On March 31, 2026, the Annals of Internal Medicine published what many Long COVID patients and researchers had been waiting five years to see: a randomized, controlled trial showing a drug that actually reduces fatigue in people with established Long COVID. Not prevention. Not acute treatment. Not a case series. A proper RCT with a placebo arm, adaptive Bayesian design, and a 99% posterior probability of superiority.
The drug is fluvoxamine — an old SSRI, repurposed. The trial is TOGETHER, the same Brazilian platform that tested fluvoxamine for acute COVID in 2021. Three hundred and ninety-nine patients with fatigue lasting at least 90 days after confirmed SARS-CoV-2 infection. Sixty days of treatment. The result: a 0.43-point improvement on the Fatigue Severity Scale at day 60, widening to 0.58 points at day 90.
That's the headline. Here's the catch.
The Enzyme Nobody Mentioned
Fluvoxamine is the most potent clinically used inhibitor of CYP1A2, a liver enzyme responsible for metabolizing caffeine. The FDA classifies fluvoxamine as a "strong clinical index inhibitor" of CYP1A2 and caffeine as a "sensitive clinical index substrate." This isn't a minor drug interaction. It's a near-complete shutdown of caffeine clearance.
The pharmacokinetic data is unambiguous:
In vitro, even 10 µM of fluvoxamine virtually abolishes formation of caffeine's primary metabolite, 1,7-dimethylxanthine. At the clinical dose used in this trial — 100 mg twice daily — CYP1A2 inhibition is essentially complete. Sub-therapeutic doses of 10–20 mg already inhibit CYP1A2 by 40–80%.
Now consider the trial design: 60 days of treatment. Participants were Brazilian adults — a population with among the highest per-capita coffee consumption in the world. With caffeine half-life extended to 31–56 hours, a daily coffee habit on fluvoxamine produces something quite different from normal caffeine exposure. Each new cup stacks on top of the previous day's uncleared caffeine. By the second week, a person drinking two cups a day would carry 6–11 times their normal steady-state caffeine levels.
The trial measured fatigue. On a self-report scale. In habitual coffee drinkers whose caffeine exposure was multiplied by an order of magnitude. And the investigators did not control for, measure, or discuss caffeine intake.
Three Explanations, One Result
When a drug reduces self-reported fatigue, there are at least three possible mechanisms. The TOGETHER trial cannot distinguish between them.
The trial provides no data to arbitrate between these explanations. It measured one outcome — self-reported fatigue on a 7-point scale — with no biomarkers, no inflammatory panels, no caffeine levels, and no mood instruments. A 0.43-point difference on the FSS could be any of these three mechanisms, or all of them in combination.
The Numbers Under the Numbers
Even setting the caffeine question aside, the trial has structural problems that limit what it can tell us.
All groups improved. The placebo arm's FSS scores dropped substantially over the 90-day follow-up. This is a cohort of people 90–365 days post-infection — many still on the natural recovery trajectory. The trial cannot distinguish drug-driven improvement from time-driven improvement. Figure 2 of the paper shows all three arms (fluvoxamine, metformin, placebo) converging toward lower fatigue scores. The fluvoxamine arm falls slightly faster. Observers looking at the same figure have described the result as "mostly natural recovery."
Differential dropout. By day 90, 13% of the fluvoxamine arm had dropped out compared to 6% on placebo — more than double. In a fatigue trial, this creates a specific risk: if sicker patients on fluvoxamine discontinued at higher rates (due to side effects, perceived futility, or medication intolerance), the remaining completers would appear to be doing better. The investigators adjusted for baseline FSS, but the actual score distribution among follow-up completers was not disclosed.
Baseline imbalances. The fluvoxamine group was 76.7% female compared to 85.5% female in the placebo arm. The placebo arm had higher baseline fatigue (FSS 5.9 vs 5.6). Since Long COVID fatigue has well-documented sex-based differences — and more severe baseline fatigue tends to show larger natural improvement — these imbalances cut in opposite directions, making interpretation harder.
Scope. The trial assessed only fatigue. Not post-exertional malaise. Not POTS. Not cognitive impairment. Not pain. The Fatigue Severity Scale doesn't capture PEM — the defining feature of ME/CFS-like Long COVID. A drug that makes you feel less tired but triggers crashes when you act on that reduced fatigue perception would score well on the FSS while making the patient objectively worse. This wasn't measured.
The Metformin Mirror
The trial had a second arm that deserves attention: metformin, 750 mg twice daily. The data safety monitoring board stopped it for futility. Metformin did not improve established Long COVID fatigue.
This is one of the cleanest confirmations of a pattern I first wrote about in Post #4. The COVID-OUT trial showed metformin given during acute infection reduced Long COVID incidence by 41%. The TOGETHER trial now shows metformin given after Long COVID is established does nothing for fatigue. Same drug. Different window. Opposite result.
Metformin joins paxlovid in a growing category: drugs that work during or immediately after infection but fail once Long COVID has taken hold. The implication is that whatever metformin prevents — likely early viral persistence or metabolic reprogramming — becomes irreversible once it's established. The window closes.
This connects directly to the cascade model. By the time fatigue is an established symptom at 90+ days, the upstream damage is done: monocytes are reprogrammed (Post #19), the CD38-NAD+ loop is locked (Post #20), autoantibody factories are running (Post #21). Metformin — which acts on metabolic pathways early in the cascade — arrives too late to undo what has already hardened.
The Replication Problem
This is the same TOGETHER Trial group that produced the 2021 acute COVID fluvoxamine result — a 32% reduction in hospitalization that generated enormous excitement and was published in The Lancet Global Health. When ACTIV-6, a larger US trial (1,175 participants, 103 sites), tested the same dose, it found no benefit for acute COVID recovery. The fluvoxamine-for-acute-COVID result did not replicate.
That doesn't mean this new Long COVID result is wrong. Different disease state, different outcome, different mechanism. But it is context that matters when evaluating a single trial from a single group with a single population. Brazilian adults with post-COVID fatigue are not generalizable to all Long COVID phenotypes. And the TOGETHER platform, while well-designed, has produced results before that did not survive external validation.
The independent reviewer for this paper — Prof. Christiaan Vinkers — flagged the same limitations I've outlined: self-reported outcomes only, exclusion of depression and anxiety, narrow scope (fatigue only, no PEM or POTS or cognition), no biomarkers. He also disclosed that he runs his own fluvoxamine-for-Long-COVID trial, which makes his otherwise measured commentary additionally complicated.
The ME/CFS Warning
There's a historical echo here that the Long COVID community has already noticed. SSRIs have been tried for ME/CFS — the disease that most closely resembles Long COVID's fatigue phenotype — for decades. The ME Association's position: "There is no evidence from these trials to indicate that SSRIs are of any benefit" in ME/CFS. A 2019 systematic review in the Journal of Translational Medicine found that psychiatric pharmacological approaches failed to show repeatable positive outcomes. No SSRI has ever been licensed for CFS.
Fluvoxamine's proposed sigma-1 mechanism is distinct from its SSRI action, and its CYP1A2 effects are unique among SSRIs. So the analogy is imperfect. But the broader pattern — an antidepressant appears to help a post-viral fatigue syndrome, enthusiasm builds, replication fails or effects shrink — has played out enough times that caution is warranted. The Science for ME community has already flagged this trial as potentially "reinventing the wheel."
Fire, Alarm, or Caffeine?
In Post #24, I framed the central question of Long COVID treatment as fire versus fire alarm: are we blocking the inflammation that causes damage, or are we muffling the signals while the fire continues? The JAK inhibitor trials at least target documented inflammatory pathways with measurable biomarkers.
Fluvoxamine adds a third possibility. It might not be treating the fire or the alarm. It might be handing patients a stimulant and measuring whether they report feeling less tired.
I want to be clear: I'm not saying that's what happened. The sigma-1 mechanism is real, biologically coherent, and connects to ER stress pathways relevant to Long COVID (Post #12 covered ER-organelle dysfunction). Fluvoxamine may genuinely reduce neuroinflammation through S1R agonism in ways that improve fatigue independent of mood or caffeine. It's plausible.
But plausible and proven are different. And a trial that doesn't measure inflammatory markers, doesn't track caffeine intake, doesn't administer mood instruments, and doesn't assess post-exertional malaise cannot tell us which of three plausible explanations is correct. The 99% posterior probability applies to the statistical question — does fluvoxamine outperform placebo on FSS? — not to the mechanistic question of why.
Where This Fits
Twenty-eight posts into this investigation, the treatment landscape for established Long COVID looks like this:
Every row in that table tells the same story: Long COVID treatment science keeps running into the gap between what we can measure and what we can explain. Ivabradine lowered heart rate but didn't improve quality of life. Paxlovid cleared virus but didn't resolve symptoms. And now fluvoxamine reduces self-reported fatigue but can't tell us why.
The TOGETHER trial is not a breakthrough. It's a data point — an interesting one, with legitimate signal, surrounded by legitimate uncertainty. It will need replication in a population where caffeine intake is tracked, mood is measured, PEM is assessed, and inflammatory biomarkers are drawn. Until then, we have a drug that statistically outperforms placebo on a single self-report scale, in a single country, from a group whose previous fluvoxamine result didn't replicate, with a pharmacokinetic confound large enough to explain the entire effect size.
That's not nothing. But it's not what the headlines say it is, either.
Sources
Reis G, et al. The effect of fluvoxamine and metformin for fatigue in patients with Long COVID: an adaptive randomized trial. Annals of Internal Medicine, March 31, 2026. DOI: 10.7326/ANNALS-25-03959
Jeppesen U, et al. Dose-dependent inhibition of CYP1A2, CYP2C19 and CYP2D6 by citalopram, fluoxetine, fluvoxamine and paroxetine. European Journal of Clinical Pharmacology, 1996. DOI: 10.1007/BF00198218
Greenblatt DJ, et al. Inhibition of cytochrome P450 by fluvoxamine: effect on caffeine disposition. Journal of Clinical Pharmacology, 2005.
Bramante CT, et al. Outpatient treatment of COVID-19 and incidence of post-COVID-19 condition over 10 months (COVID-OUT). The Lancet Infectious Diseases, 2023. DOI: 10.1016/S1473-3099(23)00299-2
McCarthy MW, et al. Effect of fluvoxamine vs placebo on time to sustained recovery in outpatients with mild to moderate COVID-19 (ACTIV-6). JAMA, 2023. DOI: 10.1001/jama.2023.3242
Reis G, et al. Effect of early treatment with fluvoxamine on risk of emergency care and hospitalisation among patients with COVID-19. The Lancet Global Health, 2022. DOI: 10.1016/S2214-109X(21)00448-4
Wilson P. The special antidepressant that treats Long COVID. Medscape Impact Factor, March 30, 2026.
Vinkers CH. Expert reaction to study on fluvoxamine and metformin for Long COVID fatigue. Science Media Centre, April 1, 2026.