metafor (version 2.0-0)

dat.ishak2007: Studies on Deep-Brain Stimulation

Description

Results from 46 studies examining the effects of deep-brain stimulation on motor skills of patients with Parkinson's disease.

Usage

dat.ishak2007

Format

The data frame contains the following columns:

study	character	(first) author and year
y1i	numeric	observed mean difference at 3 months
v1i	numeric	sampling variance of the mean difference at 3 months
y2i	numeric	observed mean difference at 6 months
v2i	numeric	sampling variance of the mean difference at 6 months
y3i	numeric	observed mean difference at 12 months
v3i	numeric	sampling variance of the mean difference at 12 months
y4i	numeric	observed mean difference at the long-term follow-up
v4i	numeric	sampling variance of the mean difference at the long-term follow-up
mdur	numeric	mean disease duration (in years)

Details

Deep-brain stimulation (DBS), which is delivered through thin surgically implanted wires in specific areas of the brain and controlled by the patient, is meant to provide relief of the debilitating symptoms of Parkinson's disease. The dataset includes the results from 46 studies examining the effects of DBS of the subthalamic nucleus on motor functioning, measured with the Unified Parkinson's Disease Rating Scale (UPDRS). The effect size measure for this meta-analysis was the mean difference of the scores while the stimulator is active and the baseline scores (before implantation of the stimulator). Since lower scores on the UPDRS indicate better functioning, negative numbers indicate improvements in motor skills. Effects were generally measured at 3, 6, and 12 months after implantation of the stimulator, with some studies also including a further long-term follow-up. However, the number of measurements differed between studies - hence the missing data on some of the measurement occasions.

Since the same patients were followed over time within a study, effect size estimates from multiple measurement occasions are likely to be correlated. A multivariate model accounting for the correlation in the effects can be used to meta-analyze these data. A difficulty with this approach is the lack of information about the correlation of the measurements over time in the individual studies. The approach taken by Ishak et al. (2007) was to assume an autoregressive (AR1) structure for the estimates within the individual studies. In addition, the correlation in the true effects was modeled, again using an autoregressive structure.

Examples

# NOT RUN {
### load data
dat <- get(data(dat.ishak2007))

### create long format dataset
dat.long <- reshape(dat, direction="long", idvar="study", v.names=c("yi","vi"),
                    varying=list(c(2,4,6,8), c(3,5,7,9)))
dat.long <- dat.long[order(dat.long$study, dat.long$time),]
rownames(dat.long) <- 1:nrow(dat.long)

### remove missing measurement occasions from dat.long
is.miss  <- is.na(dat.long$yi)
dat.long <- dat.long[!is.miss,]

### construct the full (block diagonal) V matrix with an AR(1) structure
rho.within <- .97 ### value as estimated by Ishak et al. (2007)
V <- lapply(split(with(dat, cbind(v1i, v2i, v3i, v4i)), dat$study), diag)
V <- lapply(V, function(v) sqrt(v) %*% toeplitz(ARMAacf(ar=rho.within, lag.max=3)) %*% sqrt(v))
V <- bldiag(V)
V <- V[!is.miss,!is.miss] ### remove missing measurement occasions from V

### plot data
with(dat.long, interaction.plot(time, study, yi, type="b", pch=19, lty="solid", xaxt="n",
               legend=FALSE, xlab="Time Point", ylab="Mean Difference", bty="l"))
axis(side=1, at=1:4, lab=c("1 (3 months)", "2 (6 months)", "3 (12 months)", "4 (12+ months)"))

### multivariate model with heteroscedastic AR(1) structure for the true effects
res <- rma.mv(yi, V, mods = ~ factor(time) - 1, random = ~ time | study,
              struct = "HAR", data = dat.long)
print(res, digits=2)
# }

Run the code above in your browser using DataCamp Workspace