2
$\begingroup$

I found on this paper (D'Agostino, et al. 1990) that pooled logistic regression is close to the time dependent covariate Cox regression analysis.

I would like to be able to reproduce estimates obtained from a pooled logistic model with those from a time dependent Cox regression analysis.

My data looks like this :

A

id age female time0 time1 death

1 1 20 0 0 1 1

2 2 21 1 0 1 0

3 2 21 1 1 2 0

4 3 19 0 0 1 0

5 3 19 0 1 2 1

6 4 22 1 0 1 0

7 4 22 1 1 2 0

8 4 22 1 2 3 1

9 5 20 0 0 1 0

10 5 20 0 1 2 0

11 5 20 0 2 3 0

12 5 20 0 3 4 0

13 6 24 1 0 1 0

14 6 24 1 1 2 0

15 6 24 1 2 3 0

16 6 24 1 3 4 1

I tried to fit these 2 models :

glm(death~ age + female, data= A, family=binomial)
coxph(Surv(time0,time1, death) ~ age + female, A)

Estimates of the 1st model are :

Estimate Std. Error z value Pr(>|z|)

(Intercept) -2.65425 12.06138 -0.220 0.826

age 0.08814 0.61009 0.144 0.885

female -0.59971 2.17734 -0.275 0.783

Estimates of the 2nd model are :

    coef exp(coef) se(coef)     z    p

age -0.140 0.869 0.622 -0.23 0.82

female 0.156 1.168 2.121 0.07 0.94

I dont understand why I get very different estimates with the 2 models (I also had the same problem with different datasets). Do you have any idea ?

$\endgroup$
1
  • 2
    $\begingroup$ The abstract to the paper states that the estimates are "close," not the same. It would aid potential respondents if you were to post the results from these models. $\endgroup$
    – user78229
    Commented Jul 8, 2016 at 13:54

0

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Browse other questions tagged or ask your own question.