Factors Affecting the Clinical Course of Follicular Lymphoma: A Multistate Survival Analysis Using Individual Patient Data from Eight Multicenter Randomized Clinical Trials

Abstract

Background:

Leveraging the Follicular Lymphoma Analysis of Surrogacy Hypothesis database of individual patient data from first-line clinical trials, we studied the clinical course of follicular lymphoma (FL) and investigated clinical factors associated with FL outcomes.

Methods:

We examined 2,428 patients from 8 randomized trials using multistate survival models with four states: induction treatment, progression, death from FL, and death from other causes. We utilized Aalen-Johansen estimator and Cox models to assess the likelihood of FL outcomes and quantify predictors’ effects.

Findings:

Two-year progression, FL-related death, and death from other causes estimates were 26.5%, 3.4% and 1.4%, respectively. FL-associated deaths were the primary cause of mortality within ten years of follow-up. Male sex (hazard ratio: 1.25; 95% confidence interval: 1.05–1.47), > 4 involved nodal areas (1.51; 1.23–1.86), elevated LDH (1.20; 1.01–1.43), low hemoglobin (1.44; 1.15–1.81), and elevated β-2 levels (1.23; 1.02–1.47) increased risk of progression. CD20-targeting agents reduced risks for progression (0.29; 0.22–0.39), death from FL (0.05; 0.01–0.20), and death from other causes without progression (0.13; 0.05–0.33) and following progression (0.52; 0.30–0.92). Estimated two-year progression rates were 22.3% and 43.5% with or without CD20-targeting agents, respectively. Two-year FL-associated mortality rate was 8.3% without CD20-targeting agents, 5.4% with B-symptoms, 4.9% with elevated LDH, and 9.1% low hemoglobin.

Interpretation:

This study identified independent contributions of baseline clinical factors to distinct outcomes for patients with FL following first-line therapy on a clinical trial. Similar analytical approaches are needed to increase understanding of factors that influence FL outcomes in other settings.

INTRODUCTION

Follicular lymphoma (FL) is the most common indolent lymphoma in Western countries with approximately 14,000 new cases estimated in the US per year.¹ The incidence of FL increases dramatically with age with a median age at diagnosis in the early 60’s ^2,3. Currently, FL is considered an incurable disease where the course of treatment for newly diagnosed FL depends mainly on stage and tumor burden⁴. In current practice, available first-line therapies range from less aggressive treatments, such as observation (i.e., watchful waiting) or rituximab (R) monotherapy, to more aggressive approaches, such as combination of anti-CD20 monoclonal antibody and chemotherapy (e.g., CHOP or bendamustine) with or without 2-years of maintenance treatment^5–9. However, even with a favorable response to initial therapy most patients experience recurrence in their lifetime and require multiple lines of treatments¹⁰. The duration of response reduces after initial therapy, and continues to shorten by second line and subsequent treatment, which is, on average, less than one year after the third line of treatment¹¹.

Despite recent improvements in response rates and progression-free survival (PFS), the primary cause of death for patients with FL remains lymphoma or the adverse events associated with FL therapy¹². Yet, prolonged disease-free survival is still achieved by a considerable proportion of patients with FL, who face competing risks for death based on other disease comorbidities and age. Studies have shown that approximately 20% of patients experience progression of disease within 24 months (POD24) following first line chemoimmunotherapy and have poor overall survival (OS)^13–15. POD24^7,8,15,16 was recently validated as a significant prognostic factor with individual patient data from 5,225 previously untreated patients with FL enrolled in 13 multicenter randomized controlled trials¹⁷. However, the factors affecting FL outcomes, such as disease progression or cause-specific survival, and their impacts on the clinical course of FL have not been fully explored. There are few studies that address how much well-known baseline prognostic factors, such as age, performance status (PS), and other candidate predictors contribute to the risks of progression, lymphoma-specific survival, and death from other causes.

Moreover, nearly all existing studies of FL have solely examined binary or time-to-event survival outcomes, such as POD24, PFS, or OS, in isolation without explicitly considering the various competing events that can occur during or following initial treatment of FL. To help close this gap in the literature, we performed a pooled analysis from the Follicular Lymphoma Analysis of Surrogacy Hypothesis (FLASH) database using individual patient data from randomized controlled trials of first-line therapies for patients with FL. Our objectives were to (i) study the clinical course of patients with FL as they move through the stages of first line treatment, disease progression, and death due to lymphoma or from other causes, (ii) identify the key clinical factors measured at the time of initial treatment associated with these FL events and better or worse outcomes, and (iii) quantify the impacts of the identified predictors on the clinical course of FL.

PATIENTS AND METHODS

Study Population

The population we examined in this study consisted of 2,428 patients from the FLASH database. The FLASH meta-database was originally created through a partnership of industry and academic institutions with the primary goal of establishing complete response at 30 months (CR30) as a surrogate endpoint in first-line FL clinical trials¹⁸. It contains individual patient data from randomized controlled trials of first-line FL therapies that investigated a variety of FL treatments including R monotherapy, chemo-immunotherapy, and chemotherapy alone (Table 1). Patients were not included in this analysis if they were previously treated, had histology other than FL, did not begin protocol-defined treatment, or had no clinical follow-up data available.

Table 1:

Trial Summary

Study Name	Study ID (CT.gov)	Treatment Regimens	Number of Patients (N=2,428)	Number of Treatments Including CD20-targeting Agent (N=1,928)
FIT40	NCT00185393	First line treatment followed by: R+90Y-IT consolidation vs. no further treatment	414 (17.1%)	239 (12.4%)
M3902141,42	Not Available	R-CVP vs. CVP	321 (13.2%)	162 (8.4%)
OSHO3943	NCT00269113	R-MCP vs. MCP	199 (8.2%)	106 (5.5%)
ML1686544	NCT01609010.	R vs. R+IFN	226 (9.3%)	226 (11.7%)
OSHO1945	Not Available	BOP vs. COP	73 (3.0%)	0 (0.0%)
PRIMA8	NCT00140582	R-CHOP, R-CVP, R-FCM followed by: R maintenance vs. observation	1018 (41.9%)	1018 (52.8%)
SAKK359846,47	NCT00003280	R 4 weeks vs. R 4 weeks → R × 4 doses every 2 months	51 (2.1%)	51 (2.6%)
CALGB5090448	NCT01286272	Ofatumumab + Bendamustine vs. Ofatumumab + Bendamustine and Bortezomib	126 (5.2%)	126 (6.5%)

Model Description

Our approach employed a multistate survival model to capture the clinical course of FL over time and to assess the impact of common clinical predictors on FL outcomes for patients enrolled in first-line clinical trials. The multistate model we developed for this analysis consisted of four model states: first-line therapy induction treatment, progression, death due to FL, and death from other causes. For all patients, this model defined a starting state of being alive immediately after clinical trial enrollment and initiation of induction treatment. That is, each patient began the model with time zero in the first-line treatment state ready for the initiation of induction therapy, following a patient’s registration (or randomization) date in first-line clinical trial. After the induction therapy, patients could move to other three model states from the first-line treatment state if they experience disease progression or (cause-specific) death. From the progression state, patients could move to death due to FL or death from other causes. Death from FL and death from other causes were both absorbing model states, meaning patients could not transition out of them (Figure 1). Patients who are alive at last follow-up without transition to any of these states remained in the initial state and were censored at last follow-up.

Figure 1:

State Diagram

Statistical Methods

Clinical factors of interest were age at diagnosis, gender, Ann-Arbor stage, B symptoms, nodal area involvement, bone marrow involvement, lactate dehydrogenase (LDH) status, hemoglobin level, β2-microglobulin status, CR30 status, CD20-targeting agent use, FL International Prognostic Index (FLIPI)¹⁹ and FLIPI-2²⁰ score. The levels for these factors and their proportions were summarized for both the entire patient population, and the subset of patients who received CD20-targeting agents during first-line treatment. State likelihood plots were created using Aalen-Johansen (AJ) estimators²¹ to calculate the likelihood of being in each model state at a given time and reported with 95% confidence intervals (CIs). The AJ estimator is a generalization of the nonparametric Kaplan-Meier estimator to multistate models that estimates the continuous time transition probability matrix of a finite-state Markov process such as the one modeled in this study with a multistate survival model consisting of four model states. Cox proportional hazards regression models were employed to identify the clinical factors having a statistically significant effect on the prognostic outlook of patients. A separate Cox regression model was fit for each possible transition in the multistate model to separately quantify the influence of predictors on each FL outcome, modeled by the multistate model. Hazard ratios (HRs) and 95% CIs were reported to define the effect each prognostic factor had on the FL events, captured by the transitions in our model. Analyses were conducted using the statistical computing software R version 3.6 ²² and the “survival” package²³.

Subpopulation and Sensitivity Analysis

Utilizing the four-state survival model, the clinical course of FL over time was studied for several subpopulations. In particular, the AJ estimators were compared between subpopulations as defined by age (<40 vs. 40–60 vs. 60–70 vs. 70+), sex (male vs. female), Ann-Arbor stage (I-III vs. IV), use of CD20-targeting agents (no vs. yes), B symptoms (no vs. yes), number of involved nodal areas (0–4 vs. >4), bone marrow involvement (no vs. yes), LDH (normal vs. elevated), β2-microglobulin (normal vs. elevated), and hemoglobin (≤ 12g/dL [low] vs. > 12g/dL). A sensitivity analysis was conducted with an alternative multistate model that consists of only three model states: first-line treatment, progression, and death from all causes. The results were compared the results of the original four-state model (Figure 1) that accounts for the cause of death.

RESULTS

Patient Characteristics

A total of 2,428 patients enrolled in 8 randomized multicenter international trials were identified from the FLASH database to be included in this analysis. Of these patients, 79.4% (n=1,928) received CD20-targeting agents during first line treatment. Individual treatment regimens used in these trials are listed in Table 1. Baseline characteristics are shown in Table 2 for all patients (N=2,428) and for the subset of patients who received CD20-targeting agents with or without chemotherapy (N=1,928). Approximately half of the patients were age 40–60 years (N=1,272 patients; 52.4%) and more than one-fourth of the patients were age 60–70 years (N=657; 27.1%), with the median age at diagnosis of 56 (range: 22–87). Among 2,344 patients with a calculated FLIPI, 994 patients (42.4%) were at high-risk. Only 423 patients had a calculated FLIPI-2 score and 238 of these patients (56.3%) were high-risk. A total of 770 patients (37.5%) achieved CR30 following the initiation of first line treatment; the rate of CR30 was higher (43.3%) in patients who received CD20-targeting agents.

Table 2:

Patient Baseline Characteristics

Clinical Factors	All Patients (N=2,428)	Treatment Including CD20-targeting Agent (N=1,928)
Age
< 40	244 (10.0%)	200 (10.4%)
40–60	1272 (52.4%)	969 (50.3%)
60–70	657 (27.1%)	545 (28.3%)
≥70	255 (10.5%)	214 (11.1%)
Gender
Female	1199 (49.4%)	945 (49.0%)
Male	1229 (50.6%)	983 (51.0%)
Stage
Missing	3	3
I-III	776 (32.0%)	628 (32.6%)
IV	1649 (68.0%)	1297 (67.4%)
FLIPI Category
Missing	85	64
Low	535 (22.8%)	402 (21.6%)
Intermediate	814 (34.7%)	650 (34.9%)
High	994 (42.4%)	812 (43.6%)
FLIPI 2 Category
Missing	2005	1655
Low	1 (0.2%)	0 (0.0%)
Intermediate	184 (43.5%)	110 (40.3%)
High	238 (56.3%)	163 (59.7%)
B Symptoms
Missing	374	281
No	1418 (69.0%)	1118 (67.9%)
Yes	636 (31.0%)	529 (32.1%)
Nodal Areas Involved
Missing	381	283
0–4	520 (25.4%)	465 (28.3%)
>4	1527 (74.6%)	1180 (71.7%)
Bone Marrow Involvement
Missing	701	432
No	705 (40.8%)	647 (43.2%)
Yes	1022 (59.2%)	849 (56.8%)
LDH Status
Missing	519	345
Normal	1392 (72.9%)	1124 (71.0%)
Elevated	517 (27.1%)	459 (29.0%)
Hemoglobin Status
Missing	431	241
Normal	1516 (75.9%)	1277 (75.7%)
Low	481 (24.1%)	410 (24.3%)
Beta 2 Status
Missing	1078	716
Normal	621 (46.0%)	548 (45.2%)
Elevated	729 (54.0%)	664 (54.8%)
CR 30 Months
Missing	373	269
Non-CR	1285 (62.5%)	941 (56.7%)
CR	770 (37.5%)	718 (43.3%)

Aalen Johansen (AJ) Estimates

The AJ estimates for the clinical course of FL were calculated for the entire FLASH patient population (N=2,428) and the subpopulations defined by the clinical characteristics described above. The state likelihood graph (Figure 2) illustrates the clinical course of FL over time by plotting the likelihood of a patient being in each clinical state at a given time with a median follow-up time of 8.0 years. Table 3 shows the AJ estimates for the cumulative probability of experiencing disease progression or cause-specific death by a certain time point. Calculated both for the subpopulations and the entire patient population, progression is reported at year 1 and 2, and cause-specific deaths are reported at 2 and 5 years.

Figure 2:

State Likelihood Plot

Table 3:

Aalen Johansen Estimates for All Patients and Subgroups

Clinical Factors	Levels	Progression of Disease		Death Due to FL		Death Due to Other
		1-year	2-year	2-year	5-year	2-year	5-year
All Patients		12.3% (11.1%, 13.7%)	26.5% (24.8%, 28.4%)	3.4% (2.8%, 4.3%)	7.9% (6.9%, 9.1%)	1.4% (1.0%, 1.9%)	5.1% (4.2%, 6.1%)
Age	< 40	12.2% (8.6%, 17.1%)	28.1% (22.9%, 34.4%)	0.5% (0.1%, 3.6%)	2.3% (1.0%, 5.6%)	0.5% (0.1%, 3.6%)	1.4% (0.5%, 4.4%)
	40–60	12.9% (11.2%, 14.9%)	29.4% (27.0%, 32.1%)	2.4% (1.7%, 3.5%)	6.8% (5.5%, 8.5%)	1.2% (0.7%, 2.0%)	4.1% (3.1%, 5.4%)
	60–70	12.1% (9.8%, 14.9%)	22.1% (19.1%, 25.6%)	4.5% (3.1%, 6.5%)	9.1% (7.1%, 11.7%)	1.9% (1.1%, 3.3%)	6.9% (5.1%, 9.3%)
	≥70	11.3% (8.0%, 16.0%)	22.8% (18.2%, 28.6%)	9.0% (6.1%, 13.3%)	16.7% (12.5%, 22.2%)	1.6% (0.6%, 4.2%)	9.0% (6.0%, 13.5%)
Gender	Female	11.4% (9.7%, 13.4%)	25.4% (23.0%, 28.0%)	3.4% (2.5%, 4.7%)	7.7% (6.3%, 9.5%)	0.7% (0.3%, 1.4%)	4.4% (3.3%, 5.8%)
	Male	13.4% (11.6%, 15.5%)	27.8% (25.3%, 30.4%)	3.6% (2.7%, 4.8%)	8.4% (6.9%, 10.2%)	2.0% (1.4%, 3.0%)	5.7% (4.5%, 7.3%)
Stage	I-III	10.6% (8.6%, 13.0%)	23.2% (20.4%, 26.4%)	2.5% (1.6%, 3.9%)	5.5% (4.0%, 7.5%)	0.9% (0.5%, 2.0%)	3.5% (2.4%, 5.2%)
	IV	13.2% (11.7%, 15.0%)	28.2% (26.1%, 30.5%)	4.0% (3.1%, 5.1%)	9.3% (7.9%, 10.9%)	1.6% (1.1%, 2.3%)	5.8% (4.7%, 7.1%)
CD20-Targeting Agent Included	No	26.8% (23.2%, 31.1%)	43.5% (39.2%, 48.2%)	8.3% (6.1%, 11.2%)	16.9% (13.8%, 20.9%)	3.1% (1.9%, 5.1%)	8.2% (6.0%, 11.2%)
	Yes	8.8% (7.6%, 10.1%)	22.3% (20.5%, 24.3%)	2.3% (1.7%, 3.1%)	5.8% (4.8%, 7.0%)	0.9% (0.6%, 1.5%)	4.3% (3.4%, 5.3%)
B Symptoms	No	11.1% (9.6%, 12.9%)	26.0% (23.8%, 28.4%)	1.9% (1.3%, 2.8%)	5.9% (4.8%, 7.4%)	0.9% (0.5%, 1.6%)	4.1% (3.1%, 5.3%)
	Yes	16.0% (13.4%, 19.1%)	28.0% (24.7%, 31.8%)	5.4% (3.9%, 7.5%)	11.1% (8.9%, 13.9%)	1.8% (1.0%, 3.1%)	5.9% (4.3%, 8.1%)
Number of Nodal Areas	0–4	8.1% (6.1%, 10.9%)	17.1% (14.1%, 20.7%)	2.9% (1.8%, 4.8%)	6.1% (4.3%, 8.6%)	0.6% (0.2%, 1.8%)	3.7% (2.3%, 5.8%)
	>4	14.8% (13.1%, 16.7%)	31.9% (29.7%, 34.4%)	2.9% (2.2%, 3.9%)	7.6% (6.4%, 9.1%)	1.1% (0.7%, 1.7%)	4.7% (3.7%, 5.9%)
Bone Marrow Involvement	No	9.4% (7.5%, 11.9%)	20.6% (17.8%, 23.9%)	3.2% (2.1%, 4.8%)	5.9% (4.4%, 8.0%)	1.2% (0.6%, 2.3%)	3.2% (2.1%, 5.0%)
	Yes	14.0% (12.0%, 16.3%)	28.3% (25.7%, 31.2%)	3.6% (2.6%, 5.0%)	9.8% (8.1%, 11.9%)	1.4% (0.8%, 2.3%)	5.8% (4.5%, 7.5%)
LDH Status	Normal	11.3% (9.7%, 13.1%)	26.8% (24.5%, 29.2%)	1.3% (0.8%, 2.1%)	4.9% (3.9%, 6.2%)	0.6% (0.3%, 1.2%)	3.9% (3.0%, 5.1%)
	Elevated	13.8% (11.1%, 17.1%)	26.7% (23.1%, 30.9%)	4.9% (3.3%, 7.3%)	10.9% (8.5%, 14.1%)	1.6% (0.8%, 3.3%)	5.7% (4.0%, 8.2%)
Hemoglobin Status	Normal	10.5% (9.1%, 12.2%)	24.6% (22.5%, 26.9%)	2.4% (1.7%, 3.4%)	7.2% (5.9%, 8.7%)	0.7% (0.4%, 1.4%)	4.1% (3.2%, 5.3%)
	Low	15.8% (12.8%, 19.5%)	23.6% (20.0%, 27.8%)	9.1% (6.8%, 12.1%)	14.6% (11.6%, 18.4%)	4.1% (2.6%, 6.4%)	9.5% (7.0%, 12.9%)
Beta 2 Status	Normal	7.9% (6.0%, 10.4%)	20.4% (17.5%, 23.9%)	1.5% (0.8%, 2.8%)	5.3% (3.8%, 7.5%)	0.2% (0.0%, 1.1%)	2.6% (1.6%, 4.2%)
	Elevated	12.4% (10.2%, 15.1%)	25.3% (22.3%, 28.7%)	3.4% (2.3%, 5.0%)	8.7% (6.8%, 11.1%)	2.0% (1.2%, 3.3%)	6.5% (4.9%, 8.7%)

For the entire population, 2- and 5-year mortality rates were respectively 3.4% and 7.9% due to FL, and 1.4% and 5.1% from other causes of death, whereas the cumulative progression rates were 12.3% and 26.5% by year 1 and 2. The AJ estimates for different subsets of patients indicated that patients who had FL stage I-III (as opposed to stage IV), who were younger or female, received CD20-targeting agent during first-line treatment, did not have B-symptoms, had ≤ 4 involved nodal areas, no bone marrow involvement, and had normal LDH, hemoglobin, and β-2 microglobulin levels had better FL outcomes than their counterparts (Table 3). Notably, the 1 and 2-year progression estimates were similar among the subgroups with the exception of the use of CD20-targeting agents in first-line treatment [1-year: No CD20-agent 26.8% (23.2%, 31.1%), with CD20-agent 8.8% (7.6%, 10.1%); 2-year: No CD20-agent 43.5% (39.2%, 48.2%), with CD20-agent 22.3% (20.5%, 24.3%)] and number of involved nodal areas [1-year: ≤4 areas 8.1% (6.1%, 10.9%), >4 areas 14.8% (13.1%, 16.7%); 2-year: ≤4 areas 17.1% (14.1%, 20.7%), >4 areas 31.9% (29.7%, 34.4%)]. Also, the subgroups of patients with elevated LDH [26.7% (23.1%, 30.9%)] and low hemoglobin [23.6% (20.0%, 27.8%)] had similar 2-year PFS estimates as patients in the normal LDH [26.8% (24.5%, 29.2%)] and normal hemoglobin [24.6% (22.5%, 26.9%)] subgroups, but patients with elevated LDH [10.9% (8.5%, 14.1%)] and low hemoglobin [14.6% (11.6%, 18.4%)] levels had higher 5-year rates of FL death than patients with normal LDH [4.9% (3.9%, 6.2%)] and normal hemoglobin [7.2% (5.9%, 8.7%)].

Cox Multivariable Regression Models

A separate multivariable Cox model was developed for each transition captured by the multistate survival model (Figure 1). The HRs for each prognostic factor estimated by the Cox models were reported along with 95% CIs in Table 4.

Table 4:

Multivariable Cox Models

Clinical Factors	Levels	Induction Tx → Progression	Induction Tx → Death FL	Induction Tx → Death Other	Progression → Death FL	Progression → Death Other
Age	< 40	ref	*	*	ref	ref
	40–60	0.97 (0.73, 1.29)	*	*	2.86 (0.96, 8.50)	1.54 (0.60, 3.93)
	60–70	0.97 (0.71, 1.32)	*	*	4.82 (1.60, 14.57)	3.75 (1.45, 9.74)
	≥70	0.97 (0.68, 1.39)	*	*	8.97 (2.86, 28.15)	4.70 (1.63, 13.51)
Gender	Female	ref	ref	ref	ref	ref
	Male	1.25 (1.05, 1.47)	3.93 (0.69, 22.58)	1.68 (0.79, 3.55)	1.13 (0.77, 1.65)	1.62 (1.03, 2.55)
Stage	I-III	ref	ref	ref	ref	ref
	IV	1.16 (0.86, 1.57)	0.59 (0.09, 3.84)	1.80 (0.67, 4.87)	1.32 (0.70, 2.49)	1.02 (0.45, 2.29)
Included CD20-targeting Agent	No	ref	ref	ref	ref	ref
	Yes	0.29 (0.22, 0.39)	0.05 (0.01, 0.20)	0.13 (0.05, 0.33)	0.85 (0.51, 1.42)	0.57 (0.31, 1.04)
B Symptoms	No	ref	ref	ref	ref	ref
	Yes	1.10 (0.93, 1.31)	2.31 (0.47, 11.47)	0.62 (0.28, 1.40)	1.03 (0.69, 1.52)	1.11 (0.70, 1.77)
Number of Nodal Areas	0–4	ref	*	ref	ref	ref
	>4	1.51 (1.23, 1.86)	*	0.87 (0.40, 1.89)	1.30 (0.75, 2.24)	0.84 (0.47, 1.49)
Bone Marrow Involvement	No	ref	*	ref	ref	ref
	Yes	1.13 (0.86, 1.48)	*	0.62 (0.24, 1.63)	0.71 (0.40, 1.27)	1.14 (0.55, 2.37)
LDH	Normal	ref	ref	ref	ref	ref
	Elevated	1.20 (1.01, 1.43)	0.94 (0.17, 5.36)	1.21 (0.55, 2.64)	1.69 (1.15, 2.49)	1.68 (1.08, 2.61)
Hemoglobin	Normal	ref	ref	ref	ref	ref
	Low	1.44 (1.15, 1.81)	7.59 (0.95, 60.88)	2.53 (1.07, 5.95)	1.17 (0.77, 1.79)	1.41 (0.84, 2.36)
Beta-2	Normal	ref	ref	ref	ref	ref
	Elevated	1.23 (1.02, 1.47)	2.42 (0.30, 19.36)	2.21 (1.12, 4.36)	1.14 (0.75, 1.75)	1.28 (0.78, 2.12)
Timing of Progression	<1 year	*	*	*	ref	ref
	1–2 years	*	*	*	0.51 (0.32, 0.83)	0.69 (0.37, 1.29)
	2–5 years	*	*	*	0.40 (0.23, 0.70)	0.85 (0.46, 1.59)
	5+ years	*	*	*	0.69 (0.22, 2.12)	0.55 (0.16, 1.84)

^*removed from transition model due to convergence issues

Following the initiation of first-line treatment, patients had a higher risk of disease progression if they were male (HR: 1.25; 95% CI: 1.05–1.47), had > 4 involved nodal areas (1.51; 1.23–1.86), had elevated LDH levels (1.20; 1.01–1.43), had low hemoglobin levels (1.44; 1.15–1.81), and had elevated β-2 microglobulin levels (1.23; 1.02–1.47) and a lower risk of disease progression if they received a CD20-targeting agent during first-line treatment. The risk of other-cause associated deaths was increased for patients with low hemoglobin levels (2.53; 1.07–5.95) and elevated β-2 microglobulin levels (2.21; 1.12–4.36), and patients who received a CD20-targeting agent were at a lower risk of death due to both lymphoma (0.05, 0.01–0.20) and other causes (0.13, 0.05–0.20).

After patients experienced progression, patients in the two oldest age groups were at increased risk of death due to both lymphoma (age 60–70: 4.82, 1.60–14.57; age ≥70: 8.97, 2.86–28.15) and other causes (age 60–70: 3.75, 1.45–9.74; age ≥70: 4.70, 1.63–13.51). Patients with elevated LDH were also at higher risk of death from any cause (lymphoma: 1.69, 1.15–2.49; other: 1.68, 1.08–2.61), and male patients were at increased risk of non-lymphoma death (1.62, 1.03–2.55). Compared with patients who had progression within a year of beginning first-line treatment, patients with progression 1–2 years (0.51, 0.32–0.83) and 3–5 years (0.40, 0.23–0.70) after beginning first-line treatment were at a decreased risk of death due to FL, but no statistically significant effect was seen between timing of progression and death from other causes.

DISCUSSION

The National LymphoCare Study (NLCS) showed that there has been no single standard of care for the first line treatment of FL in the US²⁴. Among 2,728 subjects enrolled at 265 sites, the initial strategy was chemotherapy in 2%, clinical trial in 6%, and radiation therapy in 6%, R monotherapy in 14%, observation in 18%, and R combined chemoimmunotherapy in 52%. There has been major progress in treatment of FL since NLCS, however, R monotherapy or R combined chemoimmunotherapy are still standard treatment options for first line treatment across countries^25,26. Long-term follow-up of prospective randomized trials involving patients with advanced-stage FL who had high tumor-burden disease demonstrated that standard chemoimmunotherapy produce median PFS of up to 10 years particularly with maintenance treatment ^27–29, and studies suggest that the median OS of patients with FL has improved from approximately six years to more than 15–20 years ^30,31.

With this improvement in survival outcomes, patients with FL go through different clinical status from active treatment to remission multiple times during the disease course and have heterogeneous outcomes. Also, patients with FL are commonly diagnosed at an advanced age where competing comorbid diseases may limit treatment options and produce competing risks for causes of death, and therefore it is important to understand what factors are associated with outcomes in each clinical stage of patients. It has been shown previously that gender, first-line rituximab use, number of nodal areas, LDH, HGB, and β-2 levels are predictors of PFS and OS.^19,20,32 Our findings take that one step further and show which clinical factors are predictors for progression and which are predictors for death both before and after progression and distinguish the influence of clinical factors on death from FL or other causes. In the multivariable Cox models to determine the impact of baseline clinical factors on each transition, sex, first-line CD20-targeted agent use, number of involved nodal areas, LDH levels, hemoglobin levels, and β-2 levels were found to be associated with higher risk of progression. However, many of these same factors were no longer statistically significant predictors of the transitions to death after patients experienced disease progression. In addition to their effects on progression, we found that elevated LDH levels increased risk of death due to both FL and other causes and first line CD20-targeting agents decreased risk of death due to other cause after progression. This helps to disentangle the role of baseline factors on downstream clinical events in FL and further highlights the need for more research into prognostic factors associated with OS after progression.

The FLIPI, FLIPI-2, and other clinical and molecular prognostic models have been developed to predict survival for FL patients and help guide treatment decisions^{19,20,33–35}. However, limited data exist regarding the role of FLIPI and other baseline prognostic factors with regard to the distinct risks of progression, death from FL, and death from other cause following first-line treatment. In analysis of the NLCS data involving 2,192 patients with FLIPI data (35% good risk, 30% intermediate risk, and 35% poor risk), FLIPI risk groups were significant predictors of OS and PFS for patients following observation, and initial therapy with non-R-containing regimens, R monotherapy, and R-chemotherapy combinations³⁶. More recently, the International FLIPI24 Consortium showed that FLIPI was strongly associated with both event free survival (c-statistic=0.61) and OS (c-statistic=0.65) in an observational cohort of >6,000 patients³⁷. The FLIPI-2 ²⁰, M7-FLIPI ³⁸, and the recent Follicular Lymphoma Evaluation Index (FLEX) ³³ risk scores have been designed to improve prediction of PFS and early treatment failure after frontline chemoimmunotherapy. However, none of these analyses have described the contributions of risk factors to the independent outcomes of progression, death from FL, and death from other causes. Such data are needed to distinguish among the available treatment approaches and optimize approaches for patient groups and individuals with the available treatment options. Our findings highlight the need for multistate models to distinguish the role of factors in predicting the separate events of progression and death.

Limitations of the current analysis include limited baseline characteristics data available in the FLASH database that were captured in the original randomized clinical trials and harmonized across trials and a lack of cause-of-death data available from several studies which were not included in this analysis. The actual state diagram used in the model is also a limitation. A more informative multistate model would have included separate states for first-line treatment and the subsequent treatments received³⁹, but such a model was not possible due to data limitations of the FLASH database that only included data on first-line regimen. Finally, only few patients (n=15) experienced FL-associated death directly after their induction treatment, and hence, we did not have sufficient data to thoroughly examine this transition with a multivariable Cox model. Despite these limitations, the analysis provided a means to disentangle the transition states that occur after first-line therapy in a well characterized cohort of patients drawn from clinical trials.

In summary, we conducted a multistate survival analysis utilizing the largest existing dataset from first-line randomized trials in FL and quantified the impact of clinical factors on early, intermediate, and late mortality by cause of death. We found that death from FL surpassed other causes of death up to 10 years following treatment. This was recently confirmed in an observational study involving both US and French cohorts¹². The current work extends these findings using data from randomized controlled trials with protocol defined ascertainment of cause of death and a multi-state statistical model that determined the independent contributions of baseline clinical factors to death following progression from FL and other causes. We showed that known prognostic factors have different impacts on outcomes across clinical disease states. Together these data suggest that the independent effects of these baseline on various intermediate and survival outcomes should be considered in planning future first-line interventions for FL. Similar approaches in additional clinical trials and observational datasets are needed to gain complete understanding of the influences of baseline clinical and biological factors on FL outcomes.