I want to conduct a meta-analysis of single means. However, these means are restricted mean survival time (RMST) of cerebrospinal fluid shunt inserted to treat hydrocephalus.
For this, I have digitized published survival curves with https://apps.automeris.io/wpd/. Then I have extracted the individual patient data with the R package IPDfromKM. Finaly, I have recontructed the survival curve and calculated the RMST at 12 months as decribed here: How to compute the mean survival time.
I have yet the following RMST and associated standard error.
study <- c("study1", "study2", "study3", "study4", "study5")
n_patients <- c(535, 209, 111, 599, 434)
rmst_12 <- c(10.54759, 11.36175, 10.50244, 10.51183, 8.716552)
se_12 <- c(0.1463532, 0.1439506, 0.3246873, 0.1471398, 0.2374582)
> data
study n_patients rmst_12 se_12
1 study1 535 10.54759 0.1463532
2 study2 209 11.36175 0.1439506
3 study3 111 10.50244 0.3246873
4 study4 599 10.51183 0.1471398
5 study5 434 8.716552 0.2374582
Now I am performing the meta-analysis of my RMST.
# Compute standard deviation (SD) from standard error (SE) ----
data $ sd_12 <- data $ se_12 * sqrt (data $ n_patients)
# Download useful package
require (meta)
# Compute the meta-analysis with the metamean function ----
mm_12 <- metamean (n = n_shunts,
mean = rmst_12,
sd = sd_12,
studlab = author_year,
data = data,
method.mean = "Luo",
method.sd = "Shi",
sm = 'MRAW',
random = TRUE,
warn = TRUE,
prediction = TRUE,
method.tau = "REML")
> mm_12
Number of studies combined: k = 5
Number of observations: o = 1888
mean 95%-CI
Common effect model 10.5757 [10.4247; 10.7268]
Random effects model 10.3373 [ 9.4868; 11.1878]
Prediction interval [ 7.0212; 13.6534]
Quantifying heterogeneity:
tau^2 = 0.8975 [0.2937; 7.7528]; tau = 0.9474 [0.5419; 2.7844]
I^2 = 95.6% [92.3%; 97.5%]; H = 4.78 [3.61; 6.34]
Test of heterogeneity:
Q d.f. p-value
91.39 4 < 0.0001
Details on meta-analytical method:
- Inverse variance method
- Restricted maximum-likelihood estimator for tau^2
- Q-Profile method for confidence interval of tau^2 and tau
- Prediction interval based on t-distribution (df = 3)
- Untransformed (raw) means
Everything works fine but, I am wondering myself if this is correct and right from a methodological point of view ?
Thank you in advance for your help.
PS. I am unsure if it is the right place for this post.
Charles
