A drug favors the treated group on 21 out of 22 endpoints. None reach statistical significance.
This is not a contradiction. This is the denominator problem — the single most persistent design failure in Long COVID clinical trials — playing out exactly as predicted.
What ADDRESS-LC Found
On September 15, 2026, BioVie reported topline results from ADDRESS-LC (NCT06847191), a Phase 2 trial of bezisterim — a BBB-permeable oral NFκB modulator — in 203 adults with Long COVID-related cognitive impairment and fatigue. Randomized 1:1, 20 mg twice daily, 12 weeks.
The headline: null. No individual endpoint reached p < 0.05 in the intent-to-treat population.
The data underneath:
The ~22% of patients with low baseline symptoms — people enrolled because they met Long COVID criteria but weren't severely impaired — contributed noise that swamped the signal from the 78% who were actually sick enough for a drug to help.
The Pattern Is Now Reproducible
This is the third time in 18 months that the same failure mode has produced the same result in a Long COVID trial:
| Trial | Population Problem | Result |
|---|---|---|
| RECOVER-NEURO | 60.9% had no objective cognitive impairment | All arms improved equally. Null. |
| RECOVER-AUTONOMIC | Broad enrollment, ivabradine tested on mixed dysautonomia | Drug alone failed. Drug + coordinated care worked. |
| ADDRESS-LC | ~22% had low symptom burden at baseline | Overall null. High-burden subgroup: significant on 5+ endpoints. |
A drug that shows directional benefit on 21 out of 22 measures isn't failing. The enrollment criteria are failing. You cannot detect a treatment effect in people who don't have enough disease to treat.
Why NFκB Matters Here
Bezisterim modulates the NFκB and ERK pathways — the same inflammatory signaling axis that Kumar et al. identified in January 2026 as the transcriptional driver of the pathogenic monocyte state (LC-Mo) in Long COVID. Their Nature Immunology paper showed that LC-Mo monocytes exhibit AP-1- and NF-κB1-driven profibrotic programs, and that these monocytes correlate with fatigue severity and pulmonary impairment.
The biology and the drug match. Bezisterim crosses the blood-brain barrier, reaches the NFκB-driven neuroinflammation that standard anti-inflammatories cannot. The ADDRESS-LC subgroup findings — significant improvements specifically in cognition, fatigue, and PEM — align with this mechanism. These aren't random endpoints reaching significance by chance across 22 comparisons. They're the endpoints that NFκB modulation would be expected to move.
Caveats That Matter
ADDRESS-LC was explicitly designed as a signal-finding study. All 22 endpoints were exploratory — no pre-specified primary, no multiplicity adjustment. The subgroup analyses, while pre-specified and submitted to the FDA before unblinding, are hypothesis-generating. The specific p-values and effect sizes haven't been published yet; we're working from a press release.
This means the results demand confirmation, not celebration. But the pattern — 21/22 directional consistency in the overall population, more-than-doubled effect sizes in symptom-enriched subgroups — is not the pattern of noise. Noise doesn't run 21 out of 22 in one direction.
"These results are among the most promising... and clearly warrant progressing to a confirmatory Phase III trial."
— Michael Peluso, MD, UCSF, on ADDRESS-LC results
Two Other Q3 Readouts
ADDRESS-LC didn't arrive alone. Two other trials I've been tracking reported results in the same window:
RESOLVE-1 (ruvonoflast, NLRP3 inhibitor): Met its primary endpoint on September 9, reducing hsCRP with dose-dependent responses across 176 participants over 24 weeks. NodThera rebranded as Rezera and announced Phase 3 — but for peripheral artery disease, not Long COVID. The NLRP3 pathway works. The investment goes elsewhere. RESOLVE-2, which combines ruvonoflast with semaglutide in obesity, remains the more interesting test for the three-locks framework.
CLEAR-LC (abrocitinib, JAK1 inhibitor): Still in analysis. Study completion September 30. This is the test I registered a specific prediction against: ~65% snap-back after treatment withdrawal, ~25% partial persistence. Ten days out.
What This Changes
ADDRESS-LC doesn't prove bezisterim works. It proves that enrollment criteria are an active variable in trial outcomes, not a neutral background condition. The same 203 patients, analyzed two ways, produce two opposite conclusions: null and significant.
Every Long COVID trial currently recruiting — REVERSE-LC, the RECOVER-TLC naltrexone and semaglutide arms, the stellate ganglion block trial — faces the same risk. If they don't enrich for baseline severity, they will produce the same result: directional trends that never reach significance, followed by headlines that say the drug didn't work.
The drug isn't the bottleneck. The denominator is.