RECOVER Research Update: December 2025

Long COVID is a complex health condition that can affect everyone differently. For that reason, no single research study can answer all questions about who Long COVID affects, how it affects them, and what healthcare providers can do to help.

Each of the 5 recent RECOVER publications discussed below has a unique focus, design, and method of advancing our understanding of Long COVID. Taken together, these 5 studies also show how RECOVER researchers can use the same data, tools, and previous findings to make more progress and achieve impact faster.

Effect of Paxlovid treatment during acute COVID-19 on Long COVID onset: An EHR-based target trial emulation from the N3C and RECOVER Consortia (PLOS Medicine, September 2025)

• Paxlovid is an antiviral medication that can help people who have COVID-19 experience less severe symptoms. Because studies show that people who become very sick with COVID are more likely to develop Long COVID, researchers wanted to know if taking Paxlovid can also help prevent Long COVID.
• Researchers used the National Clinical Cohort Collaborative’s (N3C’s) electronic health record (EHR) database to identify more than 445,000 people who had COVID between April 2022 and February 2023.
• They then created 2 smaller groups based on EHR data: people with COVID who took Paxlovid and people with COVID who did not take Paxlovid. People in both groups also had at least one health factor that put them at a higher risk of being hospitalized because of their COVID symptoms.
• This study used a method called a target trial emulation. Similar to a clinical trial, a target trial emulation evaluates the safety and effectiveness of a medical treatment. Unlike a clinical trial, which gives people medical treatments and examines how those treatments change people’s health, a target trial emulation studies EHR data to measure how treatments can affect people’s health in real-world settings.
• Researchers found that taking Paxlovid did not significantly reduce people’s risk of developing Long COVID. Only people aged 65 or older, or who had multiple pre-existing health problems were slightly less likely to develop Long COVID after taking Paxlovid.
• Researchers also found that Paxlovid was not an effective treatment for Long COVID symptoms like shortness of breath, cough, and chest pain (respiratory symptoms). However, Paxlovid was a slightly effective treatment for 2 other types of Long COVID symptoms: those affecting a person’s ability to think clearly, concentrate, and remember things (cognitive symptoms), and those making a person feel constantly tired and unable to participate in everyday activities, like going to work or school (fatigue).

Olfactory dysfunction after SARS-CoV-2 infection in the RECOVER adult cohort (JAMA Network Open, September 2025)

• Change in and loss of smell or taste are common symptoms of COVID-19. These changes can go away after a person recovers from COVID-19, or they can be persistent (long lasting).
• The technical term for these changes in smell or taste is olfactory dysfunction. Forms of olfactory dysfunction include a reduction in a person’s ability to smell or taste (hyposmia or microsmia) and the loss of the ability to detect some or all smells (anosmia).
• Olfactory dysfunction has also been associated with cognitive dysfunction in other conditions (a decline in a person’s ability to concentrate, think clearly, and remember things).
• To gain a better understanding of how olfactory dysfunction can affect people who have had COVID-19, researchers examined the results of smell tests taken by more than 3,500 adults taking part in RECOVER observational studies.
• Of study participants who had COVID-19 and reported experiencing a change in or loss of smell or taste, 80% tested positive for hyposmia. Moreover, 66% of study participants who had COVID-19 but did not report experiencing a change in or loss of smell or taste tested positive for hyposmia.
• Of study participants who had COVID-19 and reported experiencing a change in or loss of smell or taste, 66% also had trouble thinking or concentrating. In comparison, 38% who had COVID-19 but did not report experiencing a change in or loss of smell or taste also had trouble thinking or concentrating.
• These findings suggest that persistent changes in smell or taste are more common in people who have had COVID-19 than previously thought. Researchers recommend making smell tests a routine part of care plans for people living with Long COVID. They also recommend additional studies to explore the possible links between olfactory dysfunction and cognitive dysfunction.

Effect of obesity on the acute response to SARS-CoV-2 infection and development of post-acute sequelae of COVID-19 (PASC) in nonhuman primates (PLOS Pathogens, July 2025)

• People with obesity who get COVID-19 are more likely to experience worse symptoms than people without obesity. However, it is unclear whether people with obesity are also more likely to experience long-term health effects associated with COVID-19, including Long COVID.
• Researchers designed a pathobiology study using model organisms to investigate how SARS-CoV-2, the virus that causes COVID-19, may produce different changes in the body depending on a person’s weight. Learn about model organisms in this RECOVER story about pathobiology research.
• For 6 months, researchers monitored the health of macaques (a type of monkey) with and without obesity who had COVID. They monitored the macaques’ health using measurements such as weight, body temperature, lung health, and the amount of SARS-CoV-2 in their blood.
• Researchers found that the bodies of the macaques in both groups changed in response to SARS-CoV-2. Some of these changes were associated with different Long COVID symptoms and health effects. For example, the macaques with obesity were more likely to experience damage to their lungs than macaques without obesity. On the other hand, the macaques without obesity were more likely to show signs of disturbed sleep than macaques with obesity.
• These findings suggest that obesity may impact which Long COVID symptoms a person is likely to experience. Researchers believe future studies could clarify whether obesity is a risk factor for the development of Long COVID. Specifically, researchers recommend that additional studies focus on how SARS-CoV-2 might cause changes in white adipose tissue (WAT), a type of body fat more present in the bodies of people with obesity, and whether those changes can increase a person’s risk of developing Long COVID.

RECOVER researchers use EHR data to help identify Long COVID

RECOVER researchers can use electronic health record (EHR) studies to create valuable tools called computable phenotypes that can help identify Long COVID in individuals. A computable phenotype creates a definition of Long COVID by using advanced computer programs called machine-learning (ML) algorithms. These algorithms use large amounts of EHR data to find patterns about patients and their symptoms, medical test results, and diagnoses. 

Computable phenotypes offer several advantages for Long COVID research. Because researchers use very large sets of EHRs from across the US, these phenotypes allow researchers to study Long COVID faster with more efficiency, consistency, and wide applicability. For example, using a computable phenotype, researchers can identify a group of people who may be eligible for a study at one study site based on how similar that group of people is to a group at a different site. 

The next 2 studies used computable phenotypes to help improve the ability to identify children who might have Long COVID.

Identifying pediatric Long COVID: Comparing an EHR algorithm to manual review (Applied Clinical Informatics, October 2025)

• Long COVID can affect children differently than adults. These differences highlight the need for a formal definition of Long COVID (pediatric Long COVID) that researchers and healthcare providers can use.
• In this study, researchers wanted to know if a computable phenotype could produce a working definition of pediatric Long COVID. To create their computable phenotype for pediatric Long COVID, researchers used data from the National Patient-Centered Clinical Research Network (PCORnet) EHR database. This database includes the EHRs of more than 9 million children from 40 hospital systems across the United States who were tested for, diagnosed with, or vaccinated for COVID-19 between 2019 and December 2022.
• Researchers tested the accuracy of their computable phenotype by studying how well it identified children with Long COVID compared with identification by healthcare providers. Researchers created a small sample population of 651 children from the EHRs. The computable phenotype identified 318 children with Long COVID and 333 without Long COVID. Then, based on information in the EHR, a healthcare provider determined whether each child had Long COVID. Healthcare providers identified 239 children with Long COVID and 412 children without Long COVID.
• Researchers found that the computable phenotype and the healthcare providers did not always agree in determining which children had Long COVID. Specifically, the computable phenotype agreed with healthcare providers in 406 out of 651 cases.
• In 73% of cases where the computable phenotype identified a child with Long COVID and the healthcare provider didn’t, the provider associated a symptom with a pre-existing health condition instead of Long COVID.
• There were also cases in which the provider identified a child with Long COVID and the computable phenotype didn’t. In 36% of these cases, the provider based their decision on a symptom that was not included in the computable phenotype’s definition. In 33% of these cases, the provider used a longer time frame than the computable phenotype to classify a symptom as chronic (long lasting).
• These findings demonstrate that a computable phenotype can be useful in identifying children with Long COVID. However, these findings also highlight the need for healthcare providers and researchers to establish a formal definition of pediatric Long COVID. Researchers recommend that additional studies continue to refine the computable phenotype to account for pre-existing conditions, symptoms caused by those pre-existing conditions, and how pediatric Long COVID can cause symptoms to worsen over time.

A natural language processing pipeline for identifying pediatric Long COVID symptoms and functional impacts in freeform clinical notes: A RECOVER study (JAMIA Open, October 2025)

• In addition to formal diagnoses entered using codes, electronic health records (EHRs) often include notes from healthcare providers. These notes can contain important details about a person’s symptoms and how these symptoms are impacting their day-to-day activities (functional impacts).
• Although these notes contain valuable information, they can be hard to interpret and understand because each provider can use different words to describe the same idea. Each provider can also enter their notes in almost any order, so these notes do not have a standard structure.
• Researchers tested whether a form of artificial intelligence (AI) called a natural language processing (NLP) algorithm could find patterns in these notes that might help identify children with Long COVID.
• First, researchers chose sets of words (terms) from definitions of 25 symptoms and functional impacts common to Long COVID in children. Next, they had the algorithm look for these terms in more than 42,000 notes from EHRs. Finally, they evaluated the algorithm by studying how well it performed when analyzing 2 groups: children whose EHRs suggested they had Long COVID and children whose EHRs suggested they did not have Long COVID.
• Using only terms found in notes, the algorithm was consistently able to identify more children who might have Long COVID. When looking at only diagnostic codes and combinations of notes and diagnostic codes, the algorithm identified fewer children who might have Long COVID. The algorithm also consistently found higher numbers of Long COVID-related terms in the EHR notes of children who probably had Long COVID compared with those of children who probably did not.
• These findings show that AI tools can quickly and accurately gain insights about Long COVID from health data that, although valuable, can be challenging to learn from because of inconsistencies within clinical notes. Until scientists develop diagnostic tests for Long COVID, computable phenotypes and AI-powered insights can help healthcare providers make sure that the children who need Long COVID care receive it.

Continue exploring the different types of Long COVID studies RECOVER researchers are conducting on the Publications page.