Calculate Duration (Index Date to Time-to-Event) with Competing Risks

[mels9808]

Hi SAS communities, 

I am performing a survival analysis where the event of interest is outpatient prescription and competing risks include death, admission to hospital, and admission to long-term care. I have been trying to write a code to find the follow-up time for each individual in the study; however, some individuals experience more than one event, and I want to calculate time from the earliest event experienced, whether it be the event of interest or a competing risk. 

This is a simplified version of the dataset:

ID	index_date	prescription	service_date	death_date	hosp_admdate	LTC_admdate	death	hosp	LTC	censor	censdate
001	09Sep2019	0	.	16Dec2019	.	.	1	0	0
002	01Feb2019	1	19Jun2019	06Jul2019	08May2019	.	1	1	0
003	06Dec2020	0	.	.	.	.	0	0	0
004	01Mar2020	1	05Mar2020	.	.	09Feb2020	0	0	1
005	02Dec2018	0	09Feb2019	.	01Mar2019	.	0	1	0

Service_date: date of prescription. End of study date: 10Mar2020.

I found a code online which creates a censor variable and censdate based on the event, I am just not sure how to modify this code so that it captures the multiple competing risks and takes only the earliest event as the censdate.

[ballardw]

You should provide an example of what you want as output for your given example data.

Earliest date, assuming you have SAS date values, is to use Min(death_date, hosp_admdate,ltc_admdate)

If you want the interval between the earliest event and Index_date then

interval = Min(death_date, hosp_admdate,ltc_admdate) - index_date;

[mels9808]

I was hoping to obtain some sort of variable that identifies the event that occurred first for the individual. For example, if death occurred first then censor=1, if hospitalization occurred first then censor=2, and if both death and hospitalization occurred but death occurred first then censor=1. I also wanted a variable that identifies the censoring date that corresponds to the date the earliest event occurred. I was then hoping to calculate follow-up time by doing censdate-index_date for each subject.

[ballardw]

I will repeat: please show what you expect for your example data.

Your "censoring date that corresponds to the date the earliest event " is the MIN of those variables. The Min function will only return missing if all the variables have missing values.

There is a WHICHN, and for character values, WHICHC, that returns a position of a value from a given list. So that can be used. But we need to see just what you expect to see in the data set to show example code.

I have to say that I feel like you are missing something when you state: ". For example, if death occurred first then censor=1, if hospitalization occurred first then censor=2, and if both death and hospitalization occurred but death occurred first then censor=1." There is nothing in the "if both" that modifies the result given by the first two values. Plus you said you have 3 dates/ events. Do I have guess that the third event is Censor=3?? Guessing is a poor way to program.

[mels9808]

This is what I am hoping to see in my data:

ID	index_date	prescription	service_date	death_date	hosp_admdate	LTC_admdate	death	hosp	LTC	censor	censdate
001	09Sep2019	0	.	16Dec2019	.	.	1	0	0	1	16Dec2019
002	01Feb2019	1	19Jun2019	06Jul2019	08May2019	.	1	1	0	2	08May2019
003	06Dec2020	0	.	.	.	.	0	0	0	4	10Mar2020
004	01Mar2020	1	05Mar2020	.	.	09Feb2020	0	0	1	3	09Feb2020
005	02Dec2018	1	09Feb2019	.	01Mar2019	.	0	1	0	0	09Feb2019

Service_date: date of prescription. End of study date: 10Mar2020.

If first event=prescription, censor=0 and censdate=service_date.

If first event=death, censor=1 and censdate=death_date.

If first event=hospitalization, censor=2 and censdate=hosp_admdate.

If first event=LTC, censor=3 and censdate=LTC_admdate.

If no events were experienced and subjects were followed until end of study, censor=4 and censdate=end of study date.

I hope this clears things up!!

[s_lassen]

You should present your example data as a data step, like this:

data have;                                                                                                                              
length ID $3                                                                                                                            
       index_date                                                                                                                       
       prescription                                                                                                                     
       service_date                                                                                                                     
       death_date                                                                                                                       
       hosp_admdate                                                                                                                     
       LTC_admdate                                                                                                                      
       death                                                                                                                            
       hosp                                                                                                                             
       LTC  8;                                                                                                                          
  informat                                                                                                                              
       index_date                                                                                                                       
       service_date                                                                                                                     
       death_date                                                                                                                       
       hosp_admdate                                                                                                                     
       LTC_admdate  date9.;                                                                                                             
  format                                                                                                                                
       index_date                                                                                                                       
       service_date                                                                                                                     
       death_date                                                                                                                       
       hosp_admdate                                                                                                                     
       LTC_admdate date9.;                                                                                                              
infile cards truncover;                                                                                                                 
input ID--LTC;                                                                                                                          
cards;                                                                                                                                  
001 09Sep2019 0 .      16Dec2019      .      .      1      0      0                                                                     
002 01Feb2019 1 19Jun2019      06Jul2019      08May2019      .      1      1      0                                                     
003 06Dec2020 0 .      .      .      .      0      0      0                                                                             
004 01Mar2020 1 05Mar2020      .      .      09Feb2020      0      0      1                                                             
005 02Dec2018 0 09Feb2019      .      01Mar2019      .      0      1      0                                                             
;run;

You can get the result you want like this:

data want;                                                                                                                              
  set have;                                                                                                                             
  censdate=min(service_date,death_date,hosp_admdate,LTC_admdate);                                                                       
  if missing(censdate) then do;                                                                                                         
    censor=4;                                                                                                                           
    censdate='10mar2020'd; /* end of study */                                                                                           
    end;                                                                                                                                
  else if censdate=death_date then                                                                                                      
    censor=1;                                                                                                                           
  else if censdate=hosp_admdate then                                                                                                    
    censor=2;                                                                                                                           
  else if censdate=LTC_admdate then                                                                                                     
    censor=3;                                                                                                                           
  else censor=0;                                                                                                                        
  format censdate date9.;                                                                                                               
run;         

You did not describe what to do if e.g. both hospital admission and death happen on the same date - in this case death will be the end result (censor=1), as that comes first in the code. If you want different priorities, you will have to rearrange the code a bit.