Influenza Vaccine Effectiveness in the Inpatient Setting: Evaluation of Potential Bias in the Test-Negative Design by Use of Alternate Control Groups

Abstract

The test-negative design is validated in outpatient, but not inpatient, studies of influenza vaccine effectiveness. The prevalence of chronic pulmonary disease among inpatients can lead to nonrepresentative controls. Test-negative design estimates are biased if vaccine administration is associated with incidence of noninfluenza viruses. We evaluated whether control group selection and effects of vaccination on noninfluenza viruses biased vaccine effectiveness in our study. Subjects were enrolled at the University of Michigan and Henry Ford hospitals during the 2014-2015 and 2015-2016 influenza seasons. Patients presenting with acute respiratory infection were enrolled and tested for respiratory viruses. Vaccine effectiveness was estimated using 3 control groups: negative for influenza, positive for other respiratory virus, and pan-negative individuals; it was also estimated for other common respiratory viruses. In 2014-2015, vaccine effectiveness was 41.1% (95% CI: 1.7, 64.7) using influenza-negative controls, 24.5% (95% CI: -42.6, 60.1) using controls positive for other virus, and 45.8% (95% CI: 5.7, 68.9) using pan-negative controls. In 2015-2016, vaccine effectiveness was 68.7% (95% CI: 44.6, 82.5) using influenza-negative controls, 63.1% (95% CI: 25.0, 82.2) using controls positive for other virus, and 71.1% (95% CI: 46.2, 84.8) using pan-negative controls. Vaccination did not alter odds of other respiratory viruses. Results support use of the test-negative design among inpatients.

Keywords: hospitalization; influenza; test-negative design; vaccine effectiveness.

Figure 1

Flow chart of exclusions for determination of the final sample for the 2014–2015 and 2015–2016 influenza seasons, Michigan. A) 2014–2015 season; B) 2015–2016 season. “Other virus–positive” refers to participants who tested negative for influenza but positive for a different respiratory virus. “Pan-negative” refers to participants who tested negative for all respiratory viruses. In the 2014–2015 season, 7 participants who had an undeterminable polymerase chain reaction result on 1 or more respiratory virus targets and did not test positive for any viruses were classified as influenza-negative but not as either other virus–positive or pan-negative.

Figure 2

Respiratory virus counts by 2-week period during the 2014–2015 and 2015–2016 influenza seasons, Michigan. A) 2014–2015; B) 2015–2016. The percent influenza A–positive (H3N2 in 2014–2015 and H1N1 in 2015–2016) is represented by the dotted line. Each color bar represent counts of the frequency that each virus was detected in a 2-week period. The gray bars indicate counts of individuals who tested negative for all viruses. AV, adenovirus; CoV, coronavirus; EV, enterovirus; HMPV, human metapneumovirus; HPIV, human parainfluenza virus; RSV, respiratory syncytial virus; RV, rhinovirus.

Figure 3

Influenza vaccine effectiveness (VE) estimates using 3 different control groups in the 2014–2015 and 2015–2016 influenza seasons, Michigan. VE was estimated separately for each influenza season using each control group and was calculated as 100 × (1 – odds ratio) comparing case and control patients in Firth’s corrected logistic regression models. Models adjusted for sex, age group (in years: 18–49, 50–64, ≥65), hospital (A vs. B), frailty score, Charlson Comorbidity Index score, days between illness onset and specimen collection, and calendar time of illness onset measured in 2-week windows. “Other virus–positive” refers to participants who tested negative for influenza but positive for a different respiratory virus. “Pan-negative” refers to participants who tested negative for all respiratory viruses. In the 2014–2015 season, 7 participants who had an undeterminable polymerase chain reaction result on 1 or more respiratory virus targets and did not test positive for any viruses were classified as influenza-negative but not as either other virus–positive or pan-negative. CI, confidence interval.

Figure 4

Vaccine effectiveness (VE) against rhinovirus, respiratory syncytial virus (RSV), and pooled noninfluenza viruses in the 2014–2015 and 2015–2016 seasons, Michigan. VE was estimated separately for each influenza season using each virus separately, and was calculated as 100 × (1 − odds ratio) comparing case and control patients in Firth’s corrected logistic regression models. Models adjusted for sex, age group (in years: 18–49, 50–64, ≥65), hospital (A vs. B), frailty score, Charlson Comorbidity Index score, days between illness onset and specimen collection, and calendar time of illness onset measured in 2-week windows. CI, confidence interval.