Google’s online search activity up to 17 days in advance could help predict predictions in Covid-19 cases, a new study reveals.
Researchers from University College London have created computer models based on online search frequencies to gain insight into the incidence of the disease in several countries, including the United Kingdom.
Models based on online searches have successfully reported and confirmed Covid-19 cases and deaths, with 16.7 and 22.1 days, respectively.
The team’s analysis was one of the first to find an association between the appearance of Covid-19 and investigations into the symptoms of loss of sense of smell and skin rash – two symptoms of the disease listed by Public Health England.
Online search data should be used with ‘more established approaches’ to develop methods of monitoring public health for Covid and other new infectious diseases, experts say.
Google’s online search data could help inform the public health response to Covid-19, according to a report by University College London academics. Previous research has shown that different characteristics of infectious diseases can be deduced from online search behavior
“This study provides a new set of tools that can be used to detect Covid-19,” said the study’s lead author, Dr. Vasileios Lampos, told University College London.
“We have shown that our approach works in different countries, regardless of cultural, socio-economic and climate differences.”
UCL researchers used Covid-19’s symptom profile to develop models of its occurrence by looking at Google’s symptom-related searches.
They then calibrated these models to reduce the bias in these ‘signals’ caused by the public interest – in other words, the impact that media coverage has on online searches.
They developed the uncalibrated model by choosing search terms related to Covid-19 symptoms, identified by the NHS and Public Health England (PHE).
The three most common symptoms of Covid-19 are a high temperature, a new and persistent cough, and a loss or change in sense of smell or taste.
PHE also contains some less common symptoms, including aches, headaches and skin rashes.
The terms were weighted according to their occurrence ratio in confirmed Covid-19 cases.
According to UCL, this model provided ‘useful insights’, including early warnings, and shows the effects of physical distance measures.
The calibrated version, which took into account the news coverage, enabled academics to give PHE a model to more accurately predict trainings in the UK.
The model has been applied in several countries, including the United Kingdom, the USA, Italy, Australia and South Africa.
They found that the same pattern appeared, in that increases in cases were predicted by their model.
Graph shows online search scores for Covid-19 for different countries at the end of 2019 and beginning of 2020. The frequencies of queries are weighted according to the probability of the occurrence of symptoms (blue line) and have the media effects to a minimum limited (black line). Dates for physical distance or closure measures are indicated by dotted vertical lines
“Our best chance of tackling health emergencies such as the Covid-19 pandemic is to detect it early to act early,” said co-author of the study, Professor Michael Edelstein, of Bar-Ilan University, Israel, said.
“Using innovative approaches to disease detection, such as analyzing Internet activities to complement established approaches, is the best way to identify outbreaks early.”
Academics working on the models share their findings with PHE on a weekly basis to support the response to the disease, which is available online.
“We are delighted that public health organizations such as PHE have also recognized the usefulness of these new and non-traditional approaches to epidemiology,” Dr Lampos said.
The analysis of internet search activities is an established method of detecting and understanding infectious diseases, and is currently used to monitor seasonal flu. Influenza detection estimates flu-like disease rates in England based on internet searches and is included in the English flu surveillance statistics in England.
The analysis of internet activities is an established method of detecting and understanding infectious diseases.
The technique is already used monitor seasonal flu in the form of UCLs Flu Detection.
The constantly updated online tool estimates flu-like illness rates in England based on internet searches and is included in the flu surveillance standards in England.
“Previous research has shown the usefulness of online search activities for modeling infectious diseases such as influenza,” said Dr Lampos.
The study, ‘Tracking COVID-19 using online search’, was published today in Nature Digital Medicine.