Geographic Clusters in Underimmunization and Vaccine Refusal

Study Population and Data Collection

Kaiser Permanente of Northern California (KPNC) is a nonprofit, integrated health care delivery system providing care to >3 million members, including ∼30 000 annual births. KPNC provides services at 21 hospital-based medical centers, and >200 medical office buildings,¹¹ spanning 200 miles from Santa Rosa to Fresno and 100 miles from Sacramento to San Jose. In many communities, KPNC is the dominant provider, with ≥40% market share. KPNC patient demographics reflect the general population in the Northern California region, although persons with low education and income are underrepresented.¹²

Immunizations and other clinical information are kept in a fully electronic medical record system. All routine pediatric immunizations are administered to KPNC members at no charge.

In this retrospective cohort analysis, we identified all infants born between January 1, 2000, and December 31, 2009, who had continuous KPNC membership from birth until age 36 months. To focus on children using KPNC for primary care, we excluded those without ≥2 primary care visits by their first birthday. Race/ethnicity was identified via a search of electronic medical records and administrative databases. Using KPNC membership databases, we identified the home address of each child at the time of his or her 36-month birthday. Children were assigned to a census block group from the Unites States 2010 decennial census based on their home address. Census block group statistics were based on the 2006–2010 American Community Survey data collected by the US Census Bureau.^13,14

Definitions of Outcomes and Predictors

Children were defined as “underimmunized” if, by 36 months of age, they missed ≥1 of the vaccines recommended by the Centers for Disease Control’s Advisory Committee on Immunization Practices.^15–17 For years 2000 to 2003, children were expected to have received 3 doses of hepatitis B by their 36 month birthday; 4 doses of diphtheria and tetanus toxoids and acellular pertussis vaccine; 2, 3, or 4 doses (depending on vaccine manufacturer and due date of the booster dose) of Haemophilus influenza type b vaccine (Hib); 3 doses of inactivated polio virus vaccine; 1 dose of measles, mumps, rubella vaccine (MMR); and 1 dose of varicella zoster virus vaccine (VZV). In addition to these, children were expected to receive 4 doses of pneumococcal conjugate vaccine (PCV) beginning in 2004, 1 dose of hepatitis A (HepA) beginning in 2009, and 2 or 3 doses (depending on vaccine manufacturer) of rotavirus vaccine beginning in 2010. Because of vaccine shortages, we did not require children to receive the Hib booster dose if the timing of that dose fell between December 1, 2007, and December 31, 2009.^18,19 To account for the introduction of new vaccines, in our analysis, we did not count a newly introduced vaccine as required until 3 calendar years after it was added to the Advisory Committee on Immunization Practices schedule. Thus, although PCV was added in 2001, we only included it as required for children whose first PCV shot was due after December 31, 2003. Similarly, we only treated HepA vaccine as required for children whose first HepA dose was due after December 31, 2008, and rotavirus for children whose first rotavirus dose was due after December 31, 2009.

As a secondary measure of underimmunization, we used the “combination 3” performance measure used by the Healthcare Effectiveness Data and Information Set (HEDIS). This HEDIS measure included all children who were enrolled in the health plan on their second birthday and who were enrolled for at least 11 of the 12 months before this. Children were considered underimmunized if they did not receive the following immunizations by their second birthday: 4 diphtheria and tetanus toxoids and acellular pertussis vaccine, 3 inactivated polio virus vaccine, 1 MMR, 2 Hib (3 if year 2011 and 2012), 3 hepatitis B, 1 VZV, and 4 PCV.

Some families limit the number of injections at each visit but still have their children receive all required immunizations by 36 months of age. To evaluate this phenomenon, we adopted a definition used in a previous study and classified families as “shot limiting” if a child did not receive >2 injections on any day before their 36-month birthday.²⁰ “Vaccine refusal” was defined as receiving ≥1 diagnoses (International Classification of Diseases, Ninth Revision, Clinical Modification [ICD-9-CM]) denoting vaccine refusal at any time before their 36-month birthday. Vaccine refusal ICD-9-CM codes used at KPNC during this time were V6405, V6406, and V6407. Because the use of these codes was rare before 2007, analyses of vaccine refusal were restricted to children born between January 1, 2007, and December 31, 2009.

Statistical Analyses

To determine if there was a trend over time in the proportion of children who were underimmunized, we created an analytic data set with a record for each calendar year from 2003 to 2012 with a variable indicating the number of children turning 36 months of age that year who were underimmunized, a variable indicating the total number of children turning 36 months of age that year, and a numeric variable indicating the year. A Poisson regression was run in which the dependent variable was the number of underimmunized children in the year, the independent variable was the calendar year, and the offset variable was the log of the number of eligible children that year. Because trends in underimmunization could either be related to increased vaccine skepticism or to the introduction of additional vaccines into the schedule (which could increase the probability of underimmunization even if all vaccines have a fixed probability of being missed), we repeated the Poisson regression ignoring PCV, HepA, and rotavirus, which were introduced during the study period.

To assess individual-level demographic and neighborhood-level predictors of underimmunization, we performed logistic regression in which the dependent variable was whether the child was underimmunized, the individual-level independent variables were calendar year, gender, and race/ethnicity, and the neighborhood-level variables (proportions of the block group population) were race/ethnicity, Spanish speakers, households with income below the poverty level, housing units occupied by owners, and education (less than high school, and graduate or professional degree).

Geographic clusters of underimmunization, shot-limiting, and vaccine refusal were identified by using the spatial scan statistic and the free SaTScan software.²¹ The Bernoulli version of the spatial scan statistic was used to detect and evaluate geographic clusters with a high proportion of cases (eg, underimmunized children) versus controls (eg, fully immunized children) at a statistical significance level of P < .05. This is done by gradually scanning a circular variable size window across space, noting the number of cases and controls inside each of many thousands of evaluated circles, and calculating the likelihood for each. The circle with the maximum likelihood is the most likely cluster, that is, the cluster least likely to be due to chance. Secondary clusters are also identified. Using Monte Carlo hypothesis testing, the P value assigned to each cluster is adjusted for the multiple testing inherent in the large number of circles evaluated. The SaTScan option to report only clusters smaller than 10% of the population was invoked, as was the option to report only nonoverlapping clusters.

The spatial scan statistic analyses were performed unadjusted and adjusted for the child’s ethnicity (white, Asian, Hispanic, other) and census block group income (low, middle, high tertiles). For the adjusted analyses, we used the multiple data sets approach in SaTScan by using separate data files for each of the 12 race-by-income strata.