Return to Course Materials
Title: UCSF - Cohort Study Design
Slide 1: Cohort as the Basis of Design
All study design is best thought of as ways of sampling the disease experience of a cohort.
Here we have just substituted the word \x93cohort\x94 for the phrase \x93study base\x94 or, if you prefer, \x93reference population.\x94
Slide 2: Cohort Study Design
Mimics individual\x92s progress through life and accompanying disease risk.
Gold standard because exposure/risk factor is observed before the outcome occurs.
Randomized trial is a cohort design with exposure assigned rather than observed. (See also: Randomized Controlled Trial
Case-control design, in particular, is best understood by considering how the experience of a cohort is sampled. (See also: Case-Control Study)
Slide 3: Cohort Study Design - Randomized Trial
The sine qua non for causation is that the cause precede the event, and the cohort study is the gold standard because it provides this temporal sequence.
A randomized trial of whatever flavor (blinded, placebo-controlled, etc.) is a cohort study in which the exposure/treatment is determined by the investigators rather than just observed as in an observational cohort study. Even well designed clinical trials, however, do not always achieve the experimental goal of completely controlling all variables influencing the outcome because individuals may change their treatment on their own, change other behaviors during the course of the study, or be lost to observation by the investigators.
Cohort design and cross-sectional design are relatively easy to understand, but there is a lot of misunderstanding about case-control design, even in epidemiological text books. For that reason, we will spend more time on the case-control design both because it is the most misunderstood design and because it presents some options for sampling that should be understood.
Slide 4: Framingham Cohort Study
The impact of diabetes on survival following myocardial infarction in men vs women. The Framingham Study.Abbott RD, Donahue RP, Kannel WB, Wilson PW.
The impact of diabetes on recurrent myocardial infarction (MI) and fatal coronary heart disease was examined in survivors of an initial MI using 34-year follow-up data in the Framingham Study.
Among nondiabetic patients, the risk of fatal coronary heart disease was significantly lower in women compared with men (relative risk, 0.6).
In the presence of diabetes, however, the risk of recurrent MI in women was twice the risk in men.
In addition, the effect of diabetes doubled the risk of recurrent MI in women (relative risk, 2.1) but had an insignificant effect in men. JAMA, 1989
Slide 5: Framingham Study
The Framingham Study is the oldest and best known cohort study in the country;.
It is the classic cohort study of coronary heart disease, and the study that first elucidated the major risk factors of blood pressure, high cholesterol, smoking, and obesity as risk factors for coronary heart disease.
It was started in 1948 with 5,209 men and women between the ages of 30 and 62 from Framingham ;Massachusetts.
A second generation study was started in 1971 with 5,124 children, and their spouses, of the original cohort members.
A third generation is now being recruited with the goal of enrolling 3,500 grandchildren of the original cohort with an emphasis on studying genetic factors associated with heart disease.
Slide 6: Cohort Study: Main Threat to Validity
Subjects lost during follow-up
Prospective cohort thought of as best study design but poor follow-up can change that
Equally true of clinical trials and observational cohorts
Number of losses is less important than how losses are related to outcome and risk factor
\x93What is the effect of subjects who are lost to follow-up on the inferences generated by a cohort study?\x94
This is the key element in conducting a good cohort study or clinical trial.
Losses to follow-up are usually carefully reported in clinical trials, at least in more recent years in the better journals, but it is surprising how many articles from observation cohort studies give little or no attention to this crucial element.
Slide 7: Other Threats to Validity
There are, of course, other threats to validity but they are related to elements such as measurement and confounding that we will take up later in the course.
Slide 8: Cohort Lost During Follow-up - Random Losses
If losses are *random, only power is affected.
The point that random losses only affect the power of a study should be obvious since if you just randomly removed subjects from your baseline before you began the study, you would simply have a smaller study (fewer subjects=less power).
Slide 9: Cohort Lost During Follow-up - Outcome Bias
If disease incidence is research question, losses related to outcome bias results
This point may not be so obvious. If the losses are related only to the outcome, the estimate of disease incidence is biased because you will observe either fewer or more diagnoses than if you had retained everyone in the cohort (bias can go in either direction). But the association between a predictor and the outcome is not biased because the losses are balanced (proportional to the baseline proportions) in both those with and without the predictor, and therefore the ratio of the incidence in the two groups remains the same.
Slide 10: Cohort Lost During Follow-up - Question of Association
If association of risk factor to disease is focus, losses bias results only if related to both outcome and the risk factor
But if the losses are related to the predictor as well as the outcome, that ratio of the incidence in the two groups will not remain the same and you will have bias in your measure of association.
Slide 11: Who is Leaving Cohort?
Crucial issue is who is leaving cohort:
What bias do the losses to follow-up introduce?
Are disease diagnoses being missed?
Are those with a risk factor more likely to leave and then be diagnosed?
In practice, it may often not be possible to answer these questions in a cohort study for the simple reason that subjects who are lost or refuse to participate further may not be available to supply the answers.
At a minimum you need to attempt to answer them with whatever information you do have about the characteristics of those leaving the cohort by comparing those characteristics with those you retain.
Slide 12: Guarding Validity
Another strategy to guard the validity of your result is to do a sensitivity analysis in which you assume a worst case scenario for those you have lost and see how much it could affect your findings.
There are two studies in prestigious journals who were allowed to publish their results without any data on loss to follow-up or its potential effect on their results.
Something they would not have been allowed to do if these were experimental cohorts (i.e., clinical trials).
They are both treatment clinic-based cohorts, which means they counted a patient as in the cohort after two visits to the clinic (a little like a run-in design in a clinical trial where you test whether someone is going to be compliant before enrolling him or her).
It also means that their follow-up was driven by return visits to the clinic.
Since they got contrary results with almost identical methodology, one would like to know whether differing follow-up biases had anything to do with the differing results. There is no way to tell from what they published.