Validating & Measuring Influenza Vaccine Effectiveness among and against Cardiovascular Hospitalization

Abstract

Background: Influenza is a viral respiratory infection that causes serious health outcomes such as hospitalization and death and represents an important public health burden globally. Vaccination is one of the most effective interventions to prevent influenza and its complications. Although administrative databases such as pharmacy billing claims are used to measure influenza vaccination status, little is known of the validity of these databases in Ontario. Patients with cardiovascular disease (CVD) may have an altered immune response, despite being at a higher risk of influenza complications. It is not known if the Vaccine Effectiveness (VE) in this population is comparable to the general population. Further, there is a lack of research in Canada evaluating influenza vaccine effectiveness against cardiovascular outcomes, particularly using robust study designs such as the test-negative design. Three interrelated studies were carried out. First, a validation study was completed to examine the accuracy of the combination of Ontario’s administrative data from pharmacy and physician billing claims in identifying an individual’s vaccination status. The second study assessed the influenza VE against laboratory-confirmed influenza among older adults hospitalized with CVD conditions in Ontario and examined sex and age group as potential effect modifiers in the association between influenza vaccination and laboratory-confirmed influenza. The third study measured influenza VE against acute CVD outcomes using the Test-Negative Design (TND) for the first time among older adults in Ontario who were hospitalized within three days of their influenza testing. Methods: In the first study, I validated the combined physician and pharmacy billing claims within administrative databases using the linked reference standard of self-report data from the Canadian Community Health Survey (CCHS). This study estimated sensitivity, specificity, Positive Predictive Value (PPV), and Negative Predictive Value (NPV), with 95% Confidence Intervals (CI) of the estimates. In the second study, I used the TND to measure influenza VE against laboratory-confirmed influenza. The analysis included both crude and adjusted estimates accounting for potential confounders (sex, age group, neighbourhood income quintile, rurality, number of outpatient visits, beta-blocker medication use, statin medication use, receiving home care services, and influenza testing relative to the peak month of the season, and year of influenza season). In addition, I assessed effect modification of sex and age group on influenza VE by using two different methods: introducing interaction terms into the model and stratification. In the third study, I examined influenza VE against acute CVD outcomes, also using the TND. The exposure of interest was vaccination status, cases were those hospitalized for myocardial infarction, unstable angina, or stroke and testing positive for influenza, and controls were those hospitalized for a non-cardiovascular event and testing negative for influenza. Both crude and adjusted VE were estimated. Sensitivity analyses were performed to test the robustness of the findings by varying inclusion criteria, outcome definitions, and statistical adjustments. Results: In the first study, the CCHS identified 43% as vaccinated across the two survey cycles of 2013 and 2014. The sensitivity for the combined pharmacy and physician billing codes was 60.1% (95% CI 59.3%–61.0%), specificity was 98.5% (95% CI 98.3%–98.7%), PPV was 96.7% (95% CI 96.3%–97.1%) and NPV was 76.9% (95% CI 76.4%–77.5%). The second study included 1,159 patients, and almost half were vaccinated. Among vaccinated patients, 14% tested positive for influenza compared to 20% of unvaccinated patients.. Crude and adjusted VE were 32% [95% CI, 8–50%] and 43% [95% CI 20%–60%], respectively. Neither inclusion of interaction terms separately in the full model, nor stratification by sex or age group, revealed any evidence of effect modification. The third study included 33,710 hospitalized individuals who tested positive for influenza and had a CVD event (cases) or tested negative and did not have a CVD event (controls). There were 18,519 vaccinated individuals (55%). Among vaccinated patients, 0.4% tested positive for influenza and were hospitalized for a cardiovascular event, compared to 0.8% of unvaccinated patients. The adjusted influenza VE against cardiovascular outcomes was 43% [95% CI 25%–58%; p-value =0.0001]. Conclusion: Compared to past studies with only physician billing claims, the validation study provided improved performance measures of sensitivity, specificity, PPV and NPV values in the combined physician and pharmacy billing claims in identifying individual vaccination status in Ontario. The second study estimated influenza VE against laboratory confirmed infection and supports that influenza VE among older adults with CVD hospitalization is comparable to the general population. Also, no significant effect modification in VE was observed by the patient’s sex, or age. These findings suggest that the protective effect of the influenza vaccine against laboratory-confirmed influenza is consistent across key demographic and clinical subgroups within this high-risk population. The third study found that the influenza vaccine provides a significant protective effect against CVD outcomes. The global findings of this thesis emphasize the validity of administrative databases at estimating population-level vaccination rates and show the importance of influenza vaccination as an effective strategy to reduce both influenza hospitalization and CVD events. This unique research equips healthcare providers and policy makers with relevant findings to support their campaigns and recommendations on influenza VE, particularly in relation to CVD outcomes.