While observational studies of VE provide a practical alternative to randomized, controlled efficacy studies, the validity of VE estimates derived from observational studies is threatened by bias introduced because exposure to vaccine is not randomly assigned. Because the decision to vaccinate is driven by patient characteristics such as age and presence or severity of medical comorbidities and influenced by patient healthcare-seeking behaviour and provider practice, all factors which can impact clinical outcomes, the results of observational studies of VE may be influenced by uncontrolled confounding. This so-called “confounding by indication” may result in underestimation of VE if the vaccinated group is less healthy and at higher risk of adverse health outcomes than the unvaccinated group or overestimation of VE if the vaccinated group is generally healthier, the so-called “healthy-user bias.” Observational studies of VE must quantify the magnitude of the impact of confounding by indication and incorporate methods to allow appropriate adjustment of VE estimates to account for this uncontrolled bias.
Frailty is the accumulation of multiple interacting health, symptom, and functional deficits which, when viewed as a whole, creates a summary measure of vulnerability. Frailty can be measured as an index of deficits based on a Comprehensive Geriatric Assessment (CGA. Each deficit (e.g., comorbid illness, sensory deficit, functional deficit) is assigned a value of 1 if representing a deficit and 0 otherwise. The FI is the sum of deficits divided by the total number of possible deficits, in which high scores indicate greater vulnerability. The well-validated FI is predictive of health outcomes including mortality, institutionalization, and development of dementia and has consistent mathematical and statistical features. We postulate that assessment of frailty using the FI may provide a powerful tool by which to characterize the directionality and magnitude of the impact of confounding by indication on VE estimates generated by our ongoing test-negative case-control study and allow us to appropriately adjust the VE estimate to account for this bias.
In year 2, we piloted the feasibility and utility of the FI in the assessment of VE in 300 enrolled participants ≥65 years of age. The collection of required data elements for calculation of a FI was feasible at baseline, at enrollment, and at 30 days postdischarge in cases and controls. Preliminary analysis demonstrated the validity of the FI in this data set. Both influenza cases and controls were similarly frail at baseline (mean FI 0.25 and 0.26, respectively). Influenza vaccination was associated with fewer poor outcomes (death or change to palliative care) in both cases and controls (4.5% in vaccinated versus 8.9% in unvaccinated). The vaccinated participants were more frail at baseline (FI 0.27 vs. 0.21) and were older. Vaccinated and nonvaccinated participants accumulated deficits at a similar rate (mean FI increase 0.11 vs. 0.13); vaccinated participants remained more frail than non-vaccinated participants at 30 days (mean FI 0.38 versus 0.34). Thus, in this population, confounding by vaccine indication results in underestimation of VE and use of the FI to adjust for this bias results in an increase in the VE estimate.
We propose to assess frailty using a FI calculated based upon deficits collected from a CGA in all of the predicted 800 influenza cases and 1600 test-negative controls enrolled across SOS (as described above). Frailty will be compared between vaccinated and unvaccinated participants at baseline in order to assess the directionality and magnitude of the effect of differences in frailty between vaccinated and unvaccinated participants on the VE estimate. The crude VE estimate will be compared to the frailty-adjusted VE estimate to assess the magnitude of the impact of confounding by indication and VE adjusted for frailty will be compared to VE adjusted by more traditional indicators (e.g., the Barthel Index, mean comorbidity, age, gender). The Heirarchical Assessment of Mobility and Balance (HABAM) will be used to further measure the impact of mobility on VE and outcomes.