Autore: Michela Bersia, Department of Sciences of Public Health and Pediatrics, University of Torino
michela.bersia@unito.it
Sara De Luca, LINKS Foundation; Piercesare Grimaldi, Department of Sciences of Public Health and Pediatrics, University of Torino; Alessia Visconti, Centre for Biostatistics, Epidemiology and Public Health, Department of Clinical and Biological Sciences, University of Torino; Giuseppe Rizzo, LINKS Foundation; Paola Berchialla, Centre for Biostatistics, Epidemiology and Public Health, Department of Clinical and Biological Sciences, University of Torino;
Categoria Primaria: Promozione della salute e prevenzione
Categoria Secondaria: Sistemi informativi
Introduction The COVID-19 pandemic experience showed that a prompt identification of epidemic outbreaks is essential to implement public health measures mitigating infectious disease transmission.
Objective One aim of the TrustAlert project is to assess the role of real-time online media news in the early detection and monitoring of potential epidemic outbreaks to support public health policies in times of epidemiological crises.
Methods We used a transformer-based Artificial Intelligence model, MPNet (Masked and Permuted Pre-training for Language Understanding), to extract information from online news articles collected by the GDELT Project. Then, we assigned International Classification of Diseases-9 (ICD-9) codes to the extracted information using a zero-shot classification. Next, we built online news time series for a set of seven infectious diseases (namely COVID-19, childhood pneumonia, Influenza, Cholera, Dengue, West Nile, and Monkeypox) which we finally compared with the actual number of confirmed cases, as provided by the European Centre for Disease Prevention and Control. The only exception was the outbreak of childhood pneumonia in China in November 2023, for which the online news information was computed with the timing of the World Health Organization (WHO) intervention. These comparisons allowed us to measure the lead time of online news in anticipating epidemic peaks (for COVID-19) or public health interventions (for childhood pneumonia).
Results Here, we present preliminary results regarding the COVID-19 pandemic (from January 1st, 2020, to October 23rd, 2022) and the outbreak of pneumonia among Chinese children in November 2023. In the first scenario, we observed that COVID-19-related news could anticipate the first peak of the confirmed cases by 13 days. In the second scenario, we observed an anticipation of 7 days between the online news information peak (on November 16th, 2023) and WHO consultation with Chinese authorities (on November 23rd, 2023).
Conclusions Our findings showed important public health implications by highlighting the potential of online media information as a strategic data source to guide real-time epidemic surveillance. The main strengths of this approach is its favourable cost-benefit ratio, albeit users should take into account the necessity of mitigating possible false positive results, which we did not observe in our experimentation. Our approach, which combines natural language processing techniques with an epidemiological perspective, is a promising and flexible research, deserving further exploration also in regard to other emerging health issues, such as time trends in mental issues.