Background: the first confirmed cases of COVID-19 in WHO European Region was reported at the end of January 2020 and, from that moment, the epidemic has been speeding up and rapidly spreading across Europe. The health, social, and economic consequences of the pandemic are difficult to evaluate, since there are many scientific uncertainties and unknowns.
Objectives: the main focus of this paper is on statistical methods for profiling municipalities by excess mortality, directly or indirectly caused by COVID-19.
Methods: the use of excess mortality for all causes has been advocated as a measure of impact less vulnerable to biases. In this paper, observed mortality for all causes at municipality level in Italy in the period January-April 2020 was compared to the mortality observed in the corresponding period in the previous 5 years (2015-2019). Mortality data were made available by the Ministry of Internal Affairs Italian National Resident Population Demographic Archive and the Italian National Institute of Statistics (Istat). For each municipality, the posterior predictive distribution under a hierarchical null model was obtained. From the posterior predictive distribution, we obtained excess death counts, attributable community rates and q-values. Full Bayesian models implemented via MCMC simulations were used.
Results: absolute number of excess deaths highlights the burden paid by major cities to the pandemic. The Attributable Community Rate provides a detailed picture of the spread of the pandemic among the municipalities of Lombardy, Piedmont, and Emilia-Romagna Regions. Using Q-values, it is clearly recognizable evidence of an excess of mortality from late February to April 2020 in a very geographically scattered number of municipalities. A trade-off between false discoveries and false non-discoveries shows the different values of public health actions.
Conclusions: despite the variety of approaches to calculate excess mortality, this study provides an original methodological approach to profile municipalities with excess deaths accounting for spatial and temporal uncertainty.
Keywords: Bayesian models; COVID-19; attributable risk; excess mortality; Q-values; hierarchical models.