3333 - Predicting Absolute Risk of First Relapse in Classical Hodgkin Lymphoma by Incorporating Contemporary Treatment Effects

B. M. P. Aleman¹, S. Roshani², O. Visser³, M. Schaapveld², and F. E. V. Leeuwen²; ¹The Netherlands Cancer Institute (NKI-AVL), Department of Radiation Oncology, Amsterdam, Netherlands, ²The Netherlands Cancer Institute, Department of Epidemiology, Amsterdam, Netherlands, ³Netherlands Comprehensive Cancer Organisation, Department of Registration, Utrecht, Netherlands

Purpose/Objective(s): Hodgkin lymphoma (HL) is a highly curable disease. Currently, models to identify HL patients at higher risk of relapse only include pretreatment prognostic factors. Inclusion of treatment information is, however, needed to weigh relapse risk versus risks of late complications of treatment. The objective of our study is to predict absolute risk of relapse with and without including treatment effects. Materials/

Methods: Prognostic and treatment information was collected for 1,629 patients treated for newly diagnosed classical HL at ages 15-60 years between 2008-2018. Cox proportional hazard models allowing for time-varying coefficients were used to predict absolute risks of 1^st relapse up to 5 years after diagnosis. Monte Carlo cross-validation was performed for internal validation. External validation was done on patients with HL of unspecified morphology (n=717).
Results: After a median follow up time of 5.6 years, 87 out of 959 patients with early stage (Ann Arbor stage I-II) and 123 out of 667 patients with advanced stage (Ann Arbor stage III-IV) disease developed a relapse. The 5-year cumulative hazard of relapse was 9.0% (95% confidence interval (CI) 7.0%-11.0%) and 19.5% (95% CI 15.9%-23.1%) among early and advanced stage patients, respectively. Advanced stage, abnormal leukocyte counts and presence of B-symptoms predicted higher risk of 1^st relapse. Receiving >4 cycles of ABVD (doxorubicin, bleomycin, vinblastine, and dacarbazine) or ABVD+radiotherapy predicted lower risk of 1^st relapse compared to =4 cycles of ABVD in early stage patients. Among advanced stage patients, receiving (escalated) BEACOPP (bleomycin, etoposide, doxorubicin, cyclophosphamide, vincristine, procarbazine, prednisone) predicted lower risk of 1^st relapse compared to ABVD, irrespective of numbers of cycles. Externally validated 3- and 5-year Inverse Probability of Censoring Weighted (IPCW) areas under the curve (AUCs) for model without treatment information were 0.60 (95% Confidence Interval (CI) 0.53-0.66) and 0.59 (95% CI 0.52-0.66), respectively. Our tool including treatment information resulted in significantly higher IPCW AUCs (0.68 (95% CI 0.62-0.74) and 0.65 (95% CI 0.58-0.72) at 3 and 5 years, respectively).
Conclusion: Our treatment-based system showed good calibration and significantly better discrimination than a prediction model based only on pre-treatment clinical factors. This treatment-based system can help to weigh risks and benefits of contemporary treatment options.