ChatGPT: unlocking the potential of Artifical Intelligence in COVID-19 monitoring and prediction.

BACKGROUND: The COVID-19 pandemic has had an unprecedent impact of everyday life with deleterious consequences on global health, economics, and society. Thus, accurate and timely information is critical for monitoring its spread and mitigating its impact. ChatGPT is a large language model chatbot with artificial intelligence, developed by OpenAI, that can provide both textual content and R code for predictive models. It may prove to be useful in analyzing and interpreting COVID-19-related data. METHODS: This paper explores the application of ChatGPT to the monitoring of the COVID-19 pandemic, presenting R code for predictive models and demonstrating the model's capabilities in sentiment analysis, information extraction, and predictive modelling. We used the prediction models suggested by ChatGPT to predict the daily number of COVID-19 deaths in Italy. The prediction accuracy of the models was compared using the following metrics: mean squared error (MSE), mean absolute deviation (MAD) and root mean squared error (RMSE). RESULTS: ChatGPT suggested three different predictive models, including ARIMA, Random Forest and Prophet. The ARIMA model outperformed the other two models in predicting the daily number of COVID-19 deaths in Italy, with lower MSE, MAD, and RMSE values as compared to the Random Forest and Prophet. CONCLUSIONS: This paper demonstrates the potential of ChatGPT as a valuable tool in the monitoring of the pandemic. By processing large amounts of data and providing relevant information, ChatGPT has the potential to provide accurate and timely insights, and support decision-making processes to mitigate the spread and impact of pandemics. The paper highlights the importance of exploring the capabilities of artificial intelligence in the management of public emergencies and provides a starting point for future research in this area.

COVID-19 mortality rates in the European Union, Switzerland, and the UK: effect of timeliness, lockdown rigidity, and population density.

BACKGROUND: To date, the European experience with COVID-19 mortality has been different to that observed in China and Asia. We aimed to forecast mortality trends in the 27 countries of the European Union (EU), plus Switzerland and the UK, where lockdown dates and confinement interventions have been heterogeneous, and to explore its determinants. METHODS: We have adapted our predictive model of COVID-19-related mortality, which rested on the observed mortality within the first weeks of the outbreak and the date of the respective lockdown in each country. It was applied in a training set of three countries (Italy, Germany and Spain), and then applied to the EU plus the UK and Switzerland. In addition, we explored the effects of timeliness and rigidity of the lockdown (on a five-step scale) and population density in our forecasts. We report r2, and percent variation of expected versus observed deaths, all following TRIPOD guidance. RESULTS: We identified a homogeneous distribution of deaths, and found a median of 24 days after lockdown adoption to reach the maximum daily deaths. Strikingly, cumulative deaths up to April 25th, 2020 observed in Europe separated countries in three waves, according to the time lockdown measures were adopted following the onset of the outbreak: after a week, within a week, or even prior to the outbreak (r2=0.876). In contrast, no correlation neither with lockdown rigidity nor population density were observed. CONCLUSIONS: The European experience confirms that early, effective interventions of lockdown are fundamental to minimizing the COVID-19 death toll.