s-news
[Top] [All Lists]

Coxph, frailty and convergence problem?

To: <s-news@wubios.wustl.edu>
Subject: Coxph, frailty and convergence problem?
From: Mr J Gomez <ecsfi@csv.warwick.ac.uk>
Date: Thu, 22 Nov 2001 00:25:24 +0000 (GMT)
Dear SPLUS users

I am estimating coxph with frailty. When I don't introduce the frailty
term, the model takes some time to converge, but it finally does (Model
1). Nervertheless, when I introduce the frailty term I obtain the output
at the end of the message (Model 2). Could (please) anyone explain me what
this result means? Is it a convergence problem? Am I doing anything wrong?

I have tried to modify the number of iterations in "outer.max" and
"iter.max", but I do not get any different result. I have also come to the
same results with simpler versions of Model 1 with frailty when I fix the
variance of the gamma distribution (in these cases the model offered the
usual output if I did not fix the gamma variance).

Thanks a lot.

Jaime Gómez.



Model 1. MODEL WITHOUT THE FRAILTY TERM

coxph(formula = Surv(ANOI, ANOF, CENSURA) ~ log(SIZE) + INTER + PROFI +
PROXIC + PROXIE + DENSI2 + MGROW2 + DEPHA + NUMSB + CMOBC + NMARK + (
        NMARK2) + MMKC + (MMKC^2) + SYMM, data = datos, na.action =
na.omit, iter.max = 1000, method = "efron")

  n= 34529

               coef exp(coef) se(coef)        z        p
log(SIZE)   1.26096 3.53e+000  0.12694   9.9335 0.0e+000
    INTER   3.09687 2.21e+001  1.41465   2.1891 2.9e-002
    PROFI  -0.00486 9.95e-001  0.14586  -0.0333 9.7e-001
   PROXIC   2.14637 8.55e+000  0.20287  10.5802 0.0e+000
   PROXIE   1.70102 5.48e+000  0.24168   7.0382 1.9e-012
   DENSI2   0.17147 1.19e+000  0.55422   0.3094 7.6e-001
   MGROW2   0.25917 1.30e+000  2.56358   0.1011 9.2e-001
    DEPHA   0.04178 1.04e+000  0.47238   0.0884 9.3e-001
    NUMSB  -0.11842 8.88e-001  0.02586  -4.5798 4.7e-006
    CMOBC -13.42520 1.48e-006  2.97066  -4.5193 6.2e-006
    NMARK   0.02665 1.03e+000  0.04498   0.5924 5.5e-001
   NMARK2  -0.07864 9.24e-001  0.08140  -0.9662 3.3e-001
     MMKC   0.14606 1.16e+000  0.07524   1.9414 5.2e-002
I(MMKC^2)  -0.00550 9.95e-001  0.00385  -1.4279 1.5e-001
     SYMM   2.06643 7.90e+000  0.47051   4.3919 1.1e-005

Rsquare= 0.025   (max possible= 0.083 )
Likelihood ratio test= 886  on 15 df,   p=0
Wald test            = 800  on 15 df,   p=0
Score (logrank) test = 3055  on 15 df,   p=0

Model 2. EXAMPLE OF AN ESTIMATION WITH THE FRAILTY TERM

 coxph(formula = Surv(ANOI, ANOF, CENSURA) ~ log(SIZE) + INTER + PROFI +
PROXIC + PROXIE + DENSI2 + MGROW2 + DEPHA + NUMSB + CMOBC + NMARK + (
+       NMARK2) + MMKC + (MMKC^2) + SYMM + frailty(NCAJA), data = datos,
na.action = na.omit, iter.max = 1000, method = "efron")
Call:
coxph(formula = Surv(ANOI, ANOF, CENSURA) ~ log(SIZE) + INTER + PROFI +
PROXIC + PROXIE + DENSI2 + MGROW2 + DEPHA + NUMSB + CMOBC + NMARK +
 (
        NMARK2) + MMKC + (MMKC^2) + SYMM + frailty(NCAJA), data = datos,
na.action = na.omit, iter.max = 1000, method = "efron")

                     coef se(coef)    se2 Chisq   DF        p
     log(SIZE)  2.29e-299 0.1202   0.0773   0    1.0 1.0e+000
         INTER  1.15e-299 1.5005   1.3453   0    1.0 1.0e+000
         PROFI  3.65e-300 0.1742   0.1651   0    1.0 1.0e+000
        PROXIC  1.15e-299 0.2622   0.2616   0    1.0 1.0e+000
        PROXIE  2.05e-301 0.3875   0.3857   0    1.0 1.0e+000
        DENSI2  1.04e-300 0.6089   0.6088   0    1.0 1.0e+000
        MGROW2 -5.80e-299 2.4184   2.4180   0    1.0 1.0e+000
         DEPHA  3.85e-300 0.4242   0.4238   0    1.0 1.0e+000
         NUMSB -2.93e-300 0.0241   0.0241   0    1.0 1.0e+000
         CMOBC  7.14e-299 3.1358   2.0606   0    1.0 1.0e+000
         NMARK -6.46e-300 0.0844   0.0721   0    1.0 1.0e+000
        NMARK2  2.25e-299 0.2453   0.2238   0    1.0 1.0e+000
          MMKC  8.49e-300 0.1601   0.1523   0    1.0 1.0e+000
     I(MMKC^2) -5.69e-301 0.0137   0.0134   0    1.0 1.0e+000
          SYMM  4.61e-299 0.6364   0.6107   0    1.0 1.0e+000
frailty(NCAJA)                            184   52.6 2.2e-016

Iterations: 10 outer, 10000 Newton-Raphson
     Variance of random effect= 0.992   EM likelihood = -1417.8
Degrees of freedom for terms=  0.4  0.8  0.9  1.0  1.0  1.0  1.0  1.0  1.0
0.4  0.7  0.8  0.9  1.0  0.9 52.6
Likelihood ratio test=268  on 65.44 df, p=0  n= 34529
Warning messages:
  Inner loop failed to coverge for iterations 1 2 3 4 5 6 7 8 9 10 in:
coxpenal.fit(X, Y, strats, offset, init = init, iter.max = iter.max,
 outer.max = outer.max, eps = eps, toler.chol = toler.chol, weights =
...



<Prev in Thread] Current Thread [Next in Thread>
  • Coxph, frailty and convergence problem?, Mr J Gomez <=