26
Bayesian Clinical Trials: Software and Logistics John D. Cook, PhD Head of Software Development M. D. Anderson Cancer Center Division of Quantitative Sciences JSM 2007

Bayesian clinical trials: software and logistics

Embed Size (px)

Citation preview

Page 1: Bayesian clinical trials: software and logistics

Bayesian Clinical Trials:

Software and Logistics

John D. Cook, PhD

Head of Software Development

M. D. Anderson Cancer Center

Division of Quantitative Sciences

JSM 2007

Page 2: Bayesian clinical trials: software and logistics

Software developers

• Hoang Nguyen

• Kyle Wathen

• Leiko Wooten

• Clift Norris

• Odis Wooten

• John Cook

Page 3: Bayesian clinical trials: software and logistics

Outline

• Software for designing trials

• Software for running trials

Page 4: Bayesian clinical trials: software and logistics

Software for designing trials

Basic utilities Safety monitoring

Randomized trials Dose finding

Page 5: Bayesian clinical trials: software and logistics

Parameter Solver

• Solve from quantiles or mean & variance

• May also enter parameters

• Self-validating

• Graphics options

Page 6: Bayesian clinical trials: software and logistics

Parameter solver

Page 7: Bayesian clinical trials: software and logistics

Inequality Calculator

• Included inside several other applications

• Supports common distributions

• Allows a slippage term

• Customizable plots of density function

• Can specify parameters via

ParameterSolver tool

Page 8: Bayesian clinical trials: software and logistics

Inequality calculator

Page 9: Bayesian clinical trials: software and logistics

Multc Lean

• Original software Multc99, developed by

Hsi-Guang Sun

• Greatly simplified, added Windows UI

• Well-tested

• Added trial duration simulation

Page 10: Bayesian clinical trials: software and logistics

Multc Lean

Page 11: Bayesian clinical trials: software and logistics

TTEDesigner and TTEConduct

• Safety monitoring for single-arm trials withtime-to-event outcomes

• Doesn’t artificially dichotomize data

• Takes advantage of partial information

• Very robust

Page 12: Bayesian clinical trials: software and logistics

TTEDesigner and TTEConduct

Page 13: Bayesian clinical trials: software and logistics

TTE Stopping boundaries

Page 14: Bayesian clinical trials: software and logistics

Stop for futility:

Predictive Probability

• Supports binary and time-to-event outcomes

• What is likely to happen given what has

happened? Predictive probability

• Three decisions: A, B, neither

• If P(neither) is large, stop for futility

• See PP tutorial on download site

Page 15: Bayesian clinical trials: software and logistics

Predictive Probability

Page 16: Bayesian clinical trials: software and logistics

Adaptive Randomization

• Supports binary and TTE outcomes

• Up to 10 arms

• Support for covariates in development

• Separate software for conduct

Page 17: Bayesian clinical trials: software and logistics

Adaptive Randomization

Page 18: Bayesian clinical trials: software and logistics

CRMSimulator

• Emphasis on ease-of-use, not generality

• Contains features commonly used at MDACC, and no more

Page 19: Bayesian clinical trials: software and logistics

CRMSimulator

Page 20: Bayesian clinical trials: software and logistics

EffTox dose-finding

• Minimize toxicity, maximize efficacy

• Investigator specifies trade-off

• Uses twice as much data per patient

• Uses dose values, not just dose order

Page 21: Bayesian clinical trials: software and logistics

EffTox dose-finding

Page 22: Bayesian clinical trials: software and logistics
Page 23: Bayesian clinical trials: software and logistics

Trial conduct software design

• Store data on server, not on desktops

• Maintain audit trail

• Organize by protocol, not statistical method

• Hide statistical detail from nurses

• Expose statistical detail to statisticians

• Label doses with dimensional units, not ordinal numbers

• Design for sporadic use

Page 24: Bayesian clinical trials: software and logistics

Complications

• Nurses can change jobs during a trial

• Patients can withdraw from trial

• Patients can be discovered to be ineligible after being treated

• Trials may be redesigned mid-stream

• Cohorts can be complicated

• Nurses will forget to update data

• Researchers may question software output

Page 25: Bayesian clinical trials: software and logistics

MDACC trial conduct software

• Multc Lean requires no software

• Simple TTE requires no software

• TTE (Weibull with covariates)

• CRM dose-finding

• EffTox dose-finding to be added

• Pocock-Simon

• Adaptive Randomization (including covariates)

Page 26: Bayesian clinical trials: software and logistics

Resources

• http://biostatistics.mdanderson.org/SoftwareDownload/

• Watch RSS feed for news

• http://www.johndcook.com