FLIPI24: A Modern Prognostic Model and Clinical Trial Enrichment Tool for Newly Diagnosed Follicular Lymphoma

Abstract

PURPOSE

While most patients with follicular lymphoma (FL) can expect an indolent course, progressive lymphoma remains the primary cause of death during the first decade after diagnosis. Progression of disease within 24 months (POD24) of starting first-line (1L) immunochemotherapy defines a high-risk population with poor survival, but better risk stratification at diagnosis is needed.

METHODS

The FLIPI24 model was developed and internally validated to predict 24-month event rates using individual data from 4,485 patients treated with 1L immunochemotherapy from ten observational cohorts of FL. Overall and cause-specific survival was further evaluated in FLIPI24 risk groups. External validation in the 1L immunochemotherapy setting was performed using the prospective observational Lymphoma Epidemiology of Outcomes (LEO) cohort (N=565) and three randomized phase 3 trials (N=3,192); extension to all patients with FL (any 1L therapy) was performed in the LEO Cohort (N=1,445) and its MER subcohort (N=1,074).

RESULTS

The FLIPI24 model utilizes age and four blood-based variables (hemoglobin, lactate dehydrogenase, beta-2 microglobulin, white blood cell count). FLIPI24 showed consistent performance across validation and extension datasets which was superior to existing prognostic tools. Across the four external immunochemotherapy validation datasets, patients with high-risk FLIPI (23–31% of patients) had significantly higher 24-month event rates (22–35%) and inferior 5-year OS (77–83%) compared to patients with low-risk FLIPI24 (29–31% of patients, 24-month event rates:10–12%; 5-year OS:96–97%). Results were consistent when evaluating lymphoma-related death and when extended to all patients with FL.

CONCLUSIONS

The FLIPI24 model robustly stratifies, at diagnosis, patients with FL at increased risk of lymphoma-related death vs. patients with very low lymphoma-related mortality during the first decade after diagnosis. FLIPI24 can be used to enrich future clinical trial designs in newly diagnosed FL.

INTRODUCTION

Follicular lymphoma (FL) is the most common indolent non-Hodgkin lymphoma in North America and Europe.^1,2 While most patients with FL can expect their disease to follow an indolent course and have a normal life expectancy over the first decade from diagnosis with current standards of care, in some patients the disease will transform to a more aggressive lymphoma³ or become refractory to treatment. Progressive lymphoma remains the primary cause of death during the first decade after diagnosis.⁴ Progression of disease within 24 months of diagnosis (POD24) in immunochemotherapy treated FL is strongly associated with subsequent overall survival (OS).^5–7 In contrast, patients without early events have very low rates of lymphoma-related mortality in the first decade after diagnosis,⁴ with survival comparable to the background population.⁸ However, identification of patients at risk of early failure and subsequent inferior OS prior to initiating therapy and waiting two years to further risk-stratify remains a key clinical challenge.

First-line (1L) management of FL is heterogenous, ranging from observation to anti-CD20 antibody with or without combination chemotherapy (immunochemotherapy). The decision to treat and type of therapy is generally based on a combination of symptoms and disease burden,⁹ typically defined by the Groupe d’Etude des Lymphomes Folliculaires (GELF) criteria,¹⁰ but this varies widely based on provider, health system, and individual patient factors. The FLIPI model, developed in patients diagnosed between 1985 and 1992, remains the most commonly utilized prognostic tool in newly diagnosed FL. Newer models developed in the immunochemotherapy era^11–13 have been proposed but are not commonly utilized for therapeutic decision-making or determination of clinical trial eligibility. To address these needs, we developed and validated the FLIPI24 clinical prediction model for risk stratification and clinical trial enrichment in newly diagnosed patients with FL.

METHODS

Data Sources and Study Population

Individual patient data were harmonized and pooled from 10 observational cohorts of FL from Europe, North America, and Australia (Figure S1; Table S1). Eligible patients for model development were diagnosed with grades 1–3A FL and initiated 1L R-CHOP, R-CVP, R-bendamustine (B-R), or like therapies from 2002–2018. Data were collected in accordance with local guidelines including anonymization as appropriate (see supplemental materials). The study was approved by the Mayo Clinic IRB.

Outcomes

The FLIPI24 model was developed to predict outcomes in FL by leveraging the significance of early events in patients treated with 1L immochemotherapy. Event-free survival (EFS) was defined as time from the start of therapy until disease progression, relapse, initiation of second line therapy, transformation to DLBCL or high-grade lymphoma, or death due to any cause. EFS at 24 months (EFS24), which encompasses POD24 (including early transformations) and death from any cause, was the primary endpoint for model development. OS, defined as time from diagnosis (or 1L trial registration) until death from any cause, was a co-primary endpoint. OS is particularly relevant for model assessment given the heterogeneity of treatments utilized in newly diagnosed FL, the impact of anti-CD20 antibody maintenance on EFS but not OS,¹⁴ and early progression lacking surrogacy for overall survival.¹⁵ The secondary endpoint of cause of death was classified as due to lymphoma (including treatment-related), unrelated to lymphoma (including other malignancies and other causes) and unknown.⁴

Candidate Predictors

The following pre-treatment variables were harmonized for modeling based on clinical relevance and data availability: age, gender, Ann Arbor stage, ECOG performance status (PS), number of extranodal sites, number of involved nodal groups, bone marrow involvement, B symptoms, histologic grade, lactate dehydrogenase (LDH) relative to upper limit of normal, hemoglobin (HGB), beta-2 microglobulin (B2M), white blood cell count (WBC), absolute lymphocyte count, and platelets (Tables S2–S3, Figure S2).

Statistical Analysis

We followed TRIPOD guidelines for model development and validation¹⁶ (Table S4). Model development was performed on pooled individual level data from the 10 observational studies (N=3,577) with 20% (N=908) held out for internal validation (Figure S1). A model for EFS was fit using Cox proportional hazards regression. The Cox model was stratified for geographic region and type of 1L therapy and a time-dependent nuisance covariate for maintenance rituximab was used to adjust for the prolongation of EFS in patients selected for maintenance. Multiple imputation was performed to address missing data.¹⁷ Discrete risk groups (low, intermediate and high) were defined based on the distribution of FLIPI24 scores and per clinical benchmarks from recent 1L immunochemotherapy trials.^18,19 Discrimination was assessed using Harrell’s c-statistic from Cox models for EFS24 and OS, and results were compared to existing clinical prognostic tools (FLIPI²⁰ and PRIMA-PI¹²). Calibration was assessed by comparing EFS24 and 5-year OS estimates for FLIPI24 risk groups across the external datasets. External validation of the model was initially performed in the multi-center observational Lymphoma Epidemiology of Outcomes (LEO) cohort study²¹ [NCT02736357] by the LEO Statistics Core. After the initial public presentation of the model, further independent validation was performed in the international clinical trials PRIMA²² [NCT00140582], RELEVANCE¹⁹ [NCT01476787], and GALLIUM¹⁸ [NCT01332968] by the LYSARC research group. Further assessment of the model in all patients with FL was performed using the LEO and the Lymphoma Molecular Epidemiology Resource (MER)²³ cohorts.

RESULTS

Patient Characteristics

Patient characteristics are summarized in Table 1. In the model-building cohort (N=3,577), median age was 61 years (Interquartile Range [IQR] 53–69) and 1,810 (51%) were female. The immunochemotherapy backbone was CHOP-based in 1,874 patients (52%), CVP-based in 973 patients (27%), and bendamustine-based in 730 (20%). 2,164 patients (61%) received anti-CD20 maintenance. The median follow-up was 85 months (IQR: 55–124); 792 patients (22%) had an early event as defined for model development. A total of 835 patients have died during the study. The 5-year and 10-year OS from start of treatment were 84% (95%CI 83–86) and 72% (95%CI 70–73), respectively, while when measured from the time of an early event were 47% (95%CI 43–51) and 35% (95%CI 31–39) respectively (Figure S3).

Table 1.

Patient characteristics

	FLIPI24 Pooled Dataset		External Validation
	Development	Internal Validation	LEO (IC-treated)	PRIMA	RELEVANCE	GALLIUM
N	3,577	908	565	1,018 randomized	1,030	1,144
Year of enrollment	2002–2018	2002–2018	2015–2020	2004–2007	2011–2014	2011–2014
Patient characteristics
Age, years, median (Q1-Q3)	61 (53–69)	61 (53–69)	61 (52–69)	57 (48–64)	59 (50–67)	59 (50–67)
Sex, female	1,810 (51%)	442 (49%)	258 (46%)	485 (48%)	528 (51%)	609 (53%)
Grade 3A	765 (21%)	183 (20%)	101 (18%)	89 (9%)	124 (12%)	164 (14%)
B symptoms	1,090 (31%)	303 (33%)	136 (24%)	316 (31%)	275 (27%)	381 (33%)
Stage III/IV	2,999 (84%)	766 (84%)	452 (80%)	918 (90%)	960 (93%)	1,037 (91%)
ECOG PS 2–4	305 (9%)	87 (10%)	33 (6%)	36 (4%)	27 (3%)	38 (3%)
>4 nodal groups	1,697 (47%)	451 (50%)	280 (50%)	657 (65%)	679 (66%)	-
Bone marrow involvement	1,635 (46%)	403 (44%)	253 (45%)	560 (55%)	568 (56%)	597 (53%)
LDH ratio, median (Q1-Q3)	0.90 (0.74–1.11)	0.89 (0.71–1.11)	0.84 (0.72–1.02)	0.88 (0.72–1.08)	0.85 (0.72–1.04)	0.86 (0.72–1.06)
LDH > ULN	1,406 (40%)	352 (40%)	155 (27%)	337 (33%)	293 (29%)	347 (30%)
HGB, median (Q1-Q3)	13.5 (12.4–14.5)	13.6 (12.4–14.6)	13.5 (12.5–14.5)	13.5 (12.4–14.5)	13.4 (12.4–14.4)	13.4 (12.3–14.4)
HGB < 12	658 (18%)	169 (19%)	96 (17%)	196 (19%)	190 (18%)	211 (19%)
B2M, median (Q1-Q3)	2.5 (1.9–3.5)	2.5 (1.9–3.4)	2.5 (2.1–3.2)	2.4 (1.8–3.2)	2.5 (2.0–3.4)	2.5 (1.9–3.5)
WBC, median (Q1-Q3)	6.8 (5.4–8.6)	6.8 (5.5–8.5)	6.4 (5.3–7.9)	6.7 (5.3–8.3)	6.4 (5.2–8.0)	6.1 (5.0–7.8)
Prognostic models
FLIPI
Low (0–1)	717 (21%)	184 (21%)	142 (25%)	216 (21%)	153 (15%)	237 (21%)
Intermediate (2)	1,097 (32%)	267 (31%)	198 (35%)	370 (36%)	374 (36%)	421 (37%)
High (3–5)	1,643 (48%)	423 (48%)	225 (40%)	431 (42%)	503 (49%)	480 (42%)
PRIMA-PI
Low	817 (40%)	217 (42%)	255 (45%)	367 (37%)	337 (34%)	390 (36%)
Intermediate	509 (25%)	121 (24%)	144 (25%)	356 (36%)	313 (32%)	317 (29%)
High	720 (35%)	173 (34%)	166 (29%)	269 (27%)	332 (34%)	377 (35%)
Treatment
R-CVP or like	973 (27%)	237 (26%)	12 (2%)	222 (22%)	28 (3%)	113 (10%)
R-CHOP or like	1,874 (52%)	493 (54%)	217 (38%)	768 (75%)	372 (36%)	345 (30%)
B-R or like	730 (20%)	178 (20%)	336 (59%)	28 (3%)	117 (11%)	686 (60%)
R2	N/A	N/A	N/A	N/A	513 (50%)	N/A
Maintenance anti-CD20	2,164 (61%)	526 (57.9%)	130 (23%)	randomized	planned	planned

IC = immunochemotherapy; 1L = first-line therapy; PS = performance status; LDH = lactate dehydrogenase; ULN = upper limit of normal; HGB = hemoglobin; B2M = beta-2 microglobulin; WBC = white blood cell count; R-CVP = rituximab, cyclophosphamide, vincristine, and prednisone; R-CHOP = rituximab, cyclophosphamide, doxorubicin, vincristine, and prednisone; B-R = bendamustine and rituximab; R2 = lenalidomide and rituximab;

Model Development and Internal Validation

Details on model development are provided in the supplemental material. The final model was selected based on performance and ease of application, prioritizing blood-based biomarkers over clinical factors (e.g., number of nodal groups), and included age (non-linear), and LDH/ULN, B2M, HGB and WBC (all linear). The model’s numerical output is the probability of an event within the first 24 months after start of therapy based on the assumption of either R-CHOP or B-R like therapies and a 50% likelihood of receiving maintenance anti-CD20 antibody therapy (Figure 1). Discrete risk groups were identified for broad application in all newly diagnosed patients. The algorithm used to predict EFS24 is provided in Table S5. The locked model was then applied to the internal validation dataset (Figure S4). Model performance for both EFS24 and OS in the internal validation dataset was superior to FLIPI and PRIMA-PI (Table S6).

Figure 1.

FLIPI24 model nomogram.

External Validation in Patients Treated With 1L Immunochemotherapy

The locked FLIPI24 model was then applied to four external datasets of 1L IC-treated patients: LEO was an observational study managed per treating physician, and the PRIMA, RELEVANCE, and GALLIUM trials comprised patients with high tumor burden that met criteria for clinical trial entry. Patient characteristics (Table 1) were broadly similar, with trials featuring more patients with advanced stage disease. The IC-backbone varied across the studies. Anti-CD20 maintenance was planned for all patients on RELEVANCE and GALLIUM per protocol and randomized per protocol on PRIMA. Median follow-up in months (IQR) was: LEO=59 (56–60), PRIMA=112 (111–113), RELEVANCE=78 (77–78), and GALLIUM=78 (78–79).

The FLIPI24 model had superior discrimination for EFS24 compared to the FLIPI and PRIMA-PI (Table S6, Figure S5) with calibration varying by maintenance strategies (Table 2) across immunochemotherapy validation datasets. Excluding the expected lower rates of EFS24 from the non-maintenance arm of the PRIMA trial, patients with low-risk FLIPI24 (28%-34% of patients) had EFS24 rates of 88–90%; patients with intermediate-risk FLIPI24 (41%-43% of patients) had EFS24 rates of 80%-88%; patients with high-risk FLIPI24 (25%-32% of patients) had EFS24 rates of 65–78%.

Table 2.

Calibration of FLIPI24 in External Datasets

	Low-Risk FLIPI24	Intermediate-Risk FLIPI24	High-Risk FLIPI24
IC-Treated at 1L
LEO (IC)
N (%)	160 (29%)	226 (40%)	179 (32%)
EFS24 (95%CI)	89% (82–93)	87% (81–91)	67% (59–74)
5-year OS (95%CI)	96% (93–99)	91% (87–93)	77% (70–84)
5-year lymphoma-related death	3.9% (1.6–9.6)	5.5% (3.1–9.9%)	15.6% (10.8–23.0)
GALLIUM
N (%)	281 (30%)	389 (42%)	258 (28%)
R-chemo EFS24 (95%CI)	88% (80–93)	81% (75–86)	63% (54–70)
O-chemo EFS24 (95%CI)	90% (84–94)	88% (82–92)	78% (69–85)
5-year OS (95%CI)	97% (94–98)	91% (88–94)	78% (73–83)
5-year lymphoma-related death	1.8% (0.7–4.0)	3.5% (2.0–5.8)	9.3% (6.1–13)
RELEVANCE (IC Arm)
N (%)	151 (31%)	203 (42%)	129 (27%)
EFS24 (95%CI)	88% (81–92)	84% (78–88)	76% (68–83)
5-year OS (95%CI)	97% (94–99)	92% (88–94)	82% (77–86)
5-year lymphoma-related death	0.7% (0.14–2.4)	4.3% (2.6–6.7)	9.5% (6.4–13)
PRIMA
N (%)	294 (31%)	406 (43%)	242 (23%)
Maintenance arm EFS24	88% (81–92)	82% (75–86)	65% (56–73)
Non-maintenance arm EFS24	73% (65–79)	71% (65–77)	51% (42–59)
5-year OS (95%CI)	97% (94–98)	92% (89–94)	83% (78–88)
5-year lymphoma-related death	1.8% (0.7–3.9)	4.6% (2.8–6.9)	11.1% (7.5–16)
10-year OS (95%CI)	91% (86–94)	83% (78–87)	63% (55–70)
10-year lymphoma-related death	3.8% (1.8–6.9)	7.0% (4.5–10)	16% (11–21)
Non-IC-Treated at 1L
RELEVANCE (R2 Arm)
N (%)	142 (29%)	213 (43%)	135 (28%)
EFS24 (95%CI)	88% (81–92)	80% (74–85)	72% (63–78)
5-year OS (95%CI)	98% (94–99)	92% (87–95)	80% (73–86)
5-year lymphoma-related death	0.72% (0.06–3.6)	3.8% (1.8–7.1)	12% (7.2–18)
Observation or Rituximab Monotherapy at 1L
LEO
N (%)	268 (42%)	244 (38%)	122 (19%)
5-year OS (95%CI)	99% (97–100)	93% (90–97)	73% (64–83)
5-year lymphoma-related death	0.9% (0.2–3.6)	2.8% (1.1–7.0)	10.7% (5.7–20.0)
MER
N (%)	200 (39%)	208 (41%)	103 (20%)
5-year OS (95%CI)	94% (91–98)	94% (90–97)	79% (71–87)
5-year lymphoma-related death	3.1% (1.4–6.8)	2.0% (0.8–5.3)	9.1% (4.8–17.0)
10-year OS (95%CI)	93% (89–96)	79% (73–85)	59% (50–70)
10-year lymphoma-related death	3.7% (1.8–7.6)	8.9% (5.7–14.1)	16.3% (10.4–25.6)
All FL at Diagnosis (Any 1L management)
LEO
N (%)	547 (38%)	558 (39%)	340 (24%)
5-year OS (95%CI)	98% (97–99)	92% (90–95)	75% (70–81)
5-year lymphoma-related death	1.6% (0.8–3.4)	3.9% (2.5–6.2)	13.0% (5.9–17.9)
MER
N (%)	370 (34%)	433 (40%)	271 (25%)
5-year OS (95%CI)	96% (94–98)	91% (88–94)	71% (66–77)
10-year OS (95%CI)	93% (90–96)	79% (75–83)	52% (46–59)
5-year lymphoma-related death	1.9% (0.9–4.1)	4.5% (2.9–7.0)	16.2% (12.3–21.3)
10-year lymphoma-related death	3.2% (1.8–5.7)	10.1% (7.5–13.5)	24.4% (19.6–30.2)

IC = immunochemotherapy, OS = overall survival, CI = confidence interval, EFS24 = event-free survival at 24 months, 1L = first-line therapy, R = rituximab, O=obinutuzumab, R2 = lenalidomide and rituximab

The performance of FLIPI24 was even stronger for the OS endpoint. Respective c-statistics for the continuous FLIPI24 score for OS for LEO IC, PRIMA, RELEVANCE, and GALLIUM (c=0.708,0.679,0.705,0.695) were again superior to FLIPI (c=0.627,0.659,0.661,0.626) and PRIMA-PI (c=0.638,0.594,0.588,0.628) (Table S6) and there was good calibration of 5-year OS across all validation datasets (Table 2). The FLIPI24 identified 3 distinct risk groups for OS (Figure 2) in contrast to the FLIPI and PRIMA-PI risk models where there was significant overlap between risk groups (Figure 3). Patients with low-risk FLIPI24 had 5-year OS rates of 96–97%; patients with intermediate-risk FLIPI24 had 5-year OS rates of 90–92%; patients with high-risk FLIPI24 had 5-year OS rates of 67–83%. The primary cause of death in the high-risk FLIPI24 group was lymphoma in LEO, PRIMA, and RELEVANCE, while death due to non-lymphoma causes was slightly more common in GALLIUM beyond 4 years after diagnosis (Table 2, Figure S6). Notably, FLIPI24 identified a distinct group of patients with a very low mortality that was not identified in other models (Figure 3, Figure S7). Using long-term follow-up available from the PRIMA trial, the 10-year risk of lymphoma-related death for low-risk FLIPI24 was of 3.8% (95%CI: 1.8–6.9).

Figure 2.

Overall survival by FLIPI24 risk groups in the validation and extension datasets.

Figure 3.

Overall survival by FLIPI24 risk groups compared to the FLIPI and PRIMA-PI models in the validation and extension datasets.

Extension of the FLIPI24 model to low-tumor burden management approaches

We next evaluated the FLIPI24 model in patients with FL managed with rituximab monotherapy or observation from the LEO²¹ (N=634) and MER²³ (N=511) cohorts (Table S7). The discrimination of the model for OS in LEO (c=0.777) and MER (c=0.673) was again superior to the FLIPI (c=0.708,c=0.578) and PRIMA-PI (c=0.656,c=0.533) models in both cohorts (Table S6, Table 2, Figure 3).

Application of the FLIPI24 Model to All Newly Diagnosed FL Patients

Finally, we evaluated the FLIPI24 model in all patients with FL regardless of initial therapy from the LEO (N=1,445) and MER (N=1,074) cohorts (Table S7). FLIPI24 risk score was associated with both FL relapse and transformation in the LEO cohort (Table S8). A comparison of FLIPI24 risk scores to the FLIPI and GELF criteria in the LEO cohort is shown in Figure 4 while the distribution of FLIPI24 scores by FLIPI is shown in Figure S8. The discrimination of the model increased for OS in LEO (c=0.784) and MER (c=0.719) which was superior to the FLIPI (c=0.681,c=0.648), PRIMA-PI (c=0.683,c=0.599) models (Table S6; Figure 2, Figure 3). 10-year OS for FLIPI24 in the MER was 93% (95%CI: 90–96) for low-risk, 79% (95%CI: 75–83) for intermediate-risk, and 52% (95%CI: 46–59) for high-risk FLIPI24 (Table 2). When comparing MER survival to the background population (Figure S9), we observed significantly increased mortality in high-risk FLIPI24 (standardized mortality ratio=1.45, 95%CI 1.23–1.69) and to a lesser degree in intermediate-risk (1.31, 95%CI: 1.1–1.56) FLIPI24, while low-risk FLIPI24 had no excess risk (0.68, 95%CI 0.50–0.92) and a 2.3% (95%CI 1.1–4.5) risk of dying of lymphoma at 10 years from diagnosis.

Figure 4.

Comparison of FLIPI and GELF criteria with FLIPI24 in the LEO cohort.

Implications for clinical trial design

A theoretical comparison of a 1L FLIPI24 clinical trial design compared to a GELF-based design can be found in the supplemental material (Table S9).

Online tool

An interactive application for model calculation and interpretation can be found at https://rtools.mayo.edu/flipi24_calculator/.

DISCUSSION

Most patients with FL can be managed well into the second decade after diagnosis using current therapeutic options. However there remains a subset of high-risk patients who retain a significantly elevated risk of dying from transformed or treatment-resistant FL. Accurate risk prediction at diagnosis is critical for optimal management and continued 1L drug development. Here we present FLIPI24, a validated prognostic model and clinical trial enrichment tool for FL using contemporary modeling tools. The model was developed through a rigorous analytical framework that incorporated harmonized individual patient data from contemporary international cohorts. The resulting model has improved discrimination compared to existing models, identifies three distinct subsets of patients with clinically meaningful differences in outcome, and provides a framework to identify which patients can expect their disease to be well-managed with current standard of care approaches and which patients should be prioritized for clinical trials at the time of diagnosis. The model validated in all types of current 1L therapy options, including various immunochemotherapy approaches (bendamustine vs CHOP-based) as well as non-chemo approaches such as R2 and observation.

Early progression, typically defined as POD24, is the current standard most often used for defining high-risk FL.²⁴ However, POD24 has inherent limitations including restriction to those receiving 1L IC, waiting to observe the outcome, the inability to implement novel interventions before the early event, and the rarity of the population. About 40% of patients with FL initiate therapy with immunochemotherapy at diagnosis with 15%-20% expected to have a POD24 event based on recent trials. Further, a significant subset of early events will be transformation to DLBCL or high-grade lymphoma,²⁵ resulting in <5% of patients with FL captured as high-risk via the POD24 paradigm. Strategies using POD24 to define a high-risk patient population have been unsuccessful in part due to these limitations. The FLIPI24 model mitigates the limitations of POD24. It can be applied readily to all patients at diagnosis, prior to starting any treatment, and identifies a significantly larger population of high-risk patients with comparable outcomes to POD24²⁶ at our institutions. The ability of FLIPI24 to predict OS, and in particular lymphoma-related mortality, importantly demonstrates its ability to risk stratify beyond early events in newly diagnosed FL.

Patients with high-risk FLIPI24 comprise about 25% of patients with FL. This group of patients are at the highest risk of mortality in the current treatment era, with 5-year OS of 75% in the LEO Cohort and the highest rates of lymphoma related death (9.3–15.6% at 5 years) across the validation cohorts. More aggressive 1L therapeutic approaches (while being mindful of toxicity, frailty, and meeting individual patient needs) should be considered. This also applies to patients with high risk FLIPI24 without GELF criteria or who would typically be managed with rituximab monotherapy or observation. Notably, inferior outcomes were also observed in recent clinical trials that established the current standards of care, supporting prioritization of enrolling patients with high-risk FLIPI24 on clinical trials that evaluate novel 1L regimens and therapeutic approaches.

Patients with low-risk FLIPI24 comprise about 30% of patients treated with 1L immunochemotherapy and almost 40% of all patients with FL. These patients have exceptionally good outcomes, with 5-year OS of 96–98% in recent series, and 10-year survival >90% in both the MER and PRIMA datasets. These patients can expect very little mortality due to progressive lymphoma during the first decade using current standard of care approaches and should be counseled as such. This is an important reassurance to a significant percentage of patients with newly diagnosed FL. Therapy selection in low-risk FLIPI24 should meet an individual patient’s quality of life goals and clinical trial designs should seek to limit treatment burden while maintaining efficacy. Evaluation of less intensive surveillance strategies for patients with low-risk FLIPI24 is also warranted.²⁷ Outcomes for patients with intermediate-risk FLIPI24 were favorable overall, with 5-year OS of 93% in LEO and 91–92% in recent immunochemotherapy trials; however there remains a non-trivial risk of increased mortality due to FL. Any benefit in lymphoma-related mortality should be balanced against patient needs and the potential impact of any proposed treatments on QOL when designing clinical trials in these patients. Across all FLIPI24 risk groups, additional features of high-risk disease beyond the FLIPI24 model, such as widespread disease involvement, symptoms, and PET findings,²⁸ should be used in informing treatment decisions.

Trial designs in newly diagnosed FL are challenging. Registrational trials typically use GELF-based trial entry criteria; however these are absent in upwards of 50% of patients²⁹ and initiation of systemic therapy is not directly related to GELF criteria in recent observational cohorts.^29,30 Median PFS of >10 years and 10-year OS rates of >80% can be expected with these approaches which necessitates large trials with lengthy follow-up.³¹ Prognostic enrichment approaches can make trials more efficient. The FLIPI24 outperforms both GELF and FLIPI in identifying a high-risk population (25% of all patients with FL) with inferior outcomes on current standard of care. Using high-risk FLIPI24 for eligibility in 1L trials of aggressive therapeutic approaches should lead to smaller sample sizes and/or quicker trial read outs.

The FLIPI24 has clearly superior performance to the FLIPI, including distinct separation of outcomes in risk groups across all datasets and endpoints. This includes the gold-standard endpoint of OS and clinically significant death due to lymphoma. Despite its age, FLIPI remains the most featured prognostic model in clinical practice. However, it does not sufficiently risk predict for treatment selection or clinical trial enrichment and is poorly calibrated in the current treatment era.^32,33 Newer models developed in the immunochemotherapy era, including the FLIPI2,¹¹ M7-FLIPI,¹³ PRIMA-PI,¹² and FLEX³⁴ models, have been proposed. However, these models were developed on smaller series (ranging from 151–1135 patients) and lack robust validation in the current treatment era (including bendamustine). The FLIPI24 study represents the largest effort of its kind in FL with individual patient data from over 10,000 patients included across discovery and validation. FLIP24 validation was performed across the continuum of current 1L management, including bendamustine-based approaches as well as non-chemo approaches. The blood-based variables utilized in the FLIPI24 model are part of routine workup of FL and featured in other prognostic models in lymphoma^35,36 and could thus be easily implemented in most clinical settings. Further, the FLIPI24 does not necessitate a bone marrow biopsy^37–39 (PRIMA-PI/FLIPI2) or the cumbersome counting of nodal groups (FLIPI). The strength of B2M in the model supports standard assessment of this biomarker at diagnosis for all patients with FL.

Strengths of the study include the multi-center and international pooled cohort of patients used for modeling as well as robust validation of the model in multiple studies including observational cohorts and clinical trials. The modeling of early events to identify long-term outcomes is novel in this setting but builds on a strong base of clinical literature in FL.^5,8,40,41 The model also has limitations. While low-risk patients have very low rates of mortality in the first decade after diagnosis, this does not necessarily translate to cure as upwards of 30% of patients had progression or retreatment events within 6 years of initiating immunochemotherapy in the clinical trial validation datasets and longer-term data are yet to mature. The model also does not provide information regarding the risk of relapsed FL versus transformation nor inform which chemotherapy backbone is preferable for a given patient. Finally, the model is developed based on CD20 antibody based 1L IC; re-validation will be needed as novel agents are implemented into the 1L setting. In addition, the prediction of OS is sensitive to development of more effective therapies for relapsed/refractory FL.

In summary, the FLIPI24 provides improved risk stratification over existing clinical models in newly diagnosed FL. The FLIPI24 model identifies patients at an increased risk for dying from lymphoma as well as a significant percentage of patients with almost no lymphoma-related mortality during the first decade after diagnosis. FLIPI24 will allow future clinical trial designs to better balance potential benefit vs. risk in patients with follicular lymphoma.

Context Summary.

Key Objective

Using individual patient data from 10 international observational cohorts, this study describes the development of the FLIPI24 clinical prediction model for newly diagnosed FL. Performance of the FLIPI24 model is then evaluated on recent large clinical trials and observational cohorts.

Knowledge Generated

The FLIPI24 model was able to risk stratify patients with newly diagnosed follicular lymphoma more accurately than previous models. The model maintained its accuracy across clinical trials and observational data, as well as in all types of first-line management approaches.

Relevance (written by Suzanne Lentzsch)

The FLIPI24 model identifies patients with FL at increased risk of lymphoma-related death vs. patients with very low lymphoma-related mortality during the first decade after diagnosis. Its adoption may reduce overtreatment in low-risk patients and better identify candidates for early intervention strategies.

Data Sharing

These data are not publicly available. Data sharing policies and the process to request the LEO Cohort data that support the findings of this study may be found on the LEO Cohort website: leocohort.org.