Cohort studies

Question 1

Design of cohort studies - basic features, designs (2)

Answer 1

Select study population on basis of exposure status then follow up to determine which animals/herds develop the outcome. Can be prospective (disease not occurred at study start) or retrospective (disease has occurred at study start).

1. Risk-based (cumulative incidence) design: exposure groups defined at beginning of study (fixed cohorts), all subjects observed for the full risk period (closed population). LTF more problematic (>10%). Work best for diseases with relatively short risk period.
2. Rate-based (incidence density) design: new individuals added to study throughout study period and have different lengths of follow up (open source population); exposure groups contribute at-risk time until they develop disease, leave the study or period of follow up ends. If short follow-up, compare incidence density (RR). If long follow up use survival analysis to overcome assumption that disease occurs at constant rate.

NB Dohoo distinguishes between cohort (known exposure status at outset therefore can select 2 cohorts - exposed and unexposed) and longitudinal (unknown exposure status therefore select single cohort and determine exposure status).

Question 2

Sampling for cohort studies

Answer 2

Study subjects usually chosen purposively not randomly (increases risk of selection bias). Exposed and unexposed should be from one source population (helps ensure study subjects have numerous characteristics in common and can reduce the risk of unmeasured confounding). Sample size calculations performed assuming disease is measured as risk [cumulative risk] (also sufficient for rate-based [person-time] study). Also must consider unequal sample sizes (exposed/unexposed) and repeated measures.

Question 3

Measuring the exposure in cohort studies - considerations (3)

Answer 3

1. Must define what constitutes exposure. May be measured on a dichotomous (exposed, not exposed), ordinal (low, medium, high), continuous scale (ppm of toxin).
2. Consider whether lifetime exposure, historical exposure or current exposure is best measure.
3. Permanent (sex, breed) and once-off exposures (vaccination, colostrum intake) easier to measure than non-permanent exposures (ration, housing, procedure where timing might be important e.g. neutering). In latter case timing and extent of exposure need to be measured.

Question 4

Follow up in cohort studies

Answer 4

1. Must be complete as possible
2. Must be un-biased with respect to exposure status (blind person measuring outcome to exposure status).
3. Collect ancillary information to assess if LTF is related to exposure (e.g. in study of fertility, culled animals will be LTF and this may bias study findings if the reason for culling is related to exposure status).

Question 5

Measuring the outcome - considerations (4)

Answer 5

1. Must have clear diagnostic criteria.
2. Blind person measuring outcome to exposure status (reduces differential misclassification bias)
3. Test at beginning to ensure animals don’t already have disease (include only new disease events)[may not be possible in retrospective cohort]
4. Consider time of diagnosis vs time of occurrence (test regularly and use midpoint between examinations as time of occurrance)

Question 6

Analysis of cohort studies (2)

Answer 6

1. Risk-based design: report average risk of disease in exposed and unexposed and survival times during follow up, compare risk in exposed vs risk in unexposed using bi-variate analysis with statification (to control confounding), multivariate analysis using logistic regression (OR) or log-binomial or Poisson models (RR)
2. Rate-based design: survival models, Poisson models with time at risk as offset

Report findings according to STROBE

Question 7

Ensuring comparability of exposed and unexposed groups in cohort studies

Answer 7

In design phase:

1. Restricted sampling of study subjects e.g. limited to only one gender, one breed
2. Matching study subjects e.g. for every exposed male, enroll unexposed male

In analysis phase:

1. Stratification
2. Multivariate analysis

Question 8

Potential biases in cohort studies

Answer 8

1. Selection bias: in prospective cohort study is unlikely since outcome hasn’t occurred yet, however could occur if we inadvertently selected participants on the basis of them having being exposed and having a precursor to the disease of interest; LTF bias may occur if participants that are lost to follow-up in a way that is related to their exposure status and outcome status (i.e. rates of follow up differ between exposure groups, and the reason for leaving is related to the outcome.)
2. Information bias: may occur if measurement of outcome is influenced by exposure status (e.g. if diagnostician is not blinded to exposure status)
3. Confounding bias: will occur if exposed and unexposed are not comparable with respect to factors that influence likelihood of the individual developing disease (managed in design phase or analysis phase)

Question 9

Cohort - template

Answer 9

1. Target population 2. Source population 3. Unit of observation 4. Measuring/sampling based on exposure status 5. Follow-up 6. Measuring outcome 7. Potential biases

Question 10

Cohort study - advantages (4), disadvantages (3)

Answer 10

Advantages

1. Most effective among observational studies for investigation of causal associations (less subject to selection and confounding bias)
2. Provides disease incidence measures (more meaningful than prevalence)
3. Multiple outcomes can be studied simultaneously
4. Useful for rare exposures

Disadvantages

1. Long study duration, with potential for loss to follow up
2. Large sample size needed, esp for rare diseases
3. Expensive