The cost-effectiveness of alectinib versus crizotinib in first line anaplastic lymphoma kinase-positive (ALK+) advanced NSCLC
PharmacoEconomics
Josh J Carlson, Kangho Suh, William Wong, Panos Orfanos
Corresponding Author:
Josh J. Carlson, MPH, PhD
Associate Professor
University of Washington
1959 NE Pacific St., H-375I
Box 357630
Seattle, WA 98195-7630
Tele: 206.543.9649
Fax: 206.543.3835
ELECTRONIC SUPPLEMENTARY MATERIAL
1 SUPPLEMENTARY DISCUSSION, TABLES, AND FIGURES
1.1 Model Choice Discussion
We developed our model using partition survival methods after evaluating the logged cumulative hazard plots for progression free and overall survival (see Figures S1, S2). These plots are intersecting in multiple points, indicating that the proportional hazards assumption may not hold and that each treatment arm should be analyzed separately.
Data for alectinib and crizotinibprogression free survival (PFS) and overall survival (OS) were derived from the BO28984 study.[1, 2] We fit parametric survival functions to independent review committee-assessed Kaplan Meier (KM) data. Upon review, the exponential distribution for the PFS and OS curves were most appropriate based on goodness of fit (BIC) and visual fit(See Tables S1, S2; FiguresS3, S6). Clinical plausibility of the exponential distributions was also assessed with the assistance of medical experts, and based on their input and internal data, the exponential distributions were considered to align with progression and survival expectations.
1.1.1 Rationale for Progression Free Survival Fit
Based on goodness of fit alone, the log-normal distribution is the optimal fit for both alectinib and crizotinib arm in PFS. However, the log-normal distribution is prone to long tails and generally provides implausible extrapolations. This was supported by clinical advice from a Lung Cancer Advisory Board meeting consisting of 12 consultant oncologists specializing in lung cancer treatment that was held in the United Kingdom in July 2017. The different survival extrapolation models (PFS and OS) were presented to theclinicians (both in graphical form and in predicted % of patientsalive and on progression on annual intervals). The feedback was collected on open discussion and assessment was synthesized narratively for each individual and on aconsensusbasis.The plausibility of PFS extrapolations was based on the comparison between the predicted annual % of patients in the progression free state and the estimated proportions from clinicians' experience.For OS, the extrapolated estimations were judged by thecliniciansin light of the perceived 5-year survival rate, and the hazard link between alectinib and crizotinib and the subsequent treatment setting.As a result it was determined that the extrapolation of PFS with a log-normal distribution would be too optimistic. Furthermore, the log-normal fit has been shown to be less consistent with PFS for crizotinib from the PROFILE 1014 trial.[3] As a result, the most clinically plausible extrapolations were determined to be exponential or Weibull for both arms. Between the two, the exponential had the better BIC for alectinib and crizotinib. Further, the medical experts supported the exponential distribution as they believed it offered a more realistic and clinically plausible fit regarding the proportion of patients in PFS within a 5-year time frame. Along with the clinical guidance, and after review of the goodness of fit and curve fits to the KM data, we felt exponential extrapolations provided the most plausible model and clinical choice.
1.1.2 Rationale for Overall Survival Fit
The KM data for overall survival was limited as neither arm reached the median endpoint. As a result, we understood that the extrapolation of an immature dataset may lead to high uncertainty. Based on the goodness of fit tests alone, exponential was the optimal fit for alectinib and log-normal for crizotinib. The log normal was considered an implausible distribution for overall survival as well given the propensity for long tails and unrealistic mean estimates. We also, wanted to stay consistent in choosing one extrapolation for both arms when possible, and after reviewing the curve fits and considering the clinical plausibility of each, we felt the exponential extrapolation provided the best choice.
1.3Supplementary Tables
Table S1 Parametric functions fitting for overall survival
Alectinib / CrizotinibModel / AIC / BIC / AIC / BIC
Exponential / 246.59 / 249.61 / 234.27 / 237.26
Weibull / 247.98 / 254.03 / 232.79 / 238.74
Log-Normal / 247.97 / 254.02 / 230.96 / 236.91
Gamma / 249.79 / 258.86 / 232.95 / 241.84
Log-Logistic / 247.91 / 253.96 / 232.18 / 238.13
Gompertz / 248.59 / 254.63 / 234.80 / 240.76
AIC: Akaike information criteria. BIC: Bayesian information criteria
Table S2 Parametric functions fitting for progression free survival
Alectinib / CrizotinibModel / AIC / BIC / AIC / BIC
Exponential / 381.93 / 384.96 / 384.43 / 387.42
Weibull / 378.95 / 385.00 / 384.39 / 390.34
Log-Normal / 371.85 / 377.90 / 370.81 / 376.77
Gamma / 369.76 / 378.83 / 369.42 / 378.31
Log-Logistic / 376.07 / 382.12 / 375.09 / 381.04
Gompertz / 383.93 / 389.98 / 386.48 / 392.44
AIC: Akaike information criteria. BIC: Bayesian information criteria
Figure S1 Log cumulative hazard plot of progression free survival
Figure S2 Log cumulative hazard plot of overall survival
Figure S3 Exponential curve fits for overall survival
OS: overall survival. KM: Kaplain-Meier
Proportion of patients alive based on KM data and model fit
Treatment / 1 yearAlectinib (KM data) / 0.843
Alectinib (exponential fit) / 0.851
Crizotinib (KM data) / 0.825
Crizotinib (exponential fit) / 0.815
Figure S4 Weibull curve fits for overall survival
OS: overall survival. KM: Kaplain-Meier
Proportion of patients alive based on KM data and model fit
Treatment / 1 yearAlectinib (KM data) / 0.843
Alectinib (Weibull fit) / 0.844
Crizotinib (KM data) / 0.825
Crizotinib (Weibull fit) / 0.837
Figure S5 Gamma curve fits for overall survival
OS: overall survival. KM: Kaplain-Meier
Proportion of patients alive based on KM data and model fit
Treatment / 1 yearAlectinib (KM data) / 0.843
Alectinib (Gamma fit) / 0.838
Crizotinib (KM data) / 0.825
Crizotinib (Gamma fit) / 0.820
Figure S6 Exponential curve fits for progression free survival
PFS: progression free survival. KM: Kaplain-Meier
Proportion of PFS patients based on clinical feedback and model fit
Treatment / 1 year / 2 years / 3 years / 4 yearsAlectinib (clinical feedback) / 0.70 / 0.50-0.60 / - / 0.30
Alectinib (exponential fit) / 0.708 / 0.501 / 0.355 / 0.251
Crizotinib (clinical feedback) / - / 0.20 / 0.05 / 0.01-0.02
Crizotinib (exponential fit) / 0.494 / 0.244 / 0.121 / 0.060
Figure S7 Weibull curve fits for progression free survival
PFS: progression free survival. KM: Kaplain-Meier
Proportion of PFS patients based on clinical feedback and model fit
Treatment / 1 year / 2 years / 3 years / 4 yearsAlectinib (clinical feedback) / 0.70 / 0.50-0.60 / - / 0.30
Alectinib (Weibull fit) / 0.689 / 0.526 / 0.414 / 0.332
Crizotinib (clinical feedback) / - / 0.20 / 0.05 / 0.01-0.02
Crizotinib (Weibull fit) / 0.501 / 0.218 / 0.089 / 0.035
Figure S8 Gamma curve fits for progression free survival
PFS: progression free survival. KM: Kaplain-Meier
Proportion of PFS patients based on clinical feedback and model fit
Treatment / 1 year / 2 years / 3 years / 4 yearsAlectinib (clinical feedback) / 0.70 / 0.50-0.60 / - / 0.30
Alectinib (Gamma fit) / 0.657 / 0.554 / 0.498 / 0.461
Crizotinib (clinical feedback) / - / 0.20 / 0.05 / 0.01-0.02
Crizotinib (Gamma fit) / 0.460 / 0.277 / 0.198 / 0.154
Figure S9 Incremental cost-effectiveness plane
References
1.Peters S, Camidge DR, Shaw AT, Gadgeel S, Ahn JS, Kim D-W, et al. Alectinib versus crizotinib in untreated ALK-positive non-small-cell lung cancer. N Engl J Med. 2017;Jun 6.
2.Genentech. Data on File. 2017.
3.Mok T, Kim D, Wu Y, Nakagawa K, Mekhail T, Felip E, et al. Overall survival (OS) for first-line crizotinib versus chemotherapy in ALK+ lung cancer: updated results from PROFILE 1014. Annals of Oncology. 2017;28(suppl 5):v605-v49.