Patient Selection Biomarkers in Drug Development:A first step towards individualized therapyMichael OstlandGenentech BioOncologyApril 21, 2005
Outline • Background • Some Challenges of a Drug Development Program that includes Biomarkers • Decision Making • Logistical and Technical • Wide-scale Screening of Potential Biomarkers • Examples • Discussion
Background Very rough summary of “normal” drug development* :
Background (2) • Considerable inter-patient variability in treatment benefit is the norm. For many highly effective drugs, many patients won’t benefit at all • There are several examples of drugs/indications where some of this variability is account for: • Study of clinical prognostic factors • Study of clinical predictive factors • PK differences from C-P450 enzymes • Drugs targeted to a patient sub-population defined by a specific molecular biomarker (often related to the MOA of the drug)
What is a Biomarker? A biomarker is a characteristic that is objectively measured and evaluated as an indicator of normal biologic processes, pathogenic processes, or pharmacologic responses to a therapeutic intervention (Biomarkers Definitions Working Group). Our interest is in the identification of baseline-measured biomarkers that may be predictive of clinical benefit following subsequent therapy.
Background (4) Identification of molecular biomarkers may allow drugs to be targeted to treat those patients who will benefit most from the therapy. Increased benefit could come from increased efficacy and/or decreased toxicity.
Background (5) “Individualized therapy” has enormous implications for the health care system and pharmaceutical market place • Less trial-and-error in prescribing • Smaller pool of eligible patients • Greater benefit in these patients could mean • Great competitive advantage • Easier to get re-imbursement from insurers
Challenges • Decision Making • Logistical and Technical • Screening Potential Biomarkers
Decision Making Phase II trials are used to get an early indication of efficacy, further assess safety, and identify a promising dosing regimen. The overall objective is (usually) to allow a decision whether to proceed into lengthy and costly phase III trials to confirm efficacy. Assessing a potential biomarker adds another layer of complexity to the decision-making process.
Typical Oncology Randomized Phase II Trial Chemo + Placebo Compare safety and efficacy (usually tumor response rate or time to disease progression) among the three arms. Enrolled Patients randomize Chemo + low dose drug Chemo + high dose drug • Design the study so we are likely to have adequate information to choose among three possible decisions: • Proceed to phase III with low dose drug • Proceed to phase III with high dose drug • Do not proceed to phase III at this time
Decisions with a Biomarker There are seven possible decisions about Phase III when dose and biomarker questions are part of the development
Positive assay Negative Indeterminate Positive Positive assay assay Negative Negative Indeterminate Indeterminate Phase II Trial w/ Biomarker & Dose Design with retrospective Dx testing Chemo + Placebo Enrolled Patients randomize Chemo + low dose Chemo + high dose
Logistical & Technical Issues • Identification of a potential biomarker • Timely development and technical validation of a commercially viable assay (“Dx test”) • Acquiring usable patient tissues with proper informed consent from clinical trials • Dealing with indeterminate assay results • Clinical validation of the predictive value of the Dx test, including CDRH regulations.
Wide-scale Screening of Potential Biomarkers • Technologies such as DNA microarrays allow simultaneous assaying of thousands of potential biomarkers. • Approach: Assay samples from a randomized clinical trial; identify markers where assay response is associated with treatment benefit. • Examine functional data on the identified markers • Test a small number of the most promising markers in a second trial (ideally prospectively)
Screening Details • Rank all markers using a statistical model that models clinical outcome as a function of each marker (one at-a-time), treatment group, and other clinical covariates (if appropriate). • Determine a cut-off that accounts for sampling variability & multiplicity • Estimate the magnitude of the association between biomarker and treatment effect • Work with Bioinformatics group to interpret, refine, and iterate.
Screening Details (2) • Account for multiplicity using a False Discovery Rate(FDR) controlling procedure (Benjamini & Hochberg; Storey). • FDR is defined as the expected proportion of rejected hypotheses that are mistakenly rejected (“falsely discovered”). • Less conservative than FWER controlling procedures, so may be more appropriate for hypothesis generation.
Screening Example • 60 patients with refractory ovarian cancer were randomized equally to standard chemo with or without experimental drug X. • 42 had usable tissue samples that were run on Affymetrix Microarrays with ~40K mRNA probes. • Primary endpoint was duration of PFS • Prior response to first line regimen of platinum-based chemotherapy is a known prognostic factor
Ranking Genes For each of J biomarkers, fit a univariate Cox-PH model. Let represent the hazard function for the kth patient in the model for the jth biomarker: where Tk is treatment indicator, Xkj is the (log) expression measure for the jth gene in the kth subject, and greek letters are unknown parameters. Zk is an indicator that the kth subject was known to be resistant to platinum-based chemotherapy at the time of randomization.
Ranking Genes (2) Let Tj be the usual Wald test statistic of Following the development of Dudoit et al (2003), calculate unadjusted p-values, pj, with a permutation procedure. Then calculate adjusted p-values Where r1,…, rJ is a sequence that puts the unadjusted p-values in ascending order. Then selecting genes with provides strong control of the FDR (Benjamini and Hochberg, 1995)
Results • 72 genes identified at the 10% FDR • Bioinformatics analysis & follow-up ongoing • Modeling and proper interpretation require collaboration between clinical scientists, statisticians, and bioinformaticians
Tarceva Phase III Study BR.21 in NSCLC • BR.21 (NCI Canada, OSIP): Tarceva monotherapy vs. placebo in chemotherapy-relapsed (2nd/3rd line) NSCLC • The primary endpoint was survival. Secondary endpoints were tumor response, tumor response duration, progression-free survival, QoL, and to correlate the expression of EGFR levels with outcomes. • 731 patients were randomized 2:1 to Tarceva or placebo. • Designed to detect a 33% improvement in overall survival with 90% power
BR21: All patients 1.00 Tarceva Median = 6.7 mo (n=488) Placebo Median = 4.7 mo (n=243) Total N=731 0.75 1-yr Survival = 31%1-yr Survival = 21% 0.50 Survival Distribution Function 0.25 0.00 0 5 10 15 20 25 30 Survival Time (Months)
EGFR IHC and benefit from EGFR TKI • Rationale: Tumors which express target should be more likely to respond than tumors which don’t • Does the assay actually identify subgroups with differential benefit? • Can tumors which “don’t express the target” benefit from treatment?
BR.21 Survival by EGFR IHC status(data from 33% of patients) • EGFR values using “presence of staining in 10% or more cells” as positive/negative cut point - 53% were EGFR positive • EGFR IHC (+) survived significantly longer when treated with Tarceva vs. placebo in BR.21 • EGFR (-) showed no evident survival benefit with treatment in BR.21 but confidence interval for the EGFR(-) subset is wide.
Resulting Label Content on EGFR IHC Relation of Results to EGFR Protein Expression Status (as Determined by Immunohistochemistry)Analysis of the impact of EGFR expression status on the treatment effect on clinical outcome is limited because EGFR status is known for only 238 study patients (33%). EGFR status was ascertained for patients who already had tissue samples prior to study enrollment. However, the survival in the EGFR tested population, and the effect of TARCEVA were almost identical to that in the entire study population, suggesting that the tested population was a representative sample. A positive EGFR expression status was defined as having at least 10% of cells staining for EGFR in contrast to the 1% cut-off specified in the DAKO EGFR pharmDx™ kit instructions. The use of the pharmDx kit has not been validated for use in non-small cell lung cancer. TARCEVA prolonged survival in the EGFR positive subgroup (N = 127; HR = 0.65; 95% CI = 0.43 — 0.97) (Figure 3) and the subgroup whose EGFR status was unmeasured (N = 493; HR = 0.76; 95% CI = 0.61 — 0.93) (Figure 5), but did not appear to have an effect on survival in the EGFR negative subgroup (N = 111; HR = 1.01; 95% CI = 0.65 — 1.57) (Figure 4). However, the confidence intervals for the EGFR positive, negative and unmeasured subgroups are wide and overlap, so that a survival benefit due to TARCEVA in the EGFR negative subgroup cannot be excluded.
Conclusions • Patient selection via biomarkers promises to reshape the industry and patient care • There are interesting and challenging problems for drug development with a Dx test, and statisticians will play an important role in meeting these challenges • Regulators are keen to get information in the drug’s label on possible biomarkers
Acknowledgments • Xiaolin Wang • Ben Lyons • Gracie Lieber • Cheryl Jones • Alex Bajamonde