Methods
We included 27,865 cases registered between January 2020 and February 2021 in the COVID-19 Registry of Japan (COVIREGI-JP)11. We used a Bayesian hierarchical model to construct a prefecture-specific random intercept to evaluate the effect of regional heterogeneity.
The capacity of medical facilities can be evaluated by the number of hospital beds; however, the differences in health policies against COVID-19 among local municipalities should also be taken into consideration as the planning of hospital bed distribution and the priority of hospitalization indication are influenced by the policy of each local municipality. Considering this, we constructed two models: one which included the number of beds secured for COVID-19 patients in each prefecture as a fixed effect (model B) and one which did not (model A).
The independent variable was the National Early Warning Score (NEWS)12 of COVID-19 patients on the day of admission, and the explanatory variables for the fixed effect were sex, age of 65 years and older, and any risk factor of severe infection (cardiovascular disease, cerebrovascular disease, chronic respiratory disease, liver disease, diabetes, hypertension, obesity, dialysis or sever renal failure, and malignancy) for model A and the number of beds secured for COVID-19 patients in each prefecture for model B.
The independent variable, the NEWS score at the time of admission, was assumed to follow a Poisson distribution. We set four separate sampling chains, each consisting of 12,000 random samples, including 1,000 burnin. The sampling convergence was evaluated using the Gelman–Rubin statistics (R-hat below 1.1) and by visually inspecting a trace plot. All analyses were conducted using the R software version 4.0.5.13