Long COVID Score Under the Microscope: Researchers Defend Clinical Validation of RECOVER Index

A scholarly dispute over how a promising Long COVID screening tool should be judged has erupted into print, and the exchange offers a rare, candid window into one of the most consequential methodological questions in post-pandemic medicine: how do you validate a diagnostic index for a disease that has no biological gold standard? In a letter published in the Journal of General Internal Medicine, a team from Johns Hopkins University has responded point by point to critics who questioned the interpretation of their external validation study of the RECOVER PASC score, the research index developed by the National Institutes of Health’s RECOVER Initiative to identify individuals with Long COVID, formally known as Post-Acute Sequelae of SARS-CoV-2 Infection.

The controversy began when Drs. Goldman and Martin published a critique arguing that the Johns Hopkins study, led by Dr. Alba Azola along with Dr. Rebecca T. Veenhuis and Dr. Leah H. Rubin, had been framed in a way that overstated what its results could show. Their central claim was that the study, by recruiting patients from specialty clinics rather than from the general population, could not speak to how the RECOVER PASC score would perform as a population-based screening instrument. In their response, the Johns Hopkins team does not dispute the mathematical logic behind that concern. Instead, they argue that Goldman and Martin have misidentified the question the study was designed to answer in the first place.

That question, the authors explain, was deliberately clinical rather than epidemiological. Their investigation asked how well the RECOVER PASC score classifies individuals who received a clinical diagnosis of Long COVID after comprehensive, multidisciplinary evaluation in specialty clinics, compared with individuals who had documented SARS-CoV-2 infection but recovered without persistent symptoms. This is a fundamentally different exercise from estimating how the score would behave if applied to an entire community, where the mix of patients, symptom burdens, and competing diagnoses would look very different. The team emphasizes that their goal was to contribute to the ongoing independent evaluation and iterative refinement of the index, not to certify it as a definitive diagnostic test.

The technical heart of the debate concerns what statisticians call the reference standard, the benchmark against which a new test is measured. For many diseases, a laboratory assay or imaging finding can serve as an objective gold standard. Long COVID has no such benchmark. In the absence of a biological marker, the Johns Hopkins team turned to expert clinical diagnosis based on the 2024 consensus definition issued by the National Academies of Sciences, Engineering, and Medicine, which they describe as the most appropriate clinical reference standard currently available. They acknowledge candidly that this is a pragmatic comparator rather than a true gold standard, but argue that emerging research indices must be evaluated against something, and expert diagnosis grounded in a consensus definition is the best available option.

The authors also point to the evolution of the RECOVER index itself as evidence that such evaluation is expected and welcome. In 2024, the RECOVER-Adult Long COVID research index was updated to incorporate additional participant data, expanded symptom ascertainment informed by input from the patient community, and a revised symptom-weighting model and threshold. An index designed to be revised as new evidence accumulates, they argue, naturally invites the kind of external scrutiny their study provided. Testing a tool in settings that differ from those in which it was developed is a cornerstone of clinical measurement science, and the specialty referral cohort, with its rigorous phenotyping, offers exactly the kind of demanding test case that can reveal where an index succeeds and where it falls short.

One of the sharpest points of contention involved enrollment criteria. Goldman and Martin suggested that requiring participants to have at least one neuropsychiatric symptom, such as brain fog, biased the study toward higher sensitivity, inflating the apparent ability of the score to detect true cases. The Johns Hopkins team agrees that the criterion defines a specific clinical spectrum of Long COVID and must be weighed when interpreting the results, but they reject the suggestion that it contaminated the comparison. The requirement, they explain, reflected the design of the parent study funded by the National Institute of Mental Health and the clinical focus of their Brain Health Program, and it was explicitly described in the original manuscript. Crucially, participants were not selected based on their RECOVER PASC score or on meeting any component of the score’s threshold, and brain fog itself was not required for enrollment. The neuropsychiatric criterion, in other words, shaped the referral population under study rather than smuggling the index into the reference classification.

The choice of comparator group drew similar scrutiny. The Johns Hopkins study compared clinically diagnosed Long COVID patients against people who had documented SARS-CoV-2 infection and recovered without lingering symptoms, a design intended to test whether the score can discriminate between persistent illness and uncomplicated recovery. The authors concede that future studies comparing Long COVID with symptom-overlapping conditions, including other infection-associated chronic illnesses, myalgic encephalomyelitis/chronic fatigue syndrome, fibromyalgia, dysautonomia, and mood disorders, would provide important complementary information about the score’s differential diagnostic performance. Far from invalidating their findings, they argue, such studies would extend them, mapping the score’s behavior across a wider landscape of conditions that mimic or overlap with Long COVID.

On the question of predictive values, the two sides find firmer common ground. Positive and negative predictive values depend heavily on disease prevalence: the same score can yield very different predictive values in a high-prevalence specialty clinic and a low-prevalence community sample. The Johns Hopkins authors agree entirely that these measures should not be generalized beyond the sampled population, and they clarify that the predictive values in their study were presented as descriptive characteristics of the cohort rather than as estimates applicable to broader clinical or community settings. What survives this clarification, they insist, is the study’s principal observation: the RECOVER PASC score demonstrated high specificity against recovered SARS-CoV-2 controls, meaning it rarely mislabeled recovered individuals as having Long COVID, but showed limited sensitivity in a clinically characterized Long COVID cohort, meaning it missed a substantial share of expert-diagnosed cases.

That combination of high specificity and limited sensitivity carries real clinical weight. A score that rarely produces false positives but frequently produces false negatives could, if used as a gatekeeping tool, steer genuinely ill patients away from evaluation and care. The Johns Hopkins team’s willingness to highlight the score’s sensitivity limitation, even while defending their methodology, underscores that their aim is refinement rather than advocacy. They frame the exchange with Goldman and Martin as a dialogue between complementary rather than competing questions: their study characterizes performance in the specialty referral settings where patients with persistent post-COVID symptoms are actually evaluated, while population-based studies, which they call essential, would characterize performance across the full spectrum of SARS-CoV-2 recovery.

The broader lesson may outlast the dispute itself. No single study, the authors conclude, can fully characterize the performance of an emerging research index across all clinical settings; confidence is built instead through complementary studies conducted in community populations, primary care settings, specialty referral clinics, and symptom-overlapping comparator populations. For a condition as heterogeneous and contested as Long COVID, that incremental, multi-setting approach may be the only scientifically defensible path toward standardized classification. The work was supported by the National Institutes of Health and the National Institute of Mental Health, and the authors report no conflicts of interest. As research indices like the RECOVER PASC score continue to evolve, this exchange stands as a reminder that in diagnostic science, what a test is for often matters as much as how well it performs.

Subject of Research: External clinical validation of the RECOVER PASC research index for Long COVID diagnosis

Article Title: Letter to Editor External Clinical Validation of the RECOVER Research Index: A Response to Goldman and Martin

Article References: Azola, A., Veenhuis, R. T., & Rubin, L. H. (2026). Letter to Editor External Clinical Validation of the RECOVER Research Index: A Response to Goldman and Martin. Journal of General Internal Medicine. https://doi.org/10.1007/s11606-026-10851-3

Image Credits: AI Generated

DOI: 10.1007/s11606-026-10851-3

Keywords: Long COVID, RECOVER PASC score, external validation, diagnostic index, SARS-CoV-2, sensitivity, specificity, NASEM consensus definition, specialty referral cohort, predictive values, clinical research, Johns Hopkins