Evidence explainer

Heart, lung, and acute care

How Researchers Define a Moving Target Called Long COVID

Long COVID has two definitions, one broad and clinical from 2024 and one a narrower research symptom score. Because both rest on what patients feel, the same person can meet one and miss the other.

Fully reviewed by Jasaman (Jasmin) Tojjar, MD, PhD

On this page
  1. Key points
  2. Defining a condition with no test to anchor it
  3. The research tool: the RECOVER symptom index
  4. The clinical definition: NASEM 2024
  5. Why the same person can fall on both sides of the line
  6. What this does to the numbers
  7. How to read a Long COVID study

Ask two credible sources whether a given person has Long COVID and you can get two honest, opposite answers, because there is no single agreed rule and no blood test to settle the question. The condition is currently defined by what patients report rather than by anything a lab can confirm, and the two most influential definitions were written to do different jobs. Understanding why they diverge is the key to reading Long COVID research without being misled by numbers that seem to contradict each other.

Key points#

Defining a condition with no test to anchor it#

Most diagnoses eventually rest on something measurable: a culture, an antibody, an image, a biopsy. Long COVID has none of that yet. Picture trying to draw a coastline that keeps shifting with the tide. Wherever you place the line, part of it is a judgment call, and two careful people will place it differently. That is the situation researchers face, and it is why a definition is not a technicality here. The definition is the instrument, and the instrument shapes the answer.

Two efforts dominate the field, and it helps to meet the narrower one first, because it shows what a research team does when it needs precision.

The research tool: the RECOVER symptom index#

The NIH RECOVER initiative set out to answer a specific question: can we identify a group of people who are distinct enough from uninfected peers that we can study the biology and test treatments in them? Its 2023 analysis in JAMA compared self-reported symptoms in roughly 9,700 adults across 85 sites, most of whom had a prior SARS-CoV-2 infection. The team found 37 symptoms that were meaningfully more common after infection, then narrowed the list to 12 and turned them into a weighted score. Each symptom earned points according to how strongly it separated infected from uninfected participants, with post-exertional malaise and loss of or change in smell or taste among the most heavily weighted. A total of 12 points or more sorted someone into the research category of PASC, short for post-acute sequelae of SARS-CoV-2 infection.

The authors were direct about the limits. This index is a research starting point, not a bedside test, and it was never meant to tell one person whether they "have" Long COVID. Among participants first infected during the Omicron period, about 10 percent met the threshold at six months, a figure the paper reports with careful uncertainty. The value of the score is that it produces a reasonably consistent cohort. The cost is that it leaves out real patients whose symptoms do not add up to the cutoff.

The clinical definition: NASEM 2024#

A year later, at the request of federal health offices, the National Academies of Sciences, Engineering, and Medicine took the opposite approach. Its 2024 definition is deliberately wide: "Long COVID is an infection-associated chronic condition that occurs after SARS-CoV-2 infection and is present for at least 3 months as a continuous, relapsing and remitting, or progressive disease state that affects one or more organ systems." The report catalogs more than 200 documented symptoms across many organ systems, and it pointedly does not require a positive test, because so many early infections were never confirmed when testing was scarce.

The reasoning is clinical rather than statistical. As the committee argued in an accompanying New England Journal of Medicine article, a broad definition helps symptomatic people get recognized, coded, and treated instead of being turned away for lacking proof of infection. It is inclusive on purpose. It was not designed to build a clean study population, and it does not pretend to.

Why the same person can fall on both sides of the line#

Put the two definitions side by side and the friction is obvious. A person with three months of fatigue and brain fog after a probable infection clearly fits the NASEM clinical definition. That same person might score only 8 or 9 on the RECOVER index and never cross the research threshold. Reverse the case and someone who scores 14 in a study might, in a clinic, be told their symptoms have another explanation. Neither definition is wrong. They were tuned for different tasks, one to be generous so patients are not missed, the other to be selective so a research signal stays clean.

This is the classic tension between sensitivity and specificity, and no symptom-only definition escapes it. Cast a wide net and you capture the true burden of illness but mix together people who may have different underlying problems. Draw a tight boundary and you get a purer group for study while excluding patients who are genuinely affected.

What this does to the numbers#

Because the ruler shapes the measurement, three predictable problems follow.

Prevalence tracks the definition. A broad clinical definition that accepts any qualifying symptom will always count more people than a research index demanding a specific point total. Two headlines that appear to disagree are often just measuring different things with different tools, which is why no single trustworthy prevalence number exists.

The signal drifts as the virus changes. The RECOVER index was calibrated on particular populations during particular variant waves. Symptom patterns shift as variants evolve, as vaccination and prior infection accumulate, and as reinfections mount. A score fixed to one era can misjudge another.

Trials inherit the ambiguity. Case definitions and endpoints are the scaffolding of a clinical trial. When entry criteria admit a mixed group and the outcome is a symptom score that can drift, a real treatment effect is easy to dilute and hard to reproduce. A therapy that helps one biological subtype can look useless when tested in a broad group defined only by how people feel.

How to read a Long COVID study#

The practical habit is simple. Before comparing one study's numbers to another's, find out which definition each one used, and treat the figures as answers to slightly different questions. Both the NASEM committee and the RECOVER investigators say plainly that their definitions are provisional and will be revised as the biology, including candidate biomarkers and distinct subtypes, comes into focus. Reading either as the final word misreads what it was built to do.

Sources and further reading

  1. NASEM, A Long COVID Definition (2024)
  2. RECOVER, Development of a Definition of PASC, JAMA 2023
  3. NASEM committee, Long Covid Defined, NEJM 2024

Questions and answers

Is there a blood test for Long COVID?

Not yet. As of these definitions, there is no validated biomarker, so both the clinical and research definitions rely on symptoms and timing rather than a laboratory result.

Why do prevalence estimates vary so much?

Because they use different definitions. A broad clinical definition counts far more people than a narrow research score, so the same underlying reality can produce estimates that differ several times over.

Which definition should a patient use?

Neither was designed for self-diagnosis. The clinical definition supports recognition and care through a clinician, while the research score exists to build study groups. Anyone with persistent symptoms after infection should talk with a clinician.