OBJECTIVES: Hematologic response has substantial impact on progression-free survival (PFS) in patients with relapsed/refractory multiple myeloma (RRMM). Here, we investigated Bayesian parametric mixture models (B-PMMs) informed by observations stratified by complete response or better (≥CR) versus less than CR (<CR) to extrapolate PFS from a reconstructed early data cut of the phase III CASTOR study investigating daratumumab, bortezomib, and dexamethasone (DVd) in RRMM, with 7.4 months minimum follow-up. B-PMM performance was judged against observations from a later data cut of CASTOR with 40.0 months median follow-up.

METHODS: Reconstructed PFS data for carfilzomib, lenalidomide, and dexamethasone (KRd) from the phase III ASPIRE study, with 32.3 months median follow-up, were used to derive prior Weibull distributions for latent ≥CR and <CR survival functions in the B-PMM for the DVd arm of CASTOR. The latter prior distribution was adjusted by a hazard ratio calculated from the respective control arms (targeted therapy plus dexamethasone; Rd or Vd). The proportion of patients receiving KRd classed as ≥CR at 6 months in ASPIRE informed a prior distribution for this quantity in the DVd arm of CASTOR.

RESULTS: Projected 36-month PFS in the overall population was consistent with trial observations from the later data cut, albeit the estimate was cautious (15.8% [95% credible interval (CrI): 7.7-25.8%] B-PMM vs 24% observed). The B-PMM estimate for the proportion of ≥CR patients (20.7% [95% credible interval (CrI): 6.2-36.5%]) closely agreed with the CR rate reported in the initial data cut (19.2%) even though the prior expectation was conservative (12.8% [95% CrI: 3.0-21.8%]).

CONCLUSIONS: B-PMMs informed by appropriate historical trial data stratified by response status can reliably extrapolate complex survival trends borne by response-based survival heterogeneity in RRMM. Despite between-trial confounding manifesting as notably poorer outcomes in CASTOR vs ASPIRE, after a straightforward adjustment, PFS data from ASPIRE proved useful in aiding extrapolations for CASTOR.