Fact checked byShenaz Bagha

Read more

April 17, 2024
2 min read
Save

Algorithm scans outpatient letters, identifies those who must ‘shield’ against COVID-19

Fact checked byShenaz Bagha
You've successfully added to your alerts. You will receive an email when new content is published.

Click Here to Manage Email Alerts

We were unable to process your request. Please try again later. If you continue to have this issue please contact customerservice@slackinc.com.

Key takeaways:

  • An algorithm analyzed outpatient letters to quickly determine who needed to ‘shield’ against COVID-19 per national guidelines.
  • The algorithm could be applied to future public health initiatives.

An automated algorithm designed to identify rheumatology patients who need to “shield” against COVID-19 demonstrated results that matched the “gold standard” of manual review in the majority of cases, according to data.

“In April 2020, at the start of the COVID-19 pandemic, the U.K.’s Scientific Committee issued extreme social distancing measures termed ‘shielding,’” Meghna Jani, MRCP, MSc, PhD, of the University of Manchester, and colleagues wrote in Annals of the Rheumatic Diseases. “These were aimed at a subset of the U.K. population who were deemed clinically extremely vulnerable (CEV) to infection. ... In outpatient-based specialties, such as rheumatology, where communication is commonly based on unstructured outpatient clinic letters, it was not possible to run a rapid search for patients with diagnostic and medication codes.

N95 mask on face 1
An algorithm aimed at identifying rheumatology patients who need to “shield” against COVID-19 demonstrated results that matched the “gold standard” of manual review in the majority of cases, according to data. Image: Adobe Stock

“Diagnoses for outpatient visits, outpatient prescribed medications (ie, majority of high-cost drugs such as biologics) and measures such as disease severity are only recorded in semistructured letters and are thus not machine readable, searchable or analyzable across the population,” they added. “Hospital clinicians needed to manually review sequential clinic letters of patients to manually score their risk as per national guidance and identify which patients should receive communications about shielding measures. This took whole rheumatology teams many dedicated sessions amounting to scores of hours to read through recent letters for all their patient cohort.”

To create an automated algorithm to accomplish this more quickly, Jani and colleagues used outpatient letters for 803 patients at the Salford Royal Hospital, using the two most recent letters for each patient prior to April 2020.

“Outpatient letters in the U.K. are the main method of communication between hospitals and primary care physicians,” they wrote. “They are written as per the Professional Records Standards Body (PRSB) guidance and comprise of semistructured data that include relevant diagnoses and medications with additional free-text narrative.”

The researchers used Intelligent Medical Objects software to map diagnoses to Systematized Noemclature of Medicine Clinical Terms codes. They then developed a tool to retrieve medication types, doses, durations and active or past status by combining existing text mining tools.

The algorithm combined age, diagnosis and medication variables to calculate a shielding score based on British Society for Rheumatology guidelines. Its sensitivity, specificity and F1 score were compared against the “gold standard” of manual reviews by the rheumatology consultant team at the time of the pandemic, the researchers wrote.

The algorithm demonstrated a sensitivity of 80% (95% CI, 75%-85%), a specificity of 92% (95% CI, 90%-94%), and an F1 score of 0.81, according to the researchers. When applied to the records of an additional 15,865 patients, the algorithm took 18 hours to extract medication and diagnoses and 1 hour to determine shielding.

“An automated algorithm for risk stratification has several advantages, including reducing clinician time for manual review to allow more time for direct care, rapid deployment of a complex set of rules on all patient records across the whole department (rather than a subset) and improving efficiency and transparently communicating decisions based on individual risk,” Jani and colleagues wrote. “With further development, it has the potential to be adapted in other specialties that used similar risk stratification and for future public health initiatives that require prompt automated review of hospital outpatient letters.”