Cohort studies are those where people are stratified into exposed and non-exposed groups and then they are followed through in time. For each cohort, the rate of occurrrence of the outcome are noted and these are compared using the rate ratio or relative risk estimate.
Retrospective cohort studies are those where the exposure and outcomes are ascertained at the same time, although it is presumed that the exposure preceded the outcome. Retrospective cohort studies are useful study designs for occupational epidemiological studies.
A case control study can be nested within a prospective or a retrospective or even within the scope of a cross-sectional survey. This is also referred to as serological study.
Analysis of Epidemiological Studies
We analyse the results of the different epidemiological studies using different strategies. Our analysis depends on the nature of the outcomes we aim to study and the objectives of our study. In ecological study design, where air pollution and asthma health effects are studied, we use linear regression. for example, .... In case control and crosssectional surveys where we aim to regress the odds of exposure for those with and without outcomes, we use logistic regression. In logistic regression, .... In cohort studies, were time dependent events are modelled, we use survival analysis and a particular type of linear model referred to as Cox Proportional Hazards model. Here we model hazards that are defined as the instantaneous likelihood of the occurrence of an event.
The goal of each of these analyses would be to adjust for the potential confounding variables and also to understand whether the effect of an exposure would vary by the levels of a third variable.
Summary
This was a review of the essential epidemiological principles we use in Environmental Health. To recapitulate, in Environmental Health, we study environmental epidemiology. We discuss cause and causal inference as the goals of most environmental epidemiological studies. Unlike in daily life, the notion of cause and effect in health sciences is not intuitive in the sense that there can be more than one cause for a particular health outcome in populations. Hence we need to discuss the causal models in the form of sufficient and necessary causes. In order to assess whether an asociation is causal, we need to first establish that the exposure and the outcome have a valid association. This is done on the basis of ruling out the play of chance, elimination of bias in the conduct of the study, and controlling for confounding variables. Then, we can use considerations as explained by Sir Austin Bradford Hill to evaluate if the nature of association is one of cause and effect. We measure distribution of diseases using prevalence and incidence and we can measure the impact of the association between an exposure and an outcome using relative risk or risk ratios, hazard ratios, and odds ratios. We can study public health impacts by estimating attributable risks and population attributable risks or impact fractions. Impact fractions can add up to more than or less than 100% as there can be overlaps between exposure variables. In order to arrive at the measures of association, or measures of occurrence of health outcomes, we need to design a range of epidemiological studies. Ecological studies, cross-sectional surveys, case control studies, and cohort studies are used to assess the association between an exposure and an outcome. Each type of study generate data that can be analysed using a range of analytical processes: these include linear modelling and linear regression, logistic regression, Cox Proportional Hazards modelling and survival analysis.