Efficacy of COVID-19 vaccination in individuals designated as clinically extremely vulnerable in Scotland

Introduction

A primary-care based observational study that included cases diagnosed up to 15 Feb 2021 in Scotland estimated maximal efficacy of a single vaccine dose to be 70% for ChAdOx1 nCoV-19 and 85% for BNT162b2 mRNA at 28 days post-vaccination.¹ However policy on managing the epidemic depends critically upon the extent to which vaccination protects clinically vulnerable individuals, who account for most hospitalizations and deaths,² against severe COVID-19. Whet levels of vaccine efficacy pertain among those designated as clinically extremely vulnerable to COVID-19 is unclear. In Scotland early in the epidemic Public Health Scotland established a register of all those with certain conditions or on certain drugs that were likely to lead to extreme vulnerability to COVID-19 was established based on electronic health care records and primary care physician reporting.^3,4 This register was used to determine eligibility for the national shielding support programme. We previously reported on the risk of severe COVID-19 in this shielding eligibility cohort compared to the rest of the population.⁵ If vaccine efficacy were estimated to be lower in clinically extremely vulnerable individuals than in the rest of the population, additional measures to shield these individuals would be needed until overall transmission rates in the population are low. Lower vaccine efficacy is possible since many such persons are immunocompromised or on immunosuppressants. Therefore, the aim of this study was to examine the efficacy of COVID-19 vaccination in reducing the risk of severe COVID-19 among those designated as clinically extremely vulnerable in Scotland.

Methods

This analysis is based on the REACT-SCOT matched case-control study, established at the beginning of the epidemic to investigate risk factors for severe COVID-19 in the population of Scotland.² We used this study to take advantage of data linkages already established. The design has been described in detail previously.² In brief, for every incident case of COVID-19 in the population ten controls matched for one-year age, sex and primary care practice and alive on the day of presentation of the case that they were matched to were selected using the Community Health Index database. COVID-19 cases are those with a positive nucleic acid test, or a hospital admission or death with COVID-19 ICD-10 codes. The REACT-SCOT case-control dataset is refreshed regularly and is linked to the vaccination database and to the regularly updated dataset of all individuals deemed eligible for the shielding programme.

As previously,² to minimise ascertainment bias we pre-specified the primary outcome measure as severe COVID-19, defined as cases with entry to critical care within 28 days of presentation or fatal outcome (any death within 28 days of a positive test or any death for which COVID-19 was coded as underlying cause). As previously, cases and controls were classified as eligible or ineligible for shielding.⁵ For further analyses, the shielding category was subdivided as described previously into six categories: solid organ transplant, specific cancers, severe respiratory conditions, other rare conditions, on immunosuppressants, and additional conditions.⁵ Those ineligible for shielding were classified according to the presence of conditions designated by public health agencies as “moderate risk conditions” as described previously.² This analysis is based on cases presenting from 1 December 2020 to 16 March 2021 (the latest linkage dataset available at the time of preparing this report).

Vaccination status in cases was coded as having had at least one dose of any vaccine at least 14 days before presentation date. This assigns cases (and their matched controls) who were vaccinated less than 14 days prior to the case date to the unexposed category. The numbers of cases and controls were too few to allow further subdivision of time since vaccination.

The effect of vaccination in each of the clinical vulnerability categories was estimated in a conditional logistic regression model (R function survival::clogit, package version_3.2-7, R version 3.6.3) fitted to the full dataset, specifying effects ${\beta }_{R2},\dots ,{\beta }_{RJ}$ for the log rate ratio associated with risk categories 2 to $J$ (${\beta }_{R1}=0$ for the reference category $J=1$), and nested effects ${\beta }_{V1},\dots ,{\beta }_{VJ}$ for the log rate ratio associated with vaccination in each of the $J$ risk categories. With this incidence density sampling design, the conditional odds ratio is the rate ratio. The efficacy of vaccination is 1 minus the rate ratio. The strata in the conditional logistic regression model are matched sets: typically one case and up to 10 controls, but where two cases with the same age, sex and GP practice present on the same day the matched set comprises two cases and up to 20 controls. When estimating the rate ratio within a risk group such as a shielding category, only cases and controls in that risk group are retained in the matched sets. Matched sets in which all are vaccinated or none are vaccinated do not contribute to the rate ratio. We emphasize that the unconditional odds ratios calculated from summary tables of the vaccination status of cases and controls in each risk group cannot be used to estimate rate ratios.^6,7 The design controls not only for the matching factors of age, sex and general practice but also for calendar time. Because severe COVID-19 is strongly associated with care home residence and care home residents were a priority group for vaccination, the model includes care home residence as a covariate.

Severe COVID-19 is strongly associated with recent hospital admission. This might confound the association of severe COVID-19 with vaccination if hospital admission causes vaccination appointments to be missed or alternatively if hospital admission increases the probability of being vaccinated by a given date. Therefore the analysis was repeated with adjustment for any hospital discharge from 2 to 8 weeks before presentation date, and for being an inpatient for at least 8 days before presentation date (the criterion for probable/definite hospital-acquired infection⁸). We also reported the vaccine efficacy for severe COVID-19 separating out those with conditions designated previously by Public Health Scotland as moderate risk from the remainder not eligible for shielding.

To allow inference on vaccine efficacy in smaller groups, the vaccine efficacy analysis was repeated with the case definition extended to include all cases that were hospitalized or fatal.

Results

Over the study period there were 863 severe cases among those eligible for shielding of which 731 were fatal and 3482 severe cases among those not eligible for shielding of which 2641 were fatal. 97% of severe cases had a positive nucleic acid test for SARS-CoV-2 and this proportion did not differ between those eligible and ineligible for shielding. Figure 1 shows the time course of daily severe cases in Scotland and vaccination among those eligible for shielding. The proportion vaccinated at least 14 days before is estimated from those sampled as controls who were eligible for shielding. From January to February 2020 the daily number of severe cases declined rapidly and the proportion vaccinated among those eligible for shielding rose from 2% to 87%. The association between vaccination and severe disease is thus confounded by calendar time. This confounding is controlled by the matching of cases and controls on calendar date but we reiterate that because of this the unconditional odds ratios calculated from Tables 1 and 2 cannot be used to estimate rate ratios.⁷ Information about vaccine efficacy in those eligible for shielding is contributed by matched sets of cases and controls in the narrow time window during which incidence of severe COVID-19 was high and at least some of those eligible for shielding had been vaccinated at least 14 days before. Up to 16 March 2021 the proportion who had received a second dose at least 14 days before the date of sampling was very low: 1069 (0.06%) of the 1831776 controls ineligible for shielding and 33 (0.05%) of the 65573 controls eligible for shielding.

Figure 1.

(a) Daily severe cases by risk category; (b) Controls eligible for shielding: proportion vaccinated at least 14 days before, by 7-day window of date on which they were sampled as controls.

Table 1 shows the vaccination status of severe cases and their controls by risk stratum. Table 2 shows the same breakdown for hospitalised cases and controls. These tables illustrate that the data are very sparse for certain shielding groups especially solid organ transplant recipients.

The rate ratio for severe COVID-19 associated with vaccination at least 14 days before was 0.29 (95% CI 0.22 to 0.37) in those eligible for shielding, compared with 0.29 (95% CI 0.25 to 0.34) in those ineligible for shielding. When inpatient stay for at least 8 days before presentation and any hospital discharge from 15 to 56 days before presentation were included as covariates, the corresponding rate ratios were 0.40 (95% CI 0.30 to 0.53) in those eligible and 0.39 (95% CI 0.33 to 0.46) in those ineligible for shielding. Table 3 gives a breakdown of the rate ratios for severe COVID-19 associated with vaccination by shielding condition. Point estimates for the rate ratio associated with vaccination in risk groups were similar to the estimated rate ratio in those not eligible for shielding, with the exception of solid organ transplant recipients and others on immunosuppressants in whom the numbers of cases were small (Table 1) and confidence intervals were wide.

When those ineligible for shielding were separated into those with moderate risk conditions and those with no risk condition the rate ratios for severe COVID-19 associated with vaccination were 0.26 (95% CI 0.22 to 0.30) and 0.37 (95% CI 0.29 to 0.47) respectively. Among people with diabetes who were ineligible for shielding (and thus in the “moderate risk conditions” category), the rate ratios for severe COVID-19 associated with vaccination were 0.09 (95% CI 0.01 to 0.70) and 0.16 (95% CI 0.11 to 0.22) in those with Type 1 and Type 2 diabetes respectively.

For a secondary analysis with larger numbers of cases, the case definition was broadened to include all hospitalized or fatal cases. The rate ratio for hospitalised or fatal COVID-19 was 0.39 (95% CI 0.33 to 0.46) in those eligible and 0.37 (95% CI 0.33 to 0.41) in those not eligible for shielding. Table 4 shows that there was evidence for efficacy in all groups except solid organ transplant recipients in whom even with this broader case definition the numbers of cases and controls were too small for efficacy to be estimated reliably.

Conclusions

Declarations