about comparison groups, at the time of data analysis.