PPS: Cohort Studies and Case Control Studies Flashcards
What is a cause of disease?
A cause is a factor which, of itself, increases
the risk of a disease occurring
An event, condition, or characteristic without
which the disease would have been less likely
to have occurred
(Kenneth Rothman)
(in epidemiology, a potential cause is often
referred to as an `exposure’)
Define risk factor
ANY characteristic which IDENTIFIES a group of
People at increased (decreased) risk of disease, now or in the future
A risk factor need NOT be:-
- causal
- independent
- modifiable
Classic examples are AGE, SEX
How can we establish whether air travel causes VTE (venous thromboembolism)?
Need to do comparative studies, comparing:
-disease cases with non-diseased, did they have
different degrees of exposure to the potential cause?
-case control study - (observational)
-exposed with unexposed, do they have different risks
of developing disease?
-cohort (longitudinal) study - (observational)
-randomized controlled trial - (experimental)
Observational studies (not interfering with exposure)
Case control study
Cohort (longitudinal) study
Experimental studies (interfering with exposure)
Randomized controlled trial
If an association is demonstrated through a study what must you then consider?
Whether its causal association or not
How do case control and cohort studies differ?
Case control: retrospective
(Study involves selecting cases of disease and controls
(non-cases) and then studying their previous exposure)
Cohort: prospective
(Study involves following these groups forward
over time, providing disease incidence)
What makes RCT a powerful study design?
Like a cohort study, but the exposed patients are randomized and unexposed patients are exposed to see development of disease or not.
interpretation stronger than other studies.
What is the outcome of studies?
ALL of these study types give us an estimate of the
RELATIVE RISK, which in this case is:
risk of VTE in people who have flown (exposed)/
risk of VTE in people who have not flown (unexposed)
From cohort studies and from randomized controlled trials (but not from case control studies), we also obtain information on incidence rates in people exposed and unexposed, which provides `attributable
risk’
Incidence of thromboembolism
Slightly higher in afro-caribbean origin vs white europeans (racism)
strongly related tp age
What do most VTE causal factors operate through?
Virchow’s triad
- Reduced rate of blood flow
- Increased coagulability of blood
- Damage to venous endothelium
Most causal factors operate via at least one
component of this triad
How do causal factors act through reduced rate of blood flow?
• Immobility especially with serious illness
– major surgery
– serious injury - lower limb/pelvic fracture
– myocardial infarction
– neurological problem (e.g. stroke)
• Heart failure
How do causal factors act through increased coagulability?
- Severe injury (e.g. lower limb/pelvic #)
- Cancer and cancer treatment
- Pregnancy (RR 5x)
- Oral contraceptive, HRT use (RR 2-4x)
- Dehydration (haemoconcentration)
- Hereditary thrombophilias
Hereditary thrombophilias
How do causal factors act through injury to venous endothelium?
• trauma/severe injury especially to lower limbs and pelvis
OTHER RISK FACTORS
• obesity – yes at BMI > 30kg/m2
• cigarette smoking….slight increase
• varicose veins?
Does it seem likely that airline passengers are at increased risk of VTE?
• By analogy with known causes: – immobility a feature of air travel (produces low blood flow) – factors present on an airline flight which increase blood coagulability – dehydration – hypoxia
Case control studies: PROS
-quick to carry out
-relatively cheap
-a good approach where disease is uncommon
(studying existing cases is efficient)
-can look at several possible exposures
(though only one disease!
Case control studies: CONS
• CCS only provides a relative risk estimate,
no measure of incidence/attributable risk
• High risk of confounding
• High risk of bias (systematic error)
Confounding and bias can happen in both case
control studies and in cohort studies, but are
particularly likely in a case control study
Confounding factor
A confounding factor is a factor associated both with the exposure being studied and with the disease outcome, so that it can cause a spurious association
Why is systemic error (bias) a concern in CCS?
- in the selection of cases and controls
- in data collection on exposure from cases and controls
Key issues in case control design
• Hypothesis, confounding factors
• Size and statistical power of study
• Selection of cases
• Selection of controls
• Conduct of study, especially measurement of
exposure and the management of confounders
• Approach to analysis
Why is study size important?
Need to ensure that we design a study big
enough to have a good chance of finding an
association of expected strength if present –
many studies too small (scope for random error)
• To decide how big a study….
• How strong an association? (relative risk)
• How common is the exposure?
• What p value will be statistically significant?
• What chance do you want to have of detecting an
association if it is really present? (often 80-90%)
What are the guiding principles for selecting cases for case control studies?
• Standard definition of cases
• Newly diagnosed (incident) cases or
established (prevalent) cases?
– Using incident cases has advantages….
• Closer to causes of disease
• Less chance of exposure changes
• Not assessing determinants of survival
• The ideal control is a person who, were they to
develop the disease, would have become a case
• Selection – not related to key exposure (flying)
• Participation – willing to take part?
• Information gathering – can it be same as in cases?
• Confounding – similar confounder exposure as in
cases, consider `matching’ to cases
When assessing exposure in a case control study…
Ensure that opportunity to recall exposure is similar
in cases and controls
How to ensure assessment of exposure is not biased?
OBJECTIVITY is important….
-Participants should be blind to hypothesis
-Use an objective assessment method if possible
-Self-administered questionnaire ideal; if an
interviewer/observer used, should be blind to
hypothesis and case-control status if possible….
-Standard protocol, interviewer training etc
Ferrari study used single interviewer, pre-specified
questions, travel >4 hrs within 4 wks (incl air travel)
How do we deal with possible confounding at the DESIGN stage of the study?
Aim here is to make sure that confounding factor is as
evenly shared between comparison groups as
possible.
– Match each case and control so that level of
confounder exposure similar – as in Ferrari study
– Do the whole study in people at the same level of
confounder (e.g. for smoking, do the study in non-
smokers completely)
Analysis of case control study
Odds ratio
odds of being a case in those exposed/ odds of being a case in those non-exposed
= odds ratio
exposed cases x non exposed controls
OVER
exposed controls x non exposed cases
How to deal with confounding at analysis stage?
• Use logistic regression (regression with 0/1 outcome)
• Can then adjust for other confounders (assuming they were measured)….
• Analysis examines relationship between exposure
and outcome at each level (stratum) of confounder
What do case control studies NOT provide?
• Case control studies do NOT provide any indication of disease incidence (development of new disease)
• This is because they do not follow individuals over a period of time
• In contrast, both cohorts and randomized trials do
provide an indication of disease incidence
Potential biases in case control studies
information bias - In which the information on exposure obtained from
the cases and controls is not comparable and
systematically influences exposure assessment
selection bias - In which the selection of either the cases or the controls (or both) is systematically related to the
exposure being examined (here air travel)
How to minimise selection bias in CCS
Cases and controls drawn from a similar population
insofar as possible, and in a way which will not affect
their risk of exposure
How to minimise information bias in CCS
Ensure information on exposure is obtained from
cases and controls in exactly the same way, both
have the same opportunity to recall exposure
Interpreting positive results in CCS
Could be a true association (e.g. between air travel and VTE – but also consider:
• Could an association be due to chance?
• Could an association be due to bias?
• Could an association be due to confounding?
• Could there be reverse causation? (presence of
disease has affected exposure, or the estimation of
exposure)
Interpreting negative results in CCS
Could be a true null association – but also consider:
• Could the study be too small to detect an association?
(limited statistical power?)
• Could the measurement of the exposure have been
inaccurate?
• Could the absence of association reflect bias or
confounding which have obscured an association?
What is a strong RR (relative risk)?
RR of 4 reasonably strong, consistent with causality
What could suggest an association is causal?
- association is independent (not explained by confounding factors)
- consistency of results in different study types (CCs and CH studies)
- dose-response relationship
- association is biologically plausible
What is key for defining causality?
Relative risk
What is important for decision making in the individual patient?
Attribute risk (excess risk)
Attribute risk in long-haul air travel
Attributable risk (excess risk) =
Absolute risk in those exposed (to air travel)
MINUS
Absolute Risk in those not exposed (to air travel)
(baseline risk)
Attributable risk provides an indication of the
extra risk of VTE (due to air travel) in the individual
Attributable risk strongly influenced by baseline
risk in the individual (risk of VTE before air travel)
Do the increased risks of flying matter?
The relative risk of long-haul flying is fairly consistent
(~4x) for a wide range of individuals
• The key thing is the attributable risk, which is strongly influenced by baseline risk (i.e. what was the person’s
risk of VTE before getting on the plane?)
• So if your patient already has high VTE risk before
boarding a plane, that is a concern.
This was the case for Angela Robinson – she
already had a high risk of VTE before she got on
the aeroplane, because of:-
– older age
– use of hormone replacement therapy
– inherited thrombophilia
– overweight
WHAT FACTORS INFLUENCE BASELINE VTE
RISK OF A POTENTIAL AIR TRAVELLER?
• Age: older person at increased risk • Previous VTE (increases relative risk by ~ 8x) • Pre-existing illness (cardiac/cancer/stroke) • Recent recovery from trauma/surgery • Pregnancy • Use of oral contraceptives /hormone replacement • Inherited thrombophilia • Obesity • Cigarette smoker
WHAT MEASURES COULD BE TAKEN TO
PREVENT FLIGHT-RELATED VTE ? (specific)
SPECIFIC MEASURES FOR PEOPLE AT HIGH RISK
- Avoidance of flights (especially long-haul flights) where possible: discuss risk with expert
- Use of elastic compression stockings
- Consider taking anti-thrombotic medication (e.g. aspirin) in advance (?effectiveness)
- Subcutaneous low molecular weight heparin (as used in operations associated with high VTE risk)
WHAT MEASURES COULD BE TAKEN TO
PREVENT FLIGHT-RELATED VTE ? (general)
GENERAL MEASURES (ALL SUBJECTS, INCLUDING THOSE AT LOW AND HIGH RISK)
• Lower limb movement during flight
– exercise calf muscles while seated
– walking about regularly
• Keep well hydrated
– drink water
– avoid alcohol, coffee
Not easy on economy flights!
Attribute risk calculation
What is a risk factor?
Aspect personal behaviour or lifestyle, environmental exposure, inborn or inherited characteristic associated with particular disease or condition.
What is an association?
Association: Typically considered adverse (↑ risk).
- Example: Smoking risk factor for lung cancer;
- Association: May be causal.
Exposure is…
Quantification of any change in risk
Cohort study (prospective design)
Take a cohort from population
Are they exposed to risk factor (YES/NO)
Follow them through time and see if they develop disease or not
Then compare two groups in disease rates to see if there is a difference in those exposed to risk factor and those that were not. (so if you smoke, do more of these people get lung cancer?)
Must have everyone disease free at the start and follow them through time (eg you wouldnt want a patient with lung cancer joining the study, would defeat the purpose of the study)
Wisconsin Sleep Cohort study
Followed everyone up every 4 years (even though recruited at different times)
Measurements of sleep, resp and cardiovascular factors to see if association exists between sleep patterns and impact on health
Was the cohort of the Wisconsin sleep cohort study a representative sample?
They were chosen from payroll records- state of Wisconsin employees, men and women age 30-60. So automatically some difference between sample and general population = SELECTION BIAS
Also only 50% of people invited to study accepted so VOLUNTEER BIAS a systematic difference between people who accepted and those that didn’t.
Healthy entrant effect
The healthy entrant effect is a reduction in rates of morbidity and mortality in the initial stages of a longitudinal study in comparison with the general population because only healthy people were recruited to the study
Temporal association
Risk factor prior to outcome
Exposure must precede the disease, and in most epidemiological. studies this can be inferred. In studies where exposure and disease are measured simultaneously or exposure is measured after the occurrence of disease, the temporal association should be evaluated.
Comments on this data
16.2% 12.7% 19.8%
If not exposed to risk factor: 16.2% of people have CVD event
If early onset of asthma: 12.7% of people have CVD event, LOWER
If late onset of asthma: 19.8% of people have CVD event, HIGHER
So is early onset a protective factor against CVD? And late onset is detrimental exposure?
CONSIDER: 7 is a small number to be drawing conclusions about protective factors, not a big enough sample (here, adding one more person will increase the percentage by a lot- not very accurate)
What is absolute risk?
How we quantify risk
It is the ratio of people who have a medical event compared to all of the people who could have an event. For example, if 26 out of 100 people will get dementia in their lifetime, the absolute risk is 26/100 or 26%.
What do we compare in research?
Compare exposed risk to unexposed
eg, risk CVD is not-asthmatic: 0.162 (16.2%) (r1
What is an absolute comparison?
Absolute Comparison: Observational (cohort) studies- so refer to attributable risk
Attributable risk: Extra or excess (absolute) risk due to exposure;
-Risk CVD if late-onset: 0.198 (19.8%) (r3); Extra 3.6%
so the risk the individual will have if exposed
Attributable risk reduction: Reduced (absolute) risk due to exposure.
-Risk CVD if early-onset: 0.127 (12.7%) (r2); Reduction 3.5%
What is a relative comparison?
Relative Comparison: Experimental or observational studies
- Risk CVD Event (Non): r1 = 0.162
- Risk CVD Event (Early-onset): r2 = 0.127
- Risk CVD Event (Late-onset): r3 = 0.198
What do we derive from relative comparison?
Derive Relative Risk (RR):
- Ratio of absolute risks: Exposed to unexposed group.
- Compares risk in two groups as “multiples”.
How do we get relative risk of disease from this data?
Current/past = exposed Never = unexposed
10% current/past developed disease (lung cancer) compared to 5% unexposed
So the two risks are 0.10 and 0.05, can derive relative risk of disease (here lung cancer)
RRcancer = Rexp/Runexp = 0.10/0.05 = 2.0
How to interpret relative risk data?
Interpretation of Relative Risk (RR) Relative risk (RR) lung cancer if smoker (current/past) to never = 2.0
Current/past smokers are 2.0 times AS likely to develop lung cancer compared to never smokers- important to state baseline (reference category)
1.0 times MORE likely (baseline is already 1.0 times likely) because risk has increased from 5% to 10%, which is one times more
Increase is 100% from never risk- so relatively increased risk of 100% of lung cancer compared to non-smokers
Presentation of risks: relative risks
Unexposed: present as 1.0
Exposed: as calculated (0.78 etc)
up to 2 decimal places
Think about decreased and increased risks:
From 1.0 -> 0.78 (relative change of 0.22 so “relative decrease of 22%”
From 1.0 -> 1.22 (0.22 excess risk relative to unexposed) “relative increase of 22%”
Relative to unexposed
Define confounding factor
Associated with risk factor:
- Without being consequence of it. (eg, BMI does not CAUSE asthma- consequence of sampling)
Associated with disease:
- Independently of risk factor. No causative link
What happens to relative risk if there is confounding?
RRs typically adjusted for confounding (sophisticated methods)
When unadjusted: crude/raw
When adjusting for confounding what happens to all other factors?
They are all equal- average out
What’s a big problem with cohort studies and factors?
Loads of possible risk factors not measured, contributing to disease but can’t measure so a set of unknown confounders
So difficult to infer causality because these unknown risk factors leading/causing the disease not accounted for
Relative risk can be used in…
Observational and experimental studies
Length of exposure: if not fixed then what do we look at?
Incidence rates: the total number of occurrences of the outcome during follow up, divided by the total person who is at risk.
So if one person in study for 20 years, another for 40, add years of exposure together = total number of years of exposure
Then see, how many incidences of disease per 1000 etc years of exposure?
Leads to incidence rate ratios- comparative statistic
Pros and cons of cohort studies
Application and communication of knowledge: CH studies
