Recurrent Wheezing in the Third Year of Life Among Children Born at 32 Weeks' Gestation or Later:  Relationship to Laboratory-Confirmed, Medically Attended Infection With Respiratory Syncytial Virus During the First Year of Life FREE

Respiratory syncytial virus (RSV) infection is a common event in childhood; rates approach 100% by 3 years of age,¹ resulting in an innocuous upper respiratory infection in most infants. However, a significant number develop severe lower respiratory tract infections.^2- 4 Respiratory syncytial virus is responsible for 50% to 80% of bronchiolitis hospitalizations⁵ and is the leading cause of hospitalization in children younger than 1 year.⁶ Approximately 75 000 to 125 000 hospitalizations associated with RSV occur each year in the United States.²

The RSV hospitalization patterns are a good indicator of the burden of severe infection. The burden of less severe RSV infections (those only requiring outpatient care) is not trivial; it is estimated that 1 in 13 visits to a physician's office result from RSV disease.⁷ Moreover, RSV bronchiolitis during infancy is associated with subsequent development of recurrent wheezing and early childhood asthma.^8- 15 Infants at high risk of RSV disease, including the very premature, may face severe respiratory complications after RSV infection.¹⁶ Although moderate prematurity is associated with increased rates of RSV infection,¹⁷ few recent data are available on the effects of RSV infection or consequences in late-preterm infants.

This study examined the long-term effect of RSV bronchiolitis during infancy in moderately preterm and full-term infants. We hypothesized that laboratory-confirmed, medically attended RSV infection (henceforth referred to as RSV infection) during the first year of life would show a strong association with the risk of recurrent wheezing (RW) in the third year.

Our study setting was the Northern California Kaiser Permanente Medical Care Program (KPMCP). This study was approved by the Northern California KPMCP Institutional Review Board, which has jurisdiction over all of the hospitals and clinics involved in the study. The inception cohort consisted of infants who met the following criteria: birth at the KPMCP Oakland, Hayward, Sacramento, San Francisco, Santa Clara, or Walnut Creek, California, hospitals; date of birth from January 1, 1996, through December 31, 2002; 32 weeks' or later gestational age (GA) at birth; infant survival of birth hospitalization; and use or membership in Kaiser Foundation Health Plan Inc during the first year of life.

Our primary study outcome was the occurrence of RW in the third year of life (second birthday through the day before the third birthday). Before extracting electronic data, RW was defined as 3 or more outpatient encounters, at least 14 days apart, with a diagnosis of asthma (International Classification of Diseases¹⁸[ICD] code 493.xx) or wheezing (ICD code 786.07); and/or 1 or more asthma or wheezing encounter with a prescription for oral corticosteroids within 2 days before or 7 days after the encounter; and/or 1 or more hospitalizations with a diagnosis of asthma or wheezing with a total hospital length of stay of 24 hours or longer or until death occurred; and/or 4 or more dispensing events (prescriptions that were picked up), at least 14 days apart, on which selected asthma medications were prescribed in addition to 1 or more encounters with a diagnosis of asthma or wheezing; and/or death outside the hospital setting with the cause of death listed as asthma or wheezing. Selected asthma medications included rescue medications (including albuterol and theophylline), inhaled corticosteroids (including beclomethasone dipropionate, budesonide, and flunisolide), leukotriene inhibitors (montelukast), or oral corticosteroids (including dexamethasone and prednisone). The full list of 640 medications is available on request.

The primary study predictor was the occurrence of RSV infection in the first year of life. We also examined whether infants were infected with adenovirus, parainfluenza virus, influenza A or B virus, or Bordetella pertussis. We considered outpatient visits, emergency department visits, and inpatient hospitalization with diagnostic codes for each type of pathogen to be medically attended events. A medically attended event was considered to be laboratory confirmed if there was a positive result on a test for the pathogen within14 days before or after the encounter. During the period covered by this study, laboratory confirmation for these pathogens in the KPMCP was by direct fluorescent antibody testing or culture. We categorized medically attended encounters of RSV and other pathogens as outpatient (including scheduled clinic visits, unscheduled urgent care visits, and emergency department visits), uncomplicated hospitalizations (hospitalizations with a total length of stay of <96 hours without assisted ventilation), or prolonged hospitalizations (hospitalizations with a total length of stay of 96 hours or more, or in which assisted ventilation was required).

We also examined, for each infant, sex, race, maternal age, birth weight, GA, a discharge diagnosis of bronchopulmonary dysplasia (BPD) or congenital anomaly (ICD codes 425.3x, 425.4x, 425.8x, 745.xx, 746.xx, and 747.xx), discharge from birth hospitalization during the California RSV season (November through March), the number of siblings present in the home (≥1 sibling <5 years of age), and hospitalization for an unspecified respiratory illness (ICD codes 480.8, 480.9, 481.xx, 482.xx, 483.xx, 484.xx, 485.xx, and 486.xx) categorized as none, uncomplicated hospitalization, and prolonged hospitalization. For hospitalization records, we also determined the total length of stay, which included concatenating records in cases in which interhospital transport occurred, and whether the infant experienced assisted ventilation (ICD codes 93.90, 93.91, and 96.7x).

We combined electronic data elements to create 4 composite variables: small for gestational age (SGA), neonatal oxygen exposure, family history of asthma, and membership status. We classified infants as being SGA (<5th percentile) using birth weight and GA according to the algorithm developed by Brenner et al.¹⁹ We fitted a smooth cubic spline to predict the probability of RW as a function of total oxygen exposure separately for births occurring at 32 to 37 and 37 weeks' or greater GA. Both curves showed a flat effect for total oxygen exposure at less than 200 hours and a clear increase at more than 200 hours, suggesting a step function and no interaction between GA and total length of oxygen therapy. For infants treated in the neonatal intensive care unit, we created an oxygen exposure variable: no supplemental oxygen exposure during the neonatal period, no BPD; supplemental oxygen exposure of 1 to less than 200 hours during the neonatal period, no BPD; supplemental exposure 200 or more hours during the neonatal period, no BPD; and BPD. Infants who did not require intensive care as newborns were classified as having no supplemental oxygen exposure. Parental asthma history was established by scanning the parent's records for the period 18 months before to 6 months after the infant's birth and determining whether the parent had 2 or more clinical visits 14 days apart with an ICD code 493.xx for asthma and/or the parent's electronic record listed asthma on their significant problem list. The parents' significant problem list was scanned to determine whether either parent listed smoking as a problem during the infants' first year of life. We linked the children's use and membership data and created a monthly membership variable that was set equal to 1 (patient was a member during that month and/or had outpatient or inpatient use during that month) or 0 (patient was not a member during that month and did not have outpatient or inpatient use during that month) to determine eligibility for inclusion in the study cohort.

The KPMCP's information systems use a common medical record number for all inpatient, outpatient, and administrative encounters in all facilities, permitting extraction and linkage of patient data using methods that have been previously described.^20- 23 Care received outside of the program was also tracked. We identified infants using the KPMCP hospitalization database and the Neonatal Minimum Data Set,²⁴ a research database that captures supplemental oxygen use and assisted ventilation among infants receiving neonatal intensive care. These records were linked to patient demographic, membership, laboratory, pharmacy, outpatient encounter, and hospitalization databases to obtain the previously described variables. To ensure data quality, we audited any questionable laboratory test results, clinical encounters, or prescription patterns and manually corrected records when necessary. For comparison with the general US population, we randomly sampled patient records to estimate daycare use and passive smoke exposure in the first year of life. We chose 165 infants of 37 weeks' or less GA who did not have RSV infection in the first 6 months of life and 165 who did, as well as 55 infants of 38 or more weeks' GA with RSV infection and 55 without in the first 6 months of life, for a total of 440 infants. After manual medical record abstraction, sufficient data for analysis were available in 417 of the 440 records. Randomly selecting 1 infant per mother left 406 records (152/165 infants ≤37 weeks' GA with RSV; 152/165 infants ≤37 weeks' GA without RSV; 49/55 infants ≥38 weeks' GA with RSV; and 53/55 infants ≥38 weeks' GA without RSV). We estimated prevalence rates factoring the differential sampling fractions used.

Statistical analyses were conducted using SAS (SAS Institute, Cary, North Carolina). Recurrent wheezing odds ratios (OR) for all predictors were calculated using logistic regression²⁵ with RW in the third year of life as the outcome and each of the predictors as the only explanatory variable. Values of categorical variables were compared using χ² or Fisher exact test. Normally distributed continuous variables were compared using the t test. Comparisons of nonnormally distributed continuous variables used the Wilcoxon rank sum test.

We conducted 4 multivariate analyses to estimate the effect of RSV infection and other predictors on RW. Our model predictors were selected before the study began and were based on biologic plausibility and previous reports on RSV infection. These analyses used logistic regression and Cox proportional hazards modeling²⁵ (to account for censoring of infants who were Kaiser Foundation Health Plan Inc members for ≥9 months during their first year of life but who were members for <9 months during their third year) and 2 RSV exposure variables (RSV infection in the first 6 and 12 months of age).

The relative contribution of each predictor was calculated using the differences between the log likelihood of the full model and the log likelihood of a model without each of the predictors and was defined as the ratio of its log likelihood difference to the sum of the likelihood differences from all predictors times 100.²⁶

To account for the clustering of multiple deliveries and multiple births by the same mother, we calculated robust standard errors for a mother effect using the robust sandwich variance estimates from Lin and Wei²⁷ in the Cox regression. The results from this model were compared with those from a Cox regression model in which a single infant was selected from mothers with multiple deliveries during the study period and from mothers with multiple gestations. Because clustering had little effect on the final results, we present only the results from our logistic regression model involving our cohort with only 1 observation per mother and only 1 infant per mother.

To address underdiagnosis of RSV, we conducted an analysis in which we simulated what might have occurred had all infants been tested. In this analysis, we simulated the effects that universal testing might have had on our findings, assuming the same proportion of positive results among those not tested and also by varying the positivity rate by +10% and −10%. The results of this simulation can be found in eAppendix 2(http://www.archpediatrics.com).

Scanning of KPMCP databases identified 126 837 infants who survived birth hospitalization to discharge from 1996 through 2002 (Figure 1). We excluded 277 infants whose records had missing data and 1803 infants younger than 32 weeks' GA, leaving a total of 124 777 infants. To avoid statistical nonindependence owing to multiple deliveries per mother and multiple births per delivery, we identified all maternal records in which a woman gave birth to more than 1 infant during the study period and randomly selected a single delivery to be retained in our study cohort, excluding 19 472 infants. From all remaining deliveries with a multiple gestation, 1 infant was randomly selected to be retained in our study cohort, thereby excluding 1731 infants, leaving a cohort of 103 574 infants of 32 weeks' or greater GA. An additional 32 472 infants were excluded for not meeting membership criteria, leaving the final study cohort that consisted of 71 102 infants of 32 weeks' or greater GA.

Table 1 summarizes the characteristics of the study cohort and shows that it is ethnically diverse, with a slight preponderance of boys (51%). Most mothers (73.8%) were aged 18 to 34 years. From the mothers' records, 3.7% of infants had a family history of asthma.

Table 2 provides key bivariate comparisons between infants with and without RW and unadjusted odds of RW. Laboratory-confirmed, medically attended infection with RSV, prematurity, and oxygen exposure during the neonatal period were associated with increased risk of RW. Other predictors significantly associated with increased risk were male sex, SGA, African American or Asian race, family history of asthma, congenital anomaly, infection with other pathogens described previously, or hospitalization with an unspecified respiratory illness. Only 1 predictor (≥41 weeks' GA) was associated with significantly decreased risk (unadjusted OR, 0.88; 95% confidence interval [CI], 0.80-0.98). In other analyses (not shown), the season of hospital discharge was not associated with RW; therefore, it was not included in our final multivariate models. The rate of RW among premature infants in our cohort (32-36 weeks' GA) with an RSV infection was 16.23%; the rate among those who did not have RSV infection was 6.22% (unadjusted OR, 2.92; 95% CI, 1.88-4.55); and the proportions among infants of 37 weeks' or greater GA with and without RSV infection were 12.56% and 4.36%, respectively (unadjusted OR, 3.15; 95% CI, 2.61-3.80). The rate of RW among all infants with unspecified respiration-related hospitalization was 16.31%.

We performed 4 multivariate analyses based on the period for the exposure of interest (medically attended RSV infection in the first 6 or 12 months of life) and the methodology used (logistic or Cox proportional hazards regression). Because of the similarity of the results of these analyses, only the results of our logistic regression analyses that used the 12-month exposure period are presented here; the other 3 analyses are presented in eAppendix 1.

In our multivariate analyses, after controlling for the other predictors in the model, SGA status, maternal age, presence of siblings younger than 5 years in the household, and having a congenital anomaly were not significantly associated with RW (Table 3 and Table 4; Figure 2 and Figure 3). A GA of 41 weeks or greater was protective (adjusted OR [AOR], 0.90; 95% CI, 0.81-0.99). Other predictors with a statistically significant association with RW included RSV infection, decreasing gestation, any neonatal oxygen exposure, male sex, African American or Asian race, family history of asthma, and hospitalization for an unspecified respiratory illness. Infection with the other listed pathogens was not a consistent predictor for RW after controlling for covariates.

Infants with RSV infection involving uncomplicated (AOR, 4.66; 95% CI, 1.61-2.67) and prolonged hospitalization (AOR, 3.42; 95% CI, 2.01-5.82) were at greater risk for RW than those with RSV infection involving an outpatient encounter (AOR, 2.07; 95% CI, 1.61-2.67) compared with infants without any visit for RSV (Figure 2). The proportion of infants with RW increased with decreasing GA at birth, beginning at 37 weeks' GA (Figure 3).

Although many predictors reached statistical significance, the overall discrimination and explanatory power of our model were low. The area under the receiver operator characteristic curve for the model was 0.62, with a Hosmer-Lemeshow P value of .88. In terms of explanatory power, the 3 predictors that contributed the most to the total model χ² were infant sex (33%), family history of asthma (23%), and RSV infection (20%) (Table 4).

Based on our medical record review of randomly selected charts, daycare use during the first year of life was 6.4% (95% CI, 1.1-11.8%), while the prevalence of smokers in the home during the first year of life was 5.0% (95% CI, 0.0-10.3%). Review of parents' electronic records showed that 1 or more parents smoked in 3.2% of the study families during the 6 months preceding the birth of the study infant, and in 3.1% of families during the first and second 6 months of life.

In this retrospective cohort study, we found that 3 important risk factors—RSV infection, moderate prematurity, and exposure to supplemental oxygen in the neonatal period—have a statistically significant association with the development of RW in the third year of life. The effect of decreasing GA becomes apparent at 37 weeks' GA, while 41 weeks' or more GA was protective. This study supports the growing body of literature indicating that increased morbidity is associated with moderate prematurity. In addition, male sex, family history of asthma, African American or Asian race, and unspecified respiratory hospitalization were associated with an increased risk of RW. Several variables that were considered to be associated with increased risk of RW (siblings aged <5 years of age in the home, SGA status, and congenital anomalies) were not statistically significant after controlling for the other variables in our model.

Henderson et al¹³ found that approximately 28.1% of infants with RSV lower respiratory tract infection during the first few months of life developed RW by 30 to 42 months of age. Similarly, 30% and 43% of young infants with RSV had asthma by 7 and 13 years of age, respectively, in studies by Sigur et al.^8,12 A more recent study by Wu et al²⁸ demonstrated a causal association of infant age at the peak of winter virus season and the development of early childhood asthma, providing theoretical insight on RSV as an important, modifiable, environmental risk factor and highlighting the importance of preventing RSV disease.

The effect of RSV severity on asthma during early childhood has been an area of special interest. From this perspective, our study findings should be considered in light of a recent population-based retrospective study of healthy full-term infants from the Tennessee Medicaid program by Carroll et al,²⁹ which identified an association between bronchiolitis in early infancy and asthma by 5.5 years of age. Compared with infants with no bronchiolitis visit, infants with outpatient, emergency department, and hospital bronchiolitis visits had AORs of 1.86, 2.41, and 2.82, respectively, for the development of early childhood asthma. The study, however, was limited by its focus on healthy, full-term infants, excluding preterm infants. It also did not include laboratory-confirmed cases of RSV infection, nor did it address the possible effects of other viruses (eg, adenovirus). Our findings clearly complement and expand on these results by examining the contribution of medically attended, unspecified respiratory hospitalizations to the risk of RW. Many of these hospitalizations may be because of RSV infection, and the results were consistent with our observations for laboratory-confirmed, medically-attended RSV infection. Because our databases included GA in weeks, they also permitted expanding on the findings of Boyce et al,¹⁷ one of the few large studies on RSV that included late preterm infants.

To our knowledge, this is the first study to document an association between oxygen exposure in infants of 32 weeks' or greater GA and development of RW in early childhood. While the dangers of oxygen exposure among very premature infants have been known for years, the importance of oxygen toxicity in more mature newborns is only starting to be appreciated. For example, Spector et al³⁰ documented an association between neonatal oxygen exposure and cancer, and Vento et al^31- 32 found that room air resuscitation may be safer for asphyxiated newborns than that using 100% oxygen. Concern about such toxicity has led the American Academy of Pediatrics to endorse new guidelines that permit using room air for resuscitation of newborns instead of 100% oxygen,³³ part of a more widespread reappraisal of the potential toxicity of oxygen.³⁴ Given the observational nature of our study, we cannot exclude that the association with oxygen exposure is the result of confounding by severity of illness (ie, RW could be due to another mediating factor that manifests with some form of respiratory difficulty requiring oxygen in the neonatal period).

Although intriguing, the other associations we found do not constitute causality, and some are difficult to distinguish from possible confounding factors, as we note with respect to oxygen exposure. Importantly, almost two-thirds of our model's explanatory power came from 3 factors that may have a strong genetic and/or epigenetic component: sex, family history of asthma, and race. This suggests that another factor—which this study could not address—may be playing an important role: infants whose response to RSV infection is severe enough to come to medical attention may represent a population genetically predetermined to be hypersensitive to developing bronchoconstriction when confronted with a trigger.

The major limitation of our study is that not all infants were tested for RSV (for example, among hospitalized infants with respiratory illnesses, 43% were not tested). Given the absence of systematic testing for respiratory pathogens, we cannot exclude the possibility of underdiagnosis in this retrospective study. While we suspect that lack of such testing leads to underestimation of the effect of RSV, we cannot exclude the presence of other causal factors including genetic ones and ones mediated by other respiratory illnesses such as Rhinovirus infection. More research will be needed to understand the role of RSV, a potentially modifiable risk factor, in the causal pathway for RW.

Although our model's discrimination was relatively low, it considers many factors that are known predictors of asthma or that may confound the RSV effect. All of these predictors had statistically significant AORs, and because the model parameters are not routinely reported by other investigators, it is difficult to compare our results with previous studies.

Our study was conducted in a unique health care setting, which may limit generalizing our study's findings to other populations. For example, in our population, only 6.4% of the infants younger than 1 year were in some form of organized daycare, less than the 15% rate reported by the most recent US Census Bureau report.^35(pp70-86) Moreover, Northern California KPMCP is unusual in having a special program (Early Start) that targets perinatal substance abuse, including smoking.^36- 38 Our estimate of the prevalence of tobacco exposure during the first year of life (3.1%) is much lower than reported elsewhere in the country, where rates of smoking soon after pregnancy range between 7.3% and 39.3%^39- 40 but is consistent with recent internal surveys conducted by the Early Start program that reported a smoking rate of 4.2% among pregnant women in the KPMCP in 2008 (M. A. Armstrong, written communication, August 2009).

This retrospective study suggests that GA is an important risk factor for developing RW. Premature infants have a greater risk of developing RW than full-term infants after RSV infection. Furthermore, infants with more severe RSV infections, particularly those who require hospitalization, had a higher risk of developing RW in early childhood compared with infants with less severe infection. Similarly, neonatal use of supplemental oxygen was associated with increased risk of RW. Several genetic and environmental factors in our study (including race, smoking in the home, and daycare attendance) were associated with an increased risk of RW and may provide promising areas for more detailed research that should include follow-up extending beyond the third year of life.

Correspondence: Gabriel J. Escobar, MD, Division of Research, Systems Research Initiative and Perinatal Research Unit, Kaiser Permanente Medical Care Program, 2000 Broadway, 2nd Floor, Oakland, CA 94612 (gabriel.escobar@kp.org).

Accepted for Publication: March 10, 2010.

Author Contributions: Dr Escobar was the principal investigator and had full access to all of the data in the study and supervised data collection, data cleaning, and, in conjunction with Dr Kipnis, the project statistician, supervised all project analyses. Study concept and design: Escobar. Acquisition of data: Escobar, Ragins, and Li. Analysis and interpretation of data: Escobar, Li, Prager, Masaquel, and Kipnis. Drafting of the manuscript: Escobar, Ragins, Masaquel, and Kipnis. Critical revision of the manuscript for important intellectual content: Li, Prager, and Kipnis. Statistical analysis: Escobar, Masaquel, and Kipnis. Obtained funding: Escobar. Administrative, technical, and material support: Escobar, Ragins, Li, and Prager. Study supervision: Escobar, Ragins, and Kipnis.

Financial Disclosure: Dr Masaquel is an employee of MedImmune LLC and receives an incentive plan.

Funding/Support: This study was supported by a contract with MedImmune LLC.

Role of the Sponsor: MedImmune LLC was involved in the design of the study, interpretation of the data, and review of the manuscript. All authors agreed on the final text and conclusions of the manuscript, which was also reviewed by Joseph V. Selby, MD, MPH, Director of the Kaiser Permanente Division of Research, and Parthiv J. Mahadevia, MD, MPH, Senior Director, Health Outcomes and Pharmacoeconomics, MedImmune LLC.

Previous Presentations: Portions of this study were presented at the Hot Topics in Neonatology Conference; December 9, 2008; Washington, DC; the Society for Pediatric Research meetings; May 5, 2009; Baltimore, Maryland; the Kaiser Permanente Pediatrics Conference; July 15, 2009; Kauai, Hawaii; and the Pediatric Academic Society Annual Meeting; May 1-4, 2010; Vancouver, Canada.

Additional Contributions: We also wish to thank Jennifer Graff, PharmD, for assistance during the initial phases of the study, and Dennis Andaya, RHIT, for performing medical record review. We acknowledge Gerard P. Johnson, PhD, John E. Fincke, PhD, and Susan E. DeRocco, PhD, from Complete Healthcare Communications Inc for editorial support of the manuscript.