Supplementary Appendix

Transcription

1 Supplementary Appendix This appendix has been provided by the authors to give readers additional information about their work. Supplement to: Powers SW, Coffey CS, Chamberlin LA, et al. Trial of amitriptyline, topiramate, and placebo for pediatric migraine. N Engl J Med 2017;376: DOI: /NEJMoa

2 Appendix The following supplemental materials related to The Childhood and Adolescent Migraine Prevention CHAMP Study are included in this Appendix. Table of Contents 1. List of Authors and Investigators 2 2. Statistical Analysis Plan 5 3. Simulations Study and Abstract Final Study Report 29 1

3 Authors and Members of the Childhood and Adolescent Migraine Prevention (CHAMP) Study Group Authors: Scott W. Powers, PhD, ABPP, FAHS 1,2*, Christopher S. Coffey, PhD 3 *, Leigh A. Chamberlin, RD, MEd 2, Dixie J. Ecklund, RN, MSN, MBA 3, Elizabeth A. Klingner, MS 3, Jon W. Yankey, MS 3, Leslie L. Korbee, BS 4, Linda L. Porter, PhD 5, Andrew D. Hershey, MD, PhD, FAHS 1,6 * * These authors contributed equally to this work. 1. Department of Pediatrics, University of Cincinnati College of Medicine, Cincinnati, Ohio 2. Division of Behavioral Medicine and Clinical Psychology, Cincinnati Children s Hospital Medical Center, Cincinnati, Ohio 3. Department of Biostatistics, Clinical Trials Statistical and Data Management Center, University of Iowa, Iowa City, Iowa 4. Office for Clinical and Translational Research, Cincinnati Children s Hospital Medical Center, Cincinnati, Ohio 5. National Institute of Neurological Disorders and Stroke, Bethesda, Maryland 6. Division of Neurology, Cincinnati Children s Hospital Medical Center, Cincinnati, Ohio The members of the Childhood and Adolescent Migraine Prevention (CHAMP) Study Group are as follows: Executive Committee and Steering Group: C. Coffey, PhD, L. Chamberlin, RD, MEd, D. Ecklund, RN, MSN, MBA, A. Hershey, MD, PhD, L. Korbee, BS, L. Porter, PhD, S. Powers, PhD, S. Kashikar-Zuck, PhD. Clinical Coordinating Center, Cincinnati Children s Hospital Medical Center: J. Allen, MS, L. Chamberlin, RD, MEd, J. Hembree, AOS, A. Hershey, MD, PhD, M. Hoffman, RPh, M. Kabbouche, MD, S. Kashikar- Zuck, PhD, T. Knilans, MD, L. Korbee, BS, J. Kroner, BA, D. LaGory, RPh, S. LeCates, MSN, CNP, S. McMahan, RN, C. Murphy, MA, CCRP, K. Nause, BS, J. Peugh, PhD, S. Powers, PhD, K. Simmons, BS, S. Sullivan, BS, S. White, DNP, FNP. 2

4 Data Coordinating Center, Clinical Trials Statistical and Data Management Center, University of Iowa: W. Clarke, PhD, C. Coffey, PhD, M. Costigan, RN, BSN, D. Ecklund, RN, MSN, MBA, T. Huff, BS, E. Klingner, MS, K. Lilli, RN, BSN, N. Logsden-Sackett, MS, B. Pearson, BA, R. Peters, BS, H. Riss, BA, J. Yankey, MS. National Institute of Neurological Disease and Stroke, Bethesda, MD: D. Hirtz, MD, L. Porter, PhD, P. Gilbert, ScM. Data Safety Monitoring Board: E. Loder, MD (chair), Brigham and Women s Hospital, Boston; C. Moore, PhD, University of Pittsburgh; J. Scheller, MD, Children s National Medical Center, Washington, DC; S. Silberstein, MD, MPH, Jefferson Medical College, Philadelphia. Independent Medical Monitor, Mayo Clinic: D. Dodick, MD. The CHAMP Research Investigators: M. Kabbouche, MD, J. Kacperski, MD, H. O Brien, MD, Cincinnati Children s Hospital Medical Center, Cincinnati, OH*; J. Aceves, MD, Scott and White Healthcare, Temple, TX*; D. Arun, MD, Saint Louis University, St. Louis, MO*; V. Atluru, MD, Winthrop University Hospital, Mineola, NY*; S. Aurora, MD, Stanford Hospitals and Clinics, Palo Alto, CA*; N. Bennett, MD, Preferred Clinical Research, Pittsburgh, PA*; F. Berenson MD, Atlanta Headache Specialists, Atlanta, GA*; J. Bickel, MD, Children s Mercy Hospital, Kansas City, MO*; R. Bjork, MD, Colorado Springs Neurological Associates, Colorado Springs, CO*; H. Blume, MD, Seattle Children s Hospital, Seattle, WA*; J. Cohen, MD, The Headache Institute at Roosevelt Hospital, New York, NY; D. Chrusciel, MD, Oklahoma Health Sciences, Oklahoma City, OK*; M. DiSabella, DO, Children s National Medical Center, Washington, DC; L. Matthew Frank, MD, Eastern Virginia Medical School, Norfolk, VA*; A. Gelfand, MD, P. Goadsby, MD, University of California-San Francisco Headache Center, San Francisco, CA*; H. Jacobs, MD, University of Maryland School of Medicine, Baltimore, MD*; S. Kedia, MD, Children s Hospital Colorado, Aurora, CO*; L. Kerr, MD, Primary Children s Medical Center, Salt Lake City, UT*; A. LeBel, MD, Boston Children s Hospital, 3

5 Waltham, MA*; D. Lebron, MD, LeBonheur Children s Hospital, Memphis, TN*; S. Linder, MD, David B. Owen, MD, Dallas Pediatric Neurology Associates, Dallas, TX*; K. Mack, MD, Mayo Clinic, Rochester, MN; H. Markley, MD, New England Regional Headache Center, Worcester, MA*; J. McVige, MD, Dent Neurological Institute, Amherst, NY*;H. Murali, MD, Marshfield Clinic, Marshfield, WI*; A. Pakalnis, MD, Nationwide Children s Hospital, Columbus, OH*; E. Pearlman, MD, Children s Hospital at Memorial University Medical Center, Savannah, GA; K. Ridel, MD, Josephson Wallack Munshower Neurology Research, Indianapolis, IN*; D. Rothner, MD, The Children s Hospital, The Cleveland Clinic, Cleveland, OH*; J. Rothrock, MD, J. Lopez, MD, Renown Neuroscience Institute, University of Nevada, Reno School of Medicine, Reno, Nevada*; R. Simmons, MD, Schenectady Neurological Consultants, Schenectady, NY*; M. Sowell, MD, University of Louisville Health Sciences Center, Louisville, KY*; C. Szperka, MD, Children s Hospital of Philadelphia, Philadelphia, PA;M. Victorio, MD, Akron Children s Hospital, Akron, OH*; P. Winner, DO, Premiere Research Institute, West Palm Beach, FL*; M. Yonker, MD, Phoenix Children s Medical Group, Phoenix, AZ*. *denotes sites that enrolled a participant. In Memoriam: We would like to dedicate this paper to two of our colleagues, Donald W. Lewis, MD, Eastern Virginia Medical School, Norfolk, VA and Steven L. Linder, MD, Dallas Pediatric Neurology Associates, Dallas, TX, for their dedication to this study. Both of these physicians are remembered across the pediatric headache community for their advocacy and influence in the care of for children, adolescents, and young adults with headaches. 4

6 The Childhood and Adolescent Migraine Prevention (CHAMP) Study STATISTICAL ANALYSIS PLAN Principal Investigators Andrew D. Hershey, MD, PhD, FAHS Scott W. Powers, PhD, ABPP, FAHS Study Statisticians Christopher S. Coffey, PhD Jon Yankey, MS Elizabeth Klingner, MS Supported By: National Institute of Neurological Disorders and Stroke (NINDS) Grant Numbers: U01NS (Cincinnati Children s) U01NS (U. Iowa) Investigational Drug Application (IND) Holder: Andrew D. Hershey, MD, PhD, FAHS IND # 112,220 VERSION 2.2 September 02,

7 SUMMARY OF CHANGES TO THE CHAMP STATISTICAL ANALYSIS PLAN (SAP) AND KEY STUDY TIMELINES NOVEMBER VERSION 1.0 Initial version finalized prior to presenting first interim efficacy and futility analysis for the primary endpoint to the DSMB (first unblinded data presented to the DSMB). DECEMBER VERSION 2.0 Based on DSMB recommendation for early study closure, SAP was updated to clarify different populations for efficacy, tolerability, and safety analyses. MARCH VERSION 2.1 Minor edits added to clarify that a multiple comparisons adjustment will be implemented for the secondary comparison of the difference in the change of mean headache days over the 24 week treatment period, but not for any of the other secondary comparisons of interest. Updated references for International Headache Society (IHS) guidelines for controlled trials of drugs in migraine from 2 nd edition to 3 rd edition. SEPTEMBER 2016 VERSION 2.2 Minor administrative changes implemented in sections 1.2, 4.1, 9.2, 11.2, 11.3, 11.4, & 11.5 to better clarify all of the objectives for publication purposes. The rationale for these changes is to insure all of the objectives are clearly stated with more consistent ordering across other study documentation. 6

8 PREFACE This Statistical Analysis Plan (SAP) describes the planned analyses for the Childhood and Adolescent Migraine Prevention (CHAMP) study (National Institute of Neurological Disorders and Stroke (NINDS) grant # s U01NS (Cincinnati Children s) and U01NS (U. Iowa). The planned analyses identified in this SAP are intended to support the completion of the Final Study Report (FSR) and will be included in regulatory submissions and/or future manuscripts. All interim analyses will involve only the primary study endpoint, and will be performed once the specified number of subjects have been enrolled and completed the full study period. All final, planned analyses identified in this SAP will be performed only after the last subject has been enrolled and completed the full study period. Once all data have been cleaned and verified, a locked version of the data will be used for reporting the final study results. Key statistics and study results will be made available to the CCC following database lock and prior to completion of the final FSR. 1. STUDY DESIGN This is a phase III, intent to treat, three-arm, multi-center, randomized, double-blind, placebo-controlled safety and efficacy study to simultaneously assess the impact of two medication therapies (amitriptyline and topiramate ) for headache prevention in children and adolescents with a diagnosis of migraine. The global objective of the study is to determine the optimal medication for the prevention of headaches in children and adolescents. A total of 675 children and adolescents between age 8 and 17 diagnosed with migraine, migraine with aura, and/or chronic migraine will be enrolled into the trial at up to 40 sites across the United States. Subjects will be randomized in a 2:2:1 fashion to either amitriptyline (n=270), topiramate (n=270), or placebo (n=135). Stratification will occur based on age at screening (8-12, 13-17) and the baseline number of headaches per month (episodic: 4-14, chronic: 15 or more). Subjects will be involved for approximately 34 weeks in the various phases of the study, with visits, time lines, and interventions as indicated in the flow chart: Screen Randomize Titrate Treat Washout/Final V 3, 4 Visit 1 V 2 V 5, 6, 7 V 8 Phone 1 & 2 4 weeks 0 weeks 8 weeks 12 weeks 4 weeks 6 weeks After informed consent (and assent where appropriate) has been obtained, subjects will maintain a headache diary during a 4 week baseline period. If eligiblity criteria are met, subjects will be randomized to one of the three treatment arms and will enter into an 8-week titration period with study drug. This titration phase will be modifiable based on tolerability and effectiveness. This will be followed by a 16-week constant dose or maintenance phase. Subjects will complete a daily headache diary during the titration and maintenace periods. At the conclusion of maintenance, a 6 week period will ensue to wean off medication (2 weeks) and conduct follow-up safety phone calls (4 weeks). Safety labs will be collected along with repository specimens at three intervals during the study. Physical exams will be conducted at each study visit. A PedMIDAS disability score will be collected every 3 months during the study. Neurological exams will be conducted at the screening and baseline exams, at the end of dose titration, at the end of the maintenance period, and at the study endpoint. Electrocardiograms (ECG) will be collected at screening and after the completion of dose titration. A targeted symptom questionnaire will be conducted at every visit and a brief version of the symptom questionnaire will be conducted by phone between visits. Subjects describing suicidal ideation in response to the symptom questionnaires will be evaluated, and if deemed at risk to themselves or others, will be directed immediately for psychiatric evaluation and management. The Columbia Suicide Severity Rating Scale (CSSRS) instruments will be used to score subject reports of suicidality that were clinically referred for immediate psychiatric evaluation and management. Adverse events and the use of concomitant medications will be recorded at each study visit. Study drug compliance will be ascertained by pill count and subject report on the daily headache diary and batched biosample analysis. Questionnaires will be completed by subjects at screening, the end of titration, and at the study endpoint. 7

9 Based upon the findings of this 3 in 1 clinical trial, our overall goal is to determine, using a specified a priori tiered approach, which of these therapies is superior in the prevention of headaches. The tiers used to make this decision are: First Tier Primary Endpoint: 50% or greater reduction in rate of headache frequency. Second Tier Headache Disability Third Tier - Tolerability If superiority cannot be determined through this tiered approach, then the outcome of the trial would be that both therapies are efficacious and individualized clinical decisions based on subject presentation should guide the first choice in practice. All final recommendations must separately account for the safety data generated by this novel study. Safety concerns could trump the conclusions from any of the three tiers Primary Objective Primary Objective 1: Determine if amitriptyline (AMI) and/or topiramate (TPM) are superior to placebo in reducing headache frequency (defined as the number of days with headache for a given 4 week period) from 4 week baseline period to the last 4 weeks of this trial, in children and adolescents. Primary Objective 2: Conduct a comparative effectiveness study of the two therapies with respect to reducing headache frequency (defined as the number of days with headache for a given 4 week period) from 4 week baseline period to the last 4 weeks of this trial. We will determine if there is a difference in the proportion of subjects with a 50% reduction in headache frequency from the 4 week baseline period to the last 4 weeks of this 24-week trial between AMI and placebo, and between TPM and placebo. These hypotheses were powered to detect a 20 percentage point improvement from placebo, which is considered a clinically meaningful difference by headache experts (Hershey et al, 2009a). For the comparative effectiveness study, we will determine if there is a difference in the proportion of subjects with a 50% reduction in headache frequency from the 4 week baseline period to the last 4 weeks of this 24- week trial between AMI and TPM. This hypothesis was powered to detect a 15 percentage point or greater difference for the comparison of two active interventions Major Secondary Objectives Major Secondary Objective 1: Compare absolute headache disability score (measured by PedMIDAS) from the end of the 4 week baseline period to the last 4 weeks of this trial for: a) AMI vs. placebo b) TPM vs. placebo c) AMI vs. TPM Major Secondary Objective 2: Compare absolute headache days per 28 day period, measured by the change in absolute headache days from the 4 week baseline period to the last 4 weeks of the trial for: a) AMI vs. placebo b) TPM vs. placebo c) AMI vs. TPM Major Secondary Objective 3: Determine if AMI and/or TPM are well tolerated. Major Secondary Objective 4: Determine if AMI and/or TPM differ from placebo on the occurrence of treatment-emergent serious adverse events. 8

10 2. PRIMARY ENDPOINT The primary endpoint is a 50% reduction in headache frequency from the 28 days prior to randomization to the last 28 days of this 24-week trial. Headache frequency is defined as the number of days with headache for a given four week (28 day) period. A headache diary will be dispensed at each visit, and the subject and subject s parent(s) will be instructed on the completion of the daily headache diary. The headache diary will record the daily incidence, frequency, and duration of all of the subject s headaches and migraine symptoms for a period of one month. The headache diary will also record any rescue medications and daily study drug use by the subject. A headache day will be defined as any day during which any headache occurs within a 24 hour period, starting and ending at midnight. For each subject, the primary endpoint will involve a determination of whether a 50% or greater reduction in headache frequency was observed during the last 4 weeks of active treatment as compared with the headache frequency during the 4-week baseline period (Tfelt-Hansen et al, 2012). Subjects who achieve a 50% or greater reduction in headache frequency will be considered to have had a successful primary endpoint. Those subjects whose reduction in headache frequency was less than 50% will be considered to have had a failure with respect to the primary endpoint. The only exception will be that any subjects who must be removed from therapy prior to the completion of the 24-week treatment period or do not provide headache diary data at the final maintenance visit (visit 8) will be considered failures with respect to the primary endpoint. We do not anticipate issues associated with missing days for subjects with completed diaries. The daily headache diary asks a subject to respond in a yes/no manner as to whether a headache occurred on each day and to check off typical migraine symptoms when present. For days that no value is marked, study coordinators are encouraged to contact the subject to determine whether a headache occurred on that day and if the headache had migraine features. If no determination can be made, the data for these days will be marked as unobtainable. For situations where a subject completes less than 28 headache diary days at baseline or Visit 8, we will apply the following rules: 1) If at least 21 headache diary days were completed, the available information will be standardized to the number of headache days for a 28 day period. For example, if a subject only completed 21 headache diary days and reported 3 headache episodes, the outcome for that subject would be based on 4 headache episodes over a 28 day period (since 3/21 = 4/28). 2) If less than 21 headache diary days were completed, the diary information will be considered missing for the subject. Correspondingly, this subject will be handled in the analysis according to the prespecified rules for handing missing data. 3. SECONDARY ENDPOINTS 3.1. Headache Disability One major secondary objective of the CHAMP study is to examine whether the change in absolute headache disability score, as measured by the PedMIDAS, differs between treatment groups over time. The PedMIDAS scale evaluates the impact of headaches in school, home, play, and social activities. It is comprised of six items that pertain to days missed in various activities over the past 90 days. Questions are answered by the youth in consultation with their parents. The total PedMIDAS score (sum of items 1-6) will be used in this trial. The PedMIDAS examination will be dispensed at visit 2, and the subject and subject s parent(s) will be instructed on completion of the PedMIDAS to determine eligibility based on a disability score > 10, including at least mild disruption in daily activities and < 140, indicating extreme disability that may require more comprehensive, multi-component therapy. PedMIDAS data will be reviewed by study staff with the subject and parents. The PedMIDAS will also be used as a study procedure at two additional study timepoints to collect information on headache related disability that has occurred in the last three months. Thus, each subject will have a PedMIDAS score measured at baseline (covering the three months prior to enrollment), the start of the maintenance period (covering the three months since enrollment), and the 24-week visit (the end of the maintenance period, covering three months of enrollment). The main outcome measure for this comparison will be the difference in the 24-week and baseline PedMIDAS scores. 9

11 3.2. Headache Days Another major secondary objective of the study is to determine if the rate of absolute number of headache days, per 28 day period, differs between treatment groups over time. This objective will be assessed longitudinally based on the actual number of headache days from the 28 days prior to randomization to the last 28 days of this 24 week trial Tolerability An additional major secondary objective is to assess tolerability for the two active treatment groups. To assess tolerability, the percentage of subjects who complete the entire 24-week treatment period will be estimated in each of the three groups Safety The safety of study subjects is of the utmost concern to the sponsor and investigators in this study. Reporting of complete and accurate safety information according to acceptable timelines is crucial to the protection of subjects, and is required by the regulatory agencies with oversight of this study. Standard Operating Procedures for adverse event reporting have been developed by the sponsor to ensure compliance with regulatory requirements, including grading of adverse events according to the NCI Common Toxicity Criteria III, and a reporting timeline of 14 days for adverse events and one business day for SAEs. The investigator or site staff will be responsible for detecting, documenting, and reporting events that meet the definition of an Adverse Event (AE) or Serious Adverse Event (SAE). Adverse events will be collected after the ingestion of at least one dose of study drug and until the final phone contact. For subjects who are withdrawn from study drug exposure but continue on the protocol to collect diary data, adverse events will continue to be collected until 30 days after study drug exposure ends Adverse Events An adverse event (AE) is defined as any untoward medical occurrence in a subject administered an investigational pharmaceutical product and which does not necessarily have a causal relationship with this treatment. An AE can therefore be any unfavorable and unintended sign (including an abnormal laboratory finding), symptom, or disease temporally associated with the use of a medicinal (investigational) product, whether or not related to the medicinal (investigational) product. This includes any occurrence that is new in onset or increased in severity, grade, or frequency from the baseline condition, or abnormal results of diagnostic procedures including laboratory test abnormalities which are considered AEs if they: Result in discontinuation from the study Require treatment or any other therapeutic intervention, require further diagnostic evaluation (excluding a repetition of the same procedure to confirm the abnormality), are associated with clinical signs or symptoms judged by the investigator to have a significant clinical impact Decreases and increases in weight will be monitored closely during study visits. Change in weight that exceeds two isobars on the normal CDC growth chart published May 30, 2000 for children ages 2-20 will be considered a grade 2 AE. A detailed report of the reason for weight change will be documented Serious Adverse Events A serious adverse event (SAE) is any untoward medical occurrence that at any dose: Results in death Is life-threatening Requires inpatient hospitalization or prolongation of existing hospitalization Results in persistent or significant disability/incapacity Is a congenital anomaly/birth defect Is an expression of suicidal ideation or action 10

12 Or, in the opinion of the investigators, represents other significant hazards or potentially serious harm to research participants or others. The onset of suicidal thoughts with AMI or TPM usage will be carefully monitored during the study by the use of the symptom questionnaire at frequent intervals by making the subjects and family aware of self-monitoring for the onset of this possible side effect. Suicidal ideation and suicidal intent may also be reported on the CDI questionnaire as noted by a yes response to I think about killing myself but I would not do it and I want to kill myself on item #9. Any report of the expression of suicidal ideation will be considered an SAE. Inpatient or emergency department admission for the treatment of migraine headache will not be considered an SAE for this protocol Unexpected SAE An unexpected SAE is one for which the nature or severity is not consistent with the applicable product information (e.g., Investigator s Brochure) or package insert/summary of product characteristics for an approved product. Expectedness of SAEs will be reviewed by the Medical Safety Monitor (MSM) Related SAE An SAE is considered associated with use of the drug (i.e., related or treatment-emergent) if the attribution is possible, probable, or definite. Relatedness of SAEs will be reviewed by the MSM Additional Safety Endpoints The Behavioral Rating Inventory & Executive Function (BRIEF) is a measure of executive function and is a questionnaire completed by the parent/guardian. It consists of two indexes, Behavioral Regulation and Metacognition, which are then used to calculate an overall global executive composite (GEC) score. The Clinical Depression Inventory (CDI) is a reliable and valid 27-item scale assessing symptoms of depression (Kovacs, 1981; Kovacs, 1992). It yields a total score and subscale scores on negative mood, interpersonal problems, ineffectiveness, anhedonia, and negative self esteem. Separate norms are available for boys and girls ages 7-12 and 13-17, which are then converted to T-scores. The total score and a positive screen cutoff of a T-score 80 will be used. The CDI secondary measure item 9 responses will be reviewed during the visit and the results presented to the site investigator for review. Any responses that endorse suicidality require immediate action and will be acted upon by the site investigator to ensure child safety. 4. ANALYSIS POPULATIONS Due to the unexpected early stopping of the study, different populations will be used for the efficacy and safety analyses that are to be conducted for this study Primary Efficacy, Disability, Absolute Headache Days, & Tolerability Analyses All primary efficacy, as well as secondary comparisons of disability, absolute headache days, and tolerability analyses, will include all subjects who had either an observed endpoint visit (Visit 8) with complete headache diary data, or who had a target date for an expected endpoint visit on or before the target date for completion of the last weaning visit in the original closeout plan (02/04/15). Thus, any subjects who were not expected to have had their endpoint visit prior to early study weaning will not be included in the final analysis. The rationale for this is that is does not make sense to impute the missing data for these subjects since their early study completion is due solely to early study closure. 4.2 Safety Analyses All randomized subjects will be considered in all safety analyses. 5. ENROLLMENT, RANDOMIZATION, AND TREATMENT ASSIGNMENT 5.1. Enrollment of Subjects The primary methodology for identifying and recruiting subjects for this study will be through identification of potential subjects from pediatric headache medicine specialty practices, pediatric neurology practices or adult headache medicine or neurology practices that treat children and adolescents. This will include sites that are 11

13 part of the NINDS Clinical Research Consortium which have contracted to conduct the study. In order to reach our expected enrollment and based on our survey of interested sites, it is anticipated that the research sites would identify at least 4 subjects per site per month, and be able to enroll at least 1 subject every 2 months Randomization Subjects who meet the eligibility criteria and have given their consent will be randomized to one of the 3 treatment arms after the initial 4 week baseline period. Randomization will be stratified by age (8-12, 13-17) and the baseline number of headaches per month (episodic: 4-14, chronic: 15 or more). Because this strategy provides four strata, we have elected not to stratify the randomization at each site. The DCC will generate a randomization table for each of the strata using a permuted block design with random block sizes Treatment Assignment After completion of all eligibility requirements, and determination of eligibility by the investigator, the eligibility Case Report Form (ecrf) will be completed by the study coordinator. The completion of this ecrf will trigger a response from the DCC to randomly assign a treatment group in keeping with the subject s age and headache frequency stratification. The notification of treatment assignment will then be delivered electronically to the central research pharmacy and to the site study staff. The study staff will receive a blinded notification of the bottle numbers to supply the assigned treatment to the subject s family to begin study drug titration. 6. PRELIMINARY TABLULATIONS All subjects who provide informed consent will be accounted for in this study. Regularly generated enrollment reports will describe: Number of subjects consented and randomized by site Ongoing study status of all randomized subjects Reasons for ineligibility Protocol deviations Early study completers The data set will also be summarized by treatment group with respect to important confounders. The distributions of categorical variables will be tabulated by treatment group and overall. Continuous variables will be summarized as mean, median, standard deviation, minimum, and maximum by treatment group and overall. Variables to be collected will include: Gender Race Ethnicity Mother s education Father s education Annual income Age Standardized BMI (standardized to account for differential growth in pediatrics) Baseline headache frequency Baseline PedMIDAS score 7. PRIMARY EFFICACY ANALYSES Primary Hypothesis 1: Amitriptyline, at a target dose of 1 mg/kg/day, will result in an increased percentage of subjects meeting the primary endpoint compared to placebo. Primary Hypothesis 2: Topiramate, at a target dose of 2 mg/kg/day will result in an increased percentage of subjects meeting the primary endpoint compared to placebo. 12

14 Primary Hypothesis 3: Topiramate will result in an increased percentage of subjects meeting the primary endpoint compared to amitriptyline. Each of these primary hypotheses will involve a comparison of headache frequency (percentage of days with any headache) between the last 4 weeks of treatment (weeks 20-24) and the 4-week baseline period. A prior clinical trial of prevention medication in pediatric migraine found that the beneficial effects of TPM over placebo increased throughout the duration of the trial, and reached the optimal outcome during the final 4 weeks (Lewis et al, 2009). Therefore, the choice was made to evaluate the final 4 weeks versus baseline as the primary analysis for the proposed trial. The primary hypotheses will be assessed using a logistic regression model, adjusted for the two stratification variables, to estimate the log-odds of primary endpoint success in each group. The logistic regression model used for these purposes is stated here: where llllllllll(yy ii ) = ββ 0 + ββ 1 XX aaaaaa,ii + ββ 2 XX heeeeeeeeeehee,ii + ββ 3 XX AAAAAA,ii + ββ 4 XX TTTTTT,ii Y i represents the binary variable indicating whether or not the i th subject met the primary outcome requirement of a 50% or greater reduction in headache frequency, X age,i is an indicator variable for the baseline age group for the i th subject (=0 if age 8-12, =1 if age 13-17) X headache,i is an indicator variable for the baseline number of headaches per month for the i th subject (=0 if 4-14, =1 if 15 or more) X AMI,i is an indicator variable for whether the i th subject was randomized to the AMI group X TPM,i is an indicator variable for whether the i th subject was randomized to the TPM group. Correspondingly, the three primary hypotheses of interest can be assessed by performing hypothesis tests of the three pair-wise treatment comparisons of interest: AMI vs. Placebo: H 0: β 3 = 0 vs. H A: β 3 0 TPM vs. Placebo: H 0: β 4 = 0 vs. H A: β 4 0 AMI vs. TPM: H 0: β 3 = β 4 vs. H A: β 3 β 4 Each of the primary hypotheses will be tested using a Bonferroni corrected significance level of (= 0.05/3). Due to randomization, it is unlikely that important covariates will be imbalanced in this study. However, should important imbalances occur, we will control for these additional covariates in the logistic regression model. The odds ratio of primary endpoint success, and Bonferroni adjusted confidence intervals, will then be estimated: AMI relative to placebo: TPM relative to placebo: AMI relative to TPM: OR = exp(β 3), Adjusted CI: (exp( (β 3) ± 2.39 * SE(β 3)) OR = exp(β 4), Adjusted CI: (exp( (β 4) ± 2.39 * SE(β 4)) OR = exp(β 3 - β 4), Adjusted CI: (exp( (β 3 - β 4) ± 2.39 * SE(β 3 - β 4)) Results will be summarized in the following tables (number and percent of primary endpoint success are displayed in Table 6.1, odds ratios and adjusted CI s are displayed in Table 6.2): 13

15 Treatment Group N Table 7.1: Number and Percent of Primary Endpoint Success Reduction in Headache Frequency 50% N (%) Completed Visit 8 Reduction in Headache Frequency < 50% N (%) 14 Missing N (%) Early Completer/Early Drug Withdrawal N (%) AMI XXX XXX (XX%) XXX (XX%) XXX (XX%) XXX (XX%) TPM XXX XXX (XX%) XXX (XX%) XXX (XX%) XXX (XX%) Placebo XXX XXX (XX%) XXX (XX%) XXX (XX%) XXX (XX%) Table 7.2: Odds-ratios of Primary Endpoint Success Comparison Odds-ratio p-value (Adjusted 98.3% CI) AMI vs. Placebo xx.x (xx.x, xx.x) x.xxxx TPM vs. Placebo xx.x (xx.x, xx.x) x.xxxx AMI vs. TPM xx.x (xx.x, xx.x) x.xxxx If β 3 is significantly larger than zero, we will conclude AMI to be effective for the treatment of headaches relative to placebo. If β 3 is significantly less than zero, we will conclude that AMI is less effective than placebo in the treatment of headaches. If we fail to reject the first hypothesis then we will conclude that there is no difference in the percent of primary endpoint success between the AMI and placebo groups. Likewise, if we reject the second hypothesis we will conclude TPM to be effective for the treatment of headaches relative to placebo if β 4 is significantly larger than zero. If β 4 is significantly less than zero we will conclude that TPM is less effective than placebo in the treatment of headaches. If we fail to reject the second hypothesis then we will conclude that there is no difference in the percent of primary endpoint success between the TPM and placebo groups. In the comparison of AMI and TPM, if we reject the third hypothesis, we will conclude that AMI is more effective than TPM in the treatment of headaches if β 3 - β 4 is significantly larger than zero. We will conclude TPM to be more effective than AMI if β 3 - β 4 is significantly less than zero. If we fail to reject the third hypothesis then we will conclude that there is no significant difference in percent of primary endpoint success between the AMI and TPM groups. This analysis will be performed in accordance with the intention to treat (ITT) principle, that is, any subject who received a random treatment assignment will be included in the primary endpoint analysis. 8. IMPACT OF MISSING DATA The daily headache diary asks a subject to respond in a yes/no manner as to whether a headache occurred on each day and to check off typical migraine symptoms when present. All primary analyses will follow the intent-to-treat (ITT) paradigm. As such, it will be critically important to minimize the occurrence of missing data. Obviously, the optimal strategy for dealing with missing data is to make every effort to obtain complete data during the conduct of the study. Our team will use a variety of methods in order to minimize the percentage of missing data in this trial. Nevertheless, there is likely to be a small percentage of missing data. For the primary analysis, we will employ a conservative approach and impute an outcome of failure for any subject that either withdraws early from study drug or does not have observed visit 8 headache diary data. We then propose a series of sensitivity analyses to determine the potential dependence of the results of the primary analysis on missing values. This sensitivity analysis will employ multiple methods: Observed Data: Use only subjects with non-missing visit 8 headache diary data and who did not withdraw early from study drug Tolerability/Imputation: Assume all subjects terminated from the study due to tolerability issues are failures. For other subjects, the outcome will be obtained from the visit 8 headache diary data (even for subjects that were early drug withdrawals, if they remained in the study and completed the visit 8

16 headache diaries). For remaining subjects with missing visit 8 headache diary data, outcomes will be imputed using a variety of methods: o o Last Observation Carried Forward: Most recent visit with a complete 28 day headache diary period will be carried forward and used for the endpoint determination for these subjects Multiple Imputation: Impute endpoint data using a multiple imputation model based on age and headache frequency at baseline, as well as endpoint data computed at each intermediate time-point for all subjects with observed data. Imputed values will be derived using the MCMC method with multiple chains, adequate burn-in iterations, and a non-informative prior distribution. We will use five separate implementations of this approach. The logistic regression model described above will be fit to each of the five imputed data sets resulting in five separate parameter estimates, one for each of the imputed data sets. o o llllllllll(yy ii ) = ββ 0 + ββ 1,iiiiii jj XX aaaaaa,ii + ββ 2,iiiiii jj, XX heeeeeeeeeehee,ii + ββ 3,iiiipp jj XX AAAAAA,ii + ββ 4,iiiiii jj XX TTTTTT,ii where the notation imp j indicates imputation j = 1,,5. The mean of the parameter estimates from the imputed data sets will be used as the estimate for the final analysis. Variances for each of the parameter estimates will be estimated using standard formulas provided by (Little & Rubin, 2002) as a function of within imputation and between imputation variance. Best Case Scenario: Assume all subjects with missing visit 8 headache diary data in either the AMI or TPM group are successes; Assume all subjects in the placebo group with missing visit 8 headache diary data are failures. Worst Case Scenario: Assume all subjects with missing visit 8 headache diary data in either the AMI or TPM group are failures; Assume all subjects in the placebo group with missing data are successes. At the time of the interim and final analyses, results will be reported from both the primary analysis and all sensitivity analyses in order to inform how robust the overall trend observed in the study is to the missing data. For example, if an interim analysis suggests that both treatments are better than placebo based on the primary endpoint, the DSMB might be more comfortable dropping the placebo arm if that finding is supported by the majority of the sensitivity analyses. On the other hand, if the finding is not supported by the sensitivity analyses, the DSMB may choose to keep the placebo arm in the study until the time of the next planned interim assessment. These results will be displayed in the following tables: Table 8.1: Number and Percent of Subjects with Reduction in Headache Frequency 50% Reduction in Headache Frequency 50% Imputation Method AMI TPM Placebo N (%) N (%) N (%) Primary xxx (xx.x%) xxx (xx.x%) xxx (xx.x%) Observed xxx (xx.x%) xxx (xx.x%) xxx (xx.x%) Tolerability, LOCF* xxx (xx.x%) xxx (xx.x%) xxx (xx.x%) Tolerability, Multiple Imputation xxx (xx.x%) xxx (xx.x%) xxx (xx.x%) Tolerability, Best Case xxx (xx.x%) xxx (xx.x%) xxx (xx.x%) Tolerability, Worst Case xxx (xx.x%) xxx (xx.x%) xxx (xx.x%) * Last observation carried forward Table 8.2: Odds-ratios of Primary Endpoint Success Odds-ratios (Adjusted 98.3% CI) Imputation Method AMI vs. Placebo TPM vs. Placebo AMI vs. TPM Primary x.xx (x.xx, x.xx) x.xx (x.xx, x.xx) x.xx (x.xx, x.xx) Observed x.xx (x.xx, x.xx) x.xx (x.xx, x.xx) x.xx (x.xx, x.xx) Tolerability, LOCF* x.xx (x.xx, x.xx) x.xx (x.xx, x.xx) x.xx (x.xx, x.xx) Tolerability, Multiple Imputation x.xx (x.xx, x.xx) x.xx (x.xx, x.xx) x.xx (x.xx, x.xx) Tolerability, Best Case x.xx (x.xx, x.xx) x.xx (x.xx, x.xx) x.xx (x.xx, x.xx) Tolerability, Worst Case x.xx (x.xx, x.xx) x.xx (x.xx, x.xx) x.xx (x.xx, x.xx) 15

17 * Last observation carried forward 9. PRE-SPECIFIED ANALYSES OF SECONDARY OUTCOMES 9.1. Analysis of Change in PedMIDAS Score from Baseline to 24-weeks Major Secondary Hypothesis #1a: Amitriptyline will result in a decrease in absolute headache disability score (measured by PedMIDAS) compared to placebo. Major Secondary Hypothesis #1b: Topiramate will result in a decrease in absolute headache disability score (measured by PedMIDAS) compared to placebo. Major Secondary Hypothesis #1c: Topiramate will result in a decrease in absolute headache disability score (measured by PedMIDAS) compared to amitriptyline. The mean change from baseline in PedMIDAS score over time for the three groups will be assessed using a linear regression model, adjusted for the baseline PedMIDAS score and is stated here: where YY ii = ββ 0 + ββ 1 XX aaaaaa,ii + ββ 2 XX heeeeeeeeeehee,ii + ββ 3 XX AAAAAA,ii + ββ 4 XX TTTTTT,ii + ββ 5 XX bbbbbb,ii + εε ii Y i represents the change in PedMIDAS score from baseline to the end of the study period for the i th subject. X age,i is an indicator variable for the baseline age group for the i th subject (=0 if age 8-12, =1 if age 13-17) X headache,i is an indicator variable for the baseline number of headaches per month for the i th subject (=0 if 4-14, =1 if 15 or more) X AMI,i is an indicator variable for whether the i th subject was randomized to the AMI group X TPM,i is an indicator variable for whether the i th subject was randomized to the TPM group X bpm,i is the baseline PedMIDAS score for the i th subject εε ii is random error for the i th subject Correspondingly, the three hypotheses of interest can be assessed by performing a hypothesis test of the three pair-wise treatment comparisons of interest: These results will be displayed in the following table: AMI vs. Placebo: H 0,1: β 3 = 0 vs. H A: β 3 0 TPM vs. Placebo: H 0,2: β 4 = 0 vs. H A: β 4 0 AMI vs. TPM: H 0,3: β 3 = β 4 vs. H A: β 3 β 4 Table 9.1.1: Change in PedMIDAS Scores from Baseline to 24 Weeks PedMIDAS Score (24 wk baseline) Mean (SD) Min. max. Missing AMI TPM Placebo Model Adjusted Difference (95% CI) xx.x (xx.x) xx xx xx xx.x (xx.x) xx xx xx xx.x (xx.x) xx xx xx 9.2 Analysis of Change in Absolute Headache Days per 28 Day Period AMI vs. PBO: xx.x (xx.x, xx.x) TPM vs. PBO: xx.x (xx.x, xx.x) AMI vs. TPM: xx.x (xx.x, xx.x) Major Secondary Hypothesis #2a: Amitriptyline will result in a decrease in absolute headache days compared to placebo. Major Secondary Hypothesis #2b: Topiramate will result in a decrease in absolute headache days compared to placebo. 16

18 Major Secondary Hypothesis #2c: Topiramate will result in a decrease in absolute headache days compared to amitriptyline. Another major secondary objective of the CHAMP study is to determine if the change in absolute headache days from baseline to week-24 differs between treatment groups. While the a priori primary endpoint for the CHAMP study is 50% reduction in headache days from the baseline 28-day period, the absolute change in headache days from baseline to the final 28-day recording period (at 24 weeks) is an important endpoint to rigorously examine. This absolute change metric is considered by the International Headache Society (Tfelt- Hansen et al, 2012), along with the American Headache Society, to be critical for assessment of efficacy in studies of preventive interventions, it allows for comparison with the few prior pediatric preventive medication trials since absolute headache change was often chosen as a key endpoint, and it has been an endpoint examined by the FDA in many headache trials. For these reasons, the same multiple comparisons adjustment to the p-value used to determine significance of the primary endpoint will also be utilized for this secondary outcome of absolute change in headache days. This objective will be assessed using a linear regression model, adjusted for the baseline headache frequency and is stated here: where YY ii = ββ 0 + ββ 1 XX aaaaaa,ii + ββ 2 XX bbhff,ii + ββ 3 XX AAAAAA,ii + ββ 4 XX TTTTTT,ii + εε ii Y i represents the change in absolute headache days from baseline (i.e., the 28 day period prior to randomization) to the last 28 days of the 24-week follow-up for the i th subject. Positive values indicate a reduction in headache days over the course of treatment. X age,i is an indicator variable for the baseline age group for the i th subject (=0 if age 8-12, =1 if age 13-17) X bhf,i is the baseline number of headache days for the i th subject X AMI,i is an indicator variable for whether the i th subject was randomized to the AMI group X TPM,i is an indicator variable for whether the i th subject was randomized to the TPM group εε ii is random error for the i th subject Correspondingly, the three hypotheses of interest can be assessed by performing a hypothesis test of the three pair-wise treatment comparisons of interest: These results will be displayed in the following table: AMI vs. Placebo: H 0,1: β 3 = 0 vs. H A: β 3 0 TPM vs. Placebo: H 0,2: β 4 = 0 vs. H A: β 4 0 AMI vs. TPM: H 0,3: β 3 = β 4 vs. H A: β 3 β 4 Table 9.2.1: Absolute Change in Headache Days per 28 Day Period from Baseline to 24 Weeks Absolute Change in Headache Days (24 wk baseline) Mean (SD) Min. max. Missing 9.3 Analysis of Tolerability AMI TPM Placebo Model Adjusted Difference (98.3% CI) xx.x (xx.x) xx xx xx xx.x (xx.x) xx xx xx 17 xx.x (xx.x) xx xx xx AMI vs. PBO: xx.x (xx.x, xx.x) TPM vs. PBO: xx.x (xx.x, xx.x) AMI vs. TPM: xx.x (xx.x, xx.x) Major Secondary Hypothesis #3: Amitriptyline and topiramate will be well tolerated. Concerns regarding tolerability will be defined as either having a percentage of subjects who complete the entire 24 week treatment period for the two active treatment arms that is significantly lower than the percentage for placebo subjects, or significantly less than 65%.

19 To first assess tolerability, we will calculate 95% confidence intervals for the proportion of subjects that completed the 24-week treatment for each treatment group. In cases where the upper bound of the confidence interval is less than 65%, the treatment will be assumed to have tolerability issues. We will also compare the percent of subjects in the active treatment groups who complete the entire 24-week treatment period to the percent of subjects in the placebo group who complete treatment. The following hypothesis test will be performed. H 0: P AMI = P TPM = P PBO vs. H A: there is at least one pairwise difference. Here P AMI represents the proportion of subjects in the AMI group who completed the entire 24-week treatment period, P TPM and P PBO represent this proportion in the TPM and placebo groups respectively. These data will be displayed in the following table: Table 9.3.1: Number and Percent of Subjects who Completed 24-week Treatment by Group Treatment Completed 24-week Group Treatment 95% CI Percent Completed 24-week Treatment N (%) AMI xxx (xx.x%) (xx.x xx.x%) TPM xxx (xx.x%) (xx.x xx.x%) Placebo xxx (xx.x%) (xx.x xx.x%) Table 9.3.2: Odds-ratios of 24-week Study Completion Comparison Odds-ratio (Adjusted 95% CI) AMI vs. Placebo TPM vs. Placebo AMI vs. TPM x.xx (x.xx, x.xx) x.xx (x.xx, x.xx) x.xx (x.xx, x.xx) A chi-square test will be used to test the null hypothesis against the alternative. If we fail to reject H 0 then we will conclude that there is no difference in the percent of subjects completing the 24-month treatment between the treatment groups. If we reject H 0, then we will perform the following separate hypothesis tests. H 0: P AMI = P PBO vs. H A: P AMI P PBO H 0: P TPM = P PBO vs. H A: P TPM P PBO If the first hypothesis is rejected, and P AMI is less than P PBO, then we will conclude that AMI was not tolerated as well as placebo. Similarly, if the second hypothesis is rejected and P TPM is less than P PBO, we will conclude that TPM was not tolerated as well as placebo. The use of a chi-square test in this analysis assumes there are no observed cell counts less than five and no expected cell counts less than one. If we find that either one of these conditions are not met, then we will use a Fishers exact test. 9.4 Analysis of Treatment Related SAEs Major Secondary Hypothesis #4: Amitriptyline and topiramate will not differ from placebo on the occurrence of treatment-emergent serious adverse events. We will also compare the occurrence of treatment-related SAE s across the three treatment groups. This will be assessed in two ways. First, the percentage of subjects who experience any treatment-related SAE in each of the three groups will be compared using Fishers exact test. The following set of hypothesis tests will be performed to determine if proportion of subjects with any treatment-related SAE differs across group: H 0: P AMI = P PBO vs. H A: P AMI P PBO H 0: P TPM = P PBO vs. H A: P TPM P PBO H 0: P AMI = P TPM vs. H A: P AMI P TPM 18

20 Here P AMI, P TPM, and P PBO represent the proportion of subjects in the AMI, TPM, and placebo groups, respectively, who had at least one treatment related SAE. If the first hypothesis is rejected, and P AMI is greater than P PBO, we will conclude that AMI was responsible for more treatment-emergent SAE s than placebo. Similarly, if the second hypothesis is rejected and P TPM is greater than P PBO, we will conclude that TPM was responsible for more treatment-emergent SAE s than placebo. If the third hypothesis is rejected, that would imply that there was a difference between the two treatment groups with respect to the risk of treatmentemergent SAE s. The group with the lower proportion would then be determined to be the safer of the two treatment groups. These hypotheses will be repeated for assessing treatment-related SAEs by system organ class and specific type of SAE based on a classification using MedDRA preferred terms. Additionally, the rates of treatment-related SAE s across the three groups will be compared using the following Poisson regression model: where log (YY ii /TT ii ) = ββ 0 + ββ 1 XX aaaaaa,ii + ββ 2 XX heeeeeeeeeehee,ii + ββ 3 XX AAAAAA,ii + ββ 4 XX TTTTTT,ii + εε ii Y i represents the number of treatment related SAEs experienced by the i th subject. T i represents the number of the number of days between the date of randomization and the date of last follow-up for the i th subject. X age,i is an indicator variable for the baseline age group for the i th subject (=0 if age 8-12, =1 if age 13-17) X headache,i is an indicator variable for the baseline number of headaches per month for the i th subject (=0 if 4-14, =1 if 15 or more) X AMI,i is an indicator variable for whether the i th subject was randomized to the AMI group X TPM,i is an indicator variable for whether the i th subject was randomized to the TPM group εε ii is random error for the i th subject To determine if the rate of treatment related SAEs differ across group we will perform the following tests: 9.5 Additional Secondary Analyses H 0: β 3 = 0 vs. H A: β 3 0 H 0: β 4 = 0 vs. H A: β 4 0 H 0: β 3 = β 4 vs. H A: β 3 β 4 A number of additional secondary analyses are planned, but will not be included as part of the FSR. These additional secondary analyses include, but are not limited to: A comparison of the percentage of subjects in each group that achieve a 50% reduction in migraine frequency (defined as a migraine day). Two approaches will be considered to classify a headache as a migraine. o On the headache diary, the participant notes if a headache occurred as well as if they perceive the headache as a migraine. This data will be used to compare perceived migraines. o We will also employ International Classification of Headache Disorders, 3 rd Edition (beta) to review the diary pages for those headaches for which we have headache features and associated symptoms reported. The ICHD-3, beta will be matched to each headache episode reported, and this will yield a calculated (versus patient reported) migraine frequency. A comparison of absolute reduction in migraine frequency across treatment group (calculated for perceived migraine days and ICHD calculated migraine days) Improvement in the PedMIDAS Grade Score (Hershey et al, 2001), classified using a grading scale with values ranging from I (best) to IV (worst). For the purposes of our analysis, improvement will be defined as an improvement of 2 units or more from baseline or a reported score of I at the end of 19

21 treatment. The percentage of subjects with improvement in each of the three treatment groups will be compared Comparison of patient satisfaction scores among subjects within each treatment group Compare improvement in migraine-associated symptoms (nausea, vomiting, photophobia, phonophobia) across the three treatment groups 10. DECISION ALGORITHM FOR FINAL RECOMMENDATION The main objective of this project is to empirically determine the best choice prevention medication for current practice. In order to achieve this objective, we will use an a priori decision making process that incorporates the novel data that will result from this 3 trials in one project. We propose a tiered approach for using the results of the analyses from this trial to pick a winner among the three therapies tested. The first tier is based upon our primary endpoint, 50% or greater reduction rate for headache frequency (days) between the four weeks prior to baseline and the final four weeks of treatment. This endpoint is the most relevant to current practitioners and is recommended as the key variable for migraine prevention medication trials by the International Headache Society (Tfelt-Hansen et al, 2012). In our plan, there are four possible practice recommendations: 1) TPM first choice because it is superior to placebo and AMI on primary outcome 2) AMI first choice because it is superior to placebo and TPM on primary outcome 3) Both therapies first choice because both are superior to placebo but were not different on primary outcome 4) Neither therapy first choice because neither are superior to placebo If scenario (3) occurs, and a recommendation cannot be made based upon the primary endpoint, we would need to go to a tie breaker, and we will utilize tier 2, our secondary endpoint of headache-related disability. If one medication is superior to the other on this variable, it is the winner. If not, we would go to tier 3, tolerability. If one medication was better tolerated than the other, it would be the winner. Problems with tolerability would be defined as having percentage of subjects that do not complete treatment per study protocol greater than 35%, or a rate of subjects not completing the study that is statistically worse than the other therapy. If a winner cannot be identified based upon tier 3 criteria, then the outcome of the trial would be that both therapies are efficacious and it would be individualized clinical decisions based upon subject presentation that would guide first choice in practice. Of course, recommendations will also take into account the safety data generated by this novel study. 11. SAMPLE SIZE CONSIDERATIONS Primary Hypotheses 1, 2, & 3 In order to properly determine the required sample size for the proposed trial, the CCC investigators completed a review of the limited literature pertaining to pediatric headache research. This review revealed four applicable studies. Three of these studies evaluated TPM versus placebo (Winner et al, 2005; Lakshmi et al, 2007; Lewis et al, 2009), and one study examined AMI in a clinical database (Hershey et al, 2000). Based on this review, estimates of the proportion of subjects in each study group that will meet the primary endpoint of a 50% or greater reduction in headache frequency (days) can be made. Table demonstrates the percentage of subjects in existing pediatric headache prevention medication trials who achieve the primary endpoint of 50% or greater reduction in headache frequency. Table Percentage of patients in existing pediatric headache prevention medication trials who achieve the primary endpoint of 50% or greater reduction in headache frequency. Reference PBO Response Rate TPM Response Rate Winner et al, % 54% Lakshmi et al, % 95% Lewis et al, % 83% 20

22 The table shows an average response rate between 45% and 52% for the placebo group. There is generally more variability regarding the percentage of subjects on TPM who achieve the primary endpoint. The average percentage of subjects achieving the primary endpoint is 77% in the TPM group, but the effect ranges from 54% to 95% across the studies. Finally, a database from a prior study conducted at the CCC suggests that 70% of subjects taking AMI will achieve a 50% or greater reduction in headache frequency (Hershey et al, 2000). Of note, the Lakshmi sample and CCC database include representation from children with <15 and 15 headaches per month. For sample size estimation in this trial, it is critically important to keep in mind that this study essentially provides three trials in one. Hence, the chosen sample size must allow adequate power for testing all three primary hypotheses of interest. For the two comparisons with placebo, the results of our NINDS Clinical Research Collaboration survey demonstrated that the majority of responders felt that a minimum absolute increase of 20% in the percentage of subjects achieving a 50% or greater reduction in headache frequency was clinically meaningful and the threshold needed for the results of a placebo-controlled trial to change practice. Hence, the sample size has been chosen to ensure adequate power assuming that 50% of those on placebo versus 70% of those in each of the treatment groups will achieve a 50% or greater reduction in headache frequency during the final 4 week endpoint evaluation of the 24-week treatment period. Clearly, smaller differences are important for the head-to-head comparison and it becomes much more difficult to define a minimum clinically meaningful effect (since one could argue that, assuming equal cost and safety profiles, any difference between two active treatments is clinically meaningful). As shown in Table there is some preliminary evidence that suggests more than 85% of patients in the TPM group may achieve the primary endpoint of 50% reduction in headache frequency. Thus, the study is powered to detect an absolute difference of 15% or greater in the percentage of subjects achieving a 50% reduction in headache frequency for the head-to-head comparison. As stated above, since there are three primary hypotheses, a Bonferroni correction will be applied with a plan to test each hypothesis at the level. Based on the drop-out rate observed in the Lewis et al. study (2009), we assume a 15% dropout rate. Under these assumptions, enrolling a total of 675 subjects (135 placebo, 270 AMI, and 270 TPM) provides: At least 85% power to detect all treatment/placebo comparisons 90% power to detect a head-to-head difference of 70% of those on AMI vs. 85% of those on TPM will achieve a greater than 50% or greater reduction in headache frequency This demonstrates that the study has adequate power to assess each primary hypothesis of interest. However, since the main objective of the study is to provide a recommendation to clinical investigators regarding the best treatment for children with headache, the actual power of the study may be thought of as making a correct and generalizable decision using the proposed decision algorithm. To assess the ability of the trial to meet this goal, a series of simulation studies were conducted. Each simulation consisted of 10,000 replications for each of 64 conditions defined by all possible combinations of plausible values for the true response rates in the three treatment groups: placebo {40%, 45%, 50%, 55%}, AMI {50%, 60%, 70%, 80%}, and TPM {50%, 70%, 85%, 95%}. For each simulated condition, one of four possible decisions was reached: (1) AMI recommended first-line therapy, (2) TPM recommended first-line therapy, (3) both AMI and TPM are recommended as first-line therapy use secondary endpoints to decide between the two, and (4) neither can be recommended as first-line therapy. Table summarizes a few key conditions involved in the simulation. The first row in Table demonstrates that the approach maintains the type I error level at or below the desired level of The last two rows demonstrate the advantages of the overall three-in-one approach for selecting a winner. The last row, corresponding to the specific conditions used to justify the same size for the primary hypotheses, demonstrates that the overall approach provides benefits above and beyond what can be achieved with the separate hypotheses. As the table demonstrates, if this set of conditions holds true, our study would have nearly 100% power to correctly select TPM as the appropriate first-line treatment for children with migraine. The second row demonstrates that the combined approach provides a greater likelihood of correctly selecting the winner when the absolute difference between the favorable response rates for AMI and TPM is only 10%. Under these conditions, our study would have 91% power to correctly select 21

23 TPM as first-line treatment - a substantial improvement over the 44% power for the individual head-to-head comparison. Table : Probability of picking any winner and picking correct winner as a function of treatment response rate. Assumes type I error rate of 0.017, sample size of 135 in placebo group and 270 in AMI and TPM groups. PBO Rate AMI Rate TPM Rate Decision 22 Prob. of Picking Winner Prob. of Picking Correct Winner* AMI Better TPM Better Both None 50% 50% 50% 1% 1% 0% 98% 3% 97% 50% 60% 70% 0% 71% 20% 9% 91% 91% 50% 70% 85% 0% 94% 6% 0% 100% 100% *If a single drug is best, the correct winner is thought to be selected if the final decision is for either that drug or both drugs. The conclusions obtained from the simulations are only strengthened by the fact that this only examines the first tier of the decision algorithm. If the disability and tolerability trends parallel those of the primary endpoint, this should only increase the overall probability of selecting the correct winner at the end of the trial Major Secondary Hypotheses 1a-1c For major secondary hypotheses 1a-1c, it has been shown that the disability assessment (via PedMIDAS) is more sensitive to change than headache frequency (Hershey et al, 2009b; Lewis et al, 2009; Hershey, 2010). For this reason, the sample size chosen above to assure adequate power for the comparison of headache frequency should also provide sufficient power for the comparison of headache-related disability. For the active treatment versus placebo comparisons, our study has 85% power to detect effect sizes of 0.33, corresponding to a small to moderate effect size using the Cohen Classification. For example, Lakshmi et al (2007) showed a statistically significant difference in the reduction in the PedMIDAS score for pediatric subjects taking TPM versus placebo. The PedMIDAS score for subjects taking TPM was reduced from 50.7 ± 32.1 at baseline to 10.4 ± 6.4 at the end of the 12 week treatment period. The PedMIDAS score for subjects on placebo was reduced from 42.7 ± 27.5 at baseline to 23.7 ± 19.1 at the end of the treatment period. Because the population in the proposed study is thought to be similar, we conservatively assume that the baseline score will be approximately 40 for these subjects. We also assume that the PedMIDAS score will be reduced to 20 in the placebo group and to 10 in each of the active treatment groups by the end of the study. Under these assumptions, and with a significance level for each comparison of 0.05, the proposed study has greater than 85% power to detect such differences as long as the standard deviation of the change from baseline is less than or equal to 30. For the head-to-head comparison of AMI versus TPM, our study has 80% power to detect effect sizes of 0.25 on the Cohen scale. Hence, the proposed study has greater than 85% power to detect a difference of 5 in the change from baseline, as long as the standard deviation of the change from baseline is less than or equal to Major Secondary Hypotheses 2a-2c For major secondary hypotheses 2a & 2b, we computed a detectable effect given the proposed study sample size for a comparison of active and placebo treatment groups with respect to change in absolute headache days per 28 day period. One complication to that calculation is the need to specify an assumed standard deviation for the change in absolute headache days over time. For TPM, a recent meta-analysis for episodic migraine suggested that a reduction of 2-3 headache days with a standard deviation of 3 was fairly consistent (El-Chammas et al, 2013). The one published chronic migraine trial reported a larger change and standard deviation (Schwedt, 2014). For AMI, there is much less data. In a recent assessment of AMI plus weekly education sessions, Powers et al. (2013) observed a mean reduction of 6.8 +/- 8.5 headache days (from /- 5.2 at baseline to /- 9.8 at 20 weeks). However, the mean number of baseline headaches in that study is a bit larger than the expected mean number of baseline headaches in the CHAMP study. If so, this would be expected to lead to a lower standard deviation for the change in headache frequency over time in the CHAMP study. Taken together, these data suggest a typical standard deviation of around 3 for episodic migraine and 4-8 for chronic migraine. Thus, we feel that it is reasonable to expect a standard deviation in the range of 6 for the CHAMP study. Correspondingly, we then computed the detectable effect based on the

24 assumed standard deviation of 6, a desired level of 80% target power, and a significance level of (0.05/3 same as primary analysis, adjusting for three comparisons of interest). Based on these assumptions, the study is adequately powered to detect a difference in mean absolute reduction of 2.3 headache days for the treatment/placebo comparisons. Given that we expect 70% of treated subjects enrolled in this study to have a >=50% reduction in headache days, we believe that an assumed mean absolute headache reduction of 6-8 headache days seems reasonable for this group. Since we expect about half of our placebo group to achieve a >=50% reduction in headache days, an assumed mean absolute headache reduction of 3-5 headaches seems reasonable for the placebo group. Taken together, this seems to suggest that a reasonable assumption for the mean difference in the reduction over the course of the study for the treatment versus placebo groups is 2-5 headache days. Thus, with a detectable effect of 2.3 headache days, the study is adequately powered to detect changes in the reduction of absolute headache days across the treatment and placebo groups. For comparing the AMI and TPM groups, the study is adequately powered to detect a difference in mean absolute reduction of 1.9 headache days. Using the same assumptions that we considered for the primary power calculations, a 70% success rate in the AMI group and an 85% success rate in the TPM group, we would expect to see mean absolute headache reductions of 6-8 headache days and 8-10 headache days in the AMI and TPM groups, respectively. This suggests that a range of reduction in the magnitude of 1-3 headache days over the treatment period seems appropriate for the head-to-head comparison. With a detectable effect of 1.9 headache days, the study is adequately powered to detect differences in the middle to upper end of this expected range of differences. The power is much less on the lower end of the expected range of differences 27% for a difference of 1 headache day, and 61% for a difference of 1.5 headache days. Thus, we have reasonable power to detect moderate to large differences in change in absolute headache frequency between the AMI and TPM groups Major Secondary Hypothesis 3 Lewis et al. (2009) reported a 20% rate of placebo subjects not tolerating the medication. As detailed above, tolerability concerns will arise if more than 35% of subjects have to discontinue medication during the study (i.e., less than 65% tolerate the medication). With a sample size of 135 in the placebo group, 270 each in the two active treatment groups, an assumed placebo tolerability rate of 80%, and a desired significance level of 0.05 for each test, the study has 85% power to detect an absolute decrease of 15% or greater in the percentage of subjects who tolerate the treatment. Importantly, this assures that we have adequate power to detect increases if the rate of subjects who do not tolerate either of the active treatments is 35% or greater, which is the level of tolerability that is a priori thought to cause concerns in the decision algorithm Major Secondary Hypothesis 4 Because this is not a safety study, and the medications under study are currently widely used in the field, the trial was not powered to detect small differences in the safety profiles between the treatments and placebo. Nevertheless, the study will have adequate power to detect large differences that would dramatically affect the choice of medication for future subjects. With a sample size of 135 in the placebo group, 270 each in the two active treatment groups, a significance level of 0.05 for each comparison of interest, and assuming that adverse events will occur in 2% to 5% of the subjects in the placebo group, the study has at least 80% power to detect an absolute increase of 10% in the percentage of subjects who experience an adverse event. More importantly, this study will provide important information that will add to the totality of evidence regarding safety. When combined with the specified decision algorithm, this information will help to shape the final recommendations at the conclusion of this study. 12. SAFETY MONITORING Adverse Experience Reporting All AEs experience by participants during the study time frame specified in the protocol must be reported. All AE s, whether uncovered by spontaneous reporting or the structured AE interview, will be reported as quickly as possible by the clinical site. To help with this process, all AEs are collected, analyzed, and monitored using an innovative Online Adverse Event Reporting system. This system allows the site to enter data electronically 23

25 for all adverse events. If a site initiates but does no submit a serious adverse event form within 24 hours of notification, an automatic notifies the DCC staff so that follow-up may occur. At the time each AE/SAE is entered into the system, the study investigator will make a determination regarding the relatedness to therapy and severity. All AE s and SAE s are coded using the MedDRA system in conjunction with a double coding system internal to the DCC Medical Safety Monitor Dr. David Dodick will serve as the independent medical monitor for this trial. Dr. Dodick will work closely with the DCC, and will use the online AE reporting system to review all reported SAEs in near real time and evaluate them to identify the need for timely intervention. For any reported SAEs, an automatic will be sent to Dr. Dodick to prompt a review of the event for determination of whether the event is unanticipated and related to study drug and thus requires expedited reporting to the FDA. If deemed expedited, a MedWatch form is created electronically and submitted to the FDA. With the assistance of the coordinators at the DCC, Dr. Dodick has the option of requesting additional information about any SAE. He will complete a form for each review and this information will be entered into the data entry system. He may become unblinded upon request to the DCC based upon the results of any of his activities. He will also receive blinded (A, B, C) monthly tabulations of all AE s/sae s on a quarterly basis for the purpose of determining if any safety trends exist that may require a more frequent review by the DSMB. Safety will be assessed in two ways both the percentage of subjects who experience any AE and the rate of AE s across the three groups will be compared, by body system, using a Fisher s exact test. The additional questions related to whether the AE/SAE is related to treatment, is unanticipated, or is severe will be used to subset these into a series of additional tables. This quarterly review will identify any disconcerting discrepancy in the frequency of an AE/SAE between the treatment groups, with particular emphasis placed on the four AE s of most concern to practicing clinicians due to the known side effect profiles of these two treatments: (1) depression, (2) suicidal ideation, (3) fatigue, and (4) slow thinking or problems with attention. Should any discrepancy arise, or if the medical monitor feels that an SAE should be shared on an emergency basis with the participating clinical sites, he will contact the DSMB to request a special review. Data obtained using psychometrically-sound measures of depression, attention, and cognitive functioning from the current NINDS Common Data Elements will also be available for each subject in this trial at baseline, end of drug titration (Visit 5 - Week 8), and end of maintenance (Visit 8 - Week 24). To further understand the impact of the study medications on these important outcomes beyond our prospective AE monitoring approach, each quarterly report will also include a comparison of the following across the treatment groups: Mean total score from the Child Depression Inventory (CDI) Percentage of subjects with total CDI score > 80 Frequency of subjects that answer yes to intent and ideation on question #9 from the CDI Behavior Rating Inventory of Executive Function (BRIEF) global executive composite score Summary of ECG readings (at baseline and Visit 5 only) Dr. Dodick will also actively interact with the Data and Safety Monitoring Board (DSMB) and will inform the DSMB of any particular concerns that may require the urgent attention of the group. This will be coordinated by the DCC and balanced by review of the DSMB Data and Safety Monitoring Board An NIH-appointed Data and Safety Monitoring Board (DSMB) is charged with the task of providing regular oversight of the data and safety monitoring issues as detailed in the document:, NINDS Guidelines for Data and Safety Monitoring in Clinical Trials. These experts will periodically review and evaluate the accumulated data for participant safety, adverse events, study conduct, and study progress. The DSMB will make recommendations to NINDS concerning continuation, modification, or termination of the study. The Data Coordinating Center (DCC) will prepare regular reports for the DSMB to review at each meeting. These reports will include overall summaries, as well as AE & SAE summaries by treatment group in a blinded fashion. These reports will include a summary of the following topics: 24

26 Performance Monitoring: Subject recruitment, comparison with targeted recruitment, retention, protocol adherence, and quality of data collection procedures. Treatment Monitoring: Data on treatment integrity and adherence. Safety Monitoring: Data related to the safety of the subjects, including confidentiality, any adverse events or side effects related to the treatment. Efficacy monitoring: Two interim efficacy analyses are to be performed when 225 and 450 of the subjects have completed their primary endpoint. Futility monitoring: Futility will be assessed at the time of each interim analysis. At the time of each DSMB meeting, a DSMB closed session report will include all available data to allow the DSMB to monitor for trends. These unblinded reports will also include a memo from the Safety Monitor to allow the communication of any concerns or findings that may be of interest to the DSMB in their study deliberations. In general, interim safety monitoring will rely on these interim reviews to identify any potential trends of interest. However, due to the increased concern associated with the risk of suicidal ideation, we propose a more stringent guideline to trigger a concern for this variable. If the difference in the incidence in suicidal ideation between either active treatment versus placebo becomes statistically significant at the 0.05 level at any of the interim reviews, the DSMB will determine whether the trial should discontinue the affected treatment arm. 13. INTERIM ANALYSES The interim monitoring plan for this study must take into account the complex nature of the proposed project, i.e., the fact that this is really three separate trials in one. As a result, the proposed interim monitoring plan is based on the following premises: 1) The placebo arm will only be dropped if both active treatments are proven superior to placebo during the course of the study 2) The head-to-head comparison will be stopped early only if both active treatments have been proven superior to placebo during the course of the study With these premises in mind, we describe the interim monitoring plan below Interim Efficacy Analysis Two interim efficacy assessments will occur when 225 and 450 subjects have completed their 24 week visit. For these interim analyses, the Lan-DeMets alpha spending function approach with O Brien-Fleming stopping boundaries will be used. Table 12.1 below illustrates the proposed stopping boundaries under the assumption of three equally spaced analyses (two interim analyses and the planned final analysis). However, the Lan- DeMets method allows for analyses at unequal intervals and does not require pre-specifying the time of any interim analyses. Therefore, the proposed method has the flexibility to adapt should the DSMB request an interim analysis at an alternative time-point or if the DSMB is unavailable at the time of a scheduled interim analysis. Importantly, the placebo arm will be dropped only if both active treatments cross the stopping boundary at one of the pre-specified interim assessments. If only one crosses the boundary at the first look, both will be assessed again at the second look and the placebo arm dropped only if both cross the bound during the second review. Table O Brien-Fleming Stopping Boundaries Assuming Three Equally Spaced Analyses. Efficacy analysis Number of Subjects Completing 24 Week Follow-Up Nominal P-Value to Conclude Efficacy <

27 13.2 Futility Assessment Futility monitoring will be performed at each interim analysis. Due to the complex nature of the study, and the fact that definitive evidence would be required if neither of the active treatments proved superior to placebo, we propose that the trial should only be stopped for futility if both treatments were simultaneously futile (which has an exceptionally low probability). At the time of each interim analysis, we will report the conditional power based on the pre-specified effect of interest (Lan, Simon, & Halperin, 1982). If both conditional power values fall below 20%, we would recommend stopping the trial early for futility. 14. REFERENCES El-Chammas K, Keyes J, Thompson N, Vijayakumar J, Becher D, and Jackson JL (2013). Pharmacologic treatment of pediatric headaches: A meta-analysis. JAMA Pediatrics 167(3): Hershey AD (2010). Current approaches to the diagnosis and management of paediatric migraine. Lancet Neurology 9(2): Hershey A, Powers S, Bennti A, and Degrauw T. (2000). Effectiveness of amitriptyline in the prophylactic management of childhood headaches. Headache(40), Hershey AD, Powers SW, Vockell AL, LeCates S, Kabbouche MA, and Maynard MK (2001). PedMIDAS: Development of a questionnaire to assess disability of migraine in children. Neurology 57(11): Hershey A, Powers S, Winner P, & Kabbouche M (2009a). Pediatric Headaches in Clinical Practice. West Sussex, UK: John Wiley & Sons, Ltd. Hershey AD, Powers SW, Nelson TD, Kabbouche MA, Winner P, Yonker M, Linder SL, Bicknese A, Sowel MK, and McClintock W (2009b). Obesity in the pediatric headache population: A multicenter study. Headache (49): Kovacs M (1981). Rating scales to assess depression in school-age children. Acta Paedopsychiatr 46: Kovacs M (1992). Children s Depression Inventory. North Tonawanda, NY: Multi-Health System. Lakshmi C, Singhi P, Malhi P, and Ray M (2007). Topiramate in the prophylaxis of pediatric migraine: A double-blind placebo controlled trial. Journal of Child Neurology (22): Lan KK, Simon R, and Halperin (1982). Stochastically curtailed tests in long-term clinical trials theory. Communications in Statistics, Little RJ and Rubin DB (2002). Statistical Analysis with Missing Data. Hoboken, New Jersey: John Wiley & Sons, Inc. Lewis D, Winner P, Saper J, Ness S, Polverejan E, Wang S, Kurland CL, Nye J, Yuen E, Eerdekens M, and Ford L (2009). Randomized, double-blind, placebo-controlled study to evaluate the efficacy and safety of topiramate for migraine prevention in pediatric subjects 12 to 17 years of age. Pediatrics 123: Powers SW, Kashikar-Zuck SM, Allen JR, LeCates SL, Slater SK, Zafar M, Kabbouche MA, O Brien HL, Shenk CE, Rausch JR, and Hershey AD (2013). Cognitive behavioral therapy plus amitriptyline for chronic migraine in children and adolescents: A randomized clinical trial. JAMA 310(24): Schwedt TJ (2014). Chronic migraine. BMJ 348: g1416. Tfelt-Hansen P, Pascual J, Ramadan N, Dahlof C, D Amico D, Diener HC, Hansen JM, Lanteri-Minet M, Loder E, McCrory D, Plancade S, Schwedt T on behalf of the International Headache Society Clinical Trials Subcommittee (2012). Guidelines for controlled trials of drugs in migraine: Third edition. A guide for investigators. Cephalagia 32(1): Winner P, Pearlman EM, Linder SL, Jordan DM, Fisher AC, and Hulihan J (2005). Topiramate for migraine prevention in children: A randomized, double-blind, placebo-controlled trial. Headache 45(10):

28 The Childhood and Adolescent Migraine Prevention (CHAMP) Study: Randomized Clinical Trial Results for Amitriptyline, Topiramate, and Placebo Simulation Study A simulation study was initially conducted to assess the benefits of the three-in-one approach utilized in the CHAMP study. This simulation study confirmed that the complexity of the study led to greater power for detecting effects in the head-to-head comparison of amitriptyline versus topiramate, due to the ability to combine information across a number of different comparisons. The original version of the simulations did not account for the early stopping rules, since it was thought to be a low likelihood occurrence. Once the DSMB recommended the study stop early for futility, an expanded set of simulations was performed that took into account both the efficacy and futility stopping rules. These simulations confirmed that the study would only stop for futility with high probability under true assumed situations where there was truly no benefit of either intervention. 27

29 SIMULATION STUDY TO ASSESS AN A PRIORI DECISION ALGORITHM FOR A RANDOMIZED THREE ARMED CLINICAL TRIAL Zahn, Elizabeth, MS 1, Coffey, Christopher S. PhD 1 1 Department of Biostatistics, Clinical Trials Statistical and Data Management Center, University of Iowa, Iowa City, Iowa. Abstract (A07) was presented at the 35 th Annual Meeting of the Society of Clinical Trials, Philadelphia, Pennsylvania, USA, May 18-21, Background: The Childhood and Adolescent Migraine Prevention (CHAMP) trial is a threearmed randomized trial aimed at the prevention of pediatric migraine. The CHAMP trial involves two treatment arms (Amitriptyline and Topiramate), which will each be compared to placebo with respect to the primary endpoint of 50% reduction in migraine frequency. The two treatments will also be compared to each other leading to a three trials in one situation. This three armed clinical trial presents a challenge in determining the best treatment at the end of the trial. Since the objective of the study is to inform clinical practice in a novel patient population, a three-tiered a priori decision algorithm was developed in order to determine the winning treatment arm. The algorithm is based on examining the cumulative results across all three primary comparisons. Methods: To assess the performance of the three trials in one approach, a simulation study was designed to assess the probability of the trial leading to the correct first line treatment using the proposed algorithm over a range of plausible conditions of the true response rates in each of the three treatment groups. Monte Carlo simulations with 10,000 replications were conducted for each of the 64 scenarios considered. For each simulated condition, we tabulated the percentage of times that the trial would come to the correct first choice treatment using the a priori decision algorithm. Results: Simulation results provided assurance in using the decision algorithm at the conclusion of the study to provide a clinical practice recommendation for the first line of treatment in patients with pediatric migraine. Over 90% of the simulated conditions demonstrated empirical evidence for a clear winning treatment. Furthermore, this simulation study confirmed additional benefits in using this three trials in one approach over a conventional two armed study design. 28

30 The Childhood and Adolescent Migraine Prevention (CHAMP) Study FINAL STUDY REPORT September 06, 2016 Study Statisticians Christopher S. Coffey, PhD Jon Yankey, MS Elizabeth Klingner, MS University of Iowa Clinical Trials Statistical & Data Management Center Supported By: National Institute of Neurological Disorders and Stroke (NINDS) Grant Numbers: U01NS (Cincinnati Children s) U01NS (U. Iowa) VERSION

31 SUMMARY OF CHANGES TO THE CHAMP FINAL STUDY REPORT MAY VERSION 1.0 Initial version finalized based on initial data lock in January AUGUST VERSION 1.1 During a review while responding to reviewer comments on the primary study manuscript, it was confirmed that an additional subject was prescribed propranolol during their course of active treatment. Further examination confirmed that this was prescribed as a preventative medication, and thus met the criteria as an excluded medication and protocol deviation. This required a relock of the study database in August Additional information added to respond to reviewer comments (not included in original SAP) was included for completeness, and to ensure accuracy with results reported in primary study manuscript. SEPTEMBER 2016 VERSION 1.2 Minor administrative changes implemented to better clarify all of the objectives for publication purposes. The rationale for these changes is to insure all of the objectives are clearly stated with more consistent ordering across other study documentation. 30

32 This document provides a summary of the final results for the Childhood and Adolescent Migraine Prevention (CHAMP) study [National Institute of Neurological Disorders and Stroke (NINDS) grant # s U01NS (Cincinnati Children s) and U01NS (U. Iowa); funding also provided by the Eunice Kennedy Shriver National Institute of Child Health and Human Development]. An original version of this report reflected the status of the CHAMP database at the time of the final data freeze on 01/06/2016. During a review while responding to reviewer comments on the primary study manuscript, it was confirmed that an additional subject was prescribed propranolol during their course of active treatment. Further examination confirmed that this was prescribed as a preventative medication, and thus met the criteria as an excluded medication and protocol deviation. This required a relock of the CHAMP study database in August Thus, all data summarized in the current report reflect the status of the CHAMP database at the time of the final data lock on 08/05/2016. A final version of the DSMB open and closed report tables are included as appendices to this report. DEFINITIONS Headache Day: During the study, each subject was instructed to complete a headache diary in order to record the daily incidence, frequency, and duration of all of the subject s headaches and migraine symptoms. Each day during the study period was defined as either having a headache (headache day) or not. A headache day was defined as any day during which any headache occurs within a 24 hour period, starting and ending at midnight. Primary Endpoint ( 50% reduction in headache frequency): The primary endpoint is a 50% reduction in headache frequency from the 28 days prior to randomization to the last 28 days of this 24-week trial. Headache frequency is defined as the number of days with headache for a given four week (28 day) period. For each subject, the primary endpoint involves a determination of whether a 50% or greater reduction in headache frequency was observed during the last 4 weeks of active treatment as compared with the headache frequency during the 4-week baseline period. Subjects who achieve a 50% or greater reduction in headache frequency will be considered to have had a successful primary endpoint. Those subjects whose reduction in headache frequency was less than 50% will be considered to have had a failure with respect to the primary endpoint. The only exception will be that any subjects who must be removed from therapy prior to the completion of the 24-week treatment period or do not provide headache diary data at the final maintenance visit (visit 8) will be considered failures with respect to the primary endpoint. For days that no value is marked, and the study coordinators were unable to verify with the subject whether or not a headache occurred, the data were marked as unobtainable. For situations where a subject completed less than 28 headache diary days at baseline or Visit 8, the following rules were applied: 3) If at least 21 headache diary days were completed, the available information was standardized to the number of headache days for a 28 day period. For example, if a subject only completed 21 headache diary days and reported 3 headache episodes, the outcome for that subject would be based on 4 headache episodes over a 28 day period (since 3/21 = 4/28). 4) If less than 21 headache diary days were completed, the diary information was considered missing for that subject and the data for the subject was handled in the analysis according to the pre-specified rules for handing missing data. Headache Disability: Headache disability score was assessed using the PedMIDAS scale, which evaluates the impact of headaches in school, home, play, and social activities. It is comprised of six items that pertain to days missed in various activities over the past 3 months. Each subject had a PedMIDAS score measured at baseline (covering the three months prior to enrollment), the start of the maintenance period (covering the three months since enrollment), and the 24-week visit (the end of the maintenance period - covering the last three months of enrollment). The main outcome measure for the secondary comparison of interest in this report was the difference in the total PedMIDAS score (sum of items 1-6) for the 24-week and baseline PedMIDAS assessments. Tolerability: Tolerability, when specifically used as a secondary endpoint in this trial, was defined as whether or not the subject completed the entire 24-week treatment period. 31

33 ANALYSIS POPULATIONS An NINDS-appointed Data and Safety Monitoring Board (DSMB) met in Scottsdale, AZ on 11/20/2014 to review the accumulated data for the CHAMP study. At that time, the DSMB spoke very highly of the CHAMP study s success with respect to enrollment and study execution. The DSMB meeting included a planned interim analysis for the first 225 participants who had completed the CHAMP study. The futility threshold for the pre-specified comparisons of interest (each active treatment versus placebo) was based on the primary analysis, as described in the CHAMP Statistical Analysis Plan (SAP). Early stopping for futility would be recommended at any interim analysis only if the conditional power for both comparisons was less than 20%. Based on the primary results for the first interim analysis, the conditional power was 16% for the amitriptyline versus placebo comparison and 14% for the topiramate versus placebo comparison. Thus, the pre-specified futility stopping criteria had been met. Conditional power was also calculated for each of the approaches used to impute missing data in the sensitivity analyses (multiple imputation, best/worst case, & last observation carried forward). As might be expected, the conditional power varied across each sensitivity approach and, while each demonstrated low conditional power, the 20% threshold was not met in every case. The DSMB also considered the possibility of a time trend in accrual that might have influenced response rates. However, subsequent evaluation of the possibility of this type of trend found that this was unlikely to have influenced the interim findings. After considering all of the evidence, the NINDS appointed DSMB recommended early study closure at that time. After additional consideration by NINDS staff, the CHAMP study staff was advised to hold with no further enrollment and randomization of participants effective 11/26/2014. Sites were notified of the enrollment hold via on 11/24/2014. A study closure timeline and plan was developed by the CHAMP study team. The plan outlined in detail the steps for sites to implement an early completion protocol. The study closure documents, along with letters informing clinical site staff and study participants of the study s early closure were submitted to NINDS for approval on 12/12/2014. On 12/16/2014, the CHAMP study team received notification that the closure plan was approved by NINDS and the DSMB. The central pharmacy began preparing weaning kits to be shipped to sites for actively enrolled study participants. The CHAMP study team conducted conference calls with site investigators and study coordinators to inform them of the study closure on 12/22/2014, 12/23/2014, and 12/29/2014. Sites were provided with written notification of early closure and a letter to inform actively enrolled families of the early closure. All early closure documents received IRB approval at each site prior to proceeding with early study closure. Participants were brought in for their next scheduled study visit to begin safely weaning off study drug. Those participants who were scheduled for a visit 8 (endpoint visit) prior to the last projected weaning visit (02/04/2015) were kept on study drug in order to achieve the endpoint data. Weekly reports were generated by the DCC to track remaining visits with study participants. The last study visit occurred on 04/15/2015. The CCC and DCC began conducting close-out visits (remote and on-site) in January of All 22 sites with at least one active subject required an on-site close-out monitoring visit. The 9 sites with no actively enrolled subjects and/or minimal unmonitored data were closed out remotely. On-site close out monitoring visits were completed by June 2015, and interim results were presented to the clinical study sites in September In October 2015, sites were provided with guidance on the unblinding process for study participants. Due to the unexpected early stopping of the study, different populations were used for the efficacy and safety analyses that were conducted for this study. Primary Efficacy, Disability, Absolute Headache Days, & Tolerability Analyses: All primary efficacy, as well as secondary comparisons of disability, absolute headache days, and tolerability analyses will include all subjects who had either an observed endpoint visit (Visit 8) with complete headache diary data, or who had a target date for an expected endpoint visit on or before the target date for completion of the last weaning visit in the original closeout plan (02/04/15). Thus, any subjects who were not expected to have had their endpoint visit prior to early study weaning were not included in the final analysis. Since the missing data for these subjects is due solely to early study closure, it did not make sense to impute the missing data for these subjects. Safety Analyses: All randomized subjects were considered in all safety analyses. 32

34 DATA SUMMARY A total of 492 initial consent forms were completed, for 488 unique subjects. There were 4 subjects who initially failed screening, but were rescreened. After rescreening, three of these subjects were found to be eligible and were randomized into the study. One subject failed screening a second time. For the purposes of this report, only the latest screening status for each subject is shown. Of the 488 consented subjects, 361 were randomized into the study (144 to amitriptyline, 145 to topiramate, and 72 to placebo), 106 were determined to be ineligible or declined to participate, and 21 were never fully screened due to early study closure. Willingness to participate, having a baseline headache frequency outside of the allowable range, and having a baseline PedMIDAS score outside of the allowable range were the most frequent reasons for ineligibility among consented subjects. Of the 361 randomized subjects, 253 subjects completed the study per protocol, 79 were lost to early completion, and 29 were lost due to early study closure. However, 18 of the 79 subjects lost due to early completion were lost during the post-maintenance phase, which meant that they were still able to provide visit 8 headache diary data that was used to generate the observed primary endpoint of the study (i.e., they did not require imputation for the endpoint). The majority of the subjects lost to early study completion were lost during the titration phase. Excluding the subjects who terminated due to early study closure (n = 29) or early study completers not in the analysis population due to early study closure (n = 4), the early study completion rate in this study (defined as a situation where a subject is removed from treatment due to tolerability concerns, the family is unreachable by phone or mail, the subject is removed from the study for any other reason, AND the subject fails to complete the study protocol through visit 8) is 18.5% (61 out of tolerability, 17 lost to follow-up, 28 other). A total of 6 subjects withdrew early from study drug (4 due to subject or family choice, 2 due to investigator or medical monitor choice), but chose to continue in the study. Of these, four subjects completed the study and two subjects agreed to adhere to remaining visits but did not return through visit 8. Table 1 below shows the status of the 361 randomized subjects, by treatment group, with respect to their final study status. A total of 33 subjects were excluded from the primary analysis due to early study closure (12 amitriptyline, 15 topiramate, 6 placebo). As clarified above, these 33 subjects were very early in the study at the time the DSMB recommended early study closure. Since their final study visit was not expected within a window prior to study closeout, the study team made a decision to exclude them from all but the safety analyses since it did not seem reasonable to impute them as failures for the primary analysis. Thus, of these 361 randomized subjects, a total of 328 subjects are included in the primary efficacy and tolerability analysis set (132 amitriptyline, 130 topiramate, 66 placebo). Treatment Group Table 1. Summary of Final Subject Status by Treatment Group Randomized Completed Visit 8 Per Protocol* Early Completion/Drug Withdrawal Early Study Completers Not in Analysis Population Early Study Closure Amitriptyline (73.6 %) 26 (18.1 %) 1 (0.7 %) 11 (7.6 %) Topiramate (70.3 %) 28 (19.3 %) 2 (1.4 %) 13 (9.0 %) Placebo (81.9 %) 7 (9.7 %) 1 (1.4 %) 5 (6.9 %) * 3 subjects (2 in the amitriptyline group & 1 in the topiramate group) completed visit 8, but failed to provide adequate headache diary data. Sites did a good job with respect to the titration process, and the three treatment groups appear to be very comparable with respect to the percentage of subjects that did not require at least one hold or decrease during the titration process (75% for amitriptyline, 77% for topiramate, 75% for placebo), the percentage of subjects who did not require at least one decrease during the titration process (96% for amitriptyline, 96% for topiramate, 100% for placebo), and the percentage of subjects who reached the maximum target dose (90% amitriptyline, 92% topiramate, 95% placebo). Thus, although about one quarter of randomized subjects required at least one hold during the titration process, the vast majority of these subjects were still able to subsequently continue the titration process and achieve the maximum target dose. Sites also demonstrated excellent performance with respect to data completeness. Nearly 100% of all expected baseline forms were completed in the study database. Similarly, sites had excellent performance 33

35 with respect to entering forms in general with very high percentages of forms entered regardless of how this is examined ( ~90% for follow-up forms, 88% for headache diaries, & 97% for headache records). The three treatment groups were well balanced with respect to key demographic characteristic of interest. Of the 361 randomized subjects, 70% were white, 8% were Hispanic, 68% were female, 79% had a mother with some post high school education, 64% had a father with some post high school education, and 63% had a family with an annual income $35K. The mean age of the randomized subjects was 14.2 years, with a mean standardized BMI of 0.7, a mean baseline headache frequency of 11.4 days per month, and a mean baseline PedMIDAS score of During the course of the study, there were a total of 64 protocol deviations (in 55 subjects). Of these, 18 were due to Impacts Safety Monitoring, 16 were due to Impacts Inclusion/Exclusion Criteria, 14 were due to Alters Study Therapy, 8 were due to Impacts Endpoint Evaluations, and 8 were due to Consent Violations. A higher percentage of subjects in the placebo group had at least one protocol deviation, although a statistically significant difference was observed only when compared to the percent of subjects with at least one protocol deviation in the topiramate group (22% vs. 10%; p = 0.02). PRIMARY EFFICACY ANALYSES Of the 328 randomized subjects included in the analysis, 132 (40.2%) were randomized to amitriptyline, 130 (39.6%) were randomized to topiramate, and 66 (20.1%) were randomized to placebo. For the 328 randomized subjects included in the primary efficacy analyses, a summary of the observed and missing endpoint data are provided in Table 2 below. Of these 328 subjects, outcome data were missing and required imputation for 28 subjects in the amitriptyline group (2 due to missing headache diary data, 7 due to early drug withdrawal because of tolerability concerns, 9 lost to follow-up, and 10 terminated due to other reasons), 29 subjects in the topiramate group (1 due to missing headache diary data, 8 due to early drug withdrawal because of tolerability concerns, 6 lost to follow-up, and 14 terminated early due to other reasons), and 7 subjects in the placebo group (0 due to missing headache diary data, 1 due to early drug withdrawal because of tolerability concerns, 2 lost to follow-up, and 4 terminated due to other reasons). Based on the primary analysis (which assumes all subjects who did not provide endpoint data due to either an observed Visit 8 with missing headache diary data or the subject withdrawing from drug or the study prior to Visit 8 are failures), the observed estimates of success (achieving a greater than or equal to 50% reduction from baseline in the number of headache days per 28 day period) is equal to 52.3%, 55.4%, and 60.6% for the amitriptyline, topiramate, and placebo groups, respectively. Treatment Group Table 2. Number and Percent of Primary Endpoint Success Completed Visit 8 N Reduction in Reduction in Headache Headache Missing Frequency 50% Frequency < 50% N (%) N (%) N (%) Early Completer/Early Drug Withdrawal N (%) Amitriptyline (52.3 %) 35 (26.5 %) 2 (1.5 %) 26 (19.7%) Topiramate (55.4 %) 29 (22.3 %) 1 (0.08 %) 28 (21.5 %) Placebo (60.6 %) 19 (28.8 %) 0 (0.0 %) 7 (10.6 %) 34

36 Figure 1: Percent of Primary Endpoint Success by Treatment Group The results of the primary analyses are displayed in Table 3 below. Table 3. Odds-ratios of Primary Endpoint Success Comparison Odds-ratio p-value (Adjusted 98.3% CI) Amitriptyline vs. Placebo 0.71 (0.34, 1.48) Topiramate vs. Placebo 0.81 (0.39, 1.68) Amitriptyline vs. Topiramate 0.88 (0.49, 1.59) For all three comparisons, adjusted odds ratios include the value of 1.0. Thus, there were no statistically significant differences observed and we cannot conclude that any of the treatments were better with respect to achieving a successful endpoint at 28 weeks. The primary results show no significant differences for any of the comparisons, and suggest that the success rates on both amitriptyline and topiramate were below those observed for the placebo group. As specified in the statistical analysis plan (SAP), we performed a series of sensitivity analyses in order to inform how robust the overall trend observed in the study was to the missing data. These sensitivity analyses employed multiple methods: Observed Data: Use only subjects with non-missing visit 8 headache diary data and who did not withdraw early from study drug Tolerability/Imputation: Assume all subjects terminated from the study due to tolerability issues were failures. For other subjects, the outcome was obtained from the visit 8 headache diary data (even for subjects that were early drug withdrawals, if they remained in the study and completed the visit 8 headache diaries). For remaining subjects with missing visit 8 headache diary data, outcomes were imputed using a variety of methods: o o Last Observation Carried Forward: Most recent visit with a complete 28 day headache diary period was carried forward and used for endpoint determination Multiple Imputation: Imputed endpoint data using a multiple imputation model based on age and headache frequency at baseline, as well as endpoint data computed at each intermediate time-point for all subjects with observed data. Imputed values were derived using the MCMC method with multiple chains, adequate burn-in iterations, and a non-informative prior distribution. We used five separate implementations of this approach. The logistic regression model used for the primary analysis was fit to each of the five imputed data sets resulting in five separate parameter estimates, one for each of the imputed data sets. The mean of the parameter estimates from the imputed data sets was used as the estimate for the final analysis. The variance for each parameter was estimated using standard formulas provided by Little & Rubin (2002), as a function of within imputation and between imputation variance. 35

37 o o Best Case Scenario: Assumed all subjects with missing visit 8 headache diary data in either the amitriptyline or topiramate group were successes; Assumed all subjects in the placebo group with missing visit 8 headache diary data were failures. Worst Case Scenario: Assumed all subjects with missing visit 8 headache diary data in either the amitriptyline or topiramate group were failures; Assumed all subjects in the placebo group with missing data were successes. Table 4 and Figure 2 below summarize the probability of success (reduction in headache frequency 50%) for each of the methods considered. Table 4. Number and Percent of Subjects with Reduction in Headache Frequency 50% Reduction in Headache Frequency 50% Imputation Method Amitriptyline N=132 Topiramate N=130 Placebo N=66 Primary 69 (52 %) 72 (55 %) 40 (61 %) Observed* 69 (66 %) 72 (71 %) 40 (68 %) Tolerability, LOCF** 74 (56 %) 79 (61 %) 42 (64 %) Tolerability, Multiple Imputation 80 (61 %) 81 (62 %) 43 (65 %) Tolerability, Best Case 90 (68 %) 92 (71 %) 40 (61 %) Tolerability, Worst Case 69 (52 %) 72 (55 %) 46 (70 %) * Excludes 28 amitriptyline subjects, 29 topiramate subjects, and 7 placebo subjects ** Last observation carried forward Figure 2: Percent of Primary Endpoint Success by Analysis Type Similarly, Table 5 below summarizes the observed odds ratios and adjusted CI s for each comparison under each of the methods considered. Table 5. Odds-ratios of Primary Endpoint Success Odds Ratios (Adjusted 98.3% CI s) Imputation Method Amitriptyline vs. Placebo Topiramate vs. Placebo Amitriptyline vs. Topiramate Primary 0.71 (0.34, 1.48) 0.81 (0.39, 1.68) 0.88 (0.49, 1.59) Observed* 0.94 (0.41, 2.16) 1.18 (0.50, 2.75) 0.80 (0.39, 1.64) Tolerability, LOCF** 0.73 (0.35, 1.53) 0.89 (0.42, 1.87) 0.82 (0.45, 1.50) Tolerability, Multiple Imputation 0.82 (0.37, 1.80) 0.88 (0.38, 2.05) 0.92 (0.45, 1.90) Tolerability, Best Case 1.39 (0.66, 2.95) 1.57 (0.74, 3.36) 0.89 (0.47, 1.68) Tolerability, Worst Case 0.48 (0.22, 1.02) 0.54 (0.25, 1.16) 0.88 (0.49, 1.59) * Excludes 28 amitriptyline subjects, 29 topiramate subjects, and 7 placebo subjects ** Last observation carried forward The sensitivity analyses suggest that the estimated probabilities of success in the placebo group are relatively stable across the different approaches ranging from 61% to 70%. As might be expected, there is a bit more variation in the observed estimates for the amitriptyline (range from 52% to 68%) and topiramate (range from 36