Question on sample size and statistical test


I am partnering with a research hospital and am tasked with comparing surgical closure techniques and incidence of postop infection. I have data from 51 patients where a particular closure technique was used that led to 9.8% infection rate (data is binary, yes/no for infection occurrence. I am now tasked at looking 2 additional methods to see if there is a relationship between method of suture closure and infection. It will take a lot of time to acquire data for the other 2 methods so before I started I wanted to ask:

1. What tests would be appropriate for this analysis? Chi squared?

2. How will I know what the minimum n should be for the other 2 methods to make meaningful conclusions considering I have 51 for the 1st group?


Fortran must die
Sample size is complex and since I am no expert I will suggest it is chiefly critical in the context of statistical power (you need to decide what an acceptable power is, I believe the medical field has agreed on standards for this). Second is the margin of error, at least in polling. I do not know how this applies to what you are doing.

Why not make your dependent variable infection rate. Then your variable would be interval and you could use regression or ANOVA. I think interval data is generally preferable to ordinal results.


Not a robit
If it is a rate, Poisson regression can be used. If you are comparing (pairwise) more than 2 groups than you will need to correct your alpha level to address familywise error rate. Also, if treatment assignment was not randomized, you will need to control for background differences between the treatment groups. This can be done using Poisson regression.