4. Case-Control Studies Flashcards
How are case-control studies carried out?
Give some examples of cases and controls.
Why is matching important?
Identify individuals with disease (cases), identify `similar’ individuals without disease (controls), determine previous exposure, relate information on exposure to disease
Cases: incident cases from disease registry, hospital-based recruitment
Controls: same population as cases, if hospital-based ensure reason for being in hospital is not also related to exposure of interest
Potential confounders e.g. age/gender. Difficult to recruit controls
What is undermatching?
What is overmatching?
What are the types of bias in case-control studies?
Cases and controls aren’t similar enough. E.g. if don’t match on age cases may be older and therefore more likely to smoke than controls - this would give impression that smoking is related to the disease when it may not be.
Cases and controls may be too similar. E.g. if choose siblings may not differ in exposure of interest for example parental smoking. This would give impression that smoking is not related to the disease when it may be.
Recall bias (cases may remember more than controls), reverse causality (has disease caused changes in recent exposures?), case selection (are they representative of all people with the disease?), control selection (are they representative of all people without the disease, are they similar to the cases?)
Distinguish between cohort, case-control and nested case-control studies.
Cohort - ID group of people, follow-up over time. Case-control - start with cases (with disease), find similar controls, ask about their previous expxosure. Nested case-control - follow cohort over time (take serum samples) and record who gets disease, take those who get disease and match with controls from the cohort, go back to stored serum and analyse diseased ones.
Nested e.g. BUPA Cohort study took serum samples and stored in Wolfson Institute, receive death certificates as men die, after 20 years follow up look at men who died from prostate cancer, selected men who didn’t die from prostate cancer, measured PSA in stored serum samples
What are some advantages of nested case-control studies?
What are some disadvantages of nested case-control studies?
Would you use cohort or case-control studies to investigate:
a) a risk factor for a rare disease
b) a rare exposure
c) a risk factor for a disease quickly
d) Recall bias is a problem in these studies
Cheap, quick, easy, exposure before disease
Need cohort study with stored serum samples
a) case-control
b) cohort (case-control = alot smaller and if rare exposure, no one may be exposed)
c) case-control (faster than cohort)
d) case-control (ask ppl about current exposure)
Look at the data. What is the risk of a baby dying if its put on its front?
a) you can’t tell
b) 30/54 = 55%
Why is this the answer?
a) You can’t tell
B/c you can’t use RR as you don’t know the risk of the disease (as you have started with cases with the disease).
What is the formula for odds ratios?
Calculate the odds ratio of babies who died from being put on their sides rather than backs from the data below.
Used instead of RR for case-control. Odds ratios = odds exposure in cases / odds exposure in controls. If the disease is RARE the odds ratio is a good estimate of the relative risk. (In cohort study can calc RR and OR)
NB: answer is the same as saying “babies are twice as likely to die if they have been put on their sides rather than their backs”, becuase SIDs is rare and RR of a RARE disease is approximately equal to the ODDS RATIO
What is the relative risk of a rare disease approximately equal to?
How do you calculate absolute excess risk?
Can you calculate this in a case-control study?
How would you go about calculating the AER from this?
Odds ratio
AER = Risk in exposed - Risk in unexposed
No - don’t know risk in case-control study
Need to use other sources to calculate an estimate of the AER
Using the data below, and the external data that states ‘population risk of SIDs = 3.4 deaths per 10,000 liver births’, calculate the estimated absolute excess risk.
How do you calculate attributable proportion?
p = proportion exposed in the population. How would you find out p (of babies put down on their sides) using the data and example below?
p (RR - 1) / 1 + p (RR - 1)
P ≈ 241 / 774 = 31%
NB: pic = how to calc attributable proportion from it
What is a cross-sectional study?
What are the advantages and disadvantages?
Measure existing disease and current exposure. Sample at 1 point in time without knowledge of disease or exposure
Advantages
– Can look at exposures that won’t change e.g. gender
– Gives measures of prevalence and exposure rates
Disadvantages
– No use for rare exposures or rare diseases
– Not useful for assessing causality
What are the 4 types of study design?
What type of study should be used for:
a) does exposure to blue asbestos cause lung cancer
b) is back pain more prevalent in men or women
Cohort, case control, cross-sectional and clinical trial.
a) cohort - b/c exposure is rare - not many people will be exposed to blue asbestos. If do case-control, prob find noone or one or two exposed to it; case-control not good for rare exposures
b) cross-sectional survey - b/c just looking at prevalence
Give the order of the strength of proof of causality in ideally conducted studies from the worst to the best kind of study.
How do RCT, cohort, case-control and cross-sectional trials compare when looking at:
a) bias
b) confounding
c) evidence
Cross sectional (worse), case-control, cohort, clinical trials (best)
List some reasons why you might see an association between an exposure and a disease.
Bias, reverse causality, confounding, incorrect analysis, chance, causal (real association)
What kind of bias would be at work in this situation:
A case control study of patient’s with Chron’s disease showed that they were much more likely to have reported ever eating Corn flakes.
What kind of bias would be at work in this situation:
Non-randomised studies (cohort and case-control) have shown an increased risk of death in preventive (multivessel) percutaneous coronary intervention versus infarct artery only PCI in patients with STEMI
What kind of bias would be at work in this situation:
A case control study of stomach cancer patients found that they had a higher antacid usage in the 6 months prior to diagnosis than did the controls.
Recall bias - those with Chron’s disease more likely to remember everything they ate. Most people have eaten corn flakes at some time.
Selection bias - patients who had preventive PCI more likely to be sicker to start with, so more likely to die. Randomised controlled trials have shown the benefit of preventive PCI
Reverse Causality - people with undiagnosed stomach cancer tend to have the same symptoms for which antacids are taken
What kind of bias would be at work in this situation:
In a cohort study heavy drinkers were found to be more likely to develop lung cancer. Does drinking cause lung cancer?
What kind of bias would be at work in this situation:
Randomised placebo controlled trial of candesartan in patients with chronic heart failure found it increased risk; RR = 1.6 (95% CI 1.1-2.2)
Confounding - heavy smokers are more likely to be heavy drinkers. Smoking causes lung cancer
Chance - this trial was the odd one out
