Power of a study

[iheb bougmiza]

Dear list,
  I studied 40 patients dead. the death was avoidable for 28 patients and non avoidable for 12 others.
  I want to identify factors associated to the avoidable character of death.
  I've 25 independant dichotomus variables.
  I did not calculate the sample size in advance. In fact , this sample was issued from another study.
  I did a logistic regression and finally, I identified 2 factors.
  How can I calculate the power of the study after analysis in order to demonstrate that if there was not a significative association beween a factor X  for example and the avoidable character, it can be related to the sample size and the low power.
  I read that I can apply the Schoenfeld rule. Any one has more details about this rule or other method I can apply.
  Thanks for advance

  Nb: Excuse my english


M. Iheb BOUGMIZA

Adresse: 90, Rue de l'Ourcq, Bat B. 75019 Paris
Tél : [phone]
  Mailto: [hidden email]


---------------------------------
 Yahoo! Mail réinvente le mail ! Découvrez le nouveau Yahoo! Mail et son interface révolutionnaire.

[Marta García-Granero]

Hi

Your dataset has too many predictors compared to sample size. The
general rule of thumb is that you should have 10-20 cases for every
predictor you want to test. This means that your have the risk of
having found a spurious model, that fits perfectly your small sample
of data, but not generalizable to the population.

You don't mention the method you followed to identify those 2
factord. Did you have a priori sound hypotheses, or did you just use
stepwise regression (fishing expedition)?.

Anyway, you should label your results as "preliminary" until a
validation sample is used to test if the model is robust or not.

Concerning the power, post-hoc power analysis is not recommended: if a
factor has been considered significant, then it's evident that the
power was high, if a factor is not significant, then the power wasn't
enough (simplifying things a bit). The power of your study is
necessarily low, you are limited by the number of non avoidable deaths
(12).

Sincerely, I would not use a mutivariate model with such a small
sample, your model will be overparametrized.

Sorry for the "bad news" :(

Marta

ib>   I studied 40 patients dead. the death was avoidable for 28
ib> patients and non avoidable for 12 others.
ib>   I want to identify factors associated to the avoidable character of death.
ib>   I've 25 independant dichotomus variables.
ib>   I did not calculate the sample size in advance. In fact ,
ib> this sample was issued from another study.
ib>   I did a logistic regression and finally, I identified 2 factors.
ib>   How can I calculate the power of the study after analysis
ib> in order to demonstrate that if there was not a significative
ib> association beween a factor X  for example and the avoidable
ib> character, it can be related to the sample size and the low power.
ib>   I read that I can apply the Schoenfeld rule. Any one has
ib> more details about this rule or other method I can apply.

[Hal 9000]

I've studied some patients nearly to death...both theirs and mine.