[edit]
Characterizing and Understanding Temporal Effects in COVID-19 Data
Proceedings of the 1st Workshop on Healthcare AI and COVID-19, ICML 2022, PMLR 184:33-40, 2022.
Abstract
Since the global outbreak of the coronavirus 2019 pandemic, hundreds of works have been published, analyzing and modeling multiple aspects of the disease. Several of them venture into predictive and modeling tasks, such as mortality prediction and patient severity scoring, using machine-learning (ML) algorithms. An important limitation for most of these works is the fact that they do not consider the multiple temporal aspects of this pandemic, especially regarding disease profile and distributional changes over the months. Such temporal effects are mostly due to multiple interactions between different and novel viral strains, combined with mass vaccination campaigns targeting different groups or patterns (e.g., prioritizing older individuals and those with comorbidity first) and availability of different vaccines. These temporal effects result in impaired model effectiveness and classification errors. In this paper, using a large dataset with over 10,000 patients from 39 hospitals in Brazil admitted during a period of more than 20 months, we provide an overview of the multiple forms of temporal drift that happened during the pandemic and the magnitude of their effects on model effectiveness. Our analyses encompass changes in the severely ill patients’ profile as well as how mortality rates have changed over time. We also investigate how the importance of different predictive variables change and shift over time.