Six major clinical trials for Long COVID have failed in the past two years. BC007, temelimab, RECOVER-NEURO, ensitrelvir, and two separate Paxlovid trials — STOP-PASC and PAX LC — all missed their primary endpoints. Zero approved treatments exist. Over 400 million people worldwide have been affected.
But these failures are not random. They share three structural problems that, once understood, point toward what must change. The next wave of trials is already beginning to learn from them.
The Graveyard
Let me lay out what has failed and why, because each trial tells a different part of the same story.
BC007 (rovunaptabin) targeted autoantibodies against G-protein-coupled receptors — a plausible mechanism given evidence of autoimmunity in Long COVID. Berlin Cures ran a Phase IIa trial and announced failure in November 2024. Some participants improved, but the study design could not capture those changes. A Lancet Infectious Diseases commentary noted the core lesson: in a heterogeneous disease, broad enrollment dilutes the signal from the subgroup that might actually respond.
Temelimab, an antibody against human endogenous retrovirus W envelope protein, was tested by GeNeuro in 203 patients with neuropsychiatric Long COVID symptoms. No significant improvement in fatigue or most secondary endpoints. The company laid off staff. Another mechanism that might work in a subset, tested in everyone.
RECOVER-NEURO tried non-pharmacological approaches — brain training and transcranial direct current stimulation — for cognitive Long COVID symptoms. Published in JAMA Neurology in November 2025, it found that no intervention outperformed the comparison groups. All arms showed modest improvement, suggesting regression to the mean or natural fluctuation rather than treatment effect.
Ensitrelvir, a protease inhibitor, did not significantly reduce Long COVID at three months, though a higher percentage of treated participants reported returning to pre-COVID health. A trend, but not a result.
Paxlovid (STOP-PASC), the Stanford trial, randomized Long COVID patients to 15 days of nirmatrelvir/ritonavir. No benefit. A subsequent systems immunology analysis of 152 participants revealed something striking: the hypothesized mechanisms were not widespread in the trial population. Only a small subset had detectable microclots. No major differences in autoantibody profiles or EBV reactivation between Long COVID participants and controls. The drug transiently shifted the proteome at day 15, but everything normalized by week 10.
Paxlovid (PAX LC), the Yale trial published in Lancet Infectious Diseases in April 2025, confirmed the Stanford findings. One hundred participants, fully decentralized design — innovative methodology, same null result. Lead investigators Harlan Krumholz and Akiko Iwasaki acknowledged the duration may have been insufficient and selection criteria may need refinement.
Three Structural Problems
These six failures are not six separate stories. They are symptoms of three interconnected problems in how we design Long COVID trials.
1. Treating One Disease When There Are Many
Long COVID is not one disease. The NIH RECOVER cohort identified at least eight distinct symptom trajectories in 3,500+ patients. A systematic review in Lancet eClinicalMedicine (January 2026) synthesized 64 studies and identified four categorization approaches — symptom co-occurrence, organ system, severity, and clinical indicators. A Johns Hopkins analysis using machine learning on electronic health records found six distinct patient clusters, each with different phenotypic profiles.
Yet every failed trial enrolled broadly. BC007 enrolled anyone with Long COVID and autoantibodies, not specifically those whose symptoms were autoantibody-driven. Paxlovid trials enrolled anyone with persistent symptoms, not specifically those with evidence of viral persistence. Temelimab enrolled anyone with neuropsychiatric symptoms, not those with endogenous retrovirus reactivation.
The math is unforgiving. If a drug works in 20% of Long COVID patients — the subtype with the matching mechanism — a trial enrolling the full population needs enormous sample sizes to detect that signal. Most of these trials had 100–200 participants. The responders, if they existed, were drowned out.
2. Measuring With Broken Rulers
As of early 2026, no patient-reported outcome measure has been validated specifically for Long COVID. Every trial has borrowed instruments from other conditions — chronic fatigue scales, quality-of-life questionnaires, symptom checklists designed for different diseases.
A systematic review using COSMIN criteria found that only one Long COVID PROM (the LCSS) met recommended standards. Most were classified as needing further validation. None met FDA criteria for use in clinical studies. The development studies for all instruments were rated as “inadequate” due to missing patient involvement or unrepresentative populations.
This matters more than it might seem. Long COVID symptoms fluctuate — the STOP-PASC trajectory analysis found distinct patient-reported outcome trajectories within the trial itself. A measurement tool that captures a snapshot on day 15 and week 10 may miss the waxing and waning that defines the lived experience. A tool designed for fibromyalgia or chronic fatigue syndrome may not capture post-exertional malaise, the signature symptom that distinguishes post-viral conditions.
A January 2026 paper in eBioMedicine made five recommendations for trial design, with post-exertional malaise assessment as a central concern. The field is now developing Long COVID-specific instruments — including an 18-item PRO instrument with patient-involved content validation — but these were not available for the trials that have already failed.
3. Wrong Duration, Wrong Timing
The Paxlovid trials gave 15 days of antiviral treatment. If the hypothesis is viral persistence in tissue reservoirs — gut, lymph nodes, bone marrow — 15 days may be far too short. Proal et al. argued in Lancet Infectious Diseases (2025) that clearing persistent reservoirs may require 15–30 days or more, potentially in combination with other agents. A Nature case series of extended Paxlovid courses (7.5–30 days) showed variable results, suggesting duration matters but is not sufficient alone.
The STOP-PASC immunology data reinforces this: the drug transiently activated antiviral immune pathways at day 15, but these changes reversed by week 10. Whatever the drug was doing, it was not enough to produce a lasting shift.
What the Next Wave Is Doing Differently
The failure pattern is being recognized, and the next generation of trials is adapting — though not all equally well.
Mechanism-first design. The REVERSE-LC trial testing baricitinib (a JAK inhibitor) targets a specific mechanism: JAK-STAT-driven inflammation. Now expanded to 17 sites through RECOVER-TLC funding, it is enrolling 550 patients with neurocognitive data expected by November 2026. LC-REVITALIZE is combining upadacitinib (JAK inhibitor) with pirfenidone (anti-fibrotic) — a combination approach targeting two mechanisms simultaneously.
Powered for subgroups. The Scripps LoCITT trial testing tirzepatide (a GLP-1/GIP dual agonist) enrolled 1,000 participants — the largest Long COVID trial to date. It was intentionally designed to detect improvement in subgroups, acknowledging that the drug may help some phenotypes and not others. Biomarker samples are being collected from a subset for exploratory analysis.
Biomarker stratification. RECOVER-TLC is building biomarker measurement into its trial architecture. Blood-based biomarkers at baseline and 4 months. The RECOVER-VITAL trial is specifically testing whether viral antigen levels predict Paxlovid response — exactly the kind of stratification that the PAX LC and STOP-PASC trials lacked.
Novel mechanisms. Beyond JAK inhibitors and antivirals, the pipeline now includes semaglutide (GLP-1 agonist targeting immune dysregulation), low-dose naltrexone for children and young adults (targeting microglial activation), and stellate ganglion block (targeting autonomic dysfunction). Each targets a specific proposed mechanism rather than Long COVID as a whole.
What Must Still Change
The next wave is better, but the field has not yet fully internalized the lessons. Several gaps remain.
Validated outcomes are still missing. New Long COVID-specific PROMs are in development but not yet ready for pivotal trials. Until they are, trials remain vulnerable to false negatives from insensitive measurement.
Phenotyping before enrollment is rare. Most trials still define eligibility by symptom duration, not by mechanism. A trial of an antiviral should enroll patients with evidence of viral persistence. A trial of an anti-inflammatory should enroll patients with documented inflammation. The six-cluster model from Johns Hopkins, the eight-trajectory model from RECOVER, and the four-subtype framework from the Lancet eClinicalMedicine review all provide starting points — but few trials use them.
Combination approaches are underexplored. Long COVID likely involves multiple simultaneous mechanisms — viral persistence driving inflammation driving mitochondrial dysfunction driving immune exhaustion. Single-agent trials may fail even if the drug addresses one real component, because the other loops keep running. The LC-REVITALIZE trial combining a JAK inhibitor with an anti-fibrotic is a step in the right direction, but we need more.
$170 million has not been enough. RECOVER spent approximately $170 million on 13 treatments across its first round of trials. The conservative approach and limited mechanistic ambition drew criticism from patients and researchers alike. RECOVER-TLC represents a course correction, but the scale of investment still does not match the scale of the problem — 400 million affected people, zero approved treatments.
The Pattern That Gives Me Hope
Here is what I see when I look at the trail of failed trials: a field that is learning, painfully and slowly, that Long COVID cannot be brute-forced. You cannot take a drug that works for acute COVID and assume it works for the chronic aftermath. You cannot enroll everyone and expect to find anything. You cannot measure with borrowed tools and expect precision.
The STOP-PASC systems immunology analysis is perhaps the most valuable “failure” of all. It did not just show that Paxlovid does not work for Long COVID in general — it showed why. The 60-protein Long COVID Signature it identified across nine independent cohorts gives us a conserved inflammatory core to target. The absence of widespread autoantibodies and microclots in the trial population tells us these mechanisms exist in subsets, not universally. The transient proteome shift from treatment tells us 15 days is not enough to overcome the body’s return to its dysregulated equilibrium.
Every failure narrows the search space. The next trial that matches mechanism to patient, measures with validated tools, and runs long enough to overcome the body’s pathological equilibrium — that is the trial that will succeed. The question is not whether effective treatments exist. The question is whether we can design trials sophisticated enough to find them.
Sources
- Lancet Infectious Diseases — “Negative results in long COVID clinical trials: choosing outcome measures for a heterogeneous disease” (2025)
- STOP-PASC systems immunology analysis — Maestri, Kwon et al., medRxiv (Dec 2025)
- PAX LC trial — Krumholz, Iwasaki et al., Lancet Infectious Diseases (April 2025)
- BC007 Phase IIa failure — Berlin Cures (Nov 2024); Lancet ID commentary on trial design
- Temelimab Phase 2 — GeNeuro (mid-2024)
- RECOVER-NEURO — JAMA Neurology (Nov 2025)
- Lancet eClinicalMedicine — systematic review of 64 Long COVID subtyping studies (Jan 2026)
- eBioMedicine — outcome measure recommendations with PEM focus (Jan 2026)
- Proal et al. — Lancet ID (2025) on targeting SARS-CoV-2 reservoirs
- RECOVER-TLC trial designs — FNIH/NIH (2025-2026)
- LoCITT-T (tirzepatide) — Scripps Research / Schmidt Initiative (2025)
- RECOVER-Adult trajectories — Nature Communications (2025)