Pediatric Asthma Care Coordination in Underserved Communities: A Quasiexperimental Study

Abstract

Objectives. To assess the effect of care coordination on asthma outcomes among children in underserved urban communities.

Methods. We enrolled children, most of whom had very poorly or not well-controlled asthma, in medical–social care coordination programs in Los Angeles, California; Chicago, Illinois; Philadelphia, Pennsylvania; and San Juan, Puerto Rico in 2011 to 2014. Participants (n = 805; mean age = 7 years) were 60% male, 50% African American, and 42% Latino. We assessed asthma symptoms and health care utilization via parent interview at baseline and 12 months. To prevent overestimation of intervention effects, we constructed a comparison group using bootstrap resampling of matched control cases from previous pediatric asthma trials.

Results. At follow-up, intervention participants had 2.2 fewer symptom days per month (SD = 0.3; P < .01) and 1.9 fewer symptom nights per month (SD = 0.35; P < .01) than did the comparison group. The relative risk in the past year associated with the intervention was 0.63 (95% confidence interval [CI] = 0.45, 0.89) for an emergency department visit and 0.69 (95% CI = 0.47, 1.01) for hospitalization.

Conclusions. Care coordination may improve pediatric asthma symptom control and reduce emergency department visits.

Policy Implications. Expanding third-party reimbursement for care coordination services may help reduce pediatric asthma disparities.

Asthma affects more than 6.1 million children younger than 18 years in the United States¹ and results in millions of doctor visits and missed school days annually, hundreds of thousands of emergency department (ED) visits and hospitalizations, and thousands of deaths.^2,3 Despite widespread attention to disparities in pediatric asthma, significant gaps persist.^2,3 Non-Latino Black children (13.4%) are nearly twice as likely to have asthma as are non-Latino White children (7.5%) or Latino children (7.4%), with the exception of Puerto Rican children (20.7%).¹ Black and Latino children with asthma visit the ED more frequently than do White children and are more likely to die from asthma.³ Barriers to asthma management, such as lack of consistent, appropriate medical care and increased exposure to environmental triggers, disproportionately affect children living in low socioeconomic, urban areas.^4,5

Care coordination is a promising strategy to address these barriers. As defined by Brown, care coordination is

a client-centered, assessment-based interdisciplinary approach to integrating health care and social support services in which an individual’s needs and preferences are assessed, a comprehensive care plan is developed, and services are managed and monitored by a care coordinator following evidence-based standards of care.^6(p1)

Previous quasiexperimental studies have suggested that care coordination results in reduced asthma symptoms and reduced urgent health care utilization.^7–10 For example, 13% of high-risk children with asthma in the Link Line intervention group (Philadelphia) had follow-up hospitalizations over a 1-year period compared with 33% in a matched sample comparison group.⁹ In the Community Asthma Initiative (Boston), participants experienced a 68% decrease in ED visits over 12 months.¹⁰ Care coordination appears to be cost effective, with the return on investment calculated at 1.46 dollars saved per dollar spent.¹⁰

Policy developments related to the Patient Protection and Affordable Care Act increase the feasibility of care coordination models. Medicaid reimbursement is now permitted for preventive services (including chronic disease management)¹¹ that are delivered by certified, nonlicensed health care providers if they are recommended by a licensed health care practitioner. This facilitates the provision of asthma care coordination elements such as education and home visits by community health workers or certified asthma educators. The Affordable Care Act also permits state Medicaid programs to form health homes for patients with chronic illness, of which care coordination is 1 of 6 core services.¹¹

The Merck Childhood Asthma Network, Inc. (MCAN) funded a 4-year project to assess the effectiveness of pediatric asthma care coordination in 4 urban settings. These sites were the Los Angeles Unified School District (LAUSD) Asthma Program (Los Angeles, CA), the Children’s Hospital of Philadelphia Asthma Care Navigator Program (Philadelphia, PA), the federally qualified health center–based La Red de Asma Infantil de Puerto Rico (San Juan, PR), and the neighborhood-based Addressing Asthma in Englewood Project (Chicago, IL).¹² We have reported 1-year changes in participants’ daytime and nighttime symptoms and frequency of ED visits and hospitalizations.

METHODS

From 2005 to 2015, the MCAN Care Coordination Programs, phases I and II, funded diverse sites to reduce pediatric asthma morbidity in vulnerable populations. In phase I (2005–2009), 5 sites implemented evidence-based interventions to improve outcomes, explore the factors that led to successful evidence-based intervention adoption, and understand program adaptations.^13,14 In phase II (2010–2015), 4 sites implemented evidence-based pediatric asthma care coordination activities. All phase II sites adapted Yes We Can, a medical–social model of care that deploys health workers to provide asthma education, link families to health and social services, and facilitate patient–clinician communication.⁷

Details about the settings and content of the 4 programs are presented in Table 1 and are also found elsewhere.¹² Broadly, all 4 sites included intervention components in which an asthma care coordinator provided families with asthma education, including use of an asthma action plan and trigger remediation. Staff with varied professional backgrounds, including nurses, health educators, and community health workers, who were trained in delivering culturally relevant care performed the asthma care coordinator role.

TABLE 1—

Intervention: Merck Childhood Asthma Network Care Coordination Programs, Phase II, 2010–2015

Variable	Los Angeles, CA	Chicago, IL	Philadelphia, PA	San Juan, Puerto Rico
Content (delivered in-person 1-on-1 to parent or caregiver)	Asthma education	Asthma education	Asthma education	Asthma education
	Home environmental assessment	Home environmental assessment	Home environmental assessment	Home environmental assessment (received by half of participants)
	Mitigation suppliesa and education provided	Mitigation suppliesa and education provided	Mitigation suppliesa and education provided	Mitigation suppliesa and education provided
	Referrals provided	Referrals provided	Needs assessment of caregiver
	AAP given and explained	AAP given and explained	Goal setting for care coordination
	ACT administered	Follow-up reports sent to physicians	Education and links provided to meet goals
			Goals assessed
Interventionist	Asthma program nurses (RNs)	Community health educators	Asthma care navigators (clinic-based community health workers)	In clinic: health educator
				In home: community health worker
Setting	Home	Home	Clinic and home	Clinic and home
	Occasionally at school or over telephone
Contacts	3–4 home visits plus telephone calls 2 wk after each visit	≥ 1 home visit plus at least 2 additional home, clinic, or telephone contacts	3 home visits plus ≥ 4 clinic visits (if possible)	2 clinic and, for half of the participants, 2 home visits
Eligibility criteria
Asthma morbidity	Diagnosed asthma or symptoms	Diagnosis of intermittent or persistent asthma	Diagnosis of persistent asthma	Diagnosis of asthma
Age, y	4–18	0–18	0–17	0–17
Location or provider	Reside in LAUSD boundaries	Reside in Englewood, West Englewood, or in the 10 blocks surrounding	PCP in 1 of 3 urban CHOP clinics	Patient of HealthProMed FQHC
Other	ACT < 20 and any of the following:		All of the following:	Any of the following:
	≥ 10 missed school days		≥ 1 hospitalization or 2 ED visits in last year	Daily asthma symptoms in last 2 wk
	Recent ED visit or hospitalization		Prescribed controller medication	≥ 2 nights of asthma symptoms last 2 wk
	Recent diagnosis		Medical assistance as primary insurance	≥ 1 asthma hospitalizations last year
	Parent, nurse, or doctor request		English or Spanish is primary language of caregiver	≥ 2 ED asthma visits last year
				Control medication use every day last wk or rescue medicine ≥ 2 times/wk
Recruitment	Existing school district health information	Community health educators recruited at local clinics	ED and hospitalization records	Study staff recruited in FQHC waiting room
	Referrals from school nurses, doctors, and other staff, parents, Breathmobile doctors	Referrals from physicians, schools, community-based organizations; community recruitment	Chart review in participating clinics	Community recruitment
			Provider referral

Note. AAP = Asthma Action Plan; ACT = asthma control test; CHOP = Children’s Hospital of Philadelphia; ED = emergency department; FQHC = federally qualified health center; LAUSD = Los Angeles Unified School District; PCP = primary care provider; RN = registered nurse.

^aMitigation supplies include items such as roach gel and mattress covers.

Asthma care coordinators typically made 1 or more home visits to assess triggers and helped the family address social barriers to asthma management. Participants received remediation materials such as pillow covers and pest management supplies. Importantly, at each site, asthma care coordinators formed links with clinical care providers, although the extent and nature of clinical integration varied across sites.¹²

Participants and Data Collection

Eligibility criteria and recruitment procedures varied by site (Table 1). Programs recruited primarily, but not exclusively, children with poorly controlled, persistent asthma. A total of 1223 children were enrolled, exceeding the target sample size (determined by feasibility) of 800.

We administered a survey to parents or caregivers upon enrollment at baseline and 12 months later to collect data on primary intervention outcomes, asthma self-management, and satisfaction with asthma care. Across sites, we collected baseline data between January 2011 and June 2013; we collected follow-up data between January 2012 and November 2014.

Asthma care coordinators, who received training in this task and were given a detailed reference manual, administered surveys, in English or Spanish, primarily in the clinic or in the caregiver’s home. Occasionally, they conducted interviews by telephone. No incentives for survey completion were provided.

Outcomes

We assessed daytime and nighttime symptoms with 2 items adapted from the Childhood Asthma Control Test¹⁵: “During the last 4 weeks, how many days did [child’s name] have any daytime asthma symptoms (like wheezing, shortness of breath, tightness in the chest, or cough)?” and “During the last 4 weeks, how many nights did [child’s name] wake up during the night because of asthma?”

We asked parents how many times in the past 12 months their child had been “treated in the emergency department or ER for asthma” and “been admitted to a hospital for asthma and had to stay overnight for one or more nights.”

Analysis Plan

We conducted all analyses using SAS version 9.3 (SAS Institute Inc., Cary, NC). We examined all survey variables for consistency and completeness before analysis.

We calculated descriptive statistics on the demographic and health characteristics and compared those who completed the study (n = 805) with those lost to follow-up (n = 418) using the χ² test of homogeneity for categorical variables and the t test for continuous variables. We assessed changes from baseline to 12-month follow-up in mean symptom days, symptom nights, number of ED visits, and number of hospitalizations using generalized mixed-effects models with a Poisson distribution (SAS PROC GLIMMIX). We treated the family as the cluster of analysis and adjusted for age, race/ethnicity, gender, and site as fixed effects. Because the sample included children living in the same household, we used a family identifier as a random intercept to account for correlation within the 68 sets of siblings in the sample.

To enhance the rigor of the study design, we use a novel statistical technique, the Clark adjustment (named after pioneering asthma researcher Noreen Clark¹⁶), in which a comparison group is constructed using matched control group data from previous trials.¹⁷ This simulated control group is used to estimate the usual improvement over time that occurs in asthma symptoms as children age.¹⁸ Ko et al. estimate that children younger than 10 years experience an 18% per-year decrease in symptoms, with more modest decreases in older children.¹⁷ Moreover, many children enter asthma studies following exacerbations of the illness, yet symptoms naturally fluctuate over time.

Failure to account for this expected improvement over time could lead to an overestimation of the intervention effect. We used bootstrap resampling with replacement to match intervention to control cases on race/ethnicity, gender, and the dichotomized baseline value of the variable of interest to compare children at similar levels of asthma severity. We repeated this sampling 1000 times to reduce selection bias. Regression models were fit within each of the 1000 matched samples, averaging results across models to determine final estimates of the treatment group effect. A more detailed description of this technique can be found in Appendix A (available as a supplement to the online version of this article at http://www.ajph.org) and Ko et al.¹⁷

We estimated models with the following 4 outcomes: pre–post change in symptom days and symptom nights and the probability of reporting 1 or more ED visits and hospitalizations. For each model, the main predictor variable was group (treatment or control), with age and baseline value of the outcome as covariates. We obtained control cases from control group data from 4 previous randomized controlled trials of pediatric asthma interventions from 2001 or later: the Detroit Middle School Asthma Project,¹⁹ the Physician Asthma Care Education program,²⁰ the Inner City Asthma Consortium,^21,22 and the Study of Adherence Monitoring with Feedback and Asthma Basic Care interventions.²³ Details on control cohort characteristics are found in Table A (available as a supplement to the online version of this article at http://www.ajph.org).

In the 2 symptom models, we estimated general linear models with change in symptoms (daytime or nighttime) from baseline to follow-up as the dependent variable. For the 2 health care use models (probability of having at least 1 ED visit or hospitalization at follow-up, if this event was reported at baseline), we fit zero-inflated log-linear models. We excluded Latino participants (n = 339) from these 2 models only because of a lack of comparable health care use data (i.e., events not reported in the same recall period) for Latino children in the control cohorts. Analytic samples for these 2 models also excluded non-Latino participants who did not have at least 1 baseline ED visit (n = 143 excluded) or hospitalization (n = 224 excluded) at baseline.

RESULTS

A total of 1223 parents or caregivers completed a baseline interview, and 805 (66%) completed a follow-up interview. Most (67%) of the participants who were lost to follow-up came from the LAUSD site owing to the transient nature of this school population. Compared with study completers, participants lost to follow-up were older (8.7 vs 7.0 years; P < .001); more likely to be Latino (65% vs 42%; P < .001); less likely to have medical insurance (88% vs 96%; P < .001); and less likely to have Medicaid (74% vs 92%; P < .001).

Children lost to follow-up also reported more baseline symptom days (8.5 vs 7.4; P = .02) than did completers; however, symptom nights were similar between the 2 groups. Across sites, the final analytic sample was approximately equally distributed among children aged 0 to 4 years, 5 to 8 years, and 9 years or older (Table 2). Slightly more than half of participating children were male (59.5%). The vast majority of participants were either African American (50.4%) or Latino (42.1%), including those of Puerto Rican, Mexican American, and Central American origin, and recruited from the San Juan and Los Angeles sites, respectively. About one third of respondents reported speaking Spanish at home. More than 90% of the sample was insured by Medicaid.

TABLE 2—

Baseline Demographic Characteristics of Final Sample (n = 805): Merck Childhood Asthma Network Care Coordination Programs, Phase II, 2010–2014

Characteristic	No. (%)
Site (n = 805)
Chicago, IL	134 (16.6)
Los Angeles, CA	232 (28.8)
Philadelphia, PA	254 (31.6)
San Juan, Puerto Rico	185 (23.0)
Age, y (n = 803)
0–4	259 (32.2)
5–8	270 (33.5)
9–11	137 (17.0)
12–18	137 (17.0)
Male (n = 802)	479 (59.5)
Child’s race or ethnicity (n = 802)
African American	406 (50.4)
Latino	339 (42.1)
White	14 (1.7)
Other	43 (5.3)
Language spoken at home (n = 801)
English	481 (59.8)
Spanish	239 (29.7)
Other	81 (10.1)
Level of asthma control (n = 797)
Well controlled	182 (23.0)
Not well controlled	390 (49.0)
Very poorly controlled	225 (28.0)
Caregiver’s relationship to child (n = 804)
Mother	733 (91.1)
Father	26 (3.2)
Other	45 (5.6)
Highest level of school caregiver attended (n = 796)
Less than high school	121 (15.0)
High school	498 (61.9)
College	164 (20.4)
Postcollege	13 (1.6)
Other
Child has health or medical insurance (n = 794)	776 (96.4)
Child’s insurance is Medicaid (n = 791)	742 (92.2)

Pre–Post Results

As shown in Table 3, at baseline, parents reported adjusted means of 5.0 and 3.8 daytime and nighttime symptom days per month, respectively; at follow-up, these decreased to 2.1 and 1.2. Parents reported an adjusted mean of 3.2 ED visits and 2.0 hospitalizations in the past year at baseline; the adjusted means at follow-up were 1.2 and 0.6, respectively.

TABLE 3—

Self-Reported Outcomes at Baseline and 12-Month Follow-Up: Merck Childhood Asthma Network Care Coordination Programs, Phase II, 2010–2014

Outcome	Baseline (n = 805), Adjusted Meana (95% CI)	Follow-Up (n = 805), Adjusted Meana (95% CI)	Ratio of Adjusted Meansa Between Baseline and Follow-Up (95% CI)
During the past 4 wk
No. d child had any daytime asthma symptoms	5.0 (3.7, 6.8)	2.1 (1.6, 2.9)	2.4 (2.3, 2.5)
No. d child woke up during the night because of asthma	3.8 (2.6, 5.4)	1.2 (0.8, 1.7)	3.2 (3.0, 3.4)
During the past 12 mo
No. emergency department visits	3.2 (2.3, 4.5)	1.2 (0.9, 1.7)	2.6 (2.4, 2.8)
No. hospitalizations	2.0 (1.2, 3.2)	0.6 (0.4, 1.0)	3.1 (2.7, 3.5)

Note. CI = confidence interval. We used the Poisson version of generalized linear mixed-effects models with the following specifications: family as the cluster of analysis; canonical link function; gender, baseline age, time, and site as fixed effects; a random intercept for each family as a random effect; and no R-side (marginal) correlation.

^aMeans of variables of interest are for a Hispanic boy aged 7.05 years in Philadelphia, as an example case, with the mean and mode characteristics in the data set.

The pre–post changes (ratio of adjusted means) were all statistically significant. The intraclass correlation coefficient (ICC), a correlation measure available in the normal linear mixed-effects model, is not well defined in the Poisson generalized mixed-effects model we used in the analysis, and thus we did not include the ICC for family-level clustering in the table. Coefficients for all model variables are presented in Table B (available as a supplement to the online version of this article at http://www.ajph.org).

Site-Specific Results

We sought to examine the effect of care coordination across sites. However, because we found significant pre–post changes in all 4 primary outcomes in the pooled sample, as a post hoc analysis we ran models including a “site × time” interaction to determine whether the pre–post changes varied significantly by site. This coefficient was significant in the models for symptom days and hospitalizations, indicating differential change by site for these 2 outcomes.

We next examined change in symptom days within each site and found that all sites showed a statistically significant improvement in this outcome (P < .001). For hospitalizations, pre–post improvements were significant (P < .001) within all sites except LAUSD. (Results not shown.)

Clark Adjustment Results

Table 4 shows that compared with the control group, the intervention group experienced a 2.2-day (SD = 0.43) greater reduction in symptom days and a 1.9-day (SD = 0.35) greater reduction in symptom nights. Without the Clark adjustment, as shown in the example case in Table 3, participants reported 2.9 fewer symptom days and 2.6 fewer symptom nights over the project period; these represent overestimations of program effect of 130% and 136%, respectively.

TABLE 4—

Outcomes From Clark Adjustment Analyses Comparing Participants to Standardized Comparison Group: Merck Childhood Asthma Network Care Coordination Programs, Phase II, 2010–2014

Treatment Effect	Difference (95% CI) or RR (95% CI)
No. d of daytime symptoms in past 4 wka (n = 779)	–2.20 (–3.05, –1.35)
No. d of nighttime symptoms in past 4 wka (n = 782)	–1.92 (–2.61, –1.23)
No. ED visits in past 12 mob (n = 315)	0.63 (0.45, 0.89)
No. hospitalizations in past 12 mob (n = 233)	0.69 (0.47, 1.01)

Note. CI = confidence interval; ED = emergency department; RR = risk ratio. For ED visit and hospitalization analyses, results were from a zero-inflated log-linear model.

^aSamples were matched on gender, race, and whether number of baseline days or nights were ≤ 14 for the respective model.

^bSamples were matched on gender and race, including only those with a baseline event; Hispanic/Latino participants were excluded from our analysis because of the absence of comparable health care use data from matched control cases.

The relative risk (RR) associated with participation in the intervention of having 1 or more ED visits in the past 12 months, considering at least 1 baseline ED visit, was 0.63 (95% confidence interval [CI] = 0.45, 0.89) and of having 1 or more hospitalizations, considering at least 1 baseline hospitalization, was 0.69 (95% CI = 0.47, 1.01).

DISCUSSION

Overcoming persistent race/ethnicity and socioeconomic-based disparities in asthma outcomes is widely viewed as a top priority in respiratory health.²⁴ We found support for our hypothesis that structured asthma care coordination programs—which provided education and resources to families and integrated these services with clinical care—would improve key outcomes among children from low-income, urban African American and Latino communities. We found that over a 1-year period, children who participated in the care coordination programs experienced marked improvement in both daytime and nighttime symptoms and reduced their risk of reporting asthma-related ED visits.

Importantly, we were able to compare changes in program participants with matched cases in a control group derived from past trials to better isolate the effects of the interventions. This approach reduces the possibility that the observed program effect stems from typical age-related improvements in asthma symptoms or regression to the mean.

Our findings have a high degree of external validity, because our study design followed many of the principles of pragmatic trials,²⁵ for example, there were no health-related selection criteria, apart from level of asthma control at some sites; a range of practitioners, who were given flexibility in adhering to the details of implementation, applied the intervention in a variety of settings; and patient and provider adherence were minimally monitored. In other words, because the interventions were deployed under real-world conditions, improvements of a similar magnitude to those we observed are likely to result when this care coordination model is used in similar community or care settings.

Symptoms are a core outcome measure in asthma research, although evidence is lacking to identify the minimally important clinical difference in longitudinal studies.²⁶ Children who have well-controlled asthma according to National Asthma Education and Prevention Program guidelines have symptoms on 2 or fewer days per week and nighttime symptoms (awakenings) 1 or fewer days per month.²⁷ National Asthma Education and Prevention Program recommendations for asthma severity assessment specify that 2 or more ED visits or hospitalizations for asthma in the past year are associated with a higher risk of exacerbations or death. Therefore, reductions in daytime and nighttime symptoms and ED visits, as we observed, increase children’s likelihood of having a favorable clinical profile of being in the well-controlled category and at lower risk.

At the population level, reductions in expensive urgent care use will result in substantial decreases in health care costs. Demonstrating the favorable impact on costs of asthma care models may assist in advocating coverage by private health insurers.²⁸

Our results are consistent with previous research on asthma care coordination and therefore add to a growing evidence base for the effectiveness of this approach. The cross-site evaluation of the MCAN phase I sites, which employed many of the elements of care coordination that were employed in phase II, also used a pre–post comparison to assess the impact of the communities’ initiatives on asthma outcomes.⁸ Phase I sites saw a 56.1% decrease in participants’ mean symptom days over 2 weeks (compared with 57.9% in phase II sites over 4 weeks) and a 55.2% decrease in mean symptom nights over 2 weeks (compared with a decrease of 68.9% in phase II sites over 4 weeks).

An evaluation of Yes We Can also showed significant decreases in daytime and nighttime symptom days over 2 weeks (decreases of 45.1% and 46.0%, respectively).⁷ Our outcomes related to ED visits and hospitalizations were consistent with this previous work.^8,29 For example, participants in the phase I sites reported 1.1 fewer ED visits over 12 months (compared with 2.0 among phase II participants).⁸

In addition to our quantitative analysis, a comprehensive cross-site qualitative evaluation assessed the implementation process, the sustainability of these programs, and their links to key policy and systems changes.¹² The ability to demonstrate positive outcomes of the programs, such as those we have reported, also played a role in promoting program sustainment at all sites.³⁰

Our results demonstrate the value of the Clark adjustment methodology for creating a control group using matched cases from existing control cohorts. When we compared pre–post changes in daytime and nighttime asthma symptoms with the results with the Clark adjustment, we saw that the program’s effects on these 2 outcomes had been overestimated in the simpler analysis. The Clark adjustment may be useful in future evaluations of community-based studies, as well as implementation research, and pragmatic studies when randomization to a usual care condition is not feasible or acceptable.

A randomized design remains the gold standard for isolating an intervention effect; however, the Clark adjustment represents a valid, quasiexperimental alternative that can enhance the rigor of studies at little added cost, while allowing studies to be conducted under usual conditions that maximize external validity.^25,31 In addition, evaluations of community-based, asthma management support programs that employ the Clark adjustment may build a more compelling case for reimbursement for asthma educator–provided services.³²

To maximize the value of this novel methodology, we urge collaborative development and maintenance of a mechanism by which researchers and evaluators can easily access relevant control group data, for example, in a central online repository that offers an easy to use process for using the data to form a matched control group for a particular study or evaluation. This control group data should necessarily be regularly updated to reflect changes in usual care for pediatric asthma.

Control cases should be demographically and geographically representative of the entire US population of children with asthma and should represent typical settings (including community clinics, academic centers, and schools) in which children receive asthma care. Additionally, we recommend that available control group data align with the outcome measures for asthma studies recommended by the National Institutes of Health and the Agency for Healthcare Research and Quality.³³

Limitations

Several study limitations are noteworthy. First, because the control cohorts that included Latino children did not have data on ED visits and hospitalizations reported in a timeframe like ours (in the past year), we excluded Latino participants from the Clark adjustment analysis of the program’s effect on these particular outcomes. Although our analysis of the program’s effect on asthma symptoms included all sites and participants, excluding Latino participants from the 2 models of health care utilization meant that all participants at the Puerto Rico site and nearly 70% at the LAUSD site (data not shown) were not represented in the Clark adjustment analysis for ED use and hospitalizations. However, we found that within the MCAN group, non-Latino children had greater odds of reporting an ED visit at follow-up than did Latino children in the study (odds ratio [OR] = 1.44; 95% CI = 1.07, 1.94) and similar odds for reporting a hospitalization (OR = 1.08; 95% CI = 0.75, 1.54; data not shown), suggesting that the program was at least as effective among Latino children.

Second, loss to follow-up may have introduced bias; however, the children who dropped out had health characteristics that were similar to the symptoms of those who remained in the sample, and most dropouts occurred in the LAUSD, where children frequently switched schools. Third, we used control group data collected as early as 2001, when standard clinical management of asthma may have been different; for example, the use of long-term controller medications has increased over time.³⁴ Control group data from an earlier time may, therefore, be different from control group data collected contemporaneously with the MCAN interventions.

Similarly, control group data did not represent the cities included in our trial, and therefore any regional differences in asthma morbidity or treatment are not accounted for. In sum, our quasiexperimental design did not eliminate all threats to internal validity. Nonetheless, simulations with Clark adjustment methodology demonstrated that the estimates of program effects it produces are within 2.4% of those using the gold standard of a true control group.¹⁷

Last, our pre–post analysis accounted for within-family correlation but ignored within-site correlation. The small number of sites made the estimation of the intersite correlation either imprecise or not possible for each outcome. The assumption of zero intersite correlation does not affect estimates but may slightly reduce the statistical power.

Public Health Implications

We conducted an evaluation of a pediatric asthma care coordination model in 4 distinct settings—neighborhoods, schools, outpatient clinics, and a federally qualified health center—in communities strongly affected by disparities in asthma morbidity and mortality.

Using a novel statistical technique—a simulated control group made up of matched cases from previous trials—to increase the internal validity of this community-based effectiveness study, we have provided compelling evidence that this model can significantly reduce asthma symptoms and ED visits. Expanding access to care coordination services, for example, by expanding third-party reimbursement, may help reduce asthma disparities at the population level.