Stats/Ethics Flashcards
What variables to use mean for
- Interval (ordered, numerical, where zero means nothing)
- Ratio (ordered, numerical, where zero means something)
What variables to use median for
- Ordinal (ordered, named categories)
- Non-normally distributed interval/ratio
What variables to use mode for
- Nominal (named categories like gender)
Normal distribution has what percentage with 1 SD vs 2 SD
68% with mean +/- 1 SD
95% with mean +/- 2 SD
Calculation for standard error of mean
= Sample standard deviation / sq root of sample size
Relative risk calculation
- Chance of getting disease with treatment relative to placebo
(Treated with disease / all treated) / (Placebo with disease / all placebo)
Odds ratio calculation
- Number of disease in treated / number of diseased in placebo
Number needed to treat calculation
1 / risk difference
Absolute risk reduction
Difference in risk between treatment and placebo
= (Diseased placebo / all placebo) / (Diseased treated / all treated)
Sensitivity calculation
Given a patient with disease, what’s the chance the test will be positive
- True + / All with disease
- SnOUT - 100% sensitive rules OUT disease
- Good screening tests have high sensitivity
- Higher sensitivity means higher false positive
Specificity calculation
Given a healthy patient, what’s the chance test will be negative
- True - / All healthy
- SpIN - 100% specific rules IN disease
- Higher specificity is higher false negative
Positive predictive value calculation
True + / All + test
Negative predictive value calculation
True - / All - test
Likelihood ratio calculations
Chance of a positive (or negative) test result in patient with disease relative to patient without disease
LR + = (True pos / all disease) / (False positive / all healthy)
LR - = (False neg / all disease) / (True neg / all healthy)
Receiver operating curve graph ideal test
Sensitivity on y axis
1 - specificity on x axis
Best test is in the top left corner
Cohort study design
Follow exposed and non-exposed forward in time to see if they get bad outcome
- Good for rare exposures, can measure prevalence of a disease
Case control study design
- Collect cases and controls and look back in time to see if they were exposed
- Good for rare diseases, can’t measure prevalence
Which tests to use for means (with 2, before/after and more than 2)
- 2: Student T-tests
- Before/after: paired T tests
- More than 2: ANOVA
Which tests to use for medians (with 2, before/after)
- 2: Mann-Whitney-U
- Before/after: Wilcoxon signed
Which test to use for time to event
Kaplain Meier
Which test to use for proportions
Chi square
- Fischer’s test for small group
Which test to use to predict a number or yes/no
Linear regression
