Childhood Vaccinations and Risk of Asthma

Frank Destefano, MD; David Gu, PhD; Piotr Kramarz, MD; Benedict I. Truman, MD; Michael F. Iademarco, MD; John P. Mullooly, PhD; Lisa A. Jackson, MD; Robert L. Davis, MD; Steven B. Black, MD; Henry R. Shinefield, MD; S. Michael Marcy, MD; Joel I. Ward, MD; Robert T. Chen, MD; The Vaccine Safety Datalink Research Group

Pediatr Infect Dis J. 2002;21(6) 

In This Article


We analyzed data from the Vaccine Safety Datalink (VSD) project, which has been described previously.[19] Briefly VSD was created in 1991 by the National Immunization Program of the CDC. The project links medical event information, vaccine history and selected demographic information from the automated clinical databases of Group Health Cooperative (GHC) in Seattle, WA; Kaiser Permanente Northwest in Portland, OR; Kaiser Permanente Medical Care Program of Northern California (NCK) in Oakland, CA; and Southern California Kaiser Permanente Medical Care Program (SCK) in Los Angeles, CA. The time periods covered by our study were from 1991 through 1996 for GHC, Kaiser Permanente Northwest and NCK and from 1992 through 1997 for SCK. The study was restricted to children who were enrolled in one of the participating HMOs at birth and remained enrolled until at least 18 months of age.

We ascertained vaccinations received by children in the study from computerized immunization tracking systems that are maintained by each of the HMOs. Quality control comparisons of the computerized immunization data with information recorded in paper medical records have shown high levels of agreement.[20] For the common infant vaccines administered at each HMO, 85% or more of the vaccinations recorded in the medical records were usually captured by the computerized immunization tracking systems. NCK had nearly perfect agreement (98 to 99%) for all of the vaccines of interest.

We identified asthma cases using computerized medical encounter and pharmacy databases. Each of the HMOs maintains computerized databases of prescription medication dispensings and of all hospital discharges and emergency room visits. At GHC and NCK, diagnoses from outpatient clinic encounters also were available in electronic databases for certain years. To be classified as having asthma, a child had to meet one of the following criteria: (1) at least one diagnosis of asthma [International Classification of Diseases, ninth revision (ICD9) Code 493] and at least one prescription for an asthma medication; the first diagnosis and first prescription had to be within a 2-year period. Asthma medications included oral or inhaled beta-agonists, theophyllin, oral or inhaled corticosteroids, cromolyn sodium, adrenergic drugs not elsewhere specified and unclassified asthma medications; (2) at least one prescription for an inhaled beta-agonist and at least one prescription for cromolyn within a 2-year period; (3) at least five prescriptions for asthma medications during a 2-year period.

In addition to one of the above criteria, we also required that the child had to have at least one asthma diagnosis or medication prescription when 1 year of age or older. We defined the asthma incidence date as the earliest of the first asthma diagnosis date or the first date of an asthma medication prescription. A child could have had an asthma incidence date when younger than 1 year of age, but to be classified as a case the child had to have an indication that asthma was still present when he or she was older than 1 year of age. We imposed this requirement because of the difficulty in differentiating between asthma and bronchiolitis in infants.

We conducted proportional hazards regression analyses to estimate relative risks of developing asthma according to vaccination status. We used SAS software to perform the analyses (SAS Institute, Cary, NC). The outcome in the regression models was age at asthma incidence. Children were censored when they were first prescribed an asthma medication or received a diagnosis of asthma, disenrolled from HMO membership, or December 31, 1996 (1997 for SCK), whichever occurred first. We modeled vaccination status as a time-dependent variable according to date of first vaccination. We evaluated the following vaccines: DTP; oral polio vaccine (OPV); MMR; Haemophilus influenzae type b (Hib); and recombinant hepatitis B. The regression models were stratified by HMO and by month and year of birth. We included gender and low birth weight as covariables in the models, as well as receipt of acellular pertussis vaccine as the first pertussis vaccination.

We performed a subanalysis to evaluate the influence of race and ethnicity and socioeconomic factors on the results. We did not have direct measures of socioeconomic status or information on race and ethnicity. As indirect indicators of these factors, we used the characteristics of the child's Census block group of residence. From the 1990 Census data (US Census Bureau CD-ROM STF-3B), we used the following Census block variables: education (percentage of residents older than 18 years who did not graduate from high school); household income (proportion of households with annual incomes <$25 000); proportion of African-American residents; proportion of Hispanic residents; and proportion of Asian residents. Census block data were available through 1995 at NCK and through 1996 at the other three HMOs.

To evaluate the influence of our case definition, we performed an additional analysis in which we used a different asthma case definition to detect potentially more severe cases of asthma. In that analysis we required a hospital discharge or emergency room visit diagnosis of asthma as part of the case definition plus at least one prescription for an asthma medication.

We also performed subanalyses to evaluate possible medical care utilization bias. We were primarily concerned that children who had no records of receiving recommended vaccines were not actually using the HMO for most of their medical care. Thus we would be missing not only their vaccinations but also their medical care encounters. This would tend to underestimate the risk of asthma in apparently “unvaccinated” children and result in an overestimate of the relative risk associated with vaccination. To minimize this bias, at two HMOs with outpatient encounter data we performed a subanalysis restricted to children who had at least two medical care encounters during their first year of life. To further evaluate the possible influence of medical care utilization bias on the risk associated with hepatitis B vaccine, we also performed a subanalysis restricted to children who had received at least two OPV, two DTP and one MMR vaccine by 18 months of age.


Comments on Medscape are moderated and should be professional in tone and on topic. You must declare any conflicts of interest related to your comments and responses. Please see our Commenting Guide for further information. We reserve the right to remove posts at our sole discretion.