Stats/Epi Flashcards
Outline types of studies & examples?
Grade 1: Systematic review/meta-analysis
RCT
- Intervention vs placebo
- Can prove causality
- Randomisation reduces bias
i.e new asthma puffer- 1x group receives placebo, other receives drug, participants/administrators/data collectors blinded
Cohort studies
- Observational (association)- exposure vs non-exposure
- Prospective = exp then f/u
- Retrospective = outcome then look back
- RR >3
-i.e 500 children followed up to investigate effects of extreme prematurity on learning at 2,5,10yrs (compared with 500 non prems)
Case control studies
- Observational (association)
- Disease vs no disease
- Small numbers, but can be the most biased
- OR >4
i.e 100 children with obesity compared to 100 children without- environments studied
Cross sectional
- Freq of disease/risk factors at point in time
- Can develop association
- Prevalence
Ecological study
- Attempts to relate exposures in an environment to health outcomes
i.e incidence of SIDS in smoker households vs non-smoker
Survival study
- Looks at the PR(event of interest) occurring over time period
- i.e onset of disease after exposure at childcare within 2 weeks
Statistical tests and their uses?
Chi Squared (CATEGORIES): significance between two variables
T-Test (TWO GROUPS): difference in quantitative variable between two groups
ANOVA (MORE > 2): differences in quantitative variable in >2 groups
Kappa (lad- are you reliable): percentage agreement between different raters 0 = no agreement, 1 = perfect agreement
Type 1 & Type 2 error definition
Type 1 (a)= fails to recognise NO difference
- rejects null but actually no difference
Type 2 (b)= not big enough to pick up difference
- fail to reject null hypothesis as study too small
Adjustment for type 1error?
Setting p value <0.05 (5%)
- Less chance event is unexpected/due to chance
Adjustment for type 2 error?
Setting power (increase sample size)
- To determine actual effect
Sensitivity
SnOUT
- Number of true positives correctly identified by test
- Useful in ruling out
Specificity
SpIN
- Number of true negatives correctly identified by test
- Helpful in ruling disease in
Positive predictive value
The proportion of people who test positive who actually have the disease.
Negative predictive value
The proportion of people who test negative who do not have the disease.
How to calculate positive & negative LR?
Positive LR
* = sensitivity/(1-specificity).
* = (a/(a+c))/(1-(d/(d+b)))
Negative LR
* = (1- sensitivity)/specificity
* = (1-(a/(a+c)))/ (d/(d+b))