Sample size question
Locked
 
|
Hello All, I have a question regarding sample size calculation. I’m working on a project which would involve 4 different treatment sites. The treatment at each site is a little different but the ultimate goal is to see a decrease in the 30 day hospitalization rate after a procedure in the hospital. This is a prospective, non-randomized, compare to baseline study. I have performed a power analysis on the data. Let’s say that my N = 1200 people. Here is my question: do I need to have this number of people at each site or can I split this number between all 4 sites (instead have 300 people per site to equal 1200). My first instinct is to say that I need 1200 people per site but I would really appreciate a second opinion. Thanks, Johnita |
|
|
Johnita, So, you are probably working with proportions and are treating the historically recent value as population value. Do you want adequate power for each hospital or for the set of hospitals taken as one hospital? How did you set up the power analysis? Gene Maguin From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Byrd Sellers, Johnita Hello All, I have a question regarding sample size calculation. I’m working on a project which would involve 4 different treatment sites. The treatment at each site is a little different but the ultimate goal is to see a decrease in the 30 day hospitalization rate after a procedure in the hospital. This is a prospective, non-randomized, compare to baseline study. I have performed a power analysis on the data. Let’s say that my N = 1200 people. Here is my question: do I need to have this number of people at each site or can I split this number between all 4 sites (instead have 300 people per site to equal 1200). My first instinct is to say that I need 1200 people per site but I would really appreciate a second opinion. Thanks, Johnita |
|
|
In reply to this post by Byrd Sellers, Johnita
Dear Johnita,
I would want to see the output from the program that you are using to be completely sure, but power analysis software packages typically provide a figure for the total sample size that is clearly labelled as "total sample size", meaning a total summing up all four sites. Best Regards, Stephen Brand, Ph.D. www.StatisticsDoc.com ---- "Byrd Sellers wrote: > Hello All, > > > > I have a question regarding sample size calculation. I'm working on a > project which would involve 4 different treatment sites. The treatment > at each site is a little different but the ultimate goal is to see a > decrease in the 30 day hospitalization rate after a procedure in the > hospital. This is a prospective, non-randomized, compare to baseline > study. I have performed a power analysis on the data. Let's say that > my N = 1200 people. Here is my question: do I need to have this number > of people at each site or can I split this number between all 4 sites > (instead have 300 people per site to equal 1200). My first instinct is > to say that I need 1200 people per site but I would really appreciate a > second opinion. > > > > Thanks, > > Johnita > > > ===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD |
|
|
In reply to this post by Maguin, Eugene
It sounds like you are trying to decide whether the alternative hypothesis should be that the four-facility average admission rate is reduced by 25 or that the admission rate for each facility is reduced by 25%. I don’t think that is a power question per se. Sorry. Gene Maguin From: Byrd Sellers, Johnita [mailto:[hidden email]] Gene, Yes, I am working with proportions. I previously set it up to where I compared the mean admission rate for the four facilities to the National hospital admission rate. The goal is to have the mean admission rate for all centers combined reduced by 25% in comparison to the National rate. I’m thinking that I would need adequate power per hospital since each hospital’s treatment plan is slightly different. In the end, I would think that it would be important to compare each treatment type to determine which method was most effective. I’m not sure that I would be able to do that if I treated the set of hospitals as one. Please let me know what you think. Thanks for any help regarding this matter. Johnita From: SPSSX(r) Discussion [hidden email] On Behalf Of Gene Maguin Johnita, So, you are probably working with proportions and are treating the historically recent value as population value. Do you want adequate power for each hospital or for the set of hospitals taken as one hospital? How did you set up the power analysis? Gene Maguin From: SPSSX(r) Discussion [hidden email] On Behalf Of Byrd Sellers, Johnita Hello All, I have a question regarding sample size calculation. I’m working on a project which would involve 4 different treatment sites. The treatment at each site is a little different but the ultimate goal is to see a decrease in the 30 day hospitalization rate after a procedure in the hospital. This is a prospective, non-randomized, compare to baseline study. I have performed a power analysis on the data. Let’s say that my N = 1200 people. Here is my question: do I need to have this number of people at each site or can I split this number between all 4 sites (instead have 300 people per site to equal 1200). My first instinct is to say that I need 1200 people per site but I would really appreciate a second opinion. Thanks, Johnita |
|
|
Comparing the facilities to a previous national rate seems a bit
chancy, for one thing. Worse news -- when you are looking at a "treatment" for facilities, the unpleasant answer is that for doing proper test of the hypothesis, you usually ought to have dozens of facilities, so you get your error term from the consistency among facilities. With just a few facilities, perhaps you are in a better position to promote a "demonstration project" which will attempt to evaluate several "strategies", and make use of subjective follow-up information as part of the data analysis. -- Rich Ulrich Date: Mon, 5 Dec 2011 14:20:09 -0500 From: [hidden email] Subject: Re: Sample size question To: [hidden email] It sounds like you are trying to decide whether the alternative hypothesis should be that the four-facility average admission rate is reduced by 25 or that the admission rate for each facility is reduced by 25%. I don’t think that is a power question per se. Sorry.
Gene Maguin
From: Byrd Sellers, Johnita [mailto:[hidden email]]
Gene,
Yes, I am working with proportions. I previously set it up to where I compared the mean admission rate for the four facilities to the National hospital admission rate. The goal is to have the mean admission rate for all centers combined reduced by 25% in comparison to the National rate.
I’m thinking that I would need adequate power per hospital since each hospital’s treatment plan is slightly different. In the end, I would think that it would be important to compare each treatment type to determine which method was most effective. I’m not sure that I would be able to do that if I treated the set of hospitals as one.
Please let me know what you think. Thanks for any help regarding this matter.
Johnita From: SPSSX(r) Discussion [hidden email] On Behalf Of Gene Maguin
Johnita,
So, you are probably working with proportions and are treating the historically recent value as population value. Do you want adequate power for each hospital or for the set of hospitals taken as one hospital? How did you set up the power analysis?
Gene Maguin
From: SPSSX(r) Discussion [hidden email] On Behalf Of Byrd Sellers, Johnita
Hello All,
I have a question regarding sample size calculation. I’m working on a project which would involve 4 different treatment sites. The treatment at each site is a little different but the ultimate goal is to see a decrease in the 30 day hospitalization rate after a procedure in the hospital. This is a prospective, non-randomized, compare to baseline study. I have performed a power analysis on the data. Let’s say that my N = 1200 people. Here is my question: do I need to have this number of people at each site or can I split this number between all 4 sites (instead have 300 people per site to equal 1200). My first instinct is to say that I need 1200 people per site but I would really appreciate a second opinion.
Thanks, Johnita
|
| Free forum by Nabble | Edit this page |