Cohort Studies

main features of cohort studies (5)

- observational - retrospective or prospective - start with an exposure and watch for an outcome over a LONG period of time (longitudinal) (usually INCIDENCE) - no intervention - exposed and non-exposed (control) group (determined a priori)

correlation vs causation

correlation: - relationship EXISTS between two variables - there's a pattern in the data, variables move together - doesn't mean they're caused by each other! - could be seen observationally causation: - one event is due to another event - cause and effect - can't be determined from observation studies, must be experimented on

exposure vs intervention - common exposures

exposure: already decided, not assigned - still considered independent variable (effect is being studied) - how does this exposure affect the outcome? - ex: BMI, smoking status, alcohol use, sex assigned at birth intervention: assigned treatment

goal of cohort studies - key feature

goal: see if an exposure is associated with the outcome - observational - starts with exposure and watch for outcome key feature: participants DON'T have the outcome to begin with (it may happen later over time)

what's the most common bias in cohort studies? - how to avoid? (2)

loss to follow up (attrition bias) - length of cohort studies cause many dropouts (death, loss of interest, lost contact, etc.) how to avoid: - obtain info for future tracking - exclude those who will be easily lost (moving soon, unsure about participation, poor health)

are cohort studies observational or experimental?

observational - interventions are not assigned - exposure is already decided by patient factors - prospective or retrospective

positive, negative, and no correlation

positive correlation: variables move in the same direction - one increases, other does too negative correlation: variables move in opposite directions - one increases, other decreases no correlation: no pattern between variables

measure of outcomes: prevalence vs period prevalence vs incidence

prevalence: - number of people who have a disease at a GIVEN TIME - used for permanent conditions - person is only counted ONCE - ex: diabetes, CHF period prevalence: - how many people have this condition at a certain time? (ex: 5 years) incidence: - number of NEW cases in a population over time - used for transient conditions (can occur more than once) - person can be counted more than once - reported in person-years

cohort studies can be prospective or retrospective. - what does this mean?

prospective cohort - data is collected moving forward to watch for an outcome retrospective cohort - outcome has already occurred and data is collected from records - set back to a certain time point (ex: 10 years) and look forward to a certain time to determine if an exposure happened

selection bias: "healthy worker effect"

unsimilar groups are compared to each other - not accurate data --> selection bias

when are cohort studies useful? (4)

1. calculating incidence of a condition (new cases over time) 2. natural history of a condition (how condition evolves) 3. disease progression (how condition evolves) 4. rare EXPOSURES and outcomes (not really great for rare outcomes)

advantages of prospective cohorts (6)

1. can calculate incidence (new cases occurring over time) - usually the MAJOR AIM OF COHORT STUDIES 2. measuring levels of the predictor before the outcome occurs can establish a time sequence - stronger inference of CAUSE 3. predictor measurements are not influenced by knowledge of outcome occurrence - prevent OBSERVER BIAS 4. good for rare exposures (NOT OUTCOMES - take too long) 5. multiple exposures and outcomes can be studied 6. large samples

factors that decrease credibility of cohort studies (5)

1. confounders - must account for / adjust 2. selection bias (loss to follow up) - wrong people are selected 3. detection bias - data collection is aware of exposure, outcomes are assessed differently, exposure classified differently 4. missing data 5. reporting of results (observer bias)

disadvantages of a retrospective cohort (2)

1. lack of control over data (inaccurate, incomplete, missing data) 2. information bias (observer bias, reporter bias, etc.)

cohort subject considerations (7)

1. population(s) of interest - control vs exposure group 2. sampling methods (a priori, from where?) 3. timepoints, practicality 4. access to longitudinal data 5. sample size (loss to follow up bias potential) 6. inclusion/exclusion criteria 7. generalizable to population of interest

advantages of a retrospective cohort (4)

1. time efficient 2. cheaper 3. baseline measurements already made (exposures already measured and published, outcome may have already occurred) 4. follow-up already done

disadvantages of prospective cohorts (6)

1. very lengthy - susceptible to attrition (FOLLOW-UP) bias since people are dropping out, dying, etc. 2. participants may alter their behavior - ex: get more healthy, changes their exposure 3. observer bias - knowledge of exposure may result in different classification of outcome 4. blinding is difficult 5. prone to confounding variables 6. not good for rare outcomes

how are subjects selected in cohort studies? - requirements? (4)

Based on exposure and followed over time to determine outcome can be selected from: - hospital records - clinic records - large databases requirements: - inclusion/exclusion criteria must be determined a priori - exposed and control group must be selected from same source - control group must be free of exposure at baseline, but have potential to develop the outcome - timing, access, etc. must be PRACTICAL

__ variables can make two variables look like they are causative of each other, even if they aren't.

CONFOUNDING variables can make two variables look like they are causative of each other, even if they aren't.

how do confounding variables affect cohort studies? - how to control this (design vs data analysis phase) (2/3)

can lead to over- or under- estimation of an effect, can reverse direction of affect - make it look like A causes B, when really a confounder causes correlation how to control at design phase: - identify confounding variables - limit participation for those with confounders how to control at data analysis phase: - stratify during analysis (after data is collected) - adjust results with multivariate analysis (most common) - propensity score method