Hi,
You have to use the unstandardized or standardized mean gain for
before/after studies which is different from parallel designs because of
the correlation between before and after values. See Mark W. Lipsey's
book "Practical Meta-analysis" for details on how this can be done
Suhail A. R. Doi
Associate Professor of Clinical Epidemiology
University of Queensland, School of Population Health,
Brisbane, Australia
On 7/20/2013 9:00 AM, EVIDENCE-BASED-HEALTH automatic digest system wrote:
> There are 2 messages totaling 222 lines in this issue.
>
> Topics of the day:
>
> 1. Fw: Fw: Meta-analysis of before/after studies
> 2. Meta-analysis of before/after studies
>
> ----------------------------------------------------------------------
>
> Date: Thu, 18 Jul 2013 23:11:44 -0700
> From: Hamid R Baradaran <[log in to unmask]>
> Subject: Fw: Fw: Meta-analysis of before/after studies
>
>
>
>
> Hamid Baradaran M.D. (Tehran) Ph.D., (Glasgow)
> Associate Professor of Clinical Epidemiology
>
> Iran University of Medical Sciences
> Post Code : 1449614535 , Tehran
> Iran
>
> Tel : 0098-21-86702216
> Fax: 0098-21-86702237
>
> E-mail: [log in to unmask]
>
> Web Page www.tums.ac.ir
>
>
>
>
>
>
>
>
>
>
> ----- Forwarded Message -----
> From: Jalal Poorolajal <[log in to unmask]>
> To: Hamid R Baradaran <[log in to unmask]>
> Sent: Friday, July 19, 2013 9:38 AM
> Subject: Re: Fw: Meta-analysis of before/after studies
>
>
>
> Dear Colleagues,
> There is no difference between parallel
> design and before after design in meta-analysis. You can obtain a
> summary measure for the change or difference of the means using
> meta-analysis.
>
>
>
> Sincerely yours
>
> Jalal Poorolajal, MD, MPH, PhD
> Assistant Professor of Epidemiology,
>
> Department of Epidemiology &
> Biostatistics
>
> Managing Editor, Journal of Research in Health
> Sciences,
> School of Public Health,
> Hamadan University of Medical Sciences,
> Hamadan, Islamic Republic of Iran
> Postal code: 65157838695
> Tel: +98 811 8380090 [29]
> Fax: +98 811 8380509
> E-mail 1: [log in to unmask]
> E-mail 2: [log in to unmask]
> URL: http://scholar.google.com/citations?hl=en&user=LDEnqk8AAAAJ&view_op=list_works
>
>
> -----Original
> Message-----
>> From: Hamid R Baradaran <[log in to unmask]>
>> To: Dr Ali Haghdoost <[log in to unmask]>
>> Cc: Dr Jalal Poorolajal <[log in to unmask]>
>> Date: Thu, 18 Jul 2013 12:52:13 -0700 (PDT)
>> Subject: Fw: Meta-analysis of before/after studies
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>>
>> ----- Forwarded Message -----
>> From: cesar yammine <[log in to unmask]>
>> To:
>> Sent: Thursday, July 18, 2013
> 9:00 PM
>> Subject: Meta-analysis of
> before/after studies
>>
>>
>> Hello everybody,
>> I would like to know if anyone has experience with MA where the
> included studies have a before/after design.
>> Should I use the the same approach as with RCTs where I compute
> for example raw or standardized means and their SDs along with sample sizes
> or should I compute the change or difference of the means ( mean of
> intervention - mean of baseline).
>> Regards
>>
>> Kaissar Yammine, MD, MPH, PhD
>> Head of the Foot and Hand Clinic
>> Director of the Center for Evidence-Based Sport and Orthop
> ------------------------------
>
> Date: Fri, 19 Jul 2013 11:31:42 +0200
> From: Piersante Sestini <[log in to unmask]>
> Subject: Re: Meta-analysis of before/after studies
>
> On 13/07/18 18:30, cesar yammine wrote:
>> Hello everybody,
>> I would like to know if anyone has experience with MA where the included
> > studies have a before/after design.
>
>
> I recall having met this kind of problems in a meta-analysis of
> short-term beta2 agonist drugs in COPD,
> http://www.ncbi.nlm.nih.gov/pubmed/10908502. The problem then was that
> that there was a mix of parallel and cross-over studies, and in the
> latter not all the only the SD of the effect difference was given.
> You may look to the review to see what solution we used (it was quite
> long ago), but as far as I remember it involved to "guess" a reasonable
> correlation between the two measures and apply it to the SD of the
> difference to obtain the two separate SD, assuming that they were equal.
> It may look a bit adventurous, but in that case a reasonable estimate of
> the correlation between the two measures (FEV1 before and after
> treatment in the same subject) was easy to obtain from empirical data.
>
>
> Piersante Sestini
>
>> Should I use the the same approach as with RCTs where I compute for
>> example raw or standardized means and their SDs along with sample sizes
>> or should I compute the change or difference of the means ( mean of
>> intervention - mean of baseline).
>>
>> Regards
>>
>> Kaissar Yammine, MD, MPH, PhD
>> Head of the Foot and Hand Clinic
>> Director of the Center for Evidence-Based Sport and Orthopedic Research
>> Emirates Hospital
>> Dubai, UAE
>>
> ------------------------------
>
> End of EVIDENCE-BASED-HEALTH Digest - 18 Jul 2013 to 19 Jul 2013 (#2013-173)
> ****************************************************************************
>
|