Marked variation in eligibility criteria across registrational trials for relapsed diffuse large B‐cell lymphoma limits applicability to clinical practice

Summary

Therapeutic options for relapsed or refractory diffuse large B‐cell lymphoma (rDLBCL) have expanded following accelerated approvals based largely on surrogate end‐points, yet confirmatory studies have not consistently reproduced these benefits across all agents, and outcomes for many patients remain poor. Restrictive and heterogeneous eligibility criteria further limit generalisability of trial findings to routine populations. We examined eligibility criteria across seven registrational rDLBCL trials and applied them hypothetically to a real‐world cohort of 180 patients treated in Australia with standard‐of‐care therapies. The trials required a median of 39 criteria (range 31–45), with marked variability across patient and disease domains. The greatest discordance related to prior lines of therapy, definitions of measurable disease and organ function parameters. Among 320 patient relapses (median 1, range 1–8), 52% were ineligible for any trial and no patient met criteria for all seven. Eligibility ranged from 4% (SADAL) to 22% (ZUMA‐1). Common reasons for exclusion were limits on prior therapy, washout requirements and organ dysfunction. Overall survival did not differ between patients ineligible for all trials and those eligible for at least one (p = 0.97). Median progression‐free survival (PFS) was comparable to registrational studies in both groups. Substantial eligibility heterogeneity excludes most real‐world patients, limiting trial generalisability and supporting the need for more inclusive, standardised frameworks.

INTRODUCTION

Diffuse large B‐cell lymphoma (DLBCL) relapse occurs in approximately 40%–50% of patients following front‐line chemoimmunotherapy. ¹ Several novel agents have received accelerated approval for relapsed or refractory DLBCL (rDLBCL) based on landmark studies addressing this unmet need; however, outcomes remain poor. ² , ³ , ⁴ , ⁵ , ⁶ , ⁷ , ⁸ Many of these studies were non‐randomised phase 1b/2 trials with highly selective eligibility criteria, raising concerns about the applicability of their findings to the broader patient populations encountered in routine clinical practice.

For trial results to translate into meaningful clinical benefit, registrational studies must evaluate therapies in populations that reflect those treated in standard care. While some safety‐driven selection is appropriate in early phase studies, later phase trials are intended to demonstrate efficacy that is generalisable to the wider disease population. Increasingly, however, eligibility criteria have become more restrictive, often excluding patients with minimal comorbidities or mild organ dysfunction. ⁹ As a result, fewer than 5% of patients with cancer in high‐income countries enrol in clinical trials. ¹⁰ , ¹¹ , ¹²

Despite this, regulatory approvals rarely restrict access to new therapies to patients who would not have met original trial criteria. Clinicians therefore often extend the use of novel agents to broader patient populations, who are frequently older, frailer or have greater comorbidity burden than those represented in registrational studies. ⁹ This disconnect may result in unexpected toxicity and less favourable risk–benefit profiles compared with those observed in the highly selected trial cohorts. ¹³ Real‐world analysis of individual rDLBCL therapies, including polatuzumab vedotin, ¹⁴ chimeric antigen receptor T‐cell (CAR‐T) transplantation ¹⁵ , ¹⁶ and tafasitamab–lenalidomide ¹⁷ have reported eligibility rates as low as 11% among non‐trial post‐marketing populations, with consistently inferior outcomes to those reported in pivotal studies.

However, no prior study has systematically benchmarked eligibility criteria across the registrational trials and applied them to a single real‐world rDLBCL patient cohort. In the context of a rapidly evolving therapeutic landscape, with more than 1500 DLBCL trials registered on ClinicalTrials.gov since 2019, quantifying the degree of misalignment between trial populations and real‐world patients is critical. Here, we systematically evaluate eligibility criteria across registrational trials in rDLBCL supporting accelerated and prioritised approvals by the United States Food and Drug Administration (US‐FDA) in the era preceding bispecific antibody therapies. We quantify variation in eligibility criteria across trials and assess the applicability of these criteria to real‐world patients.

METHODS

This mixed‐methods study incorporated two sequential components:

1. Detailed review of similarities and differences in eligibility criteria used in recent US‐FDA registrational trials supporting accelerated approvals for rDLBCL, to characterise variability in eligibility across studies; and

2. Application of these eligibility criteria hypothetically to a multicentre, retrospective real‐world cohort of patients with rDLBCL from our routine care databases and national lymphoma registry who were treated with standard‐of‐care therapy in an Australian context.

Evaluation of clinical trials

US‐FDA registrations of new therapies for rDLBCL between 1 January 2012 and 31 December 2022 were identified through the US FDA website ¹⁸ using a combination of search terms: diagnosis related (“large B‐cell lymphoma”, “DLBCL”), treatment line (“relapsed”, “refractory”) and drug approval filters within the FDA database and medication approval announcements. For each approved agent, the corresponding registrational clinical trial was identified by cross‐referencing ClinicalTrials.gov entries and reviewing associated regulatory documentation, including prescribing information and FDA review summaries. All relevant primary publications were subsequently retrieved and reviewed. None of these studies enrolled in Australia at the real‐world cohort participating sites.

Clinical trial information was obtained from published manuscripts, ² , ³ , ⁴ , ⁵ , ⁶ , ⁷ , ⁸ ClinicalTrials.gov entries and trial protocols where available. Extracted eligibility criteria were used to inform the design of the electronic data collection tool applied to the real‐world cohort and to determine hypothetical trial eligibility at the time of each disease relapse.

For inter‐trial comparison, all inclusion and exclusion criteria were extracted and mapped to prespecified eligibility domains to facilitate structured comparison. Where multiple criteria assessed the same underlying clinical construct (e.g. cardiac function or recent ischaemic events), these were consolidated into a single composite domain (e.g. cardiac function) within the comparative matrix. Inter‐trial consistency was quantified using a per‐criterion modal agreement approach, ¹⁹ whereby the most frequently occurring eligibility criterion across trials was identified for each domain. The number of trials concordant with this modal requirement was recorded and expressed as both an absolute count and a percentage concordance. For visual interpretation, domains were colour‐coded according to the degree of concordance across trials: green, complete concordance (100%), amber, partial concordance (50%–99%) and red, low concordance (<50%).

Identification of the real‐world patient cohort and hypothetical application of eligibility criteria

Adult patients (≥18 years) with rDLBCL following at least one prior line of therapy between January 2012 and December 2022 were identified by local investigators across five Australian tertiary centres spanning four states (Victoria, Queensland, New South Wales and Western Australia). Patients experiencing relapse during this period were identified using institutional electronic medical records and the national Lymphoma and Related Diseases Registry (LaRDR). ²⁰ This period was selected to align with the timing of US Food and Drug Administration approvals and therefore reflects a standard‐of‐care cohort, as therapies evaluated in the registrational studies were not yet approved within the Australian regulatory setting.

Data were captured at the level of individual relapse episodes and included age, sex, Eastern Cooperative Oncology Group (ECOG) performance status, ²¹ comorbidities, viral and infectious disease status, organ function parameters (cardiac, renal, hepatic and pulmonary), haematological indices, histopathological findings, positron emission tomography/computed tomography (PET/CT) imaging results, prior treatments received, treatment response and clinical outcomes including survival status.

Demographic and clinical data relevant to the eligibility criteria of the selected registrational trials were extracted and entered into a standardised data collection tool by each site to determine hypothetical trial eligibility. A ‘relapse episode’ was defined as a distinct, temporally separate relapse or progression of disease within an individual patient and was assessed independently against the eligibility criteria of each clinical trial. Relapse episodes with insufficient data to permit definitive eligibility classification or survival analysis were to be censored and excluded from statistical analyses. Patients without available follow‐up data were excluded from survival analysis but were retained in reported eligibility analysis. The periods during which individual trials were open to recruitment were not considered in our assessment, as eligibility was determined solely based on the published inclusion and exclusion criteria, provided the trial fell within the prespecified study timeframe.

Data analysis

Patient characteristics were summarised using frequencies and percentages for categorical variables and medians with ranges for continuous variables. Eligibility criteria were applied to each relapse episode; episodes were classified as eligible if all trial‐specific criteria were met and ineligible if any criterion was not met. Proportions of patients eligible were reported for each trial.

Patient outcomes were analysed according to trial eligibility status to assess the association between eligibility and prognosis. Overall survival (OS) was calculated on a per‐patient basis and defined as the time from first relapse to death or last documented follow‐up. OS was estimated using the Kaplan–Meier method ²² with comparisons between patients eligible for at least one trial and those ineligible for all trials. Progression‐free survival (PFS) was assessed on a per‐relapse episode basis using Kaplan–Meier methods, recording those eligible and ineligible for each trial. Survival distributions were compared using the Gehan–Breslow–Wilcoxon test. Confidence intervals (CIs) for proportions were estimated using the Wilson score method.

All statistical analyses were performed, and figures were generated using Microsoft Excel and GraphPad Prism (V10.0).

RESULTS

Registrational clinical trial details

Seven registrational studies undertaken in patients with rDLBCL that led to accelerated US‐FDA drug approvals between 2012 and 2022 were identified—none were open at the participating Australian sites at the time of treatment of the real‐world cohort (Table 1). ² , ³ , ⁴ , ⁵ , ⁶ , ⁷ , ⁸ Eligibility criteria were extracted from published trial protocols and manuscripts; eligibility criteria are included in the Appendix S1.

TABLE 1.

Registrational clinical trials included in the comparative analysis.

Trial	Clinicaltrials.gov ID	Intervention	Phase	Randomised	Number of patients a	Date of US‐FDA registration
SADAL 2	NCT02227251	Selinexor	2b	No	127	June 2020
GO29365 3	NCT02257567	Polatuzumab vedotin + bendamustine + rituximab (BR)	1b/2	Yes	80	June 2019
L‐MIND 4	NCT02399085	Tafasitamab + Lenalidomide	2b	No	81	July 2020
LOTIS‐2 5	NCT03589469	Loncastuximab	2	No	145	April 2021
ZUMA‐1 8	NCT02348216	Axicabtagene ciloleucel	2	No	77	October 2017
TRANSCEND 7	NCT02631044	Lisocabtagene maraleucel	1b	No	269	February 2021
JULIET 6	NCT02445248	Tisagenlecleucel	2	No	111	May 2018

Abbreviation: US‐FDA, United States Food and Drug Administration.

^a Number of patients with diffuse large B‐cell lymphoma (DLBCL) who received the intended therapy.

Across the seven studies, the median number of eligibility criteria listed was 39 (range 31–45), comprising a median of 14 inclusion criteria (range 10–19) and 22 exclusion criteria (range 20–31). All eligibility criteria were collated and mapped into 21 prespecified ‘eligibility domains’ to enable structured comparison across trials (Table 2).

TABLE 2.

Comparison of eligibility criteria across all studies. [Colour table can be viewed at wileyonlinelibrary.com]

Criteria	SADAL	GO29365	L‐MIND	LOTIS‐2	ZUMA‐1	TRANSCEND	JULIET	Criterion required	Concordant eligibility	Percentage concordance (%)
Age	≥18	≥18	≥18	≥18	≥18	≥18	>18	7/7	7/7	100
ECOG	0–2	0–2	0–2	0–2	0–1	0–2	0–1	7/7	5/7	71
Estimated life expectancy	>3 months	>6 months	NS	NS	NS	NS	>3 months	3/7	2/3	67
Transformed from indolent lymphoma	Permitted	Not permitted	Permitted	NS	Permitted if transformed from follicular lymphoma b	Permitted	NS	5/7	3/5	60
Measurable disease	LA >1.5 cm, or LA 1.1–1.5 cm and SA >1.0 cm	At least one lesion >1.5 cm	At least one lesion >1.5 cm (transverse) and >1.0 cm (perpendicular)	Measurable disease as per Lugano classification	Measurable disease according to the International Working Group	PET‐positive disease	Nodal >2 cm LA, extranodal >1 cm LA and SA	7/7	0/7	0
Lines of prior therapy	2–5	≥1	1–3	≥2	≥1	≥2	≥2	7/7	3/7	43
Washout period	98 days	14 days	14 days	14 days	14 days	14 days	28 days	7/7	5/7	71
Eligibility for ASCT	Ineligible, or if relapsed post‐transplant	Ineligible, or if relapsed post‐transplant	Ineligible, or if relapsed post‐transplant	NS	NS	NS	Ineligible, or if relapsed post‐transplant	4/7	4/4	100
CNS involvement	Ineligible	Ineligible	Ineligible	Ineligible	Ineligible	Secondary/relapsed CNS disease permitted	Ineligible	7/7	6/7	86
Double or triple hit genetics	NS	Permitted	Ineligible a	Permitted	NS	Permitted	NS	4/7	3/4	75
Refractory DLBCL (PD or SD to prior line of treatment)	Eligible if washout period met c	Permitted	Ineligible	Permitted	Required	Permitted	Permitted	7/7	5/7	71
Creatinine clearance	>30 mL/min	>40 mL/min	>60 mL/min	>60 mL/min	>60 mL/min	>30 mL/min	>60 mL/min	7/7	4/7	57
Cardiac and/or LVEF (%)	AMI not within 3 months	AMI not within 6 months	Non‐severe disease	AMI not within 6 months	LVEF >50%, ineligible if AMI or cardiac stenting within 12 months	LVEF >40%, ineligible if AMI, unstable angina or cardiac stentining within 6 months	LVEF >45%, ineligible if AMI and/or unstable angina within 6 months	7/7	0/7	0
Liver function	Bilirubin <2× ULN, AST and/or ALT <2.5× ULN unless known liver involvement with lymphoma	Bilirubin ≤1.5× ULN and/or AST or ALT <2.5× ULN	Bilirubin ≤2.5× ULN and/or ALT, AST and ALP ≤3× ULN unless known liver involvement with lymphoma	Bilirubin ≤1.5× ULN and/or ALT, AST and/or GGT ≤2.5× ULN	Bilirubin ≤1.5 mg/dL, AST, ALT ≤2.5 × ULN	Bilirubin <2.0 mg/dL, ALT ≤5× ULN	Bilirubin ≤2.0 mg/dL, ALT ≤5× ULN	7/7	2/7	28
Platelet count	>75 × 109/L	>75 × 109/L	>90 × 109/L	>75 × 109/L	>75 × 109/L	‘Adequate bone marrow function’	>50 × 109/L	7/7	4/7	57
Lymphocyte count	<50 000/μL	NS	NS	NS	≥100/μL	‘Adequate bone marrow function’	≥300/μL	4/7	0/4	0
Neutrophil count	>1.0 × 109/L	≥1.5 × 109/L	≥1.5 × 109/L	≥1 × 109/L (without growth factor support)	≥1 × 109/L	‘Adequate bone marrow function’	≥1 × 109/L	7/7	3/7	43
Infection history	Ineligible if infection requiring parenteral antimicrobials within 1 week	Active infection requiring parenteral treatment within 4 weeks	Ineligible if has required parenteral antibiotics for active infection within 2 weeks	NS	Ineligible if has infection requiring parenteral antimicrobials (time not stated)	Ineligible if uncontrolled infection despite appropriate antibiotics at time of leukapheresis or liso‐cel administration	Uncontrolled infection within 3 days prior to product infusion	6/7	0/6	0
Surgery history	Ineligible if major surgery within 2 weeks	Ineligible if major surgery within 6 weeks	Ineligible if major surgery within 2 weeks	Ineligible if major surgery within 2 weeks	NS	NS	NS	4/7	3/4	75
Pregnancy status	Must not be pregnant	Must not be pregnant	Must not be pregnant	Must not be pregnant	Must not be pregnant	Must not be pregnant	Must not be pregnant	7/7	7/7	100
Hepatitis B history	Ineligible if active hepatitis B infection	Ineligible if active hepatitis B infection, eligible if prior exposure but HBV DNA is undetectable	Ineligible if active hepatitis B infection	Ineligible if seropositive and requiring anti‐viral therapy for hepatitis B (testing not mandated)	Ineligible if known hepatitis B with positive viral load	Ineligible if active hepatitis B infection	Uncontrolled active or latent hepatitis B infection	7/7	7/7	100

Note: Concordance 100%

Concordance between 50‐99%

Concordance below 50%

Abbreviations: ASCT, autologous stem cell transplant; CLL, chronic lymphocytic lymphoma; CNS, central nervous system; DLBCL, diffuse large B‐cell lymphoma; DNA, deoxyribonucleic acid; ECOG, Eastern Cooperative Oncology Group; HBV, hepatitis B virus; LA, long axis; LVEF, left ventricular ejection fraction; NA, not applicable; NS, not stated; PD, progressive disease; SA, short axis; SD, stable disease; ULN, upper limit of normal.

^a If evidence of MYC with BCL‐2 and/or BCL‐6 translocation.

^b Ineligible if transformed from CLL.

^c If PD or SD from last line, at least 98 days must have elapsed prior to enrolment.

Substantial heterogeneity in eligibility requirements was observed across the registrational trials. In 7 of the 21 eligibility domains, at least one trial did not specify whether the criterion was required at all; for example, estimated life expectancy was explicitly defined in only three of the seven clinical trials. Concordance across individual eligibility domains varied markedly, ranging from complete agreement across all trials (100% concordance) in only three domains (age, pregnancy status and active hepatitis B), to complete discordance (0% concordance) in four domains (description of measurable disease, cardiac function, lymphocyte count and description of serious infections) (Table 2).

Real‐world cohort characteristics and trial eligibility

The real‐world cohort comprised 180 patients, contributing a total of 320 distinct relapse episodes (median, 1 relapse per patient; range 1–8). At first relapse, occurring between 2012 and 2022, the median age was 65 years (range 20–98); 58% of patients were male, and 51% had a revised International Prognostic Index (R‐IPI) score of 3–5, consistent with poor‐risk disease. ²³ Basic cohort characteristics and standard of care treatment details are summarised in Table S1.

When eligibility criteria from the registrational trials were retrospectively applied to the real‐world cohort, marked variability in hypothetical eligibility was observed across the studies. Overall, 94/180 patients (52%; 95% CI: 0.45, 0.59) were not eligible for any trial at any relapse, while 86/180 (48%; 95% CI: 0.41, 0.55) met eligibility criteria for at least one trial. No patients were eligible for all trials at any time of relapse (Figure 1). When eligibility was assessed at the level of individual relapse episodes, SADAL (NCT 02227251) demonstrated the lowest proportion of eligibility, with only 12 of 320 relapse episodes (4%, 95% CI: 0.02, 0.06) meeting trial criteria. In contrast, ZUMA‐1 (NCT 02348216) was the most permissive, with 69 of 320 relapse episodes (22%; 95% CI: 0.17,0.26) deemed eligible (Table 3).

FIGURE 1.

Eligibility of patients (N = 180) for each clinical trial at any relapse episode. (A) Proportion of patients eligible (pie chart). (B) Tabulated distribution of patients by the number of trials for which they were eligible.

TABLE 3.

Percentage of patients eligible for individual trials at each relapse episode (N = 320).

Trial	Intervention	Eligibility at any relapse episode (N = 320)	Median progression‐free survival (PFS)eligible (E) vs. non‐eligible (NE) months (95% CI)	Median number of criteria failed
SADAL (NCT02227251) 2	Selinexor	4% (12/320; 95% CI: 2–6%)	E: 4.5 (0.6–10.1) NE: 6.9 (5.3–9.1)	3 (range: 0–8)
LOTIS‐2 (NCT03589469) 5	Loncastuximab	9% (29/320; 95% CI:6–13%)	E: 3.7 (2–5.18) NE: 7.9* (6.1–10)	1 (range: 0–7)
GO29365 (NCT02257567) 3	Polatuzumab + BR	11% (36/320; 95% CI:8–15%)	E: 10.2 (4.7–15.5) NE: 6.3 (4.9–8.0)	2 (range 0–9)
L‐MIND (NCT02399085) 4	Tafasitamb + lenalidomide	11% (36/320; 95% CI: 8–15%)	E: 10.75 (5.6–22.8) NE: 6.3 (4.9–8.0)	2 (range: 0–6)
JULIET (NCT02445248) 6	Tisagenlecleucel	12% (38/320; 95% CI:9–16%)	E: 9.25 (4.7–14.7) NE:6.5 (4.9–8)	2 (range: 0–6)
TRANSCEND (NCT02631044) 7	Lisocabtagene maraleucel	13% (40/320; 95% CI: 9–17%)	E: 5.7 (3.8–9.3) NE: 6.8 (5.2–9.3)	1 (range: 0–5)
ZUMA‐1 (NCT02348216) 8	Axicabtagene ciloleucel	22% (69/320; 95% CI: 17–26%)	E: 4.32 (2.6–6.1) NE: 8.6** (6.1–10.7)	1 (range: 0–6)

Note: *p = 0.002 and **p < 0.0003 Gehan–Breslow–Wilcoxon test.

The most common reason for ineligibility was failure to meet requirements for the trial‐mandated number of prior lines of therapy. In four of the seven trials, this represented the most frequent reason for exclusion, with between 41% and 70% of patients excluded for having either too few or many prior lines of treatment (Figure 2). Other frequently observed exclusion factors included poor performance status, active infection and impaired renal function. The most frequent reasons for ineligibility across trials is provided in Figure 2. Some patients reached eligibility for more trials over time and with further relapses, subject to many factors including number of prior therapies. Eligibility proportions related to the number of lines of prior treatment is shown in Table S2.

FIGURE 2.

Most common reasons for eligibility failure across all relapse episodes (N = 320). (A) The three most frequent exclusion criteria for each trial (n/320, %). (B) The stacked histogram visually depicts relative contributions. Percentages exceed 100% because patients could meet more than one exclusion criteria for each trial.

Survival outcomes

With a median follow‐up of 62.8 months (range 0.03–123), the median OS from first relapse was 21.8 months (range 0.03–123) and the median PFS across all relapse episodes was 6.8 months (range 0.03–116). In testing survival differences between patients who were hypothetically eligible versus those who were not, we identified that PFS estimates for trial‐eligible and trial‐ineligible patients in the real‐world cohort overlapped with PFS outcomes reported in the registrational clinical trials from the clinical trials, indicating no difference between the groups (Figure 3).

FIGURE 3.

Median progression‐free survival (PFS) from registrational trials compared with median PFS for relapse episodes from the real‐world cohort, stratified by trial eligibility (E, eligible; NE, not eligible), n indicates the number of real‐world patients contributing relapse episodes (N = 320). Error bars represent 95% confidence intervals.

No significant difference in OS was observed between patients who were ineligible for all trials and those eligible for at least one trial (Figure S1). At data cut‐off, 111 of 180 patients (61%, 95% CI: 0.54, 0.68) had died, including 78 deaths attributable to lymphoma and six attributed to treatment‐related toxicity. Sixty‐three patients (35%) were alive, of whom 33 had ongoing lymphoma. Ten patients (5.5%; 95% CI: 0.03, 0.10) were excluded from OS analysis due to unknown date of death, and survival status was unavailable for a further six patients (Table S3).

DISCUSSION

This analysis demonstrates that eligibility criteria in recent accelerated approval trials for rDLBCL were highly heterogeneous, with discordance observed in greater than 80% of eligibility domains examined, across trial, patient and disease‐specific characteristics. Even ostensibly universal criteria such as measurable disease were inconsistently defined, with substantial variation in lesion size and detection requirements.

Despite all trials in this analysis evaluating novel therapies in transplant‐ineligible rDLBCL, eligibility criteria varied markedly between studies. This lack of harmonisation limits meaningful cross‐trial comparisons and complicates clinical decision‐making. Even for individual regimens, determining patient suitability in routine practice to achieve outcomes comparable to those reported in trials ideally requires hypothetical eligibility for the original study, and with a median of 39 eligibility criteria per trial, this necessitates meticulous protocol review.

Furthermore, our study demonstrates that clinical trial populations are heavily selected, yet this potentially did not impact outcomes when all patients received standard of care chemotherapy‐based treatments. In our cohort, more than 50% of real‐world patients with rDLBCL did not meet eligibility criteria for any of the included trials. Eligibility rates varied substantially between studies, ranging from 4% for the SADAL trial [NCT02227251] to 22% for the ZUMA‐1 trial [NCT02348216], and no patients met eligibility criteria for all seven trials.

Prior studies have suggested that meeting eligibility criteria may carry prognostic significance; however, this was not observed in our cohort. ²⁴ There was no significant difference in OS or PFS between patients who met eligibility criteria for at least one trial compared to those who did not meet criteria for any trials. This finding suggests that, in our population, exclusion from clinical trials did not simply reflect a more frail or poor prognosis group. Rather, it highlights that restrictive eligibility criteria may limit the generalisability of trial results without clearly enriching for a population with superior outcomes.

Accelerated approvals by the US‐FDA and European Medicines Agency (EMA) are frequently based on non‐randomised phase 1b/2 studies, such as those included in this analysis. While stringent eligibility criteria in early‐phase trials facilitate attribution of outcomes to treatment effect by creating relatively homogenous cohorts, increasing eligibility stringency over time has contributed to discrepancies between trial outcomes and post‐marketing real‐world studies in rDLBCL. ⁹ , ¹⁵ , ¹⁶ , ¹⁷ , ²⁵ As observed across many studies, eligibility criteria were not consistently aligned with agent‐specific characteristics. For example, renal function thresholds varied substantially (creatinine clearance ≥30 mL/min to ≥60 mL/min) without clear justification based on drug pharmacokinetics or phase 1/preclinical data, suggesting an element of arbitrariness. Similarly, haematological thresholds were not more permissive for agents with lower expected myelotoxicity.

Our findings are consistent with prior real‐world analyses of individual regimens. In a cohort of tafasitamab–lenalidomide‐treated rDLBCL, only 12% met eligibility criteria for the registrational trial, with a median PFS of 1.9 months compared with 12.1 months reported in the pivotal study. ⁴ , ¹⁷ Similarly, in patients treated with polatuzumab–rituximab–bendamustine, 74% were ineligible for the G029365 trial (NCT02257567), with median PFS and OS of 2.3 and 3.5 months, respectively, compared with 9.5 and 12.4 months reported by the original trial. ³ , ¹⁴ These dismal outcomes defy acceptable efficacy levels needed for approvals. Analyses in indolent lymphomas show similar results with more than 50% of patients failing landmark phase III trial eligibility. ²⁶ , ²⁷

Collectively, these findings highlight a critical disconnect between regulatory trial populations and the patients for whom therapies are ultimately approved. Many real‐world patients are excluded from the trials underpinning registration and consequently do not achieve the magnitude of benefit suggested by published outcomes. This has important implications for both clinical practice and drug development. When eligibility criteria vary substantially and trial populations are highly selected, quantifying the incremental benefit of new therapies over existing standards becomes inherently challenging. Treatment selection and sequencing depend on robust, generalisable evidence to support shared decision‐making between clinicians and patients. Our findings indicate that the applicability of trial‐derived efficacy estimates to real‐world populations is diminishing as eligibility criteria in modern trials become increasingly restrictive and exclusionary.

While early‐phase studies designed to establish safety and preliminary efficacy appropriately enrol relatively fit and homogenous populations, registrational trials should more closely reflect patients in routine clinical practice. Failure to do so risks exposing patients to unanticipated toxicity and treatment futility once therapies are implemented outside this trial setting. An additional consideration is that the efficacy observed in pivotal registrational trials may, at least in part, reflect the deliberate selection of more homogeneous and clinically fit patient populations. This may be influenced by regulatory and sponsor priorities to facilitate timely drug approval, potentially resulting in an overestimation of treatment benefit when applied to broader, unselected populations encountered in routine clinical practice.

Regulatory agencies, therefore, have an important role in encouraging broader and more representative eligibility criteria during trial development. Increasingly, high quality real‐world data are available through disease‐based clinical quality registries (CQRs) such as the Lymphoma and Related Diseases Registry (LaRDR), a multicentre collaborative lymphoma database established in Australia in 2016. ²⁰ Such registries systematically capture outcomes from populations intended to receive novel therapies and provide valuable context for benchmarking trial eligibility against representative patient cohorts, thereby informing more inclusive and generalisable trial design.

This retrospective study is subject to some limitations, including missing data, and the potential for selection bias due to heterogeneity in clinician‐directed treatment decisions and variability in post‐treatment monitoring and response assessment. Although noting that there were changes to therapy and supportive care during the study period internationally for rDLBCL which may have some influence on survival in such jurisdictions, there were no new drug approvals for this group in the Australian environment; thus, the data need to be analysed in this context.

CONCLUSION

Registrational trials in rDLBCL employ stringent and highly variable eligibility criteria that exclude most real‐world patients, substantially limiting the clinical applicability of reported outcomes. Our findings add to the growing evidence supporting the need for harmonised, less restrictive eligibility frameworks across trials within the same disease indication. Integrating real‐world data early in drug development is critical to aligning study populations with routine clinical practice and ensuring that future therapies deliver meaningful benefit to the patients most likely to receive them.

AUTHOR CONTRIBUTIONS

Elizabeth Goodall: Methodology; conceptualization; investigation; validation; writing – original draft; writing – review and editing; formal analysis; data curation; project administration. Tran Binh Giang: Investigation; validation; writing – review and editing; data curation. Eliza A. Hawkes: Conceptualization; methodology; investigation; validation; writing – original draft; writing – review and editing; data curation; supervision; resources; project administration. Fiona Swain: Writing – review and editing; validation; investigation; data curation. Arina Martynchyk: Methodology; formal analysis; writing – review and editing. Gareth P. Gregory: Investigation; validation; writing – review and editing; data curation. Niamh Waters: Data curation; writing – review and editing; validation. Jodie Palmer: Formal analysis; methodology; writing – review and editing; writing – original draft. Jamie Brennan: Investigation; validation; data curation; writing – review and editing. Tamara Marconi: Investigation; validation; writing – review and editing; data curation. Colm Keane: Investigation; validation; writing – review and editing; data curation. Chan Cheah: Investigation; validation; writing – review and editing; data curation.

FUNDING INFORMATION

E.G is funded by a HSANZ/Leukaemia Foundation PhD Scholarship. E.H and C.K are funded by National Health and Medical Research Council Investigator Grants.

CONFLICT OF INTEREST STATEMENT

E.G has received honoraria and travel expenses from Abbvie, and travel expenses from Gilead. E.A.H. has received Research funding (paid to institution): Roche, Bristol Myers Squibb, Merck KgA, Astra Zeneca, TG therapeutics Merck and Abbvie; Consultant or advisory roles (*Paid to institution): Roche*, Merck Sharpe & Dohme*, Astra Zeneca*, Gilead, Antengene*, Novartis*, Regeneron, Janssen*, Specialised Therapeutics*, Sobi*, GSK* Travel expenses: Astra Zeneca, Genmab, Abbvie. C.K. has received honoraria from Roche, BeiGene, Karyopharm Therapeutics, and Gilead. C.C has received consulting/advisory/honoraria—Roche, Janssen, Gilead, AstraZeneca, Lilly, Beigene, Menarini, Dizal, Abbvie, Genmab, Sobi, CRISPR Therapeutics. BMS, Regeneron. Speaker's bureau—Janssen, AstraZeneca, Beigene, Genmab, Abbvie, Roche, MSD. Research funding—BMS, Roche, Abbvie; MSD, Lilly. Travel expenses—Lilly, Beigene. F.S has received honoraria from Gilead. G.P.G. has received research funding from Merck KgA and BeiGene (paid to institution); has held consultant or advisory roles for Roche, Merck Sharp & Dohme, AstraZeneca, Gilead, Janssen, Clinigen, and Prelude Therapeutics; received travel expenses from Novartis; and is a member of the International Burkitt Lymphoma Network.

The remaining authors declare no competing financial interests.

ETHICS STATEMENT

The study was approved by the local institutional review board (HREC approval: LNR/17/Austin/186).

PATIENT CONSENT STATEMENT

The institutional review board approved a waiver of consent in accordance with the National Health and Medical Research Council and institutional guidelines for negligible‐risk research involving de‐identified data from retrospective medical records.

Supporting information

Data S1.

DATA AVAILABILITY STATEMENT

Data are available upon request.