## Finland

### NOTE

Here the dates are those when the covid-test was taken but many sources report the date when the test result was confirmed positive. This makes the numbers differ slightly.

### Deaths ## Finnish University Hospital Catchment Areas

Finland has five university hospitals which provide advanced treatments needed for serious COVID-19 cases. In the following, we provide data on the regions for which the university hospitals are responsible.

## HYKS

### NOTE

Here the dates are those when the covid-test was taken but many sources report the date when the test result was confirmed positive. This makes the numbers differ slightly.

## TAYS

### NOTE

Here the dates are those when the covid-test was taken but many sources report the date when the test result was confirmed positive. This makes the numbers differ slightly.

## OYS

### NOTE

Here the dates are those when the covid-test was taken but many sources report the date when the test result was confirmed positive. This makes the numbers differ slightly.

## TYKS

### NOTE

Here the dates are those when the covid-test was taken but many sources report the date when the test result was confirmed positive. This makes the numbers differ slightly.

## KYS

### NOTE

Here the dates are those when the covid-test was taken but many sources report the date when the test result was confirmed positive. This makes the numbers differ slightly.

### Deaths ## Data sources

The number of confirmed cases is fetched daily from THL's public API. The ward occupancies are obtained from HS's data source.

Note that the daily confirmed cases from THL are preprocessed and adjusted with the estimated delay of the analysis times before feeding into the model to obtain improved estimates for the past few days.

## Estimation of Rt

The time-dependent reproduction number Rt is estimated from the daily number of confirmed cases using Bayesian smoothing. We set up a simple SEIR model,

\begin{align} S(t+1) &= S(t) - \frac{R_t(t)}{T_i} I(t) \frac{S(t)}{N} + \omega_S \\ E(t+1) &= E(t) + \frac{R_t(t)}{T_i} I(t) \frac{S(t)}{N} - \frac{1}{T_e} E(t) + \omega_I \\ I(t+1) &= I(t) + \frac{1}{T_e} E(t) - \frac{1}{T_i} I(t) + \omega_I \\ R_t(t+1) &= R_t(t) + \omega_{R_t}, \end{align}
where $$R_t$$ is now also a state variable, and $$\omega$$s are small Gaussian noises except for $$\omega_{R_t}$$ for which we have the variance 0.0001.

The measurement model is

\begin{align} \text{Daily New Cases} &\sim \frac{R_t(t)}{T_i} I(t) \frac{S(t)}{N} + \omega_z. \end{align}

The system state is estimated using an unscented Rauch-Tung-Striebel smoother.