For a retrospective study, I'd recommend that you start with the choices made for TCGA clinical data in Cell 173:400-416, 2018.
we chose the date of diagnosis as time zero for time-to-event calculations ...
The authors provide several reasons for that choice. It's typically the best-defined of the dates in retrospective data, and the date that probably matters most from a patient's perspective. That choice also allows you to evaluate the interval between diagnosis and start of treatment as a predictor in your model.
There might be some ambiguity in the start date of treatment if some patients receive chemoradiation adjuvant to primary surgery. You'd have to decide whether that is the date of surgery or of the start of adjuvant therapy. I'd worry about using the end date of treatment, as not all patients can tolerate a full chemoradiation series.
You could make any choice you want for time zero, however, provided that it is clear and consistent.
With respect to progression-free survival, it depends on patient tumor status at death. Quoting again from the TCGA clinical data paper, where they call this the "progression-free interval" (PFI):
PFI is the period from the date of diagnosis until the date of the first occurrence of a new tumor event (NTE), which includes progression of the disease, locoregional recurrence, distant metastasis, new primary tumor, or death with tumor. Patients who were alive without these event types, or died without tumor were censored ... The event time is the shortest period from the date of initial diagnosis to the date of an event. The censored time is from the date of initial diagnosis to the date of last contact or the date of death without disease.
You should recognize, however, that this definition of PFI can lead to bias. In practice, progression might be discovered at one of a series of clinical follow-up visits, so you only know that the progression occurred during the time interval between the immediately prior clinical visit and the visit that found progression.
Placing the date of progression at the former or the latter date will tend to bias the PFI toward shorter or longer times. It is preferable to treat the PFI in such cases as interval-censored, for which you need to record both the date of the last progression-free visit and that of the visit at which progression was detected.
