Data sources
All data were extracted from the Duke University Health System Electronic Health Record (EHR) system. Duke University Health System consists of 3 hospitals – 1 tertiary care and 2 community-based – and a network of primary care and specialty clinics. As the primary provider in Durham County, North Carolina, it is estimated that 80 percent of Durham County residents receive their care through Duke University Health System.31 For the purposes of this study we abstracted analytic data from our data warehouse covering the years January 1, 2014 to December 31, 2019.
Study population
We identified children (age 5-18), living in Durham County with asthma. Children had to be at least 5 years old to rule out unclear respiratory related diagnoses in younger children. To identify children with asthma we applied two definitions. Our first definition was encounter-based. A child had to have: 1) two outpatient encounters or one inpatient encounter with an asthma diagnosis (see Appendix Table 1 for ICD9/10 codes) and 2) a prescription for an asthma medication (see Appendix Table 2). Our second definition was problem list-based. Problem lists are an EHR feature that serves as a comprehensive list of patient diagnoses that is intended to serve as a snapshot of the patient’s health status. To be included under the problem list-based definition, a child had to have 1) asthma on their problem lists and 2) a prescription for an asthma medication. The positive predictive value of this computable phenotype is 97%.32
In order to account for differential follow-up times, person-time was calculated from time of positive asthma identification until censoring. Censoring was based on aging out of the cohort (>= 18), an indicated address outside of Durham County, or at the last known encounter.
Primary exposure
The primary exposure in our study was the residential distances to two types of roadways: roads with U.S. Census feature Class Code A1 (55+ MPH with limited access only accessible via ramps) and A2 (35+ MPH-primary road without limited access),33-35 with these speeds corresponding to highways and major roadways, respectively.15,16,18,25 We abstracted the address information of each individual in our cohort from our EHR system and geocoded them using ArcGIS (version 10.5; ESRI Inc., Redlands, CA). The accuracy of all address information was manually checked with Google Maps to verify the existence of residences at each address. Addresses were treated as a time-varying exposure based on when a child moved to a new address, as indicated in the EHR. Map figures were made with ArcGIS software.
Straight-line distance to roadways, which was calculated using ArcGIS, was used as our primary exposure (See Figure 1). Durham County has three major roadways that intersect in central Durham (Figure 1). For children who live within this triangular intersection, straight-line distance might not accurately capture exposure. We therefore also constructed radial density measures of 1.0 mile in order to evaluate exposures associated with proximity to more than one roadway (Figure 2).
Outcome of interest
The primary outcome of interest was an asthma-related exacerbation, which was defined as any encounter with an asthma-related ICD9 or -10 code and a prescription for a systemic steroid (see Appendix Table 1). We further categorized exacerbations into four different outcome tiers based on severity (listed in decreasing severity): 1) inpatient encounters lasting more than 24 hours, 2) emergency department and hospital encounters lasting less than 24 hours, 3) urgent care visits, and 4) outpatient (including telephone-based) encounters.
Covariates
We abstracted additional clinical and socio-demographic information on each child from the EHR, including sex, age, race, insurance type (public, private, self-pay), neighborhood socio-economic status (nSES), comorbidities (atopy, obesity), medication categories (only rescue, only inhaled corticosteroids or only Leukotriene receptor antagonist, or other controller medications), and number of overall encounters. Covariates were treated as time-varying and abstracted accordingly. We considered any prescription order in the past 365 days as current. To calculate nSES, we identified each child’s census tract and linked data from the American Community Survey to calculate the Agency for Healthcare Research and Quality (AHRQ) neighborhood deprivation index,36 generating a score between 0 – 100, with higher scores indicative of greater deprivation.
Air quality exposure
To further characterize potential exposures, we abstracted publicly available data from Environmental Protection Agency sensors on daily PM2.5.37 We note that Durham County only has one sensor, providing a general approximation of daily PM2.5 exposure.
Statistical analysis
We constructed a time-varying dataset in counting-process format, generating an encounter row each time a child moved (i.e., changed primary exposure) or had an asthma exacerbation. All time-varying covariates (e.g., medications) were calculated based on these encounter points. We categorized patients based on their baseline distance to roadway, compared clinical and demographic differences among these groups, and calculated the number of exacerbations per patient-year.
To assess the relationship between distance to roadway and asthma exacerbation rates we performed a survival time-to-event analysis, treating distance as a time-varying exposure. We fit separate models for straight-line distance and roadway radial density. In our cohort, about 25 percent of children had more than one asthma exacerbation over the study period. To account for multiple exacerbations, we conducted a recurrent events survival analysis using Andersen-Gill models.38 As a sensitivity analysis, we used the Prentice, Williams and Peterson model and frailty model (random effects approach) as alternative recurrent events models.39 Four different models were fit: 1) unadjusted; 2) adjusted for nSES; 3) adjusted for sex, age, race, and insurance type; and 4) adjusted for all the other covariates specified above. Additionally, we assessed the linearity of the effect of distance to roadways and roadway radial density on asthma exacerbation outcomes. We categorized our distance to roadways variables into quartiles and refitted the models. We also stratified on sex, race, and assessed time to outcome specific tiers.
All the statistical analyses were perform using R version 3.6.0.