Create a hazard graph with PROC PHREG including a random statement

LeenDS — Mon, 03 Oct 2022 15:39:18 GMT

Hi all,

I have a dataset that includes rows with

- people who viewed TV advertisements at a given time, and whether or not these people visited the website of the advertiser afterwards (+ the timestamp of this visit)

- a match for each of these people who did NOT see the TV advertisement at this same time, and whether or not these people visited the website of the advertiser afterwards (+ the timestamp of this visit)

To give a simplified idea:

On each row we also have a bunch of dummy variables for the week, weekend vs weekday, daypart (primetime etc).

I use PROC PHREG to model the duration of the treatment effect (=seeing a TV ad) on online response (web visit). We included a random term for the advertisement, because the same advertisement is shown many times on different timestamps and to different people.

We would like to produce a hazard graph including the time-related control variables + including the random term. Is that possible?

Ideally, I'd get something similar to the below output that I generated with PROC LIFETEST, but including control variables and the random term (which is not possible with PROC LIFETEST).

PROC PHREG only produces a cumulative hazard graph and a survival graph. By using the COVS statement, I can control for the time-related variables. For now, I calculated the hazard rates by doing cumhaz - lag(cumhaz) directly in the output dataset. Not sure if this is allowed?

Also, I cannot produce any output results if I include the random statement (The PLOTS=, OUTPUT=, ... options are ignored for a frailty model analysis). Does anyone know of a method to go around this?

Thank you so much in advance for any input!

current code below:

DATA covs;
FORMAT treatment best12. d_daytime_9_16 best12. d_earlyfringe_16_20 best12. d_primetime_20_23 best12. d_latefringe_23_5 best12.
d_weekend best12.
d_week14 best12. d_week15 best12. d_week16 best12. d_week17 best12. d_week18 best12. d_week19 best12. d_week20 best12.
d_week21 best12. d_week22 best12. d_week23 best12. d_week24 best12. d_week25 best12. d_week26 best12.;
INPUT treatment d_daytime_9_16 d_earlyfringe_16_20 d_primetime_20_23 d_latefringe_23_5
d_weekend
d_week14 d_week15 d_week16 d_week17 d_week18 d_week19 d_week20 d_week21 d_week22 d_week23 d_week24 d_week25 d_week26
;
DATALINES;
0 0 1 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0
1 0 1 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0
;
RUN;

ods graphics on;
PROC PHREG DATA=metrics_v4 PLOTS(overlay range=0,10080)=(survival cumhaz);
CLASS advertisement treatment(ref="0") visit(ref="0")
d_daytime_9_16(ref="0") d_earlyfringe_16_20(ref="0") d_primetime_20_23(ref="0") d_latefringe_23_5(ref="0")
d_weekend(ref="0")
d_week14(ref="0") d_week15(ref="0") d_week16(ref="0") d_week17(ref="0") d_week18(ref="0") d_week19(ref="0") d_week20(ref="0") d_week21(ref="0") d_week22(ref="0") d_week23(ref="0") d_week24(ref="0") d_week25(ref="0") d_week26(ref="0")
;
MODEL survivaltime*visit(0) = treatment
d_daytime_9_16 d_earlyfringe_16_20 d_primetime_20_23 d_latefringe_23_5
d_weekend
d_week14 d_week15 d_week16 d_week17 d_week18 d_week19 d_week20 d_week21 d_week22 d_week23 d_week24 d_week25 d_week26
;
BASELINE COVARIATES=covs OUT=base SURVIVAL=surv CUMHAZ=cumhaz / ROWID=treatment;
RUN;

DATA test_haz;
SET base;
haz = cumhaz - lag(cumhaz);
if cumhaz=0 then haz = .;
RUN;

topic Create a hazard graph with PROC PHREG including a random statement in Statistical Procedures

Create a hazard graph with PROC PHREG including a random statement