Stats Flashcards
Small group, categorical data - best test
Fischer exact
Large groups, categorical data - best test
Chi square
Continuous data, different individuals - best test
Unpaired T test
Continuous data, matched samples - best test
Paired T test
Continuous or ordinal data for non-normally distributed data - best test
“Nonparametric test to compare differences between two groups when dependent variables are ordinal or continuous”
Mann Whitney U
What is a Funnel Plot
Used to eval for bias in meta analysis/systematic reviews
Define sensitivity / specificity
Sens = TP / TP+FN = a/a+c
Spec = TN / TN+FP = d/d+b
disease t + - e + TP FP s - FN TN t
SENS/SPEC VERTICAL
What studies determine prevalence vs incidence
Prev = cross sectional survey
Incidence = new dx in a period of time = prospective cohort
95% CI is what mean/SD
Mean +/- 1.96 SD
90: 1.62
98: 2.33
99: 2.58
Define PPV and NPV
disease
t + -
e + TP FP
s - FN TN
t
PPV = TP/TP+FP
NPV = TN/TN+FN
HORIZONTAL
What is power
What is T2 error
Power = 1 - probability of T2 error (beta)
= 1- 20% = 80%
80% chance of doing study that finds p<0.05 if true association exists
T2 error = beta = study reports a negative result when a true difference exists
- False negative
- Ok 20% time
If decrease the power, increase the confidence if a negative result is found
What statistic is used for intra observer reliability
Kappa
Describe a cohort study
What statistic can be calculated
No comparison group
Know a potential risk factor - do your cohort develop the outcome if they have the risk factor
Often prospective…. INCIDENCE
Describe a case control study
ID pts with infection (cases)
Then go find their risk factor
By matching to controls
What are the 4 phases of research by FDA
1: pharmacology, det if drug is safe, dose finding
2: efficacy, does drug work?, compare to other interventions
3: confirm efficacy in larger trial
4: post marketing surveillance
What are the 5 levels of evidence?
What about systematic review/meta analysis
1 RCT
2 Cohort (prospective), poorly done RCT
3 Case control
4 Case series
5 Expert opinion
SR/MA - lowest level you incorporate into the review determines level of evidence
Define bias vs confounding
How do you combat each
Bias = you put in
- Errors in research methodology
Combat w/ better study design
Confounding = you didn’t know something was associated
- Unaccounted variables that explain associations
Combat w/ randomization
Define these types of bias
Selection
Detection
Recall
Interviewer
Performance
Hawthorne Effect
Selection - if study says “per surgeon preference” then this is selection bias
- Randomize to reduce!
Detection - tend to look more carefully for an outcomes in one group
Recall - knowledge of disease alters recall
- Pt w/ OA will recall a distant injury vs pt wo sx
Interviewer - knowledge of the treatment alters assessment
Performance - systematic, unintended difference in care
- Generalists vs trauma surgeons doing X case
Hawthorne Effect - subject behavior changes bc know being watched
- Room turnover gets better when people know turnover times are being watched
Equation for RR
How do you interpret results
RR = risk in exposed / risk in unexposed
RR = 1, risk equal exposed to non exposed
RR>1 = exposed > unexp
- Positive associated between exposure and outcome (?causal?)
1<RR = exposed < unexp
- Negative “ “ (?protective?)
Remember 95% CI for RR cannot cross 1 to be sig!
What is type 1 error
What does alpha < 0.05 mean
Saying an association exists, when it does not
False positive
Your study finds a difference when a diff doesn’t actually exist
Alpha < 0.05 = 5% of time get a difference in the data when it isn’t real
