1. COHORT AND CASE-CONTROL DESIGNS Dr. N. Birkett, Department of Epidemiology & Community Medicine, University of Ottawa SUMMER COURSE: INTRODUCTION TO EPIDEMIOLOGY AUGUST 28, 1045-1215 5/6/2014

2. Session Overview  Review basic features of these two designs  Discuss advantages and disadvantages  Key methodological features in implementation of the designs. 5/6/2014

3. General (1)  Our main focus is on etiology and studies of relationships  External validity is not big issue  A much bigger issue for cross-sectional studies  A big issue for health care delivery, etc.  Methods of Data Collection  Structured vs. unstructured interviews  Qualitative vs. quantitative methods  Personal interview  Mail & telephone surveys 5/6/2014

4. Cohort (1)  Key feature is that subjects are followed from before they became ill until they get the outcome  Two main methods of subject selection Separate exposed and unexposed groups Select a group of people with the exposure of interest and Select a group of people without the exposure of interest Select a group of people with a range of exposure experiences  Follow all subjects up to determine if they develop new cases of the outcome.  Compare the incidence of the outcome in each group. 5/6/2014

7. Prospective Concurrent Historical Retrospective Mixed Historical Ambidirectional 5/6/2014

8. Advantages of Cohort Studies  Demonstrate clear temporal relationship  Allow direct calculation of incidence rates and risks  Allow multiple outcomes to be studied in one project  Allow multiple exposures to be studied in one project.  Provides some indication of disease latency or the incubation period (sometimes)  Suitable for studying rare exposures  Reduced potential for bias  Eliminates recall bias  Controls exposure misclassification. 5/6/2014

9. Disadvantages of Cohort studies  Often require large sample sizes and long follow-up times.  High cost and complexity.  Bias due to loss to follow-up  How to handle changes in exposure status during follow-up  Also a problem in case-control studies but usually ignored.  Lack of available information for creation of exposure history in historical cohort studies.  Outcome assessment misclassification/bias if staff not blinded to exposure status  Advances in technology or medical knowledge can invalidate the outcome assessment, or even the whole study hypothesis. 5/6/2014

10. Key Design Points for Cohort Studies  Selecting the cohort  Determining exposure status  Determining outcome status  Healthy worker effect  Studies of prognosis require Inception Cohort. 5/6/2014

11. Cohorts: Selecting the Cohort (1)  Subjects must be:  Free of outcome at time of entry;  At risk of developing the study outcome  General population  Members of specific groups (e.g. unions, schools, professional organisations)  Special exposed groups (environmental) 5/6/2014

12. Cohorts: Selecting the Cohort (2)  Internal vs. External comparison group  Internal uses people within the cohort Is generally the best option.  External goes outside the cohort. External group can be hard to find due to problems in lack of comparability  General population may be a poor choice: Will include people with exposure 5/6/2014

13. Cohorts: Determining Exposure Status  Need to consider change in exposure over time.  Interviews  Self-completed questionnaires  Body specimens (e.g. blood, hair, toenail clippings)  Biomarkers (molecular epidemiology)  Administrative records  Company records  Birth records  Hospital/MD records  environmental monitoring data.  Physical Examination  Environmental tests  Job descriptions  Job exposure matrix 5/6/2014

14. Cohorts: Determining Outcome Status  Self-report  validity  completeness  Administrative records  Vital statistics  Registries  Medical records  Physical examination  Biological samples 5/6/2014

15. Cohorts: Ontario Health Study  e-mail and web sites to collect basic information of everyone  goal is to recruit all people living in Ontario  Sub-set approached for a more detailed interview  Sub-sub-set approached for physical examination, blood samples, MRI’s, etc.  Has over 200,000 registrants  Target is 1,000,000+ 5/6/2014

16. Case-control (1)  Key feature is that subjects are selected after they have developed the outcome of interest.  Interviews are done after the fact Limits the potential for some measures in etiologically relevant time periods biomarkers psychological state Subject to biases  Need a comparison group (control group or reference group)  Choosing a suitable group is a major challenge.  Can not compute incidence in case-control study. 5/6/2014

18. Advantages of case-control studies  Relatively quick and cheap  not always  depends on the design used  Appropriate for studying rare outcomes.  Require a smaller number of subjects than cohort study  assuming you can find enough cases  Allows study of multiple potential exposure factors in the same study 5/6/2014

19. Disadvantages of case-control studies  Can only study one outcome per study  Cannot determine incidence directly  Except in special circumstances.  Not appropriate for studying rare exposures.  Higher risk of biases in exposure estimation, etc.  Selection of appropriate comparison group can be hard.  They have a bad reputation  Complex design and methodological features  Finding controls is getting harder than finding cases 5/6/2014

20. Key Design Points: Case-control Studies  Selecting the cases  Selecting the controls  Determining exposure status  Sample size and power. 5/6/2014

21. Case control studies: Selecting the Cases (1)  Need a clear case definition (eligibility criteria)  How to find cases?  Hospital-based vs. population-based.  Population based in better but can be hard to find all cases in a pre-defined population.  Population-based registry can be useful. 5/6/2014

22. Case control studies: Selecting the Cases (2)  Retrospective vs. prospective case identification  retrospective is faster  prospective is better  changes in diagnostic approaches, referral patterns, etc. can be a problem with retro. method  Incident vs. prevalent cases  Incident cases are MUCH better 5/6/2014

23. Case control studies: Selecting Controls (1) Without controls, there can be no case-control studies but with the wrong controls, there can only be regrettable case-control studies. Oleckno  This is the biggest challenge is designing a case-control study!!  Should represent the source population which gave rise to the cases.  In a hospital case-control study, do not select as a control group people with diagnoses which are known to be related to the exposure.  Just finding candidate controls is becoming a big challenge  privacy laws  lack of public interest in research 5/6/2014

24. Case control studies: Selecting Controls (2)  Some key questions about control groups  Do the controls come from the same source population as the cases?  Are the controls similar to the cases with respect to potentially confounding factors? (matching; stratified analysis)  Have any restrictions or exclusions been applied to both cases and controls?  Are hospital controls taken from groups which are not associated with the exposure of interest?  Have the controls been selected from the same time period as the cases? 5/6/2014

25. Case control studies: Determining Exposure Status  Determine the etiologically relevant time period.  Can we measure exposure at that time?  Interviewer bias  Use the same method in cases and controls.  blinding  Recall Bias  Hard to prevent/control  Blind subjects to objectives  Include ‘lie detection’ questions  Use validated questionnaires  Use alternative sources of exposure information  records  biomarkers 5/6/2014

26. Case control studies: Sample Size and Power  May need large sample size, especially if studying interactions  Multiple controls per case  Useful mainly when controls are more available than cases or are cheaper to study  4 controls per case is the largest ratio usually used.  In ‘nested case-control studies in pharmacoepidemiology, can use up to 10 or 20 controls per case 5/6/2014

28. Summary  Both cohort and case-control studies have strengths and weaknesses  Case-control studies are harder to design well.  Cohort studies are generally preferred but take also much longer to complete and are much more expensive. 5/6/2014