Steps on the Road to Predictive Medicine

1. Steps on the Road to Predictive Medicine Richard Simon, D.Sc. Chief, Biometric Research Branch National Cancer Institute http://brb.nci.nih.gov

2. BRB Websitebrb.nci.nih.gov • Powerpoint presentations • Reprints & Technical Reports • BRB-ArrayTools software • Web based Sample Size Planning • Clinical Trials using predictive biomarkers • Development of gene expression based predictive classifiers

3. Many cancer treatments benefit only a minority of patients to whom they are administered • Particularly true for molecularly targeted drugs • Being able to predict which patients are likely to benefit would • save patients from unnecessary toxicity, and enhance their chance of receiving a drug that helps them • Help control medical costs • Improve the success rate of clinical drug development

4. “Hypertension is not one single entity, neither is schizophrenia. It is likely that we will find 10 if we are lucky, or 50, if we are not very lucky, different disorders masquerading under the umbrella of hypertension. I don’t see how once we have that knowledge, we are not going to use it to genotype individuals and try to tailor therapies, because if they are that different, then they’re likely fundamentally … different problems…” • George Poste

5. Biomarkers • Prognostic • Measured before treatment to indicate long-term outcome for patients untreated or receiving standard treatment • Predictive • Measured before treatment to select good patient candidates for a particular treatment

6. Prognostic and Predictive Biomarkers in Oncology • Single gene or protein measurement • e.g. HER2 protein staining 2+ or 3+ • HER2 amplification • KRAS mutation • Scalar index or classifier that summarizes contributions of multiple genes/proteins • Empirically determined based on genome-wide correlating gene expression to patient outcome after treatment

7. Prognostic Factors in Oncology • Most prognostic factors are not used because they are not therapeutically relevant • Most prognostic factor studies do not have a clear medical objective • They use a convenience sample of patients for whom tissue is available. • Generally the patients are too heterogeneous to support therapeutically relevant conclusions

8. Prognostic Biomarkers Can be Therapeutically Relevant • <10% of node negative ER+ breast cancer patients require or benefit from the cytotoxic chemotherapy that they receive • OncotypeDx • 21 gene RTPCR assay for FFPE tissue

9. Predictive Biomarkers • In the past often studied as un-focused post-hoc subset analyses of RCTs. • Numerous subsets examined • Same data used to define subsets for analysis and for comparing treatments within subsets • No control of type I error

10. Statisticians have taught physicians not to trust subset analysis unless the overall treatment effect is significant • This was good advice for post-hoc data dredging subset analysis • For many molecularly targeted cancer being developed, the subset analysis will be an essential component of the primary analysis and analysis of the subsets will not be contingent on demonstrating that the overall effect is significant

11. Prospective Co-Development of Drugs and Companion Diagnostics • Develop a completely specified genomic classifier of the patients likely to benefit from a new drug • Establish analytical validity of the classifier • Use the completely specified classifier to design and analyze a new clinical trial to evaluate effectiveness of the new treatment with a pre-defined analysis plan that preserves the overall type-I error of the study.

12. Guiding Principle • The data used to develop the classifier must be distinct from the data used to test hypotheses about treatment effect in subsets determined by the classifier • Developmental studies can be exploratory • Studies on which treatment effectiveness claims are to be based should be definitive studies that test a treatment hypothesis in a patient population completely pre-specified by the classifier

13. New Drug Developmental Strategy I • Restrict entry to the phase III trial based on the binary predictive classifier, i.e. targeted design

14. Develop Predictor of Response to New Drug Using phase II data, develop predictor of response to new drug Patient Predicted Responsive Patient Predicted Non-Responsive Off Study New Drug Control

15. Applicability of Design I • Primarily for settings where the classifier is based on a single gene whose protein product is the target of the drug • eg Herceptin • With substantial biological basis for the classifier, it may be unacceptable ethically to expose classifier negative patients to the new drug

16. Evaluating the Efficiency of Strategy (I) • Simon R and Maitnourim A. Evaluating the efficiency of targeted designs for randomized clinical trials. Clinical Cancer Research 10:6759-63, 2004; Correction and supplement 12:3229, 2006 • Maitnourim A and Simon R. On the efficiency of targeted clinical trials. Statistics in Medicine 24:329-339, 2005

17. Relative efficiency of targeted design depends on • proportion of patients test positive • effectiveness of new drug (compared to control) for test negative patients • When less than half of patients are test positive and the drug has little or no benefit for test negative patients, the targeted design requires dramatically fewer randomized patients • The targeted design may require fewer or more screened patients than the standard design

18. TrastuzumabHerceptin • Metastatic breast cancer • 234 randomized patients per arm • 90% power for 13.5% improvement in 1-year survival over 67% baseline at 2-sided .05 level • If benefit were limited to the 25% test + patients, overall improvement in survival would have been 3.375% • 4025 patients/arm would have been required

19. Web Based Software for Comparing Sample Size Requirements • http://brb.nci.nih.gov

25. DevelopPredictor of Response to New Rx Predicted Responsive To New Rx Predicted Non-responsive to New Rx New RX Control New RX Control Developmental Strategy (II)

26. Developmental Strategy (II) • Do not use the test to restrict eligibility, but to structure a prospective analysis plan • Having a prospective analysis plan is essential • “Stratifying” (balancing) the randomization is useful to ensure that all randomized patients have tissue available but is not a substitute for a prospective analysis plan • The purpose of the study is to evaluate the new treatment overall and for the pre-defined subsets; not to modify or refine the classifier • The purpose is not to demonstrate that repeating the classifier development process on independent data results in the same classifier

27. Analysis Plan A • Compare the new drug to the control for classifier positive patients • If p+>0.05 make no claim of effectiveness • If p+ 0.05 claim effectiveness for the classifier positive patients and • Compare new drug to control for classifier negative patients using 0.05 threshold of significance

28. Analysis Plan B(Limited confidence in test) • Compare the new drug to the control overall for all patients ignoring the classifier. • If poverall 0.03 claim effectiveness for the eligible population as a whole • Otherwise perform a single subset analysis evaluating the new drug in the classifier + patients • If psubset 0.02 claim effectiveness for the classifier + patients.

29. Analysis Plan C • Test for difference (interaction) between treatment effect in test positive patients and treatment effect in test negative patients • If interaction is significant at level int then compare treatments separately for test positive patients and test negative patients • Otherwise, compare treatments overall

30. Sample Size Planning for Analysis Plan C • 88 events in test + patients needed to detect 50% reduction in hazard at 5% two-sided significance level with 90% power • If 25% of patients are positive, when there are 88 events in positive patients there will be about 264 events in negative patients • 264 events provides 90% power for detecting 33% reduction in hazard at 5% two-sided significance level

34. Biomarker Adaptive Threshold Design Wenyu Jiang, Boris Freidlin & Richard Simon JNCI 99:1036-43, 2007

35. Biomarker Adaptive Threshold Design • Randomized phase III trial comparing new treatment E to control C • Survival or DFS endpoint

36. Biomarker Adaptive Threshold Design • Have identified a predictive index B thought to be predictive of patients likely to benefit from E relative to C • Eligibility not restricted by biomarker • No threshold for biomarker determined

37. Analysis Plan • S(b)=log likelihood ratio statistic for treatment versus control comparison in subset of patients with Bb • Compute S(b) for all possible threshold values • Determine T=max{S(b)} • Compute null distribution of T by permuting treatment labels • Permute the labels of which patients are in which treatment group • Re-analyze to determine T for permuted data • Repeat for 10,000 permutations • Compute point and bootstrap confidence interval estimates of the threshold b

38. DNA Microarray Technology • Powerful tool for understanding mechanisms and enabling predictive medicine • Challenges the ability of biomedical scientists to analyze data • Challenges statisticians with new problems for which existing analysis paradigms are often inapplicable • Excessive hype and skepticism

39. Good microarray studies have clear objectives, but not generally gene specific mechanistic hypotheses • Design and analysis methods should be tailored to study objectives

40. Class Prediction • Predict which tumors will respond to a particular treatment • Predict survival or relapse-free survival risk group

41. Class Prediction ≠ Class ComparisonPrediction is not Inference • The criteria for gene selection for class prediction and for class comparison are different • For class comparison false discovery rate is important • For class prediction, predictive accuracy is important • Most statistical methods were not developed for p>>n prediction problems

42. Evaluating a Classifier • “Prediction is difficult, especially the future.” • Neils Bohr • But easier than “understanding”

43. Validating a Predictive Classifier • Goodness of fit is no evidence of prediction accuracy for independent data • Demonstrating statistical significance of prognostic factors is not the same as demonstrating predictive accuracy • Demonstrating stability of selected genes is not demonstrating predictive accuracy of a model for independent data

44. Types of Validation for Prognostic and Predictive Biomarkers • Analytical validation • When there is a gold standard • Sensitivity, specificity • No gold standard • Reproducibility and robustness • Clinical validation • Does the biomarker predict what it’s supposed to predict for independent data • Clinical utility • Does use of the biomarker result in patient benefit • Depends on available treatments and practice standards

45. Internal Clinical Validation of a Predictive Classifier • Split sample validation • Training-set • Used to select features, select model type, fit all parameters including cut-off thresholds and tuning parameters • Test set • Count errors for single completely pre-specified model • Cross-validation • Omit one sample • Build completely specified classifier from scratch in the training set of n-1 samples • Classify the omitted sample • Repeat • Total number of classification errors

46. Cross validation is only valid if the test set is not used in any way in the development of the model. Using the complete set of samples to select genes violates this assumption and invalidates cross-validation • The cross-validated estimate of misclassification error is an estimate of the prediction error for model fit using specified algorithm to full dataset

48. Sample Size Planning References • K Dobbin, R Simon. Sample size determination in microarray experiments for class comparison and prognostic classification. Biostatistics 6:27, 2005 • K Dobbin, R Simon. Sample size planning for developing classifiers using high dimensional DNA microarray data. Biostatistics 8:101, 2007 • K Dobbin, Y Zhao, R Simon. How large a training set is needed to develop a classifier for microarray data? Clinical Cancer Res 14:108, 2008

49. Sample Size Planning for Classifier Development • The expected value (over training sets) of the probability of correct classification PCC(n) should be within  of the maximum achievable PCC()

50. Sample size as a function of effect size (log-base 2 fold-changebetween classes divided by standard deviation). Two different tolerances shown, . Each class is equally represented in the population. 22000 genes on an array.