Background: Observational studies are important for assessing vaccine effectiveness in the real-world, for example with new strains of pathogens or people changing their behaviour in response to vaccination. Influenza vaccine effectiveness estimates have previously been overestimated which may at least be partly due to unmeasured confounding from health-seeking behaviour and healthcare access. The test-negative-case-control design was developed to account for confounding from health-seeking behaviour and healthcare access in vaccine effectiveness research. This design requires test result data to be available and has a strong set of assumptions. It remains unclear whether confounding from health-seeking behaviour and healthcare access can be accounted for using alternative methods. Aims: Overall, the aim of this thesis was to advance methods to account for biases in observational research. The first objective was to identify and quantify biases and alternative causal pathways in a COVID-19 vaccine effectiveness test-negative-case-control study which formed the basis of UK national monitoring by the UK Health Security Agency. As alternative methods are required to account for confounding from health-seeking behaviour in other study designs, the second objective was to systematically identify a set of markers of health-seeking behaviour and healthcare access in electronic health records (EHRs) that could potentially be used to quantify and account for this type of confounding. The third objective was to quantify and account for confounding from health-seeking behaviour and healthcare access in an influenza and COVID-19 vaccine effectiveness study with a cohort design using the markers from study two. Methods: For the first objective, a questionnaire was sent to a sample of participants in one of the first UK COVID-19 test-negative-case-control vaccine effectiveness studies, which had used routinely-recorded data. Self reported information on vaccination dates, symptomatic status, comorbidities and risk behaviours was used to explore potential biases and alternative causal pathways in the original study. For the second objective, markers of health-seeking behaviour and healthcare access were identified that were appropriate to a population aged ≥65 years. These were selected based on a health behavioral model known as the Theory of Planned Behaviour. These markers were then identified in the Clinical Practice Research Datalink (CPRD), a longitudinal dataset from primary care practices, with linkages to hospital and mortality data. The prevalence of these markers in a population ≥66 years in England identified in the CPRD linked datasets were compared to national estimates. For the third objective, to quantify and account for confounding from health-seeking behaviour and healthcare access, a cohort study of COVID-19 and influenza vaccine effectiveness among older adults in England was conducted. Cox regression models were used to estimate vaccine effectiveness. The models were conducted in four sequential modelling steps – model one: adjusting for demographics, model two: additionally adjusting for ethnicity and deprivation, model three: additionally adjusting for comorbidities and model four: additionally adjusting for the health-seeking and healthcare access markers from study two. A negative control exposure cohort (history of influenza vaccination against early COVID-19 pandemic SARS-CoV2) was used to investigate the extent of residual confounding after adjustment for the markers. Results: For the first objective, there was minimal evidence of bias, and accounting for multiple potential biases only changed the estimated vaccine effectiveness after two doses of BNT162b2 decreased from 88% (95% confidence interval [CI]: 79-94%) in the original study to 85% (95% CI: 68-94%). For the second objective, fourteen markers of health-seeking behaviour and healthcare access were systematically identified. These included preventative measures where the influence of underlying health need was minimal (e.g., bowel cancer screening). They had prevalence estimates that were comparable to national estimates e.g., 73.3% for influenza vaccination in the 2018/2019 season, compared to 72.4% in national estimates. For the third objective, adjusting for these markers in the influenza vaccine effectiveness study increased vaccine effectiveness estimates against influenza infections from -1.5% (95% CI: -3.2, 0.1%) in model three (adjusting for demographics, ethnicity, deprivation and comorbidities) to 7.1% (95% CI: 5.4, 8.7%) in model four (additionally adjusting for health-seeking and healthcare access markers). Similar trends were found for more severe endpoints. For COVID-19, vaccine effectiveness estimates minimally increased from 82.7% (95% CI: 78.3, 86.2%) in model one (adjusting for demographics) to 83.1% (95% CI: 78.7, 86.5%) in model four. Adjusting for these markers in the negative control exposure analysis, increased vaccine effectiveness estimates from nearer the null (model three: -7.5% [95% CI: -10.6 - -4.5%] to model four: -2.1% [95% CI: -6.0 - 1.7%]). Conclusion: This thesis identified that when using the UK EHRs and the test-negative design the impact of potential biases on early pandemic COVID-19 observational vaccine effectiveness estimates was minimal. In instances where the test-negative-case-control design cannot be conducted, markers of health-seeking behaviour and healthcare access can be identified in EHRs. These markers can be used in other observational studies where health-seeking behaviour or healthcare access is relevant using study designs that are more broadly applicable (e.g.,cohort). The effects of confounding from health-seeking behavior is context dependent with minimal impact during early COVID-19 pandemic implementation, but more pronounced for seasonal influenza estimates.