topic Re: Missing time-dependent covariate values in Statistical Procedures

Missing time-dependent covariate values

littleka — Sun, 01 May 2022 23:52:52 GMT

Hi all,

I am performing survival analysis with time-dependent covariates.

Essentially, the values for methadone_1-methadone_12 are either OTP, OBOT, both, or none. Same for otp_methadone_bup_1-otp_methadone_bup_12. Non_med_1-non_med_12 are either 0 or 1. There were no missing values when I ran frequency tables for all of these!

I continue getting the note that 606 event observations have missing values in the time-dependent covariates.

Here is my proc phreg

proc phreg data = outlib.EXAMPLE;
class age (ref='25-34') sex (ref='F') race (ref='White') ethnicity (ref='Not Hispanic')
         rural (ref='Urban')
         
         cancer (ref='0') chronic_kidney_disease (ref='0') chronic_lung_disease (ref='0')
         chronic_liver_disease (ref='0') diabetes_mellitus (ref='0') heart_disease (ref='0')
         obesity (ref='0') smoking (ref='0') meth_use_disorder (ref='0')
         cocaine_use_disorder (ref='0');
        
model dx_survival_months*covid_dx(0)=
/*demographics*/
age race ethnicity rural
/*comorbidities/predisposing conditions*/
cancer chronic_kidney_disease chronic_lung_disease chronic_liver_disease diabetes_mellitus
heart_disease obesity smoking meth_use_disorder cocaine_use_disorder
/*time dependent vars*/
m otp non

/ ties=efron risklimits;

ARRAY methadone_(*) methadone_1-methadone_12;
m = methadone_[dx_survival_months];

ARRAY otp_methadone_bup_(*) otp_methadone_bup_1-otp_methadone_bup_12;
otp = otp_methadone_bup_[dx_survival_months];

ARRAY non_med_(*) non_med_1-non_med_12;
non = non_med_[dx_survival_months];

format age age.;
run;

Any help is appreciated, and I can try to share more if this it is not clear. I'm not sure if it's an issue with how the time-dependent covariates are formatted in the data set, or what. Thanks!

Re: Missing time-dependent covariate values

Rick_SAS — Mon, 02 May 2022 18:02:37 GMT

I suspect there are 606 missing values for dx_survival_months. (Or maybe 202 missing values?) You are using that as an index into the 0/1 variables

methadone_1-methadone_12
otp_methadone_bup_1-otp_methadone_bup_12
non_med_1-non_med_12

I think an array lookup such as methadone_[dx_survival_months] is resulting in a missing value when the index value (dx_survival_months) is missing.

Re: Missing time-dependent covariate values

littleka — Tue, 03 May 2022 04:34:06 GMT

Thank you for the response. Unfortunately in running a frequency table on the dx_survival_months variable, there are no missing values.

Re: Missing time-dependent covariate values

Rick_SAS — Tue, 03 May 2022 10:07:40 GMT

Interesting. Well, they are coming from somewhere, so I suggest you write a DATA step to manufacture the m, otp, and non variables. This will give you complete control over the variables and the ability to inspect them and, optionally, track down where the missing values are coming from.

On the other hand, PROC PHREG will drop the observations that have missing covariates, so maybe the problem will go away when you create the variables outside of the procedure.

Re: Missing time-dependent covariate values

littleka — Tue, 03 May 2022 21:43:55 GMT

I will definitely try that out, thank you.