pilot - individual participant data metaanalysis (just for CSI and GABA this time)

according to

Chen, D. G., & Peace, K. E. (2021). Applied Meta-analysis with R and Stata. Boca Raton, FL: CRC Press.

data summary (overall)

removing missing data (overall)

descriptives (by study)

meta-analysis with IPD

IPD analysis for each study

IPD analysis with pooled data

textbook: tested the treatment-study interaction

Subtopic

IPD with covariates (analogous to harmonization/residualization)

interactions of covariates and study (cov1*study, cov2*study, cov1*effect, cov2*effect, effect*study; summarily, all main effects and two-way interactions, no three-way interactions)

IPD with mixed-effect models (a type of multilevel modeling)

pooling all studies together

goal: estimate both between- and within- study effects

technical

fitting random-slope random-intercept model

random-intercept only model

model comparison (rsri/ri)

ICC (for any model)

what do we have to do in IPD?

pool correlations from all studies using harmonized (residualized) data

harmonize first where harmonization is needed

merge the data from all studies (with a study identificator)

pool between - study correlations (effect that we need)

pool within-study correlations (part of correlation that does not belong to the actual CSI-GABA relation)

how will we do it?

using multilevel correlations as implemented in the psych package

harmonization/residualization

is excluding the effects of factors that may be relevant and may effect the results, but are not the scope of analysis (e.g., vendor, age, gender...)

can be done via

ComBAT harmonization

residualization using linear models

GABA

data

Peek et al. (2021)

Aguila et al. (2016)

Terumitsu et al. (2022)

BDNF

data

Pollli et al. (2020)

Caumo et al.

control for

age

gender

group

control for

vendor

age

gender