QCovid prediction algorithm wrongly identifies high-risk patients

The reliability of an algorithm used to predict the risk of someone catching, being admitted to hospital, or dying from COVID-19 in England, UK, hit the headlines for wrongly identifying hundreds of thousands of patients as high risk.

Based on data from 'the first few months of the pandemic', the University of Oxford-developed QCovid model took into account various socio-economic indicators, underlying health conditions such as diabetes and heart disease, body-mass index, and postcode, among other factors, to return an 'absolute risk of a covid-19 associated death' or hospitalisation.

Following the launch of the tool, an additional 1.7 million people were instructed to shield, with around 800,000 people moved up the priority list to be vaccinated. These included women with previous gestational diabetes but who were healthy and could not understand why it was being recommended that they shield.

Some General Practitioners also described seeing healthy young men on the list. 

Young, healthy people are less likely to have measurements such as body weight recorded in their health records, Irene Petersen, professor of epidemiology and health informatics at University College London, told The Guardian.

The algorithm's performance led to widespread confusion and anxiety for healthy individuals, and created a massive logistical burden for General Practitioners (GPs) who had to field thousands of calls and manually correct records. 

It also risked delaying doses for those truly in need by diluting the high-priority pool. 

Timeline ⏰

• February 16, 2021. The UK Government officially introduces QCovid in England to identify a "new group" of high-risk individuals

• February 17-25, 2021. 1.7 million people are added to the Shielded Patient List (SPL). Letters are sent advising them to stay home and move up the vaccine queue

• March 2021. GPs report a surge in "false positives." Reports emerge that 400,000 to 800,000 healthy individuals were wrongly flagged due to missing BMI or ethnicity data

• April 2021. NHS Digital and GPs begin the process of "de-shielding" patients who were incorrectly identified by the algorithm

• November 25, 2021. QCovid 2 and 3 updates are integrated into clinical tools, incorporating vaccination status and refined data-handling rules.

News, commentary, analysis 🗞️

• https://www.theguardian.com/society/2021/mar/09/scientists-question-nhs-algorithm-as-young-people-called-in-for-jab

• https://www.theguardian.com/uk-news/2021/feb/16/qcovid-how-improved-algorithm-can-identify-more-higher-risk-adults

• https://www.dailymail.co.uk/news/article-9351305/Coronavirus-NHS-wrongly-marked-430-000-people-high-risk-Covid.html

• https://www.pulsetoday.co.uk/news/clinical-areas/diabetes/history-of-gestational-diabetes-prompting-patients-to-shield-under-new-algorithm/

• https://www.pulsetoday.co.uk/news/coronavirus/over-400000-patients-given-inflated-covid-risk-scores-due-to-missing-data/

• https://www.telegraph.co.uk/global-health/science-and-disease/computer-says-yes-new-covid-vaccine-algorithm-will-make-us-safer/

• https://www.politicshome.com/news/article/17m-more-people-told-to-shield-due-and-800000-moved-up-vaccine-priority-list-due-to-new-modelling

• https://www.gponline.com/qcovid-tool-reviewed-nhs-digital-healthy-women-wrongly-advised-shield/article/1708003

• https://www.pulsetoday.co.uk/news/clinical-areas/diabetes/nhs-digital-to-review-qcovid-tool-after-healthy-women-were-told-to-shield/

• https://www.digitalhealth.net/2021/02/nhs-digital-reviewing-algorithm-after-women-incorrectly-told-to-shield/

• https://www.digitalhealth.net/2021/02/new-groups-which-could-be-at-high-risk-from-covid-19-identified-with-tech/

• https://www.economist.com/graphic-detail/2021/03/11/how-we-built-our-covid-19-risk-estimator