Model-based meta-analysis of non-small cell lung cancer with standard of care PD-1 inhibitors and chemotherapy for early development decision making

Abstract

Single-arm cohorts/trials are often used in early phase oncology programs to support preliminary clinical activity assessments for investigational products, administered alone or in combination with standard of care (SOC) agents. Benchmarking clinical activity of those combinations against other treatments, including SOC, requires indirect comparisons against external trials, which presents challenges including cross-study differences in trial populations/other factors. To facilitate such nonrandomized comparisons, we developed a comprehensive model-based meta-analysis (MBMA) framework to quantitatively adjust for factors related to efficacy in metastatic non-small cell lung cancer (mNSCLC). Data were derived from 15 published studies assessing key programmed cell death protein-1 (PD-1) inhibitors pembrolizumab (n = 8) and nivolumab (n = 7), representing current SOC in mNSCLC. In the first stage, a mixed-effects logistic regression model for overall response rate (ORR) was developed accounting for effects of various population covariates on ORR. The ORR model results indicated an odds ratio (OR) of 1.02 for squamous versus non-squamous histology and OR of 1.20 for PD-ligand 1 tumor proportion score (TPS) per every 10% increase of TPS level. Next, a nonparametric mixed-effects model for overall survival (OS) was developed with ORR/other clinical covariates as input. Subsequently, MBMA simulations of relevant hypothetical scenarios involving single-arm trial design predicted OS hazard ratios as a function of ORR with matched patient characteristics. Findings from this MBMA and derived parameter estimates can be generally applied by the reader as a framework for interpreting efficacy data from early phase trials to support ORR-based go/no-go decisions and futility rules, illustrated through examples in this report.

Conflict of interest statement

D.C.T., H.Z., X.W., C.V., M.K., T.T.C., M.B., and S.A.G.V. were employees of and stockholders in GSK at the time this work was completed. D.C.T. is a former employee of GSK and currently employed by Genentech. R.W. was an employee in Certara (Certara was a consultant to GSK) at the time this work was completed, and is currently employed at QuanTx Consulting. X.W. is a former employee of GSK and currently employed by Moderna. C.V. is a former employee of GSK and currently employed by J&J. M.B. is a former employee of GSK and currently employed by Novocure. R.G., L.Z., and N.Z. are employees of Certara, and consultants for GSK.

© 2023 GSK, GlaxoSmithKline. CPT: Pharmacometrics & Systems Pharmacology published by Wiley Periodicals LLC on behalf of American Society for Clinical Pharmacology and Therapeutics.

Figures

FIGURE 1

The NSCLC MBMA framework defines ORR and ORR‐to‐OS relationships across lines of therapy and treatment types for generating synthetic control arms and projecting OS. Chemo, chemotherapy; MBMA, model‐based meta‐analysis; nivo, nivolumab; NME, new molecular entity; NSCLC, non‐small cell lung cancer; ORR, overall response rate; OS, overall survival; Ph, phase; pembro, pembrolizumab; SOC, standard of care.

FIGURE 2

Illustration of conditional probability method for modeling aggregate survival curves. Pj, conditional probability; Sj, survival up to the beginning of the interval j.

FIGURE 3

Forest plot of the impact of ORR model covariates for PD‐1 inhibitors and relevant chemotherapy comparators from selected pembrolizumab and nivolumab clinical trials (n = 15). The center line represents the modeled ORR value for a patient with 50% squamous histology and 50% PD‐L1 expression receiving 2L PD‐1 inhibitor with chemotherapy. Each point illustrates the modeled ORR rate for the univariate change described, with the indicated 90% CI. “PD‐L1 superhigh” is defined as greater than or equal to 90% PD‐L1 tumor expression and corresponds to imputed continuous PD‐L1 expression value of 95%. “PD‐L1 high” is defined as greater than or equal to 50% PD‐L1 tumor expression and corresponds to imputed continuous PD‐L1 expression value of 70%. “PD‐L1 weak” is defined as 1%–49% PD‐L1 tumor expression and corresponds to imputed continuous PD‐L1 expression value of 25%. “PD‐L1 negative” is defined as less than 1% PD‐L1 tumor expression and corresponds to imputed continuous PD‐L1 expression value of 0%. 1L, first line; ≥2L, second line or greater; CI, confidence interval; combo, combination therapy; ORR, overall response rate; PD‐1, programmed cell death protein‐1; PD‐L1, programmed death‐ligand 1.

FIGURE 4

Exploratory ORR‐to‐OS correlation of SOC PD‐1 inhibitor and chemotherapy alone for (a) median OS and (b) OS at 1‐year. Any IO, any immuno‐oncology therapy containing arm; chemo alone, chemotherapy alone; ORR, overall response rate; OS, overall survival; PD‐1, programmed cell death protein‐1; SOC, standard of care.

FIGURE 5

Forest plot of the impact of covariates on OS. The center line represents the modeled OS value for a patient with 50% squamous histology and 50% PD‐L1 tumor expression receiving 2L PD‐1 inhibitor in combination with chemotherapy. Each point illustrates the modeled OS rate for the univariate change described with whiskers representing a 90% confidence interval. “PD‐L1 superhigh” is defined as 95% PD‐L1 tumor expression. “PD‐L1 high” is defined as 70% PD‐L1 tumor expression. “PD‐L1 weak” is defined as 25% PD‐L1 tumor expression. “PD‐L1 negative” is defined as 0% PD‐L1 tumor expression. 1L, first line; ≥2L, second line or greater; IO, immuno‐oncology; OS, overall survival; PD‐L1, programmed death‐ligand 1.

FIGURE 6

ORR difference achieved for hypothetical phase I/II cohorts of IO and chemotherapy in (a) “base case” (b) “low responder” (c), and “high responder” scenarios. A “base case” scenario for patients with 20% squamous histology, 30% continuous PD‐L1 TPS. A “low responder” scenario for patients with 10% squamous histology, 15% continuous PD‐L1 TPS. A “high responder” scenario for patients with 30% squamous histology, 50% continuous PD‐L1 TPS. HR, hazard ratio; IO, immuno‐oncology; ORR, overall response rate; PD‐L1, programmed death‐ligand 1; sq, squamous histology, TPS, tumor proportion score.