Study design IV: Cohort Studies Principles of design and threats to validity Brent Hagel, PhD
Cohort study A Cohort – ‘designated group of persons’ with characteristic of interest (eg disease, class of medical students) Descriptive OR analytical Prospective OR retrospective
Cohort Study Study groups defined in terms of exposure and followed to determine frequency of outcome Prospective or concurrent cohort Study Retrospective or historical Cohort study Studies can have both prospective and retrospective components Exposure Disease ? ? Study Begins Exposure Disease ? ? Study Begins From Hennekens & Buring. Epidemiology in Medicine. Little Brown and Company, 1987; P. 24
Retrospective cohort study Identification of study groups based on exposure Some or all of the subjects may have developed outcome Look at frequency of outcome in exposed and unexposed Example question: Among factory workers, how many exposed to asbestos develop lung cancer compared with unexposed workers?
Among factory workers, how many exposed to asbestos develop lung cancer compared with unexposed workers? Lung Ca Asbestos No lung Ca Lung Ca No Asbestos No lung Ca Investigator
Retrospective cohort study design Identify cohort with exposure of interest Collect data about other potential predictor variables How do you choose predictor variables? Collect data about subsequent outcome (e.g., medical records)
Retrospective cohort study - strengths Establish that potential cause preceded outcome Potential to evaluate multiple outcomes ‘Cheap’ and quick – the outcome has already occurred
Retrospective cohort study - weaknesses Little control over cohort (what you see is what you get) No control over data originally collected What was collected How it was collected/recorded Association not necessarily causation
Prospective cohort study Identify subjects based on exposure Measurement of characteristics at baseline Follow-up over time for outcome
Among factory workers, how many exposed to asbestos develop lung cancer compared with unexposed workers? Lung Ca Asbestos No lung Ca Lung Ca No Asbestos No lung Ca Investigator
Prospective cohort study - strengths Useful for assessing incidence Establish temporal relationship Reduces recall and recording bias, because have baseline and outcome measurement at specified times More accurate and complete data collection Efficient for common outcomes
Prospective cohort study - weaknesses Expensive and inefficient for rare outcomes Prolonged follow-up may be needed Labour intensive Association not necessarily causation
Validity – Selection bias Defn. Distortion in estimate of effect based on way subjects selected for study Selection probabilities different for different cells of 2x2 table e.g., those exposed to asbestos more likely to die and more likely to be lost to follow-up Lack of association between asbestos and lung cancer Mitigate: Ensure high response rates Sensitivity analysis Carefully consider direction of bias
Validity – Misclassification bias Defn. – differential vs. non-differential Differential: knowledge of outcome influences exposure ascertainment or vice-versa Non-differential: classification errors same for cases and controls or exposed and non-exposed e.g., Differential: research assistants look harder for lung cancer outcomes among those in asbestos factories than in the unexposed industry or industries “Creates” an association between asbestos and lung cancer
Validity – Misclassification bias (cont) Mitigate by: Similar procedures for exposed and unexposed Blind interviewers to exposure status Use objective rather than subjective information Use pre-recorded data if available
Validity – Confounding bias Defn. Risk factor for outcome Associated with exposure Not affected by exposure or disease e.g., smoking an independent risk factor for lung cancer, and if asbestos workers smoke more than non-asbestos industries, could partially account for association Mitigate Matching Document and control Restriction
Design a Cohort Study . . . Research Question: “What is the risk of sustaining a head or neck injury among intercollegiate ice hockey players wearing full face shields compared with those wearing half shields?" You have 30 minutes!
Cheat sheet Choice of design reasons for using a cohort study reasons against a cohort study How would you identify and select your exposure groups? How would you define your outcome variable? When would you measure outcome? What other variables would be important to consider in this study?