* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Download Slide 1
Survey
Document related concepts
Transcript
Getting the help of SAS in Clinical Trial setting: Monitoring and Simulations Presented By: Mehmet Kocak Phase I Clinical Trials Objective is to find a maximum tolerated dose (MTD) of a new cytotoxic drug – MTD is not really the “maximum” tolerated dose but rather the highest dose that yields manageable side effects. – This dose is called the “target” dose. – Think of MTD as the target dose which is the dose that yields a specified probability of toxicity, e.g. 25%. Continual Reassesment Method(CRM) Bayesian dose-finding method developed by O’Quigley et al (Biometrics, 1990) Statistical model is used to estimate the relationship between dose and probability of toxicity (dose-toxicity) After study opens, the model is fit to the actual data and used to estimate the target dose. Continual Reassesment Method(CRM) Go to Next Dose Start with the first Dose Add More Decision Go to Previous Dose First Dose is too toxic. Continual Reassesment Method(CRM) Obs Dose n r MTD MidDose Decision 1 120 2 0 1216.67 180 GOTONEXT 2 240 2 0 1165.46 360 GOTONEXT 3 480 3 0 1243.45 600 GOTONEXT 4 720 3 2 539.85 960 GOTOPREV 5 480 1 0 560.85 600 ADDMORE 6 480 1 0 578.19 600 ADDMORE 7 480 1 0 592.62 600 ADDMORE 8 480 1 0 604.67 600 GOTONEXT 9 720 1 0 654.01 960 ADDMORE 10 720 1 0 703.34 960 ADDMORE 11 720 1 0 753.23 960 ADDMORE 12 720 1 0 803.51 960 ADDMORE 13 720 1 1 679.89 960 ADDMORE CRM – Statistical Model Logistic function is used frequently to model the dose-toxicity relationship. Don’t know the true relationship between dose and the probability of toxicity. Here are three sample logistic curves: CRM – Statistical Model Relationship Between Dose and Toxicity Based on the Logistic Function 1.0 Curve 2 Probability of Toxicity 0.8 0.6 Curve 1 0.4 Curve 3 0.2 Dose Prob. Tox. Curve 1 Prob. Tox. Curve 2 Prob. Tox. Curve 3 2% 25% 1% 235 6% 95% 3% 472 44% 100% 16% 628 78% 100% 35% 0.0 100 300 500 Dose 100 700 900 CRM – Statistical Model If you don’t know the true relationship between dose and toxicity, how do you estimate the MTD? – Use the actual data from the study to estimate the dose-toxicity curve – Borrow data from other experiences What is the target dose of interest? – Dose that has 25% toxicity What is the proposed dose-toxicity relationship? – Don’t have actual data when the study opens – Need idea about the relationship between dose and toxicity to initiate the model fitting (priors) CRM – Priors Other Studies – Adult study – Study in different population Guess – Quantify clinical intuition about drug behavior at high and low doses What dose would you guess has 90% toxicity? What dose would you guess has 10% toxicity? CRM - Example Investigator wants to open a phase I study with 4 dose levels – 100 mg/m2, 235 mg/m2, 472 mg/m2, and 628 mg/m2 Need priors to initiate model – Prior studies Has there been a previous phase I study using this drug? – Investigator’s clinical intuition about high and low doses What dose would you expect 90% toxicity? What dose would you expect 10% toxicity? – Reduce the lowest dose by half for the low prior and increase the highest dose by half for the high prior 50 for low prior and about 950 (628 + 314) for high prior Modified Continual Reassesment (CRM) Software Programmed by Dr. Steve Piantadosi – Nice interface – Has problems Required data for the model to run: – Dose – N (number of patients treated) – r (number of responses (DLTs)) Probability of toxicity – Weight Depending on the priors, our initial curve changes tremendously. Actual Patient Data Patient Dose Date on Treatment End of Dose Finding Period DLT? 1 100 2/14/03 3/14/03 No 2 100 2/23/03 3/23/03 No First two patients at Dose 100 mg/m2 did not have DLTs. DECISION: ESCALATE TO THE NEXT DOSE LEVEL Sample Patient Data (Cont.) Patient Dose Date on Treatment End of Dose Finding Period DLT? 1 100 2/14/03 3/14/03 No 2 100 2/23/03 3/23/03 No 3 235 3/19/03 4/19/03 No 4 235 4/05/03 5/05/03 No Note: Next two patients treated at Dose 235 mg/m2 did not have DLTs, either. DECISION: ESCALATE TO THE NEXT DOSE LEVEL Sample Patient Data (Cont.) Patient Dose Date on Treatment End of Dose Finding Period DLT? 1 100 2/14/03 3/14/03 No 2 100 2/23/03 3/23/03 No 3 235 3/19/03 4/19/03 No 4 235 4/05/03 5/05/03 No 5 472 4/21/03 5/12/03 Yes Note: Patient-5 had a DLT. We will immediately reestimate the MTD based on the current toxicity information. DECISION: GO BACK TO Dose Level 235. History of CRM Decision Two Step Simulation Remember that we decided to de-escalate from Dose 472 mg/m2 to 235 mg/m2. What can we say about the future decisions? Not the actual doses under investigation! Two Step Simulation with SAS Function= “move” Function= “draw” Thanks to SAS ANNOTATE Facility Simulation Study with SAS: Does CRM really Works? Go to Next Dose Start with the first Dose Add More Decision Go to Previous Dose First Dose is too toxic. Simulation Study in SAS Various Dose-toxicity relationships Iterative Procedure, which is most likely different for each simulation run; You cannot sample the whole data at once; 10,000 simulations in each setting Preserving all necessary components of runs for summarization Huge data sets, complicated algorithm. The Brain of the Simulation in SAS If the current dose is safe %next: Start with the first Dose If you need more data %addmore: %Decision: Processes… If First Dose is too toxic Or you find the MTD, %goto… %exit: If the current dose is not safe, %prev: Simulation Study in SAS %decision: --- DATA STEPS ----- SEVERAL %IF AND % GOTO STATEMENTS--%if &maxcount>=6 and &decision=GOTONEXT and &dose<&nofdl %then %goto next; %else %if &maxcount>=6 and &decision=GOTOPREV and &dose^=1 %then %goto prev; %else %if &decision=GOTONEXT and &dose=&nofdl %then %goto addmore; %else %if &decision=GOTOPREV and &dose=1 %then %goto exit; %else %if &decision=GOTONEXT %then %goto next; %else %if &decision=ADDMORE %then %goto addmore; %else %if &decision=GOTOPREV %then %goto prev; %next: %let dose=%sysevalf(&dose+1); %let ctr=%sysevalf(&ctr+1); %goto decision; %addmore: %let ctr=%sysevalf(&ctr+1); %goto decision; %prev: %let dose=%sysevalf(&dose-1); %let ctr=%sysevalf(&ctr+1); %goto decision; %exit: A paper submitted for publication Modified Continual Reassessment Method versus the Traditional Empirically-Based Design for Phase I Trials in Pediatric Oncology: Experiences of the Pediatric Brain Tumor Consortium Arzu Onar*, Mehmet Kocak, James M. Boyett Biostatistics Department, St. Jude Children’s Research Hospital, 332 North Lauderdale St. Mail Stop 768 Memphis TN 38105 * Corresponding author: Arzu Onar Biostatistics Department, St. Jude Children’s Research Hospital, 332 North Lauderdale St. Mail Stop 768 Memphis TN 38105. Email: [email protected]. Tel: 901 495 5499. Fax: 901 544 8843. References Piantadosi S, Fisher JD, Grossman S. Practical implementation of a modified continual reassessment method. Cancer Chemother Pharmacol, 41:29-436, 1998. Goodman SN, Zahurak ML, Piantadosi, S. Some practical improvements in the continual reassessment method for phase I studies. Statistics In Medicine, 14:1149-1161, 1995.