Daratumumab‐lenalidomide‐dexamethasone vs standard‐of‐care regimens: Efficacy in transplant‐ineligible untreated myeloma

Abstract Daratumumab in combination with lenalidomide‐dexamethasone (D‐Rd) recently received FDA approval for the treatment of transplant‐ineligible patients with newly diagnosed multiple myeloma (NDMM). The present PEGASUS study compared progression‐free survival (PFS) in patients treated with D‐Rd in the MAIA trial and patients treated with common standard‐of‐care regimens from the Flatiron Health electronic health record‐derived deidentified database, which has data from patients treated primarily at community‐based oncology practices in the United States. Individual‐level patient data from both data sources were used to perform an anchored indirect treatment comparison (ITC) of D‐Rd to bortezomib‐lenalidomide‐dexamethasone (VRd) and bortezomib‐dexamethasone (Vd); lenalidomide‐dexamethasone (Rd) was the common anchor for the ITC. Hazard ratios (HRs) reflecting direct comparisons of PFS within MAIA (D‐Rd vs Rd) and Flatiron Health (VRd vs Rd; Vd vs Rd) were used to make ITCs for D‐Rd vs VRd and Vd, respectively. After application of MAIA inclusion/exclusion criteria and propensity‐score weighting, the Flatiron Health patients resembled the MAIA trial population on measured baseline characteristics. Based on the direct comparison within MAIA, treatment with D‐Rd was associated with a significantly lower risk of progression or death compared to Rd (HR 0.54; 95% CI 0.42, 0.71). Based on the ITCs, D‐Rd was associated with a significantly lower risk of progression or death compared to VRd (HR 0.68; 95% CI 0.48, 0.98) and Vd (HR 0.48; 95% CI 0.33, 0.69). In the absence of head‐to‐head trials comparing D‐Rd to VRd or Vd, the present ITC may help inform treatment selection in transplant‐ineligible patients with NDMM.

(Rd) alone in patients with transplant-ineligible NDMM (hazard ratio [HR] 0.56; 95% confidence interval [CI] 0.43, 0.73; P < .001). 1,2 In the US, common standard-of-care (SOC) regimens for the treatment of transplant-ineligible NDMM include lenalidomidedexamethasone (Rd), bortezomib-lenalidomide-dexamethasone (VRd), and bortezomib-dexamethasone (Vd), which together account for at least two-thirds of treatment regimens used to treat patients with transplant-ineligible NDMM. 3,4 In recent years, VRd has become a preferred regimen in patients with sufficient fitness for triplet therapy based on results from the SWOG S0777 trial, which demonstrated superior outcomes associated with VRd relative to Rd in NDMM patients without intent for immediate transplant. 5 To date, no clinical trials have directly compared D-Rd to SOC regimens other than Rd in patients with NDMM ineligible for transplant. 2 To address this evidence gap, the present PEGASUS study compared progression-free

| Study population and design
The present study (PEGASUS) used individual-level patient data from the global MAIA phase III randomized controlled trial and the US Flatiron Health electronic health record (EHR)-derived deidentified database ( Figure S1 in Appendix S1). An anchored indirect treatment comparison (ITC) study design was used to compare relative treatment effects across the two data sources ( Figure 1). 6,7 Details on the data sources and methodology are provided below.

| MAIA trial
Patients in MAIA were enrolled between March 2015 and January 2017 and randomized to receive D-Rd or Rd. Key eligibility criteria for MAIA included ineligibility for ASCT defined by age (≥65 years) or presence of comorbidities precluding high-dose therapy, Eastern Cooperative Oncology Group (ECOG) performance status ≤2, creatinine clearance ≥30 mL/min, and adequate bone marrow reserve. 2

| Flatiron health database
Records for patients with NDMM were extracted from the Flatiron Health (FH) EHR-derived deidentified database. The FH database is a nationwide, longitudinal, demographically and geographically diverse database derived from EHR data. The database includes deidentified data from over 280 cancer clinics, primarily community-based oncology practices, visited by more than 2.4 million US cancer patients. Curated via technology-enabled abstraction, the FH database includes both structured data (eg, patient demographics, laboratory results, and coded diagnoses) and unstructured data (eg, free text from clinician notes and laboratory reports) from the EHR to define clinical measures that are often unavailable in real-world data sources, including International Staging System (ISS) stage, ECOG performance status, and disease progression. FH patients were eligible for the present study if they were diagnosed with multiple myeloma (MM) between 1 January 2011 and 30 April 2019 and initiated first-line therapy (LOT1) at an FH clinic.

| Eligibility criteria
Eligibility criteria ( Figure S1 in Appendix S1) for the present study included NDMM, age ≥ 65 years, and ineligibility for transplant (MAIA) or no transplant as part of LOT1 (FH). In the FH data, there is no variable reflecting transplant eligibility status; therefore, age ≥ 65 years and no transplant as part of LOT1 were used as a proxy for transplant ineligibility. In addition, the study was restricted to patients who initiated their assigned study regimen (MAIA) or LOT1 (FH) and had ≥1 assessment for disease response following LOT1 initiation. All patients had to meet MAIA trial inclusion criteria, F I G U R E 1 PEGASUS study design and progression-free survival hazard ratios with 95% confidence intervals for D-Rd relative to alternative standard-of-care regimens based on indirect treatment comparisons. D-Rd, daratumumab-lenalidomide-dexamethasone; HR, hazard ratio; ITC, indirect treatment comparison; Rd, lenalidomide-dexamethasone; Vd, bortezomib-dexamethasone; VRd, bortezomib-lenalidomidedexamethasone. Figure reflects results for primary on-treatment analysis of progression-free survival (PFS) with weighting of Flatiron Health treatment groups to resemble MAIA trial population including ECOG performance status ≤2, creatinine clearance ≥30 mL/ min, adequate bone marrow reserve and liver function, and absence of selected comorbidities (see Appendix S1). Finally, due to considerations of statistical power, the FH cohort was restricted to LOT1 regimens used by ≥10% of transplant-ineligible NDMM patients; VRd, Rd, and Vd satisfied this requirement.

| Baseline characteristics
Baseline characteristics captured for both MAIA and FH patients included age at LOT1 initiation, sex, race, ISS stage, cytogenetic risk stratification, ECOG performance status, laboratory measures (eg, creatinine clearance, blood counts, liver function enzymes), comorbidities that were MAIA exclusions (eg, severe cardiovascular disease, other primary malignancy), and the time interval from MM diagnosis to the start of LOT1. Details are provided in Appendix S1.

| Outcomes
The primary outcome was PFS, defined as the time interval in months between LOT1 initiation and disease progression or death. In the MAIA cohort, the definition of disease progression was based on the International Myeloma Working Group (IMWG) criteria 2,8,9 for progressive disease. In the FH cohort, a derived progression measure based on the application of IMWG criteria to the setting of real-world data was used to ascertain progression events. Details on how disease progression and mortality are ascertained in FH are provided in Appendix S1. Patients were followed until their first PFS event, or were censored due to loss of follow-up or reaching the maximum follow-up time of 48.5 months (the maximum available for MAIA participants). This analysis includes an additional 9 months of patient follow-up after the cut-off date for the first prespecified MAIA interim analysis; 2,10 details may be found in Appendix S1.
The primary analysis of PFS was an on-treatment analysis in which patients were also censored if treatment was discontinued for reasons other than disease progression or death. This approach was chosen to reduce heterogeneity in patient management across the MAIA trial and FH routine clinical practice settings. In MAIA, patients were treated until disease progression or unacceptable toxicity. However, this is not always the case in routine clinical practice, where the patient's treatment plan may not include continuous treatment to disease progression, and where patients are more likely to discontinue treatment for a variety of other reasons, including patient preference. 11 The on-treatment analysis restricted eligible follow-up to MAIA and FH patients who did not discontinue their initial treatment regimen; details are provided in Appendix S1. An intent-to-treat analysis of PFS, without censoring at treatment discontinuation, was performed as a sensitivity analysis.
Overall survival (OS), defined as the time interval in months between LOT1 initiation and death from any cause, was prespecified as an exploratory endpoint due to the immaturity of the MAIA OS data. Follow-up for OS in this study was limited to the maximum duration of follow-up for the first prespecified interim analysis for PFS of MAIA (41.4 months).
The next formal analysis of OS in MAIA based on additional follow-up will occur at the next prespecified interim analysis for OS. Because mortality often occurs after the discontinuation of LOT1 and the initiation of subsequent LOTs, only an intent-to-treat analysis was performed for OS.

| Statistical methods
The present study used an anchored ITC design (Figure 1), in which relative treatment effects were compared across the MAIA and FH study populations. The methodology follows published guidelines for an anchored matching-adjusted indirect comparison, a preferred approach for formally comparing results from two trials that share a common comparator or anchor when individual-level patient data are available from at least one of the trials. 6,7 An anchored ITC across two distinct study populations requires that the two populations be balanced on treatment effect modifiers.
To satisfy this requirement, two steps were taken. First, a common set of eligibility criteria was applied to both the MAIA and FH cohorts.
Second, FH patients treated with VRd, Vd, or Rd as LOT1 were weighted to resemble the MAIA trial population on measured baseline characteristics using propensity-score (PS) methods. 12 The purpose of the PS weighting was to ensure that (a) patients treated with VRd, Vd, or Rd in FH were balanced on measured baseline covariates, allowing for unconfounded comparisons of outcomes across the FH treatment groups, and (b) to ensure that the FH cohort resembled the MAIA trial population on possible treatment effect modifiers, allowing for a valid ITC to be performed across the two data sources.
Standardized differences were used to assess covariate balance between each of the FH treatment groups and the MAIA trial population after PS weighting. An absolute standardized difference > 0.1 was interpreted as a meaningful covariate imbalance 13  Multivariate imputation by chained equations was used to impute missing baseline covariate data, and was repeated to create 10 complete datasets. 15,16 After imputation, analyses were performed separately in each of the 10 complete datasets; the resulting parameter estimates and SEs were pooled according to Rubin's rules to obtain a summary parameter estimate and SE. 17 All data management and statistical analyses were performed using SAS version 9.4 (SAS Institute, Cary, NC). Additional details and elaboration on the statistical methods including anchored ITC methodology, PS-weighting, and multiple imputation are provided in Appendix S1.

| Subgroup analyses
In a prespecified subgroup analysis, the association between treatment and PFS was assessed in patients aged ≥75 and < 75 years. The subgroup analysis was performed by adding a covariate for age (dichotomized as ≥75 vs <75 years) and an interaction between age and treatment to the Cox regression models. 18

| Sensitivity analyses
Several sensitivity analyses were performed to evaluate the impact of loss to follow-up and early treatment discontinuation among the FH patients on estimated PFS. First, doubly robust estimation 14 incorporating both PS-weighting and regression adjustment for baseline covariates was used; adjusting for baseline covariates in the outcome model mitigates against the possibility of bias due to covariate-dependent censoring. 19 Second, we included only FH clinics with an average duration of on-treatment follow-up ≥12 months, as the treatment patterns at these clinics would more closely resemble the treat-to-progression protocol used in MAIA. Third, we included only patients with ≥12 months of ontreatment follow-up. Fourth, the association of treatment with PFS was assessed according to the intent-to-treat principle. In this analysis, patients were analyzed according to their initial treatment regimen regardless of subsequent treatment discontinuation.
An additional sensitivity analysis was performed to evaluate the possible impact of multiple imputation of missing baseline covariate data on our results. In this analysis, the FH patients were PS-weighted based on demographic characteristics alone (age, gender, and race), and clinical characteristics with higher rates of missing data (eg, ISS stage, ECOG performance status) were omitted from the PS model.  Figure S1 in Appendix S1).

| Primary outcome: Progression-free survival
In MAIA, D-Rd was associated with a significantly lower risk of disease progression or death compared to Rd over a mean follow-up for 24.9 months (HR 0.54; 95% CI 0.42, 0.71; P < .001; Figure 1).
Note that, due to the exclusion of 22 MAIA patients ( Figure S1 in Appendix S1), the censoring of patients who discontinued treatment  Table 2). Table 2 provides a summary of the results from sensitivity analyses. In the sensitivity analysis of PFS, which adjusted for only demographic characteristics (age, sex, and race), results were similar to the primary PFS analyses. Results from additional sensitivity analyses that assessed the possible impact of treatment discontinuation prior to disease progression and loss to follow-up were consistent with the principal study findings. In the intent-to-treat sensitivity analysis, the difference in PFS between FH patients treated with VRd and Rd was smaller in magnitude, which likely reflects higher rates of treatment discontinuation among patients treated with VRd in routine clinical practice ( Figure S2 in Appendix S1).  Figure S3 in Appendix S1).  Figure S1 in Appendix S1), it is possible that these conditions may be under-reported and/or undercoded in the structured FH EHR data, and thus that the included FH cohort may have had a greater comorbidity burden than the MAIA trial participants. In addition, it was not possible to assess comorbidity severity using the structured diagnosis codes available in the FH EHR data. However, as discussed in greater detail below, the anchored ITC design means that the possibility of unmeasured differences in comorbidity burden across the two populations will only cause bias if the unmeasured comorbidities are relative treatment effect modifiers.

| DISCUSSION
Patient management is also likely to differ in trial and real-world settings. 11  respect to all prognostic factors, or (b) that the absolute outcome event rates in the treatment group serving as the common anchor are identical in the two data sources. 6,7 As evidenced by the difference in PFS between the MAIA Rd and FH Rd patients (Figure 1), even after efforts to balance the patient populations there may still be unmeasured or residual differences in prognostic factors between the trial and real-world patient populations. The anchored ITC design protects against this source of bias by comparing treatment effects relative to the Rd anchor rather than directly comparing outcome rates across the data sources. Within FH, however, comparisons among patients who received VRd, Rd, or Vd may be subject to bias due to confounding by differences in baseline prognostic factors; confounding was addressed through PS weighting and in a sensitivity analysis with doubly robust estimation, which resulted in similar effect estimates.
There were missing data on key patient clinical characteristics among the study groups identified from the FH database. An explicit determination of transplant eligibility status was unavailable in the FH database; therefore, age ≥ 65 years and no receipt of transplant during LOT1 were used as a proxy for transplant ineligibility. For consistency, the age ≥ 65 restriction was also applied to the MAIA trial population. Multiple imputation was employed to account for other missing baseline covariate data; however, the validity of multiple imputation relies on the assumption that the data are missing at random (MAR)-ie, that the probability of a data point being missing is unrelated to the true value of that missing data point after conditioning on the variables included in the imputation models. 15 While the MAR assumption cannot be proven, the fact that the study results were consistent in the sensitivity analysis where only patients' demographics were adjusted for provides reassurance that the results were not sensitive to the imputation of the missing clinical characteristics.
Based on the MAIA trial data, D-Rd treatment in patients with transplant-ineligible NDMM was associated with a 46% significantly lower risk of disease progression or death relative to Rd alone. Based on the Rd-anchored ITCs performed in the present study, D-Rd treatment was associated with a 32% and 52% significantly lower risk of disease progression or death compared to the alternative SOC regimens VRd and Vd, respectively. In the absence of head-to-head trial data, the present ITC may help to inform treatment selection in patients with transplant-ineligible NDMM.

ACKNOWLEDGMENTS
The authors thank Yu-Ning Wong, MD, MSCE, for providing insights on study design, and Wenze Tang, MPH, for providing insights on study design and assistance with data management and statistical programming. Both were employed by Janssen during the conduct of the study.