Association of Late-Preterm Birth With Asthma in Young Children: Practice-Based Study

Study Population and Setting

We conducted a retrospective cohort analysis of infants born at 34 to 42 weeks of gestation between January 1, 2007, and December 31, 2007. The study was conducted at 31 practices within the Children's Hospital of Philadelphia Pediatric Research Consortium, a multistate, hospital-owned, primary care practice-based research network caring for >235 000 children and adolescents at 31 practice locations. Study sites included 4 urban teaching practices, where approximately one-third of children have private insurance, and 26 urban or suburban practices not involved in resident teaching, where the majority of children are privately insured. All practices used the ambulatory EHR EpicCare (Epic, Verona, WI) for physician documentation and order entry. The Children's Hospital of Philadelphia institutional review board approved the study.

Eligible subjects included children who were seen at a primary care site for well-child care within the first 30 days of life and subsequently were monitored for 18 months. Children who left the primary care network before 18 months of age were excluded. We also excluded children with major congenital anomalies and hereditary disorders, because their patterns of health care utilization might vary significantly from those of unaffected children. These diagnoses were identified in the chronic problem list for each patient, on the basis of International Classification of Diseases, Ninth Revision (ICD-9), codes, and were clustered into homogeneous categories by using a taxonomy developed previously for acute health problems (Appendix).²⁷

Primary Predictor Variable

Gestational age was recorded as part of routine care, on the basis of available birth hospital discharge records. We categorized infants as late preterm (34–36 completed weeks of gestation), low-normal gestation (37–38 weeks of gestation), or term (39–42 weeks of gestation). Because physicians might have been less likely to document gestational-age information for term infants, compared with preterm infants, missing gestational-age data (7.6% of the sample) were likely nonrandom. We imputed missing gestational-age data on the basis of birth weight, gender, race, insurance status, income, and delivery method, by using single imputation. For univariate and multivariate regression analyses, we then used multiple imputation, a technique that pools analyses from multiple imputed data sets and yields a more-conservative estimate by adjusting the SEs for missing data uncertainty.²⁸

Outcome Variables

We measured several outcomes to evaluate the relationship between late prematurity and the development of asthma. By using ICD-9 codes from the chronic problem lists and visit histories, we measured any diagnosis of asthma (code 493.X) within the 18-month period as a binary outcome.^25,29 We also measured the diagnosis of persistent asthma as a binary outcome, including mild, moderate, and severe persistent asthma. These classifications, which were developed by the Children's Hospital of Philadelphia in 2001 on the basis of ICD-9 codes to help physicians designate the severity of asthma, were determined through physician judgment on the basis of National Asthma Education and Prevention Program guidelines.^30,31 Asthma management tools available to all practices in the EHR included a pediatric asthma control tool to capture asthma symptom frequency, as well as standardized documentation templates and order sets to facilitate severity classification and prescription of controller medications. The internal consistency and construct validity of the pediatric asthma control tool were demonstrated in previous work.³²

We also queried the EHR for prescription information to create an indicator variable for inhaled corticosteroid use, including the following medications: beclomethasone dipropionate, budesonide, fluticasone propionate, and fluticasone-salmeterol.²⁵ For the subgroup of children with diagnosed asthma, we measured the total number of acute office visits within the 18-month period with a visit diagnosis of wheezing or asthma (Current Procedural Terminology codes 99201–99205 and 99211–99215; ICD-9 codes 786.07, 466.X, and 493.X), which were all captured in the EHR database.

Confounding Variables

Clinical and demographic covariates to control for potential confounding were chosen on the basis of previous studies of asthma risk factors and included delivery method,³³ breastfeeding,³⁴ gender,^35,36 birth weight,^16,36,37 atopy,^38,39 race,^36,40 and socioeconomic status.^37,41 We obtained information on the method of delivery (cesarean versus vaginal) and breastfeeding status (any breastfeeding versus formula feeding only) by searching EHR text for the documented birth history and visit notes. Small-for-gestational-age infants were identified by using growth curves for birth weight and gestational age.⁴² Although additional clinical factors, such as respiratory distress syndrome, surfactant use, and neonatal intensive care, might have been useful to evaluate, this type of information was documented inconsistently and data were not reliable for analyses. These factors also might be on the causal pathway between gestational age and childhood respiratory outcomes and therefore would be inappropriate for inclusion in the final regression model.⁴³ Atopy was ascertained through physician diagnoses on the problem list, including diagnoses of seasonal and food allergies and eczema (ICD-9 codes V15.01–V15.05, V15.09, 995.3, 995.6–995.7, 708.0, 693.1, 477.X, 372.14, 691.8, and 692.9). Patient race (black, white, or other) and insurance status (private insurance at any time during the study versus none) also were included. To approximate family income, we linked 5-digit zip codes to US Census tract data and grouped the zip codes on the basis of the proportions of residents living below the federal poverty level.^44,45

Statistical Analyses

First, to assess correlations between variables, we conducted bivariate analyses for all covariates. If 2 variables were colinear, then we chose the most clinically meaningful variable to include for analysis. Then, we used univariate regression analyses to analyze the independent association of each covariate with the outcome variables. We built our base multivariate model by including primary care site as a fixed effect, to adjust for clustering. Then, we added terms for interactions between gestational age and other covariates, such as gender and race, to the base model. Because no interaction term remained statistically significant after adjustment of the significance level for multiple testing, the final regression model did not include any interaction terms.

We used multivariate logistic regression analyses to analyze the association between gestational age and each binary outcome (diagnosis of asthma, diagnosis of persistent asthma, and inhaled corticosteroid use), with adjustment for all other clinical and demographic covariates. For the subgroup of patients with asthma, we used multivariate Poisson regression analysis to analyze the association between gestational age and the number of outpatient visits with a diagnosis of wheeze. Comparisons of asthma, persistent asthma, and inhaled corticosteroid use from logistic regression analyses were then transformed into standardized proportions by using marginal standardization.⁴⁶ Standardized proportions are interpreted as the outcomes expected if the entire sample alternatively was subjected to the intervention or was monitored as a control group. We repeated all analyses after exclusion of children with missing gestational-age data, and results were similar (data not shown). Analyses were performed by using Stata 11.0 (Stata Corp, College Station, TX).

Study Population

In the original data set containing data for all infants born in 2007 and seen within the practice network by 30 days of life, 9625 infants were born at gestational ages of ≥34 weeks (Fig 1). We excluded 1409 children who dropped out of the practice network before 18 months of life. In addition, we excluded 291 patients with major congenital or hereditary disorders. The remaining sample consisted of 7925 children, including 581 (7.3%) born at 34 to 36 weeks of gestation, 1716 (21.7%) born at 37 to 38 weeks, 5027 (63.4%) born at 39 to 42 weeks, and 601 (7.6%) with imputed gestational-age data (Table 1).

We conducted χ² analyses to compare our final study sample with the patients who were excluded because of dropout from the primary care network before 18 months of age. There was no statistically significant difference in gestational age or gender between our final study sample and the excluded patients. However, patients in our study sample were more likely to be white (53% vs 43%; P < .001), to have any private insurance (76% vs 62%; P < .001), and to live in zip codes with <5% of residents below the federal poverty limit (40% vs 32%; P < .001).

Unadjusted Analyses

The overall proportion of children in the sample with diagnosed asthma was 8.3%. Unadjusted analyses of gestational age and outcomes demonstrated statistically significant differences in diagnoses of asthma (P = .002), diagnoses of persistent asthma (P = .04), and inhaled corticosteroid use (P < .001) (Tables 1 and 2). Unadjusted differences in covariates across gestational-age groups are listed in Table 1. In univariate analyses, several covariates, including black race (odds ratio [OR]: 1.46 [95% confidence interval [CI]: 1.23–1.74]) and lower income (OR: 1.58 [95% CI: 1.27–1.96]), also were associated with diagnoses of asthma (Table 2).

Multivariate Analyses

Compared with gestational ages of 39 to 42 weeks, late-preterm gestation was associated with a trend toward increased diagnoses of asthma, although this difference was not statistically significant (Table 3). However, late-preterm gestation was significantly associated with increased diagnoses of persistent asthma (adjusted OR [aOR]: 1.68 [95% CI: 1.01–2.80]) and receipt of ≥1 inhaled corticosteroid prescription (aOR: 1.66 [95% CI: 1.20–2.29]). In addition, late-preterm infants with asthma had more acute respiratory visits (incident rate ratio: 1.45 [95% CI: 1.25–1.68]), compared with children born at term. Children with gestational ages of 37 to 38 weeks also had increased odds of asthma outcomes, compared with term children, including diagnoses of asthma (aOR: 1.34 [95% CI: 1.11–1.63]) and inhaled corticosteroid use (aOR: 1.41 [95% CI: 1.15–1.73]) (Table 3).

The standardized proportion of asthma in the reference group was 7.7%, compared with 9.4% for children born late preterm (P = .15) and 9.9% for children born at 37 to 38 weeks (P = .003) (Fig 2). The proportion of persistent asthma in the reference group was 2.2%, compared with 3.6% for children born late preterm (P = .04) and 2.8% for children born at 37 to 38 weeks (P = .18). The proportion of inhaled corticosteroid use in the reference group was 5.7%, compared with 9.0% for children born late preterm (P = .002) and 7.7% for children born at 37 to 38 weeks (P = .003).