We analyzed the public-use files of the National Immunization Survey (NIS) from 1999 to 2003. Problems with duplicate entries and unclear reporting rules with combination vaccines were addressed beginning with the 1999 NIS dataset.16
The National Center for Immunization and Respiratory Diseases of the Centers for Disease Control and Prevention sponsors these surveys, and the National Center for Health Statistics conducts them. The methods of these annual surveys have been published elsewhere.19
In brief, these are validated, stratified, random-digit-dialed telephone surveys of households with children 19–35 months of age. Information is collected through computer-assisted telephone interview techniques. Immunization information is collected directly from the identified immunization providers for the surveyed children. Adjustments to design variables are made by the Centers for Disease Control and Prevention for biases resulting from nonresponse and non-telephone households. The 1999–2003 surveys included 111,730 children, representing a cohort of 5,756,583 children (the average population of children 19–35 months of age during the five-year period in the U.S.).
Our main outcome variable was extra-immunization defined as present or absent. We used the provider-based record data available from NIS to assess the frequency of extra-immunization. We defined extra-immunization to allow the largest number of vaccines and based our definition on each year's published immunization schedule, including the minimum interval schedule between doses for each vaccine (i.e., the “catch-up schedule”).20–24
We defined extra-immunization with diphtheria and tetanus toxoids and pertussis vaccine, whole or acellular (DTxP) and Haemophilus influenzae type b vaccine as >4 doses each. Doses counted as DTxP included diphtheria and tetanus toxoids, diphtheria-tetanus-acellular pertussis (DTaP), and diphtheria-tetanus-pertussis (DTP). We defined extra-immunization with polio vaccine as >3 doses of either inactivated polio vaccine (IPV) or oral polio vaccine in any combination. We defined extra-immunization with hepatitis B (Hep B) vaccine as >3 doses from 1999 to 2001 and four doses from 2002 to 2003, when the first Hep B vaccine was given in the initial week of life. (Beginning in 2002, the Advisory Committee on Immunization Practices [ACIP] provided permissive language allowing four doses of Hep B vaccine to support newborn immunization and the subsequent use of combination Hep B vaccines.)
We defined extra-immunization with a measles-containing vaccine as >2 doses occurring on or after 12 months of age (discounting any dose before 12 months of age, the catch-up schedule and ACIP measles-mumps-rubella vaccine recommendation allows two doses separated by 28 days occurring after 12 months of age).25
We defined extra--immunization with varicella zoster vaccine as >1 dose. Other than the initial doses of Hep B and measles-containing vaccines, as explained previously, we did not test for initial dose or minimum interval violations. Because of the large amount of extra-immunization specific to polio vaccine that may have resulted for reasons peculiar to the polio vaccine and concurrent changes in recommendations for that vaccine during the period examined, we constructed another outcome variable that did not include polio vaccine extra-immunization.
Medical home served as the main exposure variable. As a proxy for medical home, we used the NIS-collected number and type of immunization providers. Immunization provider is not necessarily an individual clinician but, rather, one reporter of immunizations (e.g., an office or clinic). The NIS classified immunization providers by both number and facility type. We constructed a composite variable to capture both multiple providers and multiple facility types (e.g., public, private, hospital, military, mixed, other, unknown, and missing). If the record listed only one immunization provider and the provider facility type was missing (about 11% were missing), then we assumed that child had only “one provider, one facility type.” If the record indicated more than one provider but only one facility type or facility type missing, then we coded that child as having “multiple providers, same facility type.” If the record explicitly listed the child as having multiple facility types, then we coded this as “multiple facility types” regardless of how many providers were indicated.
For race/ethnicity, we used a composite race variable defined by the NIS as Hispanic, non-Hispanic white, non-Hispanic black, and non-Hispanic other. We categorized those respondents indicating multiple races/ethnicities in the “all others” category if they did not indicate Hispanic ethnicity. We also constructed a summary variable that dichotomized race/ethnicity into non-Hispanic white vs. nonwhite.
NIS data provided three potential socioeconomic status variables: poverty status, household income, and maternal education. Because maternal education had full reporting and the other variables had substantial missing values, we chose maternal education as the measure of socioeconomic status for our analyses.
The NIS data contained a variable reporting whether the child's vaccine providers reported vaccinations to a state or community immunization registry. This variable was coded in NIS data as “all providers,” “some but not all providers,” “no providers,” and “unknown.”
We modeled the effect of the presence of a medical home on extra-immunization using logistic regression. We selected the variables for further analyses based on magnitude of the effect of that variable on extra-immunization in the bivariate analyses. The final model included medical home (provider number and facility type), race/ethnicity, survey language, survey year, maternal education, and parent-held immunization records. We used adjusted odds ratios (AORs) to report the results of analyses controlling for variables in the full model. The final multivariable model did not include vaccine registry use because of the large number of unknown values over the years, averaging 21.6% and ranging from 38.6% in 1999 to 16.5% in 2001.
Each year's NIS dataset included weights appropriate for inferences to the population of children 19–35 months of age in that year in the U.S. To analyze our five-year dataset while avoiding overweighting of observations, we divided the weight for each observation by five. This had the effect of making the weighted dataset the average of the target population of children during the time period studied.
We performed our data extraction and recoding using SAS® version 9.126
and conducted analyses appropriate for this multistage, complex survey using Stata® version 8.0.27
Stata permitted the inclusion of the survey design variables into the analysis and, thus, addressed the complex sampling appropriately to achieve the best approximate variances for population estimates. All rates reported, unless otherwise stated, were weighted to reflect population-based estimates.