9 - Observational Research 1-3** Flashcards
What are observational studies?
- Studies of drugs in large groups
- Like pharmacoepidemiology
What is the Bradford Hill criteria of causality to validate cause and effect?
- Biologically plausible
- Be strong
- Reflect a biological gradient (dose response relationship)
- Be found consistently
- Holds over time (temporal incidence of the disease should reflect prevalence of offending agent)
- Confirmed by experiment
What is a case report?
- Medical hx of a single patient in “story” form
- Usually b/c some odd, one-off case stands out from typical medical case
Advantages of case reports
- Fast
- Individual detail
Disadvantages of case reports
- Not an experiment
- Not controlled
- Weak evidence, bottom of evidence hierarchy
What is a case series?
Group of case reports
What do cross sectional studies determine?
Prevalence (proportion of the population w/ a disease/risk factor at one point in time)
What is the difference between cross sectional, cohort, and case-control studies?
- Cross section = present disease and exposure
- Cohort = present exposure and future disease
- Case control = present disease and past exposure
Advantages of cross sectional studies
- Cheap, fast
- Collect info on disease and risk factors at the same time
Disadvantages of cross sectional studies
- Establishes association, not causality
- Recall bias issue (ex: surveys)
- Confounders unequally distributed
- Group sizes are unequal
Describe case control studies
- Compare an exposure or risk factor between cases/ controls
- Compute an odds ratio, 95% confidence interval, and p-value
What is an example of a question posed in a case control study?
Were individuals seen at a pediatric oncology clinic at increased odds for renal injury if they were exposed to vancomycin?
What is an odds ratio?
- An effect estimate
- Measures association between an exposure and outcome, risk factor and outcome, or a predictor and outcome
- Ex: exposure (drug) and outcome (drug response; adverse event)
What does an odds ratio of 1 mean?
- Exposure doesn’t affect odds of outcome
- Equal odds for both groups
What does an odds ratio above 1 mean?
- Exposure associated w/ higher odds of outcome
- Ex: OR = 2 means that those w/ vancomycin were at 2x increased odds for renal injury compared to those who didn’t take vancomycin
What does an odds ratio less than 1 mean?
- Exposure associated w/ lower odds of outcome
- Ex: OR = 0.5 means that those w/ vancomycin were at 50% decreased odds for renal injury compared to those who didn’t take vancomycin
What is the purpose of confidence intervals?
- Used to estimate the precision of the OR
- Large CI indicates low level of precision of OR
- Small CI indicates high precision
- Unlike p-value, 95% CI doesn’t report a measure’s statistical significance
- Often 95% CI is used as a proxy for statistical significance if it doesn’t overlap OR = 1
How does sample size affect CI?
Small sample size = wider CI and vice versa
How do you calculate odds ratio?
- OR = ad/bc
- A = # of px w/ risk factor and disease
- B = # of px w/ risk factor but no disease
- C = # of px w/ no risk factor but disease present
- D = # of px w/ no risk factor and no disease
Advantages of case control studies
- Good for rare diseases/ outcomes (able to “enrich” for outcome)
- Good for outcomes w/ long latency
- Quicker to do
- No need for follow up
- Smaller sample size
- Cheaper
Disadvantages of case control studies
- Difficult for rare exposures
- Small sample size
- Sometimes difficult to find appropriate control group
- Can’t estimate risks
- Interviewer bias
- Recall bias
Describe cohort studies
- Begin w/ sample (“healthy cohort”; subjects w/o the outcome yet)
- Start w/ exposure status, then compare subsequent disease experience in exposed vs. unexposed
What is the largest difference between case control and cohort studies?
- When the outcome is measured
- Case control = before
- Cohort = after
Describe prospective cohort studies
- Have a population (large, undefined size)
- Sample from population
- Measure predictor variable
- Measure outcome/ disease in follow up
Advantages of prospective cohort studies
- Can study multiple outcomes from a risk factor
- Prospective, so can measure variables completely and accurately
- Fatal diseases are easier to study prospectively than retrospectively
- Very useful when exposure is rare
- Time sequence strengthens inference
- Can provide a strong suggestion of causality
Disadvantages of prospective cohort studies
- Expensive and time consuming
- Need large # for a long period of time to have sufficient events to produce meaningful results
- Associations found may be due to confounding factors (ex: alcohol may be a confounder in a study of smoking)
- Loss to follow up a problem
Describe retrospective cohort studies
- Design essentially the same as prospective except that baseline measurements, follow up and outcomes have all occurred in the past
- Need adequate data about risk factors and outcomes assembled for other purposes
- Identify a cohort that has been assembled in the past, collect data on predictor variable, and collect outcome/disease data in the past or present
Advantages to retrospective cohort studies
- Measurements collected before the outcomes are known; not biased by outcome of interest
- Less costly and time consuming than prospective studies (measurements already taken and follow-up period already occurred)
Disadvantages to retrospective cohort studies
- Limited control over the nature and quality of the predictor variables
- Existing data may not include subjects or variables that are important to answering research questions
- Inaccurate or incomplete variable measurement
What is a hazard ratio?
- Measures association between an exposure and an outcome
- Ex: exposure (drug) and outcome (drug response; AE)
- Similar to odds ratio but better
- Represents instantaneous event rate (probability individual would experience an event at a particular time after intervention)
What does a hazard ratio of 0.5 mean?
At any particular time, 1/2 as many px in the tx group are experiencing an event compared to the control group
What does HR = 1 mean?
At any particular time, event rates are the same in both groups
What does HR = 2 mean?
At any particular time, twice as many px in the tx group are experiencing an event compared to the control group
Describe nested case control studies
- Start w/ a suitable cohort w/ enough cases to provide adequate statistical power
- Define criteria to identify outcome of interest
- Identify individuals in the cohort that developed the outcome of interest (called “cases”)
- Select individuals that were also part of the cohort that haven’t developed the outcome of interest (“controls”)
- Retrieve samples or records from before outcome (potential predictor variables and compares levels between cases and controls)
When is a variable considered a confounder?
If it:
- Is independently associated w/ outcome
- Is associated w/ exposure
- Doesn’t lie in the causal pathway
When does unmeasured confounding occur?
Occurs in a variable that wasn’t measured during the course of the study
Can bias be completely eliminated?
No, only minimized
What is recall bias?
People remember things more readily when they have a negative outcome (especially when recalling exposures)
What is selection bias?
Selecting groups from different places
What is channeling bias?
Giving drug to sicker px
What is misclassification bias?
- Outcome/ exposure misclassified
- Cases and controls were erroneously classified
