Bridging the evidence gap in trial inclusion with real world safety profile assessment in relapsed/refractory multiple myeloma patients

Abstract

Background

The underrepresentation of key populations in clinical trials can create critical gaps in our understanding of treatment efficacy and the generalizability of safety outcomes across broader patient populations. Ensuring that therapies are rigorously tested in diverse cohorts, while maintaining patient safety within trials, is essential for achieving optimal outcomes for all patients, regardless of medical history or comorbidities. Advancing inclusivity in clinical trials therefore requires a critical reassessment of eligibility criteria. In this context, real-world evidence provides a valuable opportunity to inform clinical trial design by assessing the safety implications of including patients that are typically excluded.

Methods

We proposed a quantitative framework leveraging real-world data and adopting a population-level approach rather than evaluating isolated eligibility criteria. The proposed method focuses on underrepresented patient subgroups, particularly those with pre-existing conditions, with applications in oncology, specifically in relapsed or refractory multiple myeloma (RRMM). Clinically relevant subpopulations and associated safety outcomes in RRMM were identified in collaboration with clinical expert. Leveraging real-world data, we assessed the risk of safety events associated with whether a patient had or did not have a specific baseline condition (e.g., renal impairment, cardiac dysfunction, etc.). These analyses were conducted separately for anti-CD38 and standard-of-care (SoC) treatments. Inverse probability of treatment weighting (IPTW) was applied to balance baseline characteristics between patient cohorts who met versus did not meet each baseline condition. IPTW-adjusted Kaplan-Meier curves and weighted Cox proportional hazards models were used to evaluate associations between key baseline conditions and the risk of safety events, expressed as hazard ratios.

Results

Our findings indicate that patients with RRMM and renal impairment face significantly increased risks of severe renal events but do not exhibit significantly elevated risks of severe hematologic events or infections. Patients with baseline cardiac dysfunction are at higher risk of both infections and cardiovascular events, with an additional increased risk of severe renal events observed in the SoC cohort. Furthermore, patients with more than two prior lines of therapy exhibited an increased risk of severe hematological events in the anti-CD38 cohort. In the SoC cohort, patients with more than two prior lines of therapy demonstrated a higher risk of infections and cardiovascular events compared with those who had received two or fewer prior lines.

Conclusion

These results provide descriptive and adjusted real-world estimates of the association between key baseline clinical conditions and the risk of safety event among patients with RRMM and may help inform clinical trial eligibility criteria. By characterizing safety outcomes in populations often underrepresented in clinical trials, this analysis offers contextual evidence to support more inclusive trial designs, while acknowledging the inherent limitations of observational data.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12885-026-16222-9.

Introduction

A persistent challenge in clinical research—and a growing regulatory priority—is enhancing trial inclusivity. To address this, study sponsors are now encouraged to submit diversity action plans aimed at increasing the enrollment of underrepresented populations, thereby ensuring that study cohorts better reflect the patient populations who will ultimately receive the intervention [1, 2]. These plans must define diversity targets and outline specific enrollment strategies. The trial exclusion criteria were often originally established to prioritize patient safety. While expanding eligibility to include certain underrepresented groups may broaden trial participation, a critical aspect involves reassessing populations historically underrepresented in trials while not compromising clinical trial scientific validity or patient safety.

The use of real-world evidence (RWE), derived from the analysis of real-world data (RWD), is increasingly informing clinical trial design and decision-making in drug development and regulatory evaluation. Several initiatives have sought to reassess clinical trial eligibility criteria using RWD reflecting routine clinical practice. Data-driven artificial intelligence (AI) approaches have been applied to relax criteria without compromising efficacy outcomes [3]. Because eligibility criteria are also designed to ensure patient safety, the impact of broadening these criteria on safety should be evaluated. Liu et al. described assessing relaxing enrollment criteria safety using a historical clinical trial database. Rogers et al. utilized electronic health record (EHR) data to analyze how individual and combined eligibility criteria affect patient counts and safety outcomes, exemplified by all-cause hospitalization risk [4]. Evaluating safety in real-world setting provides a comprehensive and robust approach by capturing diverse patient populations, comorbidities, and treatment regimens and mitigates the limitation of historical clinical trial safety database. Nevertheless, further research is needed to validate safety profiles derived from RWD and to determine their appropriateness for regulatory or clinical decision-making.

This paper explores strategies to improve inclusivity in clinical trials for relapsed or refractory multiple myeloma (RRMM). A recent systematic review by White et al. evaluated eligibility criteria across 85 multiple myeloma randomized controlled trials conducted between 2006 and 2023, highlighting the extensive use of restrictive exclusion criteria. Common exclusions included poor performance status (87%), prior malignancy (80%), hepatic (79%) or renal dysfunction (74%), and requirements for measurable disease (55%). Additional exclusions frequently applied were hematologic thresholds (59%) and peripheral neuropathy (59%). Less commonly, trials excluded patients with plasma cell leukemia (19%), CNS involvement (11%), or extramedullary disease (5%). Age limits (51%) and projected life expectancy criteria (24%) were also used [5].

Kanapuru et al. highlighted that specific eligibility criteria may indirectly contribute to enrollment disparities among racial and ethnic subgroups, with black multiple myeloma patients being more likely to be excluded due to failure to meet hematology laboratory criteria. These criteria typically include hemoglobin level, absolute neutrophil count, platelet count, and other blood cell count parameters [6]. Additionally, Jreich et al. suggested that removing selected exclusion criteria—such as a history of HIV, asthma, chronic obstructive pulmonary disease, or cardiac conditions—and relaxing certain laboratory thresholds (e.g., absolute neutrophil count, creatinine clearance, and platelet count) could potentially increase patient enrollment without compromising efficacy outcomes, as measured by the hazard ratio of progression-free survival (PFS) [7]. Nevertheless, final decisions regarding eligibility criteria should be informed by a balanced assessment of both efficacy and safety considerations.

Rather than focusing on individual eligibility criteria—which may be challenging to replicate using RWD or may exclude only a small number of patients, thereby limiting the generalizability of findings—this paper emphasizes underrepresented populations with prior complications or comorbidities from a clinical perspective. We propose a quantitative workflow to assess the associations between the inclusion of these subpopulations and the risk of safety outcomes, leveraging evidence derived from RWD. The findings aim to inform the design of future RRMM studies, with a focus on achieving broader representation of diverse patient population.

Methods

A structured analytical framework was applied to quantitatively assess the association between key baseline subpopulations and the occurrence of safety events using RWD (Fig. 1).

Fig. 1.

Structured quantitative framework to evaluate the impact of key subpopulations on safety events using Real-World-Data

Step 1: Clinical expert interview to inform inclusion framing

This structured quantitative framework was designed to evaluate the feasibility of broadening clinical trial inclusion in RRMM. The clinical expert contributing to this work brings extensive experience in global oncology development programs, providing insights highly relevant to clinical trial design and patient population considerations. Expert consultations identified four key baseline subpopulations for evaluation: renal impairment, cardiac dysfunction, Line of therapy (LoT) > 2, and patients with a history of malignancy within the past 5 years with no evidence of active cancer in the 6 months preceding the index date. For each baseline subpopulation, clinically defined safety events were assessed by considering the potential safety risks associated with including patients with each condition (Table 1). The liver impairment subpopulation was also considered but ultimately not included, as patients with mild to moderate liver impairment patients are already routinely enrolled in RRMM clinical trials, while severe liver impairment, although clinically relevant, was deemed less feasible to assess using RWD.

Table 1.

Key baseline subpopulations and associated safety concerns identified based on clinical expertise

Subpopulation /safety	Severe renal events	Cardiac events	Severe hematologcal events	Severe infections	Malignancy recurrence
Renal impairment	X		X	X
Cardiac dysfunction	X	X
Other malignancy					X
LoT  > 2	X	X	X	X

Step 2: RWD cohort selection

The real-world data source used in this framework was Flatiron Health database, a USA nationwide longitudinal EHR derived database focusing on oncology. Most of the patients in the database originate from community oncology settings; relative community/academic portions may vary depending on study cohort. The data originated from approximately 280 US cancer clinics (~ 800 sites of care). The data is de-identified and subject to obligations to prevent re-identification and to protect patient confidentiality. It comprises patient-level structured and unstructured data, curated via technology-enabled abstraction [8, 9].

Adult patients with MM were identified in the database by screening relevant diagnostic codes from the International Statistical Classification of Diseases, Ninth Revision (ICD-9: 203.0) and Tenth Revision (ICD-10: C90.00). Patients with RRMM were defined as those who received two to five LoT between 1 January 2017 and 30 April 2024 and had at least one recorded MM diagnosis prior to the index date, defined as the start of the LoT of interest. Regimens of interest included daratumumab-based combinations (with pomalidomide, carfilzomib, lenalidomide, or bortezomib, each plus dexamethasone) and non-daratumumab regimens (pomalidomide, carfilzomib, lenalidomide, or bortezomib, each combined with dexamethasone). Patients receiving any of the four daratumumab-containing regimens were classified into the anti-CD38 cohort, whereas those treated with the remaining four regimens were categorized into the standard-of-care (SoC) cohort. For patients who received one or more of the specified regimens on or after the index date, cohort assignment was based on the first regimen received. For example, if a patient received one cycle of SoC followed by an anti-CD38 regimen, the patient was assigned exclusively to the SoC cohort and not to the anti-CD38 cohort. This new-user approach ensured that cohorts remained mutually exclusive, thereby preventing double counting of individuals across treatment groups. In addition, assigning patients based on their earliest qualifying regimen helps minimize confounding by prior treatment exposure, as safety profiles and clinical outcomes may differ depending on prior exposure to SoC before initiation of subsequent therapies. Eligible patients were required to initiate one of eight treatment regimens at the index date and to maintain therapy for a minimum of 28 days post-index.

Step 3: Baseline condition and safety event derivation

As detailed in Step 1, four baseline subpopulations were evaluated based on clinical expertise, with derivation details presented in Table 2.

Table 2.

Rule-based definitions of baseline subpopulations using Flatiron Health database

Baseline Sub-populations	Time before index date	Derivation
Renal impairment	60 days	Labs (Estimated Glomerular Filtration Rate < 30mL/min/1.73m2, Creatinine Clearance < 30mL/min), ICD9-10 diagnosis codes, Dialyses
Cardiac dysfunction	24 months	ICD9-10 diagnosis codes
LoT > 2	At index date	LoT initiation at index date
Patients with history of malignancy	Within the past 5 years and no evidence of cancer in the 6 months preceding the index date	ICD9-10 diagnosis codes

For each baseline subpopulation, we evaluated related safety events, including hematological, renal, cardiovascular, infectious, and malignancy recurrence events (Further details on the ICD diagnosis codes used to define baseline conditions are provided in Table 1 of the Supplementary Material). Safety events were identified using criteria outlined in Table 2, incorporating laboratory test results, ICD9-10 diagnostic codes; additional details on the ICD diagnosis codes used to derive safety events are provided in Table 2 of the Supplementary Material. Safety events were assessed during the treatment follow-up period, defined as the time from the index date to the earliest of treatment discontinuation for the LoT, three years post-index, or death. We would like to focus particularly on the occurrence of severe safety events. Event severity was classified according to the common terminology criteria for adverse events (CTCAE) version 5.0 used in clinical trials; when unavailable, proxy measures were applied [10]. However, the severity of cardiovascular events and infections was not assessed due to the lack of relevant laboratory measurements, procedure information, medication use, and hospitalization data in the Flatiron Health Database. Severe renal and hematological events were identified using a combination of ICD9-10 codes and laboratory measurements (see Tables 3 and 3 of the Supplementary Material for additional details). For patients with a history of malignancy, malignancy recurrence was assessed from the index date until loss to follow-up, as determined by the last available information in the Flatiron Health database.

Table 3.

Safety event derivation using Flatiron Health database

Safety events	Grade	Derivation	Timeframe for assessment
Hematological	Grade 3	Labs (Hemoglobin, Platelet, Absolute Neutrophil Count), ICD-9-10 Diagnosis codes	From index date until first of: end of LoT, death, 3 years
Renal	Grade 3	Labs (Estimated Glomerular Filtration Rate, Creatinine Clearance), ICD9-10 Diagnosis codes
Cardiovascular	All grade	ICD9-10 Diagnosis codes
Infections	All grade	ICD9-10 Diagnosis codes
Malignancy recurrence	-	ICD9-10 Diagnosis codes	From index date until patient’s loss to follow-up

Step 4: Quantitative assessment

This step involves the statistical methodology development to quantitatively assess the impact of having versus not having a given baseline condition on the likelihood of experiencing safety events (Fig. 2). Comparative statistical analyses may be conducted by comparing the rate or percentage of safety event occurrence between patients with and without the baseline condition, or by estimating the hazard ratio for time to first safety event between the two groups. Regardless of the analytical approach chosen, analyses should be performed after the implementing of matching or weighting techniques to reduce confounding arising from baseline differences between groups. This ensures that observed differences in safety outcomes can be more reliably attributed to the baseline condition itself rather than to underlying confounders. Furthermore, safety outcome comparisons are preferably conducted within treatment drug classes. By isolating the effect of baseline conditions, independently of the drug class administrated during follow-up, this approach provides more robust insights into safety risks. Consequently, it supports informed decisions regarding subpopulation inclusion while ensuring patient safety, even when SoC therapies are used, thereby offering an additional layer of assurance in clinical evaluation. For our use case, we adopted a weighted time-to-safety-event analysis, which provides more informative insights than simple comparisons of event occurrence proportions. In particular, this approach captures differences in the timing of event occurrence, which may be masked when baseline conditions yield similar overall event rates during follow-up. A weighted Cox proportional hazards model was applied using inverse probability of treatment weighting (IPTW) [11, 12]. Propensity scores for having versus not having a given baseline condition were estimated using logistic regression models adjusted for prespecified baseline covariates, including age, sex, race, renal impairment, hematological abnormalities, cardiac dysfunction, and number of prior lines of therapy. IPTW was applied separately for each baseline condition of interest and within each treatment cohort (anti-CD38 and SoC).

Fig. 2.

Strategy workflow for quantitative assessment of key sub-populations impact on safety events. A Two cohorts were created based on their treatment regimens. B Within these two cohorts were captured patients with specific baseline conditions who were weighted with their counterpart. C Once balanced, within both cohorts, for each baseline condition, patients with the condition were compared with those without the condition through survival analysis

To minimize bias due to confounding by other baseline conditions, propensity score models were additionally adjusted for baseline impairments not under investigation. For example, when assessing the impact of renal impairment on safety events within the anti-CD38 cohort, the propensity score model included demographic characteristics (age, sex, race), number of prior lines of therapy, and other baseline impairments (hematological abnormalities and cardiac dysfunction), but excluded renal impairment itself from the adjustment set. Weighted Cox proportional hazards models incorporating the IPTW weights were then fitted to estimate hazard ratios comparing patients with versus without the baseline condition. These hazard ratios quantify the association between baseline condition status and the risk of safety event occurrence over time.

Standardized mean differences (SMDs) were used to assess the balance of baseline characteristics after weighting, with a threshold of 0.1 defining meaningful imbalance.

Results

Study population: demographics and clinical characteristics

A total of 18,377 patients with MM were initially identified from the Flatiron Health database. After applying the inclusion and exclusion criteria described in the Methods section, the study included 2,569 patients with RRMM who had received 2 to 5 lines of therapy (Fig. 3). Among these patients, 1,206 (46.9%) were in the anti-CD38 cohort and 1,363 (53.1%) were in the SoC cohort.

Fig. 3.

Attrition chart for the creation of the Anti-CD38 cohort and the Standard of Care cohort

Of these patients, 47.8% were female, with a median age of 70 years at the index date and 67 years at diagnosis. The average follow-up duration was 11.8 months. The racial distribution was predominantly White (60.9%), followed by Black or African American (18.8%), Hispanic or Latino (4.7%), Asian (2.1%), and Other/Unknown (13.5%) (Table 4). When comparing cohorts, the anti-CD38 cohort was slightly younger than the SoC cohort, with median ages of 68 years versus 71 years at the index date and 65 years versus 69 years at diagnosis, respectively. The anti-CD38 cohort also had a slightly lower proportion of females (45.4% vs. 50.0%). Both cohorts had similar distributions in race/ethnicity, International Staging System (ISS) stage at diagnosis, and Eastern Cooperative Oncology Group (ECOG) performance status. However, the anti-CD38 cohort had a higher proportion of patients who had received more than 2 LoT (46.4% vs. 34.6%) and who had been previously treated with lenalidomide (90.4% vs. 84.3%). The distribution of safety events observed during the follow-up period revealed that hematologic events were the most frequent, accounting 26.5% of all safety events, followed by infections (10.9%), severe renal events (8.2%), and cardiovascular events (7.8%).

Table 4.

Descriptive statistics: Socio-demographics, baseline conditions, and safety outcomes during follow-up. These abbreviations render to Poma.: pomalidomide, Carf.:carfilzomib, Lena.:lenalidomide, Bort.: bortezomib, Dexa.:dexamethasone and Dara.: daratumumab

		Real-world cohort N = 2,569	Anti-CD38 cohort N = 1,206	SoC cohort N = 1,363
Socio-demographic at baseline	Age at index date in years
	Mean (sd)	68.7 (10.6)	66.9 (10.2)	70.3 (10.7)
	Min – Max	28.0, 85.0	28.0, 85.0	30.0, 85.0
	Median (Q1 - Q3)	70.0 (61.0, 77.0)	68.0 (60.0, 75.0)	71.0 (64.0, 79.0)
	Race or ethnicity
	White	1,564 (60.9%)	734 (60.9%)	22 (1.6%)
	Asian	482 (18.8%)	223 (18.5%)	259 (19.0%)
	Black or African American	121 (4.7%)	64 (5.3%)	57 (4.2%)
	Hispanic or Latino	348 (13.5%)	153 (12.7%)	195 (14.3%)
	Other/Unknown	54 (2.1%)	32 (2.7%)	830 (60.9%)
	Sex
	Female	1,228 (47.8%)	547 (45.4%)	681 (50.0%)
	Male	1,341 (52.2%)	659 (54.6%)	682 (50.0%)
Subpopulations at baseline	Cardiac dysfunction	407 (15.8%)	198 (16.4%)	209 (15.3%)
	Liver impairment	63 (2.5%)	28 (2.3%)	35 (2.6%)
	Renal impairment	155 (6.0%)	81 (6.7%)	74 (5.4%)
	History of malignancy	192 (7.5%)	95 (7.9%)	97 (7.1%)
	Line of Therapy
	2	1,539 (59.9%)	647 (53.6%)	892 (65.4%)
	> 2	1,030 (40.1%)	559 (46.4%)	471 (34.6%)
Safety outcome during follow-up	Severe Renal Event	211 (8.2%)	94 (7.8%)	117 (8.6%)
	Severe Hematological Event	682 (26.5%)	438 (36.3%)	244 (17.9%)
	Cardiovascular event	201 (7.8%)	106 (8.8%)	95 (7.0%)
	Infection	281 (10.9%)	148 (12.3%)	133 (9.8%)
Treatment regimens	Dara.+ Bort. + Dexa.	235 (9.1%)	235 (19.5%)	-
	Dara. + Carf.+ Dexa.	138 (5.4%)	138 (11.4%)	-
	Dara. + Lena. + Dexa.	344 (13.4%)	344 (28.5%)	-
	Dara.+ Poma. + Dexa.	489 (19.0%)	489 (40.5%)	-
	Poma. + Dexa.	194 (7.6%)	-	194 (14.2%)
	Carf. + Dexa.	214 (8.3%)	-	214 (15.7%)
	Lena. + Dexa.	613 (23.9%)	-	613 (44.9%)
	Bort. + Dexa.	342 (13.3%)	-	342 (25.1%)
	1	1,539 (59.9%)	647 (53.6%)	892 (65.4%)
	2	664 (25.8%)	347 (28.8%)	317 (23.3%)
Number of Prior Treatment lines	3	254 (9.9%)	148 (12.3%)	106 (7.8%)
	4	112 (4.4%)	64 (5.3%)	48 (3.5%)
Follow-up duration (Days)	Mean (SD)	359.2 (326.7)	354.2 (320.4)	363.6 (332.2)
	Median (Q1, Q3)	225.0 (111.0, 501.0)	216.5 (118.0, 487.0)	236.0 (103.0, 521.0)

Safety outcomes in RRMM patients with renal impairment

Among the 2,569 RRMM patients included in the study, 155 (6%) were identified as having renal impairment at the index date. Of these, 81 patients were in the anti-CD38 cohort, and 74 were in the SoC cohort. Survival analyses demonstrated a significantly increased risk of severe renal events among patients with renal impairment in both the anti-CD38 cohort (HR = 17.8, 95% CI: 11.0-28.9, p < 0.05) and the SoC cohort (HR = 14.72; 95% CI: 9.78–22.16, p < 0.05), as illustrated by the IPTW-Adjusted Kaplan-Meier curves (Fig. 4). These elevated hazard ratios observed across both cohorts, regardless of treatment regimen, are likely attributable to compromised baseline renal function, as patients with renal impairment at baseline are more likely to develop severe renal events during follow-up compared to those without renal impairment. Although the difference did not reach statistical significance, a modest trend toward an increased risk of severe hematological events was observed in both cohorts.

Fig. 4.

IPTW-adjusted Kaplan-Meier survival analysis comparing patients with and without renal impairment in the anti-CD38 cohort (upper panel) and SoC (lower panel). Hazard Ratio was estimated using weighted Cox proportional hazards model. Patients were compared on infections (A), severe renal events (B), and severe hematological events (C)

Safety outcomes in RRMM patients with cardiac dysfunction

Of the 2,569 RRMM patients included in the study, 407 (15.6%) had cardiac dysfunction at the index date, with 198 in the anti-CD38 cohort and 209 in the SoC cohort. Survival analyses revealed that patients with cardiac dysfunction had an elevated risk of infections, cardiovascular events, and severe renal complications in both cohorts (Fig. 5). In the anti-CD38 cohort, patients with cardiac dysfunction demonstrated significantly higher risks of infections (HR = 3.39, 95% CI: 2.38–4.83, p < 0.001) and cardiovascular events (HR = 6.08, 95% CI: 4.00-9.22, p < 0.001). Similarly, in the SoC cohort, increased risks were observed for infections (HR = 2.58, 95% CI: 1.75–3.81, p < 0.001) and cardiovascular events (HR = 7.26, 95% CI: 4.82–10.92, p < 0.001). Additionally, patients with cardiac dysfunction in the SoC cohort exhibited a significantly higher risk of severe renal events (HR = 1.80, 95% CI: 1.20–2.71, p = 0.004).

Fig. 5.

IPTW-adjusted Kaplan-Meier survival analysis comparing patients with and without cardiac dysfunction in the anti-CD38 cohort (upper panel) and SoC (lower panel). Hazard Ratio was estimated using weighted Cox proportional hazards model. Patients were compared on infections (A), severe renal events (B), and cardiovascular events (C)

Safety outcomes in RRMM patients received more than two LoT

Among 2,569 RRMM patients included in the study, 1,030 (40.1%) had received strictly more than 2 LoT. Of these patients, 559 (46.4%) were in the anti-CD38 cohort, while 471 (34.6%) were in the SoC cohort. Survival analyses showed that patients who had received strictly more than 2 LoT had a higher risk of severe hematological events in the anti-CD38 cohort; whereas in the SoC cohort, they were at higher risk of infections, and cardiovascular events.

A significantly higher risk of severe hematological events was observed in patients who had received strictly more than 2 LoT in the anti-CD38 cohort (HR = 1.47, 95% CI: 1.22–1.78, p < 0.001) (Fig. 6). In the SoC cohort, patients who had received more than 2 LoT showed a significantly higher risk of infections (HR = 1.43, 95% CI: 1.01–2.02, p = 0.046) and cardiovascular events (HR = 1.53, 95% CI: 1.01–2.31, p = 0.045).

Fig. 6.

IPTW-adjusted Kaplan-Meier survival analysis comparing patients at LoT = 2 and LoT > 2 in the anti-CD38 cohort (upper panel) and SoC (lower panel). Hazard Ratio was estimated using weighted Cox proportional hazards model. Patients were compared on infections (A), severe renal events (B), cardiovascular events (C), and severe hematological events (D)

For the subpopulation with a history of prior non-myeloma malignancy who were recurrence-free within 6 months before the index date, analyses were limited to descriptive statistics, as the small number of patients within each prior malignancy type precluded meaningful cohort-specific estimates. These results are presented in Figs. 2 and 3 of the Supplementary Material.

Discussion

This quantitative framework evaluates safety outcomes in patient populations that are traditionally excluded from clinical trials, including those with renal impairment, cardiac dysfunction, more than 2 LoT, and evidence of prior malignancy, to inform the potential expansion of clinical trial representativeness and address current limitations in multiple myeloma patient recruitment.

The severe renal impairment subpopulation, which is typically excluded from RCTs, exhibited an elevated risk of severe renal events in both anti-CD38 and SoC cohorts, likely attributable to compromised baseline renal function. Although the increase did not reaching statistical significance, a higher risk of severe hematological events was observed in both cohorts, consistent with existing literature documenting the prevalence of hematological toxicities across multiple myeloma therapies [13, 14]. Notably, our findings corroborate previous studies showing no significant difference in infection rates between patients with and without severe renal impairment in either treatment cohort [15].

Patients with cardiac dysfunction demonstrated a significantly higher risk of infections in both the anti-CD38 and SoC cohorts. Pneumonia was the most prevalent infection, followed by upper respiratory tract infections, across both cohorts when compared with patients without cardiac dysfunction. This increased susceptibility to infections among patients with RRMM can be attributed to the natural disease course [16], underlying comorbidities [17], and immune-modulating treatments [18, 19]. Furthermore, patients with cardiac dysfunction, particularly those with heart failure, exhibit an elevated risk of respiratory infections, which may arise not only from compromised immune responses and persistent systemic inflammation [20], but also from reduced mobility leading to hypoventilation and impairment airway clearance [21], as well as shared predispoding factors such as tabacco smoking, obesity, and diabetes [22, 23]. Regardless of treatment regimen, patients with cardiac dysfunction demonstrated significantly higher risks of cardiovascular events. Notably, the hazard ratios were estimated at 6.1 in the anti-CD38 and 7.3 in the SoC cohort.

Analysis of patients who received more than two LoT revealed no significant increase in severe renal events in either cohort. However, within the anti-CD38 cohort, these heavily pretreated patients exhibited a higher risk of severe hematological events compared with those receiving second-line therapy. This observation is consistent with the known safety profile of daratumumab-containing regimens observed in RCTs and its attributable to their CD38-targeting mechanism. Recent meta-analyses have confirmed that daratumumab-containing regimens are associated with increased risks of neutropenia, leukopenia, and thrombocytopenia in both relapsed/refractory and newly-diagnosed multiple myeloma patients [24, 25]. In the SoC cohort, patients with more than two prior LoT demonstrated an elevated risk of cardiovascular events, potentially reflecting the higher proportion of patients receiving carfilzomib or bortezomib-based regimens compared with those treated in second-line setting [26, 27]. The increased incidence of severe hematological events observed among heavily pretreated patients likely reflects cumulative myelosuppression from previous therapies, resulting in enhanced susceptibility to subsequent treatment-related toxicities.

The occurrence of second primary malignancy in MM patients has been well documented in recent studies [28, 29]. However, the recurrence of prior malignancies in MM patients remains poorly understood. Malignancies observed to have higher recurrence rates in RRMM patients tend to be those with high recurrence rates in general population, such as prostate and breast cancers [30].

In the present study, safety risks associated with baseline conditions were evaluated within each treatment cohort by comparing patients with versus without the baseline condition. However, future analyses could additionally assess whether the association between baseline conditions and safety outcomes differs by treatment strategy by directly comparing anti-CD38 therapy with SoC. Such an investigation could be conducted using a two-step matching or weighting approach: first, to ensure comparability between patients receiving anti-CD38 therapy and those receiving SoC, and second, to balance patients with versus without each baseline condition. Although this approach was not implemented in the current analysis, it could provide complementary insights by disentangling treatment effects from baseline patient characteristics, thereby supporting more individualized treatment decisions and advancing precision medicine in RRMM.

Limitations

Our findings should be interpreted in light of the study limitations. In particular, we highlight the potential mis-capture of safety events and population-specific characteristics. This may be due to misclassification of safety events and potential underreporting within EHRs relying on ICD diagnosis codes. For example, reliance on ICD diagnosis codes and claims data has demonstrated limited sensitivity for identifying cancer disease status, with reported sensitivities ranging from 50% to 60% [31, 32]. In this context, Karima et al. developed and applied natural language processing to identify distant cancer recurrence using EHR data and identified a substantially larger proportion of patients with distant recurrence in a breast cancer database (11.1%) compared with International Classification of Diseases coding (2.31%).

In addition, given the oncology-specific scope of the Flatiron Health database, information on non-oncology-related diagnoses (e.g., hospitalizations) was not available, and laboratory and treatment data were limited. As a result, the prevalence of cardiac dysfunction, infections, and cardiovascular events may have been underestimated. Furthermore, the severity of infections and cardiovascular events—typically assessable using hospitalization records, procedure data, or medication use—could not be evaluated. Although Grade ≥ 3 hematological and severe renal events were captured using laboratory values and ICD codes, misclassification remains possible due to incomplete laboratory records, variability in the timing of measurements relative to clinical events, and potential inaccuracies or underreporting in diagnosis coding.

Although propensity score weighting was applied to balance key covariates, residual confounding may persist, particularly from unmeasured comorbidities or physical conditions that were not captured in the database and that would be more rigorously controlled in RCTs.

Conclusion

This study introduces a quantitative framework for evaluating the safety impact of including subpopulations that are typically excluded from clinical trials in RRMM. Clinically relevant subpopulations and safety outcomes were identified through expert input and evaluated using RWD. Our findings indicate that patients with renal impairment face a significantly increased risk of severe renal events but do not exhibit elevated risks of severe hematological events or infections. Patients with cardiac dysfunction demonstrated increased risks of both infections and cardiovascular events across treatment cohorts, as well as an additional increased risk of severe renal events in the SoC cohort. Additionally, patients who had received more than two prior lines of therapy experienced and increased risk of infections and cardiovascular events within the SoC cohort. In parallel, among patients receiving anti-CD38 therapy, a higher number of pior lines of therapy was associated with a significantly increased incidence of severe hematological events.

While RWD can help identify subgroups that may be considered for broader inclusion, these findings support a cautious yet evidence-based approach. The observation that certain populations are at higher risk for safety outcomes underscores the need for rigorous safety monitoring when considering the inclusion of underrepresented populations in RRMM trials.

Abbreviations

AI

Artificial Intelligence

CNS

Central Nervous System

ECOG

Eastern Cooperative Oncology Group

EHR

Electronic Health Records

HIV

Human Immunodeficiency Virus

HR

Hazard Ratio

ICD

International Classification of Diseases

ISS

International Staging System

LoT

Line of Therapy

MM

Multiple Myeloma

PFS

Progression-Free Survival

RCT

Randomized Clinical Trial

RRMM

Relapsed or Refractory Multiple Myeloma

RWD

Real-World Data

RWE

Real-World Evidence

SoC

Standard of Care

Authors’ contributions

R. Jreich and Z. Meng conceived the analytical scope and developed the quantitative framework. P. Loustalot supervised and conceptualized the project. K. Lounaci and S. Levy analyzed the data and performed the modeling. R. Yuan provided medical expertise for interpretation of the results and supporting references. H. Velde contributed clinical expertise and helped design the experiments. All authors reviewed and approved the final version of this manuscript.

Funding

This study was funded by Sanofi.

Data availability

The data that support the findings of this study have been originated by Flatiron Health, Inc. Requests for data sharing by license or by permission for the specific purpose of replicating results in this manuscript can be submitted to (mailto: PublicationsDataAccess@flatiron.com) .

Declarations

Ethics approval and consent to participate

This study used deidentified data, which does not constitute human subjects research according to the Common Rule and national regulations. As such, it did not require ethics approval or patient informed consent. Interviews was not applicable to this study. Study was performed on de-identified data in compliance with applicable regulations and no efforts were made to re-identify individuals.

Consent for publication

Not appliacable.

Competing interests

Rana Jreich, Zhaoling Meng and Helgi Van De Velde are employees of Sanofi and hold shares and/or stock options in Sanofi.