Table 15.1, page 548.
use https://stats.idre.ucla.edu/stat/stata/examples/alda/data/firstcocaine, clear
generate event = ~censor
stset cokeage, failure(event)
failure event: event ~= 0 & event ~= .
obs. time interval: (0, cokeage]
exit on or before: failure
------------------------------------------------------------------------------
1658 total obs.
0 exclusions
------------------------------------------------------------------------------
1658 obs. remaining, representing
382 failures in single record/single failure data
56221 total analysis time at risk, at risk from t = 0
earliest observed entry t = 0
last observed exit t = 42
/* Model A */
stcox birthyr earlymj earlyod, efron nohr
failure _d: event
analysis time _t: cokeage
Cox regression -- Efron method for ties
No. of subjects = 1658 Number of obs = 1658
No. of failures = 382
Time at risk = 56221
LR chi2(3) = 247.83
Log likelihood = -2638.6141 Prob > chi2 = 0.0000
------------------------------------------------------------------------------
_t |
_d | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
birthyr | .1550843 .0199279 7.78 0.000 .1160263 .1941422
earlymj | 1.217073 .1640307 7.42 0.000 .8955789 1.538567
earlyod | .7911743 .1962008 4.03 0.000 .4066279 1.175721
------------------------------------------------------------------------------
display -2*e(ll) " " -2*(e(ll)-e(df_m))
5277.2282 5283.2282
/* Model B */
use https://stats.idre.ucla.edu/stat/stata/examples/alda/data/firstcocaine, clear
generate event = ~censor
expand cokeage
sort id
by id: generate t = _n
generate event2 = 0
by id: replace event2 = event if _n==_N
generate usemj = 0
replace usemj = 1 if t>mjage
generate useod = 0
replace useod = 1 if t>odage
compress
stset t, fail(event2) id(id)
stcox birthyr usemj useod, nohr efron
failure _d: event2
analysis time _t: t
id: id
Cox regression -- Efron method for ties
No. of subjects = 1658 Number of obs = 56221
No. of failures = 382
Time at risk = 56221
LR chi2(3) = 855.96
Log likelihood = -2334.5481 Prob > chi2 = 0.0000
------------------------------------------------------------------------------
_t |
_d | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
birthyr | .107414 .0214486 5.01 0.000 .0653754 .1494526
usemj | 2.551764 .2809543 9.08 0.000 2.001104 3.102425
useod | 1.853868 .1292125 14.35 0.000 1.600616 2.107119
------------------------------------------------------------------------------
display -2*e(ll) " " -2*(e(ll)-e(df_m))
4669.0962 4675.0962
/* Model C */
generate soldmj = 0
replace soldmj = 1 if t>sellmjage
generate moreod = 0
replace moreod = 1 if t>sdage
stcox birthyr usemj soldmj useod moreod, nohr efron
failure _d: event2
analysis time _t: t
id: id
Cox regression -- Efron method for ties
No. of subjects = 1658 Number of obs = 56221
No. of failures = 382
Time at risk = 56221
LR chi2(5) = 944.52
Log likelihood = -2290.2684 Prob > chi2 = 0.0000
------------------------------------------------------------------------------
_t |
_d | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
birthyr | .0849289 .0218326 3.89 0.000 .0421378 .1277201
usemj | 2.459197 .283572 8.67 0.000 1.903406 3.014988
soldmj | .6898893 .1226253 5.63 0.000 .4495482 .9302304
useod | 1.251102 .1565606 7.99 0.000 .944249 1.557955
moreod | .7603747 .1306618 5.82 0.000 .5042824 1.016467
------------------------------------------------------------------------------
display -2*e(ll) " " -2*(e(ll)-e(df_m))
4580.5369 4590.5369
/* Model D */
stcox birthyr earlymj usemj soldmj earlyod useod moreod, nohr efron
failure _d: event2
analysis time _t: t
id: id
Cox regression -- Efron method for ties
No. of subjects = 1658 Number of obs = 56221
No. of failures = 382
Time at risk = 56221
LR chi2(7) = 944.75
Log likelihood = -2290.1554 Prob > chi2 = 0.0000
------------------------------------------------------------------------------
_t |
_d | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
birthyr | .0835003 .0225715 3.70 0.000 .039261 .1277395
earlymj | .0752714 .1709014 0.44 0.660 -.2596892 .4102319
usemj | 2.452513 .2842857 8.63 0.000 1.895323 3.009702
soldmj | .6788679 .1249547 5.43 0.000 .4339611 .9237747
earlyod | -.0802819 .2032497 -0.39 0.693 -.4786439 .3180802
useod | 1.25428 .1572314 7.98 0.000 .9461116 1.562448
moreod | .7637909 .1321908 5.78 0.000 .5047017 1.02288
------------------------------------------------------------------------------
display -2*e(ll) " " -2*(e(ll)-e(df_m))
4580.3108 4594.3108
Table 15.2, page 555.
We have not worked this example yet, but here is how you can get the data.
use https://stats.idre.ucla.edu/stat/stata/examples/alda/data/relapse_days, clear
Table 15.3, page 560.
Note: Uses data from Table 15.1, Model C. The unstratified model is not repeated.
/* stratified model */
stcox birthyr usemj soldmj useod moreod, nohr efron strat(rural)
failure _d: event2
analysis time _t: t
id: id
Stratified Cox regr. -- Efron method for ties
No. of subjects = 1658 Number of obs = 56221
No. of failures = 382
Time at risk = 56221
LR chi2(5) = 928.30
Log likelihood = -2135.9495 Prob > chi2 = 0.0000
------------------------------------------------------------------------------
_t |
_d | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
birthyr | .0853703 .0218743 3.90 0.000 .0424974 .1282432
usemj | 2.457945 .2837022 8.66 0.000 1.901899 3.013991
soldmj | .6847278 .1228423 5.57 0.000 .4439613 .9254944
useod | 1.251934 .1566545 7.99 0.000 .9448965 1.558971
moreod | .7468379 .1312596 5.69 0.000 .4895738 1.004102
------------------------------------------------------------------------------
Stratified by rural
display -2*e(ll)
4271.899
/* nonrural model */
stcox birthyr usemj soldmj useod moreod if ~rural, nohr efron
failure _d: event2
analysis time _t: t
id: id
Cox regression -- Efron method for ties
No. of subjects = 1316 Number of obs = 44333
No. of failures = 328
Time at risk = 44333
LR chi2(5) = 776.89
Log likelihood = -1904.6792 Prob > chi2 = 0.0000
------------------------------------------------------------------------------
_t |
_d | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
birthyr | .0812738 .0235786 3.45 0.000 .0350607 .127487
usemj | 2.436957 .3154511 7.73 0.000 1.818684 3.05523
soldmj | .7151351 .1312688 5.45 0.000 .457853 .9724171
useod | 1.272721 .1715663 7.42 0.000 .936457 1.608985
moreod | .6924939 .1410193 4.91 0.000 .416101 .9688867
------------------------------------------------------------------------------
display -2*e(ll)
3809.3584
/* rural model */
stcox birthyr usemj soldmj useod moreod if rural, nohr efron
failure _d: event2
analysis time _t: t
id: id
Cox regression -- Efron method for ties
No. of subjects = 342 Number of obs = 11888
No. of failures = 54
Time at risk = 11888
LR chi2(5) = 153.00
Log likelihood = -230.47825 Prob > chi2 = 0.0000
------------------------------------------------------------------------------
_t |
_d | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
birthyr | .1097733 .0584298 1.88 0.060 -.0047469 .2242935
usemj | 2.517957 .6487525 3.88 0.000 1.246425 3.789488
soldmj | .4541637 .352952 1.29 0.198 -.2376095 1.145937
useod | 1.145638 .3842576 2.98 0.003 .3925069 1.898769
moreod | 1.105014 .3523088 3.14 0.002 .4145017 1.795527
------------------------------------------------------------------------------
display -2*e(ll)
460.9565
Table 15.4 Page 566. Notice that the data set has a duplicated id. This will cause problem when we perform stset with the id option. We assume that this is a data entry error that the id should be recoded to other unique number.
use https://stats.idre.ucla.edu/stat/stata/examples/alda/data/lengthofstay, clear
duplicates list id
Duplicates in terms of id
+------------+ | obs: id | |------------| | 86 845 | | 87 845 | +------------+
replace id = 80000 if _n==87 (1 real change made)
stset days, failure(censor= 0) id(id)
id: id
failure event: censor == 0
obs. time interval: (days[_n-1], days]
exit on or before: failure
------------------------------------------------------------------------------
174 total obs.
0 exclusions
------------------------------------------------------------------------------
174 obs. remaining, representing
174 subjects
172 failures in single failure-per-subject data
4938 total analysis time at risk, at risk from t = 0
earliest observed entry t = 0
last observed exit t = 100
Model A:
stcox treat, nohr efron
Cox regression -- Efron method for ties
No. of subjects = 174 Number of obs = 174
No. of failures = 172
Time at risk = 4938
LR chi2(1) = 0.89
Log likelihood = -718.31392 Prob > chi2 = 0.3449
------------------------------------------------------------------------------
_t | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
treat | .1457002 .1541486 0.95 0.345 -.1564254 .4478259
------------------------------------------------------------------------------
Model B:
stsplit new, every(1) (4764 observations (episodes) created)
gen tt1 = treat*(_t-1) stcox treat tt1, nohr efron
Cox regression -- Efron method for ties
No. of subjects = 174 Number of obs = 4938
No. of failures = 172
Time at risk = 4938
LR chi2(2) = 6.15
Log likelihood = -715.68705 Prob > chi2 = 0.0463
------------------------------------------------------------------------------
_t | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
treat | .7064112 .2924036 2.42 0.016 .1333106 1.279512
tt1 | -.0208327 .0092073 -2.26 0.024 -.0388786 -.0027868
------------------------------------------------------------------------------
Model C:
recode _t (min/7 = 1) (8/14 = 2) (15/21=3) (22/28=4) (29/35=5) (35/max=6), gen(catt) (4764 differences between _t and catt)
tab catt, gen(trt)
RECODE of |
_t | Freq. Percent Cum.
------------+-----------------------------------
1 | 1,169 23.67 23.67
2 | 1,060 21.47 45.14
3 | 893 18.08 63.22
4 | 652 13.20 76.43
5 | 386 7.82 84.24
6 | 778 15.76 100.00
------------+-----------------------------------
Total | 4,938 100.00
foreach X of numlist 1/6 {
2. replace trt`X' = trt`X'*treat
3. }
(616 real changes made)
(585 real changes made)
(506 real changes made)
(398 real changes made)
(209 real changes made)
(357 real changes made)
stcox trt1 - trt6, nohr efron
Cox regression -- Efron method for ties
No. of subjects = 174 Number of obs = 4938
No. of failures = 172
Time at risk = 4938
LR chi2(6) = 19.74
Log likelihood = -708.89176 Prob > chi2 = 0.0031
------------------------------------------------------------------------------
_t | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
trt1 | 1.571139 .6406079 2.45 0.014 .3155702 2.826707
trt2 | .5677856 .4928543 1.15 0.249 -.398191 1.533762
trt3 | .8497044 .3620698 2.35 0.019 .1400607 1.559348
trt4 | -.3498585 .3641467 -0.96 0.337 -1.063573 .3638559
trt5 | -.7696889 .4159832 -1.85 0.064 -1.585001 .0456231
trt6 | -.069058 .3144735 -0.22 0.826 -.6854147 .5472987
------------------------------------------------------------------------------
Model D:
gen t_lgtrt = treat*log(_t)/log(2) stcox treat t_lgtrt, nolog nohr efron
failure _d: censor == 0
analysis time _t: days
id: id
Cox regression -- Efron method for ties
No. of subjects = 174 Number of obs = 4938
No. of failures = 172
Time at risk = 4938
LR chi2(2) = 14.46
Log likelihood = -711.53087 Prob > chi2 = 0.0007
------------------------------------------------------------------------------
_t | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
treat | 2.533511 .7603294 3.33 0.001 1.043292 4.023729
t_lgtrt | -.5301232 .1618843 -3.27 0.001 -.8474105 -.2128358
------------------------------------------------------------------------------
Figure 15.4, Page 573
Top panel:
stcox treat, nohr strata(treat) basechazard(H0) efron
Stratified Cox regr. -- Efron method for ties
No. of subjects = 174 Number of obs = 4938
No. of failures = 172
Time at risk = 4938
LR chi2(1) = 0.00
Log likelihood = -602.5695 Prob > chi2 = 1.0000
------------------------------------------------------------------------------
_t | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
treat | (dropped)
------------------------------------------------------------------------------
Stratified by treat
gen logH0 = log(H0) (528 missing values generated)
separate logH0, by(treat) gen(y)
storage display value variable name type format label variable label ------------------------------------------------------------------------------- y0 float %9.0g logH0, treat == 0 y1 float %9.0g logH0, treat == 1
line y1 y0 _t if _t<=77, sort ylab(-4(1) 2) xlab(0(7) 77)
Bottom panel using the steps for the top panel.
keep _t treat y0 y1
save whole, replace file whole.dta saved
drop if treat ==0 (412 observations deleted)
drop y0 sort _t save treat1, replace file treat1.dta saved use whole, clear
drop if treat==1 (358 observations deleted)
drop y1 sort _t
save treat0, replace file treat0.dta saved merge _t using treat1 gen diff= y1-y0 (37 missing values generated)
line diff _t if _t<=77, sort ylab(-.25(.25) 1.75) xlab(0(7) 77)
Table 15.8, Page 601
We have not worked this example yet, but here is how you can get the data.
use https://stats.idre.ucla.edu/stat/stata/examples/alda/data/doctors, clear
Table 15.9, Page 604
We have not worked this example yet, but here is how you can get the data.
use https://stats.idre.ucla.edu/stat/stata/examples/alda/data/monkeys, clear