Statistical analysis

For each socio-demographic characteristic, the median percent and interdecile range of neighbourhood composition and the crude COVID-19 incidence and mortality rates per 100,000 population in the lowest and highest deciles were estimated. Further, associations between neighbourhood-level characteristics and neighbourhood-level counts of COVID-19 cases and deaths were estimated by fitting a series of Poisson generalized linear mixed models with random effects for neighbourhood, offset for neighbourhood population. Models were assessed for zero-inflation by comparing the observed number of zeroes with model predicted number of zeroes for all models. No models were found to be underfitting zeroes. Any overdispersion present in outcomes is accounted for by the use of random effects in all models [21]. Separate crude bivariable models were used to estimate associations between each neighbourhood measure of immigration, race, housing, and socio-economic status and COVID-19 incidence and mortality. Subsequently, each of these models were adjusted for individual-level age-group (<15, 15–64, and 65+ years) and sex (male/female), and neighbourhood-level urban/rural geography. To account for uneven distribution of socio-economic characteristics across neighbourhoods, all model estimates were standardized to show relative risks and 95% confidence intervals of COVID-19 incidence and mortality rates between the 10^th (p10) and 90^th (p90) percentile of each neighbourhood socio-demographic characteristic. All 95% confidence intervals were calculated using robust standard errors. The distribution of socio-demographic characteristics were plotted against COVID-19 incidence for each neighbourhood, along with solid lines representing the model-predicted estimates (derived using ‘prediction’ package in R) and dashed lines marking p10 and p90 for each predictor’s distribution.

All analyses were conducted in R.