OBJECTIVES: In early-stage cancers, tumor resection may be followed by local recurrence (LR), distant metastases (DM), or death. When investigating DM-free survival, LR is a competing event (CE) as it modifies the risk of DM. We assessed DM-free survival according to 3 methods: the “Include” method (properly accounts for CE), as well as two simple but less optimal strategies, “Ignore” and “Censor”.

METHODS: Patients from PMSI (French national hospital discharge database) with a first resection of urothelial carcinoma in 2015 were followed-up to 12/31/2021, or death. LR, DM and death were identified with diagnosis and procedure codes. Primary outcome was DM-free survival defined as time between resection and occurrence of DM, or in-hospital death. DM cumulative incidence rate (DMCIR) was reported, with 95% confidence interval, using 3 methods. With the Ignore and Censor methods, Kaplan-Meier (KM) are estimated by either ignoring or censoring patients at CE occurrence. DMCIR are estimated as 1-KM. “Include” method is based on the Aalen-Johansen estimate, which combines a KM estimate and a risk-specific function considering multiple events, resulting in a cumulative incidence function (CIF), providing an estimate of the DMCIR.

RESULTS: Of the 3,370 patients identified, the first observed event was DM for 1,275 patients (37.8%). For 377 patients (10.0%), LR was followed by DM. For 66 patients (2.0%), LR was followed by death. Using ignore, censor, or include methods, 24-month DMCIRs were 46.4% [44.7%–48.1%], 42.6% [40.8%–44.4%] and 38.1% [36.5%–39.7%]. Furthermore, 60-month DMCIRs were 57.8% [56.4%–59.7%], 53.0% [51.4%–55.1%] and 46.2% [44.9%–48.0%].

CONCLUSIONS: With a non-negligible proportion of patients with CE, KM-based methods overestimated survival outcome compared to Aalen‑Johansen-based method as events were dependent. Differences increased with follow-up. This study emphasizes the importance of using appropriate methods to account for CE when assessing survival outcomes in early-stage cancers.