Considering the control group: The influence of follow-up assessments on asthma symptoms

Abstract

Objective:

It is unclear whether research participation effects contribute to an improvement in asthma symptoms during clinical trials in the absence of any active intervention. We examined the impact of additional follow-up surveys on caregiver-reported symptoms among control subjects in a series of randomized controlled asthma trials.

Methods:

We analyzed baseline and follow-up data for children (3–10 yrs.) with poorly controlled persistent asthma that participated as control subjects in 1 of 3 randomized trials of urban school-based asthma care (study duration: 7–10 months). We compared mean symptom-free days (SFD) per 2 weeks between baseline and final follow-up; performed bivariate regressions to explore associations between demographics and changes in SFD; and performed multivariate random-effects generalized least square regression to examine the relationship between number of follow-ups beyond baseline (range: 1–10) and changes in SFD over time.

Results:

516 children were enrolled as controls across the 3 trials (mean age 7.5 yrs., 61% Black, 28% Hispanic, 81% Medicaid). Mean SFDs increased significantly from baseline to final follow-up (7.8 to 11.4 days, P<0.001). In adjusted analyses, significant improvements in SFD were observed with all follow-up contacts in comparison with baseline. Symptom improvement showed a dose-response relationship with the number of follow-up assessments completed (1, 2–3, 4–5, and 6–10 assessments).

Conclusions:

Children with uncontrolled asthma who participate as controls in clinical trials experience a significant increase in SFD with additional follow-up assessments. This improvement should be considered when designing/analyzing asthma interventions, and may help guide clinical outreach efforts for underserved children with persistent asthma.

INTRODUCTION

Uncontrolled asthma is a leading cause of hospitalization and emergency department visits for children in the United States.^{(1, 2)} Guideline-based⁽³⁾ treatment with daily controller medication has been associated with reduced acute healthcare utilization,^{(4, 5)} yet poor adherence with controller therapy is common and contributes to the high national rates of preventable morbidity.^{(6, 7)} Asthma morbidity disproportionately impacts poor and minority children,^{(8, 9)} who are also the least likely to receive or adhere with controller treatment.^{(10, 11)} Novel approaches to asthma management that effectively address barriers to care are necessary in order to improve outcomes and reduce disparities. As clinical researchers study new interventions, however, they should take caution to avoid overestimating the treatment effect: children who participate in asthma research trials may improve over time absent any specifically delivered intervention.

This was demonstrated by the School-Based Asthma Therapy (SBAT) trial,⁽¹²⁾ a randomized study in which controller medications were administered to children by directly observed therapy in schools. Subjects randomized to the SBAT intervention reported significantly more symptom-free days relative to children in the usual care (control) arm, yet control subjects also experienced considerable improvement over the course of the study. Some of this improvement is likely due to a regression towards the mean: subjects may be more symptomatic than usual when enrolled and return to their baseline of fewer symptoms over the study period.⁽¹³⁾ Many children additionally experience a seasonal fluctuation in symptoms, with a surge during the late summer that naturally aligns with the start of the academic year. ⁽¹⁴⁾ This spike is unavoidable in a study of school-based care in which children are enrolled in the first months of the school year, and provides context for control group improvement during subsequent follow-ups. The gradual increase in subject age over the course of a trial may also be associated with improved symptoms.⁽¹⁵⁾ By including a control group, the SBAT trial was designed to isolate efficacy of the intervention itself from these other causes for symptom improvement.

It is also worth considering whether subjects improved as a result of participating in the trial, and whether the study design unintentionally contributed to this improvement. Research participation effects, variably called the “Hawthorne effect” or “measurement effect,” refer to potential changes in subject behavior and outcomes as a result of being assessed.^{(16, 17)} For example, patients with dementia randomized to a bi-monthly follow-up schedule performed better on cognitive testing than patients randomized to fewer follow-ups despite an otherwise identical intervention; however, a practice effect could not be ruled out.⁽¹⁸⁾ Similarly, home water chlorination was significantly more likely to occur in rural Kenya when survey questions about water treatment were posed on a bi-weekly basis than a bi-annual basis, with an associated significant decrease in the incidence of diarrheal illness. SBAT subjects completed monthly symptom assessments. This raises important questions, with implications for how clinical asthma research is designed and analyzed: do follow-up assessments constitute an unplanned intervention for subjects enrolled in asthma trials? Further, is there a specific change in symptoms that can be quantified and predicted based on the number of completed assessments?

To answer these questions, we conducted a study with three objectives: 1) describe longitudinal changes in symptoms among children with poorly controlled asthma who participated as control subjects in randomized trials of asthma care; 2) identify demographic or study variables associated with symptom changes; and 3) quantify the impact of each additional follow-up on reported symptoms after adjusting for covariates. We hypothesized that an increasing number of follow-up assessments would be associated with greater improvement in asthma symptoms over time, and that the amount of improvement could be predicted based on the number of assessments completed.

METHODS

Participants

We analyzed baseline and follow-up survey data for children (3–10 years old) who were enrolled as control subjects in one of three randomized controlled trials of school-based asthma care: the School-Based Asthma Therapy (SBAT) trial,⁽¹²⁾ the School-Based Preventive Asthma Care Technology (SB-PACT) trial,⁽¹⁹⁾ or the School-Based Telemedicine Enhanced Asthma Management (SB-TEAM) trial.⁽²⁰⁾ A full description of methods and primary outcomes for each trial has been reported. Enrollment criteria were consistent across the three studies, which were conducted in partnership with the Rochester City School District in urban Rochester, NY. Children were eligible for recruitment during the first four months of the school year if they had physician diagnosed asthma and recent symptoms consistent with persistent or poorly controlled asthma as defined by national guidelines,⁽²¹⁾ based on caregiver-reported daytime symptoms, nighttime symptoms, rescue medication use, and activity limitation in the previous month. We also asked about the number of exacerbations requiring oral corticosteroids in the previous year. Key exclusion criteria included an inability to speak English, lack of any telephone access for follow-up assessments (landline or mobile), and if the child had another medical condition that could impact asthma assessment (i.e., congenital heart disease, cystic fibrosis).

The control condition of usual asthma care was also similar across the three trials, with a few minor differences such as the type of educational handouts provided to patients. Importantly, none of the children randomized to a control arm and included in the present analysis received an active asthma intervention. Six subjects participated in more than one study; however these children did not receive an active intervention in either study, waited at least 1 year after participation in the first study concluded before enrolling in the second study, and met all inclusion criteria at the start of each study. Accordingly, we included all data for these 6 individuals in the full analysis. The studies were all approved by the Institutional Review Board at the University of Rochester Medical Center.

Measures

The method of data collection was identical for all included studies. The primary outcome for each of the three original trials was caregiver reported symptom-free days (SFD), defined as the number of 24-hour periods in the two weeks (14 days) prior to baseline or follow-up assessment during which a subject did not experience any asthma symptoms. This measure is also the primary outcome for the current analysis. Additional symptom outcomes reported by caregivers in the original trials included the number of days or nights with asthma symptoms, activity limitation due to asthma, and use of short-acting beta agonist (SABA) rescue medication in the previous two weeks. Pulmonary function testing was not completed during baseline visits.

Caregivers provided information on child/caregiver demographics and the child’s asthma history (including current asthma medications) during baseline in-home surveys. We assessed depressive symptoms among caregivers using either the Kessler Psychological Distress Scale⁽²²⁾ (K10; used in SBAT, SB-PACT) or the Center for Epidemiological Studies - Depression Scale⁽²³⁾ (CES-D; used in SB-TEAM); caregivers were considered to have depressive symptoms if they had a positive screen using either instrument. We also measured caregiver asthma-related quality of life with the Pediatric Asthma Caregiver’s Quality of Life Questionnaire (PACQLQ);⁽²⁴⁾ asked about caregiver smoking status; and assessed child smoke exposure by measuring salivary cotinine levels in samples⁽²⁵⁾ collected at baseline. Blinded follow-up assessments were completed by telephone either every month (SBAT) or every other month (SB-PACT, SB-TEAM) for the duration of the school year, with the exception of final assessments completed in subject homes for SB-PACT and SB-TEAM. Depending on the specific study that a subject was enrolled in, this means that the number of planned follow-ups ranged from 4 (every other month) to 10 (every month).

Statistical Analysis

Following a descriptive analysis of both the combined and component control group populations, we compared SFD and additional symptom outcomes between baseline and final follow-up assessments using paired T-tests. Next, in order to account for autocorrelation associated with serial assessments of children over time, we performed bivariate generalized least squares regressions to compare changes in reported symptoms with a number of child, caregiver, and study variables. We structured these analyses as repeated observations within individuals, in order to limit missing data for follow-up assessments that were not completed by all subjects. We utilized a feasible generalized least squares regression model as this approach has been found to provide consistent and unbiased estimates despite unbalanced panels, in which participants have unequal numbers of follow-ups.⁽²⁶⁾ Each child or caregiver variable included in bivariate analyses has been previously established as a potential factor in childhood asthma control. Study variables included season of enrollment, the number of follow-up assessments completed, and the original study that subjects participated in.

We then constructed multivariate generalized least squares regression models to examine the relationship between the number of follow-up assessments and changes in symptoms over time (reference: baseline assessment). Independent variables were included as potential covariates in multivariate models if bivariate comparisons yielded a p-value of ≤0.1. We decided a priori to include original study group in the multivariate analyses regardless of bivariate association, in order to account for differences in the control condition and periodicity of follow-up between the three randomized trials. Furthermore, we included patient age (in days) at each assessment date in all multivariate models to account for variability in the time between follow-ups. Model specification was assessed with the Hausman test and link test.

Finally, we performed additional multivariate generalized least squares regression analyses to examine whether caregiver report of recent symptoms significantly changed over the course of subsequent assessments. In this series of analyses, each follow-up assessment sequentially served as the reference for comparing symptom outcomes at all other assessments. For example, we compared the change in SFD at the first follow-up (reference: baseline) with the change in SFD at the every other follow-up for which repeated observation data was available. Then, we compared the change in SFD at the second follow-up (reference: baseline) with the change in SFD at the first follow-up and all subsequent follow-ups. This approach allowed us to group number of assessments with statistically similar improvements in SFD compared to baseline by magnitude of impact, in order to identify the minimum number of contacts required to achieve the maximum effect. All analyses were performed using Stata 12.0 (StataCorp, College Station, Tex). A 2-sided alpha <0.05 was considered statistically significant.

RESULTS

Combined control group

Across the 3 original studies, 1384 individuals met eligibility criteria and 1030 were enrolled (74%). Baseline measurements were available for the 516 subjects randomized to a control condition (Table 1), and final follow-up measurements were available for 482 individuals (93% of control subjects). Subjects in the combined control group participated for an average of 7.3 months (SD 1.1) and completed 5.7 assessments (SD 2.2). Enrolled children were 7.5 years old on average; 61% were Black, 28% were Hispanic, and 41% were female. Most were covered by public health insurance, and a current prescription for at least one preventive medication was reported by two-thirds of caregivers at baseline. The mean age of caregivers was 34 years old. One in four caregivers reported a marriage or domestic partnership, while three in four reported having a level of education greater than or equal to high school. More than one third of the caregivers had depressive symptoms, and 40% reported current cigarette smoking.

Table 1.

Baseline demographics of control subjects enrolled in one of the three randomized controlled trials^a

	Overall	SBAT	SB-PACT	SB-TEAM
Study Variables
Number of subjects at baseline	516	265	51	200
Number of subjects at final follow-up	482 (93%)	249 (94%)	47 (92%)	186 (93%)
Duration of Study (months)	7.3 (1.1)	7.6 (0.9)	5.5 (0.7)	7.5 (1.0)
Number of Completed Follow-Up Contacts	5.7 (2.2)	7.5 (1.7)	3.8 (0.6)	3.8 (0.6)
Child Variables
Age (Years)	7.5 (1.9)	7.2 (1.9)	7.0 (1.8)	7.9 (1.7)
Race
White	36 (7%)	21 (8%)	4 (8%)	11 (6%)
Black	312 (61%)	168 (63%)	26 (51%)	118 (59%)
Other	168 (33%)	76 (29%)	21 (41%)	71 (36%)
Ethnicity (Hispanic)	143 (28%)	69 (26%)	14 (28%)	60 (30%)
Gender (Female)	213 (41%)	118 (45%)	19 (37%)	76 (38%)
Public Health Insurance	418 (81.0%)	228 (86%)	41 (80%)	149 (75%)
Preventive Medication Use	343 (67%)	178 (67%)	27 (53%)	138 (69%)
Caregiver/Home Variables
Age (Years)	34.2 (8.5)	34.5 (9.1)	33.6 (9.6)	34.0 (7.3)
Relationship Status: Married/Domestic Partner	130 (25%)	68 (26%)	17 (33%)	45 (23%)
Education: ≥ GED/High School Graduate	377 (73%)	191 (72%)	36 (71%)	150 (75%)
Depressive Symptoms	188 (36%)	105 (40%)	24 (47%)	59 (30%)
Quality of Life	5.6 (1.2)	5.4 (1.3)	5.8 (1.2)	5.8 (1.1)
Smoke Exposure
Current Smoker	204 (40%)	110 (42%)	23 (45%)	71 (36%)
Quit Smoking in Past Yearb	30 (9%)	15 (9%)	4 (14%)	11 (9%)
Cotinine	1.7 (2.6)	1.6 (2.8)	2.9 (3.4)	1.5 (2.0)
Primary Language Used at Home
English	440 (85%)	222 (84%)	46 (90%)	172 (86%)
Spanish	13 (2%)	6 (2%)	0 (0.0%)	7 (4%)
English and Spanish	60 (12%)	35 (13%)	5 (10%)	20 (10%)
Other	3 (1%)	2 (1%)	0 (0%)	1 (0%)

^aData presented as n (%) or mean (SD)

^bQuestion only posed to caregivers who are not current smokers

Primary outcome: Symptom-free days

Overall, control subjects experienced a significant increase in symptom-free days (SFD) per two week period between baseline and final follow-up assessments (7.8 SFD vs. 11.4 SFD, P<0.001), as well as an improvement in all of the other symptom measures (Table 2). In bivariate regression analyses, we identified statistically significant associations between SFD and multiple independent variables, including original study group, child age at each follow-up, follow-up assessment number, public health insurance at baseline, caregiver report of a preventive medication at baseline, caregiver depressive symptoms and quality of life at baseline, and elevated salivary cotinine at baseline (Table 3). We included each of these variables in the multivariate regression analyses evaluating symptom free days over time. Of note, we did not identify an association between season of enrollment and reported symptoms.

Table 2.

Changes in caregiver reported symptoms (over the past two weeks) from baseline to final follow-up^a

	Baseline	Final Follow-Up	P-Value
Outcome Variable
Symptom-Free Days	7.8 (4.9)	11.4 (3.8)	<0.001
Symptom Days	4.1 (4.3)	1.8 (3.1)	<0.001
Symptom Nights	3.6 (4.2)	1.5 (2.7)	<0.001
Days with Activity Limitation	3.0 (3.9)	1.4 (2.9)	<0.001
Days with SABA use	3.8 (4.5)	2.1 (3.5)	<0.001

^aData presented as mean (SD); N=482

Table 3.

Changes in caregiver reported symptoms over the past two weeks (bivariate regressions, generalized least squares)^a

	Symptom-Free Days	Symptom Days	Symptom Nights	Activity Limitation	SABA Use
Study Variables
Study Group (Ref: SBAT)
SB-PACT	−1.07 (−1.98, −0.16)*	0.39 (−0.20, 0.99)	0.40 (−0.23, 1.03)	0.42 (−0.27, 1.10)	−0.09 (−0.84, 0.67)
SB-TEAM	−0.76 (−1.29, −0.22)**	0.56 (0.17, 0.96)**	−0.32 (−0.71, 0.06)	0.28 (−0.09, 0.65)	0.15 (−0.29, 0.59)
Follow-Up Number (Ref: Baseline)	(n = 516)	(n = 515)	(n = 516)	(n = 515)	(n = 516)
1	2.46 (1.99, 2.93)***	−1.66 (−2.08, −1.24)***	−1.40 (−1.80, −1.01)***	−1.02 (−1.40, −0.65)***	−0.98 (−1.42, −0.54)***
	(n = 510)	(n = 509)	(n = 509)	(n = 509)	(n = 509)
2	3.02 (2.54, 3.50)***	−1.84 (−2.26, −1.42)***	−1.65 (−2.06, −1.24)***	−1.33 (−1.71, −0.95)***	−1.62 (−2.05, −1.18)***
	(n = 501)	(n = 499)	(n = 499)	(n = 498)	(n = 500)
3	2.89 (2.38, 3.39)***	−1.89 (−2.34, −1.43)***	−1.63 (−2.05, −1.21)***	−1.25 (−1.63, −0.87)***	−1.36 (−1.82, −0.90)***
	(n = 493)	(n = 492)	(n = 493)	(n = 491)	(n = 492)
4	3.51 (3.04, 3.99)***	−2.24 (−2.67, −1.82)***	−2.12 (−2.51, −1.72)***	−1.62 (−2.00, −1.25)***	1.94 (−2.37, −1.51)***
	(n = 468)	(n = 467)	(n = 468)	(n = 468)	(n = 468)
5	3.84 (3.31, 4.37)***	−2.37 (−2.85, −1.90)***	−2.22 (−2.68, −1.77)***	−1.67 (−2.08, −1.26)***	−1.89 (−2.41, −1.36)***
	(n = 241)	(n = 241)	(n = 241)	(n = 241)	(n = 241)
6	4.23 (3.72, 4.75)***	−2.69 (−3.15, −2.22)***	−2.56 (−3.02, −2.09)***	−1.87 (−2.27, −1.46)***	−2.03 (−2.53, −1.54)***
	(n = 234)	(n = 234)	(n = 234)	(n = 234)	(n = 233)
7	4.04 (3.49, 4.59)***	−2.62 (−3.10, −2.13)***	−2.45 (−2.94, −1.96)***	−1.82 (−2.24, −1.40)***	−2.01 (−2.54, −1.48)***
	(n = 222)	(n = 221)	(n = 222)	(n = 222)	(n = 221)
8	4.16 (3.58, 4.74)***	−2.56 (−3.10, −2.02)***	−2.57 (−3.06, −2.08)***	−1.64 (−2.13, −1.16)***	−2.05 (−2.63, −1.46)***
	(n = 158)	(n = 158)	(n = 158)	(n = 158)	(n = 158)
9	3.94 (3.19, 4.68)***	−2.60 (−3.31, −1.89)***	−2.24 (−2.93, −1.55)***	−1.66 (−2.28, −1.04)***	−1.59 (−2.38, −0.80)***
	(n = 68)	(n = 68)	(n = 68)	(n = 68)	(n = 68)
10	4.13 (2.02, 6.24)***	−1.94 (−4.22, 0.33)***	−2.49 (−4.17, −0.81)***	−1.37 (−3.25, 0.51)	−1.13 (−3.08, 0.83)
	(n = 8)	(n = 8)	(n = 8)	(n = 8)	(n = 8)
Season at Baseline (Ref: Summer)
Fall	0.68 (−1.36, 2.73)	−0.08 (−1.78, 1.62)	−0.51 (−2.29, 1.27)	−0.52 (−2.65, 1.61)	−0.53 (−2.54, 1.48)
Winter	−0.07 (−2.26, 2.11)	0.05 (−1.72, 1.82)	−0.22 (−2.06, 1.62)	0.05 (−2.18, 2.27)	−0.77 (−2.85, 1.31)
Child Variables
Age at Each Follow-Up (Years)	0.40 (0.27, 0.53)***	−0.18 (−0.28, −0.08)***	−0.29 (−0.39, −0.18)***	−0.16 (−0.25, −0.06)**	−0.19 (−0.30, −0.08)***
Race (Reference = White)
Black	−0.02 (−1.06, 1.02)	−0.19 (−1.12, 0.75)	0.08 (−0.59, 0.76)	0.14 (−0.56, 0.85)	0.19 (−0.72, 1.11)
Other	−0.20 (−1.28, 0.88)	0.03 (−0.93, 1.00)	0.14 (−0.58, 0.86)	0.19 (−0.54, 0.93)	0.05 (−0.90, 0.99)
Ethnicity (Hispanic)	−0.18 (−0.74, 0.39)	0.22 (−0.20, 0.64)	0.09 (−0.34, 0.52)	0.01 (−0.37, 0.40)	0.10 (−0.36, 0.56)
Gender (Female)	0.07 (−0.43, 0.56)	−0.08 (−0.45, 0.29)	0.03 (−0.34, 0.39)	−0.13 (−0.47, 0.22)	−0.16 (−0.58, 0.25)
Public Health Insurance	−0.71 (−1.24, −0.19)**	0.32 (−0.09, 0.72)	0.73 (0.36, 1.09)***	0.63 (0.28, 0.98)***	0.74 (0.28, 1.20)**
Preventive Medication Use at Baseline	−0.58 (−1.11, −0.05)*	0.66 (0.29, 1.03)***	0.46 (0.07, 0.86)*	0.53 (0.17, 0.89)**	0.94 (0.52, 1.35)***
Caregiver/Home Variables
Age at Baseline (Years)	0.00 (−0.04, 0.03)	0.01 (−0.02, 0.03)	0.00 (−0.02, 0.03)	0.01 (−0.02, 0.04)	0.01 (−0.01, 0.04)
Relationship Status: Married/Domestic Partner	0.41 (−0.12, 0.93)	−0.17 (−0.57, 0.24)	−0.18 (−0.58, 0.22)	−0.09 (−0.48, 0.31)	−0.51 (−0.93, −0.09)*
Education: ≥ GED/High School Graduate	0.28 (−0.30, 0.87)	0.05 (−0.37, 0.47)	−0.35 (−0.79, 0.09)	−0.23 (−0.65, 0.19)	−0.31 (−0.81, 0.18)
Depressive Symptoms at Baseline	−0.78 (−1.31, −0.25)**	0.38 (−0.01, 0.77)	0.59 (0.19, 0.99)**	0.35 (−0.02, 0.72)	0.49 (0.05, 0.94)*
Quality of Life at Baseline	0.62 (0.42, 0.83)***	−0.41 (−0.57, −0.26)***	−0.50 (−0.66, −0.35)***	−0.56 (−0.72, −0.41)***	−0.55 (−0.73, −0.37)***
Smoke Exposure at Baseline
Current Smoker	0.12 (−0.39, 0.62)	−0.06 (−0.44, 0.31)	−0.35 (−0.71, 0.01)	0.16 (−0.21, 0.52)	0.05 (−0.38, 0.48)
Quit Smoking in Past Year	−0.69 (−1.72, 0.34)	0.49 (−0.34, 1.32)	0.31 (−0.46, 1.07)	0.52 (−0.37, 1.40)	0.84 (−0.26, 1.95)
Cotinine	−0.10 (−0.19, −0.01)***	0.08 (0.02, 0.14)**	0.08 (0.02, 0.14)**	0.09 (0.03, 0.15)**	0.09 (0.01, 0.17)*
Primary Language Used at Home (Ref: English)
Spanish	0.03 (−1.28, 1.35)	0.30 (−0.79, 1.38)	0.01 (−0.84, 0.86)	−0.41 (−1.24, 0.41)	0.28 (−0.78, 1.34)
English and Spanish	−0.17 (−0.99, 0.65)	0.22 (−0.42, 0.85)	0.12 (−0.48, 0.73)	0.14 (−0.42, 0.71)	−0.25 (−0.85, 0.35)
Other	0.04 (−2.84, 2.92)	1.18 (−1.49, 3.86)	0.04 (−2.18, 2.26)	0.64 (−3.23, 4.52)	0.55 (−2.31, 3.41)

^aData presented as Coefficient (95% C.I.)

^*P ≤ 0.05

^**P ≤ 0.01

^***P ≤ 0.001

In multivariate analyses, public health insurance and caregiver report of a preventive medication at baseline were significantly associated with a decrease in symptom-free days over time (Table 4), while higher caregiver quality of life score at baseline was associated with an increase in SFD. The number of completed follow-up assessments was strongly predictive of an improvement in SFD over baseline. By the time of the first follow-up, caregivers reported an additional 2.40 symptom-free days per two week period (95% CI 1.93, 2.88; P<0.001). There appeared to be a trend of increasing SFD with subsequent follow-ups: for example, by the fourth follow-up, caregivers were reporting an additional 3.49 SFD compared with baseline (95% CI 3.01, 3.97; P<0.001), or one additional 24 hour period without symptoms compared with the first follow-up.

Table 4.

Changes in caregiver reported symptoms over the past two weeks (multivariate regressions, generalized least squares)

	Symptom-Free Days	Symptom Days	Symptom Nights	Activity Limitation	SABA Use
Study Variables
Study Group (Ref: SBAT)
SB-PACT	−0.70 (−1.5, 0.11)	0.19 (−0.36, 0.75)	0.11 (−0.46, 0.67)	0.44 (−0.15, 1.03)	−0.05 (−0.71, 0.61)
SB-TEAM	−0.50 (−1.04, 0.03)	0.33 (−0.07, 0.74)	−0.51 (−0.92, −0.09)*	0.26 (−0.12, 0.64)	0.08 (−0.36, 0.53)
Follow-Up Number (Ref: Baseline)	(n = 509)	(n = 508)	(n = 509)	(n = 508)	(n = 509)
1	2.40 (1.93, 2.88)***	−1.62 (−2.04, −1.20)***	−1.36 (−1.76, −0.96)***	−0.96 (−1.33,−0.59)***	−0.93 (−1.37, −0.49)***
	(n = 504)	(n = 503)	(n = 503)	(n = 503)	(n = 503)
2	2.97 (2.48, 3.45)***	−1.79 (−2.21, −1.36)***	−1.61 (−2.02, −1.20)***	−1.28 (−1.66, −0.89)***	−1.57 (−2.01, −1.13)***
	(n = 496)	(n = 494)	(n = 494)	(n = 493)	(n = 495)
3	2.81 (2.31, 3.32)***	−1.84 (−2.30, −1.39)***	−1.57 (−2.00, −1.15)***	−1.19 (−1.57, −0.81)***	−1.32 (−1.78, −0.86)***
	(n = 489)	(n = 488)	(n = 489)	(n = 487)	(n = 488)
4	3.49 (3.01, 3.97)***	−2.25 (−2.67, −1.82)***	−2.10 (−2.50, −1.70)***	−1.61 (−1.99, −1.23)***	−1.96 (−2.39, −1.52)***
	(n = 464)	(n = 463)	(n = 464)	(n = 464)	(n = 464)
5	3.76 (3.23, 4.30)***	−2.33 (−2.81, −1.85)***	−2.29 (−2.76, −1.82)***	−1.63 (−2.05, −1.22)***	−1.91 (−2.44, −1.37)***
	(n = 241)	(n = 241)	(n = 241)	(n = 241)	(n = 241)
6	4.15 (3.63, 4.68)***	−2.64 (−3.11, −2.17)***	−2.61 (−3.10, −2.13)***	−1.83 −2.24, −1.42)***	−2.05 −2.55, −1.55)***
	(n = 234)	(n = 234)	(n = 234)	(n = 234)	(n = 233)
7	3.95 (3.39, 4.52)***	−2.57 (−3.07, −2.07)***	−2.50 (−3.01, −1.99)***	−1.78 (−2.22, −1.35)***	−2.03 (−2.58, −1.48)***
	(n = 222)	(n = 221)	(n = 222)	(n = 222)	(n = 221)
8	4.07 (3.48, 4.66)***	−2.51 (−3.06, −1.97)***	−2.62 (−3.13, −2.12)***	−1.62 (−2.11, −1.12)***	−2.06 (−2.66, −1.47)***
	(n = 158)	(n = 158)	(n = 158)	(n = 158)	(n = 158)
9	3.84 (3.09, 4.58)***	−2.54 (−3.25, −1.84)***	−2.28 (−2.97, −1.59)***	−1.62 (−2.24, −1.00)****	−1.60 (−2.38, −0.81)***
	(n = 68)	(n = 68)	(n = 68)	(n = 68)	(n = 68)
10	4.03 (1.88, 6.19)***	−1.89 (−4.19, 0.40)	−2.58 (−4.24, −0.92)**	−1.39 (−3.25, 0.47)	−1.13 (−3.17, 0.91)
	(n = 8)	(n = 8)	(n = 8)	(n = 8)	(n = 8)
Constant	4.78 (2.99, 6.58)***	6.17 (4.80, 7.54)***	6.38 (5.11, 7.66)***	5.65 (4.34, 6.96)***	6.05 (4.47, 7.63)***
Child Variables
Age at Each Follow-Up (Years)	0.09 (−0.03, 0.21)	−0.03 (−0.12, 0.06)	−0.08 (−0.18, 0.02)	−0.01 (−0.10, 0.08)	−0.03 (−0.13, 0.07)
Public Health Insurance	−0.65 (−1.13, −0.17)**	a	0.56 (0.21, 0.92)**	0.47 (0.16, 0.79)**	0.55 (0.14, 0.97)**
Preventive Medication Use at Baseline	−0.57 (−1.07, −0.06)*	0.62 (0.26, 0.98)***	0.37 (−0.01, 0.75)	0.44 (0.12, 0.77)**	0.85 (0.45, 1.25)***
Caregiver/Home Variables
Relationship Status: Married/Domestic Partner	b	b	b	b	−0.43 (−0.83, −0.03)*
Depressive Symptoms at Baseline	−0.21 (−0.78, 0.36)	0.03 (−0.38, 0.45)	0.18 (−0.24, 0.61)	−0.22 (−0.57, 0.14)	0.01 (−0.45, 0.46
Quality of Life at Baseline	0.64 (0.41, 0.87)***	−0.44 (−0.61, −0.27)***	−0.45 (−0.62, −0.28)***	−0.61 (−0.77, −0.44)***	−0.52 (−0.72, −0.33)***
Smoke Exposure at Baseline
Current Smoker	b	b	−0.56 (−0.93, −0.20)**	b	b
Cotinine	−0.05 (−0.12, 0.02)	0.05 (0.00, 0.10)*	0.08 (0.02, 0.14)**	0.06 (0.00, 0.11)*	0.05 (−0.02, 0.13)

^aData presented as Coefficient (95% C.I.)

^bNot included in multivariate model due to lack of significance in bivariate regression model

^*P ≤ 0.05

^**P ≤ 0.01

^***P ≤ 0.001

We performed additional multivariate regression analyses to determine whether the observed improvements in SFD over subsequent follow-ups were significant. In this series of models, each follow-up was sequentially used as the reference group for the other follow-ups that either preceded or followed. We identified four tiers of significant improvement in SFD based on the number of follow-ups completed (Figure 1). Caregiver-reported SFD were significantly higher at the first follow-up than at baseline (coefficient 2.40; 95% CI 1.93, 2.87). Improvements in SFD were statistically similar for 2–3 follow-ups (combined coefficient 2.89; 95% CI 2.44, 3.34) and 4–5 follow-ups (combined coefficient 3.58; 95% CI 3.13, 4.03), and maximized after 6 follow-ups with an average of 4.02 additional SFD compared with baseline (95% CI 3.54, 4.51) after controlling for child age at each follow-up, health insurance, preventive medications at baseline, and caregiver quality of life (data not shown).

Figure 1. Mean Symptom-Free Days Reported by Caregivers (Previous 2 Weeks).

Improvement in mean SFD compared with baseline. In this series of multivariate regression analyses, each follow-up was sequentially used as the reference for comparing all other follow-ups. Follow-ups were grouped for analysis when SFD were statistically similar between them.

Additional symptom outcomes

We identified similar associations between independent variables and the other four symptom outcomes using bivariate regression analyses, with the following exceptions: public health insurance was not associated with symptom days; current caregiver smoking had a non-significant association with symptom nights (P=0.056); caregiver depressive symptoms had a non-significant association with symptom days (P=0.054) and days with activity limitation (P=0.063) and caregiver marital status was associated with SABA use (Table 3). Our multivariate regression models for each additional symptom outcome included or excluded these potential covariates accordingly.

In multivariate analyses, the number of completed follow-up assessments were mostly predictive of improvements in each additional symptom outcome when compared with baseline values (Table 4); the 10th assessment was only associated with a reduction in symptom nights. As above, we iteratively ran additional regression models in which each follow-up served as the reference for all others. Each of these other symptom outcomes was associated with two or three tiers of improvement as the number of follow-up assessments increased (Figure 2).

Figure 2. Mean Change in Additional Symptom Outcomes Reported by Caregivers (Previous 2 Weeks).

Improvement in mean outcomes compared with baseline: (A) symptom days; (B) symptom nights; (C) activity limitation; (D) SABA use. In multivariate regression analyses, each follow-up was sequentially used as the reference for comparing all other follow-ups. Follow-ups were grouped for analysis when outcomes were statistically similar between them.

DISCUSSION

Children with uncontrolled persistent asthma who participate as control subjects in clinical asthma trials may experience improvements in symptoms over time, despite not having received an active intervention. Asthma control among young, inner-city children has been shown to fluctuate over a 3 to 6 month period;⁽²⁷⁾ increasing child age, seasonality, and a regression to the mean are among the previously identified reasons for symptom abatement. By demonstrating that follow-up assessments are also associated with significant increases in caregiver-reported symptom-free days, this study highlights the need to consider additional factors contributing to observed longitudinal improvement in asthma symptoms. It is notable that the follow-up number remained strongly associated with an increase in SFD after controlling for differences between parent studies (i.e., control conditions and time in trial between assessments), increasing child age, and other potential confounding variables. Further, these findings indicate that there may be a predictable degree of caregiver-reported improvement in SFD depending on the number of follow-up assessments that are completed.

The clear link between subsequent follow-up assessments and symptom improvement strongly suggests a research participation effect (RPE), or changes in participant behavior and/or outcomes attributable to the survey process.⁽²⁸⁾ A recent systematic review supports the existence of RPE,⁽¹⁶⁾ yet conclusions about size and impact are limited by heterogeneity between studies. Participation effects may be situationally dependent, and only occur when a causal pathway exists between the survey topic and the outcome of interest.⁽²⁹⁾ For children with uncontrolled asthma, increased adherence with controller medications or other self-management behaviors may explain the connection between survey assessments and improved symptoms. Patient perception of asthma as an intermittent problem rather than a chronic disease is a well-described barrier to adherence with controller medications.^{(30, 31)} The monthly/bi-monthly shared focus on asthma during follow-up surveys in the original studies might reframe asthma as a chronic process, and prompt caregivers to modify their approach to management. Qualitative data from low-income urban adults who participated as control subjects in an asthma trial support this contention: study participation, including periodic assessments, influenced subjects’ perceived understanding of asthma and the importance of medication adherence.⁽³²⁾ However, we do not have detailed information about adherence and self-management among control subjects in the trials included in this analysis, and we are therefore unable to draw firm conclusions about behavior changes.

Our findings have important implications for clinical research. If a specific degree of improvement in asthma symptoms can be anticipated based on the number of follow-up assessments completed, then sample size calculations should incorporate this improvement into the expected effect size. Analyses of pre-post trials should similarly consider the potential role of follow-up assessments in asthma symptom improvement, particularly when a comparison group is not available. Asthma intervention studies might benefit from designs with fewer follow-ups in order to limit any participation effect. While fewer follow-ups would additionally save resources, however, any reduction in measurement frequency would need to be balanced with the need for maintaining routine contact with subjects to limit attrition.

These results also suggest opportunities to reconsider how routine ambulatory care is provided for children with persistent asthma. The American Academy of Pediatrics advises primary care practices to implement a patient-centered medical home (PCMH) model of care, which includes patient outreach and care coordination services.⁽³³⁾ If asking patients/caregivers about asthma symptoms on a regular basis is associated with symptomatic improvement, it bears questioning whether routine follow-up calls about asthma from a PCMH team member might similarly improve outcomes on a broader scale. It should be noted that the control subjects in this analysis continued to experience asthma symptoms; even after the observed improvement, control subjects reported only 11.4 symptom-free days in a 14 day span. Thus telephone follow-up should not be considered sufficient for high-risk children with asthma, but may have value as part of a comprehensive management approach. Additional studies will be required in order to determine whether such a strategy would be associated with similar symptom-based improvement, lead to a reduction in acute healthcare utilization, or be cost-effective.

Strengths of this analysis include the large sample of control subjects, the identical inclusion criteria used for all three randomized trials, and the very low rates of attrition over time for each study. Our analytic approach was an additional strength: given variation between studies with respect to the follow-up period, we were able to isolate the impact of subsequent follow-up assessments from any influence of increasing subject age/time spent in the trial.

There are also several limitations to consider. The control condition was not identical for each study; we accounted for this by including indicators for the original trials as covariates within regression models. In a related limitation, the panels are unbalanced with relatively few participants having the most possible follow-ups. The unbalanced nature of our data was accounted for by utilizing a statistical model that is robust to this issue, but the lack of sample size for higher follow-up numbers (i.e., assessments 5–10) led to decreased estimation efficiency. As such, it is possible that greater numbers of follow-ups would add further benefit, but we lack the statistical power to make such a conclusion. Next, our symptom-based measures are inherently subjective. Although patient/caregiver report of recent symptoms is endorsed by national asthma management guidelines when determining severity or control,⁽³⁾ future studies will need to examine whether objective measures of asthma control (i.e., pulmonary function testing) are similarly influenced by the frequency of follow-up assessments. Finally, our findings represent symptom improvement within a specific population of children who live in urban Rochester, NY, and can only generalize to other similar populations of children with symptomatic persistent asthma.

For children with poorly controlled/persistent asthma enrolled as control subjects in clinical asthma trials, the improvement in asthma symptoms attributable to follow-up assessments is substantial. This improvement should be considered when planning asthma interventions, as it may impact study design and the interpretation of results. Future studies should determine if frequent patient contact about asthma symptoms, such as from a care coordinator within a patient-centered medical home, can lead to similar improvements among children with symptomatic persistent asthma as part of a comprehensive treatment plan.

WHAT’S NEW.

Children with uncontrolled asthma in clinical trials experience significant symptomatic improvements due to additional follow-up assessments. The degree of symptom improvement can be quantified and predicted within this population based on the number of assessments that are completed.