Protocol: generalisability of breast cancer survival prediction models across countries, and the local data needed to recalibrate them (P-CORE Paper 1)

Study protocol, version 3 (final), 2026-10-14, of a development, external validation and model-updating study in oestrogen receptor-positive early breast cancer. Six survival models of graded flexibility are developed on a West of Ireland registry cohort, frozen, and applied unchanged to two confirmatory cohorts (SCAN-B, Sweden; SEER Research Data, United States), with PREDICT Breast v3.1 as benchmark. This protocol was issued after the analysis code had been built and run on pilot cohorts (METABRIC, Rotterdam) and before any model was applied to a confirmatory cohort. Section 5 states which pilot-stage results were known when it was written. The analysis code is tagged paper1-freeze-v1 after this deposit; confirmatory analyses are run once per cohort, and any later change is a dated amendment. Funding: Health Research Board (Ireland), SDAP-2023-042. The funder had no role in the design of the study. This version adds a reading note, written on 2026-10-14 after the code freeze and before any result for hypotheses H2 or H3 in SCAN-B or SEER had been seen. The frozen code labels as primary a different one of two pre-computed options than the protocol does, for H2 (baseline recalibration method) and H3 (model form in which the range of C between models is taken). The note records that the protocol governs: the protocol's rows are the results of record and the rows the code labels primary are reported as sensitivity analyses. No analysis is added, removed or rerun. The protocol file is unchanged from the previous version.