#### Transcript Statistical Items

CBMTG 0801 MINIMIZATION / STATISTICS Tony Panzarella Princess Margaret Hospital / University of Toronto Outline Minimization Sample size Analysis CBMTG presentation April 7, 2010 A Randomized Trial of Thymoglobulin to Prevent Chronic Graft versus Host Disease in Patients Undergoing Hematopoietic Progenitor Cell Transplantation (HPCT) from Unrelated Donors CBMTG presentation April 7, 2010 Overview - Study Design CBMTG presentation April 7, 2010 “We will seek to achieve a balanced allocation of treatments over prognostic factors for cGVHD.” CBMTG presentation April 7, 2010 Prognostic factors are • Recipient age (<10; 10-30; 31-50; > 50); • Female donor for male recipient; • Donor age (< 30; ≥30) • Blood source of progenitor cells rather than bone marrow • Degree of tissue type matching (full match; one antigen/allele mismatch). CBMTG presentation April 7, 2010 “In addition to balancing for the above variables that are risk factors for cGVHD it will also be necessary to balance those variables that will influence relapse and mortality.” CBMTG presentation April 7, 2010 • Disease • stage (Early vs. Late ) • comorbidity index • type of preparative regimen (myeloablative vs. non-myeloablative) • centre (to allow for differences in clinical practice) CBMTG presentation April 7, 2010 “Given the moderate size of the trial and the numerous strata that would result from our stated aim of achieving balance upfront for the abovementioned factors it would be impractical to use stratified randomization; furthermore, potentially large overall imbalances in treatment allocation could occur, defeating the purpose of stratified randomization. ” CBMTG presentation April 7, 2010 “Instead patients will be allocated to the treatment groups based on a method of dynamic allocation referred to as minimization. As the name implies the method attempts to minimize the differences between treatment groups in terms of these factors .” CBMTG presentation April 7, 2010 “ Unlike stratified randomization, where each strata represents a combination of each of the factors identified, minimization tries to achieve overall balance by trying to achieve balance within each individual factor, not every combination of factors. This alternative approach to balancing factors between treatment groups allows the possibility of balancing over more factors. ” CBMTG presentation April 7, 2010 “Using minimization in our study the first patient will have their treatment randomly allocated (akin to flipping a fair coin). For each subsequent patient we will determine which treatment would lead to better balance between the groups with respect to the baseline prognostic variables identified a priori. ” CBMTG presentation April 7, 2010 “Each patient is then randomized using a weighting in favour of the treatment that would minimize the imbalance. A weighting of 4 to 1 will be used in this study. That is, there will be a probability of 0.8 of receiving the treatment that minimizes the imbalance. ” CBMTG presentation April 7, 2010 “Thus, the study statistician will prepare two randomization lists using a computer random number generator before the study begins: 1) a simple randomization list where both treatments occur equally often; this list will only be used when the two treatments have equal sums for the levels of the baseline prognostic factors; and 2) a list in which the treatment with the smaller total of patient levels occurs with probability 0.8 while the other treatment occurs with probability 0.2.” CBMTG presentation April 7, 2010 “Allocation will occur centrally by the Project Manager through the project management office. This approach ensures that the process of treatment allocation will be concealed from staff at the recipients centre. ” CBMTG presentation April 7, 2010 Minimization - example Suppose 16 patients have been randomized into a trial, and their characteristics are distributed as in the table CBMTG presentation April 7, 2010 Minimization - example Suppose the next patient is from hospital X, aged 38 and has stage II disease A : 4 + 3 + 3 = 10 B : 4 + 5 + 2 = 11 Assign A CBMTG presentation April 7, 2010 Minimization cards for manual allocation Stage I/II Treatment 1 Stage III/IV Treatment 1 Treatment 2 1. 1002 2. 1005 3. 1006 4. 5. 6. 7. 8. 9. 10. 1. 1001 2. 1004 3. 4. 5. 6. 7. 8. 9. 10. Treatment 2 Hospital = Z Treatment 1 Hospital = Y Treatment 1 Hospital = X Age ≥ 50 Treatment 1 Age > 50 Treatment 1 Treatment 2 1. 1002 2. 1006 3. 4. 5. 6. 7. 8. 9. 10. 1. 1004 2. 3. 4. 5. 6. 7. 8. 9. 10. Treatment 2 1. 1002 Treatment 1 2. 1005 3. 1006 1. 1005 4. 2. 1006 5. 3. 6. 4. 7. 5. 8. 6. 9. 7. 10. 8. 9. 10. CBMTG presentation April 7, 2010 Treatment 2 1. 1001 Treatment 2 2. 1004 1. 1001 2. 3. 4. 5. 6. 7. 8. 9. 10. 3. 4. 5. 6. 7. 8. 9. 10. Treatment 2 CIHR reviewer comments Three of 5 reviewers commented explicitly on the minimization approach adopted in 0801. -Two (including statistical reviewer) said it was appropriate - One would have preferred a simpler approach; namely, stratified randomization using the most important prognostic factor, and adjusting for other imbalances in the analysis CBMTG presentation April 7, 2010 Primary Endpoint “The primary endpoint is the freedom from chronic graft versus host disease, as indicated by the withdrawal of all systemic immunosuppressive agents and without resumption up to 12 months after transplantation. ” CBMTG presentation April 7, 2010 “The primary endpoint could be construed as a ‘focused stringent-positive’ variation of ‘chronic GVHD-free survival.’" CBMTG presentation April 7, 2010 Withdrawal of immunosuppressive Tx Death S 0 +1 +2 +3 +4 +5 +6 +7 +8 +9 +10 +11 +12 Months from Transplant CBMTG presentation April 7, 2010 Death (On immunosuppressive Tx) F 0 +1 +2 +3 +4 +5 +6 +7 +8 +9 +10 +11 +12 Months from Transplant CBMTG presentation April 7, 2010 Withdrawal of immunosuppressive Tx prompted by imminent death / persistent malignancy F 0 +1 +2 +3 +4 +5 +6 +7 +8 +9 +10 +11 +12 Months from Transplant CBMTG presentation April 7, 2010 Effect Size “If we assume that the no treatment group (i.e. the control group) has a probability of response of 0.4 (higher than previously described and erring on the side of caution) it would be clinically worthwhile to know if patients administered Thymoglobulin® could increase this response proportion by at least an additional absolute difference of 0.2, to 0.6.” CBMTG presentation April 7, 2010 Sample Size “To be able to detect this difference or more as statistically significant at the type I error (2-sided) level of 0.05 with a power of 0.8, a total sample size of 194 patients would be required (East version 5.1). Assuming, conservatively, that 2% of patients are lost to follow-up in each group a total of 198 patients would be recruited.” CBMTG presentation April 7, 2010 CBMTG presentation April 7, 2010 CIHR reviewer comments Definition of primary endpoint most contentious issue Sample size calculation appropriate but sample size thought to be small CBMTG presentation April 7, 2010 Analysis “The primary endpoint will be compared between treatment groups using logistic regression adjusted for covariates employed in the design. This will yield significance tests of proper size. ” CBMTG presentation April 7, 2010 Secondary endpoints • the incidence of cGVHD (regardless of need for treatment) • the incidence of “extensive” cGVHD, • time to non-relapse mortality • time to all-cause mortality • time to relapse of leukemia CBMTG presentation April 7, 2010 Secondary endpoints continued… • graft rejection or failure (Yes vs. No) • serious infection (Yes vs. No) • CMV activation (Yes vs. No) • quality of life • organ specific grading of chronic graft versus host disease • resumption of immunosuppressive agents after 12 months (Yes vs. No) • doses of immunosuppressive drugs still required at 12 months. CBMTG presentation April 7, 2010 “Comparisons of time to failure endpoints will incorporate Kaplan-Meier probability estimates, log rank testing and, when adjustment of covariates is made, use of the Cox proportional hazards model. If the analysis of a time to failure endpoint involves competing risks then the probability of failure will be estimated using the cumulative incidence function. ” CBMTG presentation April 7, 2010 Binary/categorical secondary endpoints will be compared between treatment groups using logistic regression. Doses of immunosuppressive drugs required at 12 months will be compared between treatment groups using multiple regression. CBMTG presentation April 7, 2010 Quality of life will be measured before transplant and at 6, 12, 18, and 24 months post transplant using a number of instruments. CBMTG presentation April 7, 2010 The results for each of the primary and secondary endpoints will be summarized by a significance test and 95 per cent confidence interval. CBMTG presentation April 7, 2010 “Analysis of secondary endpoints and sub-group analyses will be considered exploratory and hypothesis generating. Given that multiple comparisons increase the probability of a Type I error, adjustment of individual statistical tests using a more strict cut point (p<0.01) will be used to facilitate interpretation.” CBMTG presentation April 7, 2010 All statistical tests quoted will be 2-tailed. Analysis will follow the intention-to-treat principle. Most analyses will be conducted using SAS version 9.2. However, competing risk failure time data will be analyzed using the library cmprsk in R (http://www.r-project.org). CBMTG presentation April 7, 2010 “Sub-group analysis will be conducted as follows: The two treatment groups will be compared among subgroups based on the covariates diagnosis, gender, type of preparative regimen (myeloablative vs. nonmyeloablative), donor age and recipient age. Differences between treatments by subgroup will be tested using interaction effects. Results will be considered hypothesisgenerating. ” CBMTG presentation April 7, 2010 “One interim analysis will be conducted after half of the evaluable patients have been recruited and followed for one year. A group sequential design utilizing the Lan-DeMets spending function with an O’Brien-Fleming stopping boundary will be incorporated. As a result of the interim look the trial could be: 1) stopped early by rejecting the null hypothesis of no treatment difference; 2) stopped early by rejecting the alternative hypothesis that the difference in proportions is at least 0.2; or 3) the study is continued. ” CBMTG presentation April 7, 2010 CBMTG presentation April 7, 2010 CIHR reviewer comments The only analysis issue identified had to do with stopping early for safety rather than efficacy CBMTG presentation April 7, 2010 The foundation of success in clinical trials is teamwork…and I think the study chair has assembled a fine team! CBMTG presentation April 7, 2010