Which test to use?

#1
Thanks in advance for reading my post.

I have a year's worth data that describes the frequency of patient falls organized by the attending physician. Each physician has a unique number of cases (i.e. one doc can have 40 cases, and another has 15).

I want to know whether the monthly count of each physician's falls is significant from the overall number of patient treated for my list of physicians. I have already calculated each physicians Percent of Falls by month (Numerator = physician's falls / Denom = total physician's patients). In other words, does Dr. So&So have more falls?

I'm thinking this is a variance or standard dev. question... but I just know enough to confuse myself! :confused:

Working in MS Excel, which test(s) should I run? Tips on organizing the data too?
 

Karabiner

TS Contributor
#2
Usually I'd recommend a crosstab physician * patient fell yes/no,
with Chi² test. But for this you'd need independent observations,
i.e. each fall was from a different patient, no patient fell twice or
more times.

With kind regards

K.
 
#4
Karabiner: just curious, what would you use if they are not independent? I'm assuming they are not; seems likely to me that at least some patients will fall more than once (thought I don't know how rare that is).

Also, I assume that jaimeiam will want to do some post-hoc tests afterwards to identify which doctors have patients that fall more often. Any ideas how they could go about? I was thinking about doing a separate chi-test with one column with doctor X and the other column with all doctors aggregated? With a Bonferroni correction, i.e. alpha/N of doctors. Would that work?
 

Karabiner

TS Contributor
#5
just curious, what would you use if they are not independent? I'm assuming they are not; seems likely to me that at least some patients will fall more than once (thought I don't know how rare that is).
It depends on the quality of the records. If one could identify
the patients which fell, and how often they fell, then something
like a H-test with doctor as a grouping factor might apply.
Also, I assume that jaimeiam will want to do some post-hoc tests afterwards to identify which doctors have patients that fall more often. Any ideas how they could go about? I was thinking about doing a separate chi-test with one column with doctor X and the other column with all doctors aggregated? With a Bonferroni correction, i.e. alpha/N of doctors. Would that work?
The OP did not mention the number of doctors, so
we don't know how small alpha would be. In any case,
a Bonferroni correction will lead to a small alpha threshold
So there will be the need to decide what weights more,
possible type 1 or possible type 2 errors.

With kind regards

K.