Distinguishing Characteristics of Difficult-to-Control Asthma in Inner-City Children and Adolescents

Abstract

Background

Treatment levels required to control asthma vary greatly across a population with asthma. The factors that contribute to variability in treatment requirements of inner-city children have not been fully elucidated.

Objective

To identify the clinical characteristics which distinguish difficult-to-control asthma.

Methods

Children with asthma aged 6-17 underwent baseline assessment and bimonthly guidelines-based management visits over one year. Difficult- versus easy-to-control asthma were defined as daily therapy with fluticasone ≥500mcg +/-LABA versus ≤100mcg assigned on at least 4 visits. Forty-four baseline variables were used to compare the 2 groups using univariate analyses and identify the most relevant features of difficult-to-control asthma using a variable selection algorithm. Nonlinear seasonal variation in longitudinal measures (symptoms, pulmonary physiology and exacerbations) was examined using generalized additive mixed-effects models.

Results

Among 619 recruited participants, 40.9% had difficult-to-control asthma, 37.5% had easy-to-control asthma and 21.6% fell into neither group. At baseline, FEV₁ bronchodilator responsiveness was the most important characteristic distinguishing difficult- from easy-to-control asthma. Markers of rhinitis severity and atopy were among the other major discriminating features. Over time, difficult-to-control asthma was characterized by high exacerbation rates, particularly in spring and fall, greater day and night symptoms, especially in fall and winter, and compromised pulmonary physiology despite ongoing high dose controller therapy.

Conclusions

Despite good adherence, difficult-to-control asthma showed little improvement in symptoms, exacerbations or pulmonary physiology over the year. Besides pulmonary physiology measures, rhinitis severity and atopy were associated with high dose asthma controller therapy requirement.

Clinical Implications

Clinical baseline characteristics related to pulmonary physiology, rhinitis severity, and atopy prospectively distinguish difficult- from easy-to-control asthma.

Introduction

The concept of asthma as a heterogeneous disease is supported by several observations, particularly the variation in clinical parameters such as symptoms, pulmonary function, bronchial hyperresponsiveness and treatment responses among patients. In addition, heterogeneity of the response to treatment has become of increasing interest and importance with the development of new therapies and a movement towards precision medical approaches for asthma. The extent to which treatment response heterogeneity and associated phenotypes exist in inner-city children with asthma has yet to be defined. In an earlier, 46-week study in which guidelines-based care was provided in the context of a randomized, double-blind design testing the utility of fractional exhaled nitric oxide (FeNO) to guide therapy, inner-city children required a broad range of asthma controller treatment steps suggesting that easy and difficult-to-control asthma may, indeed, represent different disease phenotypes.¹

A considerable problem with studies that attempt to identify asthma phenotypes is that, in their majority, patient characterization occurs as a single-time evaluation and does not include a measure of phenotype stability over time. Furthermore, it is not clear whether, in most of these studies, therapy has been optimized based on guidelines and that adherence to treatment has been carefully assessed. The NIAID-funded Inner City Asthma Consortium (ICAC) designed this multi-center study with the primary objective to determine distinct characteristics that would discriminate difficult- from easy-to-control asthma in inner-city children. To achieve this goal, the study sought to enroll and evaluate children with a broad range of asthma severity. The design also aimed at addressing the above-identified deficiencies: the study involved prospective, multiple-visit evaluations of its participants for an entire year during which a standardized algorithm was used to optimize asthma management based on NAEPP Guidelines. In addition, given the known relationship between rhinitis and asthma,² this study included careful evaluation of rhinitis and an integrated rhinitis treatment plan also based on a standardized algorithm. Finally, substantial effort was devoted to participant adherence to both asthma and rhinitis treatments.

Methods

Study design

Asthma Phenotypes in the Inner City (APIC) was a prospective, longitudinal study of 6-17 year-old children with asthma living in nine urban areas. Participant selection criteria included residence in an area with >20% of residents below the poverty level, physician diagnosis of asthma, at least 2 episodes of short-acting beta-agonist administration within the past 12 months (exclusive of use associated with exercise-induced symptoms), and ≥50% adherence to controller medication between the screening and enrollment visits, as described below. Exclusion of those participants who did not achieve 50% adherence was done to limit the effect of poor adherence, so that response to treatment, rather than variation related to lack of treatment, could be studied. The study was approved by each clinical site's Institutional Review Board, and written informed consent was obtained from the parent or legal guardian of the child. The study design is illustrated in Figure 1.

Figure 1.

Schematic of APIC study design: Flow diagram illustrating: 1) screening visit; 2) clinic visits; 3) definition of easy-to-control, indeterminate and difficult-to-control groups based on controller treatment step throughout the study; and 4) domains used to group characteristics measured on each participant throughout the study.

Study evaluation and management

Participants received a comprehensive assessment at Screening and had their asthma and rhinitis managed based on NAEPP Expert Panel Report - 3 (EPR-3)³ and Allergic Rhinitis and its Impact on Asthma (ARIA) 2008⁴ guidelines-derived treatment algorithms with assessment and medication adjustments every 2 months for one year (V0, V1, V2, V3, V4, V5 and V6). To standardize evaluation and treatment across sites, clinical data about asthma and rhinitis were entered into a computer program that determined the treatment. Clinicians could override the computer-generated treatment recommendations if they deemed necessary. For asthma management, a predefined EPR-3 based treatment algorithm (Table E1 in the Online Repository) was used by the study clinicians to determine each participant's controller regimen (Table E2 in the Online Repository) based on asthma symptoms, spirometry results, and exacerbation history (Table E3 in the Online Repository). Asthma morbidity was assessed and scored using three domains from the Composite Asthma Severity Index (CASI).⁵ These domains were assessed at each study visit and included day symptoms and albuterol use in the past two weeks, night symptoms and albuterol use in the past 2 weeks and exacerbations in the past 2 months. Detailed scoring information related to these domains is shown in Table E4 in the Online Repository. For rhinitis management, an ARIA-derived algorithm was used to adjust medications based on reported severity (level of bother) and frequency of nasal symptoms, as well as on allergy status (Table E5 in the Online Repository). All study participants also received albuterol as rescue therapy and a supply of prednisone that could be used for relief of acute symptoms upon consultation with a study clinician. All asthma and rhinitis medications (except montelukast) were provided free of charge to the study participants.

Adherence to asthma therapy was assessed at each study visit for all participants requiring daily controller therapy (i.e. > Step 0). Calculation of used doses was performed after review of built-in dose counters at the visit or after receipt of devices in the mail if they were not brought to the visit. Percent adherence was determined based upon the number of doses taken divided by the number prescribed. A standardized questionnaire about adherence that was previously used in other ICAC studies was also administered at each visit. If the medications were not returned to the study site, percent adherence was calculated based upon questionnaire data. Questionnaire data were used to determine adherence in approximately 20% of study visits.

Study assessments

The visit schedule of study assessments used in this analysis is presented in Table E6 in the Online Repository. Prick skin testing was performed at Screening (unless results within one year from a prior ICAC study were available) for 12 common indoor and outdoor allergens (mouse epithelium, dog epithelium [mixed breeds], mite mix standardized [D. farinae and D. pteronyssinus mix], cat hair standardized, rat epithelium, American cockroach, German cockroach, Alternaria tenuis, Aspergillus fumigatus, ragweed mix [giant/short], Eastern 8 tree mix, K-O-T grass mix), with juniper and Bermuda grass being additionally tested at sites (Dallas and Denver) where those allergens are common. All allergen extracts were obtained from Greer Laboratories (Lenoir, North Carolina). Total serum IgE, Phadiatop inhalant allergy screen, fx5 food allergy screen, and allergen-specific IgE (sIgE) (ImmunoCAP, Phadia, Uppsala Sweden) to a panel of 20 allergens (egg white, milk, peanut, German cockroach, American cockroach, D. farinae, D. pteronyssinus, cat dander, dog dander, mouse urine proteins, rat urine proteins, Alternaria alternata, Aspergillus fumigatus, oak, pecan, birch, maple/box elder, Cladosporium herbarum, giant ragweed, timothy grass) were also measured from blood drawn at V0. Allergen sensitization was defined by a wheal ≥ 3mm larger than the saline control in the presence of a positive histamine control wheal of at least 3 mm on prick skin testing (a smaller histamine wheal invalidated skin tests from being used in the analysis) or specific IgE ≥ 0.35 kU/L for environmental and food allergens. Variables representing a positive skin prick test or positive sIgE were grouped together using a hierarchical clustering algorithm based on Pearson correlations.⁶ Using the result of this cluster analysis as a guide, the 20 sIgEs and the 12 skin test allergens were grouped into one the following 7 categories: roaches, pets, rodents, trees, foods, molds, and dust mites.

A complete blood count with differential was performed at V0 for blood eosinophil and neutrophil counts. The blood sample was also tested for total 25-hydroxyvitamin D.

Using a Jaeger spirometer, spirometry was performed at every management visit, and bronchodilator responsiveness was assessed at Screening and V6. Bronchodilator responsiveness was defined as percent change in FEV₁ as measured before and 15 minutes after 4 inhalations of albuterol. Training and supervision of the clinical site technicians was performed by pulmonary function specialists from the University of Wisconsin-Madison. Acceptability was based on the American Thoracic Society/European Respiratory Society (ATS/ERS) guidelines.⁷ For children under eight years of age, the ATS Preschool Guidelines were used.⁸

FeNO was measured at V0 and V6 using the NIOX MINO device (Aerocrine, Stockholm Sweden). Two maneuvers were performed, and if the results agreed within 5%, they were averaged. If they did not agree, a third maneuver was performed, and the two closest results were averaged.

The Asthma Control Test or childhood ACT (ACT/cACT) was administered at Screening and V6.^9-12

As a measure of household stress, perceived stress was assessed by administering the Perceived Stress Scale (PSS) to the child's primary caretaker at V0. The PSS is a 10-item scale that measures the degree to which respondents believe their lives were unpredictable, uncontrollable, and overwhelming in the preceding month (reliability=0.85).¹³

Obesity was measured at Screening by Body Mass Index (BMI) z-score based on CDC guidelines.¹⁴ Environmental tobacco exposure was evaluated at Screening and V6 by the reported number of smokers in the home and urinary cotinine, which was measured using the NicAlert™test (Nymox Pharmaceutical Corp).

The variables used in the analyses were identified in a literature review that was performed during protocol development to identify clinical and mechanistic factors that contribute to asthma severity. The study assessments served as a source of variables for analysis, which were subsequently grouped into a number of clinical domains as shown in Table E7 in the Online Repository.

Outcome definition

We defined the study groups a priori based upon asthma step level controller requirement. Difficult-to-control asthma was defined as requiring 500 mcg per day or more of fluticasone with/without long-acting beta-agonist, at four or more of the six post-baseline study visits. Of note, difficult-to-control asthma was not defined as uncontrolled or poorly controlled asthma. Participants requiring ≤ 100 mcg per day of fluticasone, montelukast only, or needing no controller medication, at four of the six post-baseline study visits were classified as having easy-to-control asthma. Those participants who did not fall into difficult- or easy-to-control groups were considered ‘indeterminate’. The decision to limit the analyses to participants who completed 4, 5 or 6 visits was pre-specified in the protocol to allow us to accurately classify difficult, easy and indeterminate participants over a 12 month period. All group assignments were made after study completion.

Statistical methods

A consensus process was used to determine 44 variables measured at Screening or V0 that were representative of the following domains (also shown in Table E7 in the Online Repository): demographics, family history, personal allergy history, asthma history, rhinitis, BMI, Vitamin D, environmental exposures, allergen sensitization, pulmonary physiology (FEV₁ bronchodilator responsiveness, FEV₁% predicted and FEV₁/FVC), allergic inflammation, and psychological stress. R version 3.2.2 was used for all analyses.

For each study group (difficult-to-control, easy-to-control, indeterminate), generalized additive mixed models were used to estimate the longitudinal pattern of treatment step throughout the study, as well as seasonal variation in symptoms, exacerbations, and pulmonary function.

To compare the 44 identified variables between study groups, analysis of variance and Kruskal-Wallis tests were used for continuous variables (normally distributed and non-normal, respectively) and chi-square or Fisher's exact tests were used for categorical variables. FEV₁ bronchodilator responsiveness, FEV₁% predicted and FEV₁/FVC were analyzed as continuous variables. Although not originally identified in the list of 44 variables, BMI percentile at Screening, controller treatment step at Screening and rhinitis medication score at Screening (categorized) were also compared. No adjustment was made to account for multiple comparisons arising from comparisons between groups or multiple variables.

Using the 44 identified variables, the Boruta variable selection algorithm was then used to determine the most relevant features distinguishing difficult- from easy-to-control asthma.¹⁵ Boruta, which adjusts for multiple comparisons and accounts for collinearity and potential interactions between variables, measures the contribution of each variable in accurately classifying an outcome by examining the effect of omitting it from the overall model. As a result, each variable is classified by the algorithm as confirmed, tentative or rejected with regard to its relevance in distinguishing difficult- from easy-to-control asthma. Prior to performing this analysis, a single imputation based on a fully conditional specification approach was performed on all variables to account for missing data.

Among difficult-to-control participants, a similar analysis was conducted to identify characteristics that were relevant to distinguishing participants with controlled asthma at V6 from participants with uncontrolled asthma. In this analysis, control status was defined as having an ACT/cACT score of 20 or higher at the last study visit.

Results

Between August 2011 and September 2013, 1195 children were recruited, 845 were eligible for screening, and 717 became part of the longitudinal cohort (Figure 2). Of these, 619 (86.3%) completed at least 4 of the 6 bimonthly visits for evaluation and management of asthma and rhinitis and were included in this analysis. Two hundred fifty-three (40.9%) participants met criteria for difficult-to-control asthma, 232 (37.5%) had easy-to-control asthma and 134 (21.6%) were indeterminate. Characteristics of the 485 participants in the difficult-and easy-to-control groups are presented in Table I. Males comprised 57.7% of the study population. Black, non-Hispanic participants represented 64.3% of the participants, followed by Hispanic participants (28.3%). The average age at Screening was 10.9 years. Characteristics of all 619 participants in the three groups, i.e. including the indeterminate group, are presented in Table E8 in the Online Repository. The majority of participants classified as indeterminate required variable treatment step levels during the year; only 18 participants consistently remained on the “intermediate” level of therapy of 200 mcg per day of fluticasone for 4 or more visits.

Figure 2.

CONSORT diagram: Flow diagram illustrating the number of individuals screened as well as the number of participants in the full APIC cohort and included in the analytic sample.

Table I.

Participant characteristics and demographics: Unless otherwise noted, characteristics are compared using chi-square or Fisher's exact test for categorical variables, and ANOVA or t-test for continuous variables. Unless otherwise noted, summary statistics are frequency (%) for categorical variables and mean and standard deviation for continuous variables.

Characteristic	N	All participants	Difficult	Easy	P-value
Demographics
Male	485	280 (57.7%)	138 (54.5%)	142 (61.2%)	0.16
Race1	484				0.07
Black (non-Hispanic)		311 (64.3%)	173 (68.7%)	138 (59.5%)
Hispanic		137 (28.3%)	67 (26.6%)	70 (30.2%)
Other/Mixed		26 (5.4%)	9 (3.6%)	17 (7.3%)
White (non-Hispanic)		10 (2.1%)	3 (1.2%)	7 (3.0%)
Age at Screening (years)	485	10.9 (3.0)	10.7 (3.0)	11.1 (3.1)	0.15
Family history
Family history of asthma	472	346 (73.3%)	186 (74.7%)	160 (71.7%)	0.54
Family history of hay fever/rhinitis/allergies	460	259 (56.3%)	133 (55.9%)	126 (56.8%)	0.92
Family history of eczema	465	263 (56.6%)	146 (60.1%)	117 (52.7%)	0.13
Allergy history
Self-report of food allergies	485	144 (29.7%)	77 (30.4%)	67 (28.9%)	0.78
History of anaphylaxis	485	35 (7.2%)	23 (9.1%)	12 (5.2%)	0.14
Eczema diagnosis	485	267 (55.1%)	150 (59.3%)	117 (50.4%)	0.06
Asthma history
Age asthma first diagnosed (years)	483	3.3 (3.2)	2.8 (2.9)	3.9 (3.4)	<0.001
Age asthma symptoms started -(years)	483	1.4 (2.6)	1.6 (2.7)	1.1 (2.4)	0.02
ACT/cACT2 at Screening	484				<0.001
Very poorly controlled		35 (7.2%)	28 (11.1%)	7 (3.0%)
Not well controlled		140 (28.9%)	103 (40.9%)	37 (15.9%)
Well controlled		309 (63.8%)	121 (48.0%)	188 (81.0%)
Controller treatment step at Screening3	485				<0.001
Steps 0-2		176 (36.3%)	23 (9.1%)	153 (65.9%)
Step 3		64 (13.2%)	27 (10.7%)	37 (15.9%)
Steps 4-6		245 (50.5%)	203 (80.2%)	42 (18.1%)
BMI
BMI z-score at Screening	485	1.0 (1.2)	1.1 (1.2)	0.9 (1.1)	0.14
BMI percentile at Screening3	485	87.3 [58.0;97.6]	88.3 [61.7;98.2]	85.5 [54.9;96.7]	0.06
Vitamin D
Total 25-hydroxyvitamin D at V0 (ng/mL)	475	19.1 (7.2)	18.7 (7.4)	19.5 (7.0)	0.25
Environmental Exposure
Any smokers in the home at Screening	482	195 (40.5%)	109 (43.4%)	86 (37.2%)	0.20
NicAlert result at Screening	476	1.3 (1.1)	1.3 (1.1)	1.3 (1.0)	0.90
Gas stove in home	475	311 (65.5%)	160 (64.3%)	151 (66.8%)	0.63
Dampness in home	473	168 (35.5%)	94 (37.8%)	74 (33.0%)	0.33
AC unit in child's bedroom	475	319 (67.2%)	170 (68.3%)	149 (65.9%)	0.66
Forced air for heat in home	476	288 (60.5%)	152 (60.8%)	136 (60.2%)	0.96
Rodents in home	475	142 (29.9%)	75 (30.1%)	67 (29.6%)	0.99
Roaches in home	473	82 (17.3%)	49 (19.8%)	33 (14.7%)	0.18
Pets in home	475	178 (37.5%)	99 (39.8%)	79 (35.0%)	0.33
Stress
Caretaker Perceived Stress Scale at V0	482	14.4 (7.5)	15.1 (7.6)	13.7 (7.3)	0.05
Allergen sensitization
Total serum IgE at V0 (kU/L)4	478	248.0 [91.0;765.2]	405.0 [101.0;1087.0]	197.0 [72.0;529.0]	<0.001
Number of allergen sensitizations (panel of 22)5 at V0	485	8.7 (6.2)	9.9 (6.3)	7.5 (5.9)	<0.001
Positive Phadiatop screen	476	374 (78.6%)	202 (81.5%)	172 (75.4%)	0.14
Positive fx5 food allergy screen	477	270 (56.6%)	154 (61.8%)	116 (50.9%)	0.02
Sensitized to molds6 at V0	485	240 (49.5%)	149 (58.9%)	91 (39.2%)	<0.001
Sensitized to dust mites7 at V0	485	287 (59.2%)	159 (62.8%)	128 (55.2%)	0.10
Sensitized to roaches8 at V0	485	272 (56.1%)	151 (59.7%)	121 (52.2%)	0.12
Sensitized to rodents9 at V0	485	208 (42.9%)	127 (50.2%)	81 (34.9%)	0.001
Sensitized to pets10 at V0	485	315 (64.9%)	176 (69.6%)	139 (59.9%)	0.03
Sensitized to pollen/peanut11 at V0	485	344 (70.9%)	197 (77.9%)	147 (63.4%)	0.001
Sensitized to foods12 at V0	479	143 (29.9%)	77 (30.7%)	66 (28.9%)	0.75
Inflammation
FeNO at V0 (ppb) 4	448	19.0 [11.0;35.5]	20.5 [12.0;38.2]	17.4 [10.7;32.0]	0.05
Blood eosinophil count at V0 (cells/mm3) 4	476	270.0 [168.0;440.0]	300.0 [190.0;500.0]	200.0 [115.0;400.0]	0.004
Blood neutrophil count at V0 (cells/mm3) 4	476	2900.0 [2000.0;3900.0]	3000.0 [2100.0;4100.0]	2700.0 [2000.0;3736.0]	0.03
Pulmonary physiology
Bronchodilator response at Screening	480	13.6 (14.1)	18.3 (15.9)	8.6 (9.6)	<0.001
FEV1 (% predicted) at Screening	485	91.4 (17.4)	86.7 (17.7)	96.6 (15.4)	<0.001
FEV1/FVC (×100) at Screening	475	77.5 (9.8)	74.2 (10.3)	81.0 (8.0)	<0.001
Rhinitis
Rhinitis classification at Screening	485				0.006
Allergic		333 (68.7%)	189 (74.7%)	144 (62.1%)
Non-Allergic		102 (21.0%)	46 (18.2%)	56 (24.1%)
No Rhinitis		50 (10.3%)	18 (7.1%)	32 (13.8%)
Rhinitis medication score13 at Screening	485	9.5 (6.2)	11.1 (5.6)	7.9 (6.3)	<0.001
Rhinitis medications at Screening3	485				<0.001
None		108 (22.3%)	37 (14.6%)	71 (30.6%)
Antihistamine only		67 (13.8%)	24 (9.5%)	43 (18.5%)
Nasal steroids only		72 (14.8%)	40 (15.8%)	32 (13.8%)
Antihistamine & nasal steroids		238 (49.1%)	152 (60.1%)	86 (37.1%)
Rhinitis symptom score14 at Screening	485	8.8 (6.0)	10.1 (6.2)	7.3 (5.3)	<0.001

¹Race is dichotomized (black vs. other) when establishing variable importance in distinguishing difficult- from easy-to-control asthma.

²ACT is categorized as very poorly controlled (≤15), not well controlled (≥16 & ≤19), and well controlled (≥20). cACT is categorized as very poorly controlled (≤12), not well controlled (≥13 & ≤19), and well controlled (≥20).

³Not included in list of variables to establish variable importance in distinguishing difficult- from easy-to-control asthma.

⁴Summarized using the median and inter-quartile range and tested using a Kruskal-Wallis test.

⁵Sensitization is based on a positive skin prick test and/or positive specific IgE to at least one of the following allergens: Alternaria tenuis (skin prick test) or Alternaria alternata (specific IgE), Aspergillus fumigatus (both skin prick test and specific IgE), Cladosporium herbarum (specific IgE only), Dermatophagoides farinae, Dermatophagoides pteronyssinus, German cockroach, American cockroach, mouse, rat, cat, dog, oak, pecan, birch, maple, Eastern 8 tree mix, ragweed mix (giant/short; skin prick test) or short ragweed (specific Ige), timothy grass, Kentucky Blue/June, Orchard and Timothy (K-O-T) grass mix, peanut, egg and milk.

⁶Sensitization is based on a positive prick skin test and/or positive specific IgE to at least one of the following allergens: Alternaria tenuis (skin prick test) or Alternaria alternata (specific IgE), Aspergillus fumigatus, and Cladosporium herbarum.

⁷Sensitization is based on a positive skin prick test and/or positive specific IgE to at least one of the following allergens: Dermatophagoides farinae and Dermatophagoides pteronyssinus.

⁸Sensitization is based on a positive skin prick test and/or positive specific IgE to at least one of the following allergens: German cockroach and American cockroach.

⁹Sensitization is based on a positive skin prick test and/or positive specific IgE to at least one of the following allergens: mouse and rat.

¹⁰Sensitization is based on a positive skin prick test and/or positive specific IgE to at least one of the following allergens: cat and dog.

¹¹Sensitization is based on a positive skin prick test and/or positive specific IgE to at least one of the following allergens: oak, pecan, birch, maple, Eastern 8 tree mix, ragweed, timothy grass, Kentucky Blue/June, Orchard and Timothy (K-O-T) grass mix, and peanut.

¹²Sensitization is based on a positive skin prick test and/or positive specific IgE to at least one of the following allergens: egg and milk.

¹³Rhinitis medication score is set to 0 for no medications, 5 for antihistamines only, 10 for nasal steroids only, and 15 for antihistamines and nasal steroids.

¹⁴Rhinitis symptom score is based on the Modified Rhinitis Symptom Utility Index (see Table E5 in the Online Repository).

The guidelines-directed asthma treatment step level of the difficult- and easy-to-control groups during the one-year follow-up period is shown in Figure 3. The assigned controller treatment step of the difficult-to-control group as a whole remained persistently elevated during the study, such that the mean controller treatment step was 4.72 (95% CI: (4.59, 4.85)) at Screening and 5.00 (4.86, 5.13) at V6. In contrast, the assigned controller treatment step for the easy-to-control group decreased, such that the mean controller treatment step was 1.81 (1.67, 1.94) at Screening and 0.76 (0.62, 0.91) at V6. This indicates that, in the context of the guidelines-based management, the easy-to-control group can improve over time, as they required gradually lower controller step level therapy while maintaining good control. In contrast, an important feature of the difficult-to-control group was the inability to taper controller therapy below Step 4 over the year. Of note, adherence to controller therapy was comparable between groups, with a median adherence of 78.4% in the difficult-to-control group, 76.2% in the easy-to-control group and 77.8% in the indeterminate group (p=0.26). While the phenotypes of difficult and easy-to-control were stable over time, they also appeared to be distinctly evident from the start of the study. Out of the 253 participants in the difficult-to-control group, 203 (80.2%) were at Steps 4-6 at Screening. Out of 212 participants in the easy-to-control group, 153 (65.9%) were at Steps 0-2 at Screening (shown in Table 1). The guidelines-directed asthma treatment step level of all 3 groups is presented in Figure E1 of the Online Repository.

Figure 3.

Controller treatment step for easy-to-control and difficult-to-control groups over 12 months. Green and blue lines (shaded areas) indicate mean (95% CI) of controller treatment step for easy-to-control and difficult-to-control groups respectively over 12 months.

The seasonal severity and patterns of conventionally assessed clinical parameters (day symptoms and albuterol use, night symptoms and albuterol use, FEV₁/FVC and frequency of asthma exacerbations) are presented in Figure 4. Compared to the easy-to-control group, the difficult-to-control group was characterized by relatively high and striking seasonally varying levels of asthma symptoms (fall and winter peaks; summer low) and exacerbations (fall and spring peaks; summer low). Persistent but variable airflow obstruction was also noted in the difficult-to-control group over the course of the study. Indeed, over the study's 12 months, difficult-to-control participants had more impaired and more variable lung function than easy-to-control participants. In the difficult-to-control group, the average within-participant mean of FEV₁ (% predicted) over 12 months was 88.9 (SD=14.8) while in the easy to control group, it was 97.7 (SD=12.7) (p<0.001). Alternatively, the average within-participant SD of FEV₁ (% predicted) was 9.7 (SD=4.8) in the difficult-to-control group and 6.0 (SD=3.0) in the easy-to-control group (p<0.001). Similarly, in the difficult-to-control group, the average within-participant mean of FEV₁/FVC (×100) over 12 months was 76.3% (SD=8.6) while in the easy-to-control group, it was 82.1% (SD=6.4) (p<0.001). Alternatively, the average within-participant SD was 5.4 (SD=2.7) in the difficult-to-control group while in the easy-to-control group, it was 3.1 (SD=1.6) (p<0.001). Notably, the clinical pattern of greater asthma severity in the difficult-to-control group occurred despite sustained use of highest step levels of guidelines-based asthma controller medication management over the course of the year. In contrast, the easy-to-control group demonstrated low levels of symptoms and exacerbations, and normal airflow by spirometry, without significant seasonal variation in symptoms or pulmonary physiology. While it appears that the easy-to-control group also exhibited spring and fall variation in asthma exacerbations, the rate of exacerbations was very low, fluctuating from 1-2%. The seasonal clinical data for all 3 groups can be found in Figure E2 of the Online Repository.

Figure 4.

Seasonal variation in clinical severity measures in easy-to-control, indeterminate and difficult-to-control groups. Green and blue lines (shaded areas) indicate mean or probability (95% CI) of clinical severity measure for easy-to-control and difficult-to-control groups respectively over 12 months. Clinical severity measures are: A) CASI component measuring day symptoms and albuterol use in the last two weeks (0 to 3 points); B) CASI component measuring night symptoms and albuterol use in the last two weeks (0 to 3 points); C) FEV₁/FVC (×100); and D) Monthly incidence of exacerbations, as defined by prednisone use.

Univariate comparisons (Table 1) between the difficult- and easy-to-control groups found that 22 out of 47 baseline variables differed. Specifically, difficult-to-control participants tended to have worse asthma control and lung function than easy-to-control participants, as well as higher levels of controller treatment, caretaker perceived stress, atopy, rhinitis severity, and inflammatory markers. To account for multiple comparisons, collinearity among variables and potential interactions, a variable importance analysis applied to 44 of the 47 baseline variables (shown in the left panel of Figure 5) found FEV₁ bronchodilator responsiveness to be the most relevant variable that distinguished difficult from easy-to-control asthma. Several additional variables obtained at Screening or V0 were confirmed as being relevant in distinguishing difficult from easy-to-control asthma, including ACT/cACT, FEV₁/FVC, rhinitis medication score, FEV₁ (% predicted), rhinitis symptom score, total serum IgE, sensitization to mold, BMI z-score, total number of allergen sensitizations and blood neutrophil count. Because lung function was one of the determinants of the treatment step level, which in turn was the basis for a participant to be categorized into one of three a priori defined groups, we repeated the variable importance analysis after removing the pulmonary physiology variables. As seen in the right panel of Figure 5, the variables identified as relevant in the first analysis remained relevant after removal of pulmonary function except for blood neutrophil count. Additionally, age at which asthma was first diagnosed and age of asthma symptom onset emerged as relevant distinguishing variables. Using an ACT/cACT score at V6 of 20 or higher as a cut-point for well-controlled asthma,^9,12 univariate analysis of the same 47 variables was performed to determine if there were any statistically significant differences at Screening or V0 between controlled and uncontrolled asthma sub-groups within the difficult-to-control group. Of the 253 participants in the difficult-to-control group, 239 had V6 ACT/cACT data for analysis; 178 (74.5%) had controlled asthma and 61 (25.5%) did not. Among 44 of the 47 variables, only the ACT/cACT at Screening was confirmed as being relevant to distinguishing these two subgroups.

Figure 5.

Variable importance plot of characteristics (with and without pulmonary physiology variables) distinguishing difficult-to-control from easy-to-control groups. Circles plot the median z-score for each variable, a measure of importance obtained from Boruta, a feature selection algorithm where higher values indicate a higher level of importance. Solid blue circles indicate variables that are confirmed as being relevant to distingushing difficult- from easy-to-control groups. Pink and green circle indicate variables that are tentative or rejected respectively and are not relevant to distingushing between difficult- from easy-to-control groups.

Discussion

This is the first study to date of a large population of children characterized prospectively while receiving guideline-based management for asthma and rhinitis. The unique and novel aspects of the study are its longitudinal assessment of participants to help characterize and establish the stability of the groups over time, inclusion of standardized, guidelines-based asthma and rhinitis assessments and computerized management algorithms, and collection of variables in a broad range of domains as reflected by the literature at the time of protocol development.

Clear clinical parameter differences were found at study entry between inner-city children with difficult-to-control asthma versus those with easy-to-control asthma. Of the 44 baseline variables that were used to establish variable importance in distinguishing difficult- from easy-to-control asthma, FEV₁ bronchodilator responsiveness was the most distinguishing characteristic. The Asthma Outcomes Workshop Report¹⁶ has recommended that assessment of bronchodilator responsiveness be used to characterize populations in clinical trials and to classify patients into clinically meaningful groups; our work provides important information that supports this endorsement. Because bronchodilator responsiveness is strongly related to baseline pulmonary function, it was not surprising that FEV₁/FVC and FEV₁ (% predicted) also differentiated these two groups.^17-20 A retrospective, cross-sectional study of low-income, mainly Hispanic children in Los Angeles found that bronchodilator responsiveness of at least 10% was significantly related to poor asthma control in controller-naïve children.²¹ Children with this degree of bronchodilator responsiveness, even in the face of normal spirometry, had more nocturnal symptoms, increased beta-agonist use and exercise limitation while also having a greater number of allergen sensitizations. Given the fact that pulmonary function was one of the determinants of our decision-making algorithm for increasing or decreasing the level of asthma controller treatment at each visit, which, in turn, determined whether a participant would be categorized as difficult or easy-to-control, the variable importance analysis was also performed after removing pulmonary physiology measures. Except for blood neutrophil count, all of the other parameters remained relevant in their ability to distinguish difficult-to-control asthma, with only two new emerging characteristics, age at asthma diagnosis and age at which asthma symptoms started.

The observation of the important contribution of rhinitis in the difficult-to-control asthma phenotype is noteworthy, highlighting the relevance of the upper airway in the heterogeneity of response to asthma therapy. Other studies have identified associations between severe asthma and rhinitis in both adults and children,^22-24 and some have suggested that treatment of allergic rhinitis with intranasal corticosteroids is associated with improved asthma control.²⁵ Our findings are consistent with the hypothesis that asthma and rhinitis constitute manifestations of the same disease in two parts of the respiratory tract² and reinforce the ARIA guidelines that rhinitis management should be incorporated as part of asthma management.^26,27

Markers of atopy, particularly total serum IgE, mold sensitization and the total number of allergen sensitizations also distinguished individuals with difficult-to-control asthma. Like rhinitis, atopy was not among the criteria on the basis of which asthma step therapy and consequently, group assignment, were determined. Although atopy and rhinitis outcomes had already emerged as relevant in the original variable importance analysis that we conducted, they remained relevant when pulmonary function was removed (Figure 5). Other studies have shown strong inverse associations between total IgE and asthma control^28,29 and our previous work has demonstrated an association between mold sensitization and increased asthma morbidity in inner-city children.³⁰ Interestingly, other established inner-city allergens, including cockroach³¹ and mouse,³² were not found to be relevant in this study. Most likely, this is because their relevance in disease is strongly linked to both sensitization and exposure. Allergen exposure outcomes were not included in this analysis because they were obtained only from a subgroup of participants.

BMI also emerged as a relevant characteristic distinguishing difficult- from easy-to-control groups. This finding is consistent with previous work that showed an association between obesity and increased morbidity and worsening pulmonary function in children with asthma,^33,34 as well as an adverse effect of obesity on corticosteroid responsiveness.^35,36

Age at asthma diagnosis and age at which asthma symptoms started both emerged as distinguishing factors between difficult and easy-to-control asthma. These factors have not been reported to be associated with the heterogeneity of response to asthma treatment.

Finally, our finding of blood neutrophil count as a factor that could differentiate difficult from easy-to-control was unexpected and its significance is unclear. This finding was no longer relevant when the variable importance analysis was performed without pulmonary measures Notably, blood eosinophil count did not prove to be a relevant discriminating characteristic.

During the 12-month longitudinal part of the study, the two groups (difficult- and easy-to-control) exhibited similarities and differences in clinical parameters measured over time. The major differences were a) the difficult-to-control group did not show any trend for improvement over one year despite guideline-based treatment, whereas the easy-to-control group did; b) the difficult-to-control group had higher peaks in exacerbation rates in the spring and fall and more daytime and nighttime symptoms in the fall and in winter. Notable similarities between the difficult and easy-to-control groups included comparable adherence rates to asthma therapy, lowest incidence of exacerbations in the summer and relatively constant pulmonary function, albeit at different levels, over the year.

The prospective nature of this study allowed us to partially examine the phenotypic stability of inner-city children with asthma. Given that the definition of our study groups included a stability criterion (to be within a specific range of therapy step levels in 4/6 study visits), the results presented in Figure 3 are not surprising. However, the fact that the easy-to-control group showed a progressive decline in medication requirements over the course of the study, whereas the difficult-to-control group did not, emphasizes the unique nature of these children's asthma. In support of this notion is the lack of any improvement in lung function that this group has exhibited, as well as the increased rate of nocturnal symptoms over the fall and winter months, which also constitutes a unique characteristic. The concept of nocturnal asthma as an indicator of poor asthma control was endorsed in the GINA report³⁷ and other investigators have previously demonstrated that nocturnal asthma is associated with more severe disease and greater bronchodilator responsiveness.²⁹ While the greater frequency of symptoms and exacerbations in the difficult-to-control group was expected, the disparity of seasonal variability in day and night symptoms and asthma exacerbations between the 2 groups was a surprising finding. A large retrospective study of children found similar seasonal variability, such that there were fewer symptoms in the summer and more symptoms from fall through spring, but did not link this variability with a specific asthma phenotype.³⁸

Our study has potential limitations. Adherence measurements may not accurately reflect actual use as they were based upon calculation of used doses from returned inhalers and/or a standardized questionnaire, which has not been validated. We acknowledge that pulmonary function was used in the determination of treatment step level and this could have enhanced the relative importance of spirometry in our findings. As a result we performed two analyses: when pulmonary physiology measures were removed, rhinitis, atopy and BMI continued to be relevant factors and age of asthma diagnosis and age of asthma symptoms onset both emerged as new relevant variables. Other factors determining treatment level requirements may exist. For the most part, APIC assessed variables that have been traditionally studied. The method used to measure stress may lack precision, which could potentially affect our findings. The domains of stress and environmental exposure were lacking comprehensive measurements that could more fully describe them. For example, we addressed caregiver-perceived stress, but not other potential sources of stress, such as neighborhood violence. New markers, particularly markers of immune cell, epithelial cell and airway smooth muscle cell biology, may be necessary to develop predictors of the response to treatment, as well as predictors of achieving asthma control. In a post-hoc analysis, conventional baseline outcomes aside from ACT/cACT did not appear to influence asthma control among difficult-to-control participants in this study.

In this large, prospective study, we found that heterogeneity of treatment response in children with asthma is associated with clinical characteristics that can be identified early in the course of treatment. Besides bronchodilator responsiveness, pulmonary physiology, rhinitis and atopy were major features that distinguished children requiring high dose asthma controller therapy. Assessments for clinicians and researchers to consider at initial evaluation are presented in Table 2. Our observations point to the importance of allergen sensitization and its sequelae in children with asthma. These findings also put into perspective how various domains may link together to describe patient phenotypes as well as potential pathways of disease.

Table II. Potential prioritization of assessments to identify difficult-to-control asthma.

FEV1 pre/post albuterol

ACT/cACT

Spirometry

Rhinitis severity1

Total serum IgE

Assessment of IgE to environmental allergens

BMI

¹See Table E5.

Supplementary Material

1

Figure E1: Controller treatment step for easy-to-control, indeterminate and difficult-to-control groups over 12 months. Green, red and blue lines (shaded areas) indicate mean (95% CI) of controller treatment step for easy-to-control, indeterminate and difficult-to-control groups respectively over 12 months.

Figure E2: Seasonal variation in clinical severity measures in easy-to-control, indeterminate and difficult-to-control groups. Green, red and blue lines (shaded areas) indicate mean or probability (95% CI) of clinical severity measure for easy-to-control, indeterminate and difficult-to-control groups respectively over 12 months. Clinical severity measures are: A) CASI component measuring day symptoms and albuterol use in the last two weeks (0 to 3 points); B) CASI component measuring night symptoms and albuterol use in the last two weeks (0 to 3 points); C) FEV₁/FVC (×100); and D) Monthly incidence of exacerbations, as defined by prednisone use.

Table E1: Asthma treatment adjustment based on control levels and adherence.

Table E2: Asthma controller treatment steps.

Table E3: Control levels of symptoms, bronchodilator usage, and FEV₁ (% personal best).

Table E4. CASI component scores.

Table E5. Rhinitis Treatment Adjustment Based on Frequency and Severity Outcomes from the Modified Rhinitis Symptom Utility Index (MRSUI)*

Table E6. Relevant study assessments by visit.

Table E7. Variables used in univariate and variable importance analyses. Unless otherwise specified, all variables were used in the variable importance analysis.

Table E8. Participant characteristics and demographics: Unless otherwise noted, characteristics are compared using chi-square or Fisher's exact test for categorical variables, and ANOVA or t-test for continuous variables. Unless otherwise noted, summary statistics are frequency (%) for categorical variables and mean and standard deviation for continuous variables.

Abbreviations

NIH

National Institutes of Health

NIAID

National Institutes of Allergy and Infectious Diseases

NAEPP

National Asthma Education and Prevention Program

EPR-3

Expert Panel Report-3

ARIA

Allergic Rhinitis and its Impact on Asthma

ICAC

Inner City Asthma Consortium

APIC

Asthma Phenotypes in the Inner City

ETS

Environmental Tobacco Smoke

CASI

Composite Asthma Severity Index

ACT/cACT

Asthma Control Test/Childhood Asthma Control Test

FeNO

Fractional exhaled Nitric Oxide

FEV₁

Forced Expiratory Volume in the first second

FEV₁/FVC

Ratio of Forced Expiratory Volume in the first second to forced vital capacity

IgE

Immunoglobulin E