Dose and Schedule Selection of the Oral Proteasome Inhibitor Ixazomib in Relapsed/Refractory

ONLINE RESOURCE

Dose and Schedule Selection of the Oral Proteasome Inhibitor Ixazomib in Relapsed/Refractory Multiple Myeloma: Clinical and Model-based Analyses

Neeraj Gupta,1Huyuan Yang,1 Michael J. Hanley,1 Steven Zhang,1 Rachael Liu,1 Shaji Kumar,2 Paul G. Richardson,3 Tomas Skacel,1,4 Karthik Venkatakrishnan1

1Millennium Pharmaceuticals, Inc., Cambridge, MA, USA, a wholly owned subsidiary of Takeda Pharmaceutical Company Limited; 2Division of Hematology, Mayo Clinic, Rochester, MN, USA; 3Dana-Farber Cancer Institute, Boston, MA, USA; 4Department of Hematology, Charles University General Hospital, Prague, Czech Republic

Corresponding author:

Neeraj Gupta, PhD

Quantitative Clinical Pharmacology, Millennium Pharmaceuticals, Inc., Cambridge, MA, a wholly owned subsidiary of Takeda Pharmaceutical Company Limited, 40 Landsdowne Street, Cambridge, MA 02139, USA

E-mail: ; Tel: +1 617 444 2119

Supplementary Methods

2.3 Statistical Methods for Exposure–Response Analyses

For the exposure–response analyses of TOURMALINE-MM1 data, the exposure–PFS and exposure–dose reduction relationships were characterized by Cox proportional hazards models relating exposure to the hazard of PFS or first ixazomib dose reduction, respectively. Additionally, Kaplan–Meier plots of PFS and time to first ixazomib dose reduction were generated and stratified by ixazomib exposure quartiles. For the Cox proportional hazards model relating exposure to the hazard of PFS, the hazard function was expressed as follows:

In this equation, λ0 (t) is the baseline hazard function and xi is a vector of predictor variables (e.g., time-averaged ixazomib exposure). The parameter vector β is estimated by maximum partial likelihood.

The exposure–clinical response analyses used logistic regression models to determine the relationship between time-averaged ixazomib exposure (to the time of first confirmed best response, or treatment discontinuation, or the start of alternative therapy) and the probability of achieving the level of best response being evaluated. For the logistic regression models relating ixazomib exposure to clinical response, the logit (log-odds) were determined by the following equation:

The logit is log-odds, P is the predicted probability for response, β0 and β are scalar and vector parameters that represent the baseline odds and the effect of Xi, such as time-averaged ixazomib exposure, respectively. This model equation assumes that the predictor variables (Xi) have proportional effects on the probability of response. The results of the logistic regression analyses are presented in terms of an odds ratio (OR), which is equivalent to eβ, and represents the odds of response rate being increased with an increase in one unit of exposure (e.g., 1 ng·h/mL/day) given the coefficient is a positive term. A similar logistic regression modelling approach was employed for the exposure–safety analyses.

For all logistic regression models, a base model was established to assess the relationship between ixazomib exposure and the event of interest. If a statistically significant (p < 0.05) relationship was identified, then covariates that may have potentially influenced the event of interest were further assessed using a multivariate logistic regression model with forward addition of significant covariates (p = 0.05) and backward elimination of nonsignificant covariates (p = 0.01). The categorical covariates examined included Eastern Cooperative Oncology Group score, International Staging System disease stage, demographics (gender, race), cytogenetic risk (efficacy analyses only), prior lines of therapy, prior proteasome inhibitor therapy, and prior immunomodulatory drug therapy; continuous covariates included creatinine clearance, hematocrit/haemoglobin (efficacy analyses only), age, and BSA.

For the exposure–lenalidomide RDI analysis, the proportion of patients with a lenalidomide RDI of ≥60% was determined for each ixazomib exposure quartile and the relationship was analyzed using logistic regression. The RDI cut-off of 60% was empirically selected as it represents a one-level lenalidomide dose reduction (from 25 to 15 mg).

1