Report

Phase I Trial Designs Jud Blatchford, PhD Colorado School of Public Health January 30th, 2013 Table of Contents 1. 2. 3. 4. Orientation Introduction Components of a Phase I Trial Phase I Trial Designs A. Rule-Based Designs B. Statistical Designs 5. References Jud Blatchford, PhD Phase I Trial Designs 2 ORIENTATION Jud Blatchford, PhD Phase I Trial Designs 3 Orientation • Features of a Clinical Trial (CT) ◦ Study of human beings ◦ Prospective ◦ Uses an intervention (i.e. changes some aspect of the subjects) ◦ Protects the safety of the subjects ◦ Follows an approved protocol Jud Blatchford, PhD Phase I Trial Designs 4 Orientation • Phases of Clinical Trials ◦ Phase I – First time an experimental drug or treatment is tested in humans to examine how well the drug is tolerated ◦ Phase II – Trials designed to examine if the drug or treatment has a biological treatment effect ◦ Phase III – Trials designed to assess the treatment effect on a clinically meaningful endpoint ◦ Phase IV – Post-marketing studies to gain additional information regarding the safety of the drug or treatment Jud Blatchford, PhD Phase I Trial Designs 5 Orientation • Components of Study Design ◦ Rationale – Establishing a legitimate reason for the study ◦ Design – Detailed description of what treatments will be administered, including a timeline of administration ◦ Subjects – Determining the group to be studied and how they will be assigned to treatment groups ◦ Data – Outcome measure(s), obtaining data, and QA ◦ Sample Size Justification – Ensuring the study will be able to answer the scientific question with adequate power ◦ Study Closure – Archiving study data, analysis files Jud Blatchford, PhD Phase I Trial Designs 6 INTRODUCTION Jud Blatchford, PhD Phase I Trial Designs 7 Introduction • Phase I Clinical Trials ◦ An experimental drug, treatment, chemotherapeutic agent, cytotoxic agent, is studied—hereafter referred to as “drug” ◦ Primary Goal: Safety Investigate whether the new drug or combination of drugs can be administered safely to subjects Investigate optimal dosing and administration of drug ◦ Secondary Goal: Efficacy Offer a treatment option to subjects who have failed other treatment regimens Jud Blatchford, PhD Phase I Trial Designs 8 Introduction • Underlying Assumptions ◦ The drug kills both cancer cells and other cells ◦ The effect is dose-dependent, therefore: 1. The efficacy of the drug increases with the dose 2. The toxicity of the drug increases with the dose ◦ Logically, it would be optimal to give the subjects the highest dose of a drug that can be administered without unacceptable toxicity ◦ Fundamental Question: What is this dose? Jud Blatchford, PhD Phase I Trial Designs 9 Maximum Tolerated Dose (MTD) • Definition of MTD: ◦ The highest dose without observing an unacceptable rate of toxicity • Aliases: ◦ Recommended Phase 2 Dose (RP2D) ◦ Phase 2 Recommended Dose (P2RD) Jud Blatchford, PhD Phase I Trial Designs 10 COMPONENTS OF A PHASE I TRIAL Jud Blatchford, PhD Phase I Trial Designs 11 Components of a Phase I Trial • Definition of a Dose-Limiting Toxicity (DLT) ◦ Clarify time-frame for experiencing a DLT • Dose Levels ◦ How many dose levels will be tested? ◦ What will the smallest dose be? ◦ What will the starting dose be? • Subjects ◦ How many subjects will be tested? ◦ Will single subjects or cohorts be tested at each dose? • What dose-escalation scheme will be employed? Jud Blatchford, PhD Phase I Trial Designs 12 Definition of a DLT • DLTs are typically defined using the National Cancer Institute’s (NCI) Common Terminology Criteria for Adverse Events (CTCAE). • DLTs are often grade ≥ 3 non-hematological and grade ≥ 4 hematological toxicities, which are definitely, probably, or possibly related to the drug. Jud Blatchford, PhD • CTCAE Grades ◦ ◦ ◦ ◦ ◦ ◦ 0 – No AE 1 – Mild 2 – Moderate 3 – Severe 4 – Life threatening 5 – Death • Degrees of Related Phase I Trial Designs ◦ ◦ ◦ ◦ ◦ Unrelated Unlikely Possibly Probably Definitely 13 Definition of a DLT • The length of observation within which a DLT occurrence is “counted” should be explicitly stated in the protocol • Typical lengths used are the first cycle of chemotherapy (often 3 weeks) • Weight the trade-off between observation time for a DLT and efficiency in enrolling subjects Jud Blatchford, PhD Phase I Trial Designs 14 Choosing the Starting Dose • Goals: ◦ Dose high enough to have chance of efficacy ◦ Dose low enough to avoid a DLT • Use data from animal pre-clinical studies • Scale dose by body surface area (mg/m2) • Studies that aren’t “first-in-human” studies may be informed from previous studies using the same drug Jud Blatchford, PhD Phase I Trial Designs 15 Choosing the Starting Dose • Choices Used: ◦ First find dose that is lethal in 10% of mice (LD10) Standard starting dose was 10% of this dose (MELD10), if no grade 4+ AEs observed in other species (rats, dogs, etc.) ◦ Find the highest dose for which the most sensitive animals investigated had no AEs Starting dose is 1/3 of this level (scaled) ◦ Find the minimal dose for which any toxicity is seen (TDL) Starting dose is 1/3 of the TDL Jud Blatchford, PhD Phase I Trial Designs 16 Choosing the Number of Dose Levels • Testing more dose levels to accurately estimate the MTD creates a more cumbersome trial, and may require more subjects • Common number of levels is 4 to 7 • Observed number has ranged from 3 to 14 Jud Blatchford, PhD Phase I Trial Designs 17 Choosing the Dose Levels • Desire to progress through possible doses in a quick (e.g. exponential) manner • Ethical considerations should guide the dose escalation scheme used • Linear sequence of numbers may be inefficient ◦ 20, 40, 60, 80, 100, 120, 140, … • Famous sequence of increasing numbers: ◦ 1, 1, 2, 3, 5, 8, 13, 21, 34, 55, 89, 144, 233, 377, 610, 987, … Jud Blatchford, PhD Phase I Trial Designs 18 The Fibonacci Sequence Term (n) Value (fn) Ratio (fn / fn-1) 1 1 - 2 1 1.000 3 2 2.000 4 3 1.500 5 5 1.667 6 8 1.600 7 13 1.625 8 21 1.615 9 34 1.619 10 55 1.618 11 89 1.618 Jud Blatchford, PhD Phase I Trial Designs 19 The Golden Ratio Jud Blatchford, PhD Phase I Trial Designs 20 The Golden Ratio Jud Blatchford, PhD Phase I Trial Designs 21 Modified Fibonacci Dose Escalation (MFDE) Ratio (fn / f1) Fib. Seq. 1 - 1.00 1 2 2.00 2.00 1 3 1.67 3.33 2 4 1.50 5.00 3 5 1.40 7.00 5 6 1.33 9.33 8 7 1.33 12.44 13 8 1.33 16.59 21 9 1.33 22.12 34 10 1.33 29.50 55 11 1.33 39.33 89 Jud Blatchford, PhD Phase I Trial Designs Comparison Conservative Ratio (fn / fn-1) Similar Term (n) 22 Ethical Considerations • Approach the MTD from below (under-estimates MTD) ◦ Bracketing the MTD is unbiased and more efficient • Expected efficacy is minimal ◦ Historical response rate is 11%; temp. stable rate is 34% ◦ 40% expect a cure • Subjects suffer significant toxicity ◦ Rate of grade 4 toxicity is 14%; death rate is 0.5% • What subjects are told is very important Jud Blatchford, PhD Phase I Trial Designs 23 PHASE I TRIAL DESIGNS Jud Blatchford, PhD Phase I Trial Designs 24 Rule-Based Designs 1. 2. 3. 4. 5. Traditional Escalation Rule Variations of the Traditional Escalation Rule Best of 5 Rule Up-and-Down Designs 2-Stage Designs Jud Blatchford, PhD Phase I Trial Designs 25 Traditional Escalation Rule (TER) 3 subjects receive dose di 0 DLTs 1 DLT 2 or 3 DLTs Escalate - 3 subjects receive dose di+1 3 more subjects at dose di Stop escalation De-escalate to di-1 0 DLTs (1/6 with DLT) 1—3 DLTs (≥ 2/6 with DLT) Escalate - 3 subjects receive dose di+1 Stop escalation De-escalate to di-1 De-escalate until a level is reached where at least 6 subjects are treated and at most 1 DLT occurs. MTD is the highest dose where at least 6 subjects were treated with at most 1 DLT. Jud Blatchford, PhD Phase I Trial Designs 26 Evaluating the TER Benefits Criticisms • Conservative escalation • Ease of implementation • Many subjects treated at low, ineffective doses • At least 2 subjects treated at level above MTD • The true MTD is underestimated ◦ Rules regarding dose assignment are clear ◦ Statistical models not fit after each subject • Design is robust • Will arrive at reasonable estimate of MTD Jud Blatchford, PhD Phase I Trial Designs 27 Variations to TER • After escalation stops, fill out all lower levels until at least 6 subjects are treated at each level • Treat subjects at a dose level between the level where escalation stopped and the next lower level Jud Blatchford, PhD Phase I Trial Designs 28 Best of 5 Rule 3 subjects receive dose di 0 DLTs 1 or 2 DLTs 3 DLTs Escalate to dose di+1 1 more at dose di Stop escalation 1/4 with DLT 2/4 with DLTs 3/4 with DLTs Escalate to dose di+1 1 more at dose di Stop escalation 2/5 with DLTs 3/5 with DLTs Escalate to dose di+1 Stop escalation MTD is the dose prior to the dose on which escalation stopped. Jud Blatchford, PhD Phase I Trial Designs 29 Up-and-Down Design (UaD) 1 subject receives dose di 0 DLT 1 DLT Escalate to dose di+1 De-escalate to di-1 Perform UaD for a pre-specified number of subjects (j). MTD is the dose that would be assigned to the j+1st subject. Jud Blatchford, PhD Phase I Trial Designs 30 Storer’s C Design (UaD-C) 1 subject receives dose di 0 DLT 1 DLT If 2 consecutive subjects with 0 DLT, escalate to dose di+1; else give dose di De-escalate to di-1 Perform UaD for a pre-specified number of subjects (j). MTD is the dose that would be assigned to the j+1st subject. Jud Blatchford, PhD Phase I Trial Designs 31 Stage 1 Storer’s Two-Stage BC Design (UaD-BC) 1 subject receives dose di 0 DLT 1 DLT Escalate to dose di+1 De-escalate to di-1 Stage 2 1 subject receives dose di-1 0 DLT 1 DLT If 2 consecutive subjects with 0 DLT, escalate to dose di; else give dose di-1 De-escalate to di-2 Perform UaD for a pre-specified number of subjects (j). MTD is the dose that would be assigned to the j+1st subject. Jud Blatchford, PhD Phase I Trial Designs 32 Accelerated Titration Designs Extension by Simon of Storer’s work Design 1: TER Designs 2—4 Stage 1: Single subjects until first DLT or second grade 2 AE Stage 2: TER Design 2: Toxicities observed in first cycle only Design 3: Toxicities may be observed in any cycle Design 4: Same as 3 except escalation factor is 2.0 Jud Blatchford, PhD Phase I Trial Designs 33 Statistical Designs Dose escalation guided by a statistical model of the relationship between dose and toxic response 1. Continual Reassessment Method 2. Modifications to the CRM 3. 2-Stage CRM Designs 4. TITE-CRM Jud Blatchford, PhD Phase I Trial Designs 34 Continual Reassessment Method (CRM) • First proposed by O’Quigley in 1990 • Subjects are enrolled individually • A dose-toxicity function is assumed ◦ f(d | α) = Pr{DLT | α} • After each patient completes observation, the estimate of α is updated • Strategy is to assign the dose closest to the estimated MTD to each subject Jud Blatchford, PhD Phase I Trial Designs 35 Considerations for the CRM • Number of dose levels • Initial estimates of toxicity rates at each dose level • Target rate of DLT (θ) • Dose-toxicity function • Escalation restrictions • Number of subjects to be treated Jud Blatchford, PhD Phase I Trial Designs 36 Considerations Number of Dose Levels Initial Estimates of Toxicity • Typically between 3 and 8 • In general, as the number of dose levels in the trial increases, the number of subjects needed to accurately estimate the MTD will increase • The estimates should bound the target rate (θ) Jud Blatchford, PhD Phase I Trial Designs ◦ The CRM is not robust when doses tested do not induce toxicity 37 Choosing a Dose-Response Function Logistic Function Logistic Regression Let p = Pr{DLT} ln = 0 + 1() 1− Solving for p we have: = 0+1() 1+ 0+1() One-parameter model: 3+α() = 1 + 3+α() Jud Blatchford, PhD Phase I Trial Designs 38 Choosing a Dose-Response Function Hyperbolic Tangent Function ℎ = Jud Blatchford, PhD − − + − = 2 −1 2 +1 Scaled Tanh Function Pr{DLT}= Phase I Trial Designs 2() −1 1 + 2 2() +2 2 −α 39 Choosing a Dose-Response Function CDF of Normal Distribution 2 P{DLT}= Jud Blatchford, PhD Phase I Trial Designs −(−) 1 22 2 −∞ 40 The Method of CRM • Dose-toxicity function and θ are chosen a-priori • Function is re-fit (i.e. new estimate of α is obtained) after each subject’s observed toxicity ◦ New function is determined from the a-priori function and the vector of observed toxicities ◦ Curve shifts to the right without toxicity; left with toxicity • Next subject is treated at the dose level whose Pr{DLT} is closest to θ Jud Blatchford, PhD Phase I Trial Designs 41 Distributions of DLT Occurrence By Dose Priors for Subject 1 Priors for Subject 26 • High degree of overlap of probabilities between doses • Separation between dose levels becoming clearer Jud Blatchford, PhD Phase I Trial Designs 42 Evaluating the CRM Benefits Criticisms • Few subjects are treated at low, ineffective doses • Subjects are treated at doses believed at the time to be the most efficacious, yet safe • Starting dose is too high • Dose escalation is too aggressive • Trial length is too long Jud Blatchford, PhD Phase I Trial Designs 43 Modified CRM • Start at the lowest dose level under consideration • Enroll two or three subjects at each cohort • Constrain dose escalation to increase by at most one dose level Jud Blatchford, PhD Phase I Trial Designs 44 “Practical” CRM • Proposed by Piantadosi • Based on pre-clinical toxicity data: ◦ Choose dose that would produce low (10%) rate of DLT ◦ Choose dose that would produce high (90%) rate of DLT ◦ Estimate dose/toxicity curve that fits these 2 points • Use the dose/toxicity curve to find dose for θ • Treat three subjects at this level, then re-estimate the dose-toxicity curve, dose for θ, and tx 3 more • Repeat until target dose changes by < 10% Jud Blatchford, PhD Phase I Trial Designs 45 2-Stage CRM Designs • Stage 1: TER ◦ “2 + 2” is a more common first stage than “3 + 3” ◦ Continue until first toxicity is observed • Stage 2: CRM ◦ After first toxicity, fit the dose-response curve using the toxicity data accrued thus far ◦ Choose dose for next cohort of 2 as dose with estimated rate of DLT closest to θ Jud Blatchford, PhD Phase I Trial Designs 46 Time-to-Event CRM (TITE-CRM) • Builds on the CRM described thus far • Uses information from subjects accrued, even if they haven’t finished observation period ◦ Subjects with DLT are given full weight ◦ Subjects without DLT are given weight t/T. • Allows subjects to be enrolled without waiting for prior cohorts to finish ◦ Benefits studies with delayed toxicity (e.g. radiation studies) Jud Blatchford, PhD Phase I Trial Designs 47 Example of a TITE-CRM Trial • Subject accrual is instantaneous • The majority of doses administered are near MTD Jud Blatchford, PhD Phase I Trial Designs 48 Additional TITE-CRM Considerations • Choice of weight function ◦ Uniform toxicities may use a linear function ◦ Expecting late toxicities may use a convex function ◦ Expecting early toxicities may use a concave function • Setting a Margin (i.e. upper limit) on toxicity ◦ If θ = 0.20 and Margin = 0.05, dose for next subject will be dose closest to 0.20 and not greater than 0.25 • Determine cumulative time exposure (B) before allowing escalation (e.g. B = 2) Jud Blatchford, PhD Phase I Trial Designs 49 Design Comparisons • Fitting a model to the data will improve the accuracy of the MTD found by rule-based designs • Model-guided designs only perform well if assumptions are met (θ in range of doses tested) • Conflicting results when designs compared • Few comparisons made on “level playing field” • Both rule-based and model-guided designs are in common use, for good reason Jud Blatchford, PhD Phase I Trial Designs 50 Important Future Work • “Individualized” Designs ◦ Development of designs allowing for within-subject dose escalation ◦ Development of designs for targeted agents • Designs for trials expecting minimal toxicity Jud Blatchford, PhD Phase I Trial Designs 51 REFERENCES Jud Blatchford, PhD Phase I Trial Designs 52 References 1989. Storer BE. Design and Analysis of Phase I Clinical Trials. Biometrics, 45, 925—937. 1990. O’Quigley J, Pepe M, and Fisher L. Continual Reassessment Method: A Practical Design for Phase I Clinical Trials in Cancer. Biometrics, 46, 33—48. 1993. Korn EL, and Simon R. Using the Tolerable-Dose Diagram in the Design of Phase I Combination Chemotherapy Trials. Journal of Clinical Oncology, 11 (4), 794— 801. 1993. Mick R, and Ratain MJ. Model-Guided Determination of Maximum Tolerated Dose in Phase I Clinical Trials: Evidence for Increased Precision. Journal of the National Cancer Institute, 85 (3), 217—223. 1994. Faries D. Practical Modifications of the Continual Reassessment Method for Phase I Cancer Clinical Trials. Journal of Biopharmaceutical Statistics, 4 (2), 147—164. 1996. Piantadosi S, and Liu G. Improved Designs for Dose Escalation Studies Using Pharmacokinetic Measurements. Statistics in Medicine, 15, 1605—1618. 1996. Smith TL, Lee JJ, Kantarjian HM, Legha SS, and Raber MN. Design and Results of Phase I Cancer Clinical Trials: Three-Year Experience at M. D. Anderson Cancer Center. Journal of Clinical Oncology, 14 (1), 287—295. 1997. Durham SD, Flournoy N, and Rosenberger WF. A Random Walk Rule for Phase I Clinical Trials. Biometrics, 53, 745—760. Jud Blatchford, PhD Phase I Trial Designs 53 References (Continued) 1997. Simon R, Freidlin B, Rubinstein L, Arbuck SG, Collins J, and Christian MC. Accelerated Titration Designs for Phase I Clinical Trials in Oncology. Journal of the National Cancer Institute, 89 (15), 1138—1147. 1998. Friedman LM, Furberg CD, and DeMets DL. Fundamentals of Clinical Trials. Springer. 1998. Whitehead J, and Williamson D. Bayesian Decision Procedures Based on Logistic Regression Models for Dose-Finding Studies. Journal of Biopharmaceutical Statistics, 8 (3), 445—467. 1999. Reiner E, Paoletti X, and O’Quigley J. Operating Characteristics of the Standard Phase I Clinical Trial Design. Computational Statistics and Data Analysis, 30, 303—315. 2000. Cheung YK, and Chappell R. Sequential Designs for Phase I Clinical Trials with Late-Onset Toxicities. Biometrics, 56, 1177—1182. 2000. Eisenhauer EA, O’Dwyer PJ, Christian M, and Humphrey JS. Phase I Clinical Trial Design in Cancer Drug Development. Journal of Clinical Oncology, 18 (3), 684—692. 2000. Wang O, and Faries DE. A Two-Stage Dose Selection Strategy in Phase I Trials with Wide Dose Ranges. Journal of Biopharmaceutical Statistics, 10 (3), 319—333. 2001. Lin Y, and Shih WJ. Statistical Properties of the Traditional Algorithm-Based Designs for Phase I Cancer Clinical Trials. Biostatistics, 2 (2), 203—215. Jud Blatchford, PhD Phase I Trial Designs 54 References (Continued) 2001. Ishizuka N, and Ohashi Y. The Continual Reassessment Method and Its Applications: A Bayesian Methodology for Phase I Cancer Clinical Trials. Statistics in Medicine, 20, 2661—2681. 2002. Potter PM. Adaptive Dose Finding for Phase I Clinical Trials of Drugs Used for Chemotherapy of Cancer. Statistics in Medicine, 21, 1805—1823. 2003. Agrawal M, and Emanuel EJ. Ethics of Phase I Oncology Studies: Reexamining the Arguments and Data. Journal of the American Medical Association, 290 (8), 1075—1082. 2003. Ivanova A, Montazer-Haghighi A, Mohanty SG, and Durham SD. Improved Up-and-Down Designs for Phase I Trials. Statistics in Medicine, 22, 69—82. 2004. Stylianou M, and Follmann DA. The Accelerated Biased Coin Up-and-Down Design in Phase I Trials. Journal of Biopharmaceutical Statistics, 14 (1), 249—260. 2005. Horstmann E, McCabe MS, Grochow L, Yamamoto S, Rubinstein L, Budd T, Shoemaker D, Emanuel EJ, and Grady C. Risks and Benefits of Phase I Oncology Trials, 1991 Through 1992. New England Journal of Medicine, 352, 895—904. 2006. Crowley J, and Ankerst DP. Handbook of Statistics in Clinical Oncology. Chapman and Hall/CRC. Jud Blatchford, PhD Phase I Trial Designs 55 References (Continued) 2006. Potter DM. Phase I Studies of Chemotherapeutic Agents in Cancer Patients: A Review of the Designs. Journal of Biopharmaceutical Statistics, 16, 579—604. DOI: 10.1080/10543400600860295. Jud Blatchford, PhD Phase I Trial Designs 56