S1782 impact of this variability on response categorization and the resulting overall response rate (ORR) in a specific trial has been practically unpredictable. Method: We built up a hierarchical model of measurement variability using a clinical trial dataset of CT scans. Simulations were then performed using the model 1) to establish the behavior of differences between the first and the hypothetical second assessments of percent change of tumor burden in various scenarios, 2) to elaborate on the probabilistic nature of decisions about categorization, and 3) to estimate the variation in the ORR due to measurement variability. Result: The extent of the discrepancies between assessments of the percent change depended on the baseline burden. Smaller differences were associated with larger shrinkage of tumor burdens. The simulated probability for a specific categorization (-30% or 20%) to result from reassessment had a sigmoid shape depending on the percent change in the first set of readings, inflecting at the cutoff point for the categorization. In 3 virtual trials having the same baseline burden and the same ORR of 50%, the presence of fewer percent changes around the cutoff in a trial resulted in a more reproducible ORR (95% central range, 35%-65% vs. 40%-60% vs. 45%-60%). Conclusion: Since determinations of partial response or progression are probabilistic outcomes due to measurement variability, quantification of the variation in the ORR by potential measurement variability is essential and will help inform decisions made on the basis of trial data. Keywords: Oncology, objective response rate, Measurement Variability
Journal of Thoracic Oncology
Vol. 12 No. 11S2
DOS correlated with DPFS (r¼0.50, p<0.0001). With CO in <20% of patients or unstated %CO (n¼144), mean DOS and DPFS were 0.93 and 0.92 months, respectively. With CO in 20% of patients (n¼57), mean DOS and DPFS were 1.29 and 1.41 months, while with CO>50% (n¼20), they were 1.4 and 1.9 months. OS HRs (mean¼0.92) were inferior to PFS HRs (mean¼0.82, n¼196, p<0.0001), although OS and PFS HRs correlated with each other (r¼0.64, p<0.0001). With CO<20% or unstated (n¼135), mean OS and PFS HRs were 0.93 and 0.83, while with CO>20% (n¼61), they were 0.90 and 0.80, and with CO>50% (n¼20), they were 0.94 and 0.71. Conclusion: OS HRs were inferior to PFS HRs, probably due to PPS, competing causes of death and CO. The better mean gains and HRs in high vs low CO trials may be due to more frequently allowing CO in trials with more effective therapies. This increases risk of false-negative OS results with effective therapies if CO is permitted, but it is potentially unethical to withhold CO of effective therapies. With PFS, clinically insignificant gains may be statistically significant. Since DOS and DPFS are similar, an alternate approach would be a primary study outcome requiring PFS HR to be statistically significant and DPFS 95% CIs in a range considered clinically relevant for OS gains. To better understand the limitations of this approach, we are analyzing examples with minimal OS gains despite DPFS>2 months and examples of DOS>2 months but no gain in PFS, and have formulated a potential biological/statistical explanation for the latter. Keywords: Progression-free survival
OA 14.03 Ontario’s Bundled Payment System for Systemic Therapy Supports Lung Cancer Trials W. Evans, T. Kais-Prial, R. Fung, L. Forbes Cancer Care Ontario, Toronto, ON/CA
OA 14.02 Rethinking Progression-Free Survival (PFS) as a Clinical Trials Surrogate for Overall Survival (OS) D. Stewart,1 D. Bosse,2 A. Ocana,3 G. Goss,1 D. Jonker1 1Medicine, University of Ottawa, Ottawa/CA, 2Harvard University, Boston/US, 3 Albacete University Hospital, Albacete/ES Background: OS assessment requires high follow-up times and patient numbers and is impacted by crossover (CO). OS hazard ratios (HRs) are generally inferior to PS HRs due to impact of post-progression survival (PPS) and CO. Some authors propose that absolute OS gains (DOS) should be similar to those in PFS (DPFS). Hence, DPFS might be a useful OS surrogate (Clin Cancer Res 2013;19:2646; Ann Oncol 2016;27:373). Method: To assess this further, we reviewed Journal of Clinical Oncology and New England Journal of Medicine 01/01/2012-06/12/ 2017 for randomized drug trials in incurable solid tumors. We extracted data for PFS and OS medians and HRs, calculated DPFS and DOS (experimental medians minus control medians), and did paired comparisons between 2-6 different arms in each study (245 comparisons across 201 trials). Result: Mean DOS across studies (1.03 months) was similar to mean DPFS (1.06 months) (n¼201 evaluable, p¼0.88).
Background: Clinical trials (CTs) are recognized as a key component of a quality cancer care system. When funding for systemic therapy (ST) services in Ontario transitioned in 2014/15 from a one-time payment for new cases to bundled payments for specific evidence-informed regimens, stakeholders expressed concern that the funding model could exclude patients from participation in CTs as treatment facilities would only receive funding when evidence-informed regimens were used. Method: A CT policy was implemented to enable public funding through the ST funding model for older and inexpensive drugs and their administration within randomized CTs at the level of the standard of care. Non-randomized CTs were to be funded at the level of best supportive care or other appropriate funding level. New and expensive drugs in a CT could be funded through a separate provincial drug reimbursement program if used according to public funding criteria with administration costs covered by the ST funding model. Each new CT is now assessed to determine the level of public funding possible. The funding model can now capture data on phase of trial, disease type, treatment regimen, trial purpose (adjuvant, palliative) and patient accrual by treatment facility Result: During 2015/16 and 2016/17, 43 and 44 lung CTs, respectively, were assessed and activated in Ontario. Trial accrual increased by 33% (from 311 to 413 patients) over the two years since the introduction of the funding model. Accrual varied by facility. In 2016/17, it ranged from a low of 0.25% to 37.5% of the new lung cancer (LC) patients seen at individual facilities. For the five largest cancer centers in Ontario, the percentage of patients recruited ranged from 5% to 18.3% and total accruals from these centers (n ¼ 257) comprised 63% of provincial LC trial recruitment. Five immunooncology trials accrued 183 patients and made up 44% of total LC trial accruals. Public funding through the ST funding model amounted to $415,000 in 2015/16 and increased to $815,000 in 2016/17. Conclusion: The new bundled payment system for ST and the CT policy have enabled public funding to support lung CTs. The ST funding model has facilitated the capture of CT data and trends not previously available for LC and other tumors. The new provincial CT policy and