1. Chapter 9 Lecture Research Techniques: For the Health Sciences Fifth Edition © 2014 Pearson Education, Inc. Conducting Analytical Epidemiologic Studies R. Eric Heidel, PhD University of Tennessee Graduate School of Medicine

2. The Nature of Epidemiology As a discipline, epidemiology knows few, if any, boundaries. Epidemiology is the study of our collective health. It compares two facets: –First, descriptive, involving the identifying of patterns, trends, and disease and injury differentials. –Second, it moves beyond the descriptive approach to embrace causation or etiology. This is also called analytic epidemiology. © 2014 Pearson Education, Inc.

3. Analytical Methodologies in Epidemiology Two major analytic methods used in epidemiologic investigations: 1.cohort 2.case control The purpose of each is to test the hypothesized cause-and-effect relationship between a suspected risk factor and a disease, injury, or even a social condition such as welfare status. © 2014 Pearson Education, Inc.

4. Analytical Methodologies in Epidemiology (cont'd) Cohort Studies –Cohort studies are sometimes called prospective studies. –They work from a postulated cause to an effect. –General design is to begin with a group of people (a cohort) and observe them over a period of time. –Selection for the group can be based upon the presence or absence of a characteristic or at random. –Individuals within the group will vary in exposure to one or more of the factors being observed. –They determine the differences in the rate at which the characteristic occurs (disease, injury, and behavioral problem) in relation to exposure to the factor(s). © 2014 Pearson Education, Inc.

5. Analytical Methodologies in Epidemiology (cont'd) Case-Control Investigations –Case-control investigations are sometimes referred to as retrospective studies. –They start with a group of people who already have the characteristic (health problem) and compare them with people who do not have that health problem (characteristic). –People who have the health problem are referred to as "cases," while those in the other group are termed "controls." –In searching for the "cause," the case-control method determines if the two groups differ in the degree of exposure to different factors. –Case-control method moves backward in time (effect to cause). © 2014 Pearson Education, Inc.

6. Timing Schematic for Cohort and Case- Control Prospective and Retrospective Investigations © 2014 Pearson Education, Inc.

7. Level of Evidence and Analytic Epidemiology Some experimental designs exert better control over variables than others, thereby providing stronger evidence for the relationship. There are five basic categories: –Level I: controlled and randomized –Level II – 1: controlled but not randomized –Level II – 2: well-designed cohort or case- controlled analytic study –Level II – 3: multiple time series –Level III: expert opinions based on clinical experience, expert committees, or descriptive studies © 2014 Pearson Education, Inc.

8. OCEBM Levels of Evidence Working Group © 2014 Pearson Education, Inc.

9. Cohort Investigations A cohort is a group of people with a common experience over a defined period of time. A current cohort study means that the subjects are selected at the beginning of the study (present time) and followed over a period of time (future time). –The cohort model can be modified by making it a retrospective in nature rather than a perspective one. © 2014 Pearson Education, Inc.

10. Cohort Investigations (cont'd) In a retrospective cohort study the cohort assembly, baseline data, and follow-up took place in the past. –This type of study demands adequate data about risk factors and outcomes on the cohort of subjects, which can be problematic. –Current cohort studies have the advantage of greater research control over data collection. © 2014 Pearson Education, Inc.

11. Cohort Investigations (cont'd) There are several ways to select a cohort, one of which is simply by accessibility; other ways include history of exposure or availability of medical records. At the conclusion of the study, the researcher compares the two groups for the incidence rate. Estimations can be made for exposure-specific incidence rates. The cohort method has five advantages and six advantages. © 2014 Pearson Education, Inc.

12. Case-Control Studies A case-control study requires the investigator to work backward in time. Part of the sample is drawn from a population of patients with the outcome (the cases). The other part is drawn from a population who does not have the outcome (the controls). The investigator then compares the predictor variables to determine which ones contributed to the outcome. Generally, it is much less expensive than cohort investigations. Most studies require fewer subjects. Sources for case groups can be records of physicians and employers, hospital records, death certificates, etc. © 2014 Pearson Education, Inc.

13. Case-Control Studies (cont'd) Information can be gathered from subjects themselves or even from proxies for the subjects or patients. –Oftentimes, a combination of records and actual subjects is used. In many case-control studies, selection of cases is often simple in that the number of sources is very limited. In control groups, the obvious point is to procure a sample from a population at risk for the outcome. Controls should be representative of the general population in terms of probability of exposure to the risk factor(s). They should have had the same opportunity to be exposed. © 2014 Pearson Education, Inc.

14. Case-Control Studies (cont'd) It is important that sources and methods of data collection are similar for both cases and controls. Case-control method has eight advantages and eight disadvantages. In comparison to the case-control study is the epidemiologic cross-sectional study. –This type of study has the investigator procuring all the measurements on a single occasion or at least within a very short period of time. –This study is descriptive in nature. © 2014 Pearson Education, Inc.

15. Establishing Causation Analytic epidemiology attempts to establish causation between the health problem and selected risk factors. The following can be of value in determining causation: –Temporal sequence –Consistency –Strength of association –Specificity of effect –Biological gradient –Existing data and theory Causation may never be fully proven in the experimental sense. Analytic epidemiology can assess the likelihood of causation. © 2014 Pearson Education, Inc.

16. Problems of Error Like other research approaches, analytic investigations can succumb to error. Three principle types of error: 1.Bias 2.Random variation 3.Random misclassification © 2014 Pearson Education, Inc.

17. Bias Refers to any error in design, conduct, or analysis of a study that makes the estimate of an exposure's effect on the risk of disease inaccurate. Bias is a threat to internal validity. Three forms of bias: 1.Selection bias Poor research design brings about a difference in comparison groups that misrepresent true results. To prevent this source of error, cohort groups should be homogenous in as many factors as possible, except for the risk factors under investigation. © 2014 Pearson Education, Inc.

18. Bias (cont'd) 2.Information bias It occurs in cohort studies when information about disease outcome is not collected uniformly across groups. Prevention of information bias is to collect information uniformly. Another way to reduce information bias is to not reveal any more about the study than you have to for both ethical and data collection reasons. © 2014 Pearson Education, Inc.

19. Bias (cont'd) 3.Confounding Confounding variables are extraneous variables that are: (1) risk factors for the health problem under investigation (2) factors associated with the exposure of interest but that are not consequence of exposure The two principal ways to control for confounding variables are by matching and by using statistical techniques. © 2014 Pearson Education, Inc.

20. Random Variation It means chance differences between groups. It affects the ability to generalize to a larger population—external validity. It generally occurs because of unrepresentativeness in the comparison groups. © 2014 Pearson Education, Inc.

21. Random Misclassification Errors can be made when determining a subject's exposure status or disease status; a subject could be misclassified. The possibility of finding a strong causal-like relationship between risk factor and health problem is greatly reduced, if not eliminated, in the presence of random misclassification. © 2014 Pearson Education, Inc.

22. Control in Epidemiological Research Matching –One of the most common methods to reduce error is to match on a subject-to-subject basis. Homogenous Grouping –Select a sample in which all the subjects would be homogenous on the variable(s) in question. Stratified Sampling –Similar to homogenous grouping in that subgroups are formed by separating the ranges of selected variables and sampling a predetermined number of cases and controls within cells made from the multiple cross-classification. © 2014 Pearson Education, Inc.

23. Control in Epidemiological Research (cont'd) Poststratification –It means that you classify unmatched cases and controls by their values on one or more variables determined during the study. Multivariate Analysis –It is conducted after the data have been collected. –Research has shown that we can rarely rely on a two-variable study to explain, predict, or control relationships or variables. Instead, we must deal with several variables simultaneously. © 2014 Pearson Education, Inc.

24. Control in Epidemiological Research (cont'd) –Multivariate procedures are: Analysis of Covariance Multiple Regression Analysis Logistic Regression Linear Structural Relation Analysis (LISREL) Proportional Hazards Regression © 2014 Pearson Education, Inc.

25. Analysis of Results in Analytic Epidemiology The purpose of both cohort and case-control studies is to test the hypothesized cause-effect relationship between a suspected risk factor(s) and a disease, injury, or social/health condition. There are measures of association for each type of investigation. –Prevalence It is a measure of the proportion of individuals within a population who have a specific health problem, disease, injury, or other health event at a particular point in time. © 2014 Pearson Education, Inc.

26. Analysis of Results in Analytic Epidemiology (cont'd) –Incidence It is a subcategory of prevalence. It is the number of new cases in a population at risk during a specified time period. Incidence rate is the probability that an individual with no prior disease will develop a new case of the disease over some specified time period. The formula is sometimes modified to fit the nature of the study. When subjects contribute unequally to the calculation of population at risk, epidemiologists use the idea of "person- time" to overcome this dilemma. The use of person-years rather than population units should bring greater accuracy to the measurement. In using person-years, three conditions must be met. © 2014 Pearson Education, Inc.

27. Analysis of Results in Analytic Epidemiology (cont'd) Analyses for Cohort Investigations –In cohort studies, investigators measure the strength of the association between exposure and the disease or health outcome by means of the rate ratio or relative risk. –Relative risk is defined as the ratio of the incidence rate for persons exposed to a risk factor to the incidence rate for those not exposed to that same risk factor. –RR, or risk ratio, shows the extent to which it is more (or less) likely that a health problem or condition will occur in the exposed group as compared to the unexposed group. © 2014 Pearson Education, Inc.

28. Analysis of Results in Analytic Epidemiology (cont'd) –Another rate that can be used in cohort studies is the attributable risk rate, which is simply the difference between incidence rates. It is the incidence rate for the exposed group minus the incidence rate for the unexposed group. This rate, also called the risk difference, indicates the magnitude of the absolute change brought about by exposure. –These two methods of calculating rate ratio or relative risk need to be modified if matching was used in the epidemiologic study design. –McNemar's test is generally used to test significance. © 2014 Pearson Education, Inc.

29. Analysis of Results in Analytic Epidemiology (cont'd) Analyses for Case-Control Investigations –The odds ratio is the most common measure of association between exposure and the health outcome in case-control or retrospective investigations. It can be seen as an estimate of the relative risk. It is the ratio of the odds of disease in exposed individuals relative to the odds of disease in unexposed individuals. –Interpretation is similar to relative risk in a cohort study. –As with cohort investigations, this formula must be modified if matching has occurred. –McNemar's test is generally used to test significance. © 2014 Pearson Education, Inc.

30. Analysis of Results in Analytic Epidemiology (cont'd) Confidence Intervals –Statistical inference can be in the form of hypothesis testing or estimation of parameters. –The two forms of estimation are point estimation and interval estimation. –Point estimation It is a single statistic (or point) to estimate the population parameter. Relative risk and odds ratios are point estimates. The problem with all point estimates is that they fail to convey the accuracy of the estimation. © 2014 Pearson Education, Inc.

31. Analysis of Results in Analytic Epidemiology (cont'd) –Interval estimation It is also referred to as a confidence interval (CI). It provides a range of values within which the population parameter has a specified probability of falling. There are several methods for determining the CIs for an odds ratio. The wider the CI, the larger is the variability of the point estimate and the less likely that the point estimate is accurate. The CI is used to ascertain statistical significance. © 2014 Pearson Education, Inc.

32. Analysis of Results in Analytic Epidemiology (cont'd) Computer programs are available to establish RR, odds ratio, and CIs. While it would be good to have that knowledge, as a minimum you should be able to interpret the results of analytical epidemiological investigations, including CIs and levels of significance. © 2014 Pearson Education, Inc.