UPDATE: EyePoint (EYPT) - LUGANO Missed. Nine Patients Broke the Thesis.

The Event: LUGANO topline landed this morning. The primary endpoint — non-inferiority in mean change in best-corrected visual acuity (BCVA, the eye-chart letters metric) at a blended Week 52/56 versus on-label aflibercept, with a -4.5 letter margin — was not achieved in the full dataset.

The Rescue: Management then presented an ad hoc analysis excluding a 9-of-211 “asymmetric cohort” who lost ≥15 letters from causes they attribute to non-wet-AMD etiologies (geographic atrophy n=6, glaucoma n=2, retinal detachment n=1). Strip those nine out and DURAVYU is non-inferior at nominal p=0.0096, with a -2.4 letter delta versus control.

The Reaction: The stock went from $12.25 to $4.87, a -60% haircut. In the original deep dive I wrote that if the Phase 3 trials missed the non-inferiority margin “the equity gets wiped out.” The market is pricing about 60% of that sentence.

• Primary endpoint: missed in the ITT-equivalent full dataset. No p-value was disclosed for the full analysis — only for the version with nine patients removed. The implication: the number that goes on an FDA statistical review is the one that failed. Everything else is argument.

• The ad hoc delta: -2.4 letters. DURAVYU +3.5 letters versus aflibercept +5.9 letters, excluding the asymmetric cohort. Compare that to DAVIO 2’s -0.3 / -0.4 letter deltas that I called “meaningful cushion” against a -4.5 margin. Even in the best-case analysis, Phase 3 slippage was roughly 2 letters worse than Phase 2. The cushion I described was real and it got mostly eaten.

• The control arm, not the treatment arm, is the statistical outlier. Only 0.5% (1 of 216) of LUGANO's aflibercept patients lost ≥15 letters, against 2.7%–5.9% in VIEW 1/2, HAWK, HARRIER, TENAYA, LUCERNE and PULSAR, and 7.7% in DAVIO 2's own control arm. Now run the same calculation on the treatment arm: DURAVYU's nine severe losers out of 211 is 4.3% — sitting squarely inside that historical band. Why it matters: this is management's strongest quantitative argument and it deserves to be stated precisely, because they didn't. DURAVYU did not produce an unusual number of severely worsening eyes. LUGANO's control arm produced an unusually small one, at roughly a tenth the historical rate, and that is a headwind nobody powered for.

• Treatment burden: 42% reduction, superiority achieved, nominal p<0.0001. 3.1 injections versus 5.3 after loading, roughly two fewer injections per patient through Week 56, against a structural ceiling of 60%. This is the commercial pitch, and it landed.

• Supplement-free rates: 76% through Week 32, 54% through Week 56; 79% got zero or one supplement. DAVIO 2’s comparable Week 32 figure was 65%. The durability improved from Phase 2, it didn’t decay.

• Anatomy: CST difference of just 4 microns versus aflibercept at Week 56. Central subfield thickness is the OCT-measured swelling of the retina — the direct physical readout of whether fluid is under control. A 4-micron gap on a ~320-micron baseline is essentially identical anatomic control. The drug is working on the plumbing.

• Safety: clean, again. No insert migration, no anterior chamber opacities, no free-floating particles, no retinal vasculitis, no severe intraocular inflammation. IOI 0.5% in both arms. Discontinuation 6% in each arm, none related to DURAVYU. The historical killers of sustained-release intravitreal inserts stayed dead.

• Calendar: LUCIA topline Q4 2026, NDA targeted 1H 2027, DME (COMO/CAPRI) Q4 2027. Unchanged. Management intends to file a pooled LUGANO+LUCIA package.

What actually happened here. DURAVYU controlled wet AMD. The 4-micron CST gap and the 54% supplement-free rate say so unambiguously — the retina stayed dry in the majority of eyes on two injections a year instead of six. What it did not do is deliver as many letters of vision as aflibercept in the full population. And the reason offered is that nine patients went blind-ish from things that aren’t wet AMD.

Is that argument credible? Partially, and this is where I have to be honest with you rather than comfortable. The mechanistic story hangs together: all nine had good anatomic control of their wet AMD, six received supplemental aflibercept and none of them recovered vision, and the attributed causes — geographic atrophy, glaucoma, retinal detachment — are not things a VEGF/PDGF/IL-6 inhibitor sitting in the vitreous plausibly causes. The AE tables back that up: cataract 4.3% vs 5.4%, IOP increase 3.8% vs 3.2%, IOI 0.5% vs 0.5%. There is no safety asymmetry driving this. Management also showed fundus autofluorescence data with no trend toward worsening geographic atrophy versus control.

How much does the healthy control arm actually explain? Management presented the comparison and stopped there, so I ran the counterfactual. Assume a patient who loses ≥15 letters finishes roughly 25 letters below a typical patient’s outcome — the actual magnitudes weren’t disclosed, so treat this as directional rather than precise.

Rerun LUGANO’s control arm at 2.7%, the bottom of the historical band, and you’d expect about six severe losers instead of one. That pulls the control-arm mean down by roughly half a letter. Run it at 5.9%, the top of the modern anti-VEGF range, and you’d expect about thirteen, dragging the control mean down closer to 1.4 letters.

Half a letter does not rescue a failed non-inferiority test. A letter and a half plausibly does. So management’s best argument lands somewhere between contributory and sufficient depending entirely on which historical comparator you pick — and they presented it without that range, which is the tell. The honest version is narrower than the way it was told this morning: the anomalous control arm is real and it genuinely contributed, but on its own it probably does not get you from a missed primary to a cleared one. Something else is doing part of the work, and the candidate is sitting in plain sight — even in the best-case ad hoc analysis, the delta was roughly two letters worse than DAVIO 2 predicted. Both things are true at once. Only one of them is somebody else’s fault.

But here’s the problem with post hoc exclusions. A 4% imbalance in catastrophic vision loss appearing in the treatment arm and zero times in the control arm is exactly the kind of finding that looks like noise when you want it to and looks like signal when a regulator wants it to. Six geographic atrophy cases is the one that would keep me up at night, and management’s rebuttal doesn’t fully answer it. They showed fundus autofluorescence data with no trend toward worsening geographic atrophy versus control. That is a real datapoint and I don’t want to wave it away — but it answers a question I didn’t ask. Population-level GA incidence tells you whether DURAVYU causes atrophy. It does not tell you whether DURAVYU produces worse functional outcomes in eyes that already have it, and those are different questions, because vision loss from atrophy depends almost entirely on whether the lesion reaches the fovea, not on whether atrophy exists somewhere in the macula.

All six of those patients had GA at baseline. So the analysis I want — and the one nobody has run publicly — is severe vision loss stratified by baseline GA status: of the patients who entered each arm with atrophy already present, how many went on to lose ≥15 letters? If that number is balanced between arms, the outlier explanation gets considerably more credible. If it isn’t, a reassuring population-level average is masking the subgroup that generated the entire result. Management has the data.

The companion question, which an analyst raised and management said they hadn’t yet analyzed, is whether additional patients lost 8–14 letters from these same non-wet-AMD etiologies, just below the ≥15 cutoff. A clean cliff at the threshold with nothing underneath it looks like bad luck. A graded excess below the line looks like something reproducible.

Until both are shown, “no trend at Week 56” means “we found no evidence at 56 weeks” — not “the drug is exonerated.” Those two get conflated constantly in press releases about diseases that play out over a decade.

The word “nominal” is doing enormous work. Every p-value in this release is nominal — p=0.0096 on the ad hoc primary, p<0.0001 on treatment burden, p=0.0035 on the supplement-free subgroup. Nominal means unadjusted for multiplicity, and in a trial where the primary endpoint failed, the entire hierarchical testing sequence collapses. Statistically, every secondary in this release is descriptive. In the original deep dive I flagged the P-Hacking Check on VERONA’s supplement-free subgroup and wrote that “leaning too heavily on subgroup data is a classic biotech distraction technique.” That warning aged into today’s press release headline.

What this means for LUCIA. LUCIA is the identical trial, n=475, reading out in Q4 2026. The asymmetric-cohort argument is testable in the cleanest possible way: if the nine-patient imbalance was random bad luck, LUCIA should hit its primary endpoint outright on the full dataset. If it was a real, low-frequency drug- or delivery-related phenomenon, it recurs and the program is finished. That’s a genuine coin flip and I’d put it slightly better than even odds given the anatomic data — but it is a coin flip, and it is now the entire company.

What this means for the FDA. Even a clean LUCIA leaves EyePoint filing a pooled package where one of two identical pivotal trials missed on the primary as pre-specified. That is not automatically fatal — agencies have approved on one-of-two before, particularly with a compelling burden-reduction story and a clean safety database — but it converts what was a routine non-inferiority filing into a live advisory-committee-shaped question. Add the unremediated July 2024 Watertown Warning Letter, which every filing including today’s still carries, and the regulatory path just got materially longer and less certain.

Share

A 60% drawdown on a trial where the drug demonstrated durable anatomic control, superiority on burden, and a clean safety profile looks excessive on the science and rational on the process. The market is not pricing “the drug doesn’t work.” It’s pricing three things: (1) the primary endpoint failed and headlines don’t have room for asymmetric cohorts, (2) every downstream statistic is now nominal and therefore rhetorically weak, and (3) at $180.5M cash against roughly $94M quarterly burn, a company that needs to fund an NDA fight now has to raise from weakness rather than from the $14.44 the ATM was printing at three weeks ago. The dilution math got ugly fast — at $4.87, raising $150M means issuing roughly a third of the current 86.2M share count.

That last point is the one I’d weight heaviest. In the August 5 note I wrote the balance sheet was “adequate, not abundant.” Adequate assumed a clean print.

• September 23–26, 2026: Retina Society 59th Annual Meeting — full LUGANO curves and subgroup analyses. The retinal community’s verdict on the asymmetric-cohort argument lands here, and it matters more than the press release did.

• Q4 2026: LUCIA topline — now the actual binary event.

• 1H 2027: Potential NDA submission with pooled LUGANO + LUCIA data.

• Q4 2027: COMO/CAPRI DME topline.

• Unresolved: Watertown Warning Letter remediation. Still no closeout letter. The factory is still the boss, and now the data package is compromised too.

• Watch for: any financing. At this price it would be painful, and the ATM is presumably unusable near-term.

Current Stance: BUY → SPECULATIVE HOLD INTO LUCIA.

• The primary endpoint missed, and I’m not going to dress that up. The BUY case I laid out was explicitly predicated on LUGANO and LUCIA clearing the -4.5 letter margin. LUGANO didn’t, as pre-specified. My own SELL scenario named this outcome. Downgrading is the honest response.

• But the asset is damaged, not dead — and that distinction is the whole post. A 4-micron CST delta, 54% supplement-free through Week 56, 42% burden reduction, and a safety database with zero vasculitis and zero migration across 211 patients is not the profile of a drug that failed biologically. It’s the profile of a drug that lost a statistical argument to nine patients and an unusually healthy control arm.

• LUCIA is now a clean, high-stakes test of management’s explanation. Identical design, n=475, Q4 2026. If the asymmetric cohort was noise, LUCIA hits on the full dataset and today’s price looks like a gift. If it recurs, the equity approaches the cash line. There is very little middle ground.

• The financing risk is now the near-term killer. $180.5M against ~$94M quarterly burn, with the stock at $4.87. Raising here is roughly 33% dilution for $150M. That arithmetic gets worse, not better, the longer they wait for LUCIA.

• Watertown remains unaddressed. A one-of-two pivotal package filed by a company with an open cGMP Warning Letter is a materially harder FDA conversation than the one I modeled in the original deep dive.

For anyone who sized this around the binary, the binary half-resolved in the worst possible way — bad enough to break the primary endpoint, good enough to keep the story alive. Positions that were sized for a wipeout scenario have now experienced most of one. What’s left is a genuine coin flip on LUCIA with a financing overhang stapled to it, and that is a materially different security than the one I wrote up. The September Retina Society presentation appears to be the next real information event — watching how retinal specialists react to the asymmetric-cohort argument may tell you more about LUCIA’s odds than anything in today’s deck.

This post is for informational and educational purposes only and does not constitute investment advice or a recommendation to buy or sell any securities. The author is a medicinal chemist, not a licensed financial advisor; the scientific analysis herein should not be interpreted as medical guidance. Biotech investing is inherently volatile — past scientific validation and early clinical data do not guarantee future late-stage clinical success or regulatory approval. Do your own due diligence.