dosresmeta - Error on restricted cubic spline modeling

[gilbertlzrus]

Hi, fellow statisticians! I am very new to R.

I am currently trying to learn to do dose-response meta-analysis using the command dosresmeta. !

I used the dataset published by Shim et al
1598549085232|675x433

When I tried to test for non-linearity with the restricted cubic spline model, the commands returned error:

knots_bin <- quantile(databin$dose, c(.05, .35, .65, .95))
> spline <- dosresmeta(formula=logrr ~ rcs(dose, knots_bin), type=type, id=id, se=se, cases=cases, n=n, method="fixed", data=databin)

## This is the error code
Error in dosresmeta.fit(X[v != 0, , drop = FALSE], Z[v != 0, , drop = FALSE],  : 
  A two-stage approach requires that each study provides at least p non-referent obs (p is the number of columns of the design matrix X)

note that databin is the dataset I am using

I tried to look up to the [source code] for dosresmeta, and the syntax states the following:

   m <- length(unique(id))
   k <- table(id)
   p <- ncol(X)
   q <- ncol(Z)
   nay <- is.na(y)
   n <- nrow(y)
   nall <- sum(!nay)
   ## names for coefficient
   np <- colnames(X)
   if (is.null(np))
      np <- paste("X", seq(p), sep = "")
   nq <- colnames(Z)
   if (is.null(nq))
      nq <- paste("Z", seq(q), sep = "")
   ny <- colnames(y)
   if (is.null(ny))
      ny <- "y"
   ylist <- lapply(unique(id), function(j) y[id == j, ][!nay[j, ]])
   Xlist <- lapply(unique(id), function(j) X[id == j, , drop = FALSE][
      !nay[j, ], , drop = FALSE])
   Zlist <- lapply(unique(id), function(j) Z[id == j, , drop = FALSE][
      !nay[j, ], , drop = FALSE])
   nalist <- lapply(unique(id), function(j) nay[id == j])

## This is the source code for the error
   if (proc == "2stage"){
      if (any(k < p)){
         stop("A two-stage approach requires that each study provides at least p non-referent obs (p is the number of columns of the design matrix X)")
      }

Does anyone know which part of my spline command caused the error? I managed to do the command with one-stage process, but I think with this model, two-stage process is more appropriate.

Please help, any help will be much appreciated. Thank you very much

Best,

[nirgrahamuk]

@gilbertlzrus

 if (any(k < p)){
         stop("A two-stage approach requires that each study provides at least p non-referent obs (p is the number of columns of the design matrix X)")
      }

looking up shows

   k <- table(id)
   p <- ncol(X)

[gilbertlzrus]

Thank you very much for your answer.

I noticed the statement that k might be smaller than p

k <- table(id)
   p <- ncol(X)

I assume the variable k returns value 4, but how can I check the value of p? And possible, which value can I modify to correct the error?

Once again, thanks

Best,

[nirgrahamuk]

i think id refers to the id you passed in, id 3 has 3 rows in your screenshot so sometimes its less than 4.
I don't know the genesis of X...
it might be

X <- model.matrix(attr(mfmX, "terms"), data = mfmX)[, 
        -1, drop = FALSE]

and it seems like

fmX <- model.frame(formulaMod, data)

[gilbertlzrus]

I see, so is it a limitation of the dosresmeta package?

I think Stata's glst command computes study-specific non-linear slopes and combines them at the end; so the number of non-referent obs does not matter as long as it is >2 obs...

I am referring to the dose-response meta-analysis tutorial in Stata by Shim et al.

Reference: Shim SR, Shin IS, Yoon BH, Bae J-M. Dose-response meta-analysis using STATA software. J Health Info Stat. 2016;41(3):351-8

Is there any alternative to solve this matter?

Thank you very much

Best,