topic how to get get AIC/BIC for frailty model model fit comparison using proc phreg in Statistical Procedures

how to get get AIC/BIC for frailty model model fit comparison using proc phreg

jonyu — Wed, 25 Apr 2018 15:20:27 GMT

I want to get the AIC/BIC from proc phreg procedure for a frailty model. A snippet of the code is seen below:

proc phreg data=local.slk;
class eth_group gender_liv region_liv/ param=ref ref=first;
model gtime_ki_n*gstatus_ki_n(0)= bmi_calc_liv eth_group GENDER_liv /risklimits ;
random region_liv;

run;

Currently, the only output is the marginal likelihood estimate. Is there a way to get the AIC/BIC criterions as well?

Re: how to get get AIC/BIC for frailty model model fit comparison using proc phreg

aa84 — Fri, 01 Nov 2019 00:50:44 GMT

I realize this was posted a while ago, but was there any resolution? I am having similar difficulties with the restriction on proc phreg capabilities when implemented a shared frailty. How can one assess model fit?

Re: how to get get AIC/BIC for frailty model model fit comparison using proc phreg

shivarajmishra — Sun, 01 Dec 2019 03:02:34 GMT

Dear Trusted SAS Users,

I am having similar issues. I want to get AIC/BIC values for my Frailty model using a random statement to account for variability between studies in proc phreg.

Does anyone have an idea of how this can be calculated?

My (tentative) syntax is below,

proc phreg data=cvd1c;
ods output ParameterEstimates = ParameterEstimates;
class meno7cat(ref='50-52 yrs' param=ref) study byc3cat racethnicity5cat education3 smoking_b bmi5cat nochildcat hrtuse_b afb2;
model YearsCVD*CVDicd_ever(0)=meno7cat byc3cat racethnicity5cat education3 smoking_b bmi5cat nochildcat hrtuse_b afb2
/ALPHA=0.05 RISKLIMITS=wald;
Random study/solution;
format meno7cat meno7catf.;
run;

I am only getting the marginal likelihood value. Can this be equated as -2LogL?

Marginal Loglikelihood

-87071.6

Full output below,

Summary of the Number of Event and CensoredValuesTotal Event Censored PercentCensored

180207

8216

171991

95.44

Convergence Status

Convergence criterion (PCONV=0.0001) satisfied.

Marginal Loglikelihood

-87071.6

Testing Global Null HypothesisTest Chi-Square AdjustedDF Pr > ChiSqLikelihood RatioWald

3291.1031	51.44	<.0001
3660.9894	51.44	<.0001

Covariance Parameter EstimatesCov Parm REML Estimate StandardErrorstudy

0.4042

0.1813