RWD-derived response in multiple myeloma

Abstract

Real-world data (RWD) are important for understanding the treatment course and response patterns of patients with multiple myeloma. This exploratory pilot study establishes a way to reliably assess response from incomplete laboratory measurements captured in RWD. A rule-based algorithm, adapted from International Myeloma Working Group response criteria, was used to derive response using RWD. This derived response (dR) algorithm was assessed using data from the phase III BELLINI trial, comparing the number of responders and non-responders assigned by independent review committee (IRC) versus the dR algorithm. To simulate a real-world scenario with missing data, a sensitivity analysis was conducted whereby available laboratory measurements in the dataset were artificially reduced. Associations between dR and overall survival were evaluated at 1) individual level and 2) treatment level in a real-world patient cohort obtained from a nationwide electronic health record-derived de-identified database. The algorithm's assignment of responders was highly concordant with that of the IRC (Cohen's Kappa 0.83) using the BELLINI data. The dR replicated the differences in overall response rate between the intervention and placebo arms reported in the trial (odds ratio 2.1 vs. 2.3 for IRC vs. dR assessment, respectively). Simulation of missing data in the sensitivity analysis (-50% of available laboratory measurements and -75% of urine monoclonal protein measurements) resulted in a minor reduction in the algorithm's accuracy (Cohen's Kappa 0.75). In the RWD cohort, dR was significantly associated with overall survival at all landmark times (hazard ratios 0.80-0.81, p<0.001) at the individual level, while the overall association was R2 = 0.67 (p<0.001) at the treatment level. This exploratory pilot study demonstrates the feasibility of deriving accurate response from RWD. With further confirmation in independent cohorts, the dR has the potential to be used as an endpoint in real-world studies and as a comparator in single-arm clinical trials.

Fig 1. Study design.

A) Landmark analysis for assessing individual-level associations and B) building cohorts for the assessment of treatment-level association between derived ORR and OS for a RWD cohort from the Flatiron Health MM database. MM, multiple myeloma; ORR, overall response rate; OS, overall survival; PD, progressive disease; RWD, real-world data.

Fig 2. Treatment-level association between OR for derived overall response rate and HR for overall survival for a RWD cohort from the Flatiron Health MM database.

Each dot in the plot represents one comparison in one cohort. For the treatment-level association analysis, only the significant ORs and HRs from the comparative analyses were used. PI+steroid was used as the comparator treatment group for all analyses. Error bars show the 95% CI of the estimated ORs and HRs. chemo, chemotherapy (cyclophosphamide, doxorubicin, melphalan, vincristine, bendamustine); CI, confidence interval; HR, hazard ratio; IMiD, immunomodulatory drug (lenalidomide, pomalidomide, thalidomide); MM, multiple myeloma; OR, odds ratio; ORR, overall response rate; OS, overall survival; PI; proteasome inhibitor (bortezomib, carfilzomib, or ixazomib); RWD, real-world data; steroid: dexamethasone, prednisone methylprednisolone, prednisolone.