The Advanced-Stage Hodgkin Lymphoma International Prognostic Index: Development and Validation of a Clinical Prediction Model From the HoLISTIC Consortium

Abstract

Purpose: The International Prognostic Score (IPS) has been used in classic Hodgkin lymphoma (cHL) for 25 years. However, analyses have documented suboptimal performance of the IPS among contemporarily treated patients. Harnessing multisource individual patient data from the Hodgkin Lymphoma International Study for Individual Care consortium, we developed and validated a modern clinical prediction model.

Methods: Model development via Transparent Reporting of a multivariable prediction model for Individual Prognosis Or Diagnosis guidelines was performed on 4,022 patients with newly diagnosed advanced-stage adult cHL from eight international phase III clinical trials, conducted from 1996 to 2014. External validation was performed on 1,431 contemporaneously treated patients from four real-world cHL registries. To consider association over a full range of continuous variables, we evaluated piecewise linear splines for potential nonlinear relationships. Five-year progression-free survival (PFS) and overall survival (OS) were estimated using Cox proportional hazard models.

Results: The median age in the development cohort was 33 (18-65) years; nodular sclerosis was the most common histology. Kaplan-Meier estimators were 0.77 for 5-year PFS and 0.92 for 5-year OS. Significant predictor variables included age, sex, stage, bulk, absolute lymphocyte count, hemoglobin, and albumin, with slight variation for PFS versus OS. Moreover, age and absolute lymphocyte count yielded nonlinear relationships with outcomes. Optimism-corrected c-statistics in the development model for 5-year PFS and OS were 0.590 and 0.720, respectively. There was good discrimination and calibration in external validation and consistent performance in internal-external validation. Compared with the IPS, there was superior discrimination for OS and enhanced calibration for PFS and OS.

Conclusion: We rigorously developed and externally validated a clinical prediction model in > 5,000 patients with advanced-stage cHL. Furthermore, we identified several novel nonlinear relationships and improved the prediction of patient outcomes. An online calculator was created for individualized point-of-care use.

FIG 1.

Relationship between continuous variables and survival. Plots of penalized smoothing splines based on partial residuals from multivariable Cox proportional hazard model for continuous variables. Plots demonstrate nonlinear associations between age with (A) PFS and (B) OS and absolute lymphocyte count with (C) PFS and (D) OS. HR, hazard ratio; OS, overall survival; PFS, progression-free survival.

FIG 2.

Distribution of predicted risk and observed outcomes stratified by predicted risk quartiles in the validation cohort. Distribution of predicted (A) PFS and (B) OS in the external validation cohort. Kaplan-Meier observed (C) PFS and (D) OS rates stratified by quartile of predicted risk in the external validation cohort. The term “event” on upper left x-axis and lower left y-axis of PFS plots indicates progression, relapse, or death. OS, overall survival; PFS, progression-free survival; Q, quarter.

FIG 3.

Calibration plots in the validation cohort. Deciles of predicted (A) PFS (B) or OS plotted against observed values in the external validation cohort. The diagonal line with a slope of 1 denotes perfect calibration. The red line shows the calibration slope using local regression (loess). Calibration curves show moderate calibration but with overestimate of risk in the highest decile of predicted risk. The term “event” on the y-axis of the PFS plot on the left indicates progression, relapse, or death. OS, overall survival; PFS, progression-free survival.