Real-world treatment patterns of patients with large B-cell lymphoma over time and into a post–CAR T approval era

Key Points

• •Real-world patients died within 12 months of initiating 1L therapy, including many patients who died within 3 months.
• •A minority of potentially eligible and fit patients received CAR Ts in the second or 3L setting, suggesting barriers to treatment.

Visual Abstract

Abstract

A significant proportion of patients with large B-cell lymphoma (LBCL) experience relapsed or refractory (R/R) disease after first-line (1L) therapy. Chimeric antigen receptor T-cell (CAR T) therapy is approved for R/R LBCL, yet real-world use remains unclear. This retrospective study evaluated treatment patterns and survival outcomes in patients potentially eligible for CAR T in the Flatiron Health Research Database. Patients diagnosed with LBCL from 2011 to 2024 who initiated 1L therapy were assessed. Predictors of mortality before the opportunity to initiate second-line (2L) therapy were evaluated using competing risk regression models. Treatment patterns among subgroups that initiated 2L and third-line (3L) therapy based timing of US Food and Drug Administration indication approvals for CAR T were evaluated and stratified by CAR T fitness using age and Eastern Cooperative Oncology Group performance status (ECOG PS) score. Among 10 016 patients who initiated 1L therapy, 13% died within 1 year without initiating 2L therapy, and 30% initiated 2L therapy. In competing risk analyses, 11% of patients with an ECOG PS score of 0/1/unknown (combined) died within 12 months of initiating 1L therapy before being able to receive 2L therapy. Among patients deemed eligible and fit for CAR T therapy, 25% received 2L CAR T therapy, and 36% received 3L CAR T therapy. Despite its curative potential, CAR T uptake remains low in potentially eligible patients, with many dying within 1 year of initiating 1L therapy and before the opportunity to receive CAR Ts. Early collaboration between the community oncologist and CAR T center, improved access, and expanded patient awareness strategies are needed to improve CAR T therapy uptake.

Introduction

Large B-cell lymphoma (LBCL) is the most prevalent subtype of non-Hodgkin lymphoma.¹ Historically, treatment options for LBCL have centered around chemotherapy, radiation, and immunotherapy, and although these treatment options have been effective as up-front therapies, approximately half of the patients experience relapse within 5 years of their initial treatment.² Although 85% of the patients with advanced-stage disease have an initial complete response, 15% of the patients either have refractory disease and do not respond to initial treatment or experience relapse early within 6 months after completing first-line (1L) R-CHOP (rituximab, cyclophosphamide, doxorubicin, Oncovin [vincristine], and prednisone) therapy,³ 70% of whom will die from their disease within the first 2 years after progression.⁴ Since 2017, new treatment options have been approved for treating LBCL, including the emergence of chimeric antigen receptor T-cell (CAR T) therapy as a promising alternative for patients with relapsed or refractory (R/R) disease.⁵

Although initial US Food and Drug Administration (FDA) approval in 2017 for the use of anti-CD19 CAR T therapy to treat patients with LBCL was based on patients who had received ≥2 lines of therapy, the indication was expanded in 2022 for 2 products, axicabtagene ciloleucel (axi-cel) and lisocabtagene maraleucel (liso-cel), to include patients with relapsed disease within 12 months after 1L therapy based on the results of phase 3 randomized clinical trials.^6,7 Despite the proven efficacy of CAR T therapies, demonstrated in both clinical and real-world studies,^8,9 including 5-year follow-up data showing sustained and durable progression-free and overall survival,¹⁰ and in updates to the National Comprehensive Cancer Network (NCCN) guidelines,¹¹ the real-world uptake and use of CAR T therapy in patients with LBCL is currently ill defined.

Challenges to delivering CAR T therapy remain a key issue because of a wide range of clinical and nonclinical factors, including rapid disease progression, logistics, high cost, and proximity to an authorized treatment center.12, 13, 14, 15, 16 As such, the rate of CAR T use among a potentially eligible population appears low, especially among older patients,^17,18 despite studies showing treatment benefit in this high-risk group.^19,20 The current treatment patterns, survival outcomes, and use of CAR T in a potentially eligible patient population are not well characterized in the US community oncology practice setting. To address these questions, the objectives of this study were to evaluate the clinical outcomes after the completion of standard 1L therapies, including survival, and to investigate the treatment patterns for second-line (2L) and third-line (3L) treatments received by patients potentially eligible for CAR T therapy.

Methods

Patient population

This retrospective study used the US-based, electronic health record–derived deidentified Flatiron Health Research Database.²¹ Patients in the Flatiron Health Research Database were eligible for inclusion if they had a chart-confirmed pathological diagnosis of LBCL between 1 January 2011 and 30 April 2024, initiated at least 1 line of therapy, and did not receive a hematopoietic cell transplant or a clinical study drug before or as a part of 1L therapy. Additional inclusion and exclusion criteria are listed in the supplemental Methods. This study did not require ethics approval or patients’ informed consent because it used deidentified data and did not constitute human subject research according to the common rule.

A subgroup of patients who were considered potentially eligible for 2L and 3L CAR T therapy was analyzed in 2 cohorts. In the first cohort, 2L treatments were assessed in a cohort of patients with R/R disease considered eligible for 2L CAR T therapy if they received 1L chemotherapy in combination with an anti-CD20 treatment and an NCCN guideline–concordant 2L treatment within 12 months of 1L therapy between 1 January 2022 and 30 April 2024, based on the publication dates of the pivotal phase 3 clinical trials for axi-cel and liso-cel.^6,7 The CAR T therapy eligibility date cutoff had to be at least 3 months before the data cutoff date (30 April 2024; see supplemental Methods for details on eligibility criteria). Patients who received 2L systemic chemotherapy (not including steroids or radiation) within 60 days preceding the 3L CAR T infusion date (ie, bridging therapy) were reclassified as having received 2L CAR T therapy and were included in this cohort. In the second cohort, 3L treatments were similarly assessed for patients eligible for CAR T therapy between January 2018 and December 2021, based on the publication of the phase 2 clinical trial and FDA approval of axi-cel for the treatment of patients who have not responded to or who have relapsed after at least 2 other kinds of treatment.^22,23 Patients who received 3L systemic chemotherapy (not including steroids or radiation) within 60 days preceding the fourth-line CAR T infusion date (ie, bridging therapy) were reclassified as having received 3L CAR T and were included in this cohort.

Patients were stratified into the following CAR T fitness categories based on age and Eastern Cooperative Oncology group performance status (ECOG PS): fit (aged ≤82 years and ECOG PS of 0/1 or aged <65 years and ECOG PS of 2), borderline (aged 65-82 years and ECOG PS of 2), and unfit (aged >82 years or any age and ECOG PS of >2), and ECOG missing (patients with missing ECOG PS data). These fitness groups leveraged expert clinical opinion and findings from the ZUMA-7 axi-cel trial, which gave insight into specific ages and performance statuses in which success in therapy was observed.⁶

Statistical analysis

Primary analysis

For patients who initiated 1L therapy, demographic and clinical characteristics were described for the following groups: (1) those who subsequently initiated 2L therapy, (2) those who did not initiate 2L therapy and died within 12 months of initiation of 1L therapy, (3) those who did not initiate 2L therapy and were followed up for <12 months (ie, last confirmed clinical activity <12 months after 1L therapy initiation and no death), and (4) those who did not initiate 2L and were followed up for ≥12 months (ie, last confirmed clinical activity ≥12 months after 1L therapy initiation and no death within 12 months). Descriptive statistics (frequencies and percentages) for variables, including age at index date, sex, race, region, practice type, insurance type, socioeconomic status (SES), year of LBCL diagnosis, ECOG PS, disease characteristics (Ann Arbor stage; presence of B symptoms, extranodal disease, and bulky disease; cell of origin), 1L therapy, and laboratory values (including lactate dehydrogenase [LDH] levels above the upper limit of normal [International Prognostic Index]) were summarized for the aforementioned groups. Treatment sequencing from 1L to 2L, including Sankey diagrams and frequency tables, was summarized. The study index date was defined as the date of initiation of 1L treatment.

To enumerate mortality outcomes in patients that did not receive a 2L therapy, we analyzed the cumulative incidence of death, with 2L initiation as a competing event, stratified by ECOG PS at landmark time points. In this analysis, patients with missing ECOG PS data were combined into a single group with patients with ECOG PS 0 or 1 because of having similar baseline characteristics and prognoses to examine their combined prognostic utility in these data. We tested the assumptions made around these patients in a separate analysis (supplemental Figure 4). We then built a competing risk regression model to identify factors predicting death before being able to initiate a 2L therapy as the primary event, with 2L initiation as a competing risk, and multivariate adjustment for clinical risk factors and patient characteristics. A competing risk model was viewed as appropriate for this analysis because 2L initiation prevents observation of the outcome of interest (death before 2L); the Fine-Gray model appropriately estimates cumulative incidence of death accounting for this competing risk, whereas standard Cox regression would bias the hazard ratios. The Fine-Gray model, an extension of the Cox proportional hazards model, was used to allow for the estimation of subdistribution hazards, representing the instantaneous likelihood of death in the presence of initiating 2L as a competing risk.²⁴

SES

Flatiron Health’s proprietary variable for SES is created using data from the American Community Survey (2015-2019) in conjunction with the Yost Index, which is based on the combination of 7 variables, including household income, poverty, rent, home value, employment, education, and working class.²⁵ The SES variable is then divided into quintiles for the population based on census tracts, which are then mapped to a patient's residential address.²⁶

Subgroup analysis

Treatment sequencing patterns and demographic and clinical characteristics were described for the patients included in the subgroup analysis. Descriptive statistics for all variables used in the primary analysis were summarized in the subgroup analysis. Treatment sequencing from 1L, 2L, and 3L was summarized through Sankey diagrams and frequency tables using the aforementioned variables. Treatment patterns among a subgroup of patients that initiated 2L and 3L therapy based on the dates of FDA indication approvals for CAR T were categorized as follows (indexed to the start of the respective line of therapy): CAR T therapy; anti-CD20 plus platinum-containing regimen (eg, R-ICE [rituximab in combination with ifosfamide, carboplatin, etoposide] or R-DHAP (rituximab in combination with dexamethasone, cytarabine, and cisplatin]); novel therapies, including polatuzumab vedotin (pola)–based regimen (pola-BR [pola plus bendamustine and rituximab] and pola monotherapy), anti-CD19–based therapy (loncastuximab tesirine and tafasitamab plus lenalidomide), and bispecific agents (glofitimab and epcoritimab); Bruton tyrosine kinase inhibitor–based therapies; R-Gem/Ox (gemcitabine-oxaliplatin plus rituximab); and other chemotherapies.

Results

Baseline clinical and demographic characteristics of patients who initiated 1L therapy

The primary analysis of this study included 10 016 patients with LBCL who initiated 1L therapy (Figure 1). Overall, the median age was 69 years (interquartile range [IQR], 58-77), 56% were male, and 30% had an Ann Arbor stage IV diagnosis (Table 1). Most patients (75%) were treated in a community oncology practice, and nearly one-third lived in census tracts within the 2 lowest SES quintiles. ECOG PS of 0 (18%), 1 (23%), and ≥2 (11%) were observed among the patients in this cohort, with approximately half of the patients missing ECOG PS data. 1L therapy consisted primarily of anti-CD20 antibody plus anthracycline-containing regimen (84%; supplemental Figure 1), with R-CHOP as the most common 1L therapy (83%; Table 1). Overall, 13% of the patients who initiated 1L therapy died within 12 months without initiating any 2L therapy (Figure 2; supplemental Table 1). Overall, 30% of the patients initiated 2L therapy at some point after 1L discontinuation; 2L therapy classes administered to patients who completed 1L (N = 2999) included anti-CD20 plus platinum-containing regimen (31%), anti-CD20 plus other chemotherapies (19%), anti-CD20 plus anthracycline-containing regimen (13%), and CAR T (2%; CAR T was not commercially available as a therapy option for the whole study period; supplemental Figure 1). Median follow-up from discontinuation of 1L to the start of 2L was 4.6 months (IQR, 1-16.1), and 26.1 months (IQR, 9.3-60.5) from discontinuation of 1L until death or loss to follow-up (for those who initiated a 2L regimen). Patients who died within 12 months of 1L start and did not initiate a 2L regimen were followed up for 2.9 months (IQR, 1.1, 5.6) until last confirmed activity (Table 1).

Figure 1.

Cohort attrition.^aPatients were considered eligible for CAR T therapy if they received 1L chemotherapy plus anti-CD20 and 2L therapy within 12 months of 1L initiation, and the patient’s CAR T eligibility/index date was after 1 January 2018. ^bPatients were considered eligible for 2L CAR T therapy if their eligibility/index date fell between January 2022 and April 2024, aligning with treatment guidelines for the period. ^cPatients were considered eligible for 3L CAR T therapy if their eligibility/index date fell between January 2018 and December 2021, aligning with treatment guidelines for the period. CSD, clinical study drug; FH, Flatiron Health; FHRD, FH Research Database; HSCT, hematopoietic stem cell transplantation.

Table 1.

Patient demographics and clinical characteristics of patients included in the study

Characteristic	Overall (N = 10 016)	Patients who initiated 2L (n = 2999)	Patients who did not initiate 2L and died within 12 mo of initiation of 1L therapy (n = 1 328)	2L CAR T eligible (January 2022–April 2024; n = 205)	3L CAR T eligible (January 2018–December 2021; n = 304)
Year of initial LBCL diagnosis, n (%)
Before 2017	4 299 (43)	1470 (49)	531 (40)	0 (0)	84 (28)
2017-2018	1 678 (17)	529 (18)	234 (18)	0 (0)	104 (34)
2019-2020	1 686 (17)	484 (16)	280 (21)	1 (0.5)	107 (35)
2021-2022	1 547 (15)	396 (13)	213 (16)	148 (72)	9 (3.0)
2023-2024	806 (8.0)	120 (4.0)	70 (5.3)	56 (27)	0 (0)
Age at index date, y
Median (IQR)	69 (58-77)	68 (58-75)	76 (68-83)	65 (55-73)	64 (55-72)
Mean (SD)	66 (14)	65 (13)	74 (10)	62 (14)	63 (13)
Range	7-86	7-85	23-86	20-85	23-83
Age at index date, y, n (%)
≤49	1 248 (13)	350 (12)	32 (2.4)	34 (17)	41 (13)
50-64	2 608 (26)	861 (29)	198 (15)	64 (31)	113 (37)
65-74	2 934 (29)	954 (32)	352 (27)	63 (31)	91 (30)
75-84	2 879 (29)	785 (26)	523 (39)	41 (20)	59 (19)
≥85	347 (3.5)	49 (1.6)	223 (17)	3 (1.5)	0 (0)
Sex, n (%)
Female/unknown	4 452 (44)	1269 (42)	564 (43)	89 (43)	108 (36)
Male	5 564 (56)	1730 (58)	764 (58))	116 (57)	196 (64)
Race, n (%)
White	6 904 (69)	2185 (73)	893 (67)	137 (67)	222 (73)
Black or African American	628 (6.3)	190 (6.3)	65 (4.9)	15 (7.3)	31 (10)
Asian	263 (2.6)	85 (2.8)	22 (1.7)	10 (4.9)	6 (2.0)
Other	962 (9.6)	266 (8.9)	143 (11)	17 (8.3)	19 (6.3)
Unknown/not documented	1 259 (13)	273 (9.1)	205 (15)	26 (13)	26 (8.6)
Region, n (%)
Midwest	1 030 (10)	294 (9.8)	145 (11)	8 (3.9)	30 (9.9)
Northeast	1 045 (10)	261 (8.7)	146 (11)	9 (4.4)	27 (8.9)
South	3 651 (36)	988 (33)	518 (39)	71 (35)	86 (28)
West	1 395 (14)	326 (11)	193 (15)	26 (13)	30 (9.9)
Other/unknown	2 895 (29)	1130 (38)	326 (25)	91 (44)	131 (43)
Practice type at index date,∗n (%)
Academic	2 270 (23)	877 (29)	235 (18)	78 (38)	104 (34)
Community	7 522 (75)	2035 (68)	1069 (81)	120 (59)	188 (62)
Both	224 (2.2)	87 (2.9)	24 (1.8)	7 (3.4)	12 (3.9)
Insurance type, n (%)
Any Medicaid/Medicare	1 377 (14)	418 (14)	234 (18)	37 (18)	33 (11)
Commercial/other	3 876 (39)	1018 (34)	502 (38)	90 (44)	99 (33)
Unknown/not documented	4 763 (48)	1563 (52)	592 (45)	78 (38)	172 (57)
SES index,†n (%)
1 (lowest SES)	1 340 (13)	388 (13)	204 (15)	26 (13)	35 (12)
2	1 655 (17)	496 (17)	226 (17)	36 (18)	48 (16)
3	1 924 (19)	579 (19)	236 (18)	37 (18)	52 (17)
4	2 187 (22)	702 (23)	257 (19)	59 (29)	78 (26)
5 (highest SES)	2 029 (20)	579 (19)	235 (18)	38 (19)	61 (20)
Unknown	881 (8.8)	255 (8.5)	170 (13)	9 (4.4)	30 (9.9)
ECOG PS at index date,‡n (%)
0	1 832 (18)	466 (16)	145 (11)	37 (18)	58 (19)
1	2 343 (23)	679 (23)	286 (22)	54 (26)	71 (23)
2	786 (7.8)	202 (6.7)	257 (19)	13 (6.3)	15 (4.9)
≥3	256 (2.6)	47 (1.6)	130 (9.8)	2 (1.0)	0 (0)
Unknown	4 799 (48)	1605 (54)	510 (38)	99 (48)	160 (53)
Ann Arbor stage at initial diagnosis, n (%)
I	1 149 (11)	210 (7.0)	67 (5.0)	10 (4.9)	12 (3.9)
II	1 578 (16)	370 (12)	137 (10)	28 (14)	26 (8.6)
III	1 833 (18)	601 (20)	199 (15)	31 (15)	68 (22)
IV	3 004 (30)	1022 (34)	458 (35)	79 (39)	127 (42)
Group stage is not reported	2 452 (24)	796 (27)	467 (35)	57 (28)	71 (23)
Documentation of B symptoms, n (%)
Presence noted	3 462 (35)	1122 (37)	573 (43)	87 (42)	128 (42)
Absence noted	5 155 (51)	1376 (46)	537 (40)	86 (42)	134 (44)
Unknown/not documented	1 399 (14)	501 (17)	218 (16)	32 (16)	42 (14)
Diagnosis of extranodal LBCL, n (%)
Documented	5 712 (57)	1747 (58)	845 (64)	120 (59)	166 (55)
Unknown/not documented	4 304 (43)	1252 (42)	483 (37)	85 (41)	138 (45)
Documentation of bulky disease, n (%)
Documented	2 146 (21)	687 (23)	259 (20)	70 (34)	87 (29)
Unknown/not documented	7 870 (79)	2312 (77)	1069 (81)	135 (66)	217 (71)
1L therapy class, n (%)
R-CHOP	8 288 (83)	2312 (77)	900 (68)	174 (85)	265 (87)
Non–R-CHOP	1 728 (17)	687 (23)	428 (32)	31 (15)	39 (13)
Cell of origin, n (%)
Germinal-center B-cell	3 513 (35)	1010 (34)	391 (30)	93 (45)	121 (40)
Activated B-cell	2 462 (25)	744 (25)	365 (28)	65 (32)	88 (29)
Unknown/not documented	4 041 (40)	1245 (42)	572 (43)	47 (23)	95 (31)
Transformation, n (%)
Yes	1 624 (16)	723 (24)	203 (15)	50 (24)	87 (29)
No/unknown	8 392 (84)	2276 (76)	1125 (85)	155 (76)	217 (71)
Patient received HSCT, n (%)
No/unknown	9 566 (96)	2549 (85)	1328 (100)	198 (97)	251 (83)
In 2L	344 (3.4)	344 (11)	0 (0.0)	5 (2.4)	33 (11)
In 3L or beyond	106 (1.1)	106 (3.5)	0 (0.0)	2 (1.0)	20 (6.6)
Double-hit or triple-hitstatus, n (%)
Triple hit§	131 (1.3)	64 (2.1)	28 (2.1)	7 (3.4)	10 (3.3)
Double hit||	415 (4.1)	171 (5.7)	72 (5.4)	20 (9.8)	25 (8.2)
Not double or triple hit	9 470 (95)	2764 (92)	1228 (93)	178 (87)	269 (88)
Time from initial LBCL diagnosis to 1L initiation, median (IQR), mo	0.7 (0.4-1.1)	0.6 (0.3-1.0)	0.7 (0.3-1.1)	0.6 (0.3-1.0)	0.5 (0.2-1.0)
Time from 1L initiation (index) to last confirmed activity, median (IQR), mo	27.6 (8.3-62.4)	30.3 (12.9-65.0)	2.9 (1.1-5.6)	12.2 (8.2-19.9)	37.3 (16.9-64.2)
1L duration (initiation to last episode), median (IQR), mo	3.5 (2.2-4.0)	3.5 (2.1-4.1)	2.0 (0.7-3.6)	3.4 (1.5-3.9)	3.5 (2.8-4.4)
Time from 1L discontinuation (last episode) to last confirmed activity, median (IQR), mo	23.4 (4.4-57.9)	26.1 (9.3-60.5)	0.5 (0.0-1.6)	9.4 (5.5-16.4)	33.3 (12.6-60.0)

HSCT, hematopoietic stem cell transplantation; SD, standard deviation.

^∗Index date was defined as the date of 1L initiation.

^†SES index is calculated using a 5-year estimate of SES variables from the Census Bureau’s American Community Survey (2015-2019).

^‡ECOG PS may have been recorded up to 30 days before the index date or up to 7 days after the index date, whichever is closest to the index date. If there are multiple ECOG PS values at the same absolute distance from the index date, priority is given to the ECOG PS value that precedes the index date. For patients with multiple ECOG PS values recorded on the same day, the highest value will be selected.

^§MYC rearrangement with both BCL2 and BCL6.

^||MYC rearrangement with BCL2 or BCL6.

Figure 2.

Events occurring after 1L therapy discontinuation. Most (∼85%) patients had either a death event, 2L initiation, or ≥12 months of follow-up from initiation of 1L (to their last clinical activity).

Survival outcomes and predictors of mortality among patients who initiated 1L therapy

Based on a competing risk regression model, the predictors of all-cause mortality before starting 2L treatment among patients with R/R disease, ordered by strongest to weakest association, included higher LDH, ECOG PS of ≥2, triple-hit status, double-hit status, presence of B symptoms, activated B-cell origin, lower SES, higher stage at initial diagnosis, and male sex (Table 2). When analyzing the cumulative incidence of death with 2L therapy initiation as a competing event, 14% of the patients had an incident death within 12 months of initiation of 1L. Overall, 7.4% of the patients in the ECOG PS of 0/1/unknown group had died at month 3, and cumulative deaths grew to 11% by month 12 for patients with ECOG PS of 0/1/unknown (n = 8974; Figure 3). Among patients with ECOG PS of 2, 24% had died by month 3% and 35% by month 12; an ECOG PS of ≥2 was a strong predictor of death before 2L therapy initiation (Figure 3).

Table 2.

Competing risk Fine-Gray subdistribution hazard regression model of death (without initiating 2L) with 2L initiation as a competing risk, overall

Adjusted covariates∗	n (%)	Subdistribution hazard ratio	95% CI	P value†
LDH/ULN IPI at index date‡				<.001
0-1 (<1 ULN; Ref)	2 632 (26.3)	—	—	—
2 (1-1.5 ULN)	999 (10.0)	1.57	1.30-1.90	—
3 (≥1.5 ULN)	436 (4.4)	3.10	2.51-3.81	—
Unknown/not documented	5 974 (59.6)	1.02	0.89-1.17	—
ECOG PS at index date				<.001
1 (Ref)	2 343 (23.4)	—	—	—
0	1 832 (18.3)	0.85	0.70-1.05	—
2	786 (7.8)	2.16	1.81-2.56	—
≥3	256 (2.6)	3.95	3.16-4.93	—
Unknown	4 799 (47.9)	0.99	0.85-1.16	—
Double- or triple-hit status				.014
Not double or triple hit (Ref)	9 470 (94.5)	—	—	—
Triple hit§	131 (1.3)	1.63	1.10-2.42	—
Double hit||	415 (4.1)	1.23	0.98-1.56	—
B symptoms				<.001
Absent (Ref)	5 155 (51.5)	—	—	—
Present	3 462 (34.6)	1.38	1.22-1.56	—
Unknown/not documented	1 399 (14.0)	1.16	0.99-1.36	—
Cell of origin				<.001
Germinal-center B cell (Ref)	3 513 (35.1)	—	—	—
Activated B cell	2 462 (24.6)	1.30	1.12-1.51	—
Unknown/not documented	4 041 (40.3)	1.32	1.15-1.52	—
Age at index date, y				<.001
Continuous	10 016 (100.0)	1.06	1.05-1.06	—
SES index¶				.2
5 (Ref)	2 029 (20.3)	—	—	—
1	1 340 (13.4)	1.07	0.93-1.22	—
2	1 655 (16.5)	1.13	1.00-1.28	—
3	1 924 (19.2)	1.10	0.98-1.24	—
4	2 187 (21.8)	1.15	1.03-1.28	—
Unknown	881 (8.8)	1.11	0.95-1.29	—
Ann Arbor stage at initial diagnosis				<.001
IV (Ref)	3 004 (30.0)	—	—	—
I	1 149 (11.5)	0.45	0.35-0.59	—
II	1 578 (15.8)	0.65	0.54-0.78	—
III	1 833 (18.3)	0.69	0.58-0.82	—
Not Reported	2 452 (24.5)	1.23	1.07-1.41	—
Sex				.015
Male (Ref)	5 564 (55.6)	—	—	—
Female	4 448 (44.4)	0.86	0.77-0.96	—
Unknown/not documented	4 (0.0)	2.97	0.46-19.3	—

CI, confidence interval; IPI, International Prognostic Index; Ref, reference; ULN, upper limit of normal.

^∗Adjusted for other variables.

^†P value derived from a Wald test assessing whether the estimated coefficient differs significantly from 0.

^‡Index date was defined as the date of 1L initiation.

^§MYC rearrangement with both BCL2 and BCL6.

^||MYC rearrangement with BCL2 or BCL6.

^¶SES index quintile for a patient’s residential block group based on 2015 to 2019 census data (1 = lowest; 5 = highest).

Figure 3.

Cumulative incidence of death from 1L initiation with 2L Initiation as a competing event, overall. CI, confidence interval.

Baseline characteristics of patients potentially eligible for 2L or 3L CAR T

A subgroup analysis of patients included in this study included 205 patients with LBCL who were considered eligible for 2L CAR T therapy in the period after January 2022 (up to 3 months before the data cutoff date) and 304 patients with LBCL who were considered eligible for 3L CAR T therapy from January 2018 to December 2021 (Figure 1; supplemental methods). In the 2L cohort, the median age was 65 years (IQR, 55-73), 57% were male, and 39% had a stage IV diagnosis (Table 1). In the 3L cohort, the median age was 64 years (IQR, 55-71), 64% were male, and 42% had stage IV disease (Table 1). Of the 205 patients considered eligible for 2L CAR T therapy, 128 (62.4%) were deemed clearly fit for CAR T therapy (median age, 61 years [IQR, 53-70]), and 15 (7.3%) were borderline fit (median age, 75 years [IQR, 72-80]).

Approximately 28% of the patients considered eligible for 2L CAR T therapy had missing ECOG PS data (supplemental Table 2). Of the 304 patients considered eligible for 3L CAR T therapy before 2L CAR T approval, 182 (59.9%) were deemed clearly fit for CAR Ts (median age, 65 years [IQR, 56-74]), with ECOG PS of 0 in 83 (46%), ECOG PS of 1 in 88 (48%), and ECOG PS of 2 in 11 (6.0%). Approximately one-third of the patients considered eligible for 3L CAR T therapy had missing ECOG PS data; however, the median age of these patients more closely resembled that of clearly fit patients than borderline and unfit patients (supplemental Table 2). Numerically lower proportions of Black patients (vs non-Black) and patients in the lowest SES indices met CAR T eligibility fitness criteria for 2L and 3L therapy.

Treatment patterns of patients potentially eligible for 2L or 3L CAR T therapies

When treatment patterns for patients considered eligible for 2L and 3L CAR T therapy were stratified by CAR T fitness, 32 (25.0%) of the 128 patients deemed clearly fit for 2L CAR T therapy received 2L CAR T therapy, and 12 (9.4%) clearly fit patients received conventional 2L salvage therapy with R-ICE or R-DHAP (Figure 4A). Other common regimens in the 2L setting included other chemoimmunotherapy (n = 33 [25.8%]) and novel therapies (n = 31 [25.0%]), which included pola, loncastuximab tesirine, tafasitamab, or bispecific antibody–based therapy. Two (13.3%) borderline-fit patients received 2L CAR T therapy. Among patients considered eligible for 3L CAR T therapy, 65 of the 182 (35.7%) clearly fit patients received CAR Ts in 3L therapy. No borderline-fit patients received 3L CAR T therapy (Figure 4B).

Figure 4.

Treatment patterns for patients eligible for CAR T therapy by fitness groups for CAR T therapy based on age and ECOG PS. (A) Patients eligible for 2L CAR T (January 2022–April 2024). (B) Patients eligible for 3L CAR T (January 2018–December 2021). Patients were indexed to the start of the respective line of therapy. BTKi, Bruton tyrosine kinase inhibitor.

Discussion

In this large, real-world study of patients with LBCL treated primarily in the community oncology setting, we found that a sizable proportion of patients died within 12 months of initiating 1L therapy, including many with ECOG PS of 0/1 or unknown. Additionally, we found that among patients who went on to receive further treatment, a minority of potentially eligible and fit patients received CAR Ts in the 2L or 3L setting, with many patients receiving additional rounds of chemoimmunotherapy instead. Although some patients will have curative benefit from 1L therapy, 35% of the patients with ECOG PS of 2 and 55% with ECOG PS of ≥3 died within 12 months of initiating 1L therapy, before the opportunity to initiate a 2L therapy. Although we are unable to determine the cause of death in these data, deaths occurring within 12 months of initiating 1L therapy have a strong potential to be related to the patient’s lymphoma or treatment complications, based on timing of events and clinical data, which highlights that standard 1L therapies are not sufficient for some patients.^27,28 Clinical trials examining novel 1L treatment strategies may provide faster access to potentially curative therapies in patients with high-risk disease who likely cannot wait until 2L to receive therapy.²⁹

The strongest predictors of all-cause death before being able to initiate a 2L therapy were the subcomponents of the International Prognostic Index score, specifically ECOG PS, higher initial stage at diagnosis, higher LDH, and older age. Patients with biologically aggressive LBCL subtypes, specifically double hit/triple hit, demonstrated significantly increased risk of mortality before initiating 2L therapy, underscoring the need for risk-adaptive treatment approaches in these high-grade subgroups. Consistent with previous studies,^30,31 we observed a heightened risk of death among patients living in low SES neighborhoods. It is unclear what is directly responsible for this observation in our study, but the lack of health insurance coverage or the resources to pay for out-of-pocket treatment expenses with coverage and limited access to health care are common obstacles for patients from low SES neighborhoods, affecting their ability to obtain high-quality treatment and follow-up care.^32,33

The large number of patients with undocumented ECOG PS (48%) is not unexpected in real-world data but presents limitations in drawing conclusions about the fitness for CAR T therapy and prognoses of these patients. To help contextualize how to consider patients with unknown ECOG PS as well as justify our decision to aggregate patients with ECOG PS of 0/1 and those with unknown ECOG PS, we assessed cumulative incidence stratified by ECOG PS (supplemental Figure 2). Patients with ECOG PS of 0/1 were found to have a cumulative incidence of death, almost identical to those with unknown ECOG PS, providing strong evidence that these patients can be considered to have a similar prognosis and likely fitness for CAR T therapy.

In 2022, the treatment indication for axi-cel and liso-cel was expanded to patients with LBCL who were primary refractory or relapsed within 12 months of 1L therapy based on phase 3 randomized trials.^6,7,34,35 Despite the efficacy demonstrated in these randomized clinical trials and the increasing availability of CAR T therapies, we observed that a large proportion of potentially eligible and otherwise fit patients did not receive CAR T therapy in the 2L setting, particularly in the community setting, suggesting that patients and providers may be considering alternative treatment approaches before or instead of CAR T therapy for reasons that may include accessibility, affordability, patient clinical factors, or patient preference. Perceived barriers to community physicians’ referral to CAR T therapy include logistics and high cost,¹³ proximity of the authorized treatment center from the physician’s practice or patient’s residence,^13,14 and patient choice.¹² A persistent digital divide among older patients with cancer treated in community settings, particularly those with lower income, lower educational attainment, and limited broadband access, may impede access to online resources needed to seek timely second opinions.³⁶ The FDA’s recent removal of the Risk Evaluation and Mitigation Strategy program requirements for approved CAR T therapies and subsequent label updates may alleviate some logistical barriers because of relaxed criteria for patient monitoring (eg, proximity of a health care facility reduced from 4 to 2 weeks, and driving restriction reduced from 8 to 2 weeks) and is anticipated to improve access to patients, especially from rural areas https://www.fda.gov/vaccines-blood-biologics/safety-availability-biologics/fda-eliminates-risk-evaluation-and-mitigation-strategies-rems-autologous-chimeric-antigen-receptor.³⁷

The socioeconomic and racial disparity in access to CAR T therapy has been well documented and may have factored into the lower uptake in the 2L and 3L setting.^38,39 There is a lack of realistic options for those patients with no or incomplete caregiver support for the duration of the CAR T treatment process. Although many patients, especially those from higher socioeconomic communities, have access to caregiver support that can facilitate outpatient care of patients in the first month during the treatment process, this level of caregiver access may be more challenging in lower socioeconomic communities because of the inability of potential caregivers to take time away from work, limitations because of disabilities, or lack of transportation. Although capture of CAR T therapy may be incomplete for some patients who leave the Flatiron Health network after completion of 1L, the low observed rate of 2L CAR T use in this cohort is unexpected given its NCCN category 1 guideline recommendation dating back to 2022; supplemental Figures 5 and 6 show treatment patterns stratified by CAR T fitness group for race and SES, respectively, and reveal that, among “clearly fit” patients, higher SES and White patients appear much more likely to receive CAR T therapy in the appropriate treatment setting.

The complexity of CAR T therapy, including referral and consultation, apheresis, manufacturing, and infusion, often takes up to 3 to 4 months.^18,40 Developments in rapid manufacturing of CAR Ts have the potential to reduce the time from leukapheresis to infusion⁴¹; however, additional considerations should be given to developing a more efficient referral system to improve patient access. A recent study from a large community-based oncology network found that the primary driver for CAR T ineligibility was disease progression and declining health while waiting between consultation and the initiation of CAR T therapy.¹⁸ Although the time from consultation to CAR T infusion has continued to decline steadily over the years, the current referral process may be inadequate because studies have shown the average time for this process to be >3 months.^18,38 This may be achieved through prioritizing early referrals, improving collaboration between community oncology practices and CAR T providers to promote access to CAR T therapy, and proactive planning for advanced therapies while patients are still getting 1L therapy.¹²

Several financial barriers may also contribute to the low uptake in community settings. Delivery of CAR T therapy remains concentrated in specialized academic centers because many community hospitals lack the infrastructure, multidisciplinary expertise, and accreditation required to safely administer therapy and manage complications such as cytokine release syndrome and neurotoxicity.⁴² Fragmentation between community oncology practices and authorized CAR T centers, combined with logistical and reimbursement challenges, can delay referrals and limit care coordination, contributing to reduced access for patients treated outside major academic institutions.

There are some limitations to our study. There is the possibility that the 2L CAR T–eligible cohort may be undercaptured because patients who initiated 1L therapy on or after February 2023 did not have a full 12-month window to meet criteria for study inclusion. Furthermore, treatment patterns in the 2L CAR T–eligible cohort may be overrepresented by treatments that are faster to receive than CAR T within a truncated follow-up period. Patients referred outside of the Flatiron Health network without supporting documentation were also not included, which may also result in the undercapture of patients who were CAR T eligible. These limitations may also be relevant to the 3L CAR T–eligible cohort. Moreover, to align with CAR T treatment guidelines, patients who had received hematopoietic stem cell transplantation or a clinical study drug in 1L or earlier settings were not included in the study; although this represented a small number (∼3%) of patients who received treatment, these patients might be healthier than patients who were not excluded, reducing the generalizability of findings. Some factors that would be important for understanding CAR T eligibility or barriers to care (eg, comorbidities and distance to CAR T clinic) or for understanding factors that could affect survival (drug toxicities and cause-specific mortality) were not available.

A portion of patients, particularly in more recent years, did not have at least 12 months of follow-up from 1L initiation; heavy censoring may be affecting cumulative incidence estimates in certain patient groups. To assess the effect of immature survival data, we stratified cumulative incidence curves by patients having earlier (2011-2017) and recent (2018-2024) index years (supplemental Figures 3 and 4). This analysis found that these 2 groups have similar cumulative incidence functions for their 6 months of follow-up. After this, the risk set decreases more rapidly in the recent index year cohort; the higher cumulative incidence observed in the 2018 to 2024 cohort likely reflects differences in follow-up maturity. Patients indexed in recent calendar years have shorter observable follow-up, resulting in earlier administrative censoring and inflation of early cumulative incidence estimates. Finally, the data used in this study originate mostly from community practices, also limiting the generalizability of our findings for patients seeking access to CAR T therapy across broader practice settings.

In conclusion, despite significant therapeutic advancements in the 2L setting, many patients with diffuse LBCL die within 12 months of 1L initiation without receipt of 2L. These findings support ongoing clinical trials using novel therapies in the frontline setting and also emphasize the systemic issues that patients face in receiving curative, FDA-approved, and guideline-recommended 2L treatment options. Further research can help understand the factors associated with barriers to CAR T therapy initiation in the hopes of removing or reducing these challenges to access.

Conflict-of-interest disclosure: M.-A.P. reports honoraria from Adicet, Allogene, Allovir, Caribou Biosciences, Celgene, Bristol Myers Squibb, Equilium, Exevir, ImmPACT Bio, Incyte, Karyopharm, Kite/Gilead, Merck, Miltenyi Biotec, MorphoSys, Nektar Therapeutics, Novartis, Omeros, Orca Bio, Sanofi, Syncopation, VectivBio AG, and Vor Biopharma; serves on data and safety monitoring boards for Cidara Therapeutics and Sellas Life Sciences; serves on the scientific advisory board of NexImmune; reports ownership interests in NexImmune, Omeros, and Orca Bio; reports institutional research support for clinical trials from Allogene, Incyte, Kite/Gilead, Miltenyi Biotec, Nektar Therapeutics, and Novartis; ownership interests in NexImmune, Omeros, and Orca Bio; and institutional research support from Allogene, Incyte, Kite/Gilead, Miltenyi Biotec, Nektar Therapeutics, and Novartis. J.P.M. reports consulting/advisory roles with Novartis, Bristol Myers Squibb, AlloVir, Autolus, Kite, Nektar, Sana Technologies, CRISPR, Envision, Caribou Bio, Legend Biotech, and Cargo Therapeutics. J.S., B. Adamson, and A.J.P. are employed at Flatiron Health, Inc, an independent member of the Roche Group; and report stock ownership in Roche. H.H., B. Adedokun, and A.R.P. report employment with Kite Pharma, Inc, an independent member of Gilead Sciences, Inc; and stock ownership in Gilead. M.R.F. declares no competing financial interests.

The current affiliation of A.R.P. is Indracon Access, Seattle, WA, and RainCity Analytics, Vancouver, British Columbia, Canada.