Statistics

 

===== Tutorial stuff =====

 

Risk ratio = occurrence in exposed group / occurrence in non-exposed group

 

 

 

Odds ratio = odds in exposed group / odds in non-exposed group

 

 

Table

	Lung cancer	No cancer
Smoker	a	b
Nonsmoker	c	d

 

Odds ratio = ad/bc
* ((a/b)/(c/d))
* i.e.

Risk ratio = (a/(a+b)) / (c/(c+d))
* i.e. chance of getting lung cancer in smokers / chance of getting lung cancer in non-smokers

 

Strong association

Temporal relationship

Biological reason?

Dose-response

 

RR only tells association, NOT causation.

 

NNT = 1 / absolute risk reduction

 

 

Questions to ask in EBM

1. Formulate question

2. Search evidence

3. Assess level of evidence

4. Apply to patients

5. Self-assessment

 

Levels of evidence

Level 1 - meta-analysis

Level 2 - randomised controlled trials

Level 3 - other well conducted experimental studies

Level 4 - expert panels

 

metaanalysis

subject to publication bias

expensive

studies may have different questions, duplications, different designs