IARC 60th Anniversary - 19-21 May 2026
Session : 20/05/26 - Posters
Recurrence After Primary Invasive Cutaneous Melanoma; Population-based Risk Prediction Tool
GIORCELLI A. 1,2, VALBERG M. 1,2, GREEN A. 3,4, VEIERØD M. 2, COSTA D. 5,6, SVENDSEN H. 7, JESPERSEN H. 8, ANDREASSEN B. 5, ROBSAHAM T. 5, GHIASVAND R. 1,5
1 Oslo Centre for Biostatistics and Epidemiology, Oslo University Hospital, Oslo, Norway; 2 Oslo Centre for Biostatistics and Epidemiology, Department of Biostatistics, Institute of Basic Medical Sciences, University of Oslo, Oslo, Norway; 3 QIMR Berghofer Medical Research Institute, Brisbane, Australia; 4 Cancer Research UK Manchester and Faculty of Biology, Medicine and Health, University of Manchester, Manchester, United Kingdom; 5 Department of Research, Cancer Registry of Norway, Norwegian Institute of Public Health, Oslo, Norway; 6 Norwegian Research Centre for Women's Health, Oslo University Hospital, Oslo, Norway; 7 Department of Plastic, Hand, and Reconstructive surgery, National Burn Centre, Haukeland University Hospital, Bergen, Norway; 8 Department of Oncology, Oslo University Hospital, Oslo, Norway
Background: The population of patients surviving a cutaneous melanoma diagnosis is steadily increasing. These melanoma survivors are at risk of recurrence, i.e. the return of the disease after the patient has been declared disease-free following the completion of treatment. The recurrence risk varies widely between individuals, and factors predicting the risk have not been sufficiently studied.
Objectives: To estimate the risk of recurrence up to five years after a first primary invasive melanoma, by patient and tumor characteristics, while accounting for the competing risk of death, and to develop a competing risk prediction model using data from national registries.
Methods: We included 25,837 patients diagnosed with a first primary invasive melanoma in TNM stages I-III, between 2008 and 2021 in Norway, using data from the Cancer Registry of Norway, the Norwegian Melanoma Registry, and the Norwegian Patient Registry. The non-parametric Aalen-Johansen estimator was applied to estimate cumulative risks of recurrence and death without recurrence, overall and by patient and tumor characteristics. A non-parametric smoothing approach using β-splines was used to estimate cause-specific hazard rates for both outcomes. Based on the Fine and Gray method, a sex-specific model was developed to predict recurrence up to five years after diagnosis, using patient and tumor characteristics, and subsequently internally validated through resampling. We generated 100 bootstrap samples to assess model calibration (via calibration plots, observed versus expected estimation, intercept and slope plots), discrimination (via area under the curve, AUC, and C-index), and prediction error (via Brier Score) ability.
Results: Among 24,095 patients who satisfied the inclusion criteria, 3,276 were detected with a recurrence. For each patient and tumor characteristics the highest 5-year recurrence risk was observed in men (17.1%, 95% confidence interval: 16.4-17.9%); age-group 70-79 (16.8%, 15.7-17.9%); TNM stage III (41.5%, 38.7-44.4%); tumor thickness >4.00 mm (40.9%, 38.8-43.1%); ulcerated tumor (33.7%, 32.1-35.4%); nodular subtype (27.7%, 26.3-29.2%); head or neck location (18.6%, 17.2-20.2%).
For a patient with all the high-risk characteristics, predicted recurrence probabilities were 34.0% (CI: 28.5-40.0%), 57.9% (50.6-65.2%), and 67.4% (40.1-74.8%) at 1-, 3-, and 5 years after the diagnosis, respectively; while the corresponding probability of dying without a recurrence were 4.2% (3.1-5.7%), 11.8% (9.0-15.7%), and 17.8% (13.6-23.3%), respectively. The 5-year prediction model demonstrated excellent performance, with strong calibration (slope range: 0.97-1.05, intercept range: - 0.07-0.06), discrimination (AUC range: 0.77-0.80, C-index range: 76.2-77.8), and prediction errors (Brier Score range: 0.08-0.12). Comparable performance was observed for the competing outcome.
Conclusions/Implications: Highest/elevated recurrence risk was associated with male sex, age over 70, TNM stage III melanoma, tumors thicker than 4.0 mm, ulceration, nodular subtype, and tumors on the head or neck. The developed risk prediction model provides a robust prognostic tool to identify high-risk patients and supports more personalized and source efficient follow-up strategies in melanoma care.