The chapter focuses on explaining epidemiologic research designs and predictive correlational designs. It starts with the explanation of epidemiology as a discipline and subdisciplines such as Descriptive and analytic epidemiology. The first is focused on the description of the distributions of determinants and diseases (Drummond & Murphy-Reyes, 2018). The second tries to quantify the association between a risk factor and a health-related state (Drummond & Murphy-Reyes, 2018). The chapter describes three research designs cross-sectional, cohort, and case-control. The cross-sectional design uses a particular point in time to investigate the prevalence of disease in an exposed and an unexposed group. Case-control design is the study where participants, also known as cases, have an outcome of interest (Drummond & Murphy-Reyes, 2018). This research design utilizes odds rations that consider odds of exposure or study-specific event by taking the ratio of cases over controls.
Cohort design utilizes potential causes of an outcome and evaluates relevant variables that could affect it. This design is divided into prospective and retrospective studies. In prospective studies, researchers recruit participants representing a target population, collect baseline measurements and evaluate exposure in the future. This design is often accompanied by statistical tests such as ANOVA or MANOVA repeated measures and linear mixed models. ANOVA compares three or more group means based on the continuous dependent variable. Linear mixed models could be interpreted as extensions of ANOVA and linear regression. It has a pattern of change on the subject level, which provides additional perspectives on the outcome. In retrospective studies, researchers identify a cohort of individuals from the past before the outcome and identify causality from previous records (Drummond & Murphy-Reyes, 2018). The authors also include the notion of relative risk, which is inseparable from the cohort design. Relative risk, also known as risk ratio, takes into consideration the probability of an outcome between two groups.
When there is a need to find a predictive link between the predictor and the outcome/criterion variable, correlational predictive design is applied. There are limited studies that rely solely on correlations. Many utilize regression analysis, which uses linear or logistic regression depending on the outcome. Linear regression is used when the variable is continuous, while logistic regression applies a dichotomous outcome (Drummond & Murphy-Reyes, 2018). Linear regression can be separated into simple and multiple linear regression that differ by the number of independent variables. For such regressions, it is important to utilize the square of the correlation coefficient or R-squared. However, R-squared requires a regression coefficient to test a hypothesis. The regression coefficient is the slope of the regression line. This coefficient with a high value demonstrates the increased effect of the independent variable on the dependent variable (Drummond & Murphy-Reyes, 2018). Other variables essential for linear regression are covariates, adjustment, and multicollinearity. Covariate or covariable has an influence solely on the dependent variable and is often considered as confounding variable. Adjustment or controlling refers to the isolation or adjustment of confounding variables. Multicollinearity expresses a high correlation of independent variables.
Similarly, logistic regression has two types — simple and multiple logistic regression. The first is used for the evaluation of the association between single independent and dichotomous dependent variables (Drummond & Murphy-Reyes, 2018). The other explores correlations between one dichotomous dependent variable and multiple independent variables. It is essential to note that the regression coefficient has a different meaning from the linear regression. Hence the odds ratio needs to be used to assess the effect of individual independent variables. In the event that the data needs to be analyzed with interest in the length of time, survival analysis is appropriate. The time from a specific point to the occurrence of the outcome is called survival time in the research frame. In order to analyze the effect of several independent variables, it is useful to utilize Cox’s proportional hazards model as it models survival rates over time. This model provides an estimate of the hazard rate ratio, the probability of an event occurring at a given time.
Reference
Drummond, K. E., & Murphy-Reyes, A. (2018). Chapter 7: Epidemiologic research designs and predictive correlational designs. In Nutrition Research: Concepts and applications (pp. 185–213). essay, Jones & Bartlett Learning.