LEARN | Module 5
Cohort Studies and Harm
Harm
Clinicians often encounter patients who face potentially harmful exposures; for example, does vasectomy increase the risk of testicular cancer? In examining such questions, health professionals must evaluate the validity of the data, strength of association between exposure and outcomes, and relevance to their patient and their circumstances. As always, please post any questions, comments or suggestions in the Disqus comment feed at the bottom of the module!
Study Design
Different study designs will have their strengths and limitations. Biases may be effectively controlled for by RCTs, but not necessarily by others. However, whilst some study designs might be useful to answer some questions, a trade-off must be made in order to other questions. For example, it is often not ethical to randomly assign a patient to something you think might cause harm!
When investigating harm, aetiology or risk factors, the best study designs are:
RCT (if it is ethical!)
Cohort study
Case-control study
This week, we will be discussing cohort studies.
Cohort Studies
Cohort = a group of individuals who share a common characteristic
Take a large group of people (e.g. all final-year high school students in Fiji)
Separate them into two cohorts:
Those who drink alcohol
Those who do not drink alcohol
Compare the average final exam score of each group
Did the exposure (drinking alcohol) affect the outcome (exam score)?
Cohort studies always move forwards in time (i.e. the exposure must come before the outcome). However:
Prospective cohort studies identify people who are being exposed at the time, and then follow up to see if they get an outcome.
Retrospective cohort studies identify people who have previously been exposed, and then follow them to see if they get an outcome.
In a cohort study, we can express results similar to RCTs if we have binary outcomes (i.e. yes, they have cancer or no, they don't have cancer).
If the follow-up time is similar between participants (outcomes occur at similar times), relative risk can be calculated by finding the risk ratio (= the risk of getting an outcome in the exposed group divided by the risk of getting an outcome in the unexposed group).
The risk of getting an outcome in the exposed group is calculated by those in the exposed group with the outcome divided by all those in the exposed group.
If the follow-up time is different between participants (if they are lost to follow-up), the relative risk must be calculated by finding the rate ratio (= the rate the outcome occurs in the exposed group divided by the rate the outcome occurs in the unexposed group).
The rate the outcome occurs in the exposed group is calculated by those in the exposed group with the outcome divided by the total time spent exposed by all people. For example, if James was exposed for 3 years and then left the study, and Sarah was exposed for 5 years, then the total time spent exposed was 8 years. These are referred to as "person-years".
To make things simpler (or more confusing!), person-years can also be calculated by finding the average number of people in the study at any one time and multiplying it by the length of the study. This is done in the following equation: (number of people in the study at the start plus number of people in the study at the end divided by 2) multiplied by number of years the study ran for.
Note that when a participant develops an outcome, they stop being "exposed".
Bias
The biases that we spoke about in week 3 are also important for cohort studies.
The most important difference between RCTs and cohort studies regarding bias is that cohort studies are far more prone to selection bias. This is because we are unable to prevent it using randomisation. Other bias includes:
Performance bias – individuals know that they have been exposed to a particular risk factor
Detection bias – exposures may not be measured consistently
Attrition bias – who knows when an outcome will occur? Cohort studies have to go for a long time, and many people will drop out
Confounding
Imagine the results of the above example were as follows:
Average score of students who drank alcohol: C+
Average score of students who did not drink alcohol: B+
One would be tempted to decry alcohol as a huge problem in schools that dramatically impacts performance!
However, just because alcohol was the only exposure we were measuring does not mean that it is the only factor having an effect. A confounder is something that can lead to the outcome and is associated with the exposure.
For example, going out three times a week partying will definitely lead to decreased grades. Going out three times a week partying will also be associated with drinking alcohol. This means that the true cause-and-effect relationship may be between [number of parties attended per week] and [exam performance] – drinking alcohol was just an association!
To prevent confounding factors from influencing out results, we need to make sure that groups are balanced. In the above example, the groups would be balanced if:
Students who drink alcohol: 30% attend parties more than twice a week, 70% do not
Students who do not drink alcohol: 30% attend parties more than twice a week, 70% do not
The optimal approach to balancing is randomisation - this should result in patient groups with similar prognostic variables.
In a cohort study, we cannot randomise. Instead, an adjusted analysis can be performed after the data has been collected. This involves creating balanced subgroups for analysis to better estimate the true effect of treatment.
Causation
An association between two factors does not imply that one has caused the other – it could have been caused by chance, bias or confounding.
In order to determine whether an association is likely to be a causal relationship, we can apply a set of rules:
Temporality: did the exposure occur before the outcome?
Dose-response: does a higher level of exposure increase the chance of an outcome?
Dechallenge-rechallenge: does the outcome disappear when the exposure is withdrawn and then return when the exposure is re-introduced?
Consistency: do multiple studies show a consistent association?
Plausibility: does it make biological sense?
Cohort Studies Video
See the video below for a summary of cohort studies.