In this issue of Blood Advances, Maurer et al1 have refined their original advanced-stage Hodgkin lymphoma International Prognostic Index (A-HIPI) risk stratification model for advanced Hodgkin lymphoma.2 In this article, they identified risk groups that may affect progression-free survival and overall survival by using standard chemotherapy-based continuous variables and developed an online calculator to measure risk. They recognized and identified some limitations of the model, namely that the A-HIPI model was “right skewed” and that there was variability in what was considered “high-risk” in the sense that risk is sometimes in the eye of the beholder.
The study of Hodgkin lymphoma has been a model for clinical and translational cancer research for decades. On the clinical side, the Stanford studies of Kaplan and Rosenberg3 first introduced algorithms in which careful anatomical staging was used as a prognostic factor (later adopted as a standard tool in other malignancies), which enabled the completion of clinical trials with cohorts of equally matched patients. This staging system was then refined and expanded by the German Hodgkin Lymphoma Group as the International Prognostic Score,4 which was used to better match risk groups and that proved to be critical for both clinical trials and day-to-day patient care. On the basic research side, first came the recognition of the mysterious Hodgkin’s cell as a crippled B cell5 that was missing the usual decorations that we have come to expect from most B cells. Then came the observations by Shipp et al that amplification of 9p24.1 led to concordant alterations of the programmed death-ligand 1 and -2 loci.6 This ultimately led to the novel and practice-changing use of checkpoint inhibitors in later lines of therapy,7 followed by concurrent use with chemotherapy as first-line therapy8 and finally establishing checkpoint inhibitors as standard of care when used in combination with chemotherapy.9
Because of these advances, we now find ourselves in a happy dilemma. Patients are doing so well that it is hard to dissect prognostic factors that might aid us in future clinical trials or that might inform the treatment decisions following clinical trials. If our new treatments are so good that they overcome the traditional prognostic factors, why do we need prognostic factors? Are there really 4 stages of classical Hodgkin lymphoma (cHL) or perhaps only 2 stages, namely (1) very favorable and (2) everything else? These questions, based on the extremely good outcomes we have come to expect for most patients with both early-stage and advanced cHL, beg for new and more robust prognostic models, and that is what Maurer et al have done with the Holistic data in this study. They have refined the A-HIPI, which may offer more clinical utility and a more robust and practical model for real-world use. This model may have the most utility for clinical trial design because the real-word use of clinical models in advanced cHL has questionable utility now. Although these older models are of interest, there are no clear differences yet based on choice of treatment regimen possibly because the models are so skewed to favorable outcomes and because outcomes are so favorable using these older prognostic models. Maurer et al have attempted to make the model more robust and relevant to both clinical trial design and real-world clinical practice.
Building on this clinical foundation, the statistical methodology in this article is noteworthy as well. A-HIPI’s reliance on continuous variables, rather than arbitrary categories, preserves the full range of information in the data and allows for more precise risk predictions. For example, variables such as age and lymphocyte count are modeled continuously, capturing subtle and nonlinear relationships in patient outcomes. This approach avoids oversimplification and enhances the model’s relevance across diverse clinical scenarios. The study also demonstrated the reliability of the A-HIPI model through rigorous validation with both internal and external data sets. Validation confirms that the model’s predictions aligned closely with the observed outcomes. Calibration, which measures how well predicted risks match actual outcomes, was carefully assessed to ensure that the model’s performance remained consistent across patient groups. These steps reinforce the model’s utility for both clinical practice and research applications. A workflow for this prognostic model is summarized (see figure).
A-HIPI model workflow: data-driven risk stratification for personalized treatment. The flowchart illustrates the A-HIPI model workflow, starting with continuous variables like age and lymphocyte count, which are analyzed without arbitrary cutoffs to preserve data integrity. Through statistical modeling, nonlinear relationships are captured to generate risk predictions. Patients are stratified into risk groups using 1 of 3 methods, namely clinical thresholds (predefined cutoffs), deviations from average (comparison to cohort norms), or rank-based stratification (percentile-based grouping). These stratification methods influence the output, predictions of PFS and OS, to guide personalized treatment and clinical trial design. OS, overall survival; PFS, progression-free survival.
A-HIPI model workflow: data-driven risk stratification for personalized treatment. The flowchart illustrates the A-HIPI model workflow, starting with continuous variables like age and lymphocyte count, which are analyzed without arbitrary cutoffs to preserve data integrity. Through statistical modeling, nonlinear relationships are captured to generate risk predictions. Patients are stratified into risk groups using 1 of 3 methods, namely clinical thresholds (predefined cutoffs), deviations from average (comparison to cohort norms), or rank-based stratification (percentile-based grouping). These stratification methods influence the output, predictions of PFS and OS, to guide personalized treatment and clinical trial design. OS, overall survival; PFS, progression-free survival.
Equally important is the exploration of different methods to define risk groups. The authors compared clinical thresholds, deviations from average patient outcomes, and rank-based stratification and ultimately found that the latter was preferred for its flexibility. Rank-based stratification enables tailored definitions of risk groups and balances statistical robustness with clinical practicality. Highlighting the trade-offs among these approaches added depth to the discussion and emphasized how statistical choices influence patient classification and treatment strategies.
A particularly practical feature of the article is its reference to an online R-Shiny application, which translates the A-HIPI model into an accessible tool for real-world use. This interactive platform allows clinicians to dynamically adjust the risk group definitions and explore individualized predictions, thereby bridging the gap between advanced analytics and clinical decision-making. The inclusion of external validation cohorts further underscored the generalizability of the A-HIPI model. Consistent performance across diverse data sets indicates its applicability in various clinical contexts. The article briefly acknowledged potential challenges, such as population differences or data quality, and that addressing these in future research will enhance the model’s impact.
In summary, this article advances our understanding of risk stratification in advanced Hodgkin lymphoma by leveraging robust statistical techniques and practical tools. The A-HIPI model’s continuous approach to prognostic variables, rigorous validation, and integration into user-friendly applications makes it a valuable resource for clinicians and researchers alike. By refining the existing prognostic models and addressing current challenges, the authors provide a framework that supports precision medicine and improved patient outcomes.
Conflict-of-interest disclosure: The authors declare no competing financial interests.