Interim findings from first-dose mass COVID-19 vaccination roll-out and COVID-19 hospital admissions in Scotland: a national prospective cohort study
Procedures
- Simpson CR
- Robertson C
- Vasileiou E
- et al.
the hospital admission data available from the Scottish Morbidity Record 01 database, and Rapid Preliminary Inpatient Data.
National Data Catalogue. Rapid preliminary inpatient data (RAPID).
Vaccination data were available from general practices and the Turas Vaccination Management Tool (TVMT),
which is a web-based tool to capture vaccinations in the community and create real-time vaccination records. Laboratory data from ECOSS included all rtPCR test results from both National Health Service laboratories (Pillar 1) and Lighthouse Government laboratories (Pillar 2).
COVID-19 testing data: methodology note.
Data were deterministically linked using the Community Health Index number, which is a unique identifier used for all health-care contact across Scotland.
- Simpson CR
- Robertson C
- Vasileiou E
- et al.
- Polack FP
- Thomas SJ
- Kitchin N
- et al.
and ChAdOx1 (AZD1222; also known as the Oxford–AstraZeneca) vaccine.
- Voysey M
- Clemens SAC
- Madhi SA
- et al.
An individual was defined as exposed if they received a single dose of vaccine between Dec 8, 2020, and Feb 22, 2021, with maximum follow-up time censored at Feb 22, 2021 (the latest event date). Vaccinated groups were stratified by time intervals including 0–6, 7–13, 14–20, 21–27, 28–34, 35–41, and 42 or more days post-vaccination, and by the type of vaccine received. Vaccination information was extracted from the general practitioner records and the TVMT system and included individuals vaccinated in general practices, community vaccination hubs, and other settings such as care homes.
Statistical analysis
The primary analyses included vaccine effect estimates for vaccination status overall and for each vaccine type. The secondary analysis included vaccine effect estimates for vaccine status overall and for each vaccine type stratified by age groups (ages 18–64, 65–79, and ≥80 years). These were grouped to the completed year (ie, 64·5 years would be categorised as 18–64 years and 79·4 as 65–79 years).
Baseline characteristics in the vaccinated and unvaccinated groups were described using proportions. We assessed the effect of one dose of either vaccine against hospital admissions due to laboratory-confirmed SARS-CoV-2 infection, or clinical diagnosis of COVID-19 on admission. Poisson regression adjusting for an offset representing the time at risk and time-dependent Cox models (considering the time at risk) were used to derive the rate ratios, hazard ratios, and 95% CIs for the association of vaccination with COVID-19 hospital admissions.
Cox models included spline terms for age and number of rtPCR tests before vaccination (a marker for health-care workers, social-care workers, and care home residents who had repeated tests). Additional adjustments were made for sex, socioeconomic status, and underlying medical conditions at risk of COVID-19 illness with vaccination groups representing a time-dependent covariate. Calendar time intervals by week were included as stratification variables because the background epidemic was changing rapidly over the observation period. Poisson regression was used for the full adjustment and propensity weighting. This regression model used age groups in 5-year intervals as well as sex, deprivation, and number of previous tests. Additionally, the following comorbidity groups, all of which are associated with an increased risk of hospital admission, were included: type 1 and type 2 diabetes, high and low blood pressure, chronic obstructive pulmonary disease, chronic kidney disease, dementia, stroke, learning disorders, fractures, neurological conditions, chronic cardiac failure, asthma, epilepsy, blood cancer, liver cirrhosis, venous thromboembolism, peripheral vascular disease, atrial fibrillation, pulmonary hypertension, Parkinson’s disease, rare pulmonary disorders, rheumatoid arthritis, and systemic lupus erythematosus.
Both the Cox models and Poisson regression used sampling weights, which were used to correct for the size of the registered general practice population being greater than the population in Scotland (some due to recently deceased patients still being recorded in the patient records and individuals who had recently moved). These weights were derived by matching the age and sex numbers in the general practice data to the Scottish population data. This adjustment ensured that the denominators in the tables matched the Scottish population.
33 834 individuals received a second dose of the vaccine. Individuals who received two doses remained in the analysis for as long as they had one dose and were then censored at the date of the second dose. All statistical tests were two-tailed with a 5% significance level of p less than 0·05.
- Simpson CR
- Robertson C
- Vasileiou E
- et al.
residential settlement measured by the urban or rural 6-fold classification (1 refers to large urban areas and 6 refers to small remote rural areas),
- Simpson CR
- Robertson C
- Vasileiou E
- et al.
and the number and types of comorbidities commonly associated with COVID-19 illness.
- Simpson CR
- Robertson C
- Vasileiou E
- et al.
To adjust for the residual confounding in which vaccines were not offered to or were declined by the most frail, we included a functional variable, namely dementia, in our covariate adjustment.
- Jackson LA
- Jackson ML
- Nelson JC
- et al.
- Polack FP
- Thomas SJ
- Kitchin N
- et al.
In particular, we carried out a falsification of exposure sensitivity analysis that involved re-running the analysis using a fictional date of vaccination 2 months before actual vaccination to determine the extent to which vaccine programme effects contributed to residual confounding. We have provided the overall predicted curves from the Cox proportional hazards model adjusting for covariates, with a stratification by calendar period (appendix pp 6–7). An over-dispersed Poisson regression model was also carried out to assess the effect of the national lockdown on Dec 26, 2020, and the additional restrictions imposed on Jan 5, 2021, on hospital admissions due to COVID-19 in adults (age 18–64, 65–79, and 80 or older).
There was missing data for deprivation, urban-rural status (patient postcode was not available), smoking status, and blood pressure (value had not been recorded). Missing data were handled by creating a separate group for these individuals.