Description of Studies

Two multi-centre, manufacturer-sponsored, double-blind RCTs met the inclusion criteria for this review. Studies AD-301 (N = 763; 48 sites) and 302 (N = 764; 42 sites) were identically designed and enrolled populations two years of age or older with mild-to-moderate AD (Investigator’s Static Global Assessment [ISGA] scoring) comparing crisaborole in a 2:1 ratio to vehicle over a 28-day treatment course. Both studies were centred entirely in the US. The primary outcome was the proportion of patients with success by ISGA at day 29, while secondary outcomes included the proportion of patients with ISGA of clear or almost clear at day 29, and the time to success in ISGA.

After a screening period which lasted up to 35 days, patients were randomized using an interactive Web response system, stratified by study centre. The screening period allowed for washout of prior drug, but if patients did not require a drug washout they could start the study treatment immediately. After completing 28 days of study therapy, patients were to complete a post-treatment follow-up visit within seven days of day 28. Patients who completed their treatment and were eligible could be enrolled into an open-label extension (AD-303) if they wished to do so.

Populations

Baseline Characteristics

The mean age across the two included studies was approximately 12 years, and the majority of patients were in the 2- to 11-year-old age range. There were more female (56% female) than male patients across both studies and the majority (approximately 60%) were white. All patients had an ISGA score of either mild or moderate, with the majority (approximately 61%) being moderate severity. There were no notable differences in baseline characteristics between groups.^5,6

Table 5

Summary of Baseline Characteristics.

Interventions

In both studies, crisaborole or vehicle (placebo) were applied twice daily to treatable areas (excluding the scalp), with enough ointment to cover each lesion. Treatment areas were designated and documented on day 1, and the patient or guardian were provided with this information about the designated treatment areas. If there were any new AD lesions then this documentation was updated, and treatment was to include these new lesions. The first application was performed by the investigators, and the patients and care providers were encouraged to participate in this initial application. Patients were told to wear loose fitting clothing, not wipe off study drug from the skin, avoid occlusion of the treated areas, and avoid swimming or bathing/washing the treated areas within four hours of application. Compliance was tracked using an electronic dosing diary. Dose modifications were not allowed during the study. Use of corticosteroids by any route was prohibited, as was the use of systemic immunosuppressants. Use of topical retinoids, benzoyl peroxide, or systemic antihistamines could not be escalated, decreased, or used on an as-needed basis (i.e., in a non-stable way).

Outcomes

The proportion of patients achieving success on the ISGA at day 29 was the primary outcome of AD-301 and AD-302, with success being defined as an ISGA of clear or almost clear and at least a 2-point improvement in scores from baseline. The ISGA employed in the included studies is a 5-point scale that provides a global clinical assessment of AD severity based on an ordinal scale, scored by an investigator or physician.²¹ There is variability in the number of points included in the ISGA scale,²¹ but the studies of interest for this review used a scale that ranges from 0 to 4.^5,6 A score of 0 corresponds to a grade of clear, 1 is almost clear, 2 is mild, 3 is moderate, and 4 is severe AD.^5,6 A 6-point scale exists, which reaches a maximum score of 5 (very severe); however, this was not used in the included studies. A decrease in score relates to an improvement in signs and symptoms. No minimal clinically important difference (MCID) was found for the ISGA.

Health-related quality of life (HRQoL) was assessed as an exploratory outcome using the Dermatology Life Quality Index (DLQI) (adults) and the Children’s Dermatology Life Quality Index (CDLQI) (children), as well as the Dermatitis Family Impact questionnaire (DFI). The DLQI is a self-reported, 10-item questionnaire that refers to the preceding week and assesses six different aspects that may affect quality of life as a result of living with a dermatological condition.^22,23 The aspects included in the questionnaire are symptoms and feelings, daily activities, leisure, work or school, personal relationships, and side effects of treatment.^22,23 Each item is scored using a Likert scale that ranges from 0 to 3.^22,23 A score of 0, 1, 2, and 3 corresponds to the following descriptions of how much an aspect is affected by the disease, respectively: “not at all/not relevant,” “a little,” “a lot,” and “very much.”²³ The scores of each of the 10 items are summed for an overall DLQI score between 0 and 30 (or a percentage of 30).²² The higher the score, the greater the impairment of quality of life. The MCID is a change in score of at least 3.3 from baseline. Further details regarding the validity of this instrument can be found in Appendix 5.

The CDLQI was developed using the same methods for the development of the DLQI, for children between the ages of 3 years and 16 years.²⁴ Like the DLQI, the CDLQI is a self-reported questionnaire that refers to the preceding week, used to assess the impact of skin disease on the quality of life but for children and may be completed with help from a parent or guardian. It also involves 10 questions that address the following topics: symptoms and feelings, leisure, school or holidays, personal relationships, sleep, and treatment. A score of 0,1,2, or 3 is assigned, respectively, to the following answers to each of the 10 questions: “not at all,” “only a little,” “quite a lot,” or “very much.”²⁴ No MCID was identified for the CDLQI. Further details regarding the validity of this instrument can be found in Appendix 5.

The DFI questionnaire was designed to assess the impact of disease on the quality of life of parents and families of children affected by AD.²⁵ It is a disease-specific, self-administered questionnaire that relies on a one-week recall, and consists of 10 items that were derived from ethnographical interviews and focus groups.²⁵ The 10 items of the questionnaire address the following topics: housework, food preparation, sleep, family leisure activity, shopping, expenditure, tiredness, emotional distress, relationships, and the impact on the carer’s life due to helping with treatment.²⁵ Each question is scored on a four-point Likert scale ranging from 0 to 3, for an overall total ranging from 0 to 30 with a higher score corresponding to a greater negative impact on the family’s quality of life due to AD.²⁵ No MCID was identified for the DFI. Further details regarding the validity of this instrument can be found in Appendix 5.

Studies AD-301 and AD-302 used a Severity of Pruritus Scale (SPS) to assess the extent or severity of itching in patients. It is scored on a 4-point numeric rating scale that ranges from 0 to 3. The ratings correspond to a grade and definition, where 0 is a grade of none or no itching; 1 is mild or occasional, slight itching/scratching; 2 is moderate or constant or intermittent itching/scratching which is not disturbing sleep; and 3 is severe or bothersome itching/scratching which is disturbing sleep. Once instructions had been provided, the scale was completed by the study participant or their parent or guardian, using an electronic diary, and based on a 24-hour recall.^5,6 The MCID was 0.20. Further details regarding the validity of this instrument can be found in Appendix 5.

Statistical Analysis

Patient Disposition

Patient disposition for the double-blind phase of the AD-301 and AD-302 trials are summarized in Table 6. Patients in the AD-301 and AD-302 trial were screened and enrolled in 48 and 42 sites in the US. In the AD-301 trial, a total of 925 patients were screened and 763 were randomized while a total of 923 patients in the AD-302 trial were screened and 764 were randomized. The manufacturer reported that the 162 screening failures in AD-301 and 159 in the AD-302 trial were the result of insufficient per cent body surface area or ISGA, voluntary withdrawal, or other reasons. Of the patients randomized in the AD-301 and AD-302 trials, 99.2% and 99.8% received at least one dose of crisaborole and 100% received at least one dose of placebo, respectively.

Table 6

Patient Disposition.

For the AD-301 and AD-302 trials, withdrawals were more common in the vehicle group (12.1% and 14.8%, respectively) as compared with the crisaborole group (5.9% and 6.0%, respectively). The manufacturer reported that the differences in withdrawals were primarily due to increases in withdrawals due to parent or guardian in the placebo group compared with the crisaborole group for both trials (7.0% versus 2.4% in AD-301 and 8.0% versus 2.7% in AD-302, respectively). The proportions of patients who withdrew as a result of AEs were similar in the vehicle (1.4% and 0.8%, respectively) and crisaborole groups (1.0% and 1.6%, respectively).

Internal Validity

Severity of AD and improvements in AD were measured with ISGA score by treating physicians or investigators. The validity of this score (range 0 to 4) for measuring mild-to-moderate severity, and judgment regarding what constitutes an improvement remain unknown, as did the MCID of the treatment effect as measured by the proportion of patients with change from baseline to day 29 in the score. Of note, there were remarkable vehicle (placebo) effects observed in both trials. There were 25% and 18%, respectively, in the two trials of patients who experienced success at the primary end point over the 29 days of the study period, which by definition was clear or almost clear with at least a 2-grade improvement in ISGA score. Also, as per protocol, no other concomitant therapies including TCS were allowed and protocol violation was low (3%). This strong placebo effect could reflect the natural variation of disease symptoms over time, or it could reflect patients’ response to the emollient effect of the vehicle itself, and emollients are used in managing mild AD. As a consequence, it could be highly unreliable to judge the difference of, for example, 1 to 2 or 3 to 2 in severity based on a score that is mostly a reflection of symptoms in an affected area. This is particularly problematic in assessing the treatment effect without considering the degree of change from baseline to day 29 in the grade of improvement. Therefore, a clear or almost clear with at least 2-grade improvement from baseline as the primary end point would be less prone to subjective bias comparing to a judgment on clear or almost clear alone, as one grade of difference would be difficult to discern (e.g., from 1 to 0).

The subjectivity is also an issue when patients suffered drug-related adverse effect, such as a higher incidence of application site pain (6.2% and 2.7% versus 1.2%, crisaborole versus vehicle, respectively, in both studies) than the vehicle group which might have led to the assessing physician becoming aware of the treatment assignment, and thus biased the estimate of ISGA score in favour of the study drug. Such situation may also apply to other drug-related adverse effects such as upper respiratory tract infection, pyrexia, nasopharyngitis, vomiting and nasal congestion, in which the incidence of the events was consistently higher in the crisaborole arm than the vehicle arm.

The manufacturer conducted calculations to determine the appropriate sample size for each of the included studies, and appeared to meet the minimum targets for enrolment in each group. The calculations were based upon assumptions of treatment success of 20% with crisaborole and 10% with vehicle, and it is not clear upon what basis these assumptions were made.

The manufacturer appeared to make adjustments for multiple statistical comparisons tests on the two secondary end points of ISGA success by applying a hierarchical testing procedure. Important outcomes such as HRQoL, assessed by the DLQI, and symptoms were not part of the statistical hierarchy and were not statistically assessed in either included study. Due to those random variability and possible subjective bias, the potential benefit of crisaborole on HRQoL improvement remains uncertain. AD can have a significant effect on quality of life, both in adults and children, thus this represents a significant gap in knowledge about crisaborole.

Subgroup analyses were planned and reported. No testing of heterogeneity on-treatment effect was reported. Additionally, there was no subgroup analysis reported based on baseline AD severity, although a post hoc analysis was provided to CADTH after a request of the manufacturer. Such subgroup analysis could be useful in revealing possible differential treatment effect by severity. In particular, it would help demonstrate the major driver (mild versus moderate AD) of treatment effect in the overall population.

The manufacturer has requested that crisaborole be reimbursed for patients who have failed or are intolerant to TCS, yet no subgroup analyses were provided assessing whether responses differed based on these subpopulations.

The included studies were both double-blinded; this was accomplished by the use of vehicle control. There were numerically more withdrawals in the vehicle-treated group than in the crisaborole group, and the difference seems to have been largely accounted for by withdrawals by parent or guardian. It is not clear whether this difference is an indication of a problem with the blinding. Unlike a typical matched control for an orally administered drug, a topically applied vehicle needs to be matched in all respects, including texture and odour, presenting additional challenges. It is not clear whether the manufacturer matched vehicle in all these respects. Additionally, there was a numerical increase in topical AEs with crisaborole such as application site pain, and this might have also made it difficult to maintain blinding in these patients.

The manufacturer used a MCMC method to impute missing data, and this method assumes data are missing at random. The fact that more patients withdrew in the vehicle group than in the crisaborole group in both studies, and that most of this difference was accounted for by lack of efficacy might indicate that an assumption of data missing at random might not be appropriate. Of note, the FDA statistical reviewer conducted a sensitivity analysis assuming all missing values constituted treatment failures and this did not change the overall conclusions on the relative effect of crisaborole versus vehicle on the primary end point.¹⁸ It is likely that missing data on the primary end point due to withdrawal was small and therefore did not have a substantial impact on the findings.

The use of concomitant TCS was prohibited in the included studies; however, there were a small percentage of patients in each of the included studies that were reported as using TCS during the study. These represent a protocol violation and potentially an important one as TCS are considered the standard therapy for treating AD. The proportion of patients using TCS was small, however; 3% in each group in AD-301 and 3% with crisaborole and 6% with vehicle in AD-302, and this may have mitigated the impact on the overall analysis.

External Validity

The populations in both studies were generalizable to the Canadian population that might use crisaborole with respect to age and sex, according to the clinical experts consulted for this review. However, both studies were conducted entirely in the US, and there was a relatively high proportion of African-Americans enrolled in each study, compared with the proportion one would expect to see in Canada. The clinical experts also noted that instruments relying on visual assessment of AD may be less reliable in patients with darker skin as changes in colour and other morphology may be more challenging to detect. The clinical experts also noted the relatively high proportion of responders in the vehicle-treated group, particularly in AD-301 (treatment success in 25% of patients), and speculated as to whether this high vehicle response might have been at least in part due to difficulties in assessing patients with more pigmented skin tones.

The primary outcome in the included studies was based on the ISGA, which relies on investigator assessment, in this case looking for responses of clear or almost clear on AD lesions. Thus this is a subjective measure; however, it is widely accepted as a standard instrument for assessment of AD in clinical trials according to the clinical experts consulted on this review. Yet, this ISGA score was not used in routine clinical practice as a measure to judge the improvement in treatment effect. This would make it uncertain whether the observed treatment effect as measured by this score could be readily interpretable to and meaningful in a real-world setting. The time frame for detecting an improvement appears short at 29 days; however, the clinical experts also believed this to be adequate for detecting a response to therapy.

There was no active comparator in either of the included studies. The two most relevant comparators would be the TCI and the TCS. The manufacturer-requested reimbursement criteria includes patients who are intolerant to or have failed TCS, thus, the efficacy and harms of crisaborole relative to TCS are unknown. TCI are typically considered to be an alternative to TCS in these unresponsive or intolerant patients, or in patients who wish to avoid using TCS due to their side effects, thus the lack of comparative data versus the TCI is a significant gap in knowledge.

Crisaborole employs a unique mechanism of action for topically applied drugs and is first in its class; therefore it is important to have long-term safety data regarding this drug. The included studies double-blind phase ended after 29 days, therefore this is not of sufficient duration to determine if there are any long-term safety issues associated with the use of this drug. AD is a chronic condition and patients would be expected to use a drug like crisaborole for long periods of time. There is data available from a longer-term extension; however, there is no longer a control group in this phase of the study.

Severity of Atopic Dermatitis and Atopic Dermatitis Lesions

Proportion of Patients Achieving Success in Investigator’s Static Global Assessment at Day 29

The primary outcome in both trials was the proportion of patients achieving success in ISGA at day 29. Success was defined as ISGA of clear or almost clear with at least a 2-grade improvement from baseline/day 1. For both trials, patients treated with crisaborole were statistically significantly more likely to have success as compared with those treated with vehicle (AD-301 = 32.8% versus 25.4%; P value: 0.038 and AD-302 = 31.4% versus 18.0%; P value < 0.001) (Table 7). The proportion of vehicle-treated patients who had a success was higher in the AD-301 trial as compared with those in the AD-302 trial (25.4% versus 18.0%). In addition, the manufacturer conducted an analysis using the per-protocol population, which consisted of 83% of the ITT population in Study 201 and 87% of the ITT in Study 202. ▬▬▬▬▬

Table 7

Key Efficacy Outcomes.

Subgroup analyses were presented based on age for the primary outcome, however, no interaction P values were reported and no formal comparisons were conducted between treatment groups. Nevertheless there was no clear indication that patient age impacts response to crisaborole. Additionally, in response to a request from CDR, a post hoc subgroup analysis with data pooled from both included studies was presented with responses broken down by baseline AD severity (ISGA, mild versus moderate). In this case the treatment effect was not statistically significant in patients with mild AD (success in 24.9% of crisaborole patients versus 21.1% of vehicle patients; mean difference between groups of 3.6 [95% CI, −3.9 to 11.2; P = 0.35]), while it was statistically significant in patients with moderate disease (36.7% versus 22.3% respectively; mean difference between groups of 14.4% [95% CI, 8.0 to 20.8; P < 0.0001]).²⁶ No interaction P values were reported.

Proportion of patients achieving ISGA of clear or almost clear (treatment success) was a secondary outcome of both studies. The proportion of patients with success was higher with crisaborole than vehicle in studies AD-301 (51.7% versus 40.6%, P = 0.005) and AD-302 (48.5% versus 29.7%, P < 0.001). Additionally, median time to treatment success was reported, and although these values could not be calculated for the vehicle group (had not reached a 50% response) the differences between groups were also reported as statistically significant (Table 7).

Health-Related Quality of Life

Symptoms

Adverse Events

AEs were reported in 29% of patients in the crisaborole group and 20% of patients in the vehicle group in AD-301, and in 29% of crisaborole and 30% of vehicle-treated patients in AD-302 (Table 8). Application site pain was the most common AEs in AD-301, occurring in 6.2% of crisaborole-treated versus 1.2% of vehicle-treated patients after 29 days. In AD-302, application site pain occurred in 2.7% of crisaborole-treated and 1.2% of vehicle-treated patients after 29 days.^5,6

Serious Adverse Events

There were few serious adverse events (SAEs) through the 29-day treatment period of either study. In AD-301, 1.0% of crisaborole-treated and 0.4% of vehicle-treated patients had an SAE, while in AD-302 0.6% of crisaborole-treated patients experienced an SAE, and none in the vehicle group (Table 8).^5,6

Withdrawals Due to Adverse Events

In AD-301, withdrawal due to adverse event (WDAE) occurred in 1.4% of crisaborole-treated patients and in 0.8% of vehicle-treated patients, and in AD-302 WDAE occurred in 1.0% of crisaborole patients and 1.6% of vehicle-treated patients. The only WDAE that occurred in more than one patient in either study was application site pain and application site urticaria, each occurring in two patients in the crisaborole group in AD-301 (Table 8).^5,6

Mortality

There were no deaths in either study.^5,6

Notable Harms

One patient had a hypersensitivity reaction, reported as a treatment-emergent AE, in the crisaborole group in AD-302. Otherwise there were no other hypersensitivity reactions reported.^5,6