11th January 2011 at 2:39 pm #3778
Hello very new to the site so hopefully posting int he right place. I am trying to look at some analysis very simple with those that have a certain procedure and either gain an infection / or have a complication – however the number of people who contracted an infection at all or any complication is so low (which is good just not for the analysis) for example out of 600 possibly only 4 had an infection and about 30 had a complication was just looking at simple chi square – procedure a vs b and infection or no infection but there seems to be no way i could rely on anything that has such low numbers for the outcome1? Is there a general consensus on the minimum split would need to see for conditons… as in no cell smaller than 5 (seem to remember reading this somewhere)?
Many thanks for any advice
M13th January 2011 at 4:16 pm #3781
Hi Many thanks for the advice – I am testing those that use a particular procedure against another so which one is best really – I was just worried as the number was so low to be able to say anything with confidence about the results. As if you only have 5 people out of 600 who had an infection, even if all five were under one procedure and no one who had the other procedure had an infection – as it is only 5 people how can I be sure this didnt just happen by chance… I mean I suppose thats why we have these tests there but just in my mind this number is so low its a bit worrying (not that i wont a greater number to have an infection) but just from a stats point of view.
thanks for advice though i may try to look at all complications as you say to see what this produces14th January 2011 at 11:26 am #3780
ah i see. This makes sense so you could only look at the analysis within one procedure (but taking the low infection rate as null prob from other procedure)… its not necessarily comparing them like in chi sqaure… but is in a round about way? what makes it more complitcated is my other procedure group is also a lot lower (only 50) which is why i think i am not getting a sig difference as 1 out of 50 is not as big a difference when comparing to 5 out of 600 although still twice the effect… I will have a look into proprtions test as above
Many thanks again for advice
Michaela7th March 2011 at 6:30 pm #3779Joseph LovettMember
You are asking a basic biostatistics/clinical evaluation question. The small cell counts are very common where a disease is rare. You need to take into account the power of the statistical test with small samples, the minimum cell sizes required by any particular test, and the sensitivity and specificity of the intervention/outcome. Small cell counts, indeed even zero counts, are not necessarily a problem. The “5” limitation you are recalling applies only to the chi-square test. Consider a different test or family of tests.
- The forum ‘Default Forum’ is closed to new topics and replies.