Step 3 biostats Flashcards
Disease
Test (or exposure
N/A
Probability of an event in a given time (ex. chance getting heart attack in 10 years)
Absolute risk
Evaluate cohort study. Compares INCIDENCE in exposed vs not exposed [chance develop dx after exposure]
Relative Risk… < 1 means less likely in exposure group (ex aspirin in MI)
compares rate of EXPOSURE in those with and w/o dx [chance group w/ dx was exposed], Used in case control (retrospective) study
Odds ratio (rare dx OR=RR)
Measures risk due to exposure (used in randomized controlled studies) [difference in risk between control and t=intervention]
ARR (absolute risk reduction or attributable risk)
Calculate RR
{a/(a+b)}/{c/(c+d)}
Calculate OR
ad/bc
Calculate PPV
a/a+b (TP/TP+FP)
Calculate NPV
c/c+d TN/TN+FN
Calculate ARR
(c/c+d) / (a/a+b)
the absolute risk (rate of adverse event) in placebo group minus absolute risk in treated patients
[ex. risk bleed w/o aspirin vs aspirin]
Calculate NNT
1/ARR
Ratio between 2 risks (used in randomized controlled studies)
RRR (relative risk reduction)
Calculate RRR
(event in control - event in experiment) / event in control
1-RR
Which is more usesful in clinical setting ARR or RRR
ARR (states difference in SE is 0.01-0.004= 0.006%)
whereas RRR states 60% decrease
“Intention to treat” does what?
Preserves randomization, decreases crossover and dropout bias (records effect of everything on both groups)
Standardized mortality ratio
Observed number of deaths/expected number of deaths (1.75 means 75% higher death rate)
Standardized incidence ratio
used to determine if occurrence of cancer is high or low relative to expected. (observed/expected)
hazard ratio
Similar to relative risks but can be obtained at multiple time intervals
Sensitivity analysis
repeating primary analysis calculations after modifying certain criteria or variable range (remove outliers)
Linear regression
models the linear relationship between a dependent variable and one or more independent variables (ex. alcohol and tobacco on gastric cancer[independent])
Propensity scoring
weighs different variables (eg, severity of different comorbidities) to balance control and tx groups
Matching
Linking a control and treatment subject based on propensity scoring
Effect modification
external variable (effect modifier) has a positive or negative impact on the observed effect of risk factor (exposure) on disease status (outcome) [ex. aspririn causes reye syndrome in kids but not adults]
How to detect effect modification
Stratification. separate measures of outcome should be reported for each stratum [ex effect on DM vs not DM]
What is a likelihood ratio (LR)
Probability of a given test result in a patient with the disorder compared to result in patient w/o disorder. Do not change with changing prevalence.
An expression of sensitivity and specificity that can be used to assess the value of a diagnostic test. The positive likelihood ratio (LR+) represents the value of a positive test result, and the negative likelihood ratio (LR-) represents the value of a negative test result. [>5 is likely to have disease
Calculate Positive likelihood ratio [basically true positive]
LR+ = sensitivity / (1 - specificity)
Calculate Negative likelihood ratio [basically true negative]
LR- = (1 - sensitivity) / specificity
What is restriction
Limiting a study to people with a specific trait
Hawthorne effect
Do better because they know they are being watched
Length-time bias
Survival benefits of screening test overstated due to detection of disproportionate number of slowly growing benign cases (whereas fast growing may not be detected)
Lead-time bias
Test diagnoses a disease earlier and therefore shows increased survival rate (not benign cases)
Type 1 error
False positive (incorrect reject the null)
Type II error
False negative (incorrect accept the null)
Standard mortality ratio
Observed/expected deaths (>1 is higher risk <1 is lower risk)
False positive rate
=1-specificity
False negative rate
=1-sensitivity
Calculation used in Cohort study
RR
Calculation used in Case Control study
OR
Bias that uses gold standard test ‘selectively’ to confirm a result (can over or underestimate)
Verification bias
Type of selection bias where control and tx groups are different in their prognostic standpoint d/t unforseen counfounders
Susceptibility bias
Bias when control group accidentally receives the tx
Contamination bias
