welcome back! what did we learn yesterday? what’s happening in nsclc – updates from asco...

Download WELCOME BACK! What did we learn yesterday? What’s happening in NSCLC – updates from ASCO Clinical trial endpoint selection Basics principles of Phase

If you can't read please download the document

Upload: eugene-goodwin

Post on 18-Jan-2018

214 views

Category:

Documents


0 download

DESCRIPTION

What did we learn yesterday? What’s happening in NSCLC – updates from ASCO Clinical trial endpoint selection Basics principles of Phase 1 trials Design considerations in Ph II studies Design considerations for biomarker-driven trials

TRANSCRIPT

WELCOME BACK! What did we learn yesterday? Whats happening in NSCLC updates from ASCO Clinical trial endpoint selection Basics principles of Phase 1 trials Design considerations in Ph II studies Design considerations for biomarker-driven trials What will we learn today? How to optimize clinical trial design in oncology The role of translational research Impact of key statistics principles and methodology How to practice good clinical trial compliance Your role as a PI How to evaluate well-designed vs poorly designed clinical trials What you are studying! Research project coaching Afternoon coaching session guidance You will have the opportunity to present and receive guidance on your ongoing research project All delegates must participate! Each participant will have 5 minutes to present Informally, with or without slides, you can use flipcharts or just speak You should include: What clinical question do you hope to answer? Project objective(s) Project planning/implementation Potential results/conclusions Consider any questions for TOP network(including faculty) or opportunities for partnership with co-delegates Following each presentation there will be 5 minutes of discussion/questions 7 David R. Gandara, MD University of California Davis Comprehensive Cancer Center Translational Research in NonSmall Cell Lung Cancer (NSCLC): What Are the Available Tools & How Can You Use Them? 8 From All Patients Are the Same to Inter- and Intra-Patient Heterogeneity in Tumour Biology & Immuno-Biology From Histology to Prognostic and/or Predictive Biomarkers Adapted from Gandara et al. Clin Lung Cancer Personalised Therapy for Individual Patients With NSCLC Translational Research Available Tools for Translational Research in NSCLC: Interaction of Pathology, Omics, and Immuno-Biology 9 Moving From Histologic Classification to Molecular Classification of NSCLC Patients Into Prognostic or Predictive Subgroups for Therapy Histologic subtyping groups tumours based on microscopic pattern recognition by a pathologist At best, Histology = crude molecular selection Positive ALK FISH EGFR Mutation 10 Li, Mack, Gandara et al. JCO (adapted from Pao et al). Evolution of NSCLC Subtyping From Histologic to Molecular-Based NSCLC as one disease ALK EGFR First Targeted Therapies in NSCLC 11 Adapted from The Cancer Genome Atlas Project: Govindan & Kondath et al. Nature Magnitude of Genomic Derangement Is Greatest in Lung Cancer 1 / Mb 10 / Mb 100 / Mb 0.1 / Mb n= Hematologic & Childhood Cancers Carcinogen-induced Cancers ?? Adenoma Squamous Ovarian, Breast, Prostate Cancers Mutations Per Mb DNA 12 Lung Cancer Complexity on an Individual Patient Basis: Squamous-Cell Lung Cancer Examples (Circos) LUSC LUSC LUSC LUSC LUSC LUSC LUSC LUSC From Ramaswamy Govindan. TCGA (The Cancer Genome Atlas). 13 From Li, Gandara et al. JCO Integration of Biomarkers Into Clinical Practice: Past, Current & Future Near-Future Approach (Patient-Based Therapy): Genomic profiling by high throughput next generation sequencing for decision-making in individual patients 1.Histomorphological Diagnosis: Cancerous Evolving Approach (Target-Based Therapy V2.0): Multiplexed molecular tests with increased sensitivity & output for decision-making in individual patients Current Approach (Target-Based Therapy V1.0): Single gene molecular testing for decision-making in individual patients 2. Molecular Diagnosis: Extract tumour nucleic acids: Archival cancer specimens Archival FFPE tumour specimens Macro- or Micro-dissection of Tumours DNA and RNA Empiric Approach (Past) (Compound-Based Therapy): Clinical-histologic factors to select drugs for individual patients Representative technologies: Single Biomarker Tests: Sanger DNA Sequencing RT-PCR FISH IHC Multiplex, Hot Spot Mutation Tests: PCR-based SNaPshot PCR-based Mass Array SNP Sequenom Initial High-Throughput Technologies: SNP/CNV DNA microarray RNA microarray Next-Generation Sequencing (NGS): Whole Genome or Exome Capture Sequencing (DNA) Whole or Targeted Transcriptome Sequencing (RNA) Epigenetic profiling Plasma cfDNA by NGS 14 Comprehensive Cancer Genomic Test: 200+ Genes 6 months. Mok et al. ESMO-Ann Oncol. 2014;25(suppl 4): abstr LBA2_PR. Primary endpoint: PFS R 1:1 PD Stage IIIB/IV NSCLC EGFR mutation positive WHO PS 01 Prior response* to 1st-line gefitinib PD 1 Trt B > Trt A Rel. difference < 1 Trt B < Trt A Effect size for sample size Minimum difference of clinical relevance Maximum difference of clinical irrelevance Abs. = absolute; Rel. = relative. Effect size Sample size Type I error False-positive rate We cannot completely eliminate such random error Lower type I error Larger sample size Usually 5% e.g. if, in fact, Trt B has the same effect as Trt A, we have 5% chance to wrongly conclude that Trt B is different from Trt A The declared difference is not always true Type I and II errors Conclusion Total No differenceDifference Truth Difference False negative = Type II error True positive = Power 100% No differenceTrue negative False positive = Type I error 100% Sample size Type II error False-negative rate = 1 power, for a given effect size We cannot completely eliminate such random error Lower type II error Higher power Larger sample size Usually 10 20% type II error (or 80 90% power) e.g. if, in fact, Trt B is different from Trt A, we have 20% chance to wrongly conclude that Trt B is the same as Trt A The true difference is not always detected Type I and II errors Conclusion Total No differenceDifference Truth Difference False negative = Type II error True positive = Power 100% No differenceTrue negative False positive = Type I error 100% Sample size Equal allocation (1:1) The total sample size is smallest Most efficient Unequal allocation Less efficient, i.e. larger total sample size Might be more attractive for patients e.g. smaller proportion to the control arm with placebo Example sample sizes Allocation ratio n A / n B nAnA nBnB n A + n B Sample size (Special cases not covered in this CONSORT statement) Sources of correlation Repeated measurements i.e. same endpoint assessed at different time points and all data are included in a single analysis Crossover i.e. each pt receives different treatments at different time periods Matched pairs e.g. different treatments for left and right hands Cluster randomisation e.g. randomise hospitals to different treatments, all patients of a given hospital receive the same treatment, outcome is measured at patient level Affects choice of statistical approach pt = patient. Correlated data Sample size Interim analyses and stopping guidelines Per-arm sample size n1n1 n2n2 nK=nnK=n (If continues) 1st interim analysis 2nd interim analysis Final ( K th ) analysis Sample size Number of interim analyses More interim analyses Larger final sample size Timing of interim analyses Allowing early stop for what? For benefit only: Stop if already see a large treatment effect For futility only: Stop if the chance to detect treatment effect is low For both: Stop if either benefit or futility criterion is reached Stopping boundary Easier to stop early Larger final sample size Interim analyses and stopping guidelines Sample size Stopping boundary Example: 3 analyses, overall type I error 0.05 Interim analyses and stopping guidelines Analysis number Stopping boundary for p-value Final analysis Easy to stop early Final boundary much smaller than 0.05 Difficult to stop early Final boundary close to 0.05 Sample size Stopping boundary Interim analyses and stopping guidelines Analysis number Stopping boundary for p-value Also other less extreme boundaries Final sample size: Sample size Independent data monitoring committee (IDMC) Avoid potential data-driven changes resulting in bias Members are independent of the trial, including medical experts of disciplines involved and statistician(s) Based on (un)blinded interim results, give sponsor a recommendation: Stop the trial Continue the trial with some modifications Continue the trial without modifications Interim analyses and stopping guidelines CONSORT 2010 Statement: Updated guidelines for reporting parallel-group randomised trials Sample size Randomisation Purpose and requirements Types of randomisation Statistical methods Outcomes and estimation Schulz KF, et al. BMC Med 2010;8:18. Biostatistics-related items Randomisation Purpose Ensure that patients assigned to different treatment arms are balanced for both known and unknown factors, which could influence their response to treatment Reduce bias Requirements Avoid systematic bias Unpredictable Purpose and requirements Randomisation Simple randomisation Restriction: No Example: Advantages: Simple to implement Straightforward for analyses Disadvantage: Cannot guarantee allocation ratio Types of randomisation ABBAAAAABA Randomisation Block randomisation Restriction: Yes, by block size Example: 1:1 ratio, block size=4 6 admissible blocks 1=AABB, 2=ABAB, 3=ABBA, 4=BBAA, 5=BABA, 6=BAAB Types of randomisation Block number4361 Treatment allocationBBAAABBABAABAABB The randomisation list is a sequence of admissible blocks in random order Randomisation Block randomisation Advantage: Allocation ratio is guaranteed within each full block Disadvantage: Introduce correlation between treatment arms Ordinary two-group tests might be biased Might require special test or adjusted analysis Considerations Block size too small Predictable Block size too large Risk of incomplete block Allocation ratio not met Use different block sizes within a trial Types of randomisation AABABABBAB Randomisation Stratified randomisation Restriction: Yes, by stratification factors Example: 2 stratification factors, 4 combinations, Separate randomisation lists for the 4 combinations Types of randomisation Previous treatment NoYes Stage IComb 1Comb 2 IIComb 3Comb 4 FactorStratumArm AArm B Stage I25%27% II75%73% Previous Trt No33%30% Yes67%70% Goal: Similar distributions between treatment arms with respect to each stratification factor Randomisation Types of randomisation Randomisation Minimisation Restriction: Yes, by stratification factors Example: 3 stratification factors Already 16 patients Next (17th) patient is stage I, not pretreated, biomarker negative Prev. = previous; Neg = negative; Pos = positive. Types of randomisation Trt ATrt B Stage I32 II56 Sum1011 Biomarker Neg44 Pos35 < Trt A to the next pt Goal: Similar distributions between treatment arms with respect to each stratification factor Prev. Trt No35 Yes53 Randomisation Minimisation Advantages Can take more stratification factors into account than stratified randomisation Try to maintain the allocation ratio within each stratum of each factor Disadvantages Dynamic allocation, no prepared randomisation lists Deterministic nature, not really random process Introduce correlation between treatment arms Ordinary two-group tests might be biased Might require special test or adjusted analysis Considerations Only take important prognostic factors as stratification factors Can build in a random element to reduce predictability Types of randomisation CONSORT 2010 Statement: Updated guidelines for reporting parallel-group randomised trials Sample size Randomisation Statistical methods Primary and secondary outcomes Multiplicity Subgroup analyses Adjusted analyses Missing values Outcomes and estimation Schulz KF, et al. BMC Med 2010;8:18. Biostatistics-related items Statistical methods For primary outcome Effect size or non-inferiority margin is pre-specified and used for sample size calculation Statistical significance Clinical relevance For secondary outcomes Effect size or non-inferiority margin is usually NOT pre-specified and NOT used for sample size calculation Statistical significance Clinical relevance (retrospectively defined) Primary and secondary outcomes Caution in interpretation! Statistical methods Indiv. = individual; Prob. = probability. Multiplicity Total number tests Indiv. type I error Prob. all tests reach correct conclusions Overall type I error 15%95% 5.00% 25% (95%) 2 = 95% 95% = 90.25% 9.75% 35%(95%) 3 = 85.74%14.26% 45%(95%) 4 = 81.45%18.55% Example Do 4 tests, each at 5% type I error Overall, 18.55% chance at least one test conclusion is wrong! There are statistical approaches to control overall type I error Statistical methods Problem More than one statistical test Increases overall type I error Want to control overall type I error for more than one test Sources of multiplicity Perform several tests for different endpoints For an endpoint, perform several tests at different times, e.g. interim analyses For an endpoint, perform several tests using different approaches For an endpoint, perform several tests in different subgroups In a 3-arm trial, perform 3 pairwise tests (A vs B, A vs C, B vs C) for a given endpoint Multiplicity Statistical methods Predefined subgroup analyses Number of analyses is known in advance Results of all analyses are reported Readers can better judge the credibility of the results, taking multiple testing into account Subgroup analyses not predefined Data-driven, fishing for significance! Tendency for biased reporting, i.e. only significant results Total number of performed analyses is unknown Readers cannot judge the credibility of the results Purely exploratory Sufficient sample size within each subgroup Subgroup analyses Statistical methods Take factors (e.g. prognostic factors, stratification factors, etc.) other than treatment group into account in the analysis As the primary analysis or sensitivity analysis for an outcome Example statistical approaches: Regressions, stratified analysis, etc. Adjusted analyses Statistical methods Missing at random? If missingness is related to the outcome measure Missing is not at random e.g. Patients with very bad performance status tend to have missing values for quality of life questionnaire Analysis using all available data is biased Other analysis approaches are needed Missing values Statistical methods Missing at random? Example NA = not applicable. Missing values Complete (sorted) data Mean=57.3 Available data set 1 (missing not at random) NA Mean=67.8 Available data set 2 (missing not at random) NA Mean=47.3 Available data set 3 (missing at random) NA45NA Mean=59.9 Can you be sure? Statistical methods Analyses using available data Results might be biased due to missing not at random Results might be biased because the balance in known and unknown factors introduced by randomisation might be destroyed Loss of efficiency In a univariate analysis UPN = unique patient number; CR = complete response; PR = partial response; SD = stable disease. Missing values UPNOutcomeAge 1CR90 2PR80 3SDMissing 4SD70 5CR80 6PRMissing Statistical methods Analyses using available data Loss of efficiency In a multivariate analysis Missing values UPNOutcomeAgeStagePack year 1CR903Missing 2PR80Missing25 3SDMissing440 4SD CR80Missing30 6PRMissing220 CONSORT 2010 Statement: Updated guidelines for reporting parallel-group randomised trials Sample size Randomisation Statistical methods Outcomes and estimation Point and interval estimates Confidence interval vs p-value Caution for final analysis after interim analyses Schulz KF, et al. BMC Med 2010;8:18. Biostatistics-related items Outcomes and estimation Point estimate Mean or median? Which one is more representative? Interval estimate Wider interval Larger variation Lower precision CI: Suitable under (approx) normal distribution, requires standard deviation IQR: 1st quartile 3rd quartile, 25th percentile 75th percentile CI = confidence interval; IQR = inter-quartile range. Point and interval estimates Mean: 4 95% CI: 3.14.9 Median: 4 IQR: 35 Mean: % CI: 1.94.2 Median: 2 IQR: 1.84.3 Outcomes and estimation CI examples Alt. hypothesis: Inequality (2-sided) Type I error: 5% Effect size: Absolute difference between treatment arms Minimum difference of clinical relevance: 2 CI vs p-value 2-sided 95% CI Absolute difference 2) The width of CI provides hints on precision 3) The estimated effect size is likely to be greater than the minimum difference of clinical relevance 1) The CI does not contain 0 Reject null hypothesis, i.e. statistical significance p-value < Type I error 5% Outcomes and estimation CI examples CI vs p-value Absolute difference Clinically relevant: Yes ? ? Statistically significant: Yes Yes No Outcomes and estimation Information provided by CI and p-value CI vs p-value CIp-value Statistical significanceYes Clinical relevance (if pre-specified) YesNo Precision of estimateYesNo p-value alone is not sufficient for good judgment! Outcomes and estimation A simplified example: 3 successes in 4 patients considered promising Without interim analyses, the chance of a promising result = 5/16 Caution for final analysis after interim analyses Outcomes and estimation A simplified example: 3 successes in 4 patients considered promising With an interim analysis after 2 patients Stopping rule: If 2 failures , then stop The chance of promising results after 4 patients = 5/12 > 5/16 Caution for final analysis after interim analyses Final p-value and CI need to be adjusted Outcomes and estimation The threshold for p-value to declare statistical significance might be lower than the type I error e.g. a p-value of 0.05 might not be sufficient for significance Caution for final analysis after interim analyses Stopping boundary for p-value Analysis number Summary Well designed and properly executed randomised controlled trials (RCTs) provide the most reliable evidence on the efficacy of healthcare interventions Different statistical methods apply when the endpoint is discrete (frequency per category), continuous (measurements), or time-to- event (survival analysis) Statistical analysis requires careful consideration of the study objectives and the nature of the endpoints Complicating factors include multiplicity, subgroup analysis and missing data Considerations for how to present trial outcomes and estimations are point and interval estimates, the use of confidence interval vs. p-values and caution for final analysis after interim analyses. Biostatistics in Clinical Trials 141 Facilitated by Prof. Shu-Fang Hsu Schmitz Discussion: Biostatistics in clinical trials COFFEE BREAK Breakout Session 1 Breakout Format Please go to the room assigned to your group on the next slide for Breakout Session 1 Group A: Phase I and Phase II trial design in oncology Dr. Shu-Fang Hsu Schmitz Group B: good clinical practice compliance Dr. Clifford Hall Once the first breakout is complete, we will have a break for lunch Following lunch, you will return to your same breakout room, but the topic and faculty will be switched for Breakout Session 2 Group A: good clinical practice compliance Dr. Clifford Hall Group B: Phase I and Phase II trial design in oncology Dr. Shu-Fang Hsu Schmitz Following Breakout Session 2, please return to this room 145 Day 2 Breakout Rooms: Eixample and Grcia Group A: Dr. Shu-Fang Hsu Schmitz, Eixample Group B: Dr. Clifford Hall, Grcia Daniel GagiannisEmanuela SalatiArik SchulzeCristina Daniela Micu Marie-Claire DesaxAija Geria-BrziaMichael SchumacherErika Korobeinikova Xu Chong-RuiPedro De MarchiZhabina AlbinaRafael Caparica Nadezhda HamrinaJoan CovesAna GelattiEsther Holgado Patricia CruzSuneil KhannaVirginia CalvoJoaqun Mosquera Martinez Natalia FernandezKrista NoonanManuel Magalhes Barbara Melosky, David Gandara, Nick Pavlakis, Ralf Schnall, Angela Mrten, Verena Zahn Thierry Le Chevalier, Rosario Garcia-Campelo, Uday Bose, Georgi Adly, Tara Regan LUNCH Breakout Session 2 149 Day 2 Breakout Rooms: Eixample and Grcia Group A: Dr. Clifford Hall, EixampleGroup B: Dr. Shu-Fang Hsu Schmitz, Grcia Daniel GagiannisKrista NoonanArik SchulzeCristina Daniela Micu Marie-Claire DesaxEmanuela SalatiMichael SchumacherErika Korobeinikova Xu Chong-RuiAija Geria-BrziaZhabina AlbinaRafael Caparica Nadezhda HamrinaPedro De MarchiAna GelattiEsther Holgado Patricia CruzJoan CovesVirginia CalvoJoaqun Mosquera Martinez Natalia FernandezSuneil KhannaManuel Magalhes Barbara Melosky, David Gandara, Nick Pavlakis, Ralf Schnall, Angela Mrten, Verena Zahn Thierry Le Chevalier, Rosario Garcia-Campelo, Uday Bose, Georgi Adly, Tara Regan 151 David R. Gandara, MD University of California, Davis Comprehensive Cancer Center Evaluating Well-designed vs Poorly- designed Randomized Trials Evaluating Good vs Poorly Designed Randomized Clinical Trials The Good, The Bad and the Ugly Why do you want to do the study? Who do you want to study? How are you going to study them? What is the study design & primary study endpoint? Where are you going to conduct the study? When do you want to look at interim results, if at all? Who, What, Where, Why, When and more Randomized Clinical Trials: The Basics Why do you want to do the study? What is the hypothesis? Will the results change SOC or lead to definitive trials? Who do you want to study? What patient population? All comer or Selected/Enriched? What stratifications (for prognostic groups)? How are you going to study them? Comparison of different treatments? (or against BSC) QOL or Comparative Effectiveness? Randomized Clinical Trials: The Basics (contd) What is the study design & primary study endpoint? Randomized Phase II, Phase II/III or Phase III? How big a patient sample size needed to address the hypothesis? If Phase II, new treatment vs SOC or pick the winner Primary Endpoint: Response, PFS, OS or Other (QOL)) Where are you going to conduct the study? Single institution, multi-site in your country or Global If Global: Will there be issues of population-related pharmacogenomics? When do you want to look at interim results, if at all? Planned interim analysis? Is the study a Phase II/III with go-no go decision? Randomized Clinical Trials: The Basics (contd) Example: QUARTZ Trial of Whole Brain Radiotherapy vs Optimal Supportive Care for NSCLC patients with brain metastases (ASCO 2015) Good, Bad or UGLY? Whole brain radiotherapy for brain metastases from non-small cell lung cancer: Quality of life and overall survival results from the UK MRC QUARTZ trial PM Mulvenna, MG Nankivell, R Barton, C Faivre-Finn, P Wilson, B Moore, E McColl, I Brisbane, D Ardron, B Sydes, C Pugh, T Holt, N Bayman, S Morgan, C Lee, K Waite, RJ Stephens, MKB Parmar, RE Langley Brain Metastases and NSCLC After radical treatment of primary Non Small Cell Lung Cancer (NSCLC), the brain remains a frequent and early site of distant relapse, affecting up to 40% of patients Patients with NSCLC and brain metastases fare poorly even if irradiated Median survival remains poor RTOG RPA prognostic classes RPA I7.1 months RPA II 4.2 months all patients received WBRT; 57% NSCLC RPA III 2.3 months In the face of modest prognosis, how do we ensure optimal quality of life? In spite of lack of randomised, controlled data, whole brain radiotherapy (WBRT) plus steroids standard care R R QUARTZ Trial Randomised Controlled Non-Inferiority Design Histologically proven NSCLC with brain metastases non-resectable and unsuitable for stereotactic radiosurgery Control Arm: Optimal Supportive Care Dexamethasone + Whole Brain Radiotherapy 20Gy in 5 daily # Investigational Arm: Optimal Supportive Care Dexamethasone Primary outcome quality adjusted life years (QALYS) Secondary outcomes overall survival symptom scores March August 2014 Main Inclusion Criteria Pragmatism, Inclusivity Histologically proven primary Non Small Cell Lung Cancer CT/MRI confirming brain metastases considered inoperable or ineligible for SRS by lung/neuro-oncology Multi-Disciplinary Teams (Tumour Boards) Previous systemic treatment allowed, at least 4 weeks prior to randomisation Subsequent/simultaneous (extra cranial) palliative RT permitted Subsequent systemic treatment permitted at clinicians discretion Adapted to changing landscape Statistical Design Non-inferiority design Aiming to exclude >1 week reduction in QALYs with omission of WBRT 80% power Sample size re-assessed in 2009 following poor recruitment Recalculated independently of results from interim analyses PatientsWBRT QALYHROne- sided Original (2006)10366 weeks1.22.5% Revised (2009)5345 weeks1.255% Challenges Treatment vs No Treatment Patient / Clinician Preferences Interim Data Release Oct 2010 538 Patients: Baseline characteristics 69 UK and 3 Australian centresOSC + WBRT (N=269) OSC Alone (N=269) AgeMedian (range)66 (38 84)67 (45 85) SexMale58% Karnofsky Performance Status 7062% 70 Controlled Primary Site Age