U.K. and U.S. residents who reported a loss of smell and taste into a COVID-19 symptom-tracker app more often tested positive for the infection, according to a study published yesterday in Nature.

The COVID Symptom Study app was developed by health technology company Zoe Global with help from King's College London and Massachusetts General Hospital. It launched in the U.K. and the U.S. in late March, and after three weeks had gained more than 2.6 million users.

By applying a predictive model incorporating these testing rates to the hundreds of thousands of users who reported symptoms without testing data, the researchers estimated that a little over 5% of those who downloaded the app "are likely to have COVID-19." However, the researchers cautioned that this is likely an overestimation due to a handful of factors, such as the use of self-reported symptom data.

TOPLINE RESULTS

A total of 2,6618,862 residents from the U.K. (n = 2,450,569) and U.S. (n =168,293) downloaded the free study app and reported potential COVID-19 symptoms. Of these, 15,638 from the U.K. and 2,763 from the U.S. paired their reported symptoms with the results of a RT-PCR SARS-CoV-2 test.

Among the U.K. cohort, nearly two thirds of those who reported positive test results also reported a loss of smell and taste, compared to the roughly 23% of negative cases who reported the same symptoms, yielding an odds ratio [OR] of 6.4 (p < .0001). Combining these rates with the U.S. cohort's OR of 10.01 (p < .0001) produced an overall OR of 6.74 (P < .0001) after adjustments.

"We report that loss of smell and taste is a potential predictor of COVID-19 in addition to other, more established symptoms including high temperature and a new, persistent cough," the researchers wrote. "COVID-19 appears to cause problems of smell receptors in line with many other respiratory viruses, including previous coronaviruses thought to account for 10%–15% of cases of anosmia."

The analysis also found 10 other symptoms such as fever, fatigue, shortness of breath and diarrhea to be associated with positive diagnostic test among the U.K. cohort, while only loss of smell and taste, fatigue and skipped meals were associated among the U.S. cohort.

From these findings, the researchers build a symptoms-prediction model to estimate the number of untested participants (n = 805,753) who likely had the disease. According to the model, 140,312 (17.42%) symptom-reporting participants likely were infected, representing 5.36% of the full study cohort.

HOW IT WAS DONE

The COVID Symptom Study app was released for free in the U.K. on March 24, and in the U.S. on March 29 to residents with or without apparent COVID-19 symptoms. Users record their location, age and core health-risk factors upon initial use, and then update the tool daily with their symptoms, healthcare encounters, diagnostic test results and quarantine behaviors.

Relationships between symptoms and COVID-19 test results were identified via multivariate logistic regression with adjustments for age, sex and BMI.

The researchers highlighted a handful of limitations inherent to their study design, chief among which were the reliance on self-reported data, rather than physiologic assessment, and participant self-enrollment. Further, because SARS-CoV-2 testing is more common among those with likely cases, the model's training sample was not representative of the general population.

THE LARGER TREND

The last couple of months have seen a wave of digital symptom-checkers, like the COVID Symptom Study app, released by technology companies, health systems and governments alike. In fact, just this weekend the World Health Organization shared plans for a symptom-checker app that it will be providing to countries without the resources to build and distribute their own digital tool.

While the debate over whether these apps and others for contact tracing are appropriate (or even effective) is ongoing, the Nature study offers an example of how large-scale data collection could be used to help public health identify predictors of the disease and understand its potential spread.

Home to the Hermitage Museum and five million residents, St. Petersburg is the second-largest city in Russia. The city also includes around 30,000 doctors, 250 public hospitals and 3,000 private clinics, translating to about 50 million healthcare encounters per year. 

“This produces a huge amount of data. This data was distributed between different healthcare organizations,” Eugene Kogan, Chief Technical Officer of the Center for Healthcare Informatics and Analytics in St. Petersburg, said during the HIMSS20 Digital webinar Problem-Oriented Regional Patient Summary. “However, this data must be shared across organizations because it is needed to achieve continuative care when the patient comes to the next organization.”

In St. Petersburg, each organization is able to implement its own electronic health record system. However, Kogan and his team began to look into ways to help link these records. 

“In 2014 we started to create a regional central database to collect the data. We started with a laboratory data exchange. Why laboratory data?” Kogan said. “Because we needed to create an incentive for doctors to use the information systems, so we had to give them something vital, which was laboratory results.” 

Kogan’s team looked into a number of options before choosing HL7 FHIR. 

“Every hospital information system is integrated by FHIR. Lab orders are sent to the central system. The laboratory receives the order,” Kogan said. “When tests are ready, the lab sends the diagnostic report to the central system. That report becomes visible to the doctor who ordered this test, every doctor who sees the patients and to the patient himself.” 

The next step, in 2018, was launching a regional integrated EHR system. 

“All healthcare institutions now must send patient health records to a patient regional database,” Kogan said. “We collect patient documents from all levels of healthcare organizations. We are therefore able to give an integrated record to a doctor and achieve the continued care, at least that is what we thought.”

This meant a lot of digital paperwork coming at the doctors left and right – and not all of it was useful. 

“We supplied our doctor with all the patient medical records. Did we help him? Not so much,” he said. “Theoretically EHRs are a great instrument for continuative care. When we look at it in detail in the real physician’s workflow we see that it is nearly impossible to use. So unfortunately, EHR itself gives nearly no difference to the paper medical record.”

The records were being organized by source – so, for example, all of the lab records were in one bucket, all of the images were in another and so on. 

“A doctor trying to find answer to specific questions sees a mixture of records, both relevant and irrelevant, because they are structured by source and not by purpose,” he said. 

In order to remedy this the team began to use an alternative approach called problem-oriented medical records. This mean that the reports would each represent a problem, in theory making it easier for a doctor to find files related to the aliment he or she is trying to treat. Historically, this type of organization has a number of hurtles, he said. The issues range from not all headings being actual problems to some information belonging in two categories.

“As we decided to make our EHR problem orientated, we decided to overcome this limitation with two instruments: an author-generated patient summary and problem summary,” he said. 

They decided to build a system that would allow for creating an automated patient summary, using the codes that a doctor uses to describe an ailment. Kogan noted that different doctors may diagnose a condition differently, but the system takes it all into account. 

“Why were the codes different? Maybe one doctor is right and the other is wrong,” he said. “Maybe both are right but the patient’s condition has changed. It is important that our algorithm doesn’t decide whether opinions are right or not. It aggregates opinions into problem.”

The team also created a problem-summary list, which needed more specifics on a condition and required a consensus on what a condition was. It was similar to a clinical registry, where there is a list of agreed parameters for each condition and the doctors registers the data. In the problem summary, it will also include a timeline and relevant history. 

“We have discussed the three tools for continuative care, integrated EHR, patient summary and patient-problem summary. We see these tools, being brought together, can support continuative care, not just in theory, in the real physician workflow,” he said.