Global outcomes and prognosis for relapsed/refractory mature T-cell and NK-cell lymphomas: results from the PETAL consortium

For common subtypes of mature T-cell lymphoma (MTCL) and mature natural killer–cell lymphomas (MNKCL), the identification of risk factors and subgroups within the relapsed and refractory (R/R) population that lead to heterogeneous outcomes to therapies remains unclear.^1-3 Histological subtype-specific prognostic scores for newly diagnosed patients with extranodal natural killer/T-cell lymphoma (ENKTCL), angioimmunoblastic T-cell lymphoma (AITL), and adult T-cell leukemia/lymphoma (ATLL) exist but not for the R/R group. Although anthracycline-based regimens are commonly used in the frontline setting for the common nodal subtypes of MTCL, the optimal strategy for the R/R patients remains ill defined. Historically, for this group, when clinical trial enrollment is not feasible, treatment with a platinum-, ifosfamide-, and gemcitabine-based cytotoxic chemotherapy (CC) is selected for the “fit” patient. In contrast, single agents (SAs) including histone deacetylase inhibitors (HDACis), antifolates (AFs) such as pralatrexate, and antibody-drug conjugates (ADCs) such as brentuximab vedotin (BV) are favored for the “unfit” patient. There is a paucity of randomized clinical data around the use of SA, and their comparative efficacy with CC outside of a clinical trial has not been well defined.^4-6 We first identified several independent predictors of overall survival (OS) for patients with R/R MTCL and MNKCL, a highly selected poor-prognosis group to begin with. Next, these pretreatment clinical and laboratory characteristics were integrated into a multivariable model for our cohort to estimate outcomes. The performance of the final multivariable model was compared with other risk scores with respect to their ability to predict, discriminate, and calibrate outcomes based on the measure of concordance index (C-index). We validated the prognostic model’s survival predictive power in test and independent study cohorts and developed a web-based calculator available for use by health care professionals. Next, we sought to contrast OS, progression-free survival (PFS), and the ability to bridge to hematopoietic stem cell transplantation (HSCT) with either SA or CC in the second-line treatment for patients with R/R MTCL and MNKCL. Finally, among patients with R/R MTCL and MNKCL, we juxtaposed the outcomes of various SA to gain more knowledge on distinct populations benefiting from specific classes of drugs.

A total of 1240 patients with R/R MTCL and MNKCL from 13 institutions participating in the Global Peripheral T-cell lymphoma (PETAL) consortium based in 10 countries (Australia, Brazil, Chile, United States, Japan, South Korea, South Africa, Saudi Arabia, Italy, and India) with lymphoma diagnosis between 1 January 2010 and 30 September 2021 were screened for eligibility. A total of 312 patients were excluded due to not receiving second-line therapy. This includes patients from Chile, who were screened but could not be included in the final analysis because they did not have details on second-line treatment. The number of patients enrolled at each site across various countries is summarized in supplemental Table 1. Further details of the methods section are included in the supplemental Data.

Of the 1240 patients screened, 925 were eligible for this study and subsequently included in the final analysis (Figure 1). Table 1 summarizes the baseline demographic and clinical characteristics of these 925 study patients with R/R MTCL and MNKCL receiving SA or CC for their second-line treatment. Most patients in the cohort were diagnosed and treated in the United States, Brazil, South Korea, and India. The median age at diagnosis was 58 years, with patients in Japan (median age, 69 years) being older. Most patients were males (578/925 [62%]), with higher proportions in Saudi Arabia (10/13 [77%]), India (56/75 [75%]), and South Africa (14/19 [74%]). Of patients whose race was known, 54% were White (453/843), 6% were Black (53/843), and 32% were Asian (273/843). Expectedly, the overall dominant lymphoma subtypes included peripheral T-cell lymphoma–not otherwise specified (PTCL-NOS; 385/925 [42%]), AITL (213/925 [23%]), anaplastic lymphoma kinase–negative anaplastic large cell lymphoma (ALK^– ALCL; 88/925 [10%]), ENKTCL (88/925 [10%]), anaplastic lymphoma kinase–positive anaplastic large cell lymphoma (ALK⁺ ALCL; 38/925 [4%]), ATLL (38/925 [4%]), enteropathy-associated T-cell lymphoma (21/925 [2%]), hepatosplenic TCL (21/925 [2%]), T-cell prolymphocytic leukemia (19/925 [2%]), and other PTCL subtypes (14/925 [2%]). Exceptions included ALCL as the dominant subtype in Saudi Arabia (7/13 [53%]) and ENKTCL and ATLL as the second most common subtype in South Korea (40/123 [33%]) and Brazil (27/138 [20%]), respectively. Most patients had Ann Arbor stage III to IV disease (723/925 [81%]), intermediate-risk (IR) to high-risk (HR) international prognostic index (IPI) from 3 to 5 (307/585 [53%]), and prognostic index for T-cell lymphoma (PIT) scores from 2 to 4 (331/636 [52%]). The median follow-up was 1.92 years (interquartile range, 0.94-3.94). At the time of data collection, of the 925 patients, 303 (33%) were alive, 515 (56%) had died, and 107 (12%) were lost to follow-up.

Table 2 summarizes the treatment characteristics of the 925 cohort patients with R/R MTCL and MNKCL receiving SA or CC for their second-line treatment. Of the 925 patients, 452 (49%) were refractory to frontline therapy with and without HSCT consolidation, and 468 (51%) had relapsed, with a higher proportion of refractory cases in Saudi Arabia (10/13 [83%]). Anthracycline-based initial chemotherapy, including CHOP (cyclophosphamide + hydroxydaunorubicin + vincristine + prednisone), CHOEP (cyclophosphamide + hydroxydaunorubicin + vincristine + etoposide + prednisone), or EPOCH (infusional etoposide phosphate + prednisone + vincristine + cyclophosphamide + hydroxydaunorubicin), remained the dominant upfront regimen globally, accounting for 69% (637/925) of all the first-line treatments. A higher number of other regimens were administered in Japan (30/38 [79%]) and India (47/75 [63%]). A table outlining the distribution of these “other” therapies is included in supplemental Table 5. Only 21% of patients (186/875) underwent autologous-HSCT (auto-HSCT) consolidation in first remission, with higher rates in Saudi Arabia (5/10 [50%]), United States (122/409 [30%]), and South Korea (32/122 [26%]) and lower rates in India (1/66 [2%]) and Japan (0/38 [0%]). Approximately 14% (121/855) of patients received radiation as part of first-line treatment. All survival estimates from the start of second-line treatment is based on 763 patients for whom the start dates of second-line treatments were available.

We compared median OS from the time of second-line treatment for patients with relapsed vs primary refractory disease, stratified by lymphoma subtype. The Kaplan-Meier estimate of OS was longer for relapsed patients than refractory patients in PTCL-NOS (1.97 years [95% confidence interval (CI), 1.27-2.69] vs 0.89 years [95% CI, 0.66-1.20]; P = .01) and ALK^– ALCL (4.22 years [95% CI, 1.52 to infinite (inf)] vs 1.36 years [95% CI, 0.38-2.84]; P = .01). The difference in OS from second-line treatment was not statistically significant in AITL (2.84 years [95% CI, 1.77-3.40] vs 2.25 years [95% CI, 1.08-5.69]; P = .98), ENKTCL (1.30 years [95% CI, 0.84 to inf] vs 1.17 years [95% CI, 0.49-7.10]; P = .35), or ALK⁺ ALCL (3.40 years [95% CI, 1.94 to inf] vs inf years [95% CI, 0.09 to inf]; P = .48; Figure 2). In addition, the OS from second-line treatment was not statistically significant (P = .065) between patients with different histological subtypes with ALK^– ALCL (2.49 years; 95% CI, 1.05-4.22), ALK⁺ ALCL (inf years; 95% CI, 1.54 to inf), PTCL-NOS (1.21 years; 95% CI, 1.04-1.65), ENKTCL (1.23 years; 95% CI, 0.77 to inf), and AITL (2.58 years; 95% CI, 1.89-3.40; supplemental Figure 1).

Although multiple prognostic factors and models have been proposed to risk stratify patients with MTCL and MKNCL, they were mostly developed for newly diagnosed patients. Thus, a new prognostic model is needed to further refine subgroups with varied outcomes within the R/R population in light of the above distinct outcomes and wide ranges of OS demonstrated above.

We thus split our global cohort of 763 patients into a training set (80%) and a test set (20%) and systematically evaluated 21 available features that were clinically relevant and easy to use in a global real-world data set. These included demographics, histological, laboratory, radiologic, and treatment characteristics (Figure 3A). We identified 11 nontreatment factors on univariable analysis to be associated with an inferior OS from the start of second-line treatment (P≤ 0.3). Based on the step-by-step selection in multivariable Cox regression and clinical considerations, our final model that combined 6 of the 11 features was chosen based on the highest training C-index among the models that incorporated covariates retaining an independent prognostic value (P ≤ .05) and based on Akaike information criterion. The final 6 features included were age >60 years, primary refractory, not AITL subtype, >1 extranodal site involvement, Ki67 ≥40%, and absolute lymphocyte count below the lower limit of normal, and each was associated with a P value of ≤.05 in the final multivariable Cox model (Figure 3B). Based on heterogeneous scoring of the Ki67 proliferation index (265 patients had Ki67 ≥40% and 94 with <40%), the cutoff of 40% was chosen, which gave us the optimal C-index in combination with other factors on the training and test cohorts. Because the relative risks associated with each of these 6 factors are similar, we assigned a score of 1 to each unfavorable feature. We then developed a novel prognostic model for 248 patients with R/R MTCL and MNKCL to estimate their OS from the start of second-line treatment and named it the prognostic index for R/R TCL (PIRT) score (Table 3). There was a clear trend of declining OS with an increasing score (Figure 4A-B). Patients were initially categorized into 6 subgroups because they had separate outcomes based on their scores (supplemental Figure 2A). For practical considerations, including ease of use as a tool to aid clinicians in the real world, we ultimately consolidated the 6 groups into 3 final risk groups: low risk (LR; 0-1), IR (2-3), and HR (4-6), with marked similarity between their predicted and actual survival (supplemental Figure 2B). On the training data set, the median 3-year OS was 75.2% for the LR group (95% CI, 29.7-85.2) vs 50.6% for the IR group (95% CI, 37.7-58.9) vs 16% for the HR group (95% CI, 7.7-26.9; Table 3). The PIRT score was associated with a lower Akaike information criterion and higher C-index when its performance was compared with the IPI and PIT scores. We were also able to segregate the test data set into 3 risk groups with disparate 3-year OS: LR (57.1%; 95% CI, 17.1-83.7) vs IR (23.3%; 95% CI, 8.7-41.9) vs HR (7%; 95% CI, 0.4-26.9; Figure 4C), similar to the training cohort. Based on the evaluation of 1000 bootstrapped test sets, our PIRT score demonstrated an improved prediction probability with a higher average C-index (0.7; 95% CI, 0.63-0.76) than IPI (0.56; 95% CI, 0.5-0.63) and PIT (0.59; 95% CI, 0.52-0.65) on the test data set (Table 3). Based on paired t tests between IPI vs PIRT and PIT vs PIRT scores, we confirmed that this distinction of performance was statistically significant.

We pooled eligible patients (not previously included in the global training and test data set) from the T-Cell Project (TCP) 2.0 registry and from 1 of the US institutions (Massachusetts General Hospital) to form the independent validation cohort. After harmonizing the data and applying the same inclusion criteria for the training and test populations, there were 102 and 45 eligible patients in the TCP 2.0 and MGH cohorts, respectively. The clinical characteristics of the combined validation cohort are summarized in the supplemental Table 6. In the combined cohort, after applying the PIRT score, patients were classified as LR (n = 7), IR (n = 60), and HR groups (n = 80). The PIRT was prognostic for 3-years OS, with a median of 85.7% (95% CI, 33.4-97.8), 29.9% (95% CI, 12.1-50.2), and 26.2% (95% CI, 15.9-37.7) for LR, IR, and HR groups, respectively (Figure 4D). To emphasize, our PIRT score demonstrated an improved prediction probability with a higher average C-index (0.62; 95% CI, 0.57-0.67) than IPI (0.51; 95% CI, 0.46-0.57) and PIT (0.55; 95% CI, 0.5-0.61) on the validation data set (P < .01). It is worth noting that the prognostic order of the risk groups defined by the score was maintained in the external validation cohort. We have developed a web-based survival risk calculator that uses the PIRT prognostic model to aid clinicians in predicting survival estimates, which will be free for use for any audience (https://www.petalconsortium.org/pirt-calculator).

Having observed heterogenous outcomes between R/R patients, we wanted to interrogate the impact of different therapies. In our cohort, R/R patients received either SA (323/925 [35%]) or CC (559/925 [60%]) for their second-line treatment, with higher proportions receiving SA in the United States (219/410 [53%]) and lower proportions in South Korea (18/123 [15%]), Saudi Arabia (1/13 [8%]), and Brazil (9/138 [7%]). Globally, 258 of 745 patients (35%) achieved a second complete response upon second-line treatment, with higher rates in South Africa (8/14 [57%]). A total of 159 of 811 patients (20%) proceeded to HSCT after second-line treatment; of these 159 patients, 78 (49%) received consolidative auto-HSCT, and 81 (51%) received allogeneic HSCT (allo-HSCT). Consolidative HSCT after second-line treatment was more common in Australia (13/34 [38%]) and South Africa (6/18 [33%]) and less common in Japan (3/38 [8%]). With the exception of the United States (27/91 [30%]), the percentage of patients bridged to auto-HSCT was higher or comparable with allo-HSCT even after second-line treatment, highlighting the differences in clinical practice globally. The distribution of patients bridged to auto-HSCT or allo-HSCT by SA or CC across various lymphoma subtypes is summarized in supplemental Table 7. Although an overall greater number of patients underwent HSCT after CC than after SA treatment, the proportion of patients undergoing allo-HSCT was comparable between the 2 treatment groups, likely secondary to heterogeneity in treatment paradigms globally.

We next sought to compare the OS of patients receiving SA vs CC across subtypes from the start of second-line treatment. The number of patients included in this analysis was 763, for whom the dates of start of second-line treatment were available. For PTCL-NOS (1.42 years [95% CI, 0.92-2.14] vs 1.14 years [95% CI, 0.91-1.59]; P = .63) and ALK⁺ ALCL (inf years [95% CI, 0.12 to inf] vs 1.94 years [95% CI, 0.09 to inf]; P = .41), OS from the time of second-line treatment was comparable between those treated with SA and CC. In AITL (3.40 years [95% CI, 2.58 to inf] vs 1.71 years [95% CI, 0.96-2.49]; P = .004) and ALK^– ALCL (3.08 years [95% CI, 1.05 to inf] vs 1.36 years [95% CI, 0.30-3.71]; P = .06), the OS for patients treated with SA was superior compared with those treated with CC (Figure 5). However, for ENKTCL, CC-treated patients (7.10 years [95% CI, 0.77 to inf] vs 0.84 years [95% CI, 0.43-1.73]; P = .04) experienced longer OS than those treated with SA. The 3-year and 5-year OS for patients treated with SA in comparison with CC confirmed these results as well and are exhibited in supplemental Table 8A-C.

We evaluated the PFS for patients who achieved response (complete response + partial response + stable disease) to SA and CC by calculating the time from the start of SA or CC treatment to the start of third-line treatment, death, or loss of follow-up (Figure 6A). We observed a statistically significant PFS advantage achieved with the use of SA over CC regardless of transplantation status after second-line therapy. This was observed even after adjusting for important prognostic factors such as PIT score, primary refractory status, and lymphoma subtype (average hazard ratio (aHR), 0.69; 95% CI, 0.51-0.95; P = .02). This was also true when transplant was considered as an event but not censored (aHR, 0.64; 95% CI, 0.49-0.85; P < .005; Figure 6B) and when transplant was used as a censoring event (aHR, 0.78; 95% CI, 0.55-1.11; P = .17; Figure 6C). When we investigated OS for patients who were administered SA vs CC but did not undergo auto-HSCT or allo-HSCT, the HR was 0.62 (95% CI, 0.42-0.91; P = .01), whereas the HR for those who proceeded to transplant receiving SA vs CC was 0.55 (95% CI, 0.29-1.03; P = .06; Figure 6D).

We then stratified analysis by specific SA subclasses based on the mechanism of action of the drugs including epigenetic modifiers (EMs), ADCs, AFs, and small-molecule signaling inhibitors (SMIs). For PTCL-NOS, there was no difference in OS from the time of second-line treatment across patients treated with CC (1.14 years; 95% CI, 0.91-1.59), EM (1.09 years; 95% CI, 0.85 to inf), ADC (2.69 years; 95% CI, 1.37 to inf), AF (0.65 years; 95% CI, 0.23-1.46), and SMI (0.87 years; 95% CI, 0.15-2.04; Figure 7A). This result was consistent when EM group only included HDACi (1.12 years; 95% CI, 0.82 to inf; supplemental Figure 3A). Patients with AITL treated with SMI had improved OS (95% CI, 8.17 to inf) compared with those treated with CC (95% CI, 0.96-2.49; P < .0005) and EM (95% CI, 1.04-3.46; P = .001) and comparable OS with those treated with ADC (95% CI, 0.01 to inf; P = .11; Figure 7B). The SMI treatments consisted mostly of alisertib, duvelisib, cerdulatinib, and cyclosporine. The distribution of SMI and number of patients specifically with AITL receiving it included anti-cyclin-dependent kinase 9 (CDK9) inhibitor (n = 1), alisertib (n = 6), duvelisib alone and its combination with romidepsin or bortezomib (n = 7), cerdulatinib (n = 4), cyclosporine (n = 4), ruxolitinib (n = 1), and tipifarnib (n = 2). In patients with AITL, patients treated with CC or ADC did not have significant difference in OS compared with the EM group, which only included HDACi (2.11 years; 95% CI, 0.88-2.84; P = .7; supplemental Figure 3B). For ENKTCL, there was no difference in OS between those treated with CC and those treated with monoclonal antibody (7.1 years [95% CI, 0.77 to inf] vs 1.07 years [95% CI, 0.33 to inf]; P = .1). For ALK⁺ ALCL, there was no difference in OS between those treated with CC and those treated with ADC (1.94 years [95% CI, 0.09 to inf] vs inf years [95% CI, 2.37 to inf]; P = .48). Importantly, for ALK^– ALCL, the OS was inferior in those treated with CC compared with ADC (1.36 years [95% CI, 0.30-3.71] vs 4.22 years [95% CI, 1.05 to inf]; P = .03; Figure 7C-E).

We also compared the median time to the next therapy for SA vs CC in the second-line setting. For the combined cohort, the time to next therapy was comparable between the 2 groups (HR, 0.81; 95% CI, 0.60-1.09; P = .16; HR with adjustment for lymphoma subtype, PIT score, and primary refractory disease, 0.83; 95% CI, 0.57-1.21; P = .34). Supplemental Figure 4 illustrates the number of patients receiving CC vs SA over time across countries. We also evaluated the OS of SA vs CC across the 3 risk groups defined by PIRT score in the combined training and test population, and once again, SA was at least comparable in all risk groups and even superior in some relative to CC (supplemental Figure 5). A summary of start and stop time points and of events for various survival analyses for the global cohort is depicted in supplemental Table 9. A summary table that highlights differences in baseline, demographic, clinical, and treatment characteristics of the SA and CC cohorts is included as supplemental Table 10. Patients in the SA cohort were older. There was a higher incidence of the AITL histological subtype, fewer patients had >1 extranodal site involvement, fewer patients had lactate dehydrogenase above the upper limit of normal at baseline, fewer patients had Ki67 ≥40%, and more patients had relapsed disease than the CC cohort, and hence, it is feasible that these factors could have contributed to their better outcomes. To assess the impact of location of diagnosis and treatment on OS, we performed additional univariable Cox analysis. When comparing being diagnosed and treated in the United States vs not, P value was .1, training C-index was 0.52, and testing C-index was 0.54, suggesting that the country of diagnosis and treatment is unlikely to affect OS (supplemental Table 11).

To the best of our knowledge, this retrospective study represents the largest cohort of patients with R/R aggressive MTCL and MNKCL and reflects outcomes for this specific subset of patients treated according to the standards of care in each country. In this study, we sought to analyze and compare the efficacy of the treatment patterns around the world between SA and CC in the second-line setting and define prognostic factors influencing the survival of patients with R/R lymphoma in a global setting. Previous reports comparing outcomes of patients with R/R lymphoma have largely focused on the combined cohort of all T-cell lymphoma subtypes, thus limiting their generalizability to individual lymphoma subtypes.^1,2 Our relatively large sample size enabled us to interrogate the common nodal lymphoma subtypes with statistical power and observe that outcomes for patients with refractory disease are not necessarily worse when estimating outcomes since the initiation of second-line treatment for subtypes such as AITL, ENKTCL, and ALK⁺ ALCL. One potential explanation could be access to contemporary drugs such as EMs, SMIs, immune checkpoint blockades, and ADCs, such as BV, that mitigate the effects of initial chemo-refractoriness to some degree.

Several research groups including us have reported on the comparable efficacy and ability to bridge R/R patients to consolidative HSCT with SA over CC. However, studies have been largely restricted to a single institution or multicenter studies within a country or metanalyzes.^4-6 Clinical observations that treatment with HDACi appears to achieve higher responses in patients with T-follicular helper (TFH) phenotype lymphoma (AITL and TFH-PTCL) than those with PTCL-NOS have triggered tremendous interest and guidance on using specific strategies more frequently and earlier in the treatment course of certain PTCL subtypes.⁷ In our global data set with well-characterized second-line treatments, we observed that, within the combined cohort, the use of SA in the second-line setting was associated with statistically significant improvement in PFS regardless of transplantation status after the treatment. Unlike the PTCL-NOS subtype, in which no difference in OS was observed between CC and EM or SMI, we observed an OS advantage with the use of SMI over CC but did not see an OS advantage of EM over CC in AITL. This intriguing finding calls for attention to continued investigation of this class of drugs in tailored clinical trials enriched with patients with specific lymphoma subtypes to fully understand its efficacy and potentially a favorable prognosis of this subtype with modern drugs. Our results confirmed in a global data set and outside of a clinical trial scenario that ADC was associated with superior OS compared with CC in the second-line setting for patients with ALK^– ALCL. Thus, we have generated, to our knowledge, the largest global evidence of real-world data that suggest that patients who receive novel agents have comparable and, in certain scenarios, superior outcomes compared with patients who receive chemotherapy-predicated treatments. This calls for the attention of clinicians, regulatory agencies, and pharmaceutical companies to facilitate discussions around expanded access to active drugs worldwide and also might inform the next generation of trial designs.

Although multiple prognostic indices such as IPI, PIT, modified version of PIT (m-PIT), International Peripheral T-Cell Lymphoma Project Score (IPTCLP), and Prognostic Index of Natural Killer Lymphoma (PINK) scores exist for patients with newly diagnosed MTCL and MNKCL, a systematic analysis of independent predictors of outcome in the R/R setting and a risk score for such patients is lacking.^8-12 This is inherently a challenging question because R/R patients are typically associated with a poor prognosis, making further stratification of this subset into defined risk groups with differing outcomes difficult. Through a comprehensive and structured approach, we analyzed the independent effects of all frequently available covariates in a real-world scenario on the outcomes for patients with R/R MTCL and MKNCL. We identified several known and new clinical features to be predictive of inferior survival. However, in the final multivariable model that was associated with the highest C-index in the training data set, age >60 years at diagnosis, primary refractory disease, histological subtype other than AITL, >1 extranodal site involvement, Ki67 ≥40%, and ALC below the lower limit of normal emerged as the most significant predictors of OS, forming the basis for our new score, the PIRT score, for patients with R/R MTCL and MNKCL. This score stratified patients into 3 distinct risk groups with differing outcomes and was validated in a second independent cohort, thereby conforming to the Transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD) guidelines. Of note, we primarily relied on results from the independent and test cohorts rather than the training cohort. To our knowledge, this is the first report of a prognostic score to estimate survival outcomes for patients with R/R MTCL and MNKCL that has been verified in training, test, and validation cohorts. We had access to a cohort of patients with MTCL and MNKCL with diverse demographic and clinical backgrounds and varying responses to heterogeneous treatment lines, which is a strength of our study and representative of scenarios outside of clinical trials. Our new prognostic model might be of use to medical professionals in real time across the globe, and hence, we developed a web-based calculator and have made it available for public use to qualified personnel with the obvious disclaimers that this is not approved for clinical use by National Comprehensive Cancer Network (NCCN) or the Food and Drug Administration (FDA) and needs to be verified in a future prospective study.

Limitations of the study include systematic error based on the retrospective nature of these analyses, specifically with varied documentation over the span of 10 years. Varied treatment practices in the second-line setting among different academic centers worldwide based on restricted access to SA might underestimate its effects. Although the high proportion of patients from the United States included in the analysis and their access to SA might bias the results of the global cohort, CC was the most commonly used second-line treatment overall, and hence, we believe that the observed benefit of SA over CC cannot be entirely attributed to geographic dominance of patients from the United States within the cohort. Although we organized various drugs as SA or CC based on their mechanism of action, certain drugs such as pralatrexate and BV were categorized under SA and not CC due to their mostly SA use in the R/R setting. Nevertheless, such a bias would equally affect all subgroups; additionally, only a small proportion of patients (n = 23) received pralatrexate in our cohort. Our study lacked central pathology and radiology review and involved data collection across multiple different sources internationally ranging from single institutions to registry cohorts. Another major limitation of our study is that we used variables from the time of diagnosis to generate the PIRT score, which was used to predict survival estimates for patients with R/R MTCL and MNKCL from the time of second-line treatment. It is feasible that variables such as ALC and the number of extranodal sites of involvement changes over time from the point of diagnosis to the start of second-line treatment, and hence, incorporating factors closer to the start of the second-line treatments could be more informative. However, re-evaluation of many of the factors in real-world practices is often not possible. Despite the potential change in some of these variables over their first-line treatment course, we observed a clear separation of survival estimates between the 3 defined risk groups as stratified by the PIRT score in all training, test, and validation cohorts. This underscores the robustness of the model and its generalizability despite the incorporation of limited information around the world. Although the PIRT score can predict survival for a combined cohort of patients with R/R lymphoma, more development, training, and fine tuning of this approach is required for subgroup analyses and individual-level predictions. Thus, personalized survival estimates to chart out patients’ trajectories to specific common interventions are the focus of our ongoing research efforts. PETAL consortium investigators are now focused on integrating clinical and molecular information and harnessing artificial intelligence for precision medicine on prospectively enrolled patients. Finally, we recognize that comparisons of treatments reported in our study may not have sufficiently accounted for potential confounders, and as such, the results are subject to potential bias. A future direction is to take an in-depth approach to carefully account for potential confounders to reduce bias. Another area of active research in our group involves the generation of novel causal inference approaches and its potential to provide a natural framework to make forecasts more accurate.

In conclusion, we characterized and benchmarked outcomes for the largest global cohort of patients with R/R MTCL and MNKCL in the era of contemporary therapies. We demonstrated that although CC remains the most frequently used treatment regimen for patients, SA is at least comparable and, in specific subtypes and scenarios, superior to CC, thereby warranting their earlier use in treatment paradigms. Our study also underscores an unmet need for expanded access to SA worldwide. A novel prognostic score such as PIRT highlights patients with poor prognosis, such as R/R MTCL and MNKCL, to have varied outcomes to conventional and recent therapies and might benefit clinicians when faced with accurately determining their prognosis.

This work was supported by Daiichi Sankyo, Acrotech Biopharma, and the Center for Lymphoma Research Funds. S.J. is supported by the National Cancer Institute K08 Career Development Award (K08CA230498) and MGH Lymphoma Research Funds. E.J. is supported by the Reid Family Fund for Lymphoma Research.

Contribution: S.J. designed the research; J.X.H., Min Jung Koh, L.B., M. Sorial, S.M.M., L.P., S. Singh, I.J.E.-U., J.G., M.J.F.T., C.T.M., A.G., A.D., K.C., A. Lenart, E.N., J.B., Min Ji Koh, E. Miranda, C.C., R.S., S.M.H., M. Merrill, E.J., M. Manni, M.C., T. Skrypets, A. Lymboussaki, M.F., Y.K., J.S.K., J.Y.C., T.E., T. Shet, E.S., A. Shetty, S. Saha, H.J., M. Sengar, C.V.D.W., H.M.P., R.H., T.M., F.F., M.G., B.C., P.L.Z, T.O., N.Y., S.E.Y., W.-S.K., G.P., Z.M., E.V., J.C.A., M.A., J.F., M.E.C., A.K., G.B., H.M., A. Sawas, K.M.K., M.I., F.B., O.A.O., C.S., E. Marchi, D.S., and S.J. performed the research; J.X.H., Min Jung Koh, L.B., M.S., C.S., E. Marchi, D.S., and S.J. analyzed data; C.C., S.M.H., E.J., Y.K., H.J., M.S., H.M.P., F.F., P.L.Z., W.-S.K., O.A.O., C.S., E. Marchi, and D.S. reviewed the manuscript; J.X.H., Min Jung Koh, L.B., D.S., and S.J. wrote the manuscript; and J.X.H., Min Jung Koh, and L.B. were assigned equal authorship based on their equal time and effort spent on this manuscript.

Conflict-of-interest disclosure: C.C. treports honoraria and consultancy fees from Roche, AbbVie, Janssen, AstraZeneca (AZ), Eli Lilly and Company (Lylli), and Takeda. S.M.H. reports consultancy fees from ONO Pharmaceuticals, Tubulis, Abcuro Inc, Cimieo Therapeutics, Auxilius Pharma, Secura Bio, Shoreline Biosciences, Inc, and Yingli Pharma Limited; research funding from Affimed, Celgene, ADC Therapeutics, Millenium, CRISPR Therapeutics, Seattle Genetics, and Verastem/Secura Bio; and consultancy fees and research funding from Daiichi Sankyo, Trillium Therapeutics, Kyowa Hakko Kirin, and Takeda. E.J. reports research funding from Celgene, Pharmacyclics, and Hoffmann-LaRoche; honoraria and research funding from Merck; honoraria from Daiichi, Bristol Myers Squibb (BMS), and Bayer; and patents and royalties with UpToDate. H.J. reports research funding from ImmunoACT, Zydus Pharmaceuticals, and Intas Pharmaceuticals. C.V.D.W. reports ending employment in the past 24 months and membership on an entity's board of directors or advisory committees of Cartherics Pty Ltd. H.M.P. reports speakers bureau fees from Takeda, Merck, Mallinkrodt, and Mundipharma. F.F. reports honoraria from Secura Bio, Daiichi Sankyo, Kyowa, Conjupro, and Astex; and speakers bureau fees from Seagen and Acrotech. B.C. reports membership on an entity's board of directors or advisory committees in Kite-Gilead, AbbVie, Janssen, Celgene-BMS, BeiGene, and Takeda; and speakers bureau fees from Kite-Gilead, AbbVie, Janssen, Novartis, Lilly, and Roche. P.L.Z. reports membership on an entity's board of directors or advisory committees in Servier, Celltrion, Sandoz, Gilead, Secura Bio, Janssen-Cilag, BMS, Novartis, Antibody-drug conjugate (ADC) Therapeutics, Incyte, AstraZeneca, Merck Sharp & Dohme (MSD), EUSA Pharma, Takeda, Roche, Kyowa Kirin, and BeiGene; speakers bureau fees from Servier, Celltrion, Gilead, Janssen-Cilag, BMS, Novartis, Incyte, AstraZeneca, MSD, Takeda, Roche, EUSA Pharma, Kyowa Kirin, and BeiGene; and consultancy fees from Novartis, MSD, and EUSA Pharma. W.-S.K. reports research funding from Sanofi, BeiGene, Boryong, Roche, Kyowa Kirin, and Donga. E.V. reports research funding from MSD. C.S. reports current employment with Biogen Digital Health. E. Marchi reports research funding from Merck, Celgene/BMS, Astex Pharmaceutical/Myeloid Pharmaceuticals, and Dren Bio; membership on an entity's board of directors or advisory committees in Dren Bio; and other interests including data safety monitoring committee fees from Everest Clinical Research. S.J. reports consultancy fees from Mersana Therapeutics, Myeloid Therapeutics, SIRPant Immunotherapeutics, and Abcuro Inc; membership on an entity's board of directors or advisory committees in Mersana Therapeutics, Myeloid Therapeutics, SIRPant Immunotherapeutics, Abcuro Inc, Secura Bio, Daiichi Sankyo, and CRISPR Therapeutics; and research funding from Daiichi Sankyo, SIRPant Immunotherapeutics, Acrotech Limited Liability Company (LLC), and Abcuro Inc. The remaining authors declare no competing financial interests.

Correspondence: Salvia Jain, Jon and Jo Ann Hagler Center for Lymphoma, Massachusetts General Hospital Cancer Center, 55 Fruit St, Boston, MA 02114-2696; email: salvia.jain@mgh.harvard.edu.