Stats Gross Flashcards
What is a case-control study
Observational, retrospective
A study done using a population of subjects that are selected based on the outcome of interest (cases) and compares them to a similar population of patients who don’t have the outcome of interest (controls). Investigators look back to identify characteristics (exposures) that are linked to the outcome
What is a cohort study
Observational, prospective OR retrospective
A study that defines two groups of patients (cohorts) - one which has been exposed (e.g., received active treatment) and one which has not been exposed (control) - and observes to see if an outcome of interest happens in either group.
Intent-To-treat vs Per-Protocol Analysis
Intent-to-Treat (ITT) – analyzing subjects according to the group to which they were randomized
Per-Protocol (PP) – analyzing only those subjects who adhered to the treatment according to the protocol
What is Type 1 Error
A Type 1 error occurs if we reject H0 when, in fact, H0 is true; If this happens, we will have a false positive
AKA say there is a difference when there really isn’t a difference
What is Type 2 Error
o A Type 2 error occurs when we accept H0 when, in fact, H0 is false; If this happens, we will have a false negative
AKA say there is no difference when there really is a difference
Probability of Type 1 Error
Alpha, normally 0.05
1 time out of 20 we will falsely reject H0
Probability of Type 2 Error
Beta, normally 0.20
Prevalence vs Incidence
Prevalence: Total number of cases (EXISTING disease) divided by total population; Tells how much disease there is right now (burden of disease)
Incidence: Measures frequency of disease in a population; How many people per year acquire the disease?
Odds Ratio calculation
Odds of outcome given EXPOSURE versus non-exposure to a substance or event (ratio of odds)
OR = Odds case was exposed/Odds control was exposed
OR=A/C / B/D
OR=AD/BC
Relative Risk Calculation
Ratio of incidence of outcome among exposed group/incidence of outcome among non-exposed
RR=Risk among exposed/Risk among non-exposed
RR=(A/A+B)/(C/C+D)
Relative Risk Reduction
Tells how much treatment reduced the risk of the outcome relative to controls who did not get treatment
RRR=1-RR
Absolute Risk Reduction
Absolute risk difference in event rate
ARR=(A/A+B) - (C/C+D)
How to calculate NNT?
1/ARR
Which test if data is:
Nonparametic
Nominal
Two groups, independent
Chi-square or Fisher exact test (if <5 in any cell or n<20)
Which test if data is:
Nonparametic
Nominal
3+ groups, independent
Chi-square or Fisher exact test (if <5 in any cell or n<20)
Which test if data is:
Nonparametic
Ordinal
2 groups, independent
- Mann-Whitney U
- Wilcoxon rank sum test
- Wilcoxon-Mann-Whitney test
Which test if data is:
Nonparametic
Continuous, not normally distributed
2 groups, independent
- Mann-Whitney U
- Wilcoxon rank sum test
- Wilcoxon-Mann-Whitney test
Which test if data is:
Nonparametic
Ordinal
3+ groups, independent
Kruskal-Wallis
Which test if data is:
Nonparametic
Continuous, not normally distributed
3+ groups, independent
Kruskal-Wallis
Nonparametric
Nominal
2 groups: Before & After a Single Treatment
McNemar test
Nonparametric
Nominal
3+ groups: Before & After a Single Treatment
Cochran Q test: Mantel-Haenszel test
Nonparametric
Ordinal
2 groups: Before & After a Single Treatment
Wilcoxon signed-rank test
Nonparametric
Ordinal
3+ groups: Before & After a Single Treatment
Friedman ANOVA by ranks
Nonparametric
Continuous, not normally distributed
2 groups: Before & After a Single Treatment
Wilcoxon signed-rank test
Nonparametric
Continuous, not normally distributed
3+ groups: Before & After a Single Treatment
Friedman ANOVA by ranks
Parametric
Continuous, normally distributed
2 groups
Student t-test
Parametric
Continuous, normally distributed
3+ groups
ANOVA followed by post hoc testing
Parametric
Continuous, normally distributed
2 groups: Before & After a Single Treatment
Paired t-test
Parametric
Continuous, normally distributed
3+ groups: Before & After a Single Treatment
Repeated measures ANOVA
What is study bias? Give examples
Systematic, nonrandom variation in study metholodgy and.or data collection process that results in an incorrect estimate of association between exposure and outcome
Selection bias Information bias Recall bias Interviewer bias Misclassification bias PUblicaiton bias Confounding variables
Categorical data: what is Nominal? Ordinal?
Nominal = categories w/o rank (i.e., sex, race, presence/absense of disease)
Ordinal = categories with rank (i.e., likert scale, pain score)
Continuous data: what is Ratio? Interval?
Ratio = HAS absolute zero where zero=none (HR, BP)
- Can calculate ratio of 2 measurements with ratio
Interval = NO absolute zero (temp)