Nested frailty analysis within multi-year clustered Randomized Control Trial (RCT) w/repeat subjects

[PeteMc]

Hi Folks,

Trying to fit a nested (shared or joint) frailty model using Penalized Likelihood estimation on the hazard function. Dataset is a multi-year clustered RCT

Variables:

Treatment: 0-placebo, 1-Intervention

riv: 0-no exposure to illness, 1-exposure to illness

tte: time to event of exposure (days)

community: cluster id 

subject_id: participant id

year: study year {1,2,3,4}

At each year, participants across 36 communities (varying in size ~100 to 150 people) are followed from April to October, monitored for exposure to illness. For the most part subjects re-enroll each year , and Treatment remains the same for all years. Exposure variable (riv) is set 0 at the beginning of each year.

Each year I obtain the HR 95% CI from Analysis of Maximum Likelihood Estimates
with Sandwich Variance Estimate using 

proc phreg data=rivdata covs(aggregate) covm ;
id community;
model tte*riv(0)=Treatment /ties=EFRON rl;
by year;
run;

But would like to fit all years together and see HR 95% CI. The two approaches I have tried are -

One:

proc phreg data=rivdata covs(aggregate) covm ;

class community subject_id;

model tte*riv(0)=Treatment /ties=EFROM rl;
/* strata year; */
random community;
run;

Two:

proc surveyphreg data=rivdata;
class Treatment;

cluster community;

/* cluster community subject_id; */

model tte*riv(0)=Treatment /ties=EFRON rl;
hazardratio Treatment;

/* strata year; */
run;

Any help would be appreciated to get correct HR 95% CI for all years combined.  

I've tried frailtypack in R, package is unstable, even after being re-released Oct 20, 2020.

Thanks

[SteveDenham]

What happens if you remove the BY year statement from your original code?  What about changing the RANDOM effect in your first PHREG statement to 

random community_year;

where community_year is created by concatenating the two variables?

One more thing - I am going to move this to the Analytics\Statistical Procedures community, where it should draw attention from people specifically knowledgeable about this.

SteveDenham

[PeteMc]

Thanks for quick replay SteveDenham.

Using your suggestion above, I get virtually same results as having.
random community;
strata year;

I need to reword my question - my explanation implies independence between years , need to adjust for repeated subject measures.

I'll reword my question and post in Analytics\Statistical procedures.

Thanks