Revised MALT-IPI: A new predictive model that identifies high-risk patients with extranodal marginal zone lymphoma

Abstract

Extranodal marginal zone lymphoma (EMZL) is a heterogeneous disease with a subset of patients exhibiting a more aggressive course. We previously reported that EMZL with multiple mucosal sites (MMS) at diagnosis is characterized by shorter survival. To better recognize patients with different patterns of progression-free survival (PFS) we developed and validated a new prognostic index primarily based on patient’s disease characteristics. We derived the “Revised mucosa-associated lymphoid tissue International Prognostic Index” (Revised MALT-IPI) in a large data set (n=397) by identifying candidate variables that showed highest prognostic association with PFS. The revised MALT-IPI was validated in two independent cohorts, from the University of Iowa/Mayo Clinic (n=297) and from IELSG-19 study (n=400). A stepwise Cox regression analysis yielded a model including four independent predictors of shorter PFS. Revised MALT-IPI has scores ranging from 0 to 5, calculated as a sum of 1 point for each of the following- age >60 years, elevated LDH, and stage III-IV; and 2 points for MMS. In the training cohort, the Revised MALT-IPI defined 4 risk groups: low risk (score 0, reference group), low-medium risk (score 1, HR=1.85, P=0.008), medium-high risk (score 2, HR=3.84, P<.0001), and high risk (score 3+, HR=8.48, P<.0001). Performance of the Revised MALT-IPI was similar in external validation cohorts. Revised MALT-IPI is a new index centered on disease characteristics that provides robust risk-stratification identifying a group of patients characterized by earlier progression of disease. Revised MALT-IPI can allow a more disease-adjusted management of patients with EMZL in clinical trials and practice.

Graphical Abstract

INTRODUCTION

Extranodal marginal zone lymphoma (EMZL) of the mucosa-associated lymphoid tissue (MALT lymphoma) is an indolent disease accounting for 7% to 8% of all B-cell lymphomas. While usually associated with long survival,¹ EMZL is a heterogeneous entity with a subset of patients exhibiting a more aggressive disease course with relapses and shorter survival.² The disease is frequently localized at diagnosis allowing the utilization of local therapies (e.g. radiation), which provides a long-term disease control in many patients.^3,4 However, even among these patients there is a subgroup that relapses and may exhibit shorter survival.^5–7 Further, 23% to 34% of the patients present with advanced stage disease that is not curable and characterized by recurrent relapses.^8,9

Overall survival (OS) is the universally accepted primary endpoint for clinical trials in patients with malignant diseases.¹⁰ However, in indolent EMZL characterized by long survival, progression-free survival (PFS) or event-free survival (EFS), requiring shorter follow-up time to observe benefit, may represent more appropriate surrogate endpoints for clinical trials.^11–14 PFS is defined from the time of diagnosis, study entry, randomization, or start of treatment, until relapse or disease progression or death from any cause. Definition of EFS varies, but often is defined as the time until any treatment failure including discontinuation of treatment for any reason (e.g., toxicity, patient preference, initiation of new treatment without documented progression), relapse or disease progression, or death from any cause. PFS and EFS may better evaluate efficacy of novel therapies and have been commonly incorporated as primary endpoint in EMZL studies.^12,13,15

Patients at risk for inferior prognosis were difficult to identify until 2017 when Thieblemont et al. developed the mucosa-associated lymphoid tissue (MALT) International Prognostic Index (IPI).¹⁶ This prognostic score, which is the sum of indicators of age ≥70 years, stage III-IV and elevated lactate dehydrogenase (LDH), was developed using data from 401 patients prospectively enrolled in an international randomized clinical trial and validated in an independent set of 633 patients. MALT-IPI grouped patients into 3 risk groups with observed 5-year EFS of 70%, 56%, and 29% in low-risk (no risk factors), intermediate-risk (1 risk factor), and high-risk (≥2 risk factors) groups, respectively. However, a recent analysis of the Surveillance, Epidemiology, and End Results (SEER) database in patients with localized EMZL demonstrated age ≥60 years as an independent factor associated with shorter lymphoma-specific survival.³ This is also the age threshold used in other lymphomas prognostic indexes.^17,18 Furthermore, disease-attributable characteristics account for 2 of the 3 (67%) of the factors in the MALT-IPI index and patients older than 70 years may experience non-lymphoma related events. This opens an opportunity to design a revised prognostic index with greater clinical utility by incorporating additional disease-associated prognostic factors.

Shorter PFS, OS and higher risk for high-grade transformation (HGT) in a subset of patients with disseminated EMZL involving multiple areas of disease was previously reported.¹⁹ This entity was named multiple mucosal sites (MMS), defined as the presence of two or more different extranodal sites independent of bone marrow and spleen involvement. This definition includes EMZL located in two different organs (e.g., ocular adnexa and lung or stomach, pancreas and liver). Bilateral involvement of the same organ (e.g. eye, lung) or involvement of multiple areas of the skin is not considered MMS. This clinical variable was independent of stage, LDH and age in predicting outcome of patients with EMZL and could separate stage IV patients into 2 groups with statistically significant different outcomes (10-year PFS 45% vs. 5%, P< 0.001).¹⁹

To better recognize high-risk disease-attributable features and identify those patients at risk for earlier disease progression, we developed a new index called Revised MALT-IPI based on significant variables derived from a multivariable Cox model for PFS. Revised MALT-IPI was built by examining multiple commonly available clinical parameters that competed for inclusion in the model and were associated with PFS. The model ultimately consisted of stage, LDH, presence of MMS, and a younger age cutpoint of 60 years. The Revised MALT-IPI index was developed based on prediction of PFS using data from the University of Miami EMZL cohort of 397 patients and was externally validated in two independent cohorts, a North American and a European cohort, aiming to demonstrate its applicability in various geographic settings and with a diversity of therapies.

METHODS

Trainning cohort (N=397)

Initially, a total of 405 patients with newly diagnosed EMZL treated at the University of Miami (UM) Health Care System during the period 1995 to 2017 were retrospectively identified by a review of the Florida Cancer Registry database. After excluding 8 patients with HGT at diagnosis, our training cohort consisted of 397 total patients with newly diagnosed EMZL with follow-up at our center. The institutional review board of the UM approved this study, which followed the tenets of the Declaration of Helsinki. Patient’s characteristics on 405 total patients were previously described.¹⁹ In our present cohort of 397 patients, treatments included radiation therapy (RT) (52.9%), chemotherapy (27.5%), chemotherapy and RT (8.3%), surgery (5.5%) and no treatment (5.8%) as per the discretion of treating physicians; and the median follow-up of patients for progression, estimated using the reverse Kaplan-Meier method, was 6.9 years (95% CI: 5.9 to 7.4 years).

Statistical analysis

PFS was defined as the time from diagnosis to HGT after diagnosis, progression/relapse, or death, whichever occurred first. OS was defined as the time from diagnosis to death from any cause. Progression-free patients were censored at the date of last follow up. PFS and OS curves were estimated using the Kaplan-Meier method and compared using the log-rank test. Competing risk analysis was used to estimate lymphoma-specific survival and incidence of HGT with non-lymphoma related death as competing risk. Univariable and multivariable analyses using Cox proportional hazards regression were conducted to evaluate the effect of the prognostic variables on PFS and OS. MALT-IPI was calculated as reported.¹⁶

The new index, Revised MALT-IPI, was derived in the UM cohort as the training set. To build the new index, a multivariable Cox model for PFS (main study endpoint) was attained using stepwise selection among candidate variables with cutoffs P≤ 0.30 to enter and P≤ 0.05 to stay in the model. The candidate variables tested as potential predictors of PFS were the following: age >60, age ≥70, sex, anemia (hemoglobin <12 g/dL), stage III-IV, Eastern Cooperative Oncology Group (ECOG) performance status (PS) 2-4, elevated serum LDH, number of extranodal sites >1, number of nodal sites >4, and presence of MMS.¹⁹ For the derived multivariable Cox model and the corresponding univariable Revised MALT-IPI score model for PFS, we report hazard ratios (HRs) with 95% confidence intervals, estimated coefficients b=ln(HR), bootstrap estimates of the model coefficients and corresponding standard errors. The bootstrap estimates are based on 1000 bootstrap samples on the same size (n=397) from the training cohort using simple random sampling with replacement.

As additional internal model validation in the UM training set, we report the following statistics for models on PFS and OS: the goodness of fit AIC statistics, the integrated time-dependent area under the curve (IAUC) using Uno’s method, and the censored-adjusted concordance c-statistics by Uno (C-Uno) which provides an overall measure of model predictive accuracy with higher value indicating better discrimination.²⁰ In addition, for PFS by Revised MALT-IPI and MALT-IPI, time-dependent areas under the receiver operator characteristic (ROC) curves (AUCs), estimated using the inverse-probability of censoring weighted (IPCW) method, are provided. Positive predictive values of the two prognostics indices as functions of time, PPV(t) are also provided. In our context, the PPV(t) is the probablility of the occurrence of the event prior to time t in the high-risk group. PPV(t) were estimated using the Kaplan-Meier method and evaluated in the high-risk group, score 3+ for Revised MALT-IPI and 2+ for MALT-IPI.²⁰ The time-dependent ROC curves and the AUC functions characterize how well the fitted model can distinguish between subjects who experience an event and subjects who do not. AUC functions or time-dependent ROC curves summarize the predictive accuracy at specific times. Model callibration was assessed by calibration plots depicting the observed PFS rates obtained by Kaplan-Meier method versus the predicted rates obtained via Cox regression.^20–23 We also tested the performance of Revised MALT-IPI to identify patients at risk for EMZL relapse or progression of disease within 24 months (POD24). POD24 was assessed in the subset of treated patients with POD24 event or with follow-up of at least 24 months.²⁴

External validation of the Revised MALTI-IPI index was obtained using data from two independent cohorts: N=297 EMZL patients from the University of Iowa/Mayo Clinic Lymphoma Specialized Program for Research Excellence (SPORE) Molecular Epidemiology Resource (MER) enrolled from 2002 to 2012, ^7,25 and N=400 patients from the International Extranodal Lymphoma Study Group 19 (IELSG-19) study.^11,15 These are the same patients used by Thieblemont et al. to build the MALT-IPI prognostic index out of the 454 initially enrolled in the IELSG-19 study.¹⁶ Patients in the MER cohort received the following treatments: observation (32%), RT (23.9%), alkylator-based chemotherapy (12.8%), single-agent rituximab (12.5%), other treatments (11.1%), anthracycline-containing chemotherapy (4.4%), and surgery (3.4%).⁷ In this data set, EFS was used instead of PFS, and was defined as time from diagnosis until relapse or progression, unplanned retreatment of lymphoma after initial management, or death due to any cause.⁷ The EFS median follow-up of patients in MER cohort was 7.9 years (95% CI: 6.9 to 8 years).

All patients enrolled in the IELSG-19 study received chemotherapy with the following distribution: chlorambucil (32.6%), rituximab and chlorambucil (33%), and rituximab (34.4%).¹¹ The PFS median follow-up of patients in IELSG-19 cohort was 7.6 years (95% CI: 7 to 8.4 years). Using a locked model from the discovery cohort, performance of the two prognostic indices were compared by using the same methods in the two validation cohorts.²⁶ All tests were two-sided, with P-value ≤ 5% considered significant. Statistical analyses were performed using SAS version 9.4 (SAS Institute Inc., Cary, North Carolina) and the R statistical software environment, version 3.4 (https://www.r-project.org).

RESULTS

The proposed index, Revised MALT-IPI, was constructed using the UM cohort training set of 397 out of 405 EMZL patients from a previously publication.¹⁹ External validation was assessed in two independent cohorts, the MER (n=297) and the IELSG-19 (n=400)^11,15 cohorts. A summary of clinical characteristics is shown in Supplemental Table 1; the cohorts were largely similar except for more extranodal site involvement in IELSG-19, more observation as initial therapeutic approach in MER and only systemic therapy in IELSG-19.

Derivation of the Revised MALT-IPI in the UM cohort (training set)

Among the candidate variables tested in univariable analysis, the following were statistically significant predictors of shorter PFS: age >60, age ≥70, elevated LDH, anemia, stage III-IV, ECOG PS 2-4, number of extranodal sites >1, number of nodal sites >4, and MMS (Supplemental Table 2). The final stepwise Cox regression analysis yielded a model with four independent predictors of PFS: age >60 years (HR= 1.47, P=0.021), elevated LDH (HR= 1.86, P=0.001), stage III-IV (HR= 2.00, P=0.0005) and presence of MMS (HR= 2.88, P<.0001) (Table 1). The new Revised MALT-IPI index was then developed by simplifying the variable coefficients of this fitted multivariable Cox model. Consequently, Revised MALT-IPI has scores ranging from 0 to 5, calculated as sum of 1 point for each, age >60 years, elevated LDH, and stage III-IV; and 2 points for MMS. In the UM cohort, the Revised MALT-IPI clearly defined 4 risk groups as follows: low risk (score 0, reference group), low-medium risk (score 1, HR=1.85, P=0.008), medium-high risk (score 2, HR=3.84, P<.0001), and high risk (score 3+, HR=8.48, P<.0001) (Table 1).

Table 1.

Estimates of two Cox models for PFS, goodness of fit AIC statistics, and bootstrap internal validation, in the University of Miami training set (155 events in n=397).

					Cox model coefficient estimate	Bootstrap coefficient estimate
Model/Variables	Category	HR (95% CI)	P	AIC	b (SE)	b (SE)
Multivariable model
Age at diagnosis	≤60	Reference
	>60	1.47 (1.06, 2.05)	0.021	1583.2	0.388 (0.168)	0.402 (0.182)
LDH	Normal	Reference
	Elevated	1.86 (1.27, 2.71)	0.001		0.619 (0.193)	0.629 (0.206)
Stage	I-II	Reference
	III-IV	2.00 (1.35, 2.94)	0.0005		0.691 (0.198)	0.700 (0.208)
MMS	Non-MMS	Reference
	MMS	2.88 (1.83, 4.55)	<.0001		1.059 (0.232)	1.075 (0.246)
Revised MALT-IPI model
	0	Reference		1584.0
	1	1.85 (1.18, 2.91)	0.008		0.616 (0.231)	0.632 (0.240)
	2	3.84 (2.27, 6.50)	<.0001		1.346 (0.268)	1.361 (0.281)
	3+	8.48 (5.26, 13.68)	<.0001		2.138 (0.244)	2.168 (0.260)

Abbreviations: LDH: lactate dehydrogenase; MMS: multiple mucosal sites.

HR: hazard ratio HR=exp(b), where b is the Cox model coefficient estimate. 95%CI: 95% confidence interval. P: p value.

AIC: Akaike information criterion. b is the variable coefficient estimate from the Cox model regression or by bootstrap estimation. The bootstrap estimates are based on B=1000 bootstrap samples of the same size n=405 from the UM training dataset using simple random sampling with replacement. SE: standard error of b.

The multivariable Cox model including 4 risk factors was obtained from a stepwise Cox model, which was used to build the new index Revised MALT-IPI by simplifying the variable coefficients of the fitted model. Revised MALT-IPI has scores ranging from 0 to 5, calculated as sum of 1 point for age >60 years, 1 point for elevated LDH (lactate dehydrogenase), 1 point for tumor stage III-IV, and 2 points for presence of MMS (multiple mucosal sites).

The results for PFS assessing the Revised MALT-IP and MALT-IPI indices in the UM cohort (derivation set), indicate that the model including the new Revised MALT-IPI index is better. This model had an AIC value lower by 26 points, compared to AIC of the MALT-IPI, and slightly higher IAUC and c-statistics. With respect to OS, there is no substantial difference between the Revised MALT-IPI and MALT-IPI model (Supplemental Table 3).

In the UM training cohort, the median PFS was 11.2 years (95%CI: 8 to 15.1) and the 5-year PFS was 65.4% (95%CI: 60% to 70.2%) (Supplemental Figure 1). The 5-year PFS rates by risk-group were: low risk 82.3% (95%CI: 74% to 88.2%), low-medium risk 69.9% (95%CI: 61% to 77.2%), medium-high risk 52.5% (95%CI: 37% to 65.8%), and high-risk 21.3% (95%CI 10.9% to 34.1%). Regarding OS, the median was not reached, and the 5-year OS was 86.1% (95%CI: 81.8% to 89.5%). The 5-year OS rates by risk-group were: low risk 93.7% (95%CI: 87.1% to 97%), low-medium risk 86.7% (95%CI: 79% to 91.7%), medium-high risk 82.6% (95%CI: 68% to 91%), and high-risk 67.1% (95%CI 50.9% to 78.9%). (Supplemental Table 4)

The new Revised MALT-IPI index clearly distributed and distinguished the patients in the UM training cohort, in terms of varying risks for shorter vs. longer PFS and OS. (Supplemental Table 3, Figures 1A and 1C). For comparison, PFS and OS by MALT-IPI are shown in Figures 1B and 1D, respectively, and the Kaplan-Meier estimates of PFS and OS by MALT-IPI risk groups are reported in Supplemental Table 4. Acknowledging a small number of lymphoma-related deaths (n=19), the Revised MALT-IPI appropriately identified those patients at risk for shorter lymphoma-specific survival (Supplemental Figure 2). When we measured lymphoma-specific mortality against the expected mortality in sex and age-adjusted general population (standardized mortality ratio SMR)), we observed statistically significant SMR<1 across all risk groups except in Revised-MALT-IPI high-risk group, where observed SMR 0.834 was not significantly different from 1 (P=0.695). In this group, considering all causes, the observed SMR=1.834 was significantly greater than 1 (P=0.012), likely due to high proportion of non-lymphoma deaths in this relatively small risk group. Similar results were observed by MALTI-IPI (Supplemental Table 5).

Figure 1. Progression-free survival (PFS) and overall survival (OS), positive predictive value (PPV) for PFS and time-dependent area under the curve for PFS in University of Miami training set.

Kaplan-Meier curves of PFS by Revised MALT-IPI (A) and MALT-IPI (B) scores. Kaplan-Meier curves of OS by Revised MALT-IPI (C) and MALT-IPI (D) scores. PPV as function of time for PFS (E). Time-dependent area under the curve (AUC) for PFS (F). Time-dependent PPV was evaluated in the high-risk group (score 3+ for Revised MALT-IPI, and 2+ for MALT-IPI) using the Kaplan-Meier estimator. Color bands are 95% confidence limits.

We also explored the ability of Revised MALT-IPI to identify those patients at risk for HGT. Thirty patients (7.4%) developed pathologically confirmed HGT to diffuse large B-cell lymphoma in the UM cohort. HGT occurred at diagnosis in 8 patients, at first relapse in 17 patients, and at second relapse in 5 patients. Based on 22 HGT events after diagnosis and accounting for 57 non-lymphoma related deaths as a competing risk, there was significant effect of Revised MALT-IPI (Gray’s test P=0.007). For Revised MALTI-IPI score 3+ patients the cumulative HGT incidence curve was statistically significantly higher than those with score 0 (HR=3.70, 95%CI: 1.30-10.56, P=0.015; Supplemental Figure 3A). Similar result was attained when including the 8 HGT that occurred at diagnosis (Gray’s test P<.001; HR=4.10, 95%CI: 1.69 – 9.95, P=0.002; Supplemental Figure 3B).

Figures 1E and 1F show positive predictive value of the prognostics indices Revised MALT-IPI and MALT-IPI for PFS as functions of time, and respective time-dependent areas under the curves (AUCs) for PFS in the UM training cohort. PPV was evaluated in the high-risk group (score 3+ for Revised MALT-IPI, and score 2+ for MALT-IPI). Although there is not statistically significant difference between the Revised MALT-IPI and MALT-IPI, the time-dependent PPV and AUC curves for Revised MALT-IPI are consistently higher than for MALT-IPI for time ≥2 years (AUC at 2 and 10-year 0.716, 95%CI 0.652 to 0.781 and 0.716, 95%CI 0.641 to 0.791 for Revised MALT-IPI and AUC at 2 and 10-year 0.663, 95%CI 0.599 to 0.727 and 0.675, 95%CI 0.595 to 0.756 for MALT-IPI, respectively). Both prognostic indices show excellent model calibration (Supplemental Figure 4A), since observed versus predicted PFS rates at all specific times fall very close to the line with slope 1 of perfect calibration (Supplemental Table 6). Revised MALT-IPI better identified those patients at risk for POD24 (AUC 0.734, 95%CI 0.668 to 0.800) compared to MALT-IPI (AUC 0.684, 95%CI 0.615 to 0.752) (Supplemental Figure 4B). Furthermore, we observed a strong correlation between Revised MALT-IPI risk groups and incidence of POD24 (scores 0 = 8.9%, 1= 16.4%, 2= 35.9%, and 3+= 56.5%, respectively) (Supplemental Figure 5). Data describing risk group reclassification from MALT-IPI to Revised MALT-IPI is shown in Supplemental Figure 4C. Importantly, 17 (12.2%) and 24 (43.6%) of patients in MALT-IPI risk group 1 (n=139) and 2 (n=55) were reclassified to high-risk group 3+ in Revised MALT-IPI, with majority of them (88.2% and 87.5%, respectively) experiencing a PFS event, indicating better identification of patients at risk for event using Revised MALT-IPI.

External validation

The locked Revised MALT-IPI index predicted PFS (or EFS) and OS in two independent validation sets: MER and IELSG-19, confirming its robust predictive power (Figure 3 and 4, Supplemental Table 4). Similar values of the concordance statistics were observed for Revised MALT-IPI and MALT-IPI in these cohorts (Supplemental Table 7). The calibration plots indicated reasonable consistency at 2, 5, and 8 years between training and validations cohorts (Supplemental Figures 6 and 7). Based on joint tests of intercept=0 and slope=1 for overall evidence of calibration, it seems that model was well calibrated in two external cohorts, except for 5 years in MER. (Supplemental Table 6,). Similarly, Revised MALT-IPI and MALT-IPI predicted risk for POD24 in both validation cohorts (Supplemental Figure 8 and 9). Reclassification of risk groups in validation cohorts is shown in Supplemental Figures 10 and 11.

Figure 3. Progression-free survival (PFS) and overall survival (OS) in the IELSG-19 validation set.

Kaplan-Meier curves of PFS for Revised MALT-IPI (A) and MALT-IPI (B) scores. Kaplan-Meier curves of overall survival (OS) for Revised MALT-IPI (C) and MALT-IPI (D) scores.

Kaplan-Meier estimates of PFS and OS by Revised MALT-IPI and MALT-IPI risk groups in the two validation cohorts are reported in Supplemental Table 4 and corresponding curves in Figures 2 and 3. The 5-year EFS in the MER cohort was 66.3% (60.4% to 71.5%). The 5-year EFS rates by risk-group in this cohort were: low risk group 78.9% (95%CI: 67.8% to 86.5%), low-medium risk group 69.4% (95%CI: 60% to 77%), medium-high risk 60.3% (95%CI: 46.2% to 71.9%), and high-risk 41.4% (95%CI 25.9% to 56.3%). The 5-year OS rates were: low risk 98.6% (95%CI 90.3% to 99.8%), low-medium risk 94.7% (95%CI: 88.6% to 97.6%), medium-high risk 76.4% (95%CI: 62.8% to 85.6%), and high-risk 79.7% (95%CI 63.4% to 89.3%).

Figure 2. Event-free survival (EFS) and overall survival (OS) in the MER validation set.

Kaplan-Meier curves of EFS for Revised MALT-IPI (A) and MALT-IPI (B) scores in MER validation set. Kaplan-Meier curves for OS for Revised MALT-IPI (C) and MALT-IPI (D) scores.

In the IELSG-19 cohort, the 5-year PFS was 65.3% (95% CI: 60.4% to 69.8%). The 5-year PFS in this cohort by risk-group were: low risk group 79.2% (95%CI: 70.3% to 85.7%), low-medium risk group 70.5% (95%CI: 62.5% to 77.2%), medium-high risk 46.1% (95%CI: 35.1% to 56.3%), and high-risk 51% (95%CI 35.7% to 64.4%). The 5-year OS were: low risk 100%, low-medium risk 92.9% (95%CI: 87.2% to 96.1%), medium-high risk 77.1% (95%CI: 66.4% to 84.8%), and high-risk 84.4% (95%CI 70.1% to 92.3%).

Discussion

The Revised MALT-IPI is a novel prognostic index specific for patients with EMZL that is validated in three independent cohorts. To develop a clinically useful tool for outcome prediction in EMZL we chose to train a prognostic model based on features associated with high-risk disease based on PFS, the most important primary endpoint in indolent lymphomas. Therefore, we constructed Revised MALT-IPI which better captures the different clinical behavior of EMZL. The Revised MALT-IPI stratified patients from the UM cohort into four distinct groups with markedly different estimated PFS: low risk (median PFS (mPFS): not estimable (NE), 95%CI: 16.3 to NE), low-medium risk (mPFS: 12.8 years, 95%CI: 8.5 to 15.8), medium-high risk (mPFS: 5.8 years, 95%CI: 2.9 to 9.1), and high risk (mPFS: 1.8 years, 95%CI: 1.3 to 2.6). Compared to MALT-IPI, the new index enriches the high-risk groups (score 2 or 3+) from 17% to 26% better identifying patients at risk for progression of disease event.

POD24 is a robust prognostic indicator of survival in follicular lymphoma.^24,27 Most recently, three independent studies demonstrated prognostic implications of POD24 in EMZL.^5,6,19 Revised MALT-IPI identifies those patients at risk for POD24 obviating the waiting period needed to recognize them. Patients from the UM cohort included in the high-risk group experienced a 2-year PFS of 44% vs >69% in the other lower risk groups (Supplemental Table 4). Results in validation cohorts revealed similar pattern of lower 2-year PFS in the high-risk group, demonstrating the fitness of the Revised MALT-IPI for this relevant endpoint. Moreover, revised MALT-IPI appropriately identified those patients at risk for HGT.

We constructed the Revised MALT-IPI based on four prognostic factors including age >60 years, elevated LDH, stage III-IV and presence of MMS. The new index shares elevated LDH and advanced stage with MALT-IPI and other prognostic scores in lymphoma such as International Prognosis Index and Follicular Lymphoma International Prognosis Index scores. The age threshold was set up at 60 years in an effort to decrease non-lymphoma-related events commonly observed in older age group. Furthermore, multivariable Cox model identified MMS as a strong factor associated with PFS that was incorporated into the Revised MALT-IPI and not present in the initial MALT-IPI. EMZL presenting with MMS is a novel entity characterized by more aggressive behavior. MMS has clinical significance and can be easily determined based on staging studies with the presence of two or more different extranodal sites of disease.¹⁹ MMS influences the Revised MALT-IPI and is the only factor providing two points to the final score. Patients with MMS are characterized by a different disease biology; however, genetic signatures associated with this entity are presently unknown underscoring the need for large collaborative efforts attempting to better characterize this entity.

Our results were validated in cohorts from the United States and a clinical trial conducted in Europe demonstrating its prognostic implications independent of a specific clinical setting (clinical trial or routine clinical practice), geography and treatment approach. Results obtained in the training set demonstrated similar survival distribution to the MER cohort, however, less clear separation of the high-risk groups was observed in the IELSG-19 cohort. These differences may be attributed to calculation of prognostic factors in the IELSG-19 cohort at the time of clinical trial enrollment rather than at lymphoma diagnosis, while some patients enrolled in this trial received prior local therapies (n= 32, 8%) and were not censored at that time. Additional explanations for different performance of the Revised MALT-IPI in the validation cohorts may be explained by differences in baseline characteristics, lack of a common criteria for treatment initiation, selection, response assessment, and follow-up. This reduction in performance is common when moving from the training to validation cohorts. Furthermore, biological factors present at different geographies may influence disease course and prognosis. For example, the association of ocular adnexal EMZL with Chlamydia psittaci has been well established in Europe,^28,29 however, this association is not confirmed in the United States.^30,31

Limitations of the present study include the heterogeneity in treatment selection in the training set. Although this limitation is important, EMZL is a rare disease without clearly defined standard therapies. The training cohort is composed by a well-weighted stage and treatment distributions representing the real-world scenario in EMZL. Moreover, Revised MALT-IPI was tested and validated in two datasets that used diverse therapeutic approaches demonstrating the prognostic value of this index independently of a specific treatment. Further, Revised MALT-IPI was validated in a clinical trial cohort treated with predetermined therapies.

Although EMZL is commonly defined as an indolent lymphoma, outcomes are heterogeneous with a subset of patients experiencing shorter PFS. MALT-IPI is a valuable model to risk stratify patients with EMZL with excellent performance across cohorts. However, the high-risk patients identified by this index are different from the high-risk patients identified by the revised MALT-IPI (Supplemental Figures 4 C, 10 and 11). Revised MALT-IPI is based more on disease characteristics and less on older patient age and thus enables early identification of high-risk patients in general and specifically patients with prediction of lymphoma progression within 2-years from diagnosis. This new index better identifies patients at risk of treatment failure enhancing the clinical utility of MALT-IPI and may help in the design of clinical trials in EMZL. Future studies will need to investigate the utility of Revised MALT-IPI in patients treated with novel agents aiming for prospective validation. Collaborative efforts are needed to better understand biologic abnormalities observed in patients with shorter survival and targeting this population in the design of clinical studies.