[R-meta] Comparison to baseline- remove intercept or keep?

[Danielle Hiam]

Hello,

I am running a repeated measures meta-analysis of miRNA expression data across multiple timepoints. Therefore I have dependent effect sizes (multiple timepoints), further some of these studies include multiple treatments (in my case exercise).
I have used the fold change (FC) of expression from baseline as my effect size and SE of FC and inserted this into vcalc as indicated below.

V <- vcalc(FC_SEM,
       cluster=StudyNUM,
       subgroup = ExTreat,
       time1=Time_NUM,
       data=dat,
       phi=0.8)
#Variance Matrix indicating multiple timepoints (Time_NUM) within studies (StudyNUM) and also cohorts within studies (ExTreat)

I then placed this matrix into my model illustrated below.

BASE=rma.mv(yi = FC_MEAN,
       V = V,
                     data = dat,
               mods = ~ factor(Time_NUM),
         random = list (~ Time_NUM|interaction(ExTreat,StudyNUM),
                                     ~ 1|StudyNUM),
        tdist = TRUE)

QUESTION ONE) Do I include the timepoint zero (being baseline) with or without the intercept in “mods=”. As I want to see if timepoint 1 (aka POST) is different from baseline, timepoint 2 (aka 1HP) is different from baseline and so on… This is my main outcome.
So would my mod be ~ factor(Time_NUM) or ~ factor(Time_NUM)-1?

Results for  mods = ~ factor(Time_NUM),
                                  estimate      se    tval  df    pval    ci.lb   ci.ub   ​
intrcpt                            0.7668  0.8175  0.9380  35  0.3547  -0.8928  2.4265
factor(Time_NUM)0    1.9611  0.9764  2.0086  35  0.0523  -0.0210  3.9432  .
factor(Time_NUM)1    1.2971  1.1548  1.1232  35  0.2690  -1.0472  3.6414
factor(Time_NUM)2    0.7478  1.3569  0.5511  35  0.5851  -2.0069  3.5026

#This is saying there are no differences at timepoint 0, 1 or 2 compared to baseline (intercept?)

Results for  mods = ~ factor(Time_NUM)-1
                    estimate      se    tval  df    pval    ci.lb   ci.ub     ​
factor(Time_NUM)-1    0.7301  0.8189  0.8915  35  0.3787  -0.9324  2.3926
factor(Time_NUM)0     2.8073  0.7704  3.6440  35  0.0009   1.2433  4.3713  ***
factor(Time_NUM)1     2.1542  0.9860  2.1848  35  0.0357   0.1525  4.1559    *
factor(Time_NUM)2     1.5666  1.2044  1.3008  35  0.2018  -0.8783  4.0116

#This is saying there are differences at timepoint 0 and 1 compared to zero?


QUESTION TWO) Is using fold change ok within the rules of the model, baseline will always be 1 and then the fold change for subsequent timepoints is based on this (i.e MEAN of timepoint 1/ MEAN of baseline timepoint)?


QUESTION THREE) How should I include studies where that particular miRNA was not detected. I.e. should this be allocated as 1 for baseline and then 1 for subsequent timepoints (as in no change from baseline). If I don’t include these studies then it will bias the results towards the studies that a) detected the miRNA and b) found changes in miR expression?

Any advice or resources would be greatly appreciated.


Important Notice: The contents of this email are intended solely for the named addressee and are confidential; any unauthorised use, reproduction or storage of the contents is expressly prohibited. If you have received this email in error, please delete it and any attachments immediately and advise the sender by return email or telephone.

Deakin University does not warrant that this email and any attachments are error or virus free.