Questions Flashcards
What is the relationship between prevalence, incidence and duration?
Prevalence = Incidence x Average Duration Incidence = Prevalence / Duration Duration = Prevalence / Incidence
What is the difference btwn odds and probability?
Odds is a ratio (A/B), probability is a proportion (A/A+B)
How do you control confounding?
At the design stage: randomization, restriction, matching
At the analysis stage: stratification, multivariable adjustment (logistic regressions)
How do you control bias?
- must control selection and measurement (information?) bias at design stage (representative sampling from same pop, and blinding)
- once study is done, it’s too late; must just assess likelihood of bias
Which study design is best for measuring prevalence?
Cross-sectional
Which study design is best for risk of harm?
cohort, case control
Which study design is best for treatment or prevention?
RCT, cohort, case control
Which study design is best for prognosis?
cohort
Which study design is best for screening?
RCT, cohort, case control
What are the strengths and weaknesses of cohort studies?
Strengths (4): good for relatively rare exposures, can minimize selection and measurement bias, can directly determine incidence rate and risk, can look at multiple outcomes from a single exposure
Weaknesses (6): ineffective for rare outcomes, often requires large sample size, takes a long time, expensive, potential ethical issues
Who was Karl Popper?
- associated with deductive approach to epidemiology: science advances only through disproofs, not proofs (rejecting null hypotheses)
- hypothesis must be testable/falsifiable; must frame a question that can be tested and rejected or not rejected by your data
What type of errors can False Positives and False Negatives create, respectively?
FP –> Type I
FN –> Type II
What determines Power?
our choice of alpha our choice of beta underlying prevalence of the condition magnitude of effect we hope to find sample size we can gather (usually more subjects --> more power; the smaller effect size you want to detect, the larger the sample size needed - must predict an effect size in order to calculate power)
What reduces the power of a study?
- inadequate sample size (hard to recruit, loss to follow up, misclassification; random misclassification biases toward the null)
- making alpha too stringent
- making beta too lax
- finding a study pop with low prevalence
- adjusting for multiple comparisons unnecessarily (Bonferroni adjustment)
How do you solve selection bias?
randomization
