Dear Statalisters,

In this first post on Statlist I apologize in advance if the format or details are not appropriate), I'd like to ask guidance for subgroup analysis in a Cox regression model using Stata v16.
I am analyzing the association between a gene variant (rsXX, with three potential genotypes: rsXX_num 1 rsXX_num 2 rsXX_num 3) and time to a composite outcome. In both uni and multivariate models, HR is significantly increased in rsXX_num 3 (vs rsXX_num 1, ref). Details are provided below:

Code:
. stcox i.rsXX_num, nolog
failure _d: Outcomerisk_combined == 8
analysis time _t: Time_PD_outcome_dayscorr
id: PatGen_ID

Cox regression -- Breslow method for ties

No. of subjects = 756 Number of obs = 756
No. of failures = 363
Time at risk = 674992
LR chi2(2) = 7.02
Log likelihood = -2033.5293 Prob > chi2 = 0.0299

------------------------------------------------------------------------------
_t | Haz. Ratio Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
rsXX_num |
2 | 1.086543 .1250257 0.72 0.471 .8671645 1.361421
3 | 1.537313 .2426225 2.72 0.006 1.128297 2.0946
------------------------------------------------------------------------------

.
Code:
. stcox DoPcreat Age Gender BMI CVD Diabetes i.rsXX_num, nolog
failure _d: Outcomerisk_combined == 8
analysis time _t: Time_PD_outcome_dayscorr
id: PatGen_ID

Cox regression -- Breslow method for ties

No. of subjects = 628 Number of obs = 628
No. of failures = 302
Time at risk = 577213.5
LR chi2(8) = 72.83
Log likelihood = -1600.2069 Prob > chi2 = 0.0000

------------------------------------------------------------------------------
_t | Haz. Ratio Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
DoPcreat240 | 2.035369 .9701659 1.49 0.136 .7996778 5.180497
Age | 1.011042 .0040155 2.77 0.006 1.003203 1.018943
Gender | 1.196518 .1479623 1.45 0.147 .9389859 1.524683
BMI | .9899537 .0135042 -0.74 0.459 .9638365 1.016779
CVD | 1.683695 .221632 3.96 0.000 1.300816 2.179268
Diabetes | 1.704068 .2312339 3.93 0.000 1.30612 2.223264
|
rsXX_num |
2 | 1.098831 .1400671 0.74 0.460 .8559113 1.410694
3 | 1.732113 .3015799 3.16 0.002 1.231324 2.436575
------------------------------------------------------------------------------


As my cohort is composed of 5 subcohorts, I would like to obtain HR for rsXX_num 3 (vs rsXX_num 1 taken as a reference) in each subgroup. Is the code
Code:
stcox i.rsXX_num#i.Cohort_number , nolog
appropriate to test this question?

Based on the results provided below, am I allowed to conclude that HR is higher in rsXX_num 3 vs rsXX_num 1 in cohorts 1/2/3/5 (not 4) ?

Thank you very much for your comments and for your help.

Johann



Code:
. stcox i.rsXX_num#i.Cohort_number , nolog
failure _d: Outcomerisk_combined == 8
analysis time _t: Time_PD_outcome_dayscorr
id: PatGen_ID

Cox regression -- Breslow method for ties

No. of subjects = 756 Number of obs = 756
No. of failures = 363
Time at risk = 674992
LR chi2(14) = 89.22
Log likelihood = -1992.4261 Prob > chi2 = 0.0000

--------------------------------------------------------------------------------------------
_t | Haz. Ratio Std. Err. z P>|z| [95% Conf. Interval]
---------------------------+----------------------------------------------------------------
rsXX_num#Cohort_number |
1 2 | .9009012 .4122355 -0.23 0.820 .3674373 2.208875
1 3 | 2.570452 .7543683 3.22 0.001 1.446111 4.568961
1 4 | 1.021353 .3630243 0.06 0.953 .5088959 2.049851
1 5 | 3.380038 .8142878 5.06 0.000 2.107938 5.419825
2 1 | 1.633623 .4229501 1.90 0.058 .983497 2.713505
2 2 | .9060641 .3421188 -0.26 0.794 .4322704 1.899163
2 3 | 2.632542 .8024708 3.18 0.001 1.448458 4.784591
2 4 | 1.409253 .4038504 1.20 0.231 .8036365 2.471259
2 5 | 3.397648 .8171751 5.09 0.000 2.120577 5.443805
3 1 | 2.292504 .9024723 2.11 0.035 1.059801 4.959022
3 2 | 3.106154 1.688919 2.08 0.037 1.070031 9.016742
3 3 | 2.831661 1.399257 2.11 0.035 1.075035 7.458648
3 4 | 1.113009 .484567 0.25 0.806 .4741479 2.612663
3 5 | 5.122214 1.407804 5.94 0.000 2.988897 8.77818
--------------------------------------------------------------------------------------------

.
Code:
. stcox DoPcreat Age Gender BMI CVD Diabetes i.rsXX_num#i.Cohort_number , nolog
failure _d: Outcomerisk_combined == 8
analysis time _t: Time_PD_outcome_dayscorr
id: PatGen_ID

Cox regression -- Breslow method for ties

No. of subjects = 628 Number of obs = 628
No. of failures = 302
Time at risk = 577213.5
LR chi2(20) = 144.90
Log likelihood = -1564.1734 Prob > chi2 = 0.0000

--------------------------------------------------------------------------------------------
_t | Haz. Ratio Std. Err. z P>|z| [95% Conf. Interval]
---------------------------+----------------------------------------------------------------
DoPcreat240 | 1.953969 .9840493 1.33 0.183 .728181 5.243196
Age | 1.011828 .0044122 2.70 0.007 1.003217 1.020513
Gender | 1.203049 .1542438 1.44 0.149 .9357293 1.546737
BMI | .9882223 .0141289 -0.83 0.407 .9609146 1.016306
CVD | 1.58262 .2104134 3.45 0.001 1.219571 2.053743
Diabetes | 1.793188 .2476533 4.23 0.000 1.367943 2.350625
|
rsXX_num#Cohort_number |
1 2 | .8634405 .4086258 -0.31 0.756 .3415063 2.183062
1 3 | 2.18686 .6931283 2.47 0.014 1.174983 4.070151
1 4 | 1.223346 .4918156 0.50 0.616 .5563458 2.690008
1 5 | 3.524228 .9636173 4.61 0.000 2.062163 6.022891
2 1 | 1.409677 .4095051 1.18 0.237 .7977179 2.491094
2 2 | .8402843 .3326644 -0.44 0.660 .3867584 1.82563
2 3 | 2.474652 .8202406 2.73 0.006 1.292342 4.738608
2 4 | 1.638635 .5346087 1.51 0.130 .8645211 3.105909
2 5 | 3.860905 1.034039 5.04 0.000 2.284114 6.526199
3 1 | 2.722979 1.122167 2.43 0.015 1.214104 6.107071
3 2 | 4.102141 2.328255 2.49 0.013 1.348632 12.4775
3 3 | 2.906612 1.618698 1.92 0.055 .975778 8.658109
3 4 | 1.436237 .705047 0.74 0.461 .548747 3.759066
3 5 | 5.211703 1.59731 5.39 0.000 2.858241 9.502992
--------------------------------------------------------------------------------------------

.