Ups and downs of COVID-19: can we predict the future? Local analysis with Google Trends for forecasting the burden of COVID-19 in Pakistan

Background We aim to study the utility of Google Trends search history data for demonstrating if a correlation may exist between web-based information and actual coronavirus disease 2019 (COVID-19) cases, as well as if such data can be used to forecast patterns of disease spikes. Patients & methods Weekly data of COVID-19 cases in Pakistan was retrieved from online COVID-19 data banks for a period of 60 weeks. Search history related to COVID-19, coronavirus and the most common symptoms of disease was retrieved from Google Trends during the same period. Statistical analysis was performed to analyze the correlation between the two data sets. Search terms were adjusted for time-lag over weeks, to find the highest cross-correlation for each of the search terms. Results Search terms of ‘fever’ and ‘cough’ were the most commonly searched online, followed by coronavirus and COVID. The highest peak correlations with the weekly case series, with a 1-week backlog, was noted for loss of smell and loss of taste. The combined model yielded a modest performance for forecasting positive cases. The linear regression model revealed loss of smell (adjusted R² of 0.7) with significant 1-week, 2-week and 3-week lagged time series, as the best predictor of weekly positive case counts. Conclusions Our local analysis of Pakistan-based data seemingly confirms that Google trends can be used as an important tool for anticipating and predicting pandemic patterns and pre-hand preparedness in such unprecedented pandemic crisis.