Effect of grass sublingual tablet immunotherapy is similar in children and adults: A Bayesian approach to design pediatric sublingual immunotherapy trials

Abstract

Background:

Large sample sizes are needed for sublingual immunotherapy (SLIT) trials because of inherent data variability secondary to inconsistent allergen exposure. Obtaining large sample sizes for pediatric SLIT trials is challenging, but a Bayesian approach using prior adult data can reduce the necessary sample size.

Objective:

We sought to describe how a Bayesian framework using prior information from adult trials can be used to improve pediatric SLIT clinical development.

Methods:

Data were compiled by using a frequentist approach (conventional clinical trial approach independent of prior data) from trials conducted during the clinical development of timothy grass SLIT-tablets.

Results:

The treatment effect of timothy grass SLIT-tablets was considered similar between pediatric (n = 795) and adult (n = 2299) data pools, with relative total combined symptom plus medication score improvement versus placebo of 21% (95% CI, 11.0% to 30.4%) and 20% (95% CI, 14.6% to 24.4%), respectively. Phleum pratense–specific IgG₄ and IgE-blocking factor increased from baseline in both children and adults treated with timothy grass SLIT-tablets. Given the reasonable assumption in similarity of treatment response between adults and children, a Bayesian approach is described to demonstrate rigorous efficacy criteria for pediatric trials incorporating information from prior adult trials and thereby reduce the sample size.

Conclusions:

Data support the similarity of efficacy and immunologic changes between children and adults treated with SLIT for allergic rhinoconjunctivitis. Therefore it is appropriate to use data from adult trials to design feasible trials in children, which might reduce unsafe off-label use by promoting more quickly proper labeling of approved products.

Pediatric trials of new allergy immunotherapy (AIT) treatments are conducted typically subsequent to establishing efficacy and safety in the adult population. There are inherent challenges in conducting large pediatric trials in a reasonable timeframe in situations in which efficacy evaluation in pediatric subjects requires the same strict bar as in adult trials. The solution for a feasible pediatric development might lie in a hybrid approach of frequentist (the standard statistical approach used in clinical trials that does not require any prior information to draw conclusions) and Bayesian framework (using prior and current data) methods in the study design and data analysis after the similarity of effect between adults and children has been reasonably established.¹

Unlike many chronic disorders, allergic rhinitis with or without conjunctivitis (AR/C) affects both children and adults. There is a large body of evidence that demonstrates the efficacy and safety of subcutaneous immunotherapy and sublingual immunotherapy (SLIT) for the treatment of AR/C in adults.^2–4 SLIT can be a particularly attractive treatment option for AR/C in children because it eliminates the need for injections, which children might fear, and it has the advantage of home administration. In the United States grass SLIT-tablets are the only SLIT products currently approved for children,^5,6 and an increasing number of physicians administer SLIT using aqueous extracts approved for subcutaneous AIT despite the lack of documented efficacy and safety data for off-label use.^7,8 Thus there is a need for more pediatric trials for SLIT products.

The US Food and Drug Administration (FDA) criteria to demonstrate the efficacy of AIT is based on patient-reported outcomes, such as symptoms and medication use. The end point must show an at least 15% difference versus placebo with a 95% CI lower limit of at least 10% for the primary efficacy outcome.⁹ Several trials show that a sample size of more than 1000 subjects might be required^10–12 because AIT field trials are complicated by uncontrollable and variable environmental allergen exposure in which patient-reported outcome parameters inherently demonstrate relatively large variability.¹³ There is a lower incidence of inhalant allergy in children, making accrual slower (years and multiple seasons) or infeasible, and obtaining such a large sample size is more challenging for pediatric than adult clinical trials. It might be impractical to achieve adequate power in the pediatric population, and thus in this case the only evidence for robust treatment efficacy can be derived from adult AIT trials.

It is natural to consider borrowing information from adult AIT trials because AR/C pathophysiology is similar in children and adults, AIT is dosed similarly in children and adults, and the mechanism of action of AIT is the same.⁴ Bayesian methodology can be applied to improve the power of clinical trials, thereby reducing the sample size needed to adequately demonstrate efficacy. The concept of Bayesian methodology borrows strength from historical information, such as that obtained from adult trials, and incorporates the information to other related populations, in this case children.¹

Use of Bayesian methodology to leverage information from prior studies is considered for FDA approval of medical devices and is supported by the European Medicines Agency but is not typically used in late-stage development of pediatric programs.^14,15 Furthermore, the FDA Pediatric Research Equity Act guidance suggests that if the course of the disease and effects of the drug are sufficiently similar in adults and pediatric patients, pediatric effectiveness can be reasonably assumed from adequate and well-controlled adult trials.¹⁶ The objective of this analysis is to present evidence suggesting that the efficacy of SLIT treatment is similar in children and adults by using a frequentist approach and discuss how a Bayesian approach can be used to improve pediatric SLIT clinical development. In this context it is important to note that the appropriateness of leveraging efficacy data from adults is independent from extrapolating for the assurance of safety. The safety assessment of SLIT products in children is not discussed in this article and should be discussed separately based on the regulatory requirements.

METHODS

Data sources

The descriptive summary of treatment effect in adults and children was compiled based on historical data from pediatric and adult phase 3 trials conducted during the clinical development of timothy grass SLIT-tablets (Merck, Kenilworth, NJ, and ALK-Abelló, Hørsholm, Denmark). The trials were randomized, double-blind, placebo-controlled trials and have been described previously.^10,17–21 Key inclusion criteria for all the trials were a clinical history of physician-diagnosed grass pollen–induced AR/C with or without controlled asthma, a history of medication use to treat AR/C symptoms during the previous pollen season, and a positive skin prick test response and serum-specific IgE level for P pratense. Data from GT-12¹⁸ and P05239¹⁷ and pediatric (<18 years of age) data from P08067¹⁰ were pooled into a "pediatric" pool, and data from GT-08 (year 1),¹⁹ P05238,²¹ GT-14,²⁰ and the adult data from P08067¹⁰ were pooled into an "adult" pool. The efficacy outcome for grass SLIT-tablets was the total combined score, which is the sum of the rhinoconjunctivitis daily symptom and daily medication scores. The treatment difference relative to placebo was defined as follows:

$$\text{(Placebo–Study drug)/Placebo}\mathrm{\ast }\text{100\%}\text{.}$$

Grass-specific IgG₄ and IgE-blocking factor (which represents competition between IgE and non-IgE antibodies for allergen binding) data from GT-08 (year 1), GT-12, GT-14, P05238, and P05239 were compiled also, and descriptive statistics were provided for different age groups.

One other marketed SLIT-tablet is available, and data for indirect comparison (meaning data were collected from separate studies) of efficacy between children and adults for the 5-grass SLIT-tablet (Stallergenes SA, Antony, France) were obtained from 2 separately conducted trials, as previously described.^22,23 Key inclusion criteria for the 2 trials were a clinical history of grass pollen–induced AR/C for 2 previous seasons, a retrospective rhinoconjunctivitis daily symptom score of 12 or greater based on recall of severe days during the previous pollen season, a positive skin prick test response to 5-grass pollen, and a positive serum-specific IgE level against P pratense. The outcome reported is the rhinoconjunctivitis total symptom score.

Bayesian framework

When appropriate, the use of available adult data can be used to facilitate a "posterior" (updated distribution after new data become available) conclusion about children. Within the Bayesian framework, there are generally 2 ways of incorporating the historical information (adult trials data) into the new trial (pediatric trial):

1. to treat the historical data as "prior" in the model and the new trial data as the "likelihood" of the parameter of interest, thereby jointly deriving the "posterior" information using both "prior" and "likelihood" and

2. to treat both the historical and new trials as independent realizations of the same underlying distribution of true treatment differences. Thus the historical trial and the new trial are indirectly linked. This is discussed in more detail in the next section.

For purposes of SLIT studies, the quantity of interest would be the treatment difference relative to placebo. It is assumed that the pooled adult data have a true treatment difference (between the study medication and placebo) of β_A, the pediatric data has a true treatment difference (between the study medication and placebo) of β_P, and that β_A and β_P are independent and have the same underlying normal distribution, with mean μ and SD ν values and a resulting formula of β_A, β_P ∼ N (μ, ν²). Note that β_A and β_P values are usually considered fixed unknown parameters in a frequentist framework, but in hierarchical models β_A and β_P are considered random variables themselves.

Denote the maximum likelihood estimators of β_A as $\widehat{{\beta }_A}$ with SE $S_A/\sqrt{n_A}$, where n_A is the sample size per arm for adult data. Also, ${\stackrel{\mathrm{ˆ}}{\mathit{\beta }}}_P$, S_P, and n_P are defined similarly for the pediatric data. In cases with multiple adult and multiple pediatric trials, such as the timothy grass SLIT-tablet program, ${\hat{\beta }}_A$ and ${\stackrel{\mathrm{ˆ}}{\mathit{\beta }}}_P$ are assumed to be estimates from a meta-analysis. Assuming normal distribution of the estimators results in the following hierarchical model:

$$\begin{gathered} {\hat{\mathit{\beta }}}_{\mathit{A}}~\mathit{N}({\beta }_A,S_A^2/n_A) \\ {\hat{\mathit{\beta }}}_P~\mathit{N}({\beta }_P,S_P^2/n_P) \\ {\beta }_A,{\beta }_P~\mathit{N}(\mu ,v^2), \end{gathered}$$

where μ has a noninformative hyper prior and the value of ν will be elicited based on the historical information and additional details provided in following subsections.

Based on the hierarchical model, one can determine the posterior distribution for ${\mathit{\beta }}_{\mathit{P}}|{\stackrel{\mathrm{ˆ}}{\mathit{\beta }}}_{\mathit{A}}\mathrm{,}{\stackrel{\mathrm{ˆ}}{\mathit{\beta }}}_P$, which will allow evaluation of the associated 95% credible interval.¹ Additional details on the model assumptions are discussed below.

Justification of model assumptions

The following 2 pivotal assumptions are made for the hierarchical model:

1. Adult and pediatric data are normally distributed around true treatment means. For the timothy grass trials, when the distribution of data deviated from the normality assumption, the nonparametric analyses were performed, and the results were compared with those from the parametric analysis under normality assumption. In these trials numerically larger treatment effects were observed for primary efficacy end points when using nonparametric compared with parametric analysis. For planning purposes and during the design stage of each trial, the assumption of underlying normal distribution could be considered conservative in the sense that even if the distributional assumption deviates from normality, the proposed sample size would lean toward a conservative direction.

2. The true placebo-adjusted treatment effects β_A and β_P come from the prior distribution N (μ, ν²). The normality assumption of the prior distribution is relatively robust for such a hierarchical model. The μ value is given a noninformative hyper prior, and usually, the format of a hyper prior is less sensitive. The value of ν is discussed in detail below.

The choice of ν

The ν value measures the SD of the 2 populations and indicates the impact for the prior adult data on the pediatric data. As suggested by Schoenfeld et al,¹ the ν value can be derived from a previous trial with both adult and pediatric data as follows:

$$|{\hat{\mathit{\beta }}}_{\mathit{A}}-{\hat{\mathit{\beta }}}_{\mathit{P}}|\mathrm{/}\sqrt{2}\mathrm{.}$$

One of the most important aspects in building a Bayesian analysis model is to avoid an overinformative prior probability distribution in which a large-sized historical trial is used to build the prior distribution and totally dominates the current smaller-sized trial. The Bayesian hierarchical model used in this study assumes that subjects are exchangeable within each population (pediatric and adult) but not across populations. This assumption enables adult studies to provide strength to the pediatric study but not as much as if the subjects in different populations are directly able to be pooled. In this sense it protects against overreliance on the adult data. However, there are additional variations of Bayesian methodologies that could be considered for incorporating adult information.

RESULTS

Evidence for similarity of SLIT efficacy in children and adults

The P08067 trial evaluated the efficacy of timothy grass SLIT-tablets in children and adults. The treatment effect with timothy grass SLIT-tablets was consistent between children and adults, with a treatment absolute score difference from placebo for TCS based on medians of −1.32 and −0.90, respectively, which translates to an improvement versus placebo of 32% and 21%, respectively (Table I).¹⁰ Numeric differences between the 2 age groups were observed; however, the results reveal a trend toward greater efficacy in children. A similar treatment effect across the age groups was confirmed when comparing the timothy grass SLIT-tablet pediatric and adult data pools (Table II). Note that the analyses in Tables I and II are based on the conventional frequentist approach.

TABLE I.

TCSs for children and adults in trial P08067 of grass SLIT-tablets

			Treatment difference		Improvement vs placebo*
End point	Grass SLIT-tablet	Placebo	Estimate (95% CI)	P value	Estimate
Children	n = 123	n = 127
Median TCS†	2.77	4.09	−1.32 (−1.9 to −0.1)‡	.020	32%
Mean TCS	3.91	5.12	−1.21 (−2.2 to −0.2)		24%
Adults	n = 506	n = 545
Median TCS†	3.33	4.24	−0.90 (−1.1 to −0.3)‡	<.001	21%
Mean TCS	4.29	5.12	−0.84 (−1.4 to −0.3)		16%

Adapted with permission from Maloney et al.¹⁰

TCS, Total combined score (defined as the sum of the rhinoconjunctivitis daily symptom score and rhinoconjunctivitis daily medication score).

^*Calculated as follows: (Placebo − Grass SLIT-tablet)/Placebo × 100%.

^†The median from the nonparametric model was the prespecified primary analysis.

^‡The Hodge-Lehmann estimate was used as the basis for 95% CIs. P values were calculated by using the Wilcoxon rank sum test.

TABLE II.

TCSs for children and adults from pooled grass SLIT-tablet trial data

	Grass SLIT-tablet	Placebo	Treatment difference*		Improvement vs placebo†
End point	Mean	Mean	Estimate (95% CI)	P value	Estimate (95% CI)
Children	n = 389	n = 406
TCS	4.63	5.87	−1.24 (−1.86 to −0.62)	<.001	21% (11.0% to 30.4%)
Adults	n = 1111	n = 1188
TCS	5.30	6.59	−1.30 (−1.65 to −0.94)	<.001	20% (14.6% to 24.4%)

TCS, Total combined score (defined as the sum of the rhinoconjunctivitis daily symptom score and rhinoconjunctivitis daily medication score).

^*Calculated by using ANOVA with treatment and study as factors in the model.

^†Calculated as follows: (Placebo − Grass SLIT-tablet)/Placebo × 100%. The bootstrap method was used to obtain 95% CIs.

Additional support for the similarity in efficacy of SLIT between adults and children is provided by results from 5-grass SLIT-tablet trials. The 5-grass SLIT-tablet groups in the pediatric and adult trials reveal a 31% and 29% reduction in the rhinoconjunctivitis total symptom score, respectively.²⁴

Evidence for similarity of immunologic response to SLIT in children and adults

P pratense–specific IgG₄ levels increase from baseline in those subjects treated with timothy grass SLIT-tablets, regardless of age in pooled timothy grass SLIT-tablet trials (Fig 1). Similarly, the IgE-blocking factor reveals increased blockade of the IgE antibody with timothy grass SLIT-tablet treatment, which was slightly higher in children versus adults (Fig 2). No statistically significant difference across age groups was found in the placebo-adjusted mean change from baseline for IgG₄- or IgE-blocking factor. These results are consistent with the clinical efficacy data and corroborate the assertion that the mechanism of action based on the adaptive immune response is analogous in young children, adolescents, and adults. Note that the analyses in Figs 1 and 2 are based on the conventional frequentist approach.

FIG 1.

Summary by age of mean change from baseline in log₁₀ IgG₄ levels at the end of the season in pooled timothy grass SLIT-tablet trials. n, Number of subjects with IgG₄ at the end of the season.

FIG 2.

Summary by age of mean change from baseline in IgE-blocking factor at the end of the season in pooled timothy grass SLIT-tablet trials. n, Number of subjects with IgE-blocking factor at the end of the season.

Example of the effect of the Bayesian model on pediatric trial design

If using the pediatric study only (frequentist approach), the total number of randomized subjects needs to be 2920 (1460 per arm) for 85% power (Table III).^11,25 Using methodology proposed by Schoenfeld et al,¹ we estimate the parameter of heterogeneity across adult and pediatric population ν values of about 0.1 and a mean treatment difference from placebo of −1.22 by using historical data, with a target relative treatment improvement of 18% with SLIT versus placebo to demonstrate how the Bayesian model can affect sample size and the power of a pediatric SLIT trial. By using these parameters, the sample size of a pediatric trial could be reduced to 290 subjects per treatment arm with 85% power to meet the FDA criteria of at least a 10% lower bound confidence interval compared with the adult trial sample size of 910 subjects per treatment arm (set from previously described house dust mite SLIT-tablet studies MT-06 and P001)^11,25 used in the Bayesian model (Table III).

TABLE III.

Power and sample size planning of a pediatric SLIT trial using a Bayesian approach

Historical mean treatment difference25	Target treatment improvement relative to placebo	Total randomized (per arm)*	ν†	Power‡
−1.22	18.1%	580 (290)	0.2	54.6%
			0.1	85.0%
		696 (348)	0.2	59.3%
			0.1	86.3%
		860 (430)	0.2	64.8%
			0.1	88.1%
		1840 (920)§	0.2	85.0%
			0.1	95.0%

Note: The sample size information for the combined adult study is set as 910 per arm (studies MT-06 and P001).^11,25

^*Assuming a dropout rate of 14%.

^†The SD between adult study and pediatric study treatment effects.

^‡The power of the test that rejects the null if the lower bound of the 95% credible interval of the true treatment difference in the pediatric study is greater than 10% of the placebo effect.

^§If using the pediatric study only (frequentist approach), the total number of randomized subjects needs to be 2920 (1460 per arm) for 85% power.

DISCUSSION

There are inherent challenges in conducting large pediatric trials in a reasonable timeframe in situations when the efficacy evaluation requires the same strict bar as adult trials. Depending on the course of disease and the similarity of treatment effect across adults and children, a potential solution for a feasible pediatric development might lie in a hybrid approach of a frequentist and Bayesian framework in study design and data analysis.

Overall, grass SLIT trial data from adults and children reveal similar efficacy and immunologic changes, providing reasonable reassurance that SLIT works in a similar manner in adults and children 5 years and older. This result is supported by a 2010 meta-analysis of SLIT for allergic rhinitis that demonstrates that the treatment effect in children is similar to that in adults.²⁶ SLIT is somewhat unique in that the SLIT dose in children and adults is the same because there is limited systemic absorption of the allergens.²⁷ In patients with AR/C, the disease pathophysiology and effect of the allergens in children and adults are the same. Allergen acts directly on the immune system through antigen-processing cells,²⁸ and the immune system is fully mature shortly after birth, with antibody response fully functional by the age of 2 years.^29,30 Therefore the response of the immune system to AIT is comparable from the age of 5 years through adulthood. Furthermore, there is no evidence to suggest that the mechanistic treatment effect obtained within 1 allergen family (among inhalant allergens) will differ between children and adults. Hence the AIT guidelines do not draw a distinction between age groups with regard to the value and efficacy of treatment nor do they recommend differing doses for adults versus children.⁴

Three trials would need to be completed before applying the suggested Bayesian framework to a future pediatric trial: 1 adult trial and 1 pediatric trial from a "similar" drug and 1 adult trial from the same drug. The first 2 trials are used to estimate/ solicit the similarity between adult and pediatric populations, and the third trial is used to plan and analyze the upcoming pediatric trial data. The Bayesian analytic approach proposed uses the available adult information for analysis of pediatric SLIT studies to demonstrate a clinically meaningful effect based on the statistical criteria required by the FDA. When there is also a pediatric study completed with the same drug, then that information could also be used during planning and analysis purposes.

There are several ways to incorporate historical information in study design and analysis.³¹ Other non-Bayesian approaches also could be considered, although the Bayesian approach is intuitive for SLIT because all Bayesian models have a structure that makes them straightforward to integrate the adult information. The proposed hybrid approach is sufficiently powered to demonstrate treatment difference relative to placebo defined as a

$$\text{(Placebo–Study drug)/Placebo}\mathrm{\ast }\text{100}\%$$

value of greater than 15% and uses the prior information from adults to provide sufficient power to demonstrate the lower bound of the 95% CI of the treatment difference relative to placebo is at least 10%. When proposing a Bayesian framework, various sensitivity analyses should be performed to examine the robustness of the underlying assumptions.

Safety is not discussed in this article, although the proposed study design allows for stand-alone safety assessment in pediatric trials. However, a smaller sample size could result in less robust safety data. Therefore additional safety studies (eg, postmarketing surveillance study) in pediatric patients might be warranted to supplement the adult existing data and provide specific safety data of clinical interest in children, such as risk of aspiration, swelling of the oral or laryngeal pharynx, or eosinophilic esophagitis.

In contrast to SLIT, using data from adults for application in children might not be appropriate for many other drugs because of pharmacologic and physiologic differences. Indeed, Janiaud et al,³² examined 89 meta-analyses representing 124 drugs and found that similar treatment effects between children and adults were observed for only 36 of the drugs, whereas 14 drugs actually had a significantly different treatment effect. For the 36 drugs with a similar treatment effect, the authors concluded that extrapolation between children and adults was reasonable. Furthermore, the large variation in dosing, extract quality, and manufacturing processes among SLIT products precludes use of the proposed Bayesian model across various SLIT products. The model should be applied only when assessing the identical SLIT product in children, as was used in the adult trials. Discussions with regulatory authorities will be needed to obtain agreement on the criteria to establish similarity between pediatric and adult populations to justify borrowing information from historical data.

Because of the inability to conduct pharmacokinetic studies (because of the lack of systemic exposure) or directly correlate biomarkers to efficacy, it is acknowledged that efficacy evaluation of SLIT in pediatric subjects is necessary to meet FDA Pediatric Research Equity Act requirements. Given the reasonable assumption in the similarity of treatment response between adults and children, the statistical success criteria based on the 95% CI for licensure in pediatric patients should not be the same as that required for the initial licensure of the product in adults.

The example implementation of the Bayesian approach for a pediatric SLIT trial presented herein has the potential to decrease the necessary sample size substantially depending on the underlying assumption of variation between the adult study and the pediatric study treatment effects. Therefore the use of adult SLIT data for pediatric drug development might benefit children as pediatric SLIT trials become more feasible and more quickly completed. Industry might be encouraged to focus on pediatric development of SLIT products because the implementation of using adult data in pediatric trials increases the efficiency of pediatric drug development.³³ Subsequently, availability of approved SLIT products will likely reduce unsafe off-label use of subcutaneous immunotherapy products for SLIT administration.

Key messages.

• Data from grass SLIT-tablet trials provide evidence that the efficacy and immunologic effects of SLIT are similar between children and adults.

• A Bayesian approach that allows using adult data provides a reasonable framework when designing future pediatric SLIT trials that are feasible to conduct with smaller sample sizes compared with stand-alone trials.

• A hybrid approach of frequentist (using current data) and Bayesian framework (using prior and current data) methods in study design and data analysis can provide a reasonable path forward to evaluate the efficacy and safety of SLIT products in the pediatric population.

Abbreviations used

AIT

Allergy immunotherapy

AR/C

Allergic rhinitis with or without conjunctivitis

FDA

US Food and Drug Administration

SLIT

Sublingual immunotherapy