ABSTRACT
Outcomes in older patients with Hodgkin lymphoma (HL) are compromised by the interplay of patient frailty and disease aggressiveness; however, current stratification tools, including the International Prognostic Score (IPS) and Comprehensive Geriatric Assessment (CGA), lack sufficient prognostic discrimination in the older HL population. We developed and validated a prognostic model integrating metabolic tumor burden and geriatric assessment. In this retrospective study across 14 centers in China between 2006 and 2024, we enrolled 306 patients aged ≥ 60 years with histologically confirmed HL. Patients were randomly partitioned into training (n = 250) and validation (n = 56) cohorts. Among the 306 patients, 98 deaths and 130 progression events were recorded during follow‐up. We analyzed 21 candidate variables. The least absolute shrinkage and selection operator (LASSO) analysis and multivariable Cox regression identified five independent predictors for 5‐year overall survival (OS): age > 73 years, baseline 18F‐FDG PET/CT‐derived total lesion glycolysis (TLG) > 200, hemoglobin ≤ 101 g/L, activities of daily living (ADL) dependence, and lymphocyte‐depleted classical Hodgkin lymphoma (LDCHL). A weighted Geriatric Risk Score stratified patients into low‐, intermediate‐, and high‐risk groups. In the validation cohort, the Geriatric Risk Score showed higher discriminatory accuracy than the CGA for both OS (C‐index, 0.770 [95% CI, 0.674–0.867] vs. 0.671 [0.581–0.762]) and PFS (0.761 [0.668–0.853] vs. 0.635 [0.535–0.735]). Among advanced‐stage (Ann Arbor stage III–IV) patients, it also outperformed the IPS‐7 and IPS‐3. By integrating metabolic tumor burden and geriatric assessment, the Geriatric Risk Score improves risk stratification over existing tools in older patients with HL.
Keywords: 18F‐FDG PET/CT, geriatric risk score, Hodgkin lymphoma, metabolic tumor burden, older adults, prognostic stratification
1. Introduction
Hodgkin lymphoma (HL) follows a characteristic bimodal age distribution, with a second peak emerging in older adulthood that accounts for 20%–25% of all cases [1, 2, 3]. Although modern therapies have rendered HL highly curable in younger populations—achieving 5‐year overall survival (OS) rates of approximately 90%—outcomes in patients aged > 60 years remain inferior, with survival rates ranging from only 49% to 65% [4, 5, 6]. This prognostic disparity is multifactorial, driven by a complex interaction between disease‐related aggressiveness and patient‐related vulnerability [7, 8, 9].
An accurate assessment of the stage of disease in patients with HL is critical for the selection of the appropriate therapy. However, no dedicated prognostic tool currently exists for the geriatric HL population. Historically, the standard 7‐factor International Prognostic Score (IPS‐7) and its subsequently proposed simplified 3‐factor variant (IPS‐3) have served as the primary risk stratification tools; however, they possess significant limitations in the contemporary treatment landscape [10, 11]. First, derived from historical cohorts treated before 1992, both the IPS‐7 and IPS‐3 incorporate an age cutoff of 45 years, rendering them inherently insensitive for distinguishing risk within an exclusively older cohort (typically > 60 years) [12]. Second, these models rely solely on basic clinical parameters, failing to capture the biological heterogeneity of the tumor or the functional reserve of the host. To address the narrowing prognostic utility of the IPS in the modern era, the advanced‐stage Hodgkin lymphoma International Prognostic Index (A‐HIPI) was recently developed by the HoLISTIC consortium [13]. While the A‐HIPI provides improved survival prediction overall, it remains fundamentally ill‐suited for the geriatric population. The model was derived and validated predominantly using clinical trial data from younger adults (median age of approximately 33 years) and generally restricted inclusion to patients under the age of 65 [13]. Furthermore, recent external validation of the A‐HIPI in real‐world registries has highlighted the challenges of translating this index into broader, unselected populations [14]. Consequently, the A‐HIPI has demonstrated suboptimal discrimination in older real‐world cohorts, primarily because it does not account for age‐related functional decline, frailty, and the competing risks of mortality that disproportionately affect older adults. Ultimately, the inability of these indices to identify patients with sufficiently poor outcomes limits their clinical relevance for guiding de‐escalation or intensification strategies in older adults.
To address the complexity of aging not captured by the IPS, the Comprehensive Geriatric Assessment (CGA) has been adopted to quantify functional, cognitive, and somatic domains [15, 16]. Although CGA‐based tools have shown utility in predicting toxicity and survival in diffuse large B‐cell lymphoma (DLBCL) and HL, they characterize the “host” rather than the “tumor.” [17, 18, 19, 20] Prognosis in older HL is not determined by frailty alone but is a composite outcome of the patient's physiological reserve and the disease's intrinsic burden. Relying exclusively on functional status risks overlooking the poor prognosis of fit patients with high tumor burden, while potentially denying curative‐intent therapy to borderline‐unfit patients who have favorable biology and limited tumor burden.
To complement the assessment of the host, fluorine‐18 fluorodeoxyglucose positron emission tomography/computed tomography (18F‐FDG PET/CT) offers a measurable index of tumor burden [21]. Quantitative metrics, such as metabolic tumor volume (MTV) and total lesion glycolysis (TLG), have emerged as powerful predictors of survival in younger HL cohorts [22, 23]. However, the prognostic value of baseline metabolic parameters specifically in the geriatric HL population remains ill‐defined. Older adults often present with distinct metabolic profiles and competing risks of death, potentially altering the prognostic weight of tumor burden markers compared to younger patients [24]. Furthermore, it remains unknown whether high tumor burden retains its negative prognostic impact in the presence of severe functional impairment.
In this study, we first sought to elucidate the specific interaction between metabolic tumor burden and functional status in older patients with HL. Building upon these biological insights, we subsequently developed and validated a novel Geriatric Risk Model.
2. Methods
2.1. Study Design and Participants
This was a multicenter, retrospective cohort study involving patients diagnosed with HL between 2006 and 2024 across 14 centers in China. We included patients aged ≥ 60 years with histologically confirmed classical Hodgkin lymphoma (cHL) or nodular lymphocyte‐predominant Hodgkin lymphoma (NLPHL). The total study population of 306 patients was randomly partitioned into a primary training cohort (n = 250) and a validation cohort (n = 56) at a ratio of 4:1. All eligible patients diagnosed during the study period were included.
2.2. CGA
A baseline CGA was performed for patients at diagnosis to evaluate functional status and comorbidities. The assessment included activities of daily living (ADL), instrumental activities of daily living (IADL), comorbidities, and age. Based on the CGA results, patients were stratified into three categories: fit, unfit, and frail, according to the criteria established by Tucci et al. [18].
2.3. PET/CT Image Acquisition and Analysis
All patients underwent baseline whole‐body 18F‐FDG PET/CT scans prior to the initiation of systemic therapy. PET/CT images were retrospectively reviewed and analyzed by three experienced nuclear medicine physicians who were unaware of patient clinical outcomes and the geriatric assessment results. Image analysis was performed using a dedicated workstation (uWS, United Imaging Healthcare, Shanghai, China). Total Metabolic Tumor Volume (TMTV) and TLG were calculated to quantify the systemic metabolic tumor burden. The semi‐automatic segmentation method was employed to define tumor boundaries. Specifically, a threshold of 41% of maximum standardized uptake value (SUVmax) was applied to separate malignant lesions from physiological background activity. TMTV was defined as the sum of the metabolic volumes of all nodal and extranodal lesions exceeding the SUV threshold. TLG was calculated as the sum of the products of the metabolic tumor volume (MTV) and the mean SUV (SUVmean) for each individual lesion (TLG = Σ(MTV × SUVmean)).
2.4. Targeted Sequencing and Bioinformatics Analysis
Genomic DNA was extracted from formalin‐fixed, paraffin‐embedded tissues and fragmented for library construction using the KAPA Hyper DNA Library Prep Kit. Target enrichment was performed via hybridization capture using a custom panel covering 475 leukemia‐ and lymphoma‐related genes. Libraries were sequenced (paired‐end 150‐bp) at a CLIA‐ and CAP‐accredited laboratory (Nanjing Geneseeq Technology) to a mean coverage depth of > 1000×. Following quality control and alignment to the human reference genome (hg19) using BWA, somatic variants were called using VarScan2 with a variant allele frequency threshold of 1%, excluding synonymous and intronic changes. Copy number variations (CNVs) and structural rearrangements were identified using ADTEx and BreakDancer, respectively.
2.5. Statistical Analysis
OS was defined as the time from diagnosis to death; progression‐free survival (PFS) was defined as the time from diagnosis to disease progression, relapse, or death. Outcomes were determined via medical records and clinical follow‐up.
2.6. Data Sourcing for Geriatric Parameters
To ensure data reliability, functional metrics were systematically extracted from standardized “Nursing Admission Assessments,” a mandatory component of the oncological care pathway across the 14 participating centers. These assessments, required for nursing care level determination and safety management, are archived in institutional electronic medical record (EMR) systems or paper‐based systems. Consistent with simplified CGA criteria [25], we retrieved functional status from medical records and these nursing protocols.
2.7. Model Development and Validation
The statistical framework was structured as a two‐stage objective pipeline to ensure prognostic rigor and clinical reproducibility. First, to identify robust prognostic predictors, we utilized a least absolute shrinkage and selection operator (LASSO) Cox regression model in the training cohort. An initial pool of 21 candidate variables—comprising demographic and histological data (age, sex, histological subtype), functional and geriatric assessments (ECOG PS, ADL status, cognitive failure, CIRS‐G), disease characteristics (Ann Arbor stage, B symptoms, bulky disease, number of extranodal sites, number of involved lymph node regions, bone marrow involvement), laboratory parameters (hemoglobin, albumin, LDH, β2‐microglobulin, HBsAg), and metabolic PET/CT metrics (TLG, TMTV, SUVmax)—was entered into the model. Patients with missing data for any of the 21 candidate variables were excluded from the analysis. To maximize the retention of predictive information and avoid the bias of premature discretization, all continuous variables were entered into the LASSO model in their original, untransformed continuous forms. We employed 10‐fold cross‐validation to determine the optimal tuning parameter, which effectively mitigated overfitting and eliminated multicollinearity by shrinking irrelevant coefficients to zero. Second, variables with nonzero coefficients identified by the LASSO regression were subsequently incorporated into a multivariable Cox proportional hazards model to confirm their independent prognostic utility for OS. To prevent multicollinearity, metabolic parameters (TLG and TMTV) were evaluated in parallel models, with TLG being prioritized for the final model integration based on its higher prognostic performance. Final model variables were selected based on their independent statistical significance (p < 0.05) in the multivariable Cox regression, clinical relevance, and consistent predictive utility across both OS and PFS endpoints. This two‐stage approach ensured that the model was grounded in objective statistical selection while maintaining clinical parsimony for subsequent score translation.
2.8. Construction of the Geriatric Risk Score
A weighted risk score was developed based on the regression coefficients of the final multivariable Cox model, following the method described by Sullivan et al. [26]. Optimal cut‐points for continuous variables were determined by identifying thresholds that minimized the Akaike information criterion (AIC). Subsequently, patients were stratified into distinct risk groups based on the total risk score. The optimal stratification cut‐points for the final score were determined using X‐tile analysis algorithms implemented in R software.
2.9. Model Performance and Comparison
Discriminatory performance was evaluated using the concordance index (C‐index; survival package) and time‐dependent area under the curve (AUC; timeROC package). Differences in the AUCs between the proposed Geriatric Risk Score and established prognostic tools were statistically compared using DeLong's test. No model recalibration or updating was performed; generalizability was instead assessed by external validation.
2.10. Software and Significance Criteria
Statistical analyses were performed using R software, version 4.3.1 (R Foundation for Statistical Computing, Vienna, Austria). A two‐sided p value of less than 0.05 was considered statistically significant.
3. Results
3.1. Baseline Characteristics of the Study Cohorts
Baseline characteristics for the primary and validation cohorts are summarized in Table 1. The training cohort comprised 250 patients with a median age of 66 years (range, 60–93) and a male predominance (65.2%). Histological subtypes included NLPHL(4.0%) and cHL, with mixed cellularity (47.2%) and nodular sclerosis (22.8%) being the most common cHL variants. CGA was available for 97.6% (n = 244) of patients at diagnosis, classifying 87.3% as fit, 9.4% as unfit, and 3.3% as frail. First‐line therapy consisted of conventional ABVD in 56.0% of patients, AVD in 20.4%, and novel agent–containing regimens (brentuximab vedotin or anti‐PD‐1 antibody combined with AVD) in 18.4% (9.6% + 8.8%). Over a median follow‐up of 51 months, 107 disease progression events and 70 deaths were documented. The estimated 5‐year OS and PFS rates were 66.1% (95% confidence interval [CI], 59.0–73.9) and 51.6% (95% CI, 45.0–59.3), respectively.
TABLE 1.
Patient characteristics in the training and validation cohort.
| Characteristics | Training cohort (n = 250) | Validation cohort (n = 56) | p value |
|---|---|---|---|
| Age (years) | 0.503 | ||
| Median (range) | 66 (60–93) | 68 (60–92) | |
| ≤ 73 | 206 (82.4%) | 44 (78.6%) | |
| > 73 | 44 (17.6%) | 12 (21.4%) | |
| Sex | 0.702 | ||
| Female | 87 (34.8%) | 21 (37.5%) | |
| Male | 163 (65.2%) | 35 (62.5%) | |
| Ann Arbor | 0.118 | ||
| I–II | 85 (34%) | 13 (23.2%) | |
| III–IV | 165 (66%) | 43 (76.8%) | |
| CGA | 0.551 | ||
| Fit | 213 (85.2%) | 46 (82.1%) | |
| Unfit | 23 (9.2%) | 8 (14.3%) | |
| Frail | 8 (3.2%) | 2 (3.6%) | |
| Missing | 6 (2.4%) | ||
| Histology | 0.856 | ||
| cHL | 240 (96%) | 54 (96.4%) | |
| NSCHL | 57 (22.8%) | 15 (26.8%) | |
| MCCHL | 118 (47.2%) | 21 (37.5%) | |
| LRCHL | 23 (9.2%) | 6 (10.7%) | |
| LDCHL | 11 (4.4%) | 3 (5.4%) | |
| cHL NOS | 31 (12.4%) | 9 (16.1%) | |
| NLPHL | 10 (4%) | 2 (3.6%) | |
| LDH | 0.239 | ||
| Normal | 168 (67.2%) | 33 (58.9%) | |
| Elevated | 82 (32.8%) | 23 (41.1%) | |
| ECOG PS | 0.220 | ||
| 0–1 | 173 (69.2%) | 34 (60.7%) | |
| ≥ 2 | 77 (30.8%) | 22 (39.3%) | |
| Extranodal sites | 0.740 | ||
| 0–1 | 197 (78.8%) | 43 (76.8%) | |
| ≥ 2 | 53 (21.2%) | 13 (23.2%) | |
| ADL | 0.564 | ||
| Independent | 226 (90.4%) | 52 (92.9%) | |
| Dependent | 24 (9.6%) | 4 (7.1%) | |
| CIRS‐G | 0.221 | ||
| < 8 | 156 (62.4%) | 30 (53.6%) | |
| ≥ 8 | 94 (37.6%) | 26 (46.4%) | |
| Cognitive failure | 0.760 | ||
| No | 235 (94.0%) | 52 (92.9%) | |
| Yes | 15 (6.0%) | 4 (7.1%) | |
| Albumin | 0.008 | ||
| < 40 g/L | 101 (40.4%) | 12 (21.4%) | |
| ≥ 40 g/L | 149 (59.6%) | 44 (78.6%) | |
| Hemoglobin | 0.724 | ||
| ≤ 101 g/L | 57 (22.8%) | 14 (25.0%) | |
| > 101 g/L | 193 (77.2%) | 42 (75.0%) | |
| β2m | 0.275 | ||
| Normal | 105 (42.0%) | 28 (50%) | |
| Elevated | 145 (58.0%) | 28 (50%) | |
| Bulky disease | 0.091 | ||
| No | 228 (91.2%) | 55 (98.2%) | |
| Yes | 22 (8.8%) | 1 (1.8%) | |
| B symptoms | 0.228 | ||
| Absent | 162 (64.8%) | 41 (73.2%) | |
| Present | 88 (35.2%) | 15 (26.8%) | |
| Lymph node sites | 0.315 | ||
| < 5 | 147 (58.8%) | 37 (66.1%) | |
| ≥ 5 | 103 (41.2%) | 19 (33.9%) | |
| TMTV | 0.745 | ||
| < 120 | 162 (64.8%) | 35 (62.5%) | |
| ≥ 120 | 88 (35.2%) | 21 (37.5%) | |
| SUVmax | 0.901 | ||
| < 11 | 87 (34.8%) | 19 (33.9%) | |
| ≥ 11 | 163 (65.2%) | 37 (66.1%) | |
| TLG | 0.847 | ||
| ≤ 200 | 52 (20.8%) | 11 (19.6%) | |
| > 200 | 198 (79.2%) | 45 (80.4%) | |
| Treatment | 0.697 | ||
| ABVD | 140 (56.0%) | 29 (53.7%) | |
| AVD | 51 (20.4%) | 9 (16.7%) | |
| BEACOPP | 4 (1.6%) | 1 (1.9%) | |
| BV‐based | 24 (9.6%) | 5 (9.3%) | |
| PD1‐based | 22 (8.8%) | 9 (16.7%) | |
| CHOP‐like | 8 (3.2%) | 1 (1.9%) | |
| Other | 1 (0.4%) | 0 (0%) | |
| IPS‐3 a | 0.987 | ||
| 1 | 51 (30.9%) | 13 (30.2%) | |
| 2 | 85 (51.5%) | 22 (51.2%) | |
| 3 | 29 (17.6%) | 8 (18.6%) | |
| IPS‐7 a | 0.138 | ||
| 1 | 2 (1.2%) | 3 (7.0%) | |
| 2 | 25 (15.2%) | 3 (7.0%) | |
| 3 | 51 (30.9%) | 11 (25.6%) | |
| 4 | 51 (30.9%) | 14 (32.6%) | |
| 5 | 29 (17.6%) | 6 (14.0%) | |
| ≥ 6 | 5 (3.0%) | 3 (7.0%) | |
| Missing | 2 (1.2%) | 3 (7.0%) |
Note: Differences in the distribution of variables between the training and validation cohorts were evaluated using Pearson's Chi‐square test. Fisher's exact test was applied instead when more than 20% of cells had an expected frequency < 5 or if any expected frequency was < 1.
IPS‐7 and IPS‐3 were calculated only for patients with advanced‐stage disease (Ann Arbor stage III–IV; n = 208 overall, 165 training, 43 validation). Percentages for these rows are based on the advanced‐stage subset rather than the entire cohort.
3.2. Construction and Prognostic Stratification of the Geriatric Risk Score
A total of 21 candidate variables were considered, including clinical and laboratory parameters (age, sex, hemoglobin, Ann Arbor stage, ECOG performance status, number of extranodal sites, B symptoms, LDH, β2‐microglobulin, albumin, bulky disease, number of involved lymph node regions, HBsAg, bone marrow involvement, and histologic subtype), geriatric assessment indices (ADL status, cognitive failure, CIRS‐G), and PET/CT–derived parameters (SUVmax, TMTV, TLG). To test whether the metabolic and geriatric domains acted independently, we first fitted a Cox model for OS including TLG, CGA, and their interaction. Both TLG (HR 7.16, 95% CI 2.24–22.90, p = 0.001) and CGA (HR 8.96, 95% CI 1.49–53.87, p = 0.017) were independently associated with OS, while the interaction term was not significant (HR 0.35, p = 0.289) (Table S1). This indicates that the two domains provide independent, nonredundant prognostic information, supporting their joint inclusion in a single model.
To identify independent prognostic factors, we performed a two‐step analysis in the training cohort. First, we utilized LASSO Cox regression to minimize overfitting and eliminate multicollinearity from an initial pool of candidate variables (Figure S1A,B). Variables with nonzero coefficients were then entered into a multivariable Cox proportional hazards model. To address potential multicollinearity, TLG and TMTV were evaluated in separate models (Tables S2–S4, Figure S2); TLG was prioritized for the final model due to its higher prognostic value. For TLG, the calculated cutoff of 192.5 was adjusted to 200 to maximize clinical convenience. The final multivariable Cox model identified five independent predictors for OS (p < 0.05): age > 73 years (HR 4.42), TLG > 200 (HR 4.64), LDCHL subtype (HR 3.91), hemoglobin ≤ 101 g/L (HR 2.27), and ADL dependence (HR 2.83).
To facilitate clinical application, we formulated a weighted Geriatric Risk Score based on the regression coefficients of these five predictors (Table 2). As detailed in our scoring system, points were assigned to each high‐risk feature: 2 points for age > 73 years, 2 points for TLG > 200, 2 points for LDCHL subtype, 1 point for hemoglobin ≤ 101 g/L, and 1 point for ADL dependence (Figure 1A). Based on the total accumulated scores, patients were stratified into three distinct risk categories using optimal cut‐points derived from X‐tile analysis: Low Risk (score 0–1), Intermediate Risk (score 2–3), and High Risk (score ≥ 4) (Figure 1B).
TABLE 2.
Multivariable cox proportional hazards analysis of prognostic factors for overall survival (OS) and point assignment for the Geriatric Risk Score.
| Variable | Multivariable HR (95% CI) | Multivariable p value | β | Points |
|---|---|---|---|---|
| Age > 73 years | 4.42 (2.04–9.56) | < 0.001 | 1.486 | 2 |
| TLG > 192.5 | 4.64 (1.31–16.40) | 0.017 | 1.534 | 2 |
| Hemoglobin ≤ 101 g/L | 2.27 (1.27–4.00) | 0.006 | 0.813 | 1 |
| Stage III–IV | 1.87 (0.92–3.83) | 0.085 | 0.629 | — |
| ECOG PS ≥ 2 | 1.60 (0.89–2.88) | 0.114 | 0.472 | — |
| ADL dependent | 2.83 (1.14–7.03) | 0.025 | 1.041 | 1 |
| Histology | ||||
| MCCHL | 1.00 (reference) | — | — | 0 |
| NSCHL | 0.71 (0.36–1.40) | 0.321 | −0.347 | 0 |
| LDCHL | 3.91 (1.46–10.46) | 0.007 | 1.363 | 2 |
| LRCHL | 1.22 (0.35–4.25) | 0.752 | 0.201 | 0 |
| CHLNOS | 0.79 (0.36–1.73) | 0.553 | −0.237 | 0 |
| NLPHL | 0.90 (0.12–6.90) | 0.921 | −0.104 | 0 |
Note: Continuous variables (Age, TLG, Hemoglobin) were dichotomized at the optimal cutoff determined by minimizing the Akaike information criterion (AIC); time was administratively censored at 5 years. Mixed cellularity classical Hodgkin lymphoma (MCCHL), the most frequent histologic subtype in our cohort, was used as the reference category; the hazard ratio of each other subtype represents its risk relative to MCCHL. Points were assigned using the Sullivan method: each point equals the regression coefficient (β) divided by a constant B (B = 0.813, the smallest coefficient, corresponding to anemia), rounded to the nearest integer. The resulting raw ratios (β/B) before rounding were as follows: Age 1.83, TLG 1.89, Hemoglobin 1.00, ADL 1.28, and LDCHL 1.68, corresponding to 2, 2, 1, 1, and 2 points, respectively. Points were assigned only to variables that remained significant in the multivariable model; nonsignificant variables (Stage, ECOG PS) were not included in the score (—).
Abbreviations: ADL, activities of daily living; CHL NOS, classical Hodgkin lymphoma not otherwise specified; CI, confidence interval; ECOG PS, Eastern Cooperative Oncology Group performance status; HR, hazard ratio; LDCHL, lymphocyte‐depleted classical Hodgkin lymphoma; LRCHL, lymphocyte‐rich classical Hodgkin lymphoma; MCCHL, mixed cellularity classical Hodgkin lymphoma; NLPHL, nodular lymphocyte‐predominant Hodgkin lymphoma; NSCHL, nodular sclerosis classical Hodgkin lymphoma; TLG, total lesion glycolysis.
FIGURE 1.
Construction and prognostic stratification of the Geriatric Risk Score. (A) The weighted point assignment for the five independent prognostic variables comprising the simplified risk scoring system: Age > 73 years, TLG > 200, anemia (HGB ≤ 101 g/L), ADL dependence, and LDCHL histology. (B) The definition of the three‐tiered risk stratification groups based on total cumulative scores: Low Risk (score ≤ 2), Intermediate Risk (score 3–4), and High Risk (score ≥ 5). (C, D) Time‐dependent area under the receiver operating characteristic curve (AUC) demonstrating the predictive stability of the Geriatric Risk Score over a 5‐year follow‐up for (C) overall survival (OS) and (D) progression‐free survival (PFS). (E, F) Kaplan–Meier survival curves demonstrating significant risk discrimination across the three Geriatric Risk Score categories for both (E) OS and (F) PFS (both log‐rank p < 0.0001).
Time‐dependent AUC analyses demonstrated the discriminatory capacity of the model, with AUC values showing stable performance for predicting both OS and PFS over a 5‐year period (Figure 1C,D). Furthermore, Kaplan–Meier survival analysis confirmed that this three‐tiered stratification significantly discriminated outcomes across the cohort. The established risk groups showed distinct survival trajectories for both OS and PFS (both Log‐rank p < 0.0001) (Figure 1E,F). In summary, this composite score serves as a systematic and granular tool for risk discrimination in older patients with HL.
3.3. Benchmarking and Risk Reclassification Against the CGA
Given the verified prognostic relevance of the CGA in HL, we first evaluated its stratification capacity within our study population. Although the conventional CGA categories achieved statistical significance for both OS and PFS (p < 0.001) (Figure S3A,B), the tool exhibited limitations in risk resolution.
To rigorously benchmark the predictive accuracy of our newly derived Geriatric Risk Model, we performed a direct comparative analysis against the CGA within the full cohort. For the primary endpoint of OS, the proposed model demonstrated a markedly superior discriminative capacity, yielding a C‐index of 0.751 (95% CI, 0.709–0.813), which significantly outperformed the CGA (C‐index: 0.577; 95% CI, 0.520–0.634). Similarly, regarding PFS, our model maintained high predictive accuracy with a C‐index of 0.724 (95% CI, 0.679–0.769), capturing a substantially greater proportion of prognostic variance than the CGA (C‐index: 0.569; 95% CI, 0.526–0.611) (Figure 2A). This superior longitudinal performance was further substantiated by time‐dependent receiver operating characteristic (ROC) analysis, where the Geriatric Risk Score sustained consistently and markedly higher AUC trajectories for both OS and PFS across a 5‐year follow‐up timeline compared with those of the CGA (Figure 2B,C; DeLong's test at 1 and 5 years: both p < 0.001 for PFS; p = 0.147 and p < 0.001 for OS, respectively).
FIGURE 2.
Predictive performance and risk reclassification against CGA. (A) Comparison of predictive accuracy (Harrell's C‐index, with 95% confidence intervals) between the Geriatric Risk Score and standard clinical indices (CGA in the full cohort; IPS‐7 and IPS‐3 in the advanced‐stage Ann Arbor III–IV cohort). (B, C) Time‐dependent AUC comparisons demonstrating the superior and stable discriminative capacity of the Geriatric Risk Score versus the conventional CGA over a 5‐year period for predicting (B) progression‐free survival (PFS) and (C) overall survival (OS). (D) A Sankey diagram illustrating the bidirectional risk reclassification dynamics of patients from the traditional CGA functional categories (Fit, Unfit, Frail) to the proposed Geriatric Score classifications (Low, Intermediate, High).
To evaluate the clinical utility of our score in refining traditional functional assignments, we cross‐tabulated the Geriatric Risk Score classifications across the established CGA categories using a reclassification framework (Figure 2D). This analysis exposed profound prognostic heterogeneity within the seemingly uniform “Fit” cohort. Notably, only a subset of CGA‐fit patients was confirmed to have a genuinely favorable prognosis and remained in the “Low‐Risk” category. Conversely, the vast majority were upstaged by the new model due to adverse clinical or tumor‐specific features, being reclassified into the “Intermediate‐Risk” or “High‐Risk” categories. For functionally impaired individuals (Unfit and Frail), the model successfully consolidated risk stratification, with a significant proportion appropriately reassigned to the “High‐Risk” category. Taken together, these findings demonstrate that our composite model successfully dissects the broad “Fit” population, capturing critical, high‐risk biological and disease‐specific attributes that are entirely overlooked by standard functional assessments alone.
3.4. Benchmarking and Risk Reclassification Against the IPS
We first evaluated the prognostic performance of the established International Prognostic Score systems IPS‐3 and IPS‐7 in our older cohort. The IPS‐7 significantly stratified patients for OS (log‐rank p = 0.024) but not for PFS (p = 0.27) (Figure S3C,D). Similarly, the IPS‐3 discriminated OS (p = 0.013), while its association with PFS did not reach statistical significance (p = 0.061) (Figure S3E,F). Notably, this overall separation in OS was largely driven by the high‐risk group (IPS‐7 scores ≥ 5; IPS‐3 scores 3), whereas the low‐ and intermediate‐risk groups showed overlap and were not well separated (IPS‐7 scores 0–2 vs. 3–4; IPS‐3 scores 1 vs. 2). These results indicate that both established indices retained prognostic value for OS but offered limited discrimination for PFS and poor resolution among lower‐risk patients in this older population.
To further evaluate the clinical utility of the Geriatric Risk Score, we benchmarked its performance against the established International Prognostic Score systems (IPS‐3 and IPS‐7) within the advanced‐stage cHL subgroup. C‐index analysis demonstrated higher discriminative performance of the Geriatric Risk Score for both OS (0.711, 95% CI 0.644–0.778) and PFS (0.647, 95% CI 0.590–0.704), compared with IPS‐7 (OS: 0.621, 95% CI 0.542–0.700; PFS: 0.592, 95% CI 0.533–0.651) and IPS‐3 (OS: 0.573, 95% CI 0.489–0.657; PFS: 0.558, 95% CI 0.496–0.620) (Figure 3A). Longitudinal time‐dependent AUC analysis over a 5‐year follow‐up period demonstrated that the AUC trajectories for the Geriatric Risk Score in predicting PFS remained superior to those of both IPS‐3 (Figure 3B; DeLong's test at 1 and 5 years: p = 0.016 and p = 0.897, respectively) and IPS‐7 (Figure 3D; DeLong's test at 1 and 5 years: p = 0.208 and p = 0.417, respectively). Similarly advantages were observed for OS, where the new model outperformed IPS‐3 (Figure 3C; DeLong's test at 1 and 5 years: p = 0.005 and p = 0.117, respectively) and IPS‐7 (Figure 3E; DeLong's test at 1 and 5 years: p = 0.021 and p = 0.059, respectively). While the discriminative capacity of traditional IPS tools remained suboptimal and fluctuated over time for both endpoints in this older demographic, the Geriatric Risk Score exhibited robust and sustained predictive accuracy across the entire timeline.
FIGURE 3.
Benchmarking and reclassification dynamics against IPS. (A–D) Time‐dependent AUC trajectories over 5 years comparing the predictive performance of the Geriatric Risk Score against established disease‐specific indices in the advanced‐stage subgroup. The plots illustrate comparisons against the IPS‐3 for (A) progression‐free survival (PFS) and (B) overall survival (OS), as well as against the IPS‐7 for (C) PFS and (D) OS. (E, F) Sankey diagrams exposing prognostic heterogeneity and depicting the risk reclassification flow of patients from the conventional (E) IPS‐3 and (F) IPS‐7 risk tiers into the newly defined Geriatric Score classifications.
We next characterized the bidirectional risk reclassification patterns between the new model and traditional IPS strata using a reclassification framework (Figure 3F,G). Sankey diagram analyses clearly revealed substantial prognostic heterogeneity within conventional IPS categories. Notably, a significant proportion of patients initially classified as “Intermediate‐Risk” by IPS‐3 or IPS‐7 were precisely redistributed across distinct risk tiers based on their individualized clinico‐biological and geriatric profiles. Crucially, the Geriatric Risk Score refined the stratification of the conventional IPS “High‐Risk” cohort, safely downstaging a substantial subset with more favorable profiles (primarily to intermediate‐ or low‐risk groups), while simultaneously upstaging underestimated patients with occult biological vulnerabilities. These reclassification dynamics confirm that the Geriatric Risk Score overcomes the limitations of standard disease‐specific indices to provide highly granular risk resolution for older patients with HL.
3.5. External Validation of the Geriatric Risk Score
To evaluate the generalizability of the Geriatric Risk Score, external validation was performed using an independent cohort. Within the overall validation cohort, the model maintained high predictive accuracy for PFS with a C‐index of 0.761 (95% CI, 0.668–0.853), markedly outperforming the CGA (C‐index: 0.635; 95% CI, 0.535–0.735). Similarly, the new model demonstrated superior prognostic discrimination for OS, yielding a C‐index of 0.770 (95% CI, 0.674–0.867) compared with 0.671 (95% CI, 0.581–0.762) for the conventional CGA (Figure 4A). Longitudinal time‐dependent ROC analysis further confirmed that the AUC trajectories for the Geriatric Risk Score remained consistently above those of the CGA for both PFS and OS over a 5‐year follow‐up period, demonstrating robust longitudinal predictive performance (Figure 4B,C; DeLong's test at 1 and 5 years: p = 0.098 and p < 0.001 for PFS; p = 0.597 and p < 0.001 for OS, respectively).
FIGURE 4.
External validation of the Geriatric Risk Score. (A) Summary table of Harrell's C‐indices (with 95% confidence intervals) assessing the predictive accuracy of the Geriatric Risk Score in an independent validation cohort. The model is benchmarked against the CGA in the full cohort and against the IPS‐7 and IPS‐3 in the advanced‐stage (Ann Arbor III–IV) subset for both overall survival (OS) and progression‐free survival (PFS). (B, C) Time‐dependent AUC plots demonstrating the sustained longitudinal predictive superiority of the Geriatric Risk Score compared with the CGA for (B) PFS and (C) OS within the validation cohort. (D, E) Time‐dependent AUC plots comparing the Geriatric Risk Score with the IPS‐3 for (D) PFS and (E) OS within the advanced‐stage validation subset. (F, G) Time‐dependent AUC plots comparing the Geriatric Risk Score with the IPS‐7 for (F) PFS and (G) OS within the advanced‐stage validation subset.
In the advanced‐stage (Ann Arbor III–IV) subgroup, the Geriatric Risk Score retained its strong clinical utility for PFS prediction, achieving a C‐index of 0.711 (95% CI, 0.600–0.823) (Figure 4A) and outperforming both IPS‐7 (0.503; 95% CI, 0.334–0.672) (Figure 4D) and IPS‐3 (0.561; 95% CI, 0.418–0.703) (Figure 4F). A consistent advantage was also observed for OS prediction, with the Geriatric Risk Score yielding a C‐index of 0.701 (95% CI, 0.586–0.817) (Figure 4A), whereas traditional clinical indices largely lost their predictive value (IPS‐7: 0.549 [95% CI, 0.408–0.689] (Figure 4E); IPS‐3: 0.533 [95% CI, 0.400–0.667]) (Figure 4G). Longitudinal AUC curves in the validation cohort tracked higher for the new model than for the traditional IPS and CGA frameworks (DeLong's test at 1 and 5 years: vs. IPS‐7, p = 0.112 and p = 0.402 for PFS, p = 0.021 and p = 0.428 for OS; vs. IPS‐3, p = 0.161 and p = 0.700 for PFS, p = 0.196 and p = 0.026 for OS).
3.6. Genomic Landscape Stratified by the Geriatric Risk Score
To explore the underlying biological heterogeneity across different risk strata, we characterized the genomic mutational landscape of the study cohort (Figure 5A). On oncoplot analysis, we revealed a highly enriched profile of somatic alterations within this older demographic, with the most frequently mutated genes being SOCS1 (25%), TET2 (22%), DNMT3A (21%), and KMT2D (20%). Other recurrently altered genes included PCLO (13%), IGLL5 (12%), TNFAIP3 (9%), and TP53 (9%). Regarding the types of variants, the vast majority of alterations were somatic mutations, whereas low‐frequency CNVs were captured in only small subset of genes, such as CDKN2A and SMARCA4.
FIGURE 5.
Genomic landscape and mutational burden stratified by the Geriatric Risk Score. (A) An oncoplot detailing the landscape of somatic alterations (mutations, structural variants [SV], and copy number variations [CNV]) across the cohort. Individual patient samples (columns) are sorted in ascending order by their calculated Geriatric Risk Score and assigned Risk Group, while genes (rows) are ranked by overall alteration frequency. (B) A stacked bar chart illustrating the asymmetric distribution of genomic burden. The graph displays the absolute number of patients harboring specific gene alterations, proportionally broken down by the newly defined Geriatric Risk Score categories (Low Risk, Intermediate Risk, and High Risk), highlighting the concentration of mutational events in higher‐risk subgroups.
We next cross‐analyzed the frequencies of these key genomic alterations across the risk groups defined by the Geriatric Risk Score. The stacked bar chart demonstrated an asymmetric distribution of the mutational burden across the distinct risk strata. Specifically, patients in the High‐Risk and Intermediate‐Risk groups, who exhibited inferior clinical outcomes, also harbored a significantly denser and complex mutational load. For instance, the absolute majority of TET2, KMT2D, and DNMT3A mutational events were densely clustered within the high‐ and intermediate‐risk populations (Figure 5B). In contrast, the Low‐Risk group carried a substantially lower overall genomic burden, retaining only sporadic alterations in genes like SOCS1, DNMT3A, and IGLL5. This distinct genomic distribution strongly indicates that the adverse clinical phenotypes identified by the Geriatric Risk Score are closely coupled with more aggressive intrinsic molecular features, further validating the capacity of this composite clinical tool to reflect deep biological risk.
4. Discussion
In this study, we developed and validated a novel, multidimensional prognostic model specifically tailored for older patients with Hodgkin lymphoma. By integrating quantitative metrics of tumor burden (TLG) with key surrogates of host vulnerability (age, hemoglobin, and ADL status) and aggressive disease biology (LDCHL subtype), our Geriatric Risk Score effectively stratifies patients into three distinct risk groups with significantly divergent outcomes. Crucially, this model overcomes the intrinsic limitations of the traditional International Prognostic Score (IPS)—which lacks discriminatory power in this older demographic—and the conventional CGA, which suffers from a pronounced “ceiling effect” by overwhelmingly classifying the majority of these patients as ostensibly “fit”.
Our findings demonstrated that CGA—originally developed and validated for risk stratification in DLBCL rather than HL [18]—exhibit limitations in risk resolution for older patients with HL. The vast majority of our cohort was categorized as CGA‐“Fit,” yet applying the Geriatric Risk Score unveiled profound prognostic heterogeneity within this group. As demonstrated by our reclassification analysis, a striking proportion of CGA‐Fit patients harbored occult high‐risk clinico‐biological features and were subsequently upstaged by our model to Intermediate‐ or High‐Risk categories. This critical lack of granularity in CGA implies that relying solely on functional metrics (e.g., comorbidity indices) without disease‐specific integration fails to capture the disease‐specific aggressive traits that drive mortality in this subset. The integration of TLG within our final model—complementing biological and metabolic factors—supports the rationale for a disease‐specific geriatric assessment system tailored to the unique landscape of older HL.
The identification of advanced age (threshold > 73 years) and functional impairment (ADL dependence) as prominent independent predictors of mortality constitutes a cornerstone of our Geriatric Risk Score. This finding strongly aligns with real‐world evidence from North American and European cohorts, which have consistently identified the dual burden of advanced chronological age and functional decline as preeminent risk factors in geriatric oncology [27, 28]. Methodologically, during the initial phase of model construction, incorporating all candidate predictors as continuous variables led to the statistical phenomenon of complete separation. This mathematical instability was primarily driven by the interaction between extreme values of hemoglobin and TLG. However, a clear distinction must be drawn: while our composite score excels in survival prognostication, the active integration of a full CGA remains an indispensable counterpart to guide toxicity management, optimize treatment tolerability, and ensure safe therapeutic delivery [16].
Then, we did not benchmark our score against the A‐HIPI due to fundamental differences in clinical design: the A‐HIPI functions as a continuous probability calculator [13], whereas our model serves as a discrete, bedside‐ready risk stratification tool. Furthermore, recent evidence suggests that the A‐HIPI lacks calibration in patients aged > 65 years [29]. Given these structural disparities and the reported performance limitations of existing indices in older populations, a direct head‐to‐head comparison would lack clinical parity. Our model aims specifically to provide a pragmatic stratification approach tailored to the unique clinical profile of geriatric patients.
Despite the widespread integration of 18F‐FDG PET/CT in lymphoma management, robust data defining the prognostic utility of baseline metabolic parameters specifically in older HL remain limited. Our study bridges this gap by identifying TLG (> 200) as a powerful independent predictor of survival. Notably, in multivariable modeling, this quantitative metric of metabolic volume surpassed traditional anatomical parameters, underscoring it as a refined biological surrogate for disease aggressiveness. When benchmarking our metabolic thresholds against the limited existing literature, such as the pivotal analysis by Albano et al. [30], notable divergences emerge. Whereas Albano et al. proposed a substantially higher TLG cutoff, our rigorous outcome‐oriented approach minimized the AIC to prioritize meaningful survival differentiation, establishing a threshold of 200. Furthermore, our model directly integrates age and functional status, explicitly addressing host frailty as a core component rather than merely a statistical confounder. By analyzing metabolic burden within this context, we successfully unmasked the independent prognostic power of TLG for OS, which was obscured in previous analyses.
The profound prognostic refinement achieved by the Geriatric Risk Score was further validated against the established IPS‐3 and IPS‐7 indices in the advanced‐stage subset. In an era where the therapeutic landscape for HL is rapidly evolving with novel agents [8]—such as brentuximab vedotin and PD‐1 inhibitors—accurate risk stratification is paramount. The conventional IPS systems exhibited a near‐total loss of discriminative capacity in our older cohort. In contrast, our model demonstrated sustained predictive superiority over a 5‐year timeline. Sankey diagram analyses revealed significant bidirectional reclassification: our model safely downstaged a substantial subset of conventional IPS “High‐Risk” patients with more favorable multidimensional profiles, while simultaneously upstaging underestimated IPS “Low/Intermediate‐Risk” patients who harbored masked biological vulnerabilities. This granular risk resolution is critical. It suggests that by “purifying” broad clinical categories, the new model can isolate the specific subset of functionally robust, truly low‐risk patients who are most likely to tolerate uncompromised curative regimens, while accurately identifying high‐risk patients who necessitate highly tailored, novel‐agent‐based interventions.
To explore the biological underpinnings driving these adverse clinical phenotypes, we characterized the targeted genomic landscape across our risk strata. Our analysis revealed a highly enriched profile of somatic alterations, dominated by frequent mutations in epigenetic regulators and signaling pathways, including SOCS1, TET2, DNMT3A, and KMT2D [31, 32]. The dominance of mutations in TET2, DNMT3A, and KMT2D—genes frequently implicated in lymphoid malignancies and Clonal Hematopoiesis of Indeterminate Potential (CHIP) [33, 34]—supports the hypothesis that epigenetic dysregulation is a hallmark of aggressive geriatric HL biology. Crucially, cross‐referencing this mutational landscape with our clinical scoring system revealed a strikingly asymmetric genomic burden. The vast majority of mutational events—particularly alterations in TET2 and KMT2D—were densely concentrated within the Intermediate‐ and High‐Risk groups. In contrast, patients in the Low‐Risk category exhibited a substantially lower overall mutational load. This distinct genomic distribution strongly indicates that the adverse clinical phenotypes identified by the Geriatric Risk Score are closely coupled with a dense, intrinsically aggressive genomic architecture.
Several limitations of this study should be acknowledged. First, dichotomizing continuous variables to facilitate clinical utility reduced predictive granularity and statistical power. Second, employing a 41% SUVmax threshold for MTV segmentation to manage cross‐institutional heterogeneity may have systematically underestimated absolute MTV. Third, the limited sample size in the validation cohort and ultra‐elderly subgroup warrants further verification in larger, prospective international registries. Due to the single‐country design, assessment of health inequalities across sociodemographic groups was not feasible. Furthermore, center effects and cross‐center performance heterogeneity were not formally assessed.
Despite these limitations, the score offers high clinical utility. Unlike time‐intensive geriatric assessments, our 0–8 point system is calculated in under 2 min. Its primary value lies in refining therapy: it identifies fit older adults for full‐intensity regimens, preventing under‐treatment, while justifying the use of targeted agents (e.g., anti‐CD30 ADCs, PD‐1 inhibitors) in high‐risk patients to overcome resistance.
In conclusion, we defined a multidimensional Geriatric Risk Score that transcends the limitations of conventional functional assessments and traditional staging indices by seamlessly integrating metabolic tumor burden, host vulnerability, and disease biology. Crucially, our findings refine the risk stratification landscape for older adults with HL, providing a highly robust and transportable framework to guide personalized therapeutic strategies in the modern era.
Author Contributions
P.Z., Y.W., M.B., and C.H. contributed equally to this work. P.Z., Y.W., M.B., and C.H. designed and performed the research, analyzed the data, were involved in data acquisition, and wrote the manuscript with input and approval of the final version from all coauthors; H.Z., Q.Z., L.S., Y.X., and H.Z. critically reviewed the manuscript; J.W., Y.L., Y.J., G.W., F.Z., S.Z., W.G., S.Q., J.H., J.M., Y.Z., L.Z., L.L., W.Y., S.C., J.X., Y.X., and O.B. designed and performed the research, analyzed the data, were involved in data acquisition, and critically reviewed and approved the manuscript. All authors approved this manuscript for publication.
Funding
This work was supported by Tianjin Key Medical Discipline Construction Project (Grant No. TJYXZDXK‐3‐003A).
Ethics Statement
The study protocol was conducted in accordance with the Declaration of Helsinki and the International Conference on Harmonization Good Clinical Practice guidelines. The study was approved by the Ethics Committee of Tianjin Cancer Hospital (Approval No. bc20255575) and by the local institutional review boards of all participating centers and by the local institutional review boards of all participating centers.
Consent
The requirement for informed consent was waived by the Ethics Committee of Tianjin Cancer Hospital and by the local institutional review board due to the retrospective nature of the study and the use of anonymized patient data.
Conflicts of Interest
The authors declare no conflicts of interest.
Supporting information
Table S1: Cox regression of TLG and CGA with their interaction (Training cohort).
Table S2: Univariate and multivariable Cox proportional hazards analysis of prognostic factors for overall survival (OS), including TMTV.
Table S3: Univariate and multivariable Cox proportional hazards analysis of prognostic factors for progression‐free survival (PFS), including TMTV.
Table S4: Univariate and multivariable Cox proportional hazards analysis of prognostic factors for progression‐free survival (PFS), including TLG.
Figure S2: Correlation analysis between metabolic tumor burden metrics. Scatter plot illustrating the strong linear relationship between Total Metabolic Tumor Volume (TMTV) and Total Lesion Glycolysis (TLG) (r = 0.92). The high correlation confirms significant collinearity between these metabolic parameters, justifying their separate evaluation in independent multivariable models.
Figure S3: Prognostic discrimination of established indices in the study cohort. Kaplan–Meier survival curves demonstrating the performance of conventional prognostic tools: (A, B) Comprehensive Geriatric Assessment (CGA) for progression‐free survival (PFS) and overall survival (OS); (C, D) International Prognostic Score (IPS‐7) for PFS and OS; (E, F) International Prognostic Score (IPS‐3) for PFS and OS.
Figure S1: Feature selection and model development via LASSO regression analysis. (A) LASSO coefficient profiles of 21 candidate prognostic variables. Each curve represents a coefficient for a specific variable as a function of the tuning parameter (λ). (B) Determination of the optimal tuning parameter (λ) using 10‐fold cross‐validation. The vertical dashed line indicates the optimal λ values (λ min and λ 1se), corresponding to the selection of 8 nonzero coefficients used for subsequent multivariable Cox modeling.
Acknowledgments
We thank the patients and their families for their participation in this study, as well as the staff at the participating centers for their assistance with data collection. We also acknowledge the support from the Tianjin Key Medical Discipline Construction Project (Grant No. TJYXZDXK‐3‐003A).
Contributor Information
Hui Zhou, Email: zhouhui@hnca.org.cn.
Qingyuan Zhang, Email: zqyhmu1965@163.com.
Liping Su, Email: sulp2005@sohu.com.
Yuanlin Xu, Email: xuyuanlin620@zzu.edu.cn.
Huilai Zhang, Email: zhlwgq@126.com.
Data Availability Statement
The data are available from the corresponding author on reasonable request. The analysis code is available from the corresponding author on reasonable request.
References
- 1. Ansell S. M., “Hodgkin Lymphoma: 2025 Update on Diagnosis, Risk‐Stratification, and Management,” American Journal of Hematology 99, no. 12 (2024): 2367–2378, 10.1002/ajh.27470. [DOI] [PubMed] [Google Scholar]
- 2. Zhou L., Deng Y., Li N., et al., “Global, Regional, and National Burden of Hodgkin Lymphoma From 1990 to 2017: Estimates From the 2017 Global Burden of Disease Study,” Journal of Hematology & Oncology 12, no. 1 (2019): 107, 10.1186/s13045-019-0799-1. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3. Shenoy P., Maggioncalda A., Malik N., and Flowers C. R., “Incidence Patterns and Outcomes for Hodgkin Lymphoma Patients in the United States,” Advances in Hematology 2011 (2011): 1–11, 10.1155/2011/725219. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4. Shah H., Jang H., Kim S., and Halwani A. S., “A Comprehensive SEER Registry Analysis of Elderly Patients With Classical Hodgkin Lymphoma Based on Treatment Era and Race,” British Journal of Haematology 200, no. 5 (2023): 579–586, 10.1111/bjh.18564. [DOI] [PubMed] [Google Scholar]
- 5. Cheng P. T. M., Villa D., Gerrie A. S., et al., “The Outcome of Older Adults With Classic Hodgkin Lymphoma in British Columbia,” Blood Advances 6, no. 22 (2022): 5924–5932, 10.1182/bloodadvances.2022008258. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6. Engert A., Ballova V., Haverkamp H., et al., “Hodgkin's Lymphoma in Elderly Patients: A Comprehensive Retrospective Analysis From the German Hodgkin's Study Group,” Journal of Clinical Oncology 23, no. 22 (2005): 5052–5060, 10.1200/JCO.2005.11.080. [DOI] [PubMed] [Google Scholar]
- 7. Galli E., Cuccaro A., Maiolo E., et al., “Comorbidity Assessment to Determine Prognosis in Older Adult Patients With Classical Hodgkin Lymphoma,” Hematological Oncology 38, no. 2 (2020): 153–161, 10.1002/hon.2715. [DOI] [PubMed] [Google Scholar]
- 8. Barrett A. and Collins G. P., “Older Patients With Hodgkin Lymphoma: Walking the Tightrope of Efficacy and Toxicity,” Frontiers in Oncology 12 (2023): 1017787, 10.3389/fonc.2022.1017787. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9. Park C. H., Cho H., and Kim S. J., “Treatment of Older Patients With Hodgkin Lymphoma,” Blood Research 60, no. 1 (2025): 35, 10.1007/s44313-025-00084-4. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10. Moccia A. A., Donaldson J., Chhanabhai M., et al., “International Prognostic Score in Advanced‐Stage Hodgkin's Lymphoma: Altered Utility in the Modern Era,” Journal of Clinical Oncology 30, no. 27 (2012): 3383–3388, 10.1200/JCO.2011.41.0910. [DOI] [PubMed] [Google Scholar]
- 11. Hasenclever D., Diehl V., Armitage J. O., et al., “A Prognostic Score for Advanced Hodgkin's Disease,” New England Journal of Medicine 339, no. 21 (1998): 1506–1514, 10.1056/NEJM199811193392104. [DOI] [PubMed] [Google Scholar]
- 12. Diefenbach C. S., Li H., Hong F., et al., “Evaluation of the International Prognostic Score (IPS‐7) and a Simpler Prognostic Score (IPS‐3) for Advanced Hodgkin Lymphoma in the Modern Era,” British Journal of Haematology 171, no. 4 (2015): 530–538, 10.1111/bjh.13634. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13. Rodday A. M., Parsons S. K., Upshaw J. N., et al., “The Advanced‐Stage Hodgkin Lymphoma International Prognostic Index: Development and Validation of a Clinical Prediction Model From the HoLISTIC Consortium,” Journal of Clinical Oncology 41, no. 11 (2023): 2076–2086, 10.1200/JCO.22.02473. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14. Buccheri V., Moreira F. R., Biasoli I., et al., “External Validation and Calibration of the holistic Consortium's Advanced‐Stage Hodgkin Lymphoma International Prognostic Index (A‐hipi) in the Brazilian Hodgkin Lymphoma Registry,” British Journal of Haematology 206, no. 1 (2025): 144–151, 10.1111/bjh.19824. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15. Mohile S. G., Dale W., Somerfield M. R., et al., “Practical Assessment and Management of Vulnerabilities in Older Patients Receiving Chemotherapy: ASCO Guideline for Geriatric Oncology,” Journal of Clinical Oncology 36, no. 22 (2018): 2326–2347, 10.1200/JCO.2018.78.8687. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16. Mohile S. G., Mohamed M. R., Xu H., et al., “Evaluation of Geriatric Assessment and Management on the Toxic Effects of Cancer Treatment (GAP70+): A Cluster‐Randomised Study,” Lancet 398, no. 10314 (2021): 1894–1904, 10.1016/S0140-6736(21)01789-X. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17. Orellana‐Noia V. M., Isaac K., Malecek M. K., et al., “Multicenter Analysis of Geriatric Fitness and Real‐World Outcomes in Older Patients With Classical Hodgkin Lymphoma,” Blood Advances 5, no. 18 (2021): 3623–3632, 10.1182/bloodadvances.2021004645. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18. Tucci A., Martelli M., Rigacci L., et al., “Comprehensive Geriatric Assessment Is an Essential Tool to Support Treatment Decisions in Elderly Patients With Diffuse Large B‐Cell Lymphoma: A Prospective Multicenter Evaluation in 173 Patients by the Lymphoma Italian Foundation (FIL),” Leukemia & Lymphoma 56, no. 4 (2015): 921–926, 10.3109/10428194.2014.953142. [DOI] [PubMed] [Google Scholar]
- 19. Lia K., Jørgensen R. R. K., Wikman P., et al., “A Simplified Frailty Score Predicts Outcome in Curatively Treated Older Patients With Classical Hodgkin Lymphoma,” Haematologica 110, no. 10 (2025): 2356–2366, 10.3324/haematol.2025.287509. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20. Kumar A. J., Nelson J., Rodday A. M., et al., “Development and Validation of a Prediction Model for 1‐Year Mortality Among Older Adults With Hodgkin Lymphoma Who Receive Dose‐Intense Chemotherapy,” Journal of Geriatric Oncology 12, no. 8 (2021): 1233–1239, 10.1016/j.jgo.2021.07.003. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21. Al‐Ibraheem A., Mottaghy F. M., and Juweid M. E., “PET/CT in Hodgkin Lymphoma: An Update,” Seminars in Nuclear Medicine 53, no. 3 (2023): 303–319, 10.1053/j.semnuclmed.2022.10.006. [DOI] [PubMed] [Google Scholar]
- 22. Kanoun S., Rossi C., Berriolo‐Riedinger A., et al., “Baseline Metabolic Tumour Volume Is an Independent Prognostic Factor in Hodgkin Lymphoma,” European Journal of Nuclear Medicine and Molecular Imaging 41, no. 9 (2014): 1735–1743, 10.1007/s00259-014-2783-x. [DOI] [PubMed] [Google Scholar]
- 23. Herraez I., Bento L., Daumal J., et al., “Total Lesion Glycolysis Improves Tumor Burden Evaluation and Risk Assessment at Diagnosis in Hodgkin Lymphoma,” Journal of Clinical Medicine 10, no. 19 (2021): 4396, 10.3390/jcm10194396. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24. Scheepers E. R. M., Vondeling A. M., Thielen N., Van Der Griend R., Stauder R., and Hamaker M. E., “Geriatric Assessment in Older Patients With a Hematologic Malignancy: A Systematic Review,” Haematologica 105, no. 6 (2020): 1484–1493, 10.3324/haematol.2019.245803. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25. Merli F., Luminari S., Tucci A., et al., “Simplified Geriatric Assessment in Older Patients With Diffuse Large B‐Cell Lymphoma: The Prospective Elderly Project of the Fondazione Italiana Linfomi,” Journal of Clinical Oncology 39, no. 11 (2021): 1214–1222, 10.1200/JCO.20.02465. [DOI] [PubMed] [Google Scholar]
- 26. Sullivan L. M., Massaro J. M., and D'Agostino R. B., “Presentation of Multivariate Data for Clinical Use: The Framingham Study Risk Score Functions,” Statistics in Medicine 23, no. 10 (2004): 1631–1660, 10.1002/sim.1742. [DOI] [PubMed] [Google Scholar]
- 27. Wahlin B. E., Övergaard N., Peterson S., et al., “Real‐World Data on Treatment Concepts in Classical Hodgkin Lymphoma in Sweden 2000–2014, Focusing on Patients Aged > 60 Years,” eJHaem 2, no. 3 (2021): 400–412, 10.1002/jha2.202. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28. Evens A. M., Helenowski I., Ramsdale E., et al., “A Retrospective Multicenter Analysis of Elderly Hodgkin Lymphoma: Outcomes and Prognostic Factors in the Modern Era,” Blood 119, no. 3 (2012): 692–692, 10.1182/blood-2011-09-378414. [DOI] [PubMed] [Google Scholar]
- 29. Koca O., Ozyurt B., Umar A., et al., “The Validation of Advanced‐Stage Hodgkin Lymphoma International Prognostic Index (A‐HIPI) in Turkish Patients With Classical Hodgkin Lymphoma,” Annals of Hematology 104 (2025): 1765–1775, 10.1007/s00277-025-06292-3. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30. Albano D., Alessandro R., Rosa D., Alessandra T., and Francesco B.. “The Predictive and Prognostic Role of Baseline 2‐[18F]FDG PET/CT Volumetric and Dissemination Features in Classical Hodgkin Lymphoma,” European Journal of Nuclear Medicine and Molecular Imaging 53, no. 9 (2026): 5243–5250, 10.1007/s00259-026-07862-x. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31. Brune M. M., Juskevicius D., Haslbauer J., et al., “Genomic Landscape of Hodgkin Lymphoma,” Cancers 13, no. 4 (2021): 682, 10.3390/cancers13040682. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32. Harrop S., Yannakou C. K., Van Der Weyden C., et al., “Epigenetic Modifications in Lymphoma and Their Role in the Classification of Lymphomas,” Hemato 3, no. 1 (2022): 174–187, 10.3390/hemato3010015. [DOI] [Google Scholar]
- 33. Fend F., van den Brand M., Groenen P. J., et al., “Diagnostic and Prognostic Molecular Pathology of Lymphoid Malignancies,” Virchows Archiv 484, no. 2 (2024): 195–214, 10.1007/s00428-023-03644-0. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 34. von Beck K., von Beck T., Ferrell P. B., et al., “Lymphoid Clonal Hematopoiesis: Implications for Malignancy, Immunity, and Treatment,” Blood Cancer Journal 13, no. 1 (2023): 5, 10.1038/s41408-022-00773-8. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Table S1: Cox regression of TLG and CGA with their interaction (Training cohort).
Table S2: Univariate and multivariable Cox proportional hazards analysis of prognostic factors for overall survival (OS), including TMTV.
Table S3: Univariate and multivariable Cox proportional hazards analysis of prognostic factors for progression‐free survival (PFS), including TMTV.
Table S4: Univariate and multivariable Cox proportional hazards analysis of prognostic factors for progression‐free survival (PFS), including TLG.
Figure S2: Correlation analysis between metabolic tumor burden metrics. Scatter plot illustrating the strong linear relationship between Total Metabolic Tumor Volume (TMTV) and Total Lesion Glycolysis (TLG) (r = 0.92). The high correlation confirms significant collinearity between these metabolic parameters, justifying their separate evaluation in independent multivariable models.
Figure S3: Prognostic discrimination of established indices in the study cohort. Kaplan–Meier survival curves demonstrating the performance of conventional prognostic tools: (A, B) Comprehensive Geriatric Assessment (CGA) for progression‐free survival (PFS) and overall survival (OS); (C, D) International Prognostic Score (IPS‐7) for PFS and OS; (E, F) International Prognostic Score (IPS‐3) for PFS and OS.
Figure S1: Feature selection and model development via LASSO regression analysis. (A) LASSO coefficient profiles of 21 candidate prognostic variables. Each curve represents a coefficient for a specific variable as a function of the tuning parameter (λ). (B) Determination of the optimal tuning parameter (λ) using 10‐fold cross‐validation. The vertical dashed line indicates the optimal λ values (λ min and λ 1se), corresponding to the selection of 8 nonzero coefficients used for subsequent multivariable Cox modeling.
Data Availability Statement
The data are available from the corresponding author on reasonable request. The analysis code is available from the corresponding author on reasonable request.