Local mortality estimates during the COVID-19 pandemic in Italy

Area of Science:

• Epidemiology
• Data Science
• Public Health

Background:

• Accurate COVID-19 death toll estimation is challenging, especially at local levels.
• Official methods for excess mortality rely on historical averages, which may not be robust.
• Granularity and data accuracy are critical for reliable local mortality assessments.

Purpose of the Study:

• To develop a more realistic counterfactual mortality scenario in the absence of COVID-19.
• To improve the accuracy of local excess mortality estimates.
• To provide insights into the demographic impact of the first COVID-19 wave in Italy.

Main Methods:

• Application of a machine learning control method for mortality prediction.
• Utilizing supervised machine learning techniques for enhanced prediction accuracy.
• Derivation of local excess mortality estimates using best-performing algorithms.

Main Results:

• Machine learning methods significantly outperform the official approach in predicting local mortality.
• Improved prediction accuracy was particularly noted in small and medium-sized municipalities.
• Dataset of local excess mortality estimates for February-September 2020 is provided.

Conclusions:

• Machine learning offers a superior approach to estimating local excess mortality during pandemics.
• The study provides valuable data for understanding the demographic impact of the COVID-19 pandemic's first wave.
• Freely available dataset supports improved diagnostic and monitoring efforts in public health.