Abstract P015

Identification of Risk Categories and Stratification from the Advanced-Stage Hodgkin Lymphoma (AS-HL) International Prognostic Index (A-HIPI) Model

Background: Predictive modeling yields personalized risk prediction for individual patients (pt). The A-HIPI model for AS-HL (Rodday A. JCO 2023) leverages continuous variables to generate individualized probability of progression-free survival (PFS) events or death (OS) within the first 5 years (y) from diagnosis. Risk groups have clinical utility in informing the stratification of pt populations for future clinical trials. We examined approaches using the A-HIPI model to generate varied risk groups with detailed analyses of strengths & limitations.

Methods: Three approaches were examined for the generation of risk groups. Proposed cutoffs were defined using the distribution of A-HIPI risk scores & data from the clinical-trial-based development cohort. Validation was done via the A-HIPI validation cohort from cancer registries.

Results: Approach 1: Risk groups based on clinical thresholds. Clinicians provided estimates of PFS5 constituting high vs low risk. The skewed distribution of risk scores from the A-HIPI model limited this approach, as cutoffs of PFS5<70 and PFS>90 only identified 15% & <1% of pts, respectively. Approach 2: Risk groups based on deviation from “average” pt. The 5y PFS was 77% (95% CI: 76-78). We explored defining “standard risk” based on this CI with pts above or below thresholds classified as decreased or increased risk, respectively. This classified ~20% of patients into decreased and increased risk groups. Approach 3: Risk groups based on “ranking” of pts. We ranked the A-HIPI risk scores of the 4,022 pts in the development cohort and used the distribution of the risk scores as a benchmark. The risk profile for a future pt was then compared to this distribution (eg, how you rank compared to your peers). This approach allows flexibility for the user to define the tradeoff between size of the risk groups and magnitude of difference in predicted outcomes (Figure). Application of this approach also showed good alignment between the predicted model percentiles and the observed distribution of scores in the validation cohort. Additionally, this approach is more dynamic as it is agnostic to historical clinical benchmarks and allows for use of the model as treatments change.

Conclusions: We assessed 3 varied approaches to define risk groups from the A-HIPI individual risk prediction model. A flexible “rank-based” approach provided the most clinical utility, which may be leveraged for clinical trial design and AS-HL pt stratification.

Authors

Matthew Maurer, Susan K. Parsons, Jenica Upshaw, Angie Mae Rodday, Jonathan W. Friedberg, Andrea Gallamini, Massimo Federico, Eliza Hawkes, David Hodgson, Peter Johnson, Eric Mou, Kerry Savage, Pier Luigi Zinzani, Andrew Evens