Quantifying and adjusting for confounding from health-seeking behaviour and healthcare access in observational research
Listed in
This article is not in any list yet, why not save it to one of your lists.Abstract
Objective
To assess the feasibility and effect of using proxy markers of health-seeking behaviour and healthcare access to quantify and adjust for confounding in observational studies of influenza and COVID-19 vaccine effectiveness (VE).
Design
Cohort study for influenza VE in the 2019/2020 influenza season and for early COVID-19 VE (December 2020 – March 2021).
Setting
Primary care data pre-linked to secondary care and death data in England.
Participants
Individuals aged ≥66 years on 1 September 2019.
Interventions
Vaccination with any influenza vaccination in the 2019/2020 season or with either a BNT162b2 or ChAdOx1-S vaccination from 08/12/2020 to 31/03/2021.
Main outcome measures
Influenza or COVID-19 specific infections, hospitalisation and death. VE was estimated with sequential adjustment for demographics, underlying health conditions, and 14 markers reflecting uptake of public health interventions (screenings, vaccinations and NHS health checks), active healthcare access/use (prostate antigen testing, bone density scans, GP practice visits, low value procedures and blood pressure measurements) and lack of access/underuse (hospital visits for ambulatory care sensitive conditions and did not attend primary care visits). Influenza vaccination in the 2019/2020 season was also considered as a negative exposure intervention against the first wave of COVID-19.
Results
We included 1,991,284, 1,796,667, and 1,946,943 individuals in the influenza, COVID-19 and negative exposure VE populations, respectively. Vaccinated individuals were more likely to display active health-seeking behaviour, including participation in UK national screening programmes, compared with unvaccinated individuals. In the 2019/2020 influenza season, adjusting for health-seeking markers increased VE against infection from −1.5% (95%CI: −3.2,0.1) to 7.1% (95%CI: 5.4,8.7), but this trend was less apparent for more severe outcomes. For COVID-19 during early vaccine roll out, adjusting for health-seeking markers in addition to demographics and underlying health conditions did not change VE estimates against infection or severe disease (e.g., two doses of BNT162b2 against infection: from 82.8% [95%CI: 78.4,86.3] to 83.1% [95%CI: 78.7,86.5]). Adjusting for health-seeking markers removed bias in the negative exposure analysis of influenza vaccination against SARS-CoV-2 infection (−7.5% [95%CI: −10.6,-4.5] vs −2.1% [95%CI: −6.0,1.7] before vs after adjusting for health-seeking markers).
Conclusions
Markers of health-seeking behaviour and healthcare access can be identified in electronic health records, are associated with vaccination uptake, and can be used to quantify and account for confounding in observational studies.
What is already known on this topic
Health-seeking behaviour and healthcare access are recognised confounders in observational studies, but are not directly measurable in electronic health records (EHRs). Previously we systematically identified 14 markers in UK EHRs that reflect different aspects of health-seeking behaviour and healthcare access. We do not know if these markers can be utilised to quantify and account for this type of confounding in observational research using influenza and COVID-19 vaccine effectiveness as examples.
What this study adds
This study demonstrated using the proxy markers that confounding from health-seeking behaviour and healthcare access underestimates influenza VE estimates, but has negligible impact on COVID-19 VE estimates during early vaccine roll out. We also demonstrated via a negative exposure analysis that residual confounding can be removed by adjusting for these proxy markers.
