Trajectory of treatment response in the child and adolescent migraine prevention (CHAMP) study: A randomized clinical trial

Abstract

Objective

Identify preventive medication treatment response trajectories among youth participating in the Childhood and Adolescent Migraine Prevention study.

Methods

Data were evaluated from 328 youth (ages 8–17). Childhood and Adolescent Migraine Prevention study participants completed headache diaries during a 28-day baseline period and a 168-day active treatment period during which youth took amitriptyline, topiramate, or placebo. Daily headache occurrence trajectories were established across baseline and active treatment periods using longitudinal hierarchical linear modeling. We tested potential treatment group differences. We also compared final models to trajectory findings from a clinical trial of cognitive behavioral therapy plus amitriptyline for youth with chronic migraine to test for reproducibility.

Results

Daily headache occurrence showed stability across baseline. Active treatment models revealed decreases in headache frequency that were most notable early in the trial period. Baseline and active treatment models did not differ by treatment group and replicated trajectory cognitive behavioral therapy plus amitriptyline trial findings.

Conclusions

Replicating headache frequency trajectories across clinical trials provides strong evidence that youth can improve quickly. Given no effect for medication, we need to better understand what drives this clinically meaningful improvement. Results also suggest an expected trajectory of treatment response for use in designing and determining endpoints for future clinical trials.

Trial Registration. ClinicalTrials.gov Identifier: NCT01581281

Introduction

Consensus-based practice guidelines for preventive migraine treatment encourage providers to start patients on a low medication dose, slowly increase dose until clinical benefits are achieved, and then allow for an adequate medication trial (i.e., a minimum of 2 months, up to 6–12 months) to assess full drug benefit (1–3). Research suggests that some guidelines do not accurately reflect the course of preventive treatment response in pediatric populations.

In the Childhood and Adolescent Migraine Prevention (CHAMP) study – a randomized clinical trial of amitriptyline, topiramate, and placebo – commonly used preventive medications worked similarly to a placebo pill in producing clinically significant reductions (≥50%) in headache days among youth with migraine (4). A separate randomized clinical trial revealed that the combination of cognitive behavioral therapy (CBT) and amitriptyline produced significant and sustained changes in headache days among youth with chronic migraine (15 or more headache days in a 28 day period) that began in the first month of treatment and were more notable earlier in the trial period. CBT plus amitriptyline was superior to an amitriptyline and education treatment group in that trial (5,6).

The current study examined headache frequency trajectories among youth in the CHAMP study, and tested whether this independently developed trajectory model replicated best-fitting trajectory models in the CBT plus amitriptyline trial (6). Replicating trajectory findings across clinical trials will provide more conclusive evidence to inform provider expectations regarding rates of change for preventive migraine treatment response, and will inform novel approaches to establishing primary endpoints in future trials. Given the results of the CHAMP study, we hypothesized that there would be no between-group (amitriptyline, topiramate, and placebo) trajectory differences.

Methods

Study design and participants

The current project included data from 328 pediatric patients with migraine (8–17 years old) who were included in primary efficacy analyses for the CHAMP study, a multi-site, phase III clinical trial funded by the National Institutes of Health (NIH; Grant numbers: U01NS076788 and U01NS077108). Participants were enrolled from 31 sites across the United States. Eligible participants met International Classification of Headache Disorders (ICHD) criteria for migraine with or without aura or chronic migraine without continuous headache or medication overuse. More details regarding CHAMP inclusion and exclusion criteria can be found in the published study protocol (7).

The CHAMP study was approved by the Cincinnati Children’s Hospital Medical Center institutional review board (IRB); enrolling sites complied with Cincinnati Children’s IRB or local IRB regulations. Written informed consent was obtained from the guardians of all participants in the study; written assent was obtained from study participants as required per local/Cincinnati Children’s IRB regulations.

Procedures

Participants completed a prospective headache diary during a 28-day baseline period. They were then randomized in a 2:2:1 ratio to receive one of two preventive migraine medications – amitriptyline or topiramate – or placebo, which they subsequently took each day over a 24-week active treatment period. Participants maintained a daily calendar diary throughout the study, which was collected at study visits that occurred approximately monthly (7). Information regarding CHAMP trial procedures was previously reported in line with the Consolidated Standards for Reporting Trials (CONSORT) (4).

Outcomes

Our current primary outcome measure, daily headache occurrence, was captured using responses to the prospective daily headache diaries that participants completed throughout baseline and active treatment phases of the study. Participants received monthly calendars and were instructed to document whether (“yes” = 1) or not (“no” = 0) a headache occurred each calendar day. Daily headache occurrence was determined using the proportion of participants completing a diary on each baseline or active treatment period day who endorsed experiencing a headache divided by the total participant sample. Here, the range would be 0% – 100% on a given day, where 100% would correspond to all participants with a diary entry reporting a headache that day. Daily headache occurrence was calculated for each treatment group and for the combined CHAMP sample.

Statistical analyses

The analytic plan focused on trajectories of daily headache occurrence in the baseline and active treatment study phases. Because the CHAMP protocol allowed active treatment visits to occur within a flexible window (i.e., each visit was scheduled 4 ± 1 weeks after the previous visit), participants differed in the number of days in which headache frequency information was collected during the active treatment phase of the trial. To address this, active treatment trajectory analysis only included headache diary data from the first 24 weeks (168 days) post-randomization. Data were first aggregated to visualize observed daily headache occurrence in each treatment group across each of the 28 days in the baseline period and each of the 168 days in the active treatment period (See Figure 1, Panel 1). Baseline and treatment response trajectories were then estimated using longitudinal hierarchical linear modeling for our binary outcome variable of daily headache occurrence. Study sites were controlled for in tested models to correct standard error estimates and avoid inferential errors.

Figure 1.

CHAMP daily headache occurrence: Visualized data by treatment group and modeled trajectories in combined sample.

Analyses were conducted to identify the best fitting and most parsimonious longitudinal trends for the combined CHAMP sample for both baseline and active treatment trial periods. To determine whether baseline and active treatment trajectory models corroborated findings of no between-group differences in headache day reduction at the end of the CHAMP trial, we tested whether being randomized to amitriptyline or topiramate resulted in significant differences in baseline and/or active treatment trajectories of daily headache occurrence when compared to placebo. To examine the influence of participant gender and age on our findings, we tested these variables as predictors of headache day occurrence across CHAMP baseline and active treatment periods.

Subsequently, we examined the extent to which independently developed trajectory models replicated trajectory models established using data from a clinical trial of CBT plus amitriptyline for chronic migraine. It should be noted that the best fitting trajectory model for baseline daily headache occurrence in the CBT plus amitriptyline trial was intercept-only, indicating no significant change in headache frequency (6). Across treatment groups, best fitting models in the active treatment period of the CBT plus amitriptyline trial were characterized by a significant negative linear slope, and a significant positive quadratic change. Only the linear slope showed significant variation across participants (linear slope was predicted by treatment group membership, such that the magnitude of headache reduction was greater for youth in the CBT plus amitriptyline group across the active treatment period versus the comparison group of amitriptyline plus education attention control).

Unconditional modeling of CHAMP baseline and active treatment trajectories was conducted using a model-building procedure, where fixed effects were added to the model first, and then random effects were added. Linear and non-linear (e.g. quadratic, cubic) polynomial trend components were added sequentially to the model until no more parameters could be added. The model-building process followed four established analysis guidelines to determine the final analysis model (see Table 1). Models were further subjected to likelihood ratio nested model testing to confirm that the most parsimonious baseline and treatment models were obtained using the above guidelines. These tests determined the extent to which model fit improved with additional parameter estimates in the trajectory model.

Table 1.

Established statistical guidelines for parameter retention in longitudinal models (8).

1). If the fixed effect and random effect for a given a polynomial trend component were both statistically significant, then both were retained in the longitudinal trajectory model.

2). If a fixed effect for a given trend component was significant but the random effect for that trend component was not, only the fixed effect was retained in the model.

3). If the fixed effect for a given trend component was not statistically significant, but the random effect was, both components were retained in the trajectory model.

4). If neither fixed nor random effects for a given trend component were statistically significant, or if their inclusion resulted in a non-significant increase in model fit, both trend components were removed from the trajectory model and model-building stopped.

Once the most parsimonious baseline and active treatment trajectory models were established, dummy-coded treatment group variables for topiramate and amitriptyline (placebo = reference group) were added to the models to test for significant headache trajectory differences. Final trajectory models were then compared to those established by Kroner and colleagues for the CBT plus amitriptyline trial (6). There was minimal missing data (<.001%) in the current dataset, which was handled using maximum likelihood estimation.

To contextualize our trajectory findings, secondary analyses were conducted to examine the proportion of participants who met two clinically meaningful benchmarks: 1) a ≥50% reduction in headache frequency as compared to baseline frequency, and 2) ≤4 headache days per month. These were computed for each month of the active trial period and compared across treatment groups using χ² analyses; false discovery rate correction was used to control family-wise Type I errors. The proportion of CHAMP participants meeting the benchmark of ≤ 4 headache days per month was also calculated during the baseline period for comparison. Significance was determined via p < .05 (2-tailed).

To quantify the proportion of delayed treatment responders in our sample, we conducted an unconditional discrete time survival analysis among youth who had not shown a clinically meaningful initial treatment response to determine if they met clinical benchmarks in later trial periods. Specifically, we first removed all initial “responders” (i.e., those individuals who had an initial ≥ 50% reduction in headache frequency as compared to baseline, or those who had reached a headache frequency of ≤ 4 headache days per month) from our dataset and treated clinical benchmark attainment at subsequent trial months as a survival “event” of interest. This allowed conditional probabilities of clinical benchmark attainment for initial treatment non-responders at later months in the CHAMP trial to be computed. Because of the inherently categorical nature of our clinical benchmark data (i.e., we had to account for all headache days in a given “month” or 28 day period before calculating whether or not a participant met a clinical benchmark), continuous survival analysis could not be completed. All analyses were completed using Mplus Version 8.6 (9) and SPSS Version 25 (10).

Results

Information regarding participant demographics can be found in the CHAMP primary efficacy results (4). Females reported more headache days than males over the course of the baseline (b = .578, p = .001; 95% CI: 0.239 – 0.916) and active treatment periods (b = 0.609, p = .003; 95% CI: 0.203 – 1.015) of the CHAMP study.

Figure 1 (Panel 2) shows the best fitting and most parsimonious trajectory models identified for the CHAMP baseline and active trial periods. The final baseline trajectory model of daily headache occurrence was characterized by a significant negative linear slope (b = −.035; p < .001; 95% CI: −.049 – −0.020) and a positive quadratic change term (b = 0.001; p < .001; [95% CI: .001–.002]). For the active treatment phase of the CHAMP study, the best fitting longitudinal model was characterized by a significant and negative random linear slope (b = −.132, p < .001; 95% CI: −.312 – −0.189) and positive random quadratic change term (b = .004; p = .023; 95% CI: <.001–.008). Subsequent calculations showed that model-predicted quadratic change was zero at day 16.5 of the active trial period. For both baseline and active treatment phases, treatment group membership did not show superiority to placebo in terms of predicting change in daily headache occurrence. Neither age, gender, the interaction of age and gender, nor the interaction of these variables by treatment group significantly predicted changes in daily headache occurrence in the baseline and active trial periods of the CHAMP study.

When compared to the best fitting trajectory models for the CBT plus amitriptyline trial, model differences were minimal for baseline and active treatment periods of the two trials. For the baseline period, daily headache occurrence initially decreased but then began to increase toward initial daily headache levels. Similar to findings by Kroner and colleagues, best fitting trajectories in the active treatment period of CHAMP revealed a negative linear effect and a positive quadratic effect. In the current study however, the quadratic effect varied significantly across participants.

As shown in Figures 2 and 3, the proportion of participants who achieved clinically meaningful benchmarks of treatment response increased across the trial period. Over half of the CHAMP sample experienced a ≥50% reduction in headache frequency in the second month of the active trial period; over half of the sample had ≤ 4 headache days per month in the third month of the trial period. There were no treatment-group differences in the percentage of participants who achieved either of these clinical benchmarks across the active trial period.

Figure 2.

Percentage of participants who reported a ≥50% reduction in monthly headache frequency, as compared to baseline.

Figure 3.

Percentage of participants who reported less than 4 headache days per month of the CHAMP trial (baseline included for reference).

Survival analyses were conducted to determine conditional probabilities of clinical benchmark achievement at 12, 16, 20, and 24 weeks of the active trial period among those who failed to experience a ≥50% reduction in headache frequency and/or ≤4 headache days per month in the second month of the CHAMP trial. For both clinical benchmarks, the conditional probabilities for a CHAMP non-responder at 8 weeks attaining a clinical benchmark in a given (subsequent) trial month was roughly 20% (Range: 15–30%; see Table 2).

Table 2.

Survival analysis results showing conditional probabilities of attaining a clinical benchmark after failing to do so after 8 weeks in the CHAMP trial.

	Clinical Benchmarks
Trial Month	Headaches ≤4 / Month	≥50% Headache Reduction
3 (week 12)	26.25% (NTOT = 141)	27.51% (NTOT = 120)
4 (week 16)	18.63% (NTOT = 102)	15.29% (NTOT = 85)
5 (week 20)	18.51% (NTOT = 81)	30.01% (NTOT = 70)
6 (week 24)	21.87% (NTOT = 64)	21.28% (NTOT = 47)

Discussion

The current project investigated trajectories of headache frequency across the one-month baseline and six-month active treatment periods of the CHAMP study; findings revealed no meaningful change in headache frequency during the prospective baseline period. During the active trial period, notable decreases in headache days began early in treatment and continued throughout each trial month. Quadratic trends were found across treatment groups in our headache frequency data, suggesting that although there may be some variability across participants, the most significant improvements in headache days occur in the first “month” (28 day period) or two of the trial period and then become more gradual.

Taking initial mean CHAMP headache frequency data into account, these findings suggest that over half of the sample had a reduction in headache frequency by the end of the second month (56 days) of the trial that was equivalent to going from ∼ 11.5 to 5 or 6 headache days per month. Over 40% of the sample experienced a headache frequency reduced to ≤4 headache days per month by this time. Treatment that leads to experiencing a migraine less than once a week – when our patients often present with a 3 to 4 year history of headache and a frequency of 3 to 4 per week – shows the promise of caring for youth with migraine.

Importantly, our trajectory findings replicated those of a separate trial of two preventive treatments (CBT plus amitriptyline and amitriptyline plus education attention control) for youth with chronic migraine. Replication is one of the great challenges in scientific research and is the foundation for making inferences about what findings are real, meaningful, and generalizable to the population at large (11). Although there are always limitations in generalizing findings from clinical trials to everyday practice, replicating this pattern of treatment response across five randomized treatment groups provides strong evidence of an expected preventive treatment response for clinicians to use in their work.

As hypothesized, neither amitriptyline nor topiramate showed greater clinical benefit over placebo in these trajectory analyses. The current findings highlight that the quick and notable improvement in headache frequency seen in this clinical trial was not due to the pharmacological/chemical aspect of the pill-based treatment provided (12). Given that preventive medications for youth with migraine have bothersome and sometimes serious side effects, there have been increasing calls in the literature to take the lack of evidence for prevention medication seriously and to focus on non-pharmacological and coping-based treatments for youth with migraine (3,12–15). The 2020 practice guidelines from the American Academy of Neurology and the American Headache Society now recommend that, ‘clinicians should engage in shared decision making with patients and caregivers regarding the use of preventive treatments for migraine, including discussion of the limitations in the evidence to support pharmacologic treatments (3, p. 500).’ These clinical guidelines suggest that behavioral treatments, despite the challenges of access, be combined with pill-based therapies for prevention of migraine in youth (3,16–18). Notably, our CHAMP trajectory results add to the active debate about the clinical use of preventive medications for youth with migraine and have implications for possible use of newer classes of medications (e.g., calcitonin gene-related peptide (CGRP) drugs) in clinical practice (3,12–14,16,19–21).

More targeted research is needed to understand why so many youth can demonstrate relatively rapid and clinically meaningful improvement from headache treatments (12). Future studies should investigate potential mechanisms associated with the act of pill-taking (e.g., expectations of treatment benefit, increased pain coping self-efficacy, changes in associated brain activity)(22,23). Several other potential mechanisms of improvement are worth exploring as the CHAMP protocol also included a tailored, evidence-based acute treatment plan (often including non-steroidal anti-inflammatory drugs (NSAIDs) and/or triptan medications); behavioral plans to increase healthy habits (e.g., adequate sleep duration, regular exercise and daily meals, and increased hydration); and regular contact with supportive research staff (about once per month over the course of the trial; included discussions to promote treatment adherence as needed).

Given that migraine can often be a lifelong condition, identifying biopsychosocial dimensions of treatment that can be utilized to empower children to better manage pain may be a preferred alternative (or addition) to a trial of prophylactic pharmacological treatment (15,18). The current literature suggests that cognitive behavioral therapy – a time limited psychological treatment package that includes training in relaxation skills, activity pacing, and cognitive reappraisal skills – may be a particularly beneficial treatment option for youth with migraine. Psychological interventions (including cognitive behavioral therapy) used as monotherapy have been shown to decrease headache days in this patient population (24), and CBT has been shown to improve the clinical status of youth with migraine over and above that of prophylactic pill-based treatment alone (5).

That said, additional research is needed to test the question of treatment sequencing as a means of maximizing the impact of prophylactic headache interventions. Current guidelines for pediatric controlled trials of preventive migraine treatments suggest sequential, multiple assignment, randomized trial (SMART) designs as a novel methodology to utilize in future research (25). In SMART designs, participants are randomized to a given intervention, and then go through additional stages of the trial where they are randomly (re)assigned to one of several intervention options based on pre-determined decision points regarding their initial treatment response (26). This trial design could be used to identify patient variables that influence initial treatment response, compare the relative effectiveness of multi-component interventions, and identify the optimal time to introduce new biologics (e.g., CGRP antibodies) and/or neuromodulator devices as a preventive treatment. The results of the current study – when taken together with the CBT plus amitriptyline trial – provide the data needed to inform 8 weeks as a tailoring variable (or time point variable for clinical decision making) to be used to determine whether a pediatric participant is a treatment responder or non-responder (and therefore needs to be re-randomized in the trial). The current work represents a critical step towards incorporating adaptive trial designs in migraine research.

Current findings also suggest that measures of time to clinical improvement in headache frequency could be used as an alternative measure for assessing primary endpoints in future clinical trials, as opposed to using a set trial period (e.g., six months). This “survival” analysis approach would minimize participants’ exposure to ineffective treatments or those with potential adverse events and allow for a more efficient identification of new treatments and is already being widely used in trials for other disease processes. For example, clinical trials of juvenile arthritis utilize primary outcome measures such as % of the sample with an arthritic flare, or % of sample that show clinical improvements per American College of Rheumatology Ped-30 criteria (27). Future clinical trials of youth with migraine could assess the time it takes participants to achieve a clinically meaningful benchmark (e.g., ≥50% reduction in headache days, ≤4 headache days per month) as a means of evaluating response to preventive treatment, and determining when participants complete the active phase of a clinical trial. Results could be compared to an expected treatment response trajectory born out of our current work.

While such novel endpoint approaches would not have demonstrated a medication effect in the CHAMP study, we hope that they could provide a more precise and cost-effective means of evaluating preventive treatment efficacy for youth with migraine in future trials of other interventions (new biologics, devices, behavioral and mind and body therapies, or their combination). Our group has demonstrated that clinical improvements from pill-based preventive migraine treatments are maintained for several years after discontinuing medication treatment (28). Given this, as the rate of improvement levels off, patients could be taken off preventive migraine medications and the maintenance of clinical benefit could be tested in this “survival” analysis approach.

As a final note, the current replication of rapid and sustained improvement in the clinical status of youth with migraine across two independent trials raises caution regarding the common practice of extrapolating the results and implications from adult studies and practice guidelines to children and adolescents. The field needs to take a more developmental approach to researching and treating youth with migraine, and should utilize research designs that inform targeted and adaptive interventions for youth that do not respond to a treatment as anticipated. A nuanced, proactive, and efficient approach, as well as additional rigorous research, is needed to maximize treatment benefit for this pediatric patient population.

Clinical implications

Youth who receive pill-based preventive therapy improve, particularly over the first 8 weeks of treatment

In line with primary CHAMP study findings, preventive medications do not show greater clinical benefit over placebo in terms of trajectory of treatment response.

A proactive, multi-modal treatment approach is indicated if this initial response pattern is not observed.

Guided by the current study, future clinical trials can utilize adaptive designs, as well as time to clinical endpoint as a novel clinical outcome measure for new biologics.