Dear all,
I thank you in advance.
I have a question.
It is the first time I run a frailty survival model to account for the clustering of patients within different hospitals.
I used "linktest" to assess the model specification and I got bad results (I wished for a significant _hat).
So I tried to look at the distribution of deviance residuals against predicted values.
I got this strange pattern.
How would you interpret it? In other survival models I got more homogeneous distribuitons of deviance reisduals.
The model I ran was the following:
I thank you in advance.
I have a question.
It is the first time I run a frailty survival model to account for the clustering of patients within different hospitals.
I used "linktest" to assess the model specification and I got bad results (I wished for a significant _hat).
Code:
------------------------------------------------------------------------------
_t | Haz. ratio Std. err. z P>|z| [95% conf. interval]
-------------+----------------------------------------------------------------
_hat | 3.442834 2.606742 1.63 0.103 .7806006 15.1846
_hatsq | 2.431289 12.82696 0.17 0.866 .0000785 75264.97
_cons | .9794605 .0972407 -0.21 0.834 .8062687 1.189855
------------------------------------------------------------------------------
Note: _cons estimates baseline hazard.
So I tried to look at the distribution of deviance residuals against predicted values.
I got this strange pattern.
How would you interpret it? In other survival models I got more homogeneous distribuitons of deviance reisduals.
The model I ran was the following:
Code:
. xi: streg ib3.AED_aggiunto Età i.Sesso i.EventiYN i.dicOutcome N_AED_prima_reclutamento Mesi_da_diagnosi_epilessia,
> distribution(exponential) frailty(gamma) shared(Prefisso_Centro) nolog forceshared
i.Sesso _ISesso_0-1 (naturally coded; _ISesso_0 omitted)
i.EventiYN _IEventiYN_0-1 (naturally coded; _IEventiYN_0 omitted)
i.dicOutcome _IdicOutcom_0-1 (naturally coded; _IdicOutcom_0 omitted)
Failure _d: OutYN==0
Analysis time _t: cs
Exponential PH regression
Gamma shared frailty Number of obs = 807
Group variable: Prefisso_C~o Number of groups = 23
Obs per group:
No. of subjects = 807 min = 1
No. of failures = 157 avg = 35
Time at risk = 159.0404 max = 80
LR chi2(10) = 1.81
Log likelihood = -483.1438 Prob > chi2 = 0.9976
--------------------------------------------------------------------------------------------
_t | Haz. ratio Std. err. z P>|z| [95% conf. interval]
---------------------------+----------------------------------------------------------------
AED_aggiunto |
1 | .9841484 .2479838 -0.06 0.949 .6005877 1.612667
2 | 1.018387 .3468174 0.05 0.957 .5224341 1.985155
4 | .9716402 .2861077 -0.10 0.922 .5455844 1.73041
5 | .9941576 .3782271 -0.02 0.988 .4716456 2.095534
|
Età | .9993997 .0059959 -0.10 0.920 .9877168 1.011221
_ISesso_1 | .9968113 .1654504 -0.02 0.985 .7199972 1.380051
_IEventiYN_1 | 1.193334 .2336239 0.90 0.367 .8130531 1.75148
_IdicOutcom_1 | .8259511 .1780857 -0.89 0.375 .5412828 1.26033
N_AED_prima_reclutamento | .9920639 .0687574 -0.11 0.908 .8660541 1.136408
Mesi_da_diagnosi_epilessia | 1.000015 .0002245 0.07 0.947 .9995748 1.000455
_cons | 1.06081 .4599163 0.14 0.892 .4535211 2.481291
---------------------------+----------------------------------------------------------------
/lntheta | -1.039067 .5173458 -2.01 0.045 -2.053046 -.0250876
---------------------------+----------------------------------------------------------------
theta | .3537847 .183029 .1283434 .9752245
--------------------------------------------------------------------------------------------
Note: Estimates are transformed only in the first equation to hazard ratios.
Note: _cons estimates baseline hazard.
LR test of theta=0: chibar2(01) = 16.01 Prob >= chibar2 = 0.000
Attached Files
