Can follow up time be zero for proc phreg?

[SeaMoon_168]

I am not sure if proc phreg would count any observations with follow-up time = 0. Shall I change the follow-up time = end date - start date + 1 (Not follow-up time= end date - start date) so that even the start and end date are the same, the follow up time is 1 instead of 0.

[FreelanceReinh · Accepted solution]

Hello @SeaMoon_168,

I assume that you will use the "follow-up time" variable as the "response" variable in a MODEL statement of the first form (see documentation), i.e.

MODEL response <*censor (list)> = effects </ options>;

In this case only negative or missing response values would be excluded from the analysis, whereas response=0 is no problem. Shifting all response values in the analysis dataset by a constant positive amount (e.g. response=response+100; in a DATA step) should leave all statistics in the PROC PHREG output unchanged and thus confirm that any zeros (before the shift) had no special bearing on the results.

[SeaMoon_168]

Thank you for your advice. However, I am still not sure if the model will not include any observations with follow up time = 0 ? 

[FreelanceReinh]

---

@SeaMoon_168 wrote:

Thank you for your advice. However, I am still not sure if the model will not include any observations with follow up time = 0 ?

---

Do you see a note like this in the log?

NOTE: n observations were deleted due either to missing or invalid values for the time, censoring, frequency or explanatory variables or to invalid
      operations in generating the values for some of the explanatory variables.

Or a difference between the "Number of Observations Read" and the "Number of Observations Used" in the PROC PHREG output (section "Model Information") due to zeros in the follow-up times?

Or a difference in the model statistics after shifting the follow-up times as I suggested?

Or are you using the "Counting Process Style of Input"? That is, the second form of the MODEL statement:

MODEL (t1, t2)<*censor(list)> = effects </ options>;

In the latter case, yes, observations with t1=t2=0 would be excluded because PROC PHREG uses semiclosed time intervals (t1, t2] and (0, 0] equals the empty set.