cran / saslm Goto Github PK

Hello,
Many thanks to the authors of this new package.

I have been testing it against the reference datasets provided as supplementary materials to this paper:
https://link.springer.com/article/10.1208%2Fs12248-014-9661-0

The results agree with those from the software packages commonly used in the statistical analysis of bioequivalence studies (table 1 of the paper). However, CPU time is much longer (see below) using sasLM than when using author's function "Analyse222BE" which only relies on lm() and drop1(). I noticed that there are some ancillary functions in sasLM which have not been vectorized.

Whereas sasLM may be useful for more complex study designs, there is no clear advantage in using it (only speaking here about its current version i.e. 0.6.3) for balanced and unbalanced 2x2x2 crossover designs.

On a Windows 10 x64 computer with an i7-2600K
File "12248_2014_9661_MOESM7_ESM.txt" contains 2000 rows of data.
##########################################################
library(sasLM)
library(data.table)
BEsasLM <- function (file, alpha=0.05) {
dta <- fread(file=file,header=TRUE)
mescols <- c("Subj","Seq","Per","Trt")
dta <- dta[,mescols:=lapply(.SD,as.factor),.SDcols=mescols]
f1 <- log(Var) ~ Seq/Subj + Per + Trt
print(T3test(f1, dta, "Seq:Subj"))
ci0 <- CIest(f1, dta, "Trt", c(-1, 1),1-2*alpha) ; ci0 # 90% CI
print(exp(ci0[1,c("Estimate", "Lower CL", "Upper CL")])) # 90% CI of GMR
}

system.time(BEsasLM("12248_2014_9661_MOESM7_ESM.txt", alpha=0.05))
$Seq
Error: 1Seq:Subj + 0MSE
Df Sum Sq Mean Sq F value Pr(>F)
Seq 1 129.18 129.176 426.75 < 2.2e-16
Error 998 302.09 0.303

$The Rest Terms
Df Sum Sq Mean Sq F value Pr(>F)
Per 1 221.566 221.566 719.2442 < 2.2e-16
Trt 1 3.342 3.342 10.8472 0.001024
Seq:Subj 998 302.090 0.303 0.9826 0.609149
RESIDUALS 998 307.438 0.308

Estimate Lower CL Upper CL
0.9215024 0.8846037 0.9599403

  user         system   elapsed 
2083.15        2.89     2089.66

============ to compare with ====================================
system.time(Analyse222BE("12248_2014_9661_MOESM7_ESM.txt", alpha=0.05))

222-BE analysis executed for '12248_2014_9661_MOESM7_ESM.txt' on 22-NOV.-2021 20:03:50.
_
Type I sum of squares: Analysis of Variance Table

Response: log(Var)
Df Sum Sq Mean Sq F value Pr(>F)
Trt 1 3.3415 3.3415 10.84720 0.0010242
Per 1 221.5658 221.5658 719.24417 < 2.22e-16
Seq 1 129.1755 129.1755 419.32802 < 2.22e-16
Subj 998 302.0898 0.3027 0.98261 0.6091493
Residuals 998 307.4376 0.3081

Type III sum of squares: (mimicking SAS/SPSS but with Seq tested against Subj)

log(Var) ~ Trt + Per + Seq + Subj
Df Sum Sq Mean Sq F value Pr(>F)
Trt 1 3.3415 3.3415 10.84720 0.0010242
Per 1 221.5658 221.5658 719.24417 < 2.22e-16
Seq 1 129.1755 129.1755 426.75115 < 2.22e-16
Subj 998 302.0898 0.3027 0.98261 0.6091493
Residuals 998 307.4376 0.3081

Back-transformed PE and 90% confidence interval
CV (%) ..................................: 60.06
Point estimate (GMR).(%).................: 92.15
Lower confidence limit.(%)...............: 88.46
Upper confidence limit.(%)...............: 95.99

   user       system    elapsed 
   5.32        0.04        5.41