Binary Outcome Data/NNT
Summary
Relative Measures of effect Compare 2 groups by dividing the risk/odds in intervention group by risk/odds in control group
Summary Statistics
Risk Ratio (probability of event occurring between groups)
RISK RATIO = risk of MI in new treatment/risk of MI in current treatment (example 0.257/0.314 = 0.82 (0.82-1 = -0.18 or 18% reduction
'those randomised to the new treatment had 0.82 times of 18% reduction in the risk of MI compared to those on the current treatment'
Relative measures (Odds & Ratio) are considered alongside absolute measures (Risk Difference)
Odds Ratio (odds of an event occurring between groups)
ODDS RATIO = odds of MI in new treatment/odds of MI in current treatment (example 0.35/0.46 = 0.76 (0.76-1 = -0.24 or 24% reduction
'those randomized to the new treatment had 0.76 times or 24% reduction in the odds of MI compared to those on the current treatment'
Risk/Odds Ratio
Odds - statically more appealing
Risk - more interpret-able than odds
1 = no evidence of difference between groups
<1 = evidence of lower risk/odds of MI in new treatment
1 = evidence of higher risk/odds of MI in new treatment
Absolute measures of effect (compare 2 groups by difference in probability
Summary Statistic
Risk Difference
RISK DIFFERENCE = risk of MI in new treatment - risk of MI in current treatment (example 0.257-0.314 = 0.057 or -5.7%
'those randomized to the new treatment had a 5.7% lower rick of MI compared to those on the current treatment
0 = no evidence of difference
<0 = evidence of lower rick of MI in new
.>0 = evidence of higher rick of MI in new
:
Two values (YES/NO) (event occurred or not?)
The proportion of those with the event of interest (risk/probability) 0-1 or 0-100%
RCTs looks at new events from randomization (incidence proportion)
The odds of having the event of interest( ratio of probability of an event to the probability of not having the event) 0- infinity, no upper bound
Can be misleading
Risk ratio of 0.5 or 50%, and risk difference of -3.4 or 3%
50% vs 3% (very different values)
NNT (number needed to treat) - average no of people to be treated with new treatment in order for one person to benefit
Ideal NNT = 1, 1 person to treat for 1 person to benefit
NNT 20 = 20 treated, 1 benefitted
Estimate NNT
NNT= 1/risk difference
1/0.057 = 17.5 (or 18 people, round up)
On average 18 pts need to be treated with new treatment in order to prevent 1 MI over 1 year period
Doesn't tel us what happened to the other 17 pts?
Expresses the effect of treatment in a way that is easily understood by clinicians and patients