Временная изменчивость результатов одинаковых режимов в ходе недавно диагностированных испытаний миэломы: систематический обзор.

Research in context

Evidence before this study

Cross-trial comparisons are frequently performed in oncology despite well-recognised methodological limitations. While individual trials have established bortezomib/lenalidomide/dexamethasone (VRd) and lenalidomide/dexamethasone (Rd) as standard-of-care regimens for newly diagnosed multiple myeloma, the longitudinal performance of these identical regimens across different trials has not been systematically evaluated. This gap is particularly important given the 2014 revision of myeloma diagnostic criteria, which reclassified ultra-high-risk smoldering myeloma patients as having active myeloma, changing the natural history of the disease due to inclusion of patients with more indolent disease. Furthermore, the potential impact of informative censoring on myeloma trial outcomes has not been previously analysed. A systematic search of PubMed, Embase, and ClinicalTrials.gov from database inception to March 18th 2026 revealed no existing studies that have comprehensively analysed temporal variability in outcomes, and censoring patterns, for myeloma regimens.

Added value of this study

To the best of our knowledge, this is the first systematic analysis examining temporal trends in efficacy outcomes for VRd and Rd across their complete portfolio of randomised controlled trials. Our analysis of seven VRd trials and seven Rd trials reveals substantial outcome variability, with median progression-free survival (PFS) for VRd ranging from 33.6 to 54.3 months and Rd from 18.3 to 31.9 months. Trials enrolling after the 2014 diagnostic reclassification demonstrated the longest PFS outcomes. Through reconstruction of Kaplan–Meier curves, we identified differential censoring patterns in most trials, predominantly favoring control arms. Our analysis demonstrates that while statistical differences between arms were generally preserved after censoring adjustment, the absolute magnitude of outcomes varied substantially across trials conducted in different time periods, likely reflecting diagnostic evolution, stage migration, and improvements in supportive care.

Implications of all the available evidence

These findings challenge the validity of cross-trial comparisons commonly used to guide clinical decision-making. Although advances in treatment in myeloma are undeniable, the temporal drift in outcomes for identical regimens demonstrates that apparent improvements in efficacy may also reflect changing patient populations, diagnostic criteria, and censoring patterns. This has implications for interpreting contemporary trials as well as in power calculations in the design of these trials. Recognition of these confounding factors could inform future trial design and clinical decision making.

Introduction

The triplet regimen of bortezomib, lenalidomide, and dexamethasone (VRd) has been a cornerstone of myeloma therapy for over a decade, 1 having been preceded by the doublet of lenalidomide/dexamethasone (Rd). 2 VRd was initially studied in Phase II trials with promising efficacy 3 and its survival benefit was confirmed in a pivotal Phase III study by the Southwest Oncology group, demonstrating superiority over lenalidomide and dexamethasone alone. 1 Consequently, VRd has been widely recommended in clinical guidelines and frequently used as a control arm in trials, 4 , 5 , 6 both with and without autologous transplantation.

While therapeutic advances have undoubtedly improved myeloma outcomes, evolving diagnostic criteria and advancements in imaging may have also played a role. 7 Highly sensitive imaging modalities such as PET/CT and diffusion-weighted whole-body MRI now enable earlier detection of myeloma bone disease, leading to more diagnoses than when skeletal surveys were the primary tool. 8 , 9 Additionally, the 2014 revision of myeloma diagnostic criteria reclassified ultra-high-risk smoldering myeloma as active myeloma, even in the absence of end-organ damage (known as SliM only myeloma), further improving patient outcomes due to the inclusion of more indolent disease. 10

In addition to these shifts, common methodological challenges in clinical trials—such as informative censoring—can substantially influence the apparent effect size and complicate interpretation across studies, including key myeloma trials. 11 , 12 Most censoring is administrative and uninformative, occurring late in the study for patients for whom no event has been observed by the data cutoff. 13 Standard Kaplan–Meier analysis assumes that censoring is random (non-informative), meaning that censored patients have the same risk of experiencing the event as those who remain under observation. When this assumption is violated, censoring is considered informative.

When censoring occurs at different rates between treatment arms (differential censoring), it indicates that the censoring is non-random and informative, since random censoring would be expected to occur at comparable rates across all arms. Well-recognised mechanisms may drive such patterns: excess censoring in the control arm can occur when patients on a perceived-inferior arm selectively discontinue, deflating control arm efficacy while excess censoring in the intervention arm can occur when frailer patients discontinue due to toxicity, inflating intervention arm efficacy. 14 This can lead to an inflated treatment effect or even cause a completely inert substance to demonstrate a benefit. 15 This complexity, amongst others undermines straightforward cross-trial comparisons and can hinder the objective appraisal of evidence needed to inform clinical decision-making. Regulatory bodies including the FDA and EMA encourage sensitivity analyses for censoring in oncology trials, and oncology guidelines such as Common Sense Oncology provide specific recommendations for such analyses. 16 , 17 , 18

As supportive care and optimized drug dosing continue to enhance treatment efficacy, evaluating the long-term performance of the same regimens (VRd and Rd) offers insight into the relative impact of diagnostic reclassification and imaging advances on survival improvements. In this systematic review, we identify and analyse all randomised controlled trials in which VRd or Rd without transplant served as either a control or intervention arm, aiming to assess its outcomes over time. Such an analysis allows us to evaluate the true extent of how much variability exists for the same regimen in different trials, highlighting the problem of cross-trial comparisons. As censoring has never been previously assessed in a systematic fashion in a cohort of myeloma trials, this analysis also allowed us to evaluate the role of informative censoring in the results of these trials.

Methods

Study design and ethics

This systematic review is a secondary analysis of a previously published systematic review and did not have a separately registered protocol. 19 The protocol for the original systematic review was registered on Open Science Framework (OSF) and is accessible at https://doi.org/10.17605/OSF.IO/MXK8T . This study was not separately registered. This report follows PRISMA guidelines for transparent and comprehensive reporting.

As this systematic review was conducted using data from previously published studies, it was considered exempt from institutional board review and further ethical approval and written informed consent were not required. All data analysed were publicly available.

Search strategy

We systematically searched three databases—MEDLINE/PubMed, Embase, and the Cochrane Central Register of Controlled Trials (CENTRAL)—from database inception to March 18, 2026 for randomised controlled trials (RCTs) involving patients with multiple myeloma (MM). Our search strategy incorporated both free-text and MeSH terms related to MM and applied filters for RCTs published in English. The search strategy is listed in the supplement. Two authors performed the initial search strategy, and two authors performed a search update in March 2026 (GRM, MM). We included both full-text publications and conference abstracts. Additionally, we searched ClinicalTrials.gov for completed studies with reported results.

Objectives

The objective of this analysis was to assess the variability in progression-free survival outcomes for identical therapeutic regimens (VRd and Rd) across multiple myeloma trials and identify factors contributing to this variation. Key factors analysed were the impact of diagnostic reclassifications and the role of differential censoring.

Eligibility criteria

We included randomised Phase II or III trials that had efficacy results reported. Studies had to be published after January 1, 2001, to align with the emergence of immunomodulatory drugs (IMiDs) and proteasome inhibitors, ensuring that VRd and Rd could be adequately evaluated. Eligible trials had to include VRd and Rd (without the use of transplant) as either a control or intervention arm and provide sufficient data to infer efficacy outcomes in order to standardize results. Trials that reported efficacy results but without a median progression-free survival (PFS) estimate were excluded, such as KEYNOTE-185, 20 or trials that were terminated prematurely, such as the EMN20 trial of carfilzomib/lenalidomide/dexamethasone versus Rd. 21 We also excluded treatments that utilized high-dose dexamethasone (40 mg Day 1–4, 9–12 or similar dosing), leading to S0232 being excluded. 22 We also excluded trials that did not utilize a starting dose of 25 mg of lenalidomide, such as Mookerjee et al. for VRd. However, Moojerkee et al. was included as a study for Rd, since the Rd arm utilized 25 mg of lenalidomide. 23 Finally, the IFM 2017-03 trial was excluded as a Rd study since it exclusively enrolled a frail population limiting ability to compare to other trials, and utilized lower doses of dexamethasone. 24

Although Mookerjee was included in our comparisons of median estimates, due to lack of a presented survival curve, it was not included in censoring analysis. Trials in which both arms received autologous stem cell transplantation were not included due to sheer heterogeneity of conditioning regimens; consolidation strategies and maintenance approaches present in such trials. For trials in which one arm received transplant, we included the arm that did not receive transplant in our analysis, such as in the DETERMINATION 25 and IFM-2009 trial. 26

Data synthesis and analysis

Considering that the diagnostic reclassification of myeloma occurred in November 2014, 10 we categorised them as to whether they began enrolling before this date, or after this date.

Trial characteristics, including median duration of follow-up, were abstracted directly from the primary publications. To assess the potential impact of differential censoring, PFS Kaplan–Meier curves were extracted using WebPlotDigitizer (v5.0) and reconstructed with the reconstructKM (v0.1.0) package in R. Numbers at risk were available for all included curves and were incorporated into the reconstruction process. Accuracy of reconstructed data was assessed by comparing reconstructed hazard ratios and 95% confidence intervals against published values ( Supplementary Table S1 ). Reverse Kaplan–Meier curves were generated by flipping the status indicator, such that censoring events are treated as the outcome of interest and progression events are treated as censoring events. By utilising reverse Kaplan–Meier for this goal, focusing on censoring as the event of interest over time (rather than progression), we can analyse differences in censoring rates over time while using the same common assumptions in standard survival analysis. Importantly, the excess censoring analysis using the reverse Kaplan–Meier does not compare total numbers of censored patients, which are expected to differ between arms (for example, due to administrative censoring at data cutoff in a positive trial), but rather the rate of censoring among patients still at risk at each timepoint, allowing identification of excess censoring at specific timepoints. For each plot, an accompanying figure displays the censoring excess between arms, where positive values indicate higher censoring in the intervention arm and negative values indicate higher censoring in the control arm. Sensitivity analysis adjusting for excess censoring in the control or intervention arm was performed. 27 In summary, sensitivity analysis was performed according to the main arm with excess censoring. When excess censoring in the control group of an open-label trial was seen, the hypothesis was that the censored patients were better-performing patients and that their distribution of survival was similar to that of the same number of the longest surviving patients who remained on trial in that group. When excess censoring in the intervention group was seen, we hypothesized that the censored patients were poorly performing and had experienced an event. Reverse restricted mean survival times (RMSTs), an alternative to the HR that does not assume proportional hazards, were also utilized to assess censoring across treatment arms. Briefly, reverse RMST describes the average time a subject allocated to a given group spends without experiencing the event of interest. 28

For the purposes of this work, the reverse HR quantifies the instantaneous rate of censoring between groups, 13 whereas the reverse RMST difference (RMST-D) estimates the difference in the average time remaining uncensored over the follow-up period. We utilized reverse Kaplan–Meier strictly as a comparative diagnostic tool to detect differential censoring between trial arms, rather than to establish absolute median follow-up times. We have therefore not assumed independence 29 ; rather, we have used the reverse Kaplan–Meier as a diagnostic for departures from it, which in turn justified the use of a sensitivity analysis. Forest plots were created to display the HRs derived from the reconstructed Kaplan–Meier curves (black), reverse HRs (blue), and HRs adjusted for excessive censoring (rightmost column of the figures). Forest plots for RMST-D followed the same structure. Hazard ratios were estimated using Cox proportional hazards models, and between-arm comparisons were performed using the log-rank test. RMST was estimated using the survRM2 (v1.0–4) package. As a post-hoc analysis, we performed a per-arm random-effects meta-analysis (DerSimonian–Laird) of median PFS for VRd and Rd separately using reconstructed Kaplan–Meier data, pooling within each regimen to characterise heterogeneity. Between-trial heterogeneity was quantified using the I 2 statistic and Cochran’s Q test. A p-value of <0.05 was considered statistically significant, and all p-values were two-tailed. Statistical analyses were conducted using RStudio (Version 4.5.2). A risk of bias assessment was performed using the Cochrane Risk of Bias tool.

Role of the funding source

This study had no external funding.

Results

Trials characteristics

A total of seven VRd arms and seven Rd arms across thirteen trials were identified. One trial had both a VRd arm and a Rd arm, SWOG 0777. 1 Fig. 1 highlights our study selection process and the reasons for exclusion for studies. Nine trials had a low risk of bias, while four had an intermediate risk of bias ( Supplementary Fig. S1 ).

Amongst the seven VRd trials, enrolment start dates ranged from April 2008 to December 2018. Of these, five trials began enrolling patients before the 2014 diagnostic criteria reclassification, and two trials began enrolling after the diagnostic reclassification. Median duration of follow-up ranged from 26 (15–39) to 76 months ( Table 1 ).

Seven randomised controlled trials with an Rd regimen were included with enrolment start dates ranging from April 2008 to January 2019. Five trials began enrolling before the diagnostic revision and two after. Median follow-up in the Rd arms went from 17.1 (range 1–33) months to 56 (IQR 50–70) months, with one trial (FIRST) not reporting the duration of follow-up specific for this treatment arm ( Table 1 ).

Progression-free survival

Reconstructed Kaplan–Meier curves showed high concordance with published data across all trials ( Supplementary Table S1 ). The median PFS with VRd varied across studies, ranging from 33.6 months (95% confidence interval (CI) 19.55- not reported [NR]) to 54.3 months (95% CI 45.2-NR). The two studies with the longest reported PFS were IMROZ (54.3 months) and CEPHEUS (52.6 months), 5 , 6 both of which enrolled after the diagnostic changes. Confidence intervals for median PFS were inconsistently reported, with five studies providing full confidence intervals for PFS with VRd. The hazard ratios and their confidence intervals for the PFS estimate for these trials are listed in Supplementary Table S1 .

Median PFS in the Rd studies ranged from 18.3 months (95% CI NR) to 31.9 months (95% CI 28.9-NR). Reporting of CI’s was not consistent as two Rd studies did not report them and the study arm with the longest PFS was conducted after diagnostic criteria changes ( Fig. 2 ).

In addition, we performed a per-arm meta-analysis. For VRd, pooled median PFS was 42.4 months (95% CI 36.8–48.8; I 2 = 76.9%; Q = 26.3, p < 0.001). For Rd, pooled median PFS was 27.1 months (95% CI 22.7–32.5; I 2 = 77.4%; Q = 21.9, p < 0.001).

Impact of diagnostic criteria changes

No study (either with VRd or Rd) provided a percentage of patients meeting only SLiM criteria at diagnosis. Additionally, no study reported the number or proportion of patients classified as having a progression event in the PFS analysis based on SLiM criteria.

Heterogeneity of high-risk classification

Amongst the thirteen included trials, nine trials used definition of high-risk consistent with the IMWG definition of high-risk MM, defined as the presence of del (17p), t (4; 14) and/or t (14; 16), whereas three trials used varying definitions, and one did not describe high-risk prevalence or definition.

The definitions of high-risk disease and the percentages of patients with high-risk disease in each trial are listed in the Table 2 . For the eight trials that used this definition of high-risk disease, the prevalence of patients with high-risk disease ranged from 13.6% to 23%.

Differential censoring

Overall, differential censoring was present in 11 of 13 comparison groups, with only IFM-2009 and SWOG S0777 showing no evidence of differential censoring ( Figs. 3 and 4 ). Among comparison groups with differential censoring, seven had excessive censoring in the control arm, two (DETERMINATION and REAL-MM) in the experimental arm, and the remainder (FIRST trial [lenalidomide 18 cycles arm] and Larocca et al.) had excess censoring in both ( Fig. 3 ). In most cases, differential censoring was evident in the earlier stages of the follow-up period ( Fig. 3 ). Supplementary Fig. S2 demonstrates numbers at risk in these curves.

Examples of significant differential censoring, as measured by the reverse hazard ratio were MAIA (reverse HR 0.64, p < 0.001), FIRST (reverse HR 0.75, p = 0.01), ENDURANCE (reverse HR 0.84, p = 0.01), DETERMINATION (reverse HR 1.7, p < 0.001), and Larocca et al. (reverse HR 1.9, p = 0.02). Additionally, reverse-KM plots suggest differential censoring in other trials, including the ELOQUENT trial, though statistical significance was not reached, likely due to excessive censoring co-occurring in both study arms ( Fig. 4 ).

After adjusting HR estimates for excessive censoring, treatment effect attenuation was observed in most trials ( Fig. 4 ). For example, MAIA showed attenuation from HR 0.55 to 0.69 (45%–31% reduction in apparent relative benefit), IMROZ from HR 0.59 to 0.71 (41%–29%), and CEPHEUS from HR 0.58 to 0.67 (42%–33%). However, these were all numerical reductions, and statistical significance was lost only in DETERMINATION.

Discussion

To our knowledge, no previous study has analysed the longitudinal performance of a single regimen in myeloma. In oncology trials, sorafenib has been studied where overall survival in clinical trials improved over time, while PFS remained unchanged. 30 It is intuitive to see a difference in overall survival due to advances in therapies received after the trial period, but to see such a dramatic difference in PFS with the same regimen, as seen with VRd here, remains unusual. While some of the improvement in PFS may be attributed to enhanced supportive care, the magnitude of the change suggests additional contributing factors. 13

We speculate that some of the improvement seen with VRd in recent trials could be driven by an increasing proportion of patients being enrolled in trials without having CRAB features at diagnosis, as well as increased usage of advanced imaging over time, leading to the discovery of smaller lesions, and patients who previously would have been diagnosed as smoldering myeloma had a skeletal survey been done, now being diagnosed as myeloma. As a result, stage migration and the Will Rogers phenomenon may be influencing outcomes, giving the appearance of improved efficacy. 7 , 13 Unfortunately, none of the trials in our dataset reported the proportion of patients diagnosed with myeloma solely based on CRAB features, and as a result, we cannot prove this hypothesis. We tried requesting this information from the sponsors of the recently published IMROZ and CEPHEUS trial but were told this information was unavailable. To test this hypothesis, future studies should clearly report this percentage and analyse outcomes separately for patients with and without traditional CRAB myeloma. Notably, the trials with longest PFS (IMROZ and CEPHEUS, both with PFS >52 months) also demonstrated significant differential censoring in the control arm potentially inflating effect size estimates in the intervention arm (whereas VRd was the control arm). Sensitivity analysis for excessive censoring yielded hazard ratios, closer to 1 (0.59 to 0.71 for IMROZ and 0.58 to 0.67 for CEPHEUS respectively), indicating a decrease in effect size for the intervention arm. This further emphasizes that at least for IMROZ and CEPHEUS, diagnostic reclassifications rather than censoring may have played a role in the better performance of VRd.

This finding has implications for power calculations for ongoing clinical trials where VRd without transplant continues to serve as the control arm—for example, CARTITUDE-5, which compares VRd to ciltacabtagene autoleucel ( NCT04923893 ). Given that the trial was conceived prior to the results of the most recent VRd trials (CEPHEUS and IMROZ), its PFS estimate of the control arm may be inaccurate.

Our findings reveal differential censoring in the majority of included trials showing excess censoring in the control arm. Notably, much of this differential censoring occurred early in follow-up, when administrative censoring would not be expected, further supporting that it is informative in nature. Excess censoring in the control arm can inflate intervention arm efficacy when better-performing patients dropout due to perceived-inferior arm, 13 , 23 while excess censoring in the intervention arm, as seen in DETERMINATION, can similarly inflate efficacy through early dropout of frailer patients. Sensitivity analysis modeling the potential outcomes of excessively censored patients attenuated or negated treatment benefit. In the PANORAMA-1 trial, excessive early censoring was among the concerns that prompted an FDA Oncologic Drugs Advisory Committee review of whether the observed benefit reflected a true treatment effect or was driven by informative censoring. 31 Ultimately, the degree of censoring in these trials might contribute to the poor reproducibility of the results of these regimens, both in subsequent trials, and in the real-world.

Our study emphasizes the perils of cross-trials comparisons. The widely varying performance of the same regimen across these trials suggests that differences in enrolled patient populations and differential censoring patterns, rather than the regimen itself, account for much of the observed outcome variability. Some of these factors cannot be evaluated without detailed access to patient level-data, such as what fitness (or lack of) the enrolled participants had, the social background of study participants and their access to healthcare resources and the length and rigor of the screening phase (which may exclude patients with the most aggressive disease).

Our study has several important limitations. We were unable to obtain patient-level data on the proportion of patients diagnosed solely based on SLiM criteria versus traditional CRAB features, limiting our ability to definitively test the stage migration hypothesis. The analysis relies on reconstructed Kaplan–Meier curves from published figures, which may introduce minor inaccuracies compared to individual patient data. While we identified systematic patterns of differential censoring across trials, we cannot fully determine the underlying reasons for censoring in each study, as this information was inconsistently reported in the original publications. Our assumptions regarding the reasoning for censoring may not necessarily be true, as patients who are censored in any of the study arms may either be at a higher or lower risk of progression. As such, the true treatment effect likely falls between the reported and sensitivity-analysis estimates. Nevertheless, our results highlight that regardless of the reason, censoring can contribute to wide variability in observed results. Our analysis focused specifically on patients receiving VRd and Rd regimens without transplant, which may limit generalizability to other multiple myeloma treatments or patient populations. Several well-recognised sources of trial heterogeneity could not be formally adjusted for in our analysis, including differences in patient fitness and frailty, transplant eligibility criteria, the use and duration of maintenance therapy, dose intensity, risk stratification methodology, and advances in supportive care and salvage therapies available during and after each trial. Notably, our study design did address some sources of heterogeneity by restricting inclusion to non-transplant arms (hence no post-transplant maintenance) and requiring standard pre-specified doses of medications. Nevertheless, the remaining factors likely contribute to the outcome variability we observe, and definitive conclusions regarding their relative contribution would require individual patient-level data, which were not available for this analysis. The relatively small number of trials (n = 13) limits the power to detect all potential sources of heterogeneity, though our findings demonstrate clear patterns of bias that warrant attention in future trial design and interpretation. While we performed a per-arm meta-analysis to quantify between-trial heterogeneity in median PFS, we did not pool hazard ratios across trials. Our objective was to characterise variability in outcomes for the same regimen rather than derive a single comparative treatment effect, and pooled hazard ratios would have been uninterpretable given that the comparator arms differed across trials. Furthermore, the regimens we analysed are now considered outdated as the field has moved onto four drug therapies, and future studies should re-evaluate our findings with those regimens, although we would expect similar findings would apply to those regimens. A final limitation is that even though we chose regimens that have stayed the standard of care for a long time, the number of included studies was small, limiting the power of our analysis.

In summary, using VRd and Rd as a case study, we highlight that improvements in myeloma outcomes may not be solely driven by therapeutic advances, and that comparing the PFS across different regimens is problematic, as the same regimen can provide widely varying results. Diagnostic reclassifications, improvements in supportive care, varying proportions of informative censoring, differences in characteristics of enrolled patients, and stage migration likely contribute as well to these results, although definitive conclusions are limited without patient level data. Future studies should report the prevalence of patients diagnosed without CRAB features (SLiM only myeloma) and provide clear rationale of, and analysis adjusting for, censoring to ensure more accurate evaluations of efficacy.

Contributors

GRM and TM conceived the idea. GRM and TM accessed and verified the underlying data. HDK and TM performed statistical analysis. GRM, TM and HAV wrote first draft of manuscript. GRM, MM and HAV collected data. GRM and MM completed the study search. GRM, HAVA, AG, AG, MM, RC, HDK, and TM all critically reviewed the manuscript and study methodology. All authors read and approved the final version of the manuscript.

Data sharing statement

Data can be shared upon reasonable request to the corresponding author.

Declaration of interests

GRM reports institutional research funding from Janssen and Sanofi, and honoraria from MashupMD and Medscape for writing. RC reports honoraria from Janssen, Adaptive and Sanofi. AG reports consulting from Janssen. The other authors have no conflicts.

Acknowledgements

None.

Footnotes

Supplementary data related to this article can be found at https://doi.org/10.1016/j.eclinm.2026.104064 .

Supplementary data

Supplementary Figs. S1 and S2 and Table S1

References

1. Durie B.G.M., Hoering A., Abidi M.H., et al. Bortezomib with lenalidomide and dexamethasone versus lenalidomide and dexamethasone alone in patients with newly diagnosed myeloma without intent for immediate autologous stem-cell transplant (SWOG S0777): a randomised, open-label, phase 3 trial. Lancet. 2017;389(10068):519–527. doi: 10.1016/S0140-6736(16)31594-X.

2. Rajkumar S.V., Hayman S.R., Lacy M.Q., et al. Combination therapy with lenalidomide plus dexamethasone (Rev/Dex) for newly diagnosed myeloma. Blood. 2005;106(13):4050–4053. doi: 10.1182/blood-2005-07-2817.

3. Richardson P.G., Weller E., Lonial S., et al. Lenalidomide, bortezomib, and dexamethasone combination therapy in patients with newly diagnosed multiple myeloma. Blood. 2010;116(5):679–686. doi: 10.1182/blood-2010-02-268862.

4. Kumar S.K., Jacobus S.J., Cohen A.D., et al. Carfilzomib or bortezomib in combination with lenalidomide and dexamethasone for patients with newly diagnosed multiple myeloma without intention for immediate autologous stem-cell transplantation (ENDURANCE): a multicentre, open-label, phase 3, randomised, controlled trial. Lancet Oncol. 2020;21(10):1317–1330. doi: 10.1016/S1470-2045(20)30452-6.

5. Usmani S.Z., Facon T., Hungria V., et al. Daratumumab plus bortezomib, lenalidomide and dexamethasone for transplant-ineligible or transplant-deferred newly diagnosed multiple myeloma: the randomized phase 3 CEPHEUS trial. Nat Med. 2025;31(4):1195–1202. doi: 10.1038/s41591-024-03485-7.

6. Facon T., Dimopoulos M.A., Leleu X.P., et al. Isatuximab, Bortezomib, lenalidomide, and dexamethasone for multiple myeloma. N Engl J Med. 2024;391(17):1597–1609. doi: 10.1056/NEJMoa2400712.

7. Chakraborty R. Survival in smouldering myeloma and symptomatic myeloma: is there an emerging will Rogers phenomenon? Eur J Cancer. 2022;172:234–236. doi: 10.1016/j.ejca.2022.05.041.

8. Hillengass J., Moulopoulos L.A., Delorme S., et al. Whole-body computed tomography versus conventional skeletal survey in patients with multiple myeloma: a study of the International Myeloma Working Group. Blood Cancer J. 2017;7(8) doi: 10.1038/bcj.2017.78.

9. Siontis B., Kumar S., Dispenzieri A., et al. Positron emission tomography-computed tomography in the diagnostic evaluation of smoldering multiple myeloma: identification of patients needing therapy. Blood Cancer J. 2015;5(10) doi: 10.1038/bcj.2015.87.

10. Rajkumar S.V., Dimopoulos M.A., Palumbo A., et al. International Myeloma Working Group updated criteria for the diagnosis of multiple myeloma. Lancet Oncol. 2014;15(12):e538–e548. doi: 10.1016/S1470-2045(14)70442-5.

11. Popat R., Beksac M., Dimopoulos M.A., et al. Venetoclax-dexamethasone versus pomalidomide-dexamethasone in t(11;14)-Positive relapsed/refractory multiple myeloma: primary results of the randomized, phase III CANOVA study. J Clin Oncol. 2026;44(3):164–175. doi: 10.1200/JCO-25-00924.

12. Mohyuddin G.R., Meirson T. A critical appraisal of the ATLAS trial of maintenance therapy for multiple myeloma: end points, censoring and equipoise. Nat Rev Clin Oncol. 2023;20(11):735–736. doi: 10.1038/s41571-023-00801-x.

13. Gilboa S., Pras Y., Mataraso A., Bomze D., Markel G., Meirson T. Informative censoring of surrogate end-point data in phase 3 oncology trials. Eur J Cancer. 2021;153:190–202. doi: 10.1016/j.ejca.2021.04.044.

14. Hsu E.J., Lin T.A., Dabush D.R., et al. Association of differential censoring with survival and suboptimal control arms among oncology clinical trials. J Natl Cancer Inst. 2024;116(6):990–994. doi: 10.1093/jnci/djae028.

15. Tannock I.F., Pond G.R., Booth C.M. Biased evaluation in cancer drug trials-how use of progression-free survival as the primary end point can mislead. JAMA Oncol. 2022;8(5):679–680. doi: 10.1001/jamaoncol.2021.8206.

16. FDA . 2008. Guidance for sponsors, clinical investigators, and IRBs data retention when subjects withdraw from FDA-regulated clinical trials. https://www.fda.gov/media/75138/download#:∼:text=Page%206-,Contains%20Nonbinding%20Recommendations,developed%20to%20maintain%20study%20validity Available from:

17. EMA . 2012. Appendix 1 to the guideline on the evaluation of anticancer medicinal products in man. https://www.ema.europa.eu/en/documents/scientific-guideline/appendix-1-guideline-evaluation-anticancer-medicinal-products-man-methodological-consideration-using-progression-free-survival-or-disease-free-survival-confirmatory-trials_en.pdf Available from:

18. Gyawali B., Eisenhauer E.A., van der Graaf W., et al. Common sense oncology principles for the design, analysis, and reporting of phase 3 randomised clinical trials. Lancet Oncol. 2025;26(2):e80–e89. doi: 10.1016/S1470-2045(24)00451-0.

19. Mainou M., Tsapa K., Michailidis T., et al. Outcomes in randomized controlled trials of therapeutic interventions for multiple myeloma: a systematic review. Crit Rev Oncol Hematol. 2024;204 doi: 10.1016/j.critrevonc.2024.104529.

20. Usmani S.Z., Schjesvold F., Oriol A., et al. Pembrolizumab plus lenalidomide and dexamethasone for patients with treatment-naive multiple myeloma (KEYNOTE-185): a randomised, open-label, phase 3 trial. Lancet Haematol. 2019;6(9):e448–e458. doi: 10.1016/S2352-3026(19)30109-7.

21. Bringhen S., Cani L., Antonioli E., et al. Carfilzomib-lenalidomide-dexamethasone versus lenalidomide-dexamethasone in patients with newly diagnosed myeloma ineligible for autologous stem-cell transplantation (EMN20): a randomised, open-label, multicentre, phase 3 trial. Lancet Haematol. 2025;12(8):e621–e634. doi: 10.1016/S2352-3026(25)00162-0.

22. Zonder J.A., Crowley J., Hussein M.A., et al. Lenalidomide and high-dose dexamethasone compared with dexamethasone as initial therapy for multiple myeloma: a randomized Southwest Oncology Group trial (S0232) Blood. 2010;116(26):5838–5841. doi: 10.1182/blood-2010-08-303487.

23. Mookerjee A., Gupta R., Jasrotia S., et al. Bortezomib, lenalidomide and low-dose dexamethasone (VRD) versus lenalidomide and low-dose dexamethasone (Ld) for newly-diagnosed multiple Myeloma- a randomized phase III study. Blood. 2017;130(Supplement 1):906.

24. Manier S., Lambert J., Hulin C., et al. Safety and efficacy of a dexamethasone-sparing regimen with daratumumab and lenalidomide in patients with frailty and newly diagnosed multiple myeloma (IFM2017-03): a phase 3, open-label, multicentre, randomised, controlled trial. Lancet Oncol. 2025;26(10):1323–1333. doi: 10.1016/S1470-2045(25)00280-3.

25. Richardson P.G., Jacobus S.J., Weller E.A., et al. Triplet therapy, transplantation, and maintenance until progression in myeloma. N Engl J Med. 2022;387(2):132–147. doi: 10.1056/NEJMoa2204925.

26. Attal M., Lauwers-Cances V., Hulin C., et al. Lenalidomide, bortezomib, and dexamethasone with transplantation for myeloma. N Engl J Med. 2017;376(14):1311–1320. doi: 10.1056/NEJMoa1611750.

27. Meirson T., Ofer J., Zimhony-Nissim N., et al. Reanalysis of urothelial cancer chemoimmunotherapy trials with differential censoring. JAMA Netw Open. 2025;8(1) doi: 10.1001/jamanetworkopen.2024.55630.

28. Royston P., Parmar M.K. Restricted mean survival time: an alternative to the hazard ratio for the design and analysis of randomized trials with a time-to-event outcome. BMC Med Res Methodol. 2013;13:152. doi: 10.1186/1471-2288-13-152.

29. Xue X., Agalliu I., Kim M.Y., et al. New methods for estimating follow-up rates in cohort studies. BMC Med Res Methodol. 2017;17(1):155. doi: 10.1186/s12874-017-0436-z.

30. Tan D.J.H., Tang A.S.P., Lim W.H., et al. Survival trends in sorafenib for advanced hepatocellular carcinoma: a reconstructed individual patient data meta-analysis of randomized trials. Liver Cancer. 2023;12(5):445–456. doi: 10.1159/000529824.

31. Ayon Verduzco A., Yeboah K. Disagreements between FDA and its oncologic Drugs Advisory Committee (ODAC) Ther Innov Regul Sci. 2021;55(1):98–110. doi: 10.1007/s43441-020-00194-4.

Associated Data

Supplementary Materials

Supplementary Figs. S1 and S2 and Table S1