Comparing binary variables between groups Flashcards
What is a binary variable?
categorical variable with 2 categories
eg mortality status, diabetes status
How is proportion used to summarise binary variables for a single group
number of participants of interest / total number
0-1 value range (% = x100)
How are odds used to summarise binary variables for a single group and why is it useful?
% category of interest / 100 - % category of interest
for comparing binary variables between groups in case-controlled studies
What is the exposure variable?
exposure is the potential cause of outcome
What absolute measure is used to compare binary variables between independent groups?
risk difference (% in group one - % in group 2)
What relative measures are used to compare binary variables between independent groups?
Risk ratio and odds ratio
%/odds R with disease in intervention group / %/odds R with disease in control group
what do different risk & odds ratio values mean?
= 0, > 0, < 0
(OR always further from 1 then RR)
=0 = groups equally likely to have disease
> 0 = first group more likely
<0 = second group more likely
What does number needed to treat (NNT) show?
the number of people than need to receive an intervention before 1 person benefits as a result of treatment
how is NNT calculated?
100 / risk difference (%)
What does number needed to harm show
the number of people that receive the intervention until somebody is negatively effected
How is relative risk reduction (RRR) calculated ?
1 - risk ratio x 100 = RRR %
How is absolute risk reduction (ARR) calculated?
one percentage - the other = ARR %
When is using risk difference and NNT better?
quantify the impact of an intervention for reducing disease in a given setting
When is using Risk ratio & odds ratio better?
quantifying the strength of association between intervention and disease status
When estimating confidence intervals what assumptions must be met?
total sample size is at least 40
at least 5 subjects in each category of the binary variable in each group
How is expected value (if hypothesis were true) calculated from a 2x2 table
row total x column total / overall total
can also be worked out as %
How is P value calculated when comparing binary variables between two independent groups
Chi-squared test (parametric (assumption based)) or Fisher’s exact test (non-parametric)
Tests quantify evidence against null hypothesis
When is Fisher’s exact test used? (when chi-squared test assumptions are not met)
when there are fewer than 20 participants or when between 20-39 participants w/ expect value <5 in at least one cell
Which test should be used for larger contingency tables?
Chi-squared if expected value is >5 in at least 80% of cells if not use Fisher