Follicular lymphoma in the modern era: survival, treatment outcomes, and identification of high-risk subgroups

Patients with follicular lymphoma (FL) frequently require multiple treatments during their disease course; however, survival based on lines of treatment remains poorly described in the post-rituximab era. Also, the Follicular Lymphoma International Prognostic Index (FLIPI) score was developed to predict survival at diagnosis, yet it remains unknown whether increase in FLIPI score following an initial observation period is associated with less-favorable outcomes. To address these knowledge gaps, we retrospectively studied 1088 patients with FL grade 1–3A managed between 1998 and 2009 at our institution. Median overall survival (OS) and progression-free survival (PFS) after first-line treatment were not reached and 4.73 years, respectively. Following successive lines of treatment, years of median OS and PFS were, respectively: after second-line, 11.7 and 1.5; third-line, 8.8 and 1.1; fourth-line, 5.3 and 0.9; fifth-line, 3.1 and 0.6; sixth-line, 1.9 and 0.5. In initially observed, subsequently treated patients, FLIPI score increase after observation was associated with inferior survival following first-line treatment. The reduced survival we observed after second-line and later therapy supports the development of new treatments for relapsed patients and benchmarks historical targets for clinical endpoints. This study also highlights the utility of changes in FLIPI score at diagnosis and after observation in identifying patients likely to have worse outcomes.


Introduction
Follicular lymphoma (FL) is the second most common lymphoma in the United States, with approximately 14,000 patients being diagnosed each year 1,2 . While FL remains incurable, overall survival (OS) continues to improve due to improvements in diagnostic tools and supportive care, the development of the monoclonal anti-CD20 antibody rituximab, and the increasing number of FDA-approved therapies [3][4][5] . Current first-line regimens for FL typically achieve high response rates [6][7][8][9][10][11][12] . As the disease recurs, patients are treated with multiple lines of therapies during their lifetime. The outcome of these different regimens and the impact on patient survival remain understudied in the modern era. A secondary analysis from the LymphoCare study recently reported progression-free survival (PFS) but not OS for firstthrough fifth-line therapy 13 . We investigated how the survival outcomes OS, PFS, and event-free survival (EFS) evolved after multiple lines of therapy, information that should aid in estimating clinical endpoints when designing clinical trials for multiply-relapsed patients. Another aim of this study was to provide treating physicians with additional biomarkers predictive of high-risk patients that might permit identification for early treatment intervention.
Prior studies have shown patients with FL with low tumor burden can be initially observed without impacting survival 14,15 . Even so, patients who undergo initial observation follow a heterogenous clinical course. The FL International Prognostic Index (FLIPI), a five-factor risk model based on age, stage, lactate dehydrogenase and hemoglobin levels, and number of nodal areas, has been validated as a diagnostic model in both the pre-and postrituximab eras. It is used to predict patient survival and to stratify patients in clinical trials 4,16 . However, many patients are initially observed and have prolonged lead time from diagnosis to first treatment. Moreover, there are no data on the stability of FLIPI score in initially observed patients, nor the impact of changes in FLIPI score during initial observation on survival or the rate of subsequent histological transformation to other types of lymphoma. Here we investigate, for patients who were initially observed, whether FLIPI risk group had changed between time of diagnosis and the time of initial treatment, and whether increased FLIPI score impacted outcomes after treatment.

Study design and patients
We retrospectively examined outcomes for patients diagnosed during the years 1998−2009 with de novo FL managed at Memorial Sloan Kettering Cancer Center (MSK). The beginning year was selected to capture patients with exposure to rituximab, FDA-approved for non-Hodgkin's lymphoma in 1997; end year was chosen to ensure at least 10-year follow-up. The institutional review board approved this retrospective analysis; all patients had given written informed consent to biospecimen protocols. We excluded patients who were <18 years old; were diagnosed with an active concurrent malignancy; had grade 3B FL at diagnosis because of it's management similar to diffuse large B cell lymphoma; or whose pathology showed composite histology at diagnosis. We also excluded patients who were managed with fewer than three visits at our institution, indicative of a consultative role; but not patients who had died before their third visit. At least one pathology specimen for each patient with FL was centrally reviewed at our institution. Documentation of transformations to other forms of lymphoma was based on biopsy confirmation. Clinical staging and best overall response, as assessed by treating physicians, were extracted from chart review. Ambiguity in staging or response to therapy was settled by review of available radiographic and pathologic reports.
Intention of treating physician to manage certain patients with active surveillance was assessed by medical record review. Patients whose first-line treatment began ≥12 months after diagnosis were considered initially observed. For all other patients, charts were reviewed to determine the physician's intent of observation or initial treatment.

Statistical methods
Patient outcome was analyzed by OS, PFS, and EFS. OS was calculated either from time of diagnosis or from time of treatment commencement, as indicated on the figure, until last follow-up or death. PFS and EFS times for sequential lines of therapy were calculated from treatment commencement until qualifying event (progression or death for PFS; progression, change of treatment, or death for EFS). EFS12 failures were defined as patients with disease progression, change of therapy or death within 12 months of treatment initiation (EFS12). Date of progression was assessed based on chart review; for cases of unclear documentation, radiographic imaging was reviewed. Patients with inadequate response to treatment, change of treatment, or stable disease without subsequent documented progression were censored in the PFS analysis. Overall survival from time of transformation to diffuse large B-cell lymphoma (DLBCL) or DLBCL with features of Burkitt lymphoma was calculated from the time of first recorded pathologic transformation. OS, PFS, and EFS were evaluated using the Kaplan −Meier method. The OS, PFS, and EFS times for sequential lines of therapy were compared across lines of therapy using the log-rank test with adjustment for within-patient correlation. PFS and EFS were compared to other clinical variables using the log-rank test. Chisquared method was used to compare PFS curves between groups by stage.
The risk of biopsy-proven transformation to DLBCL or DLBCL with features of Burkitt lymphoma was assessed using a competing risk analysis wherein patients can experience either transformation or death without transformation. Time origin was set at time of diagnosis; rates of transformation and death without transformation at specific time points were calculated.
We assessed patients' FLIPI score at diagnosis and at initiation of first treatment 16,17

Data-sharing statement
For de-identified original data, please contact leec@mskcc.org. Requests would be submitted to our institutional review board for consideration and review prior to any data sharing.

Patient characteristics
We identified 1446 consecutive adult patients (≥18 years) diagnosed with FL at MSK from January 1, 1998 to December 31, 2009. We excluded 358 patients who had active concurrent malignancy, grade 3B FL, mixed histology at initial diagnosis, or fewer than three visits to our institution ( Fig. 1). Median age of the resulting 1088 patients was 57 years (range, 20-94) with median followup of 8.3 years (range, 0.2-17.5). Clinical characteristics are provided in Table 1. Two-thirds of patients had stage III or IV disease at diagnosis. Bone marrow biopsy results were available for 79% of patients; therefore, patients with stage III disease were provisionally staged dependent on status of bone marrow biopsy. FLIPI risk score at diagnosis was available for 851 (78%) patients, of whom almost one-third were low-risk. Positron emission tomography with 2-deoxy-2-[fluorine-18] fluoro-D-glucose integrated with computed tomography (PET/CT) was available for 60% of patients at diagnosis. Of the 924 patients who required treatment, 468 (51%) patients were managed with an anthracycline during their treatment course and 739 (80%) were treated with rituximabcontaining therapy.

Initial observation vs treatment
Of the 1088 patients included in this analysis, 461 were initially observed, 617 required initial therapy with systemic therapy or radiation, and 10 patients underwent a diagnostic and therapeutic surgery (Fig. 1). Eight of the 10 surgery patients never required additional treatment. In the initial observation group, 156 patients never required treatment. In total, 164/1088 (15%) patients never required treatment beyond their diagnostic and therapeutic procedure. We excluded two patients with incomplete first-line treatment information, leaving 922 patients who required systemic therapy or radiation and had available treatment data.
Among the patients who were initially observed rather than receiving immediate treatment (n = 461), median duration of observation was 3.9 years (95% CI, 3.4-4.5). For patients who were initially observed and subsequently treated (n = 305), median time to first treatment was 2.3 years (range, 0. 27-13.33). In the initially treated population (n = 617), median time from diagnosis to first treatment was 0.14 year (95% CI, 0.13-0.15). Overall survival was not adversely affected by observation strategy (p = 0.206) ( Supplementary Fig. 1).

Survival outcomes by time period
Management of FL has evolved over the years; therefore, we evaluated the effect of modern FL treatments on clinical outcomes. Patients were distributed between three time frames across two decades

Patterns of treatment in FL
Among the 922 patients who received treatment, firstline therapy was rituximab in combination with chemotherapy in 52.1% (n = 480) and chemotherapy alone in 10.3% (n = 95) (Supplementary Table 1). Anthracyclinebased chemotherapy (with or without rituximab) represented 45% of the first-line treatment. Rituximab single-agent therapy and radiation alone were first-line treatment in 18.1% (n = 167) and 15.5% (n = 143) of the population, respectively. Recurrent uses of single-agent rituximab and radiotherapy were common throughout multiple lines of therapy. Alkylator-based chemotherapy represented 22.3% of second-line therapy. Radioimmunotherapy made up 3-9% of treatments in secondto sixth-line therapy. Ten percent of treated patients (91/ 922) received stem cell transplants during their course of therapy, 6% (n = 54) autologous stem cell transplant and 4% (n = 37) allogeneic stem cell transplant. Investigational therapies were uncommon in first-line therapy but increased to 8-22% with increasing lines of therapy.  (31/164) and after first-line therapy in 81% (133/164) of patients. For patients with a transformation event after firstline therapy but after diagnosis of FL, anthracycline was a component of prior therapy in 71/133 (53%) events. Competing risk analysis showed risk of histological transformation at 2, 5, and 10 years after diagnosis to be 3%, 8%, and 16%, respectively. Risk of death without transformation at 2, 5, and 10 years after diagnosis was 2%, 6%, and 14%, respectively ( Supplementary Fig. 3A). We compared OS in patients whose transformation event was before or after first-line therapy. The latter was associated with increased risk of death (HR 3.35; 95% CI, 1.34-8.39; p = 0.010) (Supplementary Fig. 3B). Median OS after transformation was not reached (95% CI, 10.1 years − not reached [NR]) in patients who transformed prior to frontline therapy (n = 31), whereas median OS in patients who transformed after frontline therapy (n = 133) was 7.6 years (95% CI, 2.4-NR; log-rank p = 0.01). Fiveyear OS from time of transformation was 83% (95% CI, 61-93%) for patients who transformed prior to frontline therapy and 55% (95% CI, 45-64%) for patients who transformed after frontline therapy.

OS, PFS, and EFS by line of therapy
We evaluated 922 patients for OS, PFS, and EFS by line of therapy. Despite FL treatment options increasing between 1998 and 2009, individual patients' treatment outcomes nevertheless declined with increasing lines of therapy (Fig. 3a). Among the 922 patients receiving firstline therapy, median OS was not reached ( Table 2). After second-line therapy, median OS was 11.67 years (95% CI, 9.67−NR). Median OS further decreased with each line of therapy and decreased to 3.13 years (95% CI, 2.22-6.13) after fifth-line therapy. At four or more lines of therapy, median PFS was 0.9 years (95% CI, 0.6-1.1), and median EFS was 0.6 years (95% CI, 0.5-0.8) (Fig. 3b, c).
PFS and EFS outcomes after first-line therapy were affected by stage at treatment: patients with stage I FL enjoyed a prolonged remission after first-line therapy (Supplementary Table 2). Median PFS was 8.8 years (95% CI, 6.7-11.4) for stage I patients vs <5.1 years for stage II −IV patients (chi-squared p = 0.006). EFS after first-line therapy was also statistically significantly longer for stage I patients (chi-squared p = 9 × 10 -4 ). However, stage at first treatment was not prognostic for PFS or EFS following second-line or later therapy (Supplementary Table 2).

Prognostic value of changes in FLIPI score at diagnosis and treatment
Advanced-stage FL is commonly managed with active observation. We sought to understand the clinical impact when FLIPI worsens during observation, between diagnosis and initial treatment. We identified 684 patients  with FLIPI available at diagnosis and first treatment and excluded 496 patients who initiated therapy within 6 months (Fig. 4a) For the 164 patients who never required therapy, we reviewed medical records to determine FLIPI scores for the first 5 years of their observation (Fig. 4b). We were able to ascertain FLIPI score from diagnosis and at least one other time point for 105 patients. For patients with multiple FLIPI scores over the 5-year period, the maximum FLIPI was used for comparison to FLIPI at diagnosis. Sixty-one (58%) of these patients had a stable FLIPI, 27 (26%) patients had a increased FLIPI, and 17 (16%) patients were excluded for decreased or continued high FLIPI. Among the 164 patients, median duration of follow-up was 7.3 years (range, 0.2-16.7 years).
In patients observed ≥6 months from their diagnosis, increased FLIPI was associated with an inferior OS (p = 0.011) (Fig. 5b). Median OS was not reached for either the increased or stable FLIPI groups. PFS was also notable for a negative association with increased FLIPI (p = 0.002) (Fig. 5c). The median PFS was 3.14 years (95% CI, 2.48-6.79) in the increased FLIPI group and 6.25 years (95% CI, 5.05−NR) in the stable FLIPI group. Similarly, in patients who were observed ≥12 months, increased FLIPI was a marker for inferior OS and PFS. While the median OS was not reached for either increased or stable FLIPI groups, OS was statistically significantly different (log rank p = 0.006) (Fig. 5d). The median PFS was 2.98 years (95% CI, 2.48-6.79) in the increased FLIPI group and 6.25 years (95% CI, 3.62−NR) in the stable FLIPI group (logrank p = 0.011) (Fig. 5e).
Increased FLIPI during observation was also associated with a higher rate of EFS12 failure after treatment initiation. For 76 patients who were observed ≥6 months and subsequently treated with increased FLIPI, 21 (27.6%) failed EFS12 after treatment. In comparison, for 68 patients who were treated with stable FLIPI after ≥6 months of observation, only 8 (11.8%) failed EFS12.
In patients observed without therapy, clinical parameters seen at diagnosis such as SUV, Ki-67, and abnormal LDH were similar across all groups (Supplementary Table 4). Patients with increased vs stable FLIPI were similarly treated with rituximab at first-line therapy. Patients with increased FLIPI had a higher risk of transformation compared to patients with stable FLIPI. Using a competing risk analysis, in patients who were observed ≥6 months, transformation rate at 10 years after diagnosis was 28.1% in patients with increased FLIPI vs 14.5% in patients with stable FLIPI. Similarly, in patients who were observed ≥12 months, transformation rate at 10 years after diagnosis was 27.6% in patients with increased FLIPI vs 14.4% in patients with stable FLIPI.

Discussion
Despite the commonly indolent nature of FL, a subpopulation of patients harbors aggressive disease. Moreover, the heterogenous nature of FL makes approval of novel therapies challenging. Therefore, the identification of patients with high-risk biology is important for the development of future treatments. Existing markers of high risk include failure to achieve a complete response to initial chemoimmunotherapy, and retreatment within 2 years of initial treatment. In this study, we describe the outcomes of patients with FL treated at our center, with the aim of providing additional biomarkers to identify high-risk patients for early treatment intervention. This population was diagnosed after FDA approval of rituximab; therefore, rituximab was readily available to manage the disease. Despite this, 20% (185/922) of the patients who required therapy never had rituximab exposure throughout their course of therapy; however, 52% (97/185) of these patients were treated with radiotherapy.
We found outcomes of this generation of patients are improving. Median OS for all patients has not been reached even with nearly 20 years follow-up on the earlier patients. Five-year OS was 92% for all patients, highlighting the good outcomes of most patients with FL. Progression-free survival improved during the study period, likely through improved therapies. This study supports earlier studies demonstrating patients receiving initial observation have the same OS as those treated promptly after diagnosis 15,18,19 .
The Stanford University FL experience demonstrated improved OS over multiple eras, with stable PFS of approximately 2 years across all eras between 1960 and 2003 20 . Our study reports median OS is not reached for patients diagnosed during time periods 1998-2000, 2001-2005, and 2006-2009; and median PFS improved from 2.5 years for 1998-2000 to a range of 4.8-6.0 years in the latter two time frames. We note this improved PFS despite the lack of maintenance rituximab in 80% of our patients 6 . We found 5-year PFS for patients after first-line therapy to range from 40 to 55%. In comparison, the PRIMA study demonstrated a 6-year PFS estimate of 43% in the no-maintenance arm and 69% in the rituximab maintenance arm 11,12 . The FOLL05 study of advancedstage FL did not incorporate maintenance rituximab and had an 8-year PFS of 48% 7 . The StiL study demonstrated median PFS not reached, with median 3.75 years of follow-up in R-bendamustine arm and 3.4 years in the RCHOP arm 10 .
Transformation to aggressive lymphoma has been linked to adverse prognosis [21][22][23][24][25][26] . Similar to results from the PRIMA trial where 194/1018 (19%) patients had documented histologic transformation with a median follow-up of 6 years 21 , we identified 167 histologic transformation events in 1088 patients (15%) with a median follow-up of 8.9 years. The data support an adverse prognosis in patients with FL who transform after first-line therapy. In PRIMA, 5-year OS of the transformed population was approximately 40% at 5 years 21 , similar to our 5-year OS estimate of 55% in patients who transformed after first-line therapy.
Outcomes in FL after multiple lines of therapy have been incompletely described in the literature, another reason for our current study. Follicular lymphoma is typically responsive to frontline chemotherapy but may later become refractory 27,28 . Prior to the widespread use of rituximab, retreatment after multiple lines of therapy was required every 2.75 years and median OS after first relapse was 5 years 27 . The development of more effective FL therapies has improved outcomes. In our dataset, median OS for patients who received second-line treatment was >10 years.
The availability of new treatments for FL also reflects a need for new benchmarks to identify effective treatment strategies. In many tumors, improvement in OS remains the gold standard for clinical approval; however, the long heterogenous course of FL makes OS a challenging primary endpoint. Progression at 24 months is a potential endpoint for FL; however, its impact on prognosis is less validated [29][30][31] . PFS remains a reasonable clinical trial endpoint. We demonstrate that at fourth-line therapy and beyond, PFS is less than 1 year, providing a clinical endpoint that is robust and feasible to assess within a clinical trial context. This is supported by the National LymphoCare study that also showed PFS decreasing with increasing lines of therapy: 6.6, 1.5, 0.83, 0.69, and 0.68 years after first, second, third, fourth, and fifth-line therapy, respectively 13 . Both our and the LymphoCare analyses provide a benchmark for future drug approval in relapsed FL.
This study also assessed the impact of changes to the FLIPI score during observation as a marker for adverse prognosis. Patients observed for ≥6 or ≥12 months were evaluated for increased FLIPI between diagnosis and prior to first-line treatment. We found that subsequently treated patients whose FLIPI increased during observation had inferior OS and PFS. FLIPI increase was most commonly associated with increased nodal involvement, progression to advanced stage, and abnormal LDH. Of course, the increased FLIPI population may reflect patients who had adverse biology at diagnosis, especially since we continue to demonstrate that observation had no impact on OS. Nonetheless, in this era of readily available, easily tolerated therapies, it may be questionable whether initial observation is the ideal strategy. Future randomized trials comparing observation to upfront treatment may be warranted.
The strength of this analysis lies in the large cohort. However, it is limited by its single-center and retrospective nature. Though our patients received heterogenous treatments reflecting a real-world strategy, they lack the greater diversity of patients treated by multiple unaffiliated practices.
This study benchmarks a single institution's outcomes of FL in the post-rituximab era. The work identifies higher number of lines of therapy and increased FLIPI score as markers for high-risk biology in FL. Based on EFS and PFS observed in this large retrospective series, we can consider designing clinical trials for fourth-line treatment of FL. Future prospective studies are needed to assess a possible correlation between increased FLIPI score and inferior outcomes.