| |
 |
|
|
Science Forum Index » Statistics - Education Forum » HELP: meta-analysis on only two studies
Page 1 of 1
|
| Author |
Message |
| SmileDomain |
 Posted: Sat Apr 26, 2008 11:17 pm |
|
|
|
Guest
|
Recently I am working to do a meta-analysis on some kind of gene-
disease. Since there are only two studies include in my work, it is
impossible to do a meta-regression. And the fixed effected model and
radom affected model are useless too. There are few studies focusing
on this topic, can any one give a sugguestion?
ps. I have not only the published data, but also the original data.
But they are so different, one meet the HWE perfectly, while the Chi-2
of another study is larger than 10. |
|
|
| Back to top |
|
| Bruce Weaver |
| Posted: Mon Apr 28, 2008 3:44 am |
|
|
|
Guest
|
On Apr 27, 5:17 am, SmileDomain <enthume...@gmail.cn> wrote:
Quote: Recently I am working to do a meta-analysis on some kind of gene-
disease. Since there are only two studies include in my work, it is
impossible to do a meta-regression. And the fixed effected model and
radom affected model are useless too. There are few studies focusing
on this topic, can any one give a sugguestion?
What prevents you from computing a pooled estimate in the usual
fashion? If I'm not mistaken, the usual chi-square test of
heterogeneity would be equivalent to the square of a z-test comparing
the estimates from the two individual studies.
Quote: ps. I have not only the published data, but also the original data.
But they are so different, one meet the HWE perfectly, while the Chi-2
of another study is larger than 10.
What is HWE? What kind of data are you talking about? Odds ratios
from 2x2 tables? Or something else?
--
Bruce Weaver
bweaver@lakeheadu.ca
www.angelfire.com/wv/bwhomedir
"When all else fails, RTFM." |
|
|
| Back to top |
|
| SmileDomain |
| Posted: Tue Apr 29, 2008 6:22 pm |
|
|
|
Guest
|
I will take an example to explain what have prevented me from doing
an estimate.
the form of data I receive can be described as the following table:
GENE I GENE II
study Type AA AG GG TT TG GG
1 case c11 c12 c13 k11 k12 k13
control t11 t12 t13 l11 l12 l13
2 case...
cnotrol...
First of all, HWE(Hardy-Weinberg Equilibrium) should be checked. But
in this case, the Chi-2 statistic of study 2(GENE I control ONLY) is
7.3, almost 3 times of chi-2(90%)=2.7. This is rather strange. (Maybe
the missing data do effects the result greatly.)
Besides, as most researchers do, I use Q-statistic to do a test about
the heterogeneity of the two study in my work. But Q-statistic is not
a good choice when the numer of study is too small.
These days I am trying to use Random Effects Model to finish my
thesis. I hope it can work. Can someone give a sugguestion?
Yifan Yang
yyf1986@mail.ustc.edu.cn |
|
|
| Back to top |
|
| David Duffy |
| Posted: Wed Apr 30, 2008 1:41 am |
|
|
|
Guest
|
SmileDomain <enthumelon@gmail.cn> wrote:
Quote: I will take an example to explain what have prevented me from doing
an estimate.
the form of data I receive can be described as the following table:
GENE I GENE II
study Type AA AG GG TT TG GG
1 case c11 c12 c13 k11 k12 k13
control t11 t12 t13 l11 l12 l13
2 case...
cnotrol...
Quote: First of all, HWE(Hardy-Weinberg Equilibrium) should be checked. But
in this case, the Chi-2 statistic of study 2(GENE I control ONLY) is
7.3, almost 3 times of chi-2(90%)=2.7. This is rather strange. (Maybe
the missing data do effects the result greatly.)
Besides, as most researchers do, I use Q-statistic to do a test about
the heterogeneity of the two study in my work. But Q-statistic is not
a good choice when the numer of study is too small.
These days I am trying to use Random Effects Model to finish my
thesis. I hope it can work. Can someone give a sugguestion?
Are the two study samples from greatly different populations ethnically? Are
allele frequencies in controls in the two studies comparable? You
test HWE because it may i) reflect laboratory error ii) population substructure etc.
If you are happy that i) is not the case, "statistically significant" disequilibrium
may not be large enough to prevent valid inferences about gene-disease
association and interstudy differences ie go ahead and fit a log-linear model
to the 2x3x2 table.
David Duffy.
--
| David Duffy (MBBS PhD) ,-_|\
| email: davidD@qimr.edu.au ph: INT+61+7+3362-0217 fax: -0101 / *
| Epidemiology Unit, Queensland Institute of Medical Research \_,-._/
| 300 Herston Rd, Brisbane, Queensland 4029, Australia GPG 4D0B994A v |
|
|
| Back to top |
|
| |
|
Page 1 of 1
All times are GMT - 5 Hours
The time now is Sat Jul 26, 2008 11:23 pm
|
|
