Evidence-based Medicine Flashcards
For prognosis questions, the perfect study is a large
Prospective cohort study
For diagnosis questions, the perfect study for “instant” diagnostics is a large
Cross-sectional study
For “follow-up” diagnostic questions, the perfect study is a large
Prospective cohort
For intervention/exposure questions, the perfect study is a large
RCT
How are diagnostic, prognostic study outcomes reported?
Sensitivity, specificity, LR, and ROC statistic
We can also use predictive value in certain populations to report the diagnostic/prognostic study outcomes. But this will be
Population-specific
Estimate of precision of results
Confidence interval (CI)
How are the intervention/exposure study outcomes reported?
ARR, RRR, NNT/H/S, and also P value
Probability of finding this level of association when the null hypothesis is true, due completely to CHANCE
P value
The conventional cut-off value for statistical significance of P is?
Less than 0.05
Provides an estimate of the precision of the results AND statistical significance (how far away is the interval from the null result)
Confidence interval
Refers to how well an experiment is done, especially whether it avoids confounding (more than one possible independent variable [cause] acting at the same time)
Internal validity
Are your patients and situation similar enough that you think you can apply the results to them? I.e. is there generalizability?
External validity
What are three major flaws to an observational study?
Confounding, selection bias, recall bias
What are some of the problems we see with RCT’s?
Expensive and time-consuming, randomization/concealment issues, and inadequate blinding/placebo
If patients are taking a drug in a trial but have to drop out because of the side effects, we need to perform
Intention-to-treat analysis
Good for hypothesis-generating. Generates “translational” research
Bench/animal research
Points out rare or new threats, e.g. AIDS, Zika, unknown drug effects
-Hypothesis generating
Case series
Snapshot of individuals in a population
–Can directly measure or survey
Cross-sectional study
Survey outcomes subject to recall bias, and all are subject to confounding
Cross-sectional study
Best for evaluating a test with instant reference, establishing prevalence
-Can look for time trends
Cross-sectional study
Snapshot of population inputs and outcomes
Ecological study
Best for hypothesis generation and risky exposures
Ecological study
Best for - uncommon outcomes (e.g. folate decreasing neural tube defects); and risky exposures
Case control study
Good for hypothesis generation and can also be hypothesis-testing
Case-control study
One of the major disadvantages of a case-control study is that it is probably most subject to
Confounding/bias
Best for – diagnosis or prognosis questions based on information usually captured in routine care; risky or rare exposures; long time-frame issues (e.g. radiation exposure)
Retrospective Cohort study
Subject to confounding, missing information, and loss to follow-up
Retrospective cohort study
Best for – diagnostic test or prognosis questions; risky or rare exposures
-Ex: Framingham heart study
Prospective cohort
Expensive/time-consuming and subject to confounding and loss to follow-up
Prospective cohort study
Really a type of retrospective cohort, but without any individual records review
-Best for – rare outcomes, post-marketing surveillance
“Big Data” studies (Ex: pharmacoepidemiology)
Best for – common illnesses; efficacy and common harms of interventions: treatment, screening, prevention, or diagnostic/therapeutic strategy
RCT
Disadvantages – therapeutic equipoise limits; expensive/time-consuming; external validity; surrogate or composite outcomes; cross-overs, non-adherence, co-interventions
RCT
Best for – questions lacking a good large study, or with multiple high-quality studies
Systematic review