Alixorexton, Alkermes’ once-daily orexin-2 receptor (OX2R) agonist, appears to work. The stock assumes it works for ~31,000 patients; we can support ~12,300 (§1). Alkermes (ALKS) closed October at $45.34, a ~$7.8B equity value, and our company DCF and peer multiples put the price’s requirement at ~31,000–36,000 patients at peak.

Narcolepsy type 1 (NT1) cannot plausibly close that gap: even at a generous 40% share of the NT1 patients we expect the class to treat, the stock would still need narcolepsy type 2 (NT2) to deliver several times the patients we model (§2). That leaves the price resting on one positive NT2 Ph2, an 18 mg Epworth Sleepiness Scale (ESS) improvement of 3.6 points that we cannot statistically separate from the Takeda NT2 trial that failed. If the NT2 Ph3 succeeds, as we think more likely than not, we expect it to confirm a ~3-point drug: a real effect, but well short of the ~5 points we would want before underwriting broad switching from cheap generic wake promoters.

VIBRANCE-3, the idiopathic hypersomnia (IH) readout due between late December and early 2027, is worth ~$1 a share and tests none of this (§4, §5). We stay Neutral and would not buy ahead of it. That is a judgment about what the price requires, not a short call or a forecast of the catalyst-day move.

Alkermes spent most of the past five years as a CNS cash-flow business: long-acting injectables in addiction and schizophrenia (Vivitrol, Aristada), a slow-building Lybalvi launch, and a stock that traded between ~$26 and ~$35 from Oct 2024 to Apr 2026, a range consistent with a market paying for cash flow rather than a pipeline. Two developments changed that, though neither lifted the shares when it landed. The Avadel acquisition, announced in Oct 2025, brought LUMRYZ and a sleep-specialist sales force. Alixorexton produced Ph2 data in NT1 (VIBRANCE-1: all three doses met the maintenance of wakefulness test, or MWT, primary) and NT2 (VIBRANCE-2: MWT significant at 14 and 18 mg, and ESS at 18 mg only, after multiplicity).

The VIBRANCE-2 topline came on Nov 12, 2025, and ALKS then spent six months between $28 and $35. The re-rating came later: a 55% rise from Apr to Jun 2026, to $52, while short interest fell (A1). We do not attribute that move to any one event; the price it left makes us uncomfortable. At $45.34, the October close, the market has paid for a narcolepsy franchise that needs ~2.5x the patients our forecast supports, and, on any plausible NT1 share, most of the gap sits in NT2.

Not on the evidence so far. NT1 cannot plausibly close the gap, the NT2 data point to ~3 points against our ~5-point commercial bar, and VIBRANCE-3 tests neither. Orzeyful’s first US quarters and the Brilliance NT2 readout are what would change our view.

What we said in June. Our June note called oveporexton a validation of orexin replacement in NT1, not yet of orexin as a platform, and rated ALKS Neutral at a ~$7.5B market value. The FDA approved Orzeyful on Aug 5, ahead of the Sep 30 PDUFA date we expected, and the DEA scheduling step we flagged is still holding back US dispensing. Our doubt about NT2 has hardened: the FDA review says Takeda’s NT2 patients were withdrawn from the extension study because their trial showed no clinical benefit. The ADHD Ph1b we listed for 4Q26 came early, on Sep 21, and without a placebo comparison. We got two things wrong. The 11,075 Xywav patients we labeled NT1 were all narcolepsy patients. And the pollakiuria and insomnia rates we cited (13–33% and 9–21%) were Ph2 numbers; Orzeyful’s label reports urinary frequency of 53–58% and insomnia of 55–60% at the approved doses in pooled Ph3 data. The stock has added ~$0.3B since, to ~$7.8B. We remain Neutral.

1. At $45, the stock needs ~31,000 alixorexton patients; we can support ~12,300

Orexin-2 agonists have produced compelling results in narcolepsy. In NT1, alixorexton lengthened the time patients stayed awake on the MWT from about 3 minutes at baseline to 24–28 minutes at week 6, at all three doses [S33]. In NT2, the 18 mg dose beat placebo on the ESS by 3.6 points, significant after multiplicity adjustment at that dose only [S3, S34]. Takeda’s oveporexton, a rival orexin agonist, won FDA approval in NT1 on Aug 5, 2026 (§3). VIBRANCE-3 asks whether the same mechanism works in IH, a different disease in which patients sleep long hours and wake unrefreshed (§4).

What we can support. We forecast ~12,300 alixorexton patients at peak (2037) across NT1, NT2 and IH, within a range of ~5,700 to ~23,100. Our earlier top-down forecast, labeled “original model” in the exhibits, had 28,620. The forecast applies our estimates of treatment rates and alixorexton’s share to company estimates of ~100K diagnosed narcolepsy and ~40K diagnosed IH patients [S2], at ~$100K of net revenue per patient a year (§5). About 69% of the forecast comes from wake-promoting drugs, 19% from oxybates and 12% from people not yet treated.

For scale, Xywav and LUMRYZ together treat 21,025 patients today: ~17,125 on Xywav, ~5,850 of them in IH, and ~3,900 on LUMRYZ [S26, S27]. Undisclosed Xyrem and authorized-generic patients make the oxybate class larger, and oxybate use is not a ceiling for an oral drug.

Exhibit 1

Rebuilt peak patients are ~12,300, 43% of the original 28,620

The original NT1 figure implies the OX2R class treating 136% of treated NT1 patients. Most rebuilt patients come from those now on wake promoters, not oxybate switchers.

IndicationDiagnosedClass patients (share of diagnosed)Alixorexton share of classAlixorexton peak patients, base (low–high)Sourced from oxybate / wake promoter / untreatedOriginal model (implied class reach)
NT155,00021,150 (38%)22%4,653
2,122–8,445
1,078 / 2,970 / 60513,200
136% of treated
NT245,00013,700 (30%)30%4,110
2,025–7,900
450 / 3,255 / 4057,920
73% of treated
IH40,00010,050 (25%)35%3,518
1,512–6,795
788 / 2,205 / 5257,500
71% of treated
Total140,00012,280
5,660–23,140
19% / 69% / 12%28,620

US peak (2037) on-therapy patients. Diagnosed pools are company estimates (ALKS JPM 2026) [S2]; treated shares (80% narcolepsy, 75% IH), oxybate users at peak, class on-therapy shares and alixorexton shares of class are Clinaptis assumptions. Implied class reach = original-model patients ÷ base-case alixorexton share ÷ treated patients.

Source: Clinaptis model [S18]. Modeled; pools reported as company estimates.


Lapsed patients are mostly counted already. Our June note sized ~55–60K narcolepsy patients who started a premium therapy and later dropped back to generics. That pool is the most plausible route to a much larger franchise, but it is not an untreated pool: these patients largely sit on generic wake promoters, which is where 69% of our forecast already comes from. Winning them back is a contest for share among wake-promoter users, which our share assumptions already cover (Exhibit 1).

What the stock needs. At ~$45 (a ~$7.8B equity value), our company DCF needs ~31,000 alixorexton patients at peak, at $100K net revenue per patient (27.7K at $112K, 36.5K at $85K) [S23]. That is 2.5x our forecast, 1.5x the 21,025 reported oxybate patients, and above our high case of ~23,100. A7 tests the base-business cash flows and valuation conventions behind that figure; across those sensitivities, the price still needs more patients than our high case. We could be wrong in three ways: penetration above that high case, a base business materially stronger than A7 tests, or options we cannot size, such as the ADHD and fatigue programs or a sale of the company. The peer valuation below is an independent check.

The pace the price needs. A 2030 launch, as the company DCF assumes, needs 1,033 net additions every quarter to reach ~31,000 average patients in 2037; our ~12,300 needs 409. The best branded sleep drug added ~525 in 2Q26 (Xywav), with Wakix at ~450 and LUMRYZ at ~300, and Xywav’s since-launch pace is flattered by conversions of Jazz’s own Xyrem patients [S38]. Our base case sits within that range; the stock needs nearly twice the best of it for seven and a half years (Exhibit 2). A new mechanism could outrun these launches, and Orzeyful’s first US quarters will be the first test.

Exhibit 2

A $45 stock needs ~31K patients, at twice today’s best sleep-launch pace

Peak US alixorexton patients (2037) that make the company DCF equal the ~$45 close at $100K net, against our rebuilt cases; and the quarterly net additions each implies, against branded sleep launches.

Peak US patients, thousands

0K10K20K30K40KRebuilt low5.7KRebuilt base12.3KRebuilt high23.1KOriginal model28.6KImplied by ~$45 stock31.0K(27.7–36.5K)Xywav + LUMRYZ patients today: 21.0K

Net patient additions per quarter

03006009001,200What the ~$45 price needs1,033Our base case (~12.3K)4092.0×the pace of Xywav, the best launch today1,033 needed a quarter vs 525 added in 2Q262Q26 QUARTERXywav, 2Q26525Wakix, 2Q26450LUMRYZ, 2Q26300SINCE LAUNCH, PER QUARTERXywav, since Nov 2020 launch745Wakix, since Nov 2019 launch331LUMRYZ, since Jun 2023 launch300

ImpliedThe rebuilt-base mix scaled uniformly (NT1/NT2/IH, 65% narcolepsy and 36% conditional IH PoS) until the DCF reads $45.3 with opex held fixed. Whisker: $112K–$85K net.

SensitivityAt $100K net, ±10% on ex-alixorexton FCF moves the implied figure to 28.1–33.9K; across all price and base cases, 25.1–39.9K.

TodayXywav 11,275 narcolepsy and 5,850 IH patients [S26] plus LUMRYZ ~3,900 [S27]. Xyrem and authorized-generic patients are not disclosed, so the oxybate class is larger than shown.

Net additionsA straight-line ramp from a 2030 launch, the company DCF's first alixorexton revenue year, to the 2037 peak (30 quarters). 2Q26: Xywav ~525 net patients (Jazz), Wakix +450 average patients (Harmony), LUMRYZ +300 (3,600 to 3,900).

Since launchPatients at 2Q26 divided by quarters on market; Xywav's figure includes conversion of Jazz's own Xyrem patients.

SourceClinaptis model [S18] and company DCF [S23]; company reports [S26, S27, S38]. Modeled; reported inputs as labeled.


Peer multiples leave ~$19 a share for orexin. A DCF can be argued with, so we also valued ALKS without its orexin franchise against six specialty pharma peers: NBIX, HRMY, the pending Supernus–Indivior combination, JAZZ, ACAD and INCY [S39, S40]. Four methods (a base-business DCF, a DCF to 2029 with an exit multiple, EV to forward sales and forward P/E), weighted 25/20/25/30, give $26.18 a share, within $20.31–30.55 (Exhibit 3). That base business is Vivitrol, Aristada, Lybalvi, LUMRYZ and manufacturing and royalty income, and its two psychiatry growth products, Lybalvi and Aristada, are stalling (A7). We credit the base business with every dollar of alixorexton cost and add the 2027 program R&D back to earnings; both choices favor the bull case. At $45.34 that leaves $19.16 a share, ~$3.3B, for the orexin franchise. On our rNPV economics that residual needs ~36,000 peak patients (~29–45K across the range), close to the ~31,000 from the DCF reverse-solve: the two routes broadly agree. As a whole company ALKS trades at 4.6x forward EV/sales and 29x forward earnings, against peer medians of 2.8x and 12.7x. Harmony, the closest sleep comparable and itself developing an orexin agonist, trades at 1.5x.

Exhibit 3

Peers value ALKS without orexin at ~$26; the price leaves ~$19 for orexin

Value per diluted share of ALKS excluding the orexin franchise, low to high by method; the tick is the central case. Weighted central value $26.18 against a $45.34 price.

$10$15$20$25$30$35$40$45CentralWeightBase-business DCFOpex scales with revenue; 9–12% rate$19$25$21.8225%DCF to 2029 + exit multiple6x / 8x / 10x 2029 base EBITDA$23$36$29.3720%EV / forward sales25th / median / 75th pct, six peers$16$31$25.3925%Forward P/EEPS ex 2027 alixorexton R&D$23$31$28.3230%Hybrid, ALKS excluding orexinWeighted across the four methods$20.31$30.55$26.18100%Memo: growth-adjusted EV / sales$22Memo: DCF, company opex as forecast$14not weightedHybrid $26.18Price $45.34$19.16 a share (~$3.3B) left for orexinRange $14.79–$25.03 across the hybrid range

Method. Base business: alixorexton revenue removed and all program costs credited back (no R&D or stock-compensation add-backs); both credits favor the bull case. DCF: base opex held at the 2029 base EBIT margin (29.6%), 12% / 10.5% / 9% discount rates, −2% terminal growth. Exit: 6x / 8x / 10x 2029 base EBITDA ($759M). Multiples applied to forward consensus revenue ($1.87B) and EPS ($1.56 plus $0.66 of after-tax 2027 alixorexton R&D). Weights are editorial.

Peers, EV/sales / P/E. NBIX 3.4x/12.5x; HRMY 1.5x/9.5x; SUPN+INDV pro forma 2.3x/8.9x; JAZZ 3.3x/12.9x; ACAD 1.9x/25.8x; INCY 3.4x/14.2x.

Source. Clinaptis Analytical Model; S&P Global consensus and market data via stockanalysis.com, Oct 19–Nov 2, 2026 [S39, S40]. Price is the Oct 30, 2026 close. Modeled; not observed data.


The gap is a narcolepsy question. On any plausible NT1 share, it is mostly an NT2 one. It rests on our treated-share and class-share assumptions, which is why we show a range. Nothing in it depends on IH: our IH forecast is ~3,500 patients, and VIBRANCE-3 tests whether the mechanism transfers, not how many patients follow. §2 asks whether the NT2 evidence can carry the gap, §3 sets out the competition and price around it, and §4 and §5 size what VIBRANCE-3 can add.

2. NT2 must close most of the gap, and the NT2 data point to ~3 points

NT1 could close the gap only by taking almost the whole class. We expect the OX2R class to treat ~21,150 NT1 patients at peak and give alixorexton 22% of them (Exhibit 1). Narcolepsy patients carry higher approval odds and launch earlier than IH patients, so an NT1-heavy mix needs fewer patients than the ~31,000 base-mix figure: with NT2 and IH at our forecast, NT1 alone closes the gap at ~19,600 patients, 93% of the class. At a generous 40% NT1 share, NT2 would need ~15,300 patients: 3.7x our forecast, 111% of the NT2 class we model and 34% of the ~45,000 diagnosed NT2 patients. Scaling every indication evenly instead puts ~10,400 in NT2, 2.5x our forecast. Either way, NT2 must deliver well beyond our forecast. A 2026 claims study puts diagnosed NT2 prevalence at roughly three times our pool (51.1 against 15.5 per 100,000 for NT1, unadjusted) [S44]. A larger pool would lower that last ratio and lift our own forecast with it, but it would not change the binding constraints: how many patients leave generics, and on what effect size.

Two NT2 trials point to the same ~3-point effect. VIBRANCE-2 met its ESS endpoint at 18 mg with a 3.6-point placebo-adjusted improvement (95% CI 0.5 to 6.7) [S3]. Takeda’s TAK-861-2002 missed with 3.06 points at its higher oveporexton regimen (−0.06 to 6.18) [S6], and the FDA review records that its NT2 patients were withdrawn from the extension study because the trial showed no clinical benefit [S30]. The difference between the two is 0.54 points (95% CI −3.86 to +4.94; p=0.81). Placebo arms moved alike (−3.7 and −3.39 points), so the difference is not a placebo artifact. Pooled, the two trials estimate an NT2 class effect of 3.33 points (1.13 to 5.53). The results that decided which program advanced landed either side of p=0.05; they are not evidence that alixorexton does something oveporexton cannot (Exhibit 4).

MWT is where a real difference could sit. Alixorexton’s MWT gains were 6.7–9.3 minutes against 2.4 for oveporexton’s higher regimen [S41], a 4.3-minute lead at 18 mg (95% CI −4.2 to +12.8). That fits the hypothesis in our June note that oveporexton was dose-capped in NT2, and Takeda has since moved NT2 to a different molecule, TAK-360. But MWT is a laboratory measure, while ESS is what patients report and what payers will read, and on ESS the two programs are indistinguishable. Brilliance NT2 uses MWT as its primary endpoint and ESS as a key secondary [S43], so alixorexton can win on the measure where it may be better while the commercial measure stays near 3 points.

A Ph3 win would still confirm a ~3-point drug. Brilliance NT2 enrolls ~180 patients across three regimens and placebo for 12 weeks, ~45 per arm. Taking alixorexton’s 18 mg result at face value, the MWT primary has a 75% chance of success; treating Takeda’s trial as evidence about the same mechanism, 57%. The ESS secondary has a ~75% chance either way, because both priors sit near 3.3–3.6 points and the trial detects 3.2 points with 80% power.

The number that matters for patients is different. The probability that the true NT2 ESS effect reaches 5 points, the modafinil-like bar we use for commercial relevance (§4), is 19% on alixorexton’s data alone and 7% with Takeda’s added. The alixorexton-only prior rests on the best of three arms, so it flatters the drug.

What NT2 would need. A ~3-point drug can still sell. Alixorexton carries no sodium and needs no middle-of-the-night dose, advantages that matter most against oxybates, which supply only ~11% of our NT2 forecast. Most NT2 patients we expect it to win are on modafinil or stimulants, where a branded orexin must beat cheap generics after step edits. Reaching ~15,000 NT2 patients would take a drug that payers accept as clearly better than generics, and the data so far describe one that is not clearly better. We keep alixorexton at 30% of the NT2 class (Exhibit 1) and see NT2 uptake above our high case as the likeliest way we could be wrong.

Exhibit 4

Alkermes’ NT2 win and Takeda’s NT2 miss point to the same ~3-point effect

Placebo-adjusted week-8 effects in NT2 with 95% CIs. On ESS the two programs differ by 0.5 points (95% CI −3.9 to +4.9); on MWT alixorexton leads by 4.3 minutes (−4.2 to +12.8). A Ph3 win would still confirm a drug unlikely to reach 5 points.

AlixorextonOveporextonPooled and differenceBars are 95% confidence intervals.

Swipe sideways to see the MWT panel.

−8−6−4−20+2+4−50+5+10+15Placebo-adjusted ESS, pointsPlacebo-adjusted MWT, minutes← Favors drugFavors drug →5-point ESS differentiation thresholdAlixorexton 10 mgVIBRANCE-2−2.1+9.3Alixorexton 14 mg−3.2+6.7Alixorexton 18 mg−3.6+6.7Oveporexton lowTAK-861-2002−0.3Oveporexton highTAK-861-2002−3.1+2.4not reportedPooled, both programsESS −3.3 (−5.5, −1.1)MWT +4.7 (+0.5, +9.0)Difference:18 mg minus highESS −0.5 (−4.9, +3.9)MWT +4.3 (−4.2, +12.8)

NT2 efficacy estimates are statistically compatible across programs; a Phase 3 success would not, by itself, establish a commercially differentiated ESS effect.

The two programs ran separate trials; this is not a randomized head-to-head comparison. Pooled and difference rows use independent normal approximations from published 95% CIs. The 5-point ESS line is a Clinaptis-defined commercial hurdle, not an FDA efficacy requirement or an established clinically important difference. Source: VIBRANCE-2 [S3]; TAK-861-2002 [S6, S41]; Clinaptis calculations.

View underlying data & statistical assumptions
NT2, week 8ESS, points (95% CI)MWT, minutes (95% CI)
Alixorexton 10 mg (VIBRANCE-2)−2.1 (−5.3, +1.1)+9.3 (+3.3, +15.3)
Alixorexton 14 mg−3.2 (−6.3, −0.1)+6.7 (+0.8, +12.6)
Alixorexton 18 mg−3.6 (−6.7, −0.5)+6.7 (+0.9, +12.5)
Oveporexton low (TAK-861-2002)−0.3 (−3.6, +2.9)—
Oveporexton high (TAK-861-2002)−3.1 (−6.2, +0.1)+2.4 (−3.9, +8.7)
Pooled, both programs−3.3 (−5.5, −1.1)+4.7 (+0.5, +9.0)
Difference: 18 mg minus high−0.5 (−4.9, +3.9)+4.3 (−4.2, +12.8)

Brilliance NT2 Ph3 (~45 per arm, 12 weeks, MWT primary): probability of success and of a commercial-grade effect

Evidence used as the priorP(MWT primary succeeds)P(ESS secondary succeeds)P(true ESS effect ≥5)
Prior: alixorexton 18 mg as observed75%75%19%
Prior: pooled with TAK-861-200257%74%7%
Smallest effect detected with 80% power6.0 min3.2 points—

Assurance is the probability of p<0.05 at the top regimen averaged over the prior; SDs from VIBRANCE-2 (MWT 10.2 min) and 5.5 (ESS). With Bonferroni across three regimens MWT assurance is 67% and 45%. The first prior is the best of three VIBRANCE-2 arms and is biased upward. P(true ESS ≥5) uses flat-prior posteriors. Oveporexton low-regimen MWT was not reported in the sources we used.

Source: VIBRANCE-2 [S3]; TAK-861-2002 [S6, S41]; Brilliance NT2 design [S43]; Clinaptis calculations (closed form, no simulation).


3. Takeda’s Orzeyful sets the pricing and tolerability reference for alixorexton

Orzeyful sets the regulatory and pricing precedent. The FDA approved oveporexton on Aug 5, 2026 for NT1 only, dosed twice daily with the second dose three hours after the first. Takeda expects a US launch in early Nov once DEA scheduling is complete and is expected to disclose pricing then [S19, S24]. The approval de-risks the class in NT1. Takeda’s launch price becomes the practical reference, since ALKS has said its own pricing will be influenced by it, and the DEA schedule will show how far OX2R agonists can be positioned as easier to prescribe than oxybates. It says nothing about efficacy or uptake in IH or NT2.

Once-daily dosing may be an edge. Alixorexton is taken once a day; Orzeyful needs two morning doses. ALKS is not taking the edge for granted: it is testing a split-dose regimen (N=30). Management frames evening coverage as a class-wide hypothesis that began with NT1 clinician requests and a planned post-approval top-up, and expects once daily to remain the modal regimen; VIBRANCE-3 is the first clinical test.

Insomnia is where alixorexton may differ. In NT1, alixorexton’s ESS improvements of 6.4–8.7 points at week 6 [S42] sit close to Orzeyful’s 8.0–9.75 at week 12 [S30]. The adverse-event profiles split. Insomnia affected 28% of alixorexton-treated NT1 patients against 55% and 60% on Orzeyful’s two approved doses, with 1% on placebo [S37]; Orzeyful’s label tells prescribers to warn patients before starting and to consider a lower dose if insomnia persists beyond a week. If a gap of that size survives Ph3, it is a real reason to prefer alixorexton, and our 22% NT1 share could prove conservative (Exhibit 5). Three cautions keep it out of our base case: the comparison crosses a 6-week Ph2 and 12-week Ph3s, the label groups several insomnia terms, and alixorexton’s per-dose rates are not published. A larger NT1 share still cannot close the gap without NT2 (§2).

Urinary effects are class-wide. Urinary frequency affected 53–58% of Orzeyful patients (5% on placebo), and pollakiuria 55% of alixorexton NT1 patients and 46–73% of NT2 patients [S3]. The label attributes these symptoms to OX2R agonism on central micturition pathways and asks prescribers to screen for lower urinary tract symptoms. Takeda has set prescribers’ expectations for the class; urinary effects will not differentiate either drug.

Exhibit 5

Alixorexton reports half Orzeyful’s insomnia rate, but the same urinary frequency

Insomnia is 28% on alixorexton against 55–60% on Orzeyful at broadly similar NT1 ESS gains; urinary frequency is 53–58% on both. Cross-trial comparison.

Orzeyful (1 and 2 mg)Alixorexton NT1 (pooled)Alixorexton NT2 (by dose)Placebo

Swipe sideways to see the ESS column.

0%20%40%60%80%Patients reporting the adverse event (drug arm)Placebo-adjustedESS gain, ptsInsomniaUrinary frequency (pollakiuria)Orzeyful, NT1 Ph31 and 2 mg BID, 12 wkAlixorexton, NT1 Ph24–8 mg pooled, 6 wkAlixorexton, NT2 Ph210, 14, 18 mg, 8 wk55–60%28%27–32 pp below Orzeyful13–36%8.0–9.756.4–8.72.1–3.6Orzeyful, NT1 Ph31 and 2 mg BID, 12 wkAlixorexton, NT1 Ph24–8 mg pooled, 6 wkAlixorexton, NT2 Ph210, 14, 18 mg, 8 wk53–58%55%Similar incidence: 55% vs 53–58%46–73%8.0–9.756.4–8.72.1–3.6

Drug-arm rates, not placebo-subtracted. Placebo: Orzeyful 1% insomnia, 5% urinary frequency; VIBRANCE-2 0% and 13%; VIBRANCE-1 placebo rates not in the published summary. Alixorexton NT1 rates are pooled across 4, 6 and 8 mg because per-dose rates are not published. NT2 rates are by dose and not dose-ordered (insomnia 17%, 36%, 13% at 10, 14, 18 mg). Cross-trial: Ph2 vs Ph3, 6 or 8 vs 12 weeks, and the Orzeyful label groups initial and middle insomnia. ESS: placebo-adjusted; Orzeyful TAK-861-3001; alixorexton week 6 (NT1) and week 8 (NT2).

Source: Orzeyful label and FDA review [S30, S37]; VIBRANCE-1 [S42]; VIBRANCE-2 [S3]. Reported values; no adjustment.


Takeda is not the only rival. Lilly’s cleminorexton (ORX750, acquired with Centessa) is in a registrational Ph2/3, SAPPHIRE, with 222 NT1 and NT2 patients on the registry, and the protocol can add IH by amendment [S21]. That would erode alixorexton’s trial-stage lead in IH. The largest tranche of Lilly’s Centessa CVR, $5.00 a CVR (~$0.83B in total, calculated), pays on US approval of cleminorexton or ORX142 in IH by mid-2031 [S22]. It puts Lilly’s IH program on a defined clock; it is not an estimate of the IH market. Nor is Lilly’s $6.3B Centessa purchase a clean price for an orexin asset: it bought a discovery platform and earlier molecules (ORX142, ORX489) alongside ORX750.

Catalysts to watch. Two rival IH reads could land around VIBRANCE-3. Lilly’s CRYSTAL-1 Ph2a in NT1, NT2 and IH lists completion in Dec 2026, after an interim that showed an IH MWT signal but no IH ESS contrast [S14]; Takeda’s TAK-360 study lists Nov 30, 2026 primary completion, with Week-4 ESS and IHSS secondary [S12, S13]. Neither date promises a release, but a full IH ESS read from either would be the first controlled test of whether the class’s ESS effect transfers to IH, and a benchmark for alixorexton’s.

Pricing is not yet observed. Takeda has not disclosed a US Orzeyful price; Jefferies has guessed anywhere from ~$142K a year, level with Xywav, to more than $250K [S45]. A price near the top of that range would lower the patients the stock needs, which already fall to 27.7K at $112K net. Alkermes management describes Takeda’s pricing as a “price corridor” for a “typical orphan drug”. We infer that an orphan-range price could sit above our $85–112K deck in NT1, though not in NT2 or IH, and will test that when Takeda prints. ICER’s $50.4–59.4K health-benefit benchmark is a cost-effectiveness estimate, not a forecast of realized price [S20]. Realized branded sleep pricing supports a ~$100K net base: by our calculation from 2Q26 revenue, Wakix nets ~$117K per average patient, Xywav ~$110K per active patient and LUMRYZ ~$92K. Our base sits inside that range (Exhibit 6). Takeda’s first price anchors every number in it.

Exhibit 6

Base case nets $100K, inside the $92–117K that branded sleep drugs realize

WAC, gross-to-net and net are kept separate. All four cases are assumptions; no US OX2R agonist price has been observed.

Alixorexton price deck, annual, per patient

CaseWACGTNNet
Upside$144K22%$112K
Base$132K24%$100K
Downside$120K29%$85K
Severe payer pressure$120K54%$55K

Reference points

AnchorValueBasis
Wakix net revenue per average patient~$117K netCalculated: 2Q26 $261.3M × 4 ÷ 8,950 average patients [S38]; 1Q26 on the same basis ~$101K
Xywav net revenue per active patient~$110K netCalculated: 2Q26 $471M × 4 ÷ 17,125 patients [S26]
LUMRYZ net revenue per patient~$92K netCalculated: 2Q26 $96.6M less ~$7M inventory benefit, × 4 ÷ 3,900 [S27]
ICER health-benefit price benchmark$50.4–59.4K netCost-effectiveness benchmark, not a price forecast [S20]
Orzeyful US priceNot disclosedTakeda to publish at US launch, after DEA scheduling [S24]

Net = WAC × (1 − GTN), rounded to the nearest $1K. Wakix divides by average patients, the closest basis to realized net per treated patient; Xywav and LUMRYZ divide by period-end patients and are scale checks. ALKS has said its pricing will be influenced by Takeda's [S27].

Source: Clinaptis assumptions; reference points as labeled.


4. VIBRANCE-3 tests orexin in IH, not the stock’s patient numbers

Some IH patients sleep more than eleven hours a night and still wake unrefreshed. One drug is approved for them, at ~$110K a year net, and ~5,850 people take it.

IH is small, cheap to treat first, and under-served on symptoms ESS misses. We use ~40K diagnosed US patients (company estimate) and assume 75% treated (~30K); the haircut is ours. Management applies the 40K figure to treated, eligible and diagnosed patients at different points and calls it undersized. Xywav is the only approved IH drug, and LUMRYZ cannot be marketed for IH before Mar 1, 2028 [S10]. Generic modafinil and stimulants are cheap and used first. The unmet need is sleep inertia and unrefreshing sleep, which ESS does not measure.

NT2 sets the plausibility, and §2 shows how thin it is. VIBRANCE-2 is the closest positive read-through to IH: OX2R agonism improved sleepiness in a hypersomnia without frank orexin loss, which supports the mechanism outside NT1. The IH question is how much of a ~3-point NT2 effect transfers. NT2-to-IH transfer is the largest unresolved variable: IH differs in sleep inertia, long and unrefreshing sleep, and higher baseline objective wakefulness. Every probability below is conditional on a retained fraction of the NT2 effect, and the Ph1b crossover shows target engagement in IH and little more (A2).

Modafinil is the only parallel-group IH comparator, and it sets ~5 points as the upper reference. Inoue 2021 (n=71, 3 weeks, 200 mg) gives a baseline-adjusted ESS difference of −5.01 (95% CI −7.23 to −2.78), with MWT the primary [S8]. Mayer 2015 (31 analyzed, 100 mg twice daily) gives −3.64 on week-3 final scores [S9, S29]. Cochrane’s pooled −5.08 is a final-score estimate inflated by Inoue’s 1.5-point higher placebo baseline, so we use −5.0 (A4, Exhibit A4). Neither transfers directly: both ran 3 weeks, enrolled IH without long sleep time, and modafinil acts downstream of orexin. Randomized-withdrawal results (Xywav’s ~6.5-point ESS separation [S4]; REVITALYZ) measure deterioration after withdrawal in tolerant patients and do not belong on the same axis. Pitolisant’s INTUNE missed its withdrawal-phase ESS primary despite a −9.4-point open-label improvement [S11].

The commercial bar is modafinil-level efficacy. We expect payers to place a branded OX2R agonist after generics. A three-point effect, replicated across doses, justifies a Ph3 and little else: it gives a stable oxybate patient no reason to switch and a payer no reason to prefer it to generics. A four- to five-point true effect with supportive IHSS would support a competitive claim. That threshold is our underwriting judgment, not a regulatory requirement or a validated IH MCID. Our valuation does not distinguish the tiers: the clean-win state is worth the same $0.81/share (rNPV) at three or five points, and effect size enters only through conditional PoS (§5). A larger true effect would raise share and patient numbers; the clean state holds patient numbers fixed, and the 5,500-patient stretch case in §5 brackets that upside. Statistical success alone leaves the commercial case open.

VIBRANCE-3 is small, and VIBRANCE-2 supplies the variance. The design poster shows ~96 patients randomized 1:1:1:1 to placebo or 10, 14 or 18 mg once daily for 8 weeks, stratified by region and nighttime sleep (<9, 9–11, >11 hours) [S1]; the exploratory split-dose component adds no primary power (A3). VIBRANCE-2’s 18 mg confidence interval implies a residual SD of ~5.5, the figure the Xywav IH statistical plan also assumed [S3]. That supports the variance assumption and says nothing about the IH effect size. At 24 patients per arm, each dose-versus-placebo contrast carries an SE of 1.59.

Exhibit 7

VIBRANCE-3 has a ~60% chance of a win, and a win would overstate the effect

At a true 3.6-point effect, at least one dose has a 62% chance of significance (58% under Bonferroni), and a significant 18 mg arm would print 4.57 points on average. Weighted across outcomes, IH is worth $0.41 a share before the readout.

Fixed sequence, descending dose (central)Bonferroni α/3SD 5.0–6.5

Probability at least one dose is significant

0%20%40%60%80%100%23456True IH 18 mg effect, ESS pointsNT2 mean 3.0NT2 18 mg 3.662%58%

Expected printed 18 mg effect if significant, ESS points

2345623456True IH 18 mg effect, ESS pointsPrinted = truePrinted effect if significant4.344.57true 3.0true 3.6Selectioninflation, +1.3+1.0

What each VIBRANCE-3 outcome is worth (24 per arm, fixed sequence)

Readout outcomeChance of each outcome, by share of NT2 effect IH keepsApproval odds if it occursValue if it occurs, $/sh
75%100% (central)125%
Clean winTwo or more doses significant20%40%63%65%$0.81
MarginalExactly one dose significant19%22%18%35%$0.23
MissNo dose significant60%38%19%5%$0.10
Any dose significant (clean win + marginal)40%62%81%
IH approval odds, weighted23%36%48%
IH value before the readout, $/sh$0.27$0.41$0.57

Left panel scales the NT2 dose pattern (2.1 / 3.2 / 3.6 points); 3.6 is the best of three NT2 arms and is upwardly biased. Right panel: single 18 mg test at α=0.05, SE 1.59. Clean win is a statistical proxy; IHSS, tolerability and usable dosing are not jointly modeled. Approval odds by outcome are our judgment; outcome values are rNPV at $100K net with ~9% LUMRYZ switching, so each weighted row is the sum of chance × value. Other multiplicity procedures are in A3; original-volume values in §5.

Source: Clinaptis calculations [S5] and rNPV model [S18]. Modeled; not observed data. 24 per arm, SD 5.5, shared placebo (ρ=0.5); exact numerical integration.


The odds are ~60% for any dose and ~40% for a clean win. An NT2-sized effect gives a 62% chance that at least one dose is significant and a 40% chance that two are. The latter is our clean-win proxy, which does not jointly model IHSS, tolerability or usable dosing. Retention moves the odds most: at 75% of the NT2 effect, any-dose power falls to ~40%; at 125%, it rises to ~81%. Reaching 70% needs ~3.9–4.1 points at 18 mg, ~110% of the NT2 18 mg result. Sample size matters more than multiplicity (62%, 67% and 72% at 24, 27 and 30 per arm), and the procedure changes which dose wins more than whether any does. The SAP is not public, so we assume a descending-dose fixed sequence from precedent and test seven alternatives in A3. The 3.6-point anchor is the best of three small NT2 arms and makes IH’s hurdle look easier. These are conditional calculations, not probabilities that the drug works in IH.

A win would print ~4.5 points even from a 3-point drug. Conditional on p<0.05, a significant 18 mg arm prints 4.34 points on average if the true effect is 3.0, and 4.57 if it is 3.6 (Exhibit 7). A single contrast carries a 95% CI of ±3.1 points, so the printed best-arm number cannot separate a three-point drug from a four-point one. Agreement between 14 and 18 mg is more informative; a pooled 14+18 mg contrast (SE 1.38) is a useful descriptive check but is neither prespecified nor multiplicity-controlled. Nor would a win settle transfer: with equal starting odds across our three retention cases, a clean win lifts the chance that IH keeps at least the full NT2 effect from 67% to 84%, and a miss cuts it to 48%.

A winning dose has to be usable. VIBRANCE-2’s tolerability was nonmonotonic: pollakiuria and insomnia peaked at 14 mg (73% and 36%) rather than 18 mg (46% and 13%) (Exhibit 5). At ~22–24 patients per arm that does not support dose-level safety rankings. Management says pollakiuria “doesn’t tend to lead to discontinuation,” extension retention is 90% in NT1 and 81% in NT2, and VIBRANCE-2 discontinuations for adverse events were 9%, 5% and 0% at 10, 14 and 18 mg (A6). But in a chronic disorder with disturbed sleep, persistence will shape the dose, and our consistency read leans on 14 mg corroborating 18 mg. The dose whose replication we need most is the messiest, and ~22 patients an arm will not resolve it.

Sleep-duration subgroups will not help: at ~8 patients per arm per stratum the 95% CI is ±5.4 points, and external evidence points both ways [S28], so we assign no subgroup premium or discount.

What the readout is worth. Weighting the three outcomes by their odds gives IH $0.41 a share of value before the readout: a clean win is worth ~$0.81, a marginal result ~$0.23 and a miss ~$0.10 (Exhibit 7). The $0.41 is a probability-weighted value, not the value of a win. Exhibit A5 maps each result to its state for readout day; agreement between usable 14 and 18 mg doses, supported by IHSS, earns more weight than the best-arm number.

5. VIBRANCE-3 is worth ~$1 a share

We value the readout two ways, both at ~$100K net revenue per patient a year and on 172.763M diluted shares. The standalone rNPV discounts at 11%, charges IH Ph3 spending only after a supportive readout and deducts ~9% IH cannibalization from LUMRYZ; narcolepsy approval probability is 65%. The DCF convention applies the company DCF’s 10.5% discount rate, tax and terminal value to the same IH revenue (A7): “net” charges IH Ph3, IH SG&A and cannibalization as risked costs, and “gross” holds company costs fixed. Exhibit 8 is a fundamental sensitivity, not a stock forecast.

Exhibit 8

The IH readout is worth ~$1 a share; $2–3 needs 5,500–7,000 patients

At $100K net on our rebuilt forecast, a clean win adds $0.39 and a miss removes $0.32 against the $0.41 pre-readout IH value (rNPV). Clean win minus miss is $0.71 by rNPV, $0.98 by DCF net and $1.81 gross.

Rebuilt base (~3,500 IH patients)Original model (7,500 IH patients)

Change vs pre-readout IH value (zero line), rNPV, $ per share

−1.0−0.50.0+0.5+1.0+$0.39+$1.15Clean win−$0.18−$0.20Marginal−$0.32−$1.11Miss

Clean win minus miss (not stock upside), rNPV, by net price

0.00.51.01.52.02.53.0$55Ksevere$85K$100Kbase$112K$0.84$1.79$2.27$2.65$0.04$0.49$0.71$0.89OriginalmodelRebuiltbase

Clean win minus miss by IH peak patients and valuation method, $ per share, $100K net

IH peak patientsrNPV (primary)DCF net (~99% terminal value)DCF gross (upper bound, opex fixed)
Low (~1,500)−$0.07$0.02$0.78
Rebuilt base (~3,500)$0.71$0.98$1.81
Stretch (5,500)$1.48$1.92$2.82
Rebuilt high (~6,800)$1.99$2.54$3.49

A fundamental sensitivity, not a predicted stock move. rNPV: 11% discount rate, explicit IH Ph3 spend, ~9% LUMRYZ cannibalization. DCF net: the company DCF's 10.5% WACC, charging IH Ph3 spend, IH SG&A and cannibalization as risked costs; DCF gross holds company opex fixed with no cannibalization. Original model (7,500 patients): $2.27 rNPV, $2.87 DCF net, $3.85 gross. Conditional IH PoS 65 / 35 / 5%. Stretch case (5,500) is our scenario.

Source: Clinaptis model [S5, S18, S23]. Modeled; not observed data.

View reconciliation & method detail

rNPV uses 172.763M diluted shares. DCF net also applies the company DCF's 22% tax rate and terminal value. Franchise rNPV at $100K net is $3.8 per share on rebuilt volume and $14.6 on the original; the legacy DCF carries ~$23 per share for alixorexton (Exhibit A7). State weights: power model (fixed sequence, 100% retention) times judgment conditional IH PoS of 65 / 35 / 5%.


The win-minus-miss spread is about $1 a share at ~$100K net on our patient forecast: $0.71 by rNPV, $0.98 by DCF net, $1.81 gross. Across $85K/$100K/$112K the rNPV spread is $0.49/$0.71/$0.89 and the DCF-net spread $0.71/$0.98/$1.19. That is a spread between outcomes, not upside from today’s value: a clean win adds $0.39 (rNPV) to $0.52 (DCF net) to the modeled pre-readout IH value, and a miss removes $0.32 to $0.45. DCF net is ~99% terminal value, capitalizing 2036 cash flow in perpetuity while IH is still ramping (A7), so we treat it as the generous bound, not as independent corroboration of the rNPV. Rebuilt value falls to about zero at $55K. In our low case (~1,500 patients) the rNPV spread turns negative (−$0.07): a standalone IH launch cannot cover its fixed costs, so a clean win is worth less than a miss.

$2–3 needs 5,500–7,000 IH patients, 1.6–2.0x our base. At 5,500 patients the spread is $1.5 (rNPV), $1.9 (DCF net) and $2.8 (gross); at our 6,800-patient high case, $2.0 / $2.5 / $3.5. Our base is ~3,500 peak patients, about 12% of ~30K treated IH patients and 60% of the ~5,850 Xywav already treats. We read uptake as slower than that pool implies, because IH patients are sequenced behind wake promoters and stimulants: Xywav is five years past its IH approval and treats 5,850 of ~40K diagnosed patients. That read is ours. No sponsor baseline is published, and Jazz reports strong new-patient growth in IH and coverage for ~90% of commercial lives [S36]. We carry 5,500 as a stretch case, not the central one. Reaching it would take consistent 14/18 mg efficacy with IHSS support, usable dosing and persistence, payer access beyond generic step edits, and a net price near $100K; VIBRANCE-3 cannot tell us which case is right.

What moves the modeled value. Outcome chances come from the power model (Exhibit 7); approval odds by outcome are judgment: 65% after a clean win (~70% Ph3 replication times ~93% approval on an existing narcolepsy label), 35% after a marginal result, which needs a larger, redesigned Ph3 from a weaker prior, and 5% after a miss, which likely ends the program. At full retention that gives a ~36% pre-readout IH PoS, slightly below the 40% in our original model; we have not fitted the weights to it. At 75% retention the PoS is ~23% and the pre-readout value $0.27 rather than $0.41, and across the eight multiplicity procedures it is $0.34–0.41 ($1.14–1.34 on the original forecast), with no change to the win-minus-miss spread.

A marginal result costs more than our base model charges. It delays IH and enlarges the Ph3, leaving the premium unchanged on the data, but our marginal state keeps the clean state’s 2032 launch and R&D schedule. Delaying launch to 2034 and adding two years of $35M Ph3 spend cuts the marginal state from $0.23 to −$0.11 and the pre-readout IH value from $0.41 to $0.34 (rNPV), without changing the clean-minus-miss spread. A clean win adds ~$0.4–0.5 and confirms assumed efficacy while leaving the patient numbers unresolved. A miss, possible even with full transfer, removes ~$0.3–0.45 and is weak evidence against the drug: our retention grid excludes no-effect states, so a miss does not rule out efficacy.

LUMRYZ cuts both ways. It gives ALKS a sleep-specialist field force and payer relationships, which we assume keeps incremental alixorexton SG&A to ~$120M rather than a standalone build. That $120M is our assumption: management says it will be building out the sales force, and psychiatry and sleep forces overlap little. It also exposes ALKS to cannibalization. LUMRYZ can be marketed for IH from Mar 2028 (§4), years before alixorexton. ALKS reports physician interest in combining orexins with oxybates, but management says combination “will not be the modal treatment,” that wakefulness alone may suffice for many, and that combination evidence is aimed at payers. It does not rebut cannibalization, so we keep the derived ~9%.

Peak LUMRYZ revenue in our DCF is ~41% of the peak oxybate pool, and 22% of alixorexton IH patients come from oxybate. That puts ~9% of alixorexton IH patients as LUMRYZ switchers (calculated). At the base forecast and $100K net, pre-readout IH value is $0.48 / $0.44 / $0.41 / $0.37 / $0.25 per share at no, 5%, ~9% (central), 15% and 30% switching. The derived figure may be high for IH, since LUMRYZ is unlikely to hold its overall oxybate share there.

IH cannot validate the narcolepsy premium, but the stock can move more than the IH value at stake. The stock pays ~$9–15/share above our rebuilt DCF, depending on embedded spend. The trial informs IH retention odds and sentiment without testing NT1/NT2 class share; the marginal buyer at $52 was paying for an orexin platform, not an IH program (our read), and the print tests only the program. Positioning adds leverage. ALKS rose 55% from Apr to Jun 2026 versus XBI’s +20%, then fell 23% by Sep while XBI was flat; short interest fell from 19.8M (Apr 30) to 14.4M (Jul 31) before rebuilding to 19.0M (Oct 15; 12.3% of float, ~9 days to cover), enough to amplify a win or a miss (Exhibit A1). Investor attention is elsewhere: across the last three calls, 7 of 47 analyst questions touched VIBRANCE-3, against ~9 on ADHD or fatigue and ~11 on LUMRYZ or oxybate.

The risk-reward is unfavorable: about $1 a share between a win and a miss ($2–3 only on IH patient numbers we cannot yet support), no validation of the narcolepsy patient forecast, and a premium exposed to disappointment. The patient-forecast rebuild moves value ~7–15x the IH spread, depending on method. That is our read of the risk-reward, not a forecast of the print: a low drug value does not imply a small price reaction.

6. What would change our view: Orzeyful’s launch and Brilliance NT2

Orzeyful’s launch and Brilliance NT2 are the real tests of the patient numbers. Takeda’s first US quarters will show how fast an orexin agonist adds patients against the pace in Exhibit 2, what it costs (Exhibit 6) and how prescribers handle insomnia and urinary effects (Exhibit 5). A launch adding well above ~500 patients a quarter within a year would weaken our pace objection; one that tracks Wakix and LUMRYZ would strengthen it. Brilliance NT2, with primary completion estimated for May 2027, decides the NT2 claim, and its data would show it before sales do. An ESS effect near 5 points at more than one regimen would falsify our NT2 leg on its own; a ~3-point effect that wins on MWT would confirm our base case. Per-dose NT1 data confirming the insomnia advantage would lift our 22% NT1 share, but neither that nor a fast Orzeyful launch closes the gap without NT2 (§2).

VIBRANCE-3 can move our IH assumptions, not the thesis. A multiplicity-controlled win with consistent 14 and 18 mg efficacy and IHSS support would raise our conditional IH assumptions; below ~2 points at both doses, or efficacy confined to poorly tolerated exposure, would lower them sharply. Failed NT2-to-IH transfer remains the main IH risk, and it is small in dollars. Only evidence that IH peak patients exceed ~5,500, such as accelerating oxybate IH uptake or a management IH patient estimate we can reconcile, would move the spread toward $2–3.

The valuation case could also fail outside the clinic. A weak IH result would not validate a short: LUMRYZ revenue 25% above our forecast adds ~$7 a share to the DCF, more than the IH swing. Our variant view would also fail on patients above our high case, materially stronger base cash flow, friendlier DCF conventions, or unmodeled ADHD, fatigue, capital-return or strategic value.

ALKS: Neutral. The valuation is ahead of the evidence.

At ~$45, ALKS needs ~31,000 alixorexton patients; our forecast supports ~12,300. NT2 must carry most of the difference, and its data point to an ESS benefit near 3 points, short of what it takes to pull patients off cheap generics at scale. VIBRANCE-3 can show alixorexton works in IH, but not that the stock’s patients exist, and we put its fundamental value at ~$1 a share against a far larger narcolepsy premium. We stay Neutral: the price is hard to defend, but a weak IH result alone is not a short thesis. Orzeyful’s launch and Brilliance NT2 are the tests that matter.

Appendix

A1. Short interest could amplify the move on either VIBRANCE-3 outcome

Exhibit A1

ALKS ran 55% Apr–Jun as shorts fell; shorts are back at 19M

Indexed to Oct 2024, ALKS is up 76% against 61% for XBI. Short interest fell from 19.8M shares on Apr 30 to 14.4M on Jul 31, and was back at 19.0M on Oct 15.

Month-end closes, indexed (Oct 2026 month to date)

80100120140160180200Vibrance-1NT1 toplineAvadel dealannouncedVibrance-2NT2 toplineLilly–Centessa$6.3B dealREVITALYZLUMRYZ IHOrzeyfulFDA approvalIBB 144XBI 161ALKS 176Oct 2024Jan 2025Apr 2025Jul 2025Oct 2025Jan 2026Apr 2026Jul 2026Oct 2026Indexed, Oct 2024 = 100

ALKS shares sold short, FINRA settlement dates

81216202421.8M, 5-year high14.4M19.0MShares short12.3% of floatOct 2024Jan 2025Apr 2025Jul 2025Oct 2025Jan 2026Apr 2026Jul 2026Oct 2026Millions

Event lines mark co-movements, not attributed causes. IBB is used as a proxy for the Nasdaq Biotechnology Index. Short interest: 9.1 days to cover on Oct 15; five-year low 5.8M (Jun 2023).

Source: month-end closes via digrin.com [S31]; FINRA consolidated short interest via ChartExchange and CurvedTrading [S32]; event dates from company releases.


Monthly closes indexed to Oct 2024, through Oct 2026. ALKS +76% vs XBI +61% over the period [S31]. The lower panel shows FINRA short interest on biweekly settlement dates [S32]. The 19.0M shares of Oct 15 is close to the 21.8M five-year high of Dec 31, 2025; five-year context is ~12M in Oct 2021, a 5.8M low in Jun 2023, and a step up from Dec 2023. These are co-movements around dated events, not attributed causes.

A2. Ph1b shows target engagement in IH, not efficacy

The Ph1b crossover shows target engagement in IH, and little more. Placebo-adjusted MWT gains at 5, 12 and 25 mg were +8.1, +11.1 and +17.7 minutes in eight IH patients, against +11.6, +18.6 and +21.0 minutes in nine NT2 patients [S7]; we do not apply those 60–84% ratios to ESS. Post-dose absolute latencies were similar (~29.0 / 31.9 / 38.6 min in IH vs ~26.0 / 32.9 / 35.3 min in NT2; digitized baseline plus reported change), but IH started higher (~22.6 vs ~14.3 min), so its smaller change reflects less room under the 40-minute cap. The Ph1b NT2 cohort also had no MWT entry cap, whereas VIBRANCE-2 required MSL ≤15 min and averaged 5.6 min [S3]. Single-dose MWT measured 2–8 hours after dosing cannot calibrate an 8-week, 7-day-recall ESS difference.

Exhibit A2

Alixorexton takes IH and NT2 patients to similar wakefulness

Ph1b crossover, mean MWT sleep latency in minutes. The IH–NT2 gap narrows from 8.3 minutes at baseline to 3.3 at 25 mg, and reverses at 12 mg.

010203040MinutesBaseline5 mg12 mg25 mg40-min MWT ceilingVIBRANCE-2 NT2baseline, 5.6 min14.326.032.935.322.629.031.938.6IH (n=8)+16.0 min from baselineNT2 (n=9)+21.0 min from baseline

Single doses, latency measured 2–8 hours after dosing. Ph1b shows IH target engagement; it does not supply an ESS translation factor.

Source: Alkermes SLEEP 2025 Ph1b poster [S7]. Baselines digitized; post-dose values are baseline plus reported LSM change (calculated).


A3. VIBRANCE-3 power: the split-dose cohort adds nothing to the primary

VIBRANCE-3 enrolls adults with ESS >12 after screening and a two-week washout; Week-8 ESS is primary and IHSS key secondary [S1]. The SAP is unpublished.

Documented. VIBRANCE-2 used graphical testing across MWT, ESS and three doses: adjusted 18 mg ESS p was 0.046 against ~0.02 unadjusted, and the largest MWT effect, at 10 mg, was untestable. Orzeyful’s Ph3s used graphical hierarchies [S30].

Inferred from precedent. VIBRANCE-2 is consistent with α split across endpoints and descending doses within each. With one primary, we give 18 mg full α, then 14, then 10.

Unknown. The actual procedure, α weights, IHSS hierarchy, estimand and missing-data handling; dual-primary precedent cannot settle single-primary testing.

Split dosing. An added ~30-patient cohort, randomized to alixorexton or matching split-dose placebo, began enrolling in Q2 2026 and was still enrolling at 2Q26; management says the whole-study topline includes it, so it gates timing. Regimen and alpha status are undisclosed, so we treat the cohort as descriptive, assign it no primary power and would not let an exploratory signal offset a failed once-daily primary. The 27/30-per-arm sensitivities assume extra once-daily enrollment, which is undisclosed; dropout works against them (VIBRANCE-2: 90/93 Week-8 completers).

Power inputs and method. Four equal arms, n=24, residual SD 5.5, contrast SE 1.588, shared-placebo ρ=0.5; NT2 effects of 2.1/3.2/3.6 scaled by retention. Fixed sequence 18→14→10 at two-sided α=0.05 is central; the alternatives are two graphical schemes, Dunnett step-down and single-step, Hochberg, Holm and Bonferroni α/3. Correlated contrasts give Dunnett ~2 points over Bonferroni on any-dose power, and a graphical split ~1 point over fixed sequence. We use exact numerical integration (120-node Gauss–Hermite, conditional on placebo noise), not Monte Carlo; the central case reproduces 62.1% [S5], and the significance filter uses the truncated-normal mean above 1.96 × SE. Uncertainty in true NT2 efficacy, dropout, non-normality and small-sample t effects is unmodeled, and the actual SAP may differ from every procedure tested.

Exhibit A3

Procedure moves P(18 mg wins) 17 points, P(any dose) 5

At the NT2 effect size and 24 per arm, at least one dose wins 58–63% of the time across eight procedures; 18 mg alone wins 45–62%. Three to six extra patients per arm add 5–10 points.

24 per arm; % of NT2 effect retained in IH≥1 dose significant, 100%
ProcedureBasis18 mg significant, 100%≥1 dose, 75%≥1 dose, 100%≥1 dose, 125%≥2 doses, 100%27 per arm30 per arm
Fixed sequence, descending dose central18, then 14, then 10 mg at full α. Central case, inferred from the VIBRANCE-2 dose-ordered scheme62%40%62%81%40%67%72%
Graphical, 18 mg first18 mg at full α, then 14 and 10 mg split α/262%40%62%81%37%67%72%
Graphical, 18/14 split18 and 14 mg at α/2 each, 10 mg last (Orzeyful Ph3 style)55%40%63%82%39%68%73%
Dunnett step-downStandard for doses against a common placebo50%38%60%80%35%66%70%
HochbergStep-up; valid under positive correlation49%36%59%79%36%65%69%
HolmStep-down48%36%58%79%34%64%69%
Dunnett single-stepCommon placebo, no ordering47%38%60%80%30%66%70%
Bonferroni α/3Conservative bound45%36%58%79%28%64%69%

The VIBRANCE-3 procedure is undisclosed; the central case is an inference from precedent. 27 and 30 per arm are sensitivities, not disclosed arm sizes; the 30 split-dose patients add no primary power.

Source: Clinaptis calculations [S5]. Modeled; not observed data. Two-sided familywise α 0.05, SD 5.5, shared placebo; exact enumeration with 1-D quadrature, no simulation.


A4. Modafinil in IH sets the ~5-point efficacy reference

Inoue 2021: n=71, 3 weeks, 200 mg each morning, Japanese ESS ≥11. MWT was primary (+5.02 minutes; 95% CI 3.26–6.77); the ANCOVA ESS difference was −5.01 (95% CI −7.23 to −2.78), and ESS normalized to ≤10 in 55% versus 8% [S8].

Mayer 2015: 31 analyzed, 3 weeks, 100 mg twice daily. Cochrane’s final-score ESS difference is −3.64 (95% CI −6.93 to −0.35); MWT did not separate [S9, S29].

Cochrane: pooled final scores give −5.08 (95% CI −7.16 to −3.01), with Inoue carrying 65% of the weight at −5.86. Inoue’s placebo arm started 1.5 points higher (18.22 vs 16.71), which inflates its final-score difference over its change difference (−5.01). For VIBRANCE-3’s change-from-baseline endpoint, Inoue’s −5.0 is the like-for-like anchor and Mayer’s −3.6 the smaller, less precise trial; an inverse-variance combination of the two gives ~−4.6 (calculated; it mixes a change and a final-score estimate). We use −5.0 as the upper reference. Both trials enrolled IH without long sleep time, whereas VIBRANCE-3 stratifies on sleep duration up to >11 hours.

Exhibit A4

NT2 ESS effects sit near 3 points; IH modafinil sets ~5

Alixorexton 18 mg moved ESS by 3.6 points in NT2. Modafinil's best like-for-like IH estimate is Inoue's change-from-baseline −5.0; Cochrane's −5.1 pools final scores.

−8−6−4−2024Placebo-adjusted ESS, points (negative favours drug)Difference (95% CI)NT2, 8 weeksAlixorexton 10 mg (VIBRANCE-2)−2.1 (−5.3, +1.1)Alixorexton 14 mg (VIBRANCE-2)−3.2 (−6.3, −0.1)Alixorexton 18 mg (VIBRANCE-2)−3.6 (−6.7, −0.5)Oveporexton low (TAK-861-2002)−0.32 (−3.6, +2.9)Oveporexton high (TAK-861-2002)−3.06 (−6.2, +0.1)IH, modafinil, 3 weeksInoue 2021, change from baseline−5.01 (−7.2, −2.8)Inoue 2021, final score−5.86 (−8.1, −3.6)Mayer 2015, final score−3.64 (−6.9, −0.3)Cochrane pooled, final scores−5.08 (−7.2, −3.0)

Filled markers: alixorexton (blue) and the change-based modafinil anchor (black). Hollow: comparators and final-score estimates; the diamond is a pooled estimate. Panels are not pooled with each other. Inoue's placebo arm started 1.5 points higher (18.22 vs 16.71), which inflates its final-score difference. Both modafinil trials enrolled ICSD-2 IH without long sleep time. Randomized-withdrawal programs (Xywav, LUMRYZ, pitolisant) measure deterioration off drug and are not plotted.

Source: VIBRANCE-2 [S3]; TAK-861-2002 [S6], low-regimen CI calculated from its SE; Inoue 2021 [S8]; Cochrane Analysis 1.1 [S9, S29].


A5. What management says VIBRANCE-3 must show matters more than the multiplicity detail

Management has not disclosed the multiplicity procedure and is unlikely to before topline, so we carry it as a modeled sensitivity (§4, A3). The more useful disclosure is the efficacy bar management sets.

Efficacy bar. Any significant 18 mg arm clears ~3.1 points at 24 per arm (calculated), so a “~3-point” bar is mechanical. A modafinil-like ~5 points appears in only ~22–30% of wins for true effects of 3.0–3.6 (calculated). The signals that matter are 14/18 mg consistency and IHSS (§4). Management has pre-framed heterogeneity (“average ESS doesn’t tell the whole story”); we grade on the prespecified primary, not on responder reframing.

Scope and timing. Management says the topline covers the whole study, including the split cohort; completion is guided to 4Q26, with data possibly in early 2027. A once-daily primary reported separately, with IHSS and sleep-duration strata, would make the result easier to read; a slipping split cohort would delay it.

Patient numbers. Peak patients, class share and oxybate switching by indication are the inputs our thesis needs, and VIBRANCE-3 cannot supply them.

Exhibit A5

Dose consistency says more than the printed 18 mg number

What each VIBRANCE-3 result would establish, and the valuation state it maps to.

ResultWhat it establishesImplication for the thesis
14 and 18 mg both significant, IHSS concordantReal, reproducible IH effect at a usable doseClean-win state (~$0.81/share on rebuilt volume; ~$2.5 on original volume). Does not resolve the narcolepsy volume question
18 mg significant at a printed ≥4–5 points, 14 mg weakStatistical win; true effect could be ~3Marginal-to-clean. Discount the printed number; dose consistency matters more than size
18 mg nominal, multiplicity-negative; 14 mg directionally supportiveConsistent with a real NT2-sized effect in an underpowered trialMarginal state (conditional IH PoS ~35%). Not a falsification; delays IH and enlarges the Ph3
No dose above ~2 pointsWeak IH ESS effectMiss state. IH value ~$0.1/share; narcolepsy thesis unaffected
Efficacy only at a dose with insomnia >30% or D/C >10%Narrow usable windowMarginal at best; split-dose data become the key follow-up
Split-dose signalRegimen-optimization information; coverage hypothesis supported, not confirmedExploratory. Informs the Ph3 regimen; no credit toward the once-daily primary

State values from Exhibit 8 at $100K net. Thresholds are Clinaptis underwriting judgments, not regulatory criteria.

Source: Clinaptis analysis [S3, S5].


A6. Narcolepsy evidence leaves IH transfer unresolved

VIBRANCE-1 (NT1). A randomized, double-blind, six-week Ph2 in 92 patients at 4, 6 and 8 mg once daily. All doses met the MWT primary, with mean week-6 latency of ~24 / 26 / 28 minutes from a ~3-minute baseline [S33]. ESS (baseline 18.5) improved by 6.4, 8.7 and 8.3 points versus placebo; the weekly cataplexy rate reached significance on the prespecified analysis at 6 mg only [S42]. Pollakiuria (55%) and insomnia (28%) are reported across doses through week 6; per-dose rates are not published.

VIBRANCE-2 (NT2). Week-8 ESS fell 3.7 points on placebo and 5.8 / 6.9 / 7.2 points at 10 / 14 / 18 mg. Adjusted ESS p-values were 0.195 / 0.077 / 0.046 (unadjusted ~0.20 / ~0.04 / ~0.02, calculated from CIs); the company reports MWT significant at 14 and 18 mg and ESS at 18 mg only after multiplicity adjustment [S34]. MWT contrasts were +9.3 / +6.7 / +6.7 minutes, with 10 mg untestable under the graphical procedure. Pollakiuria ran 57 / 73 / 46% (placebo 13%), insomnia 17 / 36 / 13% (0%) and urinary urgency 30 / 14 / 21% (0%); D/C for adverse events was 9 / 5 / 0% (0%).

Responder conversion (modeled). Assuming normally distributed Week-8 ESS changes with the trial’s implied SD of 5.5, 62% of placebo patients improve by ≥2 points against 76 / 81 / 83% on drug (differences of 13 / 19 / 21 points); at ≥5 points the rates are 41% against 56 / 64 / 66% (15 / 23 / 25 points). Takeda’s NT2 trial through the same arithmetic gives 60% against 79% at the 2-point bar, so the responder view does not separate the programs either, and placebo already converts most patients at 2 points. These rates are calculated from published means and an assumed SD, not observed responder data.

A7. Company DCF: $23 of the $43 rests on alixorexton

This part of the appendix holds the DCF reasoning behind §1; the peer valuation in Exhibit 3 is the independent check.

$23 of our $43 DCF is alixorexton, on patient numbers we cannot support. Our company DCF values ALKS at $43.17/share [S23]. Its hard-coded, risk-adjusted alixorexton revenue rises from $100M in 2030 to $1.25B in 2036, and only COGS scales with it. Removing the line at the DCF’s conventions (10.5% WACC, 22% tax, −2% terminal growth on 2036 FCF) removes $23.2/share, $13.5 through terminal value. With no attached program cost, the model keeps ~68% of each revenue dollar after tax; if rNPV-sized costs (~$90–140M a year risk-adjusted) sit inside company opex, net embedded value is ~$17–19/share. Whether they do is unknown.

Two FCF add-backs and the share base flatter the DCF. The company DCF adds back ~17% of R&D ($84M a year) and all stock-based compensation, and divides by 172.763M diluted shares that include ~5M existing options but no future grants; our rNPV uses the same base. Discounted, the R&D add-back is worth $4.85/share and SBC $4.39 (calculated). We do not subtract them, because they are the model’s conventions and their economic treatment is unresolved. Removing both would take the DCF to ~$34, our rebuilt reference to ~$21 and the ex-alixorexton value to ~$11, and would raise the patients $45 requires from ~31K to ~42K at $100K net (~37K without the R&D add-back alone). That supports our view, but it also means neither $43.17 nor $30.0 is a validated absolute.

The original forecast has 1.4x today’s reported oxybate patients. Its 28,620 peak US patients (NT1 13,200, NT2 7,920, IH 7,500) compare with 21,025 on Xywav and LUMRYZ today, and at our ~22% NT1 share imply an OX2R class reaching 136% of treated NT1 patients (Exhibit 1). The reported-patient comparison does not depend on our assumptions; the share check does.

Exhibit A7

Alixorexton is $23 of the $43 DCF; the rebuilt patient forecast backs $10

Removing the embedded alixorexton line at the DCF's own conventions takes $23.2 off $43.2; the rebuilt-patient line adds back $10.0, leaving $30.0 against a ~$45 share price.

$0$10$20$30$40$50$43.2Repairedlegacy DCF−$23.2Legacyalixorexton line$20.0DCF ex-alixorexton+$10.0Rebuilt-volumeline, $100K net$30.0DCF onrebuilt volume$45.3ALKS sharepriceTV $13.5−$13.2

DCF value per share with the alixorexton line replaced, by volume and annual net price (change vs $43.17)

Alixorexton volume$85K net$100K net$112K net
Rebuilt low$24.0 (−19.2)$24.7 (−18.5)$25.2 (−17.9)
Rebuilt base$28.5 (−14.7)$30.0 (−13.2)$31.2 (−12.0)
Rebuilt high$35.9 (−7.2)$38.8 (−4.4)$41.0 (−2.2)
Original model volume$40.3 (−2.9)$43.9 (+0.7)$46.7 (+3.6)

Calculated on the company DCF at its own conventions: 10.5% WACC, 22% tax, COGS as a share of total revenue, −2% terminal growth on 2036 FCF, 172.8M diluted shares. R&D and SG&A are hard-coded company totals and are held fixed. If alixorexton costs of the size in our rNPV sit inside them, the net embedded value is $16.9–19.0 rather than $23.2. Replacement lines are risk-adjusted at 65% narcolepsy PoS and 36% conditional IH PoS. A sensitivity, not a price target; the standalone rNPV is not added to the DCF. Share price is the Oct 30, 2026 close.

Source: Clinaptis calculations on our company DCF [S23] and volume rebuild [S18].


Patient definition. Patients mean average on-therapy patients over a year, so annual net revenue per patient applies directly; peak year-end counts would overstate revenue without a persistence adjustment. Exhibits label peak patients on treatment “volume.”

The rebuilt patient forecast removes ~$13/share. We substitute risk-adjusted revenue (65% narcolepsy PoS; 36% conditional IH PoS) and hold company opex fixed, because launch cost does not scale with patients; the company forecast trims R&D and SG&A as revenue falls in 2033–36, and we model no further response. At $100K net, 2036 revenue is ~$570M and the DCF ~$30.0 (−$13.2). The base forecast gives $28.5–31.2 across $85–112K net; the low-to-high range gives $24.7–38.8 at $100K. The original forecast at $100K gives $43.9 (+$0.7), matching the embedded line within rounding.

The rebuilt line is worth more inside the DCF (~$10/share) than as a standalone rNPV ($3.8). The DCF capitalizes 2036 cash flow in a perpetuity while the franchise is still ramping, discounts at 10.5% rather than 11%, and attaches no program cost. We do not add the rNPV to the DCF or net one against the other. ~$30 is our DCF reference; it is not a price target.

What the inference assumes. Ex-alixorexton DCF value is ~$20/share, up to ~$26 if embedded program spend also goes, so at ~$45 the residual is ~$19–25/share against our ~$10. Three conditions govern that reading.

The base-business forecast is right. The base business is Vivitrol, Aristada, Lybalvi, LUMRYZ and a manufacturing and royalty line that loses Consta royalties in 2027. Its psychiatry growth engines are flattening. Lybalvi has printed $92–98M a quarter for four quarters while prescriptions grew 18%, with gross-to-net rising from 29% toward the high 30s, yet the DCF takes it from $390M in 2026 to $685M in 2030. Aristada fell 4.5% Y/Y in 2Q26 on 7% unit growth as net price eroded [S27]. ±10% on ex-alixorexton FCF moves the ex-alixorexton value to $17.6–22.4 and the ascribed value to $22.9–27.7. The LUMRYZ line (~$1.05B at peak, ~11,400 patients at ~$92K net against ~3,900 today) is the largest single swing: ±25% on it moves the ascribed value to ~$18 or ~$33. LUMRYZ patients rose 3,500 → 3,600 → 3,900 as growth slowed to 25%, and guidance implies flat H2 revenue; the 11,400 peak is 2.9x today’s base, 23% of management’s ~50K oxybate-eligible narcolepsy pool and 1.3x the ~9K cycling segment (calculated). The 3Q26 print (Nov 3) did not change this: ~$17M of its $480M revenue was one-time gross-to-net and inventory benefit, Lybalvi was ~$94M underlying on 18% prescription growth, and LUMRYZ added ~200 patients to ~4,100 on flat underlying revenue. Vivitrol is smaller: the DCF fades it from $465M (2027) to $270M (2036), and holding it flat at $470M adds at most ~$4.0/share (calculated). For our ~$10 to be consistent with the price, ex-alixorexton FCF would have to be 1.6–1.7x the DCF’s; our high case with the base business 10% higher still reads only $41–43.

The DCF conventions are neutral. The 10.5% WACC and −2% terminal growth on a still-ramping 2036 apply equally to the base business and to the alixorexton line, so the residual is sensitive to them in both directions. The R&D and SBC add-backs above are not neutral: they raise the base business, and removing them widens the residual.

Nothing else explains the residual. An ADHD or fatigue option, capital return or strategic value would sit in the price but in neither our work nor most of the DCF, which holds only $150M (~$0.87/share) of residual pipeline value against an ascribed ~$3.3–4.3B. The non-sleep pipeline is ALKS 7290 in ADHD, whose Ph1b (n=50) reported only within-arm changes on Sep 21 with the 10-patient placebo arm undisclosed, and whose ~312-patient Ph2 reads out in 2027; a Ph2 in major depressive disorder (MDD) that began in 3Q26; a fatigue program; and unnamed ideas. The $172M of remaining buyback authorization is ~$1.00/share of deployable cash, not added value. Asked about a sale process on the 1Q26 call, management called it “a little premature” and did not rule it out; that is strategic value we cannot size.

The price requires more patients than any orexin product has yet treated, unless one of those three conditions breaks.

IH catalyst under DCF conventions (calculated). Same states and weights as the rNPV, with the rebuilt IH line risked at 65% narcolepsy approval times each state’s conditional IH PoS (65 / 35 / 5%), COGS following revenue, and the DCF’s WACC, tax, terminal growth and share base. “Net” charges IH Ph3 R&D (unrisked before launch in clean and marginal states, nil in a miss), IH SG&A at 20% of incremental SG&A and LUMRYZ cannibalization (~9% of IH patients at $92K net and 75% margin) as risked costs; “gross” leaves company opex fixed and ignores cannibalization. At $100K net, the clean-win-minus-miss spread by IH peak patients is:

IH peak patientsrNPVDCF netDCF gross
~1,500−$0.07$0.02$0.78
~3,500 (base)$0.71$0.98$1.81
5,500 (stretch)$1.48$1.92$2.82
~6,800 (high)$1.99$2.54$3.49
7,500 (original)$2.27$2.87$3.85

Pre-readout IH value at the base forecast is $0.41 (rNPV), $0.57 (DCF net) and $1.08 (gross).

Method notes. Company-DCF calculations hold opex fixed, with COGS following revenue. The reverse-solve in §1 scales the base NT1/NT2/IH mix uniformly; base-business sensitivity scales ex-alixorexton FCF through the terminal year, and LUMRYZ sensitivity scales revenue and COGS. DCF net is ~99% terminal value ($1.08 of $1.10 in the clean state), and the rNPV runs to 2045 with no explicit patent or revenue fade. The delayed-marginal sensitivity (§5) moves the marginal launch to 2034, with peak in 2039. One base-business check is not run: management guides Ph3 completion in 2027 while the DCF alixorexton line starts in 2030.

Sources and provenance

Primary trial and regulatory material

Internal models and data

  • [S18] Clinaptis alixorexton rNPV model (patient forecast rebuild, price deck, readout-state valuation). Modeled.
  • [S23] Clinaptis ALKS company DCF model and the alixorexton reconciliation. Calculated.
  • [S31] ALKS, XBI and IBB monthly closes via digrin.com.
  • [S32] FINRA consolidated short interest, 121 settlement dates Oct 2021–Oct 2026, no gaps. Exact figures and float from ChartExchange (Oct 2025 on); earlier rows from CurvedTrading, rounded to 0.1M. Overlapping dates agree within 0.05M.

[S15], [S16], [S17] and [S25] are retired.