Flashcards in Evidence Based Medicine Deck (27)
Case control study
-retrospective, observational study which selects patients with the outcome (cases) and patients without the outcome (controls) and attempts to find exposures linked to the outcome
--> recall bias
-possible in rare outcomes/illnesses
-observation does not equal causationStrength of association: Relative Risk 1.01-1.15 weak, 1.51-3 moderate, >3 strong
-observational study which has two groups (cohorts) that are observed over time for an outcome (prospective), with one cohort having an exposure (condition or treatment or risk) that the other doesn't
-takes a long time, expensive
-not good for rare things
-good for things that can't be made into a RCT
Strength of association: Relative Risk 1.01-1.15 weak, 1.51-3 moderate, >3 strong
each patient receives both treatments in two phases, separated by a wash out period. Each patient serves as his own control, thus less variability in outcomes and smaller sample sizes are required, period effects may limit findings
Randomized Control Trial (RCT)
-prospective study in which patients are randomized to treatment or control groups (equal chance of being assigned to any), groups are followed for outcome of interest
-systematic collection, review and presentation of available questions addressing a clinical question using specific criteria and methods, may include meta-analysis
combining studies meeting pre-specified criteria and addressing a clinical question. Results are calculated from the data of each study. Data is pooled. Higher sample size and statistical power is useful if the individual trials are underpowered or subgroup analysis
-design flaws leading to over or under estimation of treatment effect
-eg recall bias, selection bias, publication bias, confounding factors--> especially in observational studies
if investigators or patients are unaware of who receives treatment vs control, they are less likely to inappropriately report better results with treatment
the number of people experiencing the event as a proportion of the total number of people in the population or group
-experimental ER= (number of events in exp group)/(total in exp group)
-control ER=(number of events in ctrl group)/(total in ctrl group)
RR= Experimental event rate/control event rate
Relative Risk Reduction
= 1-relative risk
-more consistent between populations than absolute risk reduction
Absolute risk reduction
ARR= control event rate- exp group event rate
if the risk increases- absolute risk increase
-more variable than relative risk reduction depending on population treated
Number needed to treat
the number of patients who would need to be treated with the study intervention for the study time for 1 of them to benefit
NNT= 100/(absolute risk reduction %)
Number needed to harm
number of patients who would have to be treated with the study intervention for the study time for 1 extra person to experience an adverse event
NNH= (100/ (absolute risk increase %)
OR= (experimental events odds)/(control events odds)
-can be used in obersvational studies where base risk is unknown, also used in meta-analysis.
-when event are rare, is similar to RR, but exaggerated relative to RR with common events
trial result is used as best estimate of true effect
like RR but more accurate-- accounts for the time each participant was in the study before having event or withdrawing
a 95% CI provides the range of values that we are 95% certain the true value is within-- indicates the precision of the estimate
-for ratios-- CI including 1 means possibility of no difference
-for absolute risk reduction, absolute risk increase, NNT/NNH-- a CI including 0 means possibility of no difference
-even if data is not significant, trends can suggest areas for future investigation
type 1 error (alpha)
the false positive-- ie finding a difference where there is none
p-value is a measure of how likely a type 1 error is
p< 0.05 means that there is a < 1/20 chance that the difference is due to chance
-smaller p values means that a type 1 error is less likely (less likely the result is due to chance)
Type 2 error (beta)
the false negative-- to conclude there is no difference where there really is a difference
-- eg inadequate power (not enough patients enrolled to detect a difference)
when studies within a meta-analysis have more variation than expected
-may be inappropriate to combine study
1. Is the study valid?
-randomization and treatment allocation concealed?
-controlled? was it appropriate?
-were the treatment and control groups similar? any differences contributing to confounding?
-were all patients accounted for?
-was data analyzed based on intent to treat (more real world), or per protocol?
-were the groups treated similarly except for the intervention?
-was there bias from the authors? or with funding?
2. What are the study results?
-what was the primary and secondary endpoints? were they pre-determined?
-was there a difference between groups? benefits? harms? -were the differences statistically significant? clinically significant?
-what are the relative and absolute risk reductions/increases?
-what is the number needed to treat/harm?
3. Does this study matter to my patients?
-are the outcomes clinically relevant/important?
-were the patients similar to mine? consider inclusion and exclusion criteria-- often complex or very sick patients excluded
-do the treatment benefits outweigh the risks, costs and impact on life?
-% of positive tests of everyone who has the disease
= true positive/ (true positive + false negative)
-snout-- how useful is the test to rule out (how good is it at picking up a positive)
% of negative tests of everyone who does not have the disease
=true negatives/ (true negatives + false positives)
-spin= rule in-- how good is the test a picking ups true negative
Positive predictive value
% of positive results that are true-- ability to predict a positive result
(true positive)/(true positive + false positive)
Negative predictive value
% of negative results that are true-- ability to predict a negative result
(true negatives)/(true negative + false negative)