Last data update: May 06, 2024. (Total: 46732 publications since 2009)
Records 1-2 (of 2 Records) |
Query Trace: Flannery BL[original query] |
---|
Relative and absolute effectiveness of high-dose and standard-dose influenza vaccine against influenza-related hospitalization among older adults - United States, 2015-2017
Doyle JD , Beacham L , Martin ET , Talbot HK , Monto A , Gaglani M , Middleton DB , Silveira FP , Zimmerman RK , Alyanak E , Smith ER , Flannery BL , Rolfes M , Ferdinands JM . Clin Infect Dis 2020 72 (6) 995-1003 BACKGROUND: Seasonal influenza causes substantial morbidity and mortality in older adults. High-dose inactivated influenza vaccine (HD-IIV), with increased antigen content compared to standard-dose influenza vaccines (SD-IIV), is licensed for use in people aged >/=65 years. We sought to evaluate the effectiveness of HD-IIV and SD-IIV for prevention of influenza-associated hospitalizations. METHODS: Hospitalized patients with acute respiratory illness were enrolled in an observational vaccine effectiveness study at eight hospitals in the United States Hospitalized Adult Influenza Vaccine Effectiveness Network during the 2015-2016 and 2016-2017 influenza seasons. Enrolled patients were tested for influenza, and receipt of influenza vaccine by type was recorded. Effectiveness of SD-IIV and HD-IIV was estimated using a test-negative design (comparing odds of influenza among vaccinated and unvaccinated patients). Relative effectiveness of SD-IIV and HD-IIV was estimated using logistic regression. RESULTS: Among 1487 enrolled patients aged >/=65 years, 1107 (74%) were vaccinated; 622 (56%) received HD-IIV and 485 (44%) received SD-IIV. Overall, 277 (19%) tested positive for influenza, including 98 (16%) who received HD-IIV, 87 (18%) who received SD-IIV, and 92 (24%) who were unvaccinated. After adjusting for confounding variables, effectiveness of SD-IIV was 6% (95% confidence interval [CI] -42%, 38%) and that of HD-IIV was 32% (95%CI -3%, 54%), for a relative effectiveness of HD-IIV versus SD-IIV of 27% (95%CI -1%, 48%). CONCLUSIONS: During two U.S. influenza seasons, vaccine effectiveness was low to moderate for prevention of influenza hospitalization among adults aged >/=65 years. High-dose vaccine offered greater effectiveness. None of these findings were statistically significant. |
Re: "Invited commentary: Beware the test-negative design"
Ferdinands JM , Foppa IM , Fry AM , Flannery BL , Belongia EA , Jackson ML . Am J Epidemiol 2017 185 (7) 1 In their commentary, Westreich and Hudgens (1) discussed the article by Sullivan et al. (2), in which the authors offered a theoretical framework for examining mechanisms via which bias and confounding may arise in a vaccine effectiveness (VE) study using a test-negative design (TND). We agree that illustrating the TND using directed acyclic graphs provides useful insight, although Sullivan et al. cautioned that they did not address the real-world magnitude or relative importance of the potential biases. Unlike Westreich and Hudgens (1), we found the TND to be valid under a wide range of conditions, a conclusion that is supported by multiple theoretical, simulation-based, and comparative studies (3–5). | To support their opinions, Westreich and Hudgens conducted a simulation study in which individuals with a certain characteristic C = −1 were more likely to be vaccinated (62% vs. 18%). Those with C = −1 could, for example, be people having a pre-existing condition who are specifically targeted for influenza vaccination. In addition, persons in the C = −1 group were much more likely to become infected than were those in the C = 1 group (92% vs. 3% among the unvaccinated). As a result, the risk of influenza infection was greater among vaccinated individuals (23%) than among unvaccinated individuals (7%), and the “crude VE” was strongly negative, illustrating the extreme confounding designed into the example. Using inputs identical to those of Westreich and Hudgens, we found the same likelihood of infection of 11% and the same likelihood of being tested (and therefore observed in the TND study) of 17%. However, in contrast to Westreich and Hudgens, we found a true causal odds ratio of 0.37, not 0.74, using both simulation and calculation (Web Table 1, available at http://aje.oxfordjournals.org/) (6). After adjustment for the characteristic C in a multivariate logistic regression model, we found estimated odds ratios of 0.40 and 0.38 in simulated analyses using the entire cohort and the TND sample, respectively. Thus, even under the extreme conditions of the example used by Westreich and Hudgens, the TND yields an estimated odds ratio similar to the true causal odds ratio, which, unlike the causal relative risk, is independent of C. Finally, given the implausibly high risk of infection among those with C = −1 in this contrived example, the odds ratio is a poor approximation of the relative risk of disease, and 1 minus the odds ratio is not an asymptotically consistent estimator of VE. Fortunately, real-world applications of the TND, which are unlikely to be conducted under such improbable conditions, rarely suffer the same limitation. |
- Page last reviewed:Feb 1, 2024
- Page last updated:May 06, 2024
- Content source:
- Powered by CDC PHGKB Infrastructure