Influenza Vaccine Effectiveness Against Pediatric Deaths: 2010–2014

Influenza-Associated Pediatric Deaths

To estimate influenza vaccine effectiveness (VE) against death, we used a case–cohort design (or screening method).^9,10 Cases were deaths in US residents aged <18 years with laboratory-confirmed influenza virus infection that were reported to the Influenza-Associated Pediatric Mortality Surveillance System.³ Data obtained from standardized case report forms included demographic characteristics, state of residence, results of laboratory testing, underlying high-risk medical conditions, and influenza vaccination status. Confirmation of influenza virus infection consisted of clinical diagnostic testing during illness or postmortem examination including antigen detection, viral culture, or nucleic acid amplification. We included children ≥6 months old on November 1 of the influenza season during which death occurred who were eligible to receive ≥1 dose of seasonal influenza vaccine and had complete medical history; 10 deaths among children who turned 6 months old after November 1 were excluded. For cases without illness onset date, we estimated onset as 4 days before death, based on the mean duration of illness among cases with known onset date. We excluded 7 children who received their first dose of current season vaccine <14 days before illness onset because vaccine-induced protection may be incomplete.¹¹

Health department personnel determined vaccination status based on a review of available information from the child’s medical records, health care provider, state or county immunization information systems, parental report, or coroner’s report. Children were considered vaccinated (either fully or partially) if records indicated receipt of ≥1 doses of current season influenza vaccine ≥14 days before onset of illness. Partial vaccination was defined as receipt of 1 dose when health department personnel determined that 2 doses were indicated. Vaccination status was listed as unknown if surveillance personnel could not determine from available records whether the child had received influenza vaccine. If immunization records indicated vaccination for previous seasons but no current season vaccination, children were assumed to be unvaccinated. A sensitivity analysis explored the effects of assuming that children with unknown vaccination status were unvaccinated. Presence of underlying medical conditions was recorded on case report forms according to high-risk conditions for influenza-related complications identified by the Advisory Committee on Immunization Practices (ACIP)¹²: asthma, chronic lung disease, neurologic or neurodevelopmental disorders, heart disease (including congenital heart disease), blood disorders, endocrine disorders, metabolic disorders, kidney disorders, liver disorders, immunosuppression, and pregnancy. For consistency with vaccination coverage estimates, age groups were defined based on child’s age on November 1 for influenza season of fatal illness.

Influenza Vaccination Coverage in Cohorts

We used 3 sources of influenza vaccination coverage among comparison cohorts of US children and adolescents: National Immunization Survey–Flu (NIS-Flu),¹³ National Health Interview Survey (NHIS),¹⁴ and the MarketScan Commercial Claims and Encounters database (Truven Health Analytics, Ann Arbor, MI). NIS-Flu is a national telephone survey of households with children aged 6 months through 17 years that produces national and state-level estimates (and for some cities, including Chicago and New York City) of influenza vaccine coverage for children. NHIS is an in-person household survey that provides nationally representative estimates of influenza vaccination for children aged 6 months through 17 years with and without specified high-risk medical conditions (cystic fibrosis, sickle cell anemia, diabetes, congenital heart disease or other heart condition, cerebral palsy, muscular dystrophy, or seizures, or an asthma episode in the past 12 months).¹⁵ In NIS-Flu and NHIS, seasonal influenza vaccination coverage is estimated from parental report of child’s receipt of ≥1 dose of seasonal influenza vaccine. Data from NIS-Flu and NHIS are weighted by age, sex, race or ethnicity, and geographic area to represent the US population. Monthly coverage estimates and SEs for vaccination received during July or August (depending on survey and season) through May were calculated by Kaplan–Meier survival analysis of data from interviews conducted from August, September, or October through June.¹⁶ Age group was based on the child’s age on November 1 of each influenza season.

In the MarketScan Commercial Claims and Encounters database, coverage estimates were based on commercial insurance claims for influenza vaccination; children with no vaccination claim were assumed unvaccinated. We analyzed data from July 2010 through June 2014 among children continuously enrolled in a health plan from January preceding each influenza season. We determined underlying high-risk medical conditions based on International Classification of Diseases, Ninth Revision codes associated with hospitalizations and outpatient visits during a 12-month period.¹⁷ We calculated cumulative vaccination coverage from August through May each season among children aged <18 years with and without underlying high-risk medical conditions. Age categories were based on child’s age at vaccination or at the beginning of each calendar year; children in the youngest age category (0–4 years) were ≥10 months old on November 1. For MarketScan comparison cohorts, state-level coverage estimates were used for children without high-risk conditions, and national coverage estimates were used for children in high-risk groups because of their smaller sample sizes.

Analysis

The case–cohort method^9,10 produces an odds ratio (OR), which estimates the relative risk for influenza-associated death among vaccinated versus unvaccinated children. Influenza VE was estimated as (1 − OR). ORs were obtained from logistic regression models where vaccination status of the case was the dependent variable and the log odds of vaccination (proportion vaccinated/1 − proportion vaccinated) in the comparison cohort from NIS-Flu, NHIS, or MarketScan data were entered as an offset. With this offset in the model, the intercept provides an estimate of the log(OR). Cases and corresponding coverage estimates were stratified by influenza season, age, and state of residence (for NIS-Flu and MarketScan) or high-risk condition (for NHIS and MarketScan) for the month ending ≥2 weeks before case illness onset. For average coverage estimates by age group, season, or influenza virus type, we derived confidence limits by averaging the variance of independent, normally distributed coverage estimates. To incorporate uncertainty in coverage estimates from survey data, 10 000 estimates were sampled from normally distributed vaccination coverage estimates for NIS-Flu and NHIS comparison cohorts to calculate 95% credible intervals (CIs) for VE estimates. Because MarketScan estimates were derived from all observations without sampling, confidence limits for VE estimates based on MarketScan data were estimated from logistic regression assuming fixed coverage estimates. We assumed that cases did not contribute uncertainty to VE estimates. Sensitivity analyses were conducted under the assumption that cases with unknown vaccination status were unvaccinated and that coverage estimates based on parental report underestimated coverage by 10% or overestimated coverage by 10% or 20%.^18,19 χ² tests were used to evaluate proportions, and 2-sided P values <.05 were considered significant. Analyses were conducted in SAS version 9.4 (SAS Institute, Inc, Cary, NC) and R (version 3.1.1; R Core Team, Vienna, Austria).