Biostats - First Aid Flashcards
Cross Sectional Study
Observational.
Collect data from a group of people to assess frequency (f) of disease at a particular time.
It’s a screenshot in time that looks at disease prevalence.
Can show RF assoc. w/ disease but doesn’t establish causality.
Case Control Study
Observational. Retrospective.
Compares group of people with disease to group without disease, then looks for prior exposure/RF.
measures odds ratio.
Cohort study
Observational.
Prospective: groups with RF/exposure compared to group w/o RF/exposure. Later down the line look for disease vs. no disease development.
Retrospective: looks at exposure vs. no exposure, then looks forward to see who eventually developed the disease vs did not develop the disease.
measures relative risk.
Twin concordance study
compares frequency w/which both MZ twins or DZ twins develop same disease.
Measure heritability and influence of environmental factors (nature v. nurture)
Adoption study
compares siblings raised by biological vs adoptive parents.
Measure heritability and influence of environmental factors.
Ecologic Study
The unit of analysis in ecological studies are populations, not individuals!
Studied using population data - they are useful to generate hypotheses but should not be used to make conclusions regarding individuals w/in these populations (ecological fallacy)
Clinical Trial
Experimental Study.
Compares benefits of 2 or more treatments or tx vs. placebo.
Study quality increases if randomized, controlled, and double blinded.
Triple blinding offers additional benfit, in with researchers analyzing data are also unaware.
Phases of Clinical Trials
Pre-clinical: tested on animals
Phase 1: Small number of healthy volunteers to test safety, toxicity, PKs, and PDs
Phase 2: small # pts with disease of interest. Tests if drug works - assess efficacy optimal dosing, and AEs.
Phase 3: large # pts randomly assigned to either tx under investigation or best available tx/placebo. Asks if the drug is good or better.
Phase 4: Postmarketing surveillance of pts after tx approved. Asks if the drug can stay on market, bc it can detect rare/long term AEs. Can result in drug withdrawal if harmful.
Sensitivity
Given you have disease, what’s the probability that you test positive?
High sensitivity when negative, rules disease out (SNOUT).
Sensitivity is fixed property of a test (stays fixed regardless of prevalence).
Specificity
Given you do not have the disease, what’s the probability that you test negative?
High specificity when positive, rules disease in (SPIN).
Specificity is fixed property of a test (stays fixed regardless of prevalence)
Positivity predictive value (PPV)
Given a positive test result, what’s the probability that you truly have the disease?
PPV varies directly with prevalence - as the prevalence of disease increases, so does the PPV.
Negative predictive value (NPV)
Given a negative test result, what’s the probability that you truly do not have the disease?
NPV varies directly with prevalence - as the prevalence of disease increases, the NPV decreases!
Incidence
# new cases / # people at risk Looks new cases during a period of time.
Prevalence
#existing cases / #people at risk Looks at all current cases. *For a short duration disease (like flu), prevalence can equal incidence
Odds ratio (OR)
Quantifying Risk.
Typically used in case - control studies.
Odds that the group with the disease (cases) was exposed to a RF, divided by odds that group w/o disease (control) was exposed to the same risk factor.
OR = (a/c) / (b/d)
Relative Risk (RR)
Quantifying Risk.
Typically used in cohort studies.
Risk of developing disease in the exposed group divided by risk in unexposed group.
Note that if prevalence is low, then OR = RR.
RR = (a/a+b) / (c/c+d)
Ex. If 2% of pts who get flu shot develop the flu vs. 8% of pts who do not get the flu shot develop the flu, then RR = 2/8 = .25
Ex. If 21% smokers develop lung cancer vs. 1% nonsmokers develop lung cancer, then RR = 21/1 = 21.
Attributable Risk (AR)
Difference in risk between exposed and unexposed groups, or the proportion of disease occurrence that are attributable to the exposure.
AR = (a/a+b) - (c/c+d) AR = (RR - 1)/RR
Ex. If 21% smokers get lung cancer, vs. 1% nonsmokers, then 21-1 = 20% lung cancer in smokers is attributable to smoking
Relative Risk Reduction (RRR)
Proportion of risk reduction attributable to the intervention compared to control.
Ex. If 2% of pts who get flu shot develop the flu vs. 8% of pts who do not get the flu shot develop the flu, then RR = 2/8 = .25, and RRR = .75
RRR = 1 - RR
Absolute Risk Reduction (ARR)
Difference in risk attributable to intervention as compared to control. Ex. if 8% people who get placebo vaccine get flu, vs 2% who get flu vaccine get flu, then ARR = 6%.
ARR = (c/c+d) - (a/a+b)
Number needed to treat (NNT)
Number of pts who need to be treated for 1 pt to benefit.
NNT = 1/ARR
Number needed to harm (NNH)
Number of pts who need to be exposed to risk factor for 1 pt to be harmed.
NNH = 1/AR
Precision
Reliability.
The consistency and reproducibility of a test.
*The absence of random variation in a test.
Note that presence increases as standard deviation decreases, and as statistical power increases.
Random error decreases precision of a test.
Accuracy
Validity.
The trueness of test measurements (as compared to the “gold standard”)
The absence of systemic error bias in a test.
Note that systemic error decreases accuracy of a test.
Selection Bias
Sampling/Referral bias: Sample doesn’t represent the population.
Susceptibility bias: Intervention based on disease severity.
Attrition bias: Different growth withdrawals - ex. if one intervention gives bad SE, then that group may have greater # participants leave.
Berkson bias: study population selected from hospital is less healthy than general population
Healthy worker effect: study population is healthier than general population.
Non-response bias: participating subjects differ from nonrespondants in meaningful ways.
Recall bias
Inaccurate recall of past exposures by subject.
Awareness of d/o alters recall by subjects - this is common in retrospective studies
Pts with disease recall exposure after learning of similar cases!
Measurement bias
Info is gathered in a way that distorts it - miscalibrated scale consistently overstates weight, etc.
Need to use a standardized data collection method.
Procedure bias
Subjects in different groups are not treated the same.
Ex. pts in treatment groups may spend more time in highly specialized hospital units
Observer expectancy bias/Pygmalion effect
Self fulfilling prophecy.
Researcher’s beliefs in efficacy of treatment can be potentially affect the outcome.
Or if observer expects treatment group to show signs of recovery, then he is more likely to document those types of outcomes.
Ex. teachers told students have higher IQs and expect more from the group, but the kids in “higher IQ” groups performed better probably bc the teachers unconsiously behaved in a way that would facilitate their success.
Hawthorne Effect / Observer Effect
Tendency of study subjects to change their behavior as result of their awareness that they are being studied. Affects the validity of the study.
Commonly seen in studies concerning behavioral outcomes or outcomes that can be influenced by behavioral changes.
Confounding bias
When a factor is related to both the exposure and outcome, but not on the causal pathway…the factor distorts or confuses effect of exposure.
Ex. pulmonary disease is more common in coal workers than general population, but people who work in coal mines also smoke more frequently than general population.
Lead time bias
Apparent prolongation of survival after applying screening test, but really just detects disease earlier w/o adding to prognosis
