CDC Says About 20% Get Long COVID. New Models Try Define It
As the number of people reporting persistent, and sometimes debilitating, symptoms from COVID-19 increases, researchers have struggled to pinpoint exactly how common so-called “long COVID” is, as well as how to clearly define exactly who has it or who is likely to get it.
Now, Centers for Disease Control and Prevention (CDC) researchers have concluded that 1 in 5 adults aged 18 and older have at least one health condition that might be related to their previous COVID-19 illness; that number goes up to 1 in 4 among those 65 and older. Their data was published last week in the CDC’s Morbidity and Mortality Weekly Report.
The conditions associated with what’s been officially termed post-acute sequelae of COVID-19, or PASC, include kidney failure, blood clots, other vascular issues, respiratory issues, heart problems, mental health or neurologic problems, and musculoskeletal conditions. But none of those conditions is unique to long-COVID.
Another new study, published in The Lancet Digital Health, is trying to help better characterize what long COVID is, and what it isn’t.
The research team, supported by the National Institutes of Health, used machine learning techniques to analyze electronic health record data to identify new information about long COVID and detect patterns that could help identify those likely to develop it.
CDC Data
The CDC team came to its conclusions by evaluating the EHRs of more than 353,000 adults who were diagnosed with COVID-19 or got a positive test result, then comparing those records with 1.6 million patients who had a medical visit in the same month without a positive test result or a COVID-19 diagnosis.
They looked at data from March 2020-November 2021, tagging 26 conditions often linked to post-COVID issues.
Overall, more than 38% of the COVID patients and 16% of those without COVID had at least one of these 26 conditions. They assessed the absolute risk difference between the patients and the non-COVID patients who developed one of the conditions, finding a 20.8 percentage point difference for those 18-64, yielding the 1 in 5 figure, and a 26.9 percentage point difference for those 65 and above, translating to about 1 in 4.
“These findings suggest the need for increased awareness for post-COVID conditions so that improved post-COVID care and management of patients who survived COVID-19 can be developed and implemented,” said study author Lara Bull-Otterson, PHD, MPH, co-lead of data analytics at the Healthcare Data Advisory Unit of the CDC.
Pinpointing Long COVID Characteristics
Long COVID is difficult to identify, because many of its symptoms are similar to those of other conditions, so researchers are looking for better ways to characterize it to help improve both diagnosis and treatment.
Researchers on the Lancet study evaluated data from the National COVID Cohort Collaborative, N3C, a national NIH database that includes information from more than 8 million people. The team looked at the health records of 98,000 adult COVID patients and used that information, along with data from about nearly 600 long-COVID patients treated at three long-COVID clinics, to create three machine learning models for identifying long-COVID patients.
The models aimed to identify long-COVID patients in three groups: all patients, those hospitalized with COVID, and those with COVID but not hospitalized. The models were judged by the researchers to be accurate because those identified at risk for long COVID from the database were similar to those actually treated for long COVID at the clinics.
“Our algorithm is not intended to diagnose long COVID,” said lead author Emily Pfaff, PhD, research assistant professor of medicine at the University of North Carolina Chapel Hill. “Rather, it is intended to identify patients in EHR data who ‘look like’ patients seen by physicians for long COVID.”’
Next, the researchers say, they will incorporate the new patterns they found with a diagnosis code for COVID and include it in the models to further test their accuracy. The models could also be used to help recruit patients for clinical trials, the researchers say.
Perspective and Caveats
The figures of 1 in 5 and 1 in 4 found by the CDC researchers don’t surprise David Putrino, PT, PhD, director of rehabilitation innovation for Mount Sinai Health System in New York City and director of its Abilities Research Center, which cares for long COVID patients.
“Those numbers are high and it’s alarming,” he said. “But we’ve been sounding the alarm for quite some time, and we’ve been assuming that about 1 in 5 end up with long COVID.”
He does see a limitation to the CDC research — that some symptoms could have emerged later, and some in the control group could have had an undiagnosed COVID infection and gone on to develop long COVID.
As for machine learning, “this is something we need to approach with caution,” Putrino said. “There are a lot of variables we don’t understand about long COVID,” and that could result in spurious conclusions.
“Although I am supportive of this work going on, I am saying, ‘Scrutinize the tools with a grain of salt.’ Electronic records, Putrino points out, include information that the doctors enter, not what the patient says.
Pfaff responds: “It is entirely appropriate to approach both machine learning and EHR data with relevant caveats in mind. There are many clinical factors that are not recorded in the EHR and the EHR is not representative of all persons with long COVID.” That data can only reflect those who seek care for a condition, a natural limitation.
When it comes to algorithms, they are limited by data they have access to, such as the electronic health records in this research. However, the immense size and diversity in the data used “does allow us to make some assertations with much more confidence than if we were using data from a single or small number of healthcare systems,” she said.
MMWR Morb Mortal Wkly Rep. Published online May 24, 2022. Full text
For more news, follow Medscape on Facebook, Twitter, Instagram, YouTube, and LinkedIn
Source: Read Full Article