Small Area Vulnerability Index (SAVI)

Summary

The SAVI is an empirically informed measure of COVID-19 vulnerability for each Middle Super Output Area (MSOA) in England. The SAVI index investigates the association between each predictor (proportion of the population from Black, Asian and Minority Ethnic (BAME) backgrounds, living in care homes, living in overcrowded housing and having been admitted in the past 5 years for a long-term health condition) and COVID-19 mortality using a multivariable Poisson regression, whilst adjusting for the age profile of each area and accounting for the regional spread and duration of the epidemic.

Technical description

We used multivariable Poisson regression to investigate the independent association between each predictor and COVID-19 mortality. All predictors including income deprivation (IMD) and aged over 80 years group were initially included in the model and then tested for all possible two-way interactions. A parsimonious criterion-based model building approach was used for including interactions in the final model informed by the Akaike Information Criterion (AIC).

Factors related to the distribution of the initial wave (time to first 10 cases and region dummies) were then set to zero. This removes the effect of the duration of the epidemic in each area and regional factors that are likely due to idiosyncratic characteristics of the transmission dynamics in the initial wave. The model was then used to predict the number of deaths in each MSOA based on the selected 4 vulnerability measures for each MSOA. This was divided by the number of deaths that would have been expected of each area had the average predicted mortality to give the vulnerability index. This essentially provides a measure for each MSOA that indicates the relative increase in COVID-19 mortality risk that results from the level of each of the 4 vulnerability measures for each area. To further improve the interpretability of the index, we also provide a ‘shrunken’ version of the index for suppressing the extreme high values using the methodology adopted in the IMD 2019 index.