lme() does not work with all fixed effects [🇷 for BE/BA]

posted by Astea – Russia, 2016-11-04 01:13 (2701 d 22:06 ago) – Posting: # 16771
Views: 19,412

Dear all!

Trying to understand lm and lme I found a simple and very clear instruction (Lme tutorial). Hope it would be useful for someone not familiar with such models.

I try to evaluate Data set I from QA in R. The model is as follows:

> mod.lme<-lme(log(Cmax) ~ seq +  prd + drug, random=~drug|subj/seq, data=QA)
summary(mod.lme)

Linear mixed-effects model fit by REML
 Data: QA
     AIC    BIC  logLik
  548.86 589.38 -263.43

Random effects:
 Formula: ~drug | subj
 Structure: General positive-definite, Log-Cholesky parametrization
            StdDev   Corr 
(Intercept) 0.750263 (Intr)
drug        0.044904 -0.885

 Formula: ~drug | seq %in% subj
 Structure: General positive-definite, Log-Cholesky parametrization
            StdDev   Corr 
(Intercept) 0.530707 (Intr)
drug        0.037234 -0.844
Residual    0.393976       

Fixed effects: log(Cmax) ~ seq + prd + drug
              Value Std.Error  DF t-value p-value
(Intercept)  7.4432   0.32947 219 22.5912  0.0000
seq         -0.0248   0.19682  75 -0.1261  0.9000
prd          0.0468   0.02052 219  2.2811  0.0235
drug         0.1465   0.04627 219 3.1665  0.0018
 Correlation:
     (Intr) seq    prd   
seq  -0.909             
prd  -0.204  0.053       
drug -0.251  0.006  0.000

Standardized Within-Group Residuals:
      Min        Q1       Med        Q3       Max
-3.001657 -0.408221 -0.023744  0.340791  5.038066

Number of Observations: 298
Number of Groups:
         subj seq %in% subj
           77            77


Then I try to interpret the data in order to estimate CI. I took 0.1465 as point estimation (exp(0,1465)=1,1578 and calculate CI as PE+- SE*t(0,1;219), where SE=0.04627.
The result is 115,78: 107,26 - 124,97 (comparing with method B from QA might be 115,73: 107,17- 124,97). Not ideal but seems to be similar. What is the reason for difference: rounding or different model, or am I do something wrong? :confused:

Later I try to recalculate it using bear and get

> Fixed effects: log(Cmax) ~ seq + prd + drug
   Value Std.Error  DF t-value p-value
  7.6332  0.151688 217  50.321  0.0000
 -0.0196  0.197671  75  -0.099  0.9212
  0.0003  0.064043 217   0.004  0.9965
  0.0381  0.062413 217   0.610  0.5425
  0.1474  0.063881 217   2.308  0.0219
  0.1449  0.046986 217   3.083  0.0023

 Point Estimate   CI90 lower   CI90 upper
          115.587      106.955      124.916


And finally if I try to use lme just from yjlee168's (post#14730),
< mod.lme<-lme(log(Cmax) ~ seq +  prd + drug, random=~drug -1|subj, data=TotalData, method="REML")

I got another result:
Fixed effects: log(Cmax) ~ seq + prd + drug
              Value Std.Error  DF t-value p-value
(Intercept)  7.4154  0.225498 219  32.885  0.0000
seq         -0.0007  0.141455  75  -0.005  0.9959
prd          0.0467  0.028185 219   1.658  0.0988
drug          0.1415 0.081819 219    1.730  0.0850

However it seemed to me strange, what model is in bear now?

And I still wonder how does the model affect the Point Estimation? For QA data set 1 I've got manually Tmean=7,830303 and Rmean=7,676036, T-R=0,1543 far from the estimated via lme PE.

Trying to understand afermentioned Phoenix results I see non-integer df. How can they be achieved manually?

"Being in minority, even a minority of one, did not make you mad"

Complete thread:

UA Flag
Activity
 Admin contact
22,957 posts in 4,819 threads, 1,636 registered users;
82 visitors (0 registered, 82 guests [including 7 identified bots]).
Forum time: 23:20 CET (Europe/Vienna)

Nothing shows a lack of mathematical education more
than an overly precise calculation.    Carl Friedrich Gauß

The Bioequivalence and Bioavailability Forum is hosted by
BEBAC Ing. Helmut Schütz
HTML5