PHEPB: Case control studies in cancer research Flashcards
Describe the design of a case control study
Case study → group of individuals that actually have the disease
Case control → those who do not have the disease
Compare the 2 to past exposure and postential risk factors - this is why it is a retrospective study
What is the difference between casae control and cohort studies?
Cohort → you are looking at incidence - you are lookimg forward in time [Prospective study]
Case control → you already have people w disease - the control should not have the disease but should be comparable - you want them to be similar though (like age, sex, socioeconmic status) [Retrospective study]
How are controls selected for case control studies?
Controls must not have the disease but are otherwise comparable to cases. Main sources inc:
General population
-Random sample from the pop that gave rise to the cases.
-Not always practical.
-Subjects non-co-operative, low response rate.
Hospital
-Readily available, time to spare, far more co-operative.
-Controls must not have a condition that has causes in common with the study disease- eg do not use coronary patients in a study about lung cancer as they both have RF in common like smoking
What is confounding and what can it lead to?
Does H pylori cause stomach cancer, or is it that age is cuases both stomach cancer and h pylori?? (Age= common confounder)
How does matching adjust for confounding factors and what are its disadvantages?
Collect info on potential confounders and adjust for them at the analysis stage. And/or take account at the design stage by matching:
Each case is randomly matched to 1 or more controls (e.g. matched for age and sex). Not always practical.
As you increase the number of matching variables it becomes more difficult to find a match – loss of cases.
Can introduce bias due to over matching. This occurs if the matching variable is on the causal pathway between exposure and disease
What can we not calculate in a case control study?
CANNOT CALCULATE RELATIVE RISK IN CASE CONTROL STUDY! Instead, we estimate odds
“did you smoke prior to diganosis” - pt vs “ did you smoke prior to this date” - control
This is data on a study for lung cancer and cigarette smoking
Use this table to estimate the odds ratio, and otutline the step by step of how you do this
Here the odds ratio is 9.08. Work out the statistical significance of this AND the expected frequencies
Null hypothesis: if there is no differences in cases and controls, then the true odds ratio will equal 1
Expected frequencies= values in the table if we expected the null hypothesis to be true
- yes - case: (1357 x 2646) / 2714 = 1323
- no - case: (1357 x 2646) / 2714 = 1323
- yes - control: (1357 x 68) / 2714 = 34
- no - control: (1357 x 68) / 2714 = 34
How do we interpret odds ratios?
Dose response can emphasise an association between the variable being tested in case-control studies. Use the data below to calculate the odds ratios
e.g. in the 5-14 category ⇒ 489 x61 / 570 x 7 = 7.48 odds ratio
in the 15-24 category ⇒ 475 x61 / 431 x 7= 9.60
Describe bias in case control studies
In designing and analysing a case-control study we aim to ensure:
Comparisons are not explained by confounding.
Comparisons are not explained by bias (systematic error).
Confounding and bias are particularly likely in a case control study
We really do not want a bias in the estimate: odds ratio → this could be a bias leading to systematic error
What are the types of error?
If exposure variable is measured w a random error, then it is kinda okay but not really
But
We really do not want a bias in the estimate: odds ratio → this could be a bias leading to systematic error
Even random error can cause bias → this is called bias towards the null - this means that our estimate to odds ratio will be closer to 1 than when it should be
What are the different sources of bias in case control studies?
Selection Bias: If the control group is not comparable (e.g. if older)
Methods of data collection:
Information bias: If the way we collect info from cases and controls differs (e.g. If case is too ill family gives info instead).
Random misclassification error: Errors in classification of exposure status occur to a similar extent in both cases and controls (generally biases associations towards the null).
Recall bias: Recall of past info may differ between cases and controls.
Interviewer bias. Must ask questions in a standardised form
What are historical case control studies?
historical case-control studies: In some case-control studies exposure info obtained from historical records
Advantage: info recorded prior to disease onset (may reduce the possibility of reverse causation; reduce recall bias)
Disadvantages: Records may be incomplete (loss of cases / loss of controls).
Those extracting the data may introduce bias unless ‘blind’ to case-control status.
What is reverse causation?
Reverse causation → in response to early symptoms, they may change their behaviour - this change in behaviour may be mistaken for causing the disease rather than the idea that it is just an effect of the disease