Estimated half with likely long COVID at 12 months fell below a symptom index by 36 months

A prospective cohort analysis in the Annals of Internal Medicine followed 1,019 adults at 83 U.S. sites who still met a research definition of likely long COVID 12 months after a first SARS-CoV-2 infection. Harrison T. Reeder and colleagues, in the NIH RECOVER-Adult program, estimated time to symptom improvement rather than testing a treatment. The primary endpoint was not a clinic diagnosis or a hospitalization. It was two consecutive visits with a Long COVID Research Index below 11.

Using time-to-event methods, the authors estimated that 36.2 percent met that primary definition by 24 months after first infection and 50.1 percent by 36 months. By 36 months, 56.4 percent and 26.9 percent met two secondary outcomes; the news summary does not say which of three listed endpoints those figures map to. Prespecified secondaries were at least a 50 percent index drop, good or better quality of life, and an index of zero. People who still had likely long COVID at 18 or 24 months had similar later improvement timing.

These estimates are conditional on still meeting the index at 12 months, and about half had not crossed below 11 on two visits by three years after infection. The index, from earlier RECOVER work in JAMA in 2023, is a weighted symptom score, not a validated surrogate for organ recovery, work capacity, or health-care use. Regression to the mean, reinfection, vaccination, variant era, and loss to follow-up can move a score without establishing a disease course. A threshold crossing is not the same outcome as a 50 percent reduction or an index of zero.

For care planning, the useful number is the unresolved half, not only the improving half. Disability assessments and trial sample sizes should assume that many people with prevalent long COVID at one year will still score in the likely range two years later. This design can describe timing under routine conditions. It cannot show that any therapy caused improvement, cannot separate subtypes, and cannot show that a lower index means recovered function. Next studies need clinical outcomes, reinfection handling, and randomized tests powered against this background rate.