Comparison of staging systems for extraocular retinoblastoma: analysis of 533 patients.

IMPORTANCE Different staging systems for extraocular retinoblastoma have been published, but to date they have not been validated in large cohorts. OBJECTIVE To review 533 patients (and pathology slides) with retinoblastoma included in 4 protocols (January 1, 1988, to December 31, 2009) who received uniform treatment. DESIGN AND SETTING Retrospective review in a hospital setting. A critical analysis for detecting inconsistencies and omissions was performed. PARTICIPANTS Patients were reclassified according to the modified St Jude Children's Research Hospital staging system, Grabowski-Abramson staging system, International Retinoblastoma Staging System (IRSS), and American Joint Committee on Cancer TNM staging system. MAIN OUTCOME AND MEASURE The main outcome measure was disease-free survival (DFS), considering only extraocular relapse as an event. RESULTS In the IRSS and the St Jude system, higher stages correlated with poorer DFS. For intraocular disease, only the TNM system and the IRSS included pathological definitions, and all systems except for the IRSS included substages without differences in DFS. Omissions of factors significantly associated with lower DFS included scleral invasion by the TNM system and massive choroidal invasion by the Grabowski-Abramson system. The St Jude system omits postlaminar optic nerve involvement, but this omission did not correlate significantly with lower DFS because these patients received intensive therapy. No differences in DFS were observed among substages for metastatic disease except for the presence of central nervous system involvement. All staging systems had inconsistencies in definitions of extent of disease. No system provides guidelines for imaging. CONCLUSIONS AND RELEVANCE Only the IRSS and the St Jude system allowed for grouping of patients with increasing risk of extraocular relapse. For lower stages, only the IRSS considers all unequivocal pathological prognostic factors. For higher stages, all systems had redundant information, resulting in an excess of substages.