Abstract

The delay in reporting deaths or births can be an obstacle to publishing the most up-to-date statistical data possible. It means that the cases are not available in their entirety for a certain period of time and thus do not provide any information on the current situation regarding deaths and births. A certain amount of time passes until a case is reported to the registry office and this report is processed there in such a way that it is registered in the official statistics. In Berlin in particular, several weeks sometimes elapse before all cases are fully recorded. Thus, there is a great deal of uncertainty in Berlin regarding the current case numbers. A model was sought that is able to extract knowledge from available but incomplete data and use this for an estimate. From this knowledge, a forecast can be made for the very recent past. Such forecasts, which refer to current rather than future points in time, are called nowcasts. They can give a better indication of the current situation than raw, unfinished case counts.

For the approach presented here to estimate death case numbers at the recent edge in Berlin, elements of machine learning (ML) are used. The presented method learns from reliable registration offices of the near past, which allow a good forecast on the basis of cases reported up to a certain point in time on the final total number. The selection criterion for determining these reliable reporting offices is based on a strong correlation between cases available at a point in time and the past final number of cases. For the nowcast, therefore, an estimate can be made using a regression approach.

Although estimated figures can by no means replace correct and verified results from official statistics, they can at least provide support during the period when the truth is not yet known. Similar to the weather forecast, an estimate gives us a sense of security, predictability and control.

The reason for the investigation is the observation of the current events of the persons who died from COVID-19. However, it is not only the numbers of those individuals who have died directly from COVID-19 that are of interest. It is the death rates overall that reveal much: including how changes and actions related to the Corona pandemic are affecting our health and, consequently, our mortality. Since there is a great deal of uncertainty in the current case numbers, not only for deaths in Berlin, the procedure was also extended to include births.