Bristol Myers Squibb (BMY) reported that admilparant 60 mg BID met the Ph3 ALOFT-IPF primary endpoint and 120 mg did not, without disclosing a single effect size. BMY’s testing procedure and model SE are also undisclosed, so we work through the plausible procedures one at a time. Depending on the assumed testing procedure, the minimum observed 60 mg effect needed to pass the test is ~38–50 mL. The topline also allows the two doses’ observed results to be nearly identical, or even to numerically favor 120 mg, under one of the procedures we consider (a 60 mg-first graph). It does not by itself establish which dose has the larger true effect. This note interprets the clinical and statistical disclosure; it does not address valuation or positioning. Information is through the Oct 13 close.

What BMY disclosed

On October 12, BMY said 60 mg produced a statistically significant improvement in absolute FVC change at Week 52. It said 120 mg “slowed FVC decline” but missed significance. BMY described discontinuations due to adverse events (D/C) as “low and similar to placebo” in both arms. There were no point estimates, confidence intervals, p-values, placebo decline, secondary endpoints or liver data.

The registrational cohort randomized 1,191 patients (~397/arm), and the statistical analysis plan is not public. The baseline data cover all 1,251 treated patients, including a 60-patient safety cohort: median age 71, FVC 78.9% and DLCO 51.9% of predicted, and 64.1% on background antifibrotics. The abstract describes “on average, moderately impaired lung function” in a population “similar to other recent phase 3 trials.” That is an established IPF population, not early disease. Prior treatment histories are not disclosed.

LPA1 signaling drives fibroblast recruitment and vascular leak after lung injury, a credible antifibrotic rationale. An earlier LPA1 antagonist, BMS-986020, reached Ph2 but was stopped for cholecystitis and dose-related liver enzyme elevations, which BMS authors hypothesize reflects off-target transporter inhibition by the compound, not LPA1 antagonism, based on retrospective nonclinical work. Admilparant’s Ph2 described it as safe and well tolerated with no clinically meaningful laboratory findings, and BMY’s Oct 12 release did not mention liver safety.

We use FIBRONEER-IPF as the design template: a Week 52 MMRM with a death penalty and an effective SD of ~275 mL calibrated from its published CIs, which gives a contrast SE of ~19.5 mL per dose. Every threshold below is a proxy built on that template, not a back-calculation of BMY’s result.

The doses share one placebo arm, so a direct 60-vs-120 contrast has the same ~19.5 mL SE and would need an observed gap of ~38 mL to clear a nominal 0.05. ALOFT does not appear to have prespecified that test; treat ~38 mL as a reference point, not a hurdle the trial set.

The testing procedure decides what the labels mean

Without the testing graph, “60 mg won, 120 mg failed” has no fixed quantitative meaning.

The bar for a 60-only win depends on how BMY split alpha between the doses. FIBRONEER started its higher dose at α=0.04 and its lower dose at α=0.01, recycling alpha after each success.

In plain terms: if the plan favored the high dose, the low dose faced a harder first test, and once it passed, the high dose got a second chance under an easier test. If ALOFT copied that graph and prioritized 120 mg, 60 mg had to clear its initial α=0.01 test, which needs an observed effect of ~50 mL. Once 60 mg clears its test, that alpha passes to 120 mg, which is retested at α≈0.05. The 120 mg miss therefore means an observed effect below ~38 mL, not the ~40 mL its initial α=0.04 would imply; a 120 mg result of ~38–40 mL would still count as a win.

Under other procedures the bars are lower and closer together (Exhibit 1). The smallest observed gap consistent with the topline ranges from ~12 mL to ~−2 mL. If BMY prioritized 60 mg, 120 mg could have been numerically larger by up to ~2 mL and still “failed”, because 120 mg’s recycled alpha is only ~0.03 under the recycling weight we take from FIBRONEER’s published graph. Every other procedure we consider requires 60 mg to be at least equal.

Exhibit 1

Of the procedures we consider, only a 60 mg-prioritized graph lets 120 mg be higher and still miss

Observed results consistent with “60 mg won, 120 mg missed”, in three of the six testing procedures we consider. Only when 60 mg is prioritized does the shaded region cross the 60 = 120 line (minimum gap −2.3 mL).

30 40 50 60 70 Observed 120 mg effect vs placebo (mL) FIBRONEER graph, 120 mg prioritized 60 mg ≥ 50.3 · 120 mg < 38.3 Min observed gap: +12.0 mL 30 40 50 60 70 Permitted results, not disclosed results Bonferroni 60 mg ≥ 43.7 · 120 mg < 43.7 Min observed gap: 0 mL 30 40 50 60 70 Permitted results, not disclosed results FIBRONEER graph, 60 mg prioritized 60 mg ≥ 40.1 · 120 mg < 42.4 Min observed gap: −2.3 mL 30 40 50 60 70 Permitted results, not disclosed results 60 = 120 60 = 120 60 = 120 120 mg numerically higher by up to 2.3 mL Observed 60 mg effect vs placebo (mL), same axes in every panel

Read with care. The axes show observed contrasts that a trial could report, built on an assumed effective SD of 275 mL and ~397 patients per arm. ALOFT’s testing procedure and model SE are undisclosed, so each panel is a scenario, not a reconstruction. No ALOFT effect size has been disclosed.

If BMY copied FIBRONEER’s graph with 120 mg first, a 60 mg win needs an observed effect of at least ~50.3 mL while the 120 mg miss caps its observed effect near 38.3 mL, a gap of at least 12 mL. Under Bonferroni the two bars meet at ~43.7 mL. Only when 60 mg is prioritized can the 120 mg arm be numerically higher, by up to ~2.3 mL, and still miss. The topline does not by itself rank the doses.

Panels show three of the six procedures in the note: the FIBRONEER graph with 120 mg first, Bonferroni, and the FIBRONEER graph with 60 mg first. Hochberg / Holm and the two fixed sequences are in the data table. Under fixed sequence 120 → 60, 60 mg is not tested if 120 mg fails, so “60 mg won, 120 mg missed” cannot occur. The 60 mg-first result rests on 120 mg’s recycled alpha being only ~0.03, the weight we take from FIBRONEER’s published graph; with full recycling the minimum gap would be +1.8 mL, not −2.3 mL.

View data
Procedure60 mg observed ≥ (mL)120 mg observed < (mL)Min observed gap (mL)
FIBRONEER graph, 120 mg prioritized50.338.3+12.0
Hochberg / Holm43.738.3+5.5
Bonferroni43.743.70
FIBRONEER graph, 60 mg prioritized40.142.4−2.3
Fixed sequence 60 → 12038.338.30
Fixed sequence 120 → 60Outcome impossible––

Observed contrasts; effective SD 275 mL, ~397 per arm.

Source: Clinaptis calculations (modeled); FIBRONEER-IPF supplement and protocol. Each row is a scenario, not a reconstruction. Thresholds scale with the SE: at an effective SD of 250 mL they fall by ~9%. ALOFT’s statistical analysis plan is not public. Not disclosed results.

Equal doses can produce this headline

In a hypothetical world where each dose truly delivers 45 mL, “60 mg wins, 120 mg fails” occurs 7% of the time under the FIBRONEER graph with 120 mg first, 17% under Bonferroni and 19% under the FIBRONEER graph with 60 mg first. These are probabilities of the headline pattern, not of the doses being equal. A split result needs no real dose difference. A true 20 mL gap (60 mg at 30–60 mL, 120 mg 20 mL lower) makes the pattern ~2–5x more likely than equal effects, so the split is modest evidence of a dose difference; the multiple rises with the true effect size. The split is much harder to reconcile with a large common effect: if both doses truly delivered ~88 mL (our illustrative unshrunk Ph2 bridge, not a disclosed result), it would occur ~0.4–1% of the time.

Selection alone then opens an apparent gap. Given the split and a true 45 mL for both doses, we expect published estimates of ~53–60 mL for 60 mg and ~29–32 mL for 120 mg, an apparent gap of ~22–31 mL (Exhibit 2). The 60 mg estimate would also run ~8–15 mL above its true value, because it was reported as the winner.

Exhibit 2

If the doses are equal, this split implies a 22 to 31 mL gap on average

Assume both doses truly deliver 45 mL and condition on the reported split (60 mg wins, 120 mg misses). The expected published estimates then differ by ~22 to 31 mL on average. Selection does it: 60 mg is reported because it won, 120 mg because it missed. Equal doses produce this split only 7 to 19% of the time.

0 20 40 60 80 True effect, both doses: 45 mL FIBRONEER graph, 120 mg prioritized P(split | equal doses) 7% ~31 mL gap (80% range 19 to 46) ~29 ~60 120 mg 60 mg Bonferroni P(split | equal doses) 17% ~23 mL gap (80% range 9 to 40) ~32 ~56 FIBRONEER graph, 60 mg prioritized P(split | equal doses) 19% ~22 mL gap (80% range 7 to 39) ~31 ~53 Expected published effect vs placebo (mL), given the reported split

Read with care. This is a scenario, not a forecast of BMY’s numbers: both true effects are set to 45 mL, shared-placebo correlation 0.5, effective SD 275 mL. It shows what the published estimates would look like conditional on the split, not what ALOFT found. Gaps are computed from unrounded values, so a gap can differ by 1 mL from the difference of the rounded estimates.

Under the FIBRONEER graph with 120 mg prioritized, the expected published pair is ~60 and ~29 mL, a ~31 mL gap on average, with equal true effects. Bonferroni gives ~56 and ~32 mL, and the graph that prioritizes 60 mg gives ~53 and ~31 mL. In every case the 60 mg estimate sits above the true 45 mL and the 120 mg estimate below it. Conditional on the split, a 20 to 30 mL gap is what equal doses would be expected to produce, although it would still add some evidence for a real difference.

View data
Procedure60 mg observed (mL)120 mg observed (mL)Gap, mean (80% range)P(split | equal doses)
FIBRONEER graph, 120 mg prioritized~60~29~31 (19 to 46)7%
Bonferroni~56~32~23 (9 to 40)17%
FIBRONEER graph, 60 mg prioritized~53~31~22 (7 to 39)19%

Expected published estimates if both doses truly deliver 45 mL, given the reported split. P(split | equal doses): chance that equal 45 mL doses produce “60 mg wins, 120 mg fails” under that procedure.

Source: Clinaptis calculations (modeled). Bivariate normal with shared-placebo correlation 0.5; effective SD 275 mL; ~397 per arm. Split probabilities are closed-form; conditional estimates use deterministic grid integration, no simulation. Scenario, not a forecast of BMY’s numbers. Not disclosed results.

Conditional on the split, a 20–30 mL gap at the medical meeting is what equal doses would be expected to produce, though it would still add some evidence for a difference, and more as the gap grows. Judging dose response needs both point estimates, their covariance or a direct 60-vs-120 contrast, and the sensitivity analyses, not the headline gap alone.

Ph2 efficacy and its limitations

The Ph2 randomized 276 IPF patients to placebo, 30 mg or 60 mg for 26 weeks. The 60 mg arm declined 1.4 ppFVC less than placebo (95% CI −0.1 to +3.0), about +45 mL at Week 26 and a 54% relative reduction. The 30 mg arm showed no benefit versus placebo. No patient received 120 mg. A separate PPF cohort of 123 patients showed a larger but less precise 60 mg effect of +3.2 ppFVC (95% CI 0.7 to 5.7).

The evidence is weaker than the headline. The IPF interval crosses zero, and the positive reading rests on a Bayesian primary analysis with a historical-trial prior. The model had no multiplicity adjustment, and the progression hazard ratio of 0.54 is post hoc. The observed completer gap does not widen cleanly (+59 mL at Week 4, +27 to +42 mL through Week 16, +58 mL at Week 26, ~70 patients/arm). The 30 mg arm tracked 60 mg through Week 16 (−41 vs −46 mL) before separating at Week 26 (−117 mL), so the 30 mg null rests on the last visits.

Earlier IPF programs show that Ph2 effects shrink by very different amounts. Nerandomilast 18 mg kept ~78% of its Week 12 Ph2 gap at Week 52 in FIBRONEER-IPF, nintedanib kept ~85% from TOMORROW to INPULSIS, and pamrevlumab kept ~39% from PRAISE to ZEPHYRUS-1. Ziritaxestat’s early signal vanished in ISABELA. This is a selected case series with different durations, endpoints and populations, not a fitted prior, so it justifies skepticism but not a specific haircut. Ph2 supports a positive 60 mg effect of uncertain size. That is why our prior below is ~45 mL and not the unshrunk ~88 mL, which applies the Ph2 relative effect to a ~170 mL placebo decline.

An illustrative two-dose sensitivity analysis for 60 mg

We combine a Ph2-based prior with both disclosed outcomes. This is a sensitivity analysis, conditional on assumptions BMY’s data could overturn. Exhibit 3 shows how far the answer moves with them.

Our prior comes from the Ph2 treatment-policy analysis (+1.4 ppFVC above), which keeps data after dose reduction. The while-on-treatment analysis, the figure BMY’s ALOFT release cites, was larger at +1.8 ppFVC (95% CI 0.2 to 3.4), a 62% reduction. We anchor on treatment-policy because ALOFT’s primary estimand is more likely to retain data after dose changes; that is unverified. We then shrink the Ph2 effect to Week 52, which gives a prior of ~45 mL. That matches the Ph2 Week 26 mL estimate only by coincidence: it is the Ph2 relative effect projected onto a ~170 mL Week 52 placebo decline and then shrunk toward zero. We apply it to both doses because 120 mg has no efficacy data, and we vary the correlation between the doses’ true effects from 0.5 to 0.95.

Conditioning on the reported 60 mg win alone lifts our model estimate of the underlying 60 mg effect to ~69–75 mL. Adding the 120 mg miss brings it down to ~44–68 mL (Exhibit 3), because with correlated doses a 120 mg miss is also evidence against a large 60 mg effect. That span covers model means across correlation settings and procedures. Each row assumes a different similarity between the doses, and the topline does not estimate it. At a correlation of 0.8 it is ~52–58 mL. The dose gap moves mostly with the correlation we assume, so the model does not resolve it. A dose-specific efficacy or exposure prior could move the 60 mg estimate further.

Exhibit 3

Our model estimate of the underlying 60 mg effect runs ~44 to 68 mL, depending on how alike we assume the doses are

A two-dose sensitivity analysis (illustrative), conditioned on the reported split (60 mg wins, 120 mg misses). The assumed correlation between the doses’ true effects lifts the 60 mg estimate from ~44 to 50 mL (correlation 0.95) to ~61 to 68 mL (0.50) and widens the dose gap from ~8 to 10 mL to ~40 to 47 mL. A model estimate, not a disclosed ALOFT effect.

20 40 60 80 Jascayd 9 mg: +48 mL cross-trial reference Jascayd 18 mg: +64 mL not a validated bar Dose gap, mean (80% range), mL P(60 mg ≥ 64 mL) Correlation 0.95 doses nearly alike ~44 to 50 60 mg ~36 to 40 120 mg ~8 to 10 (−7 to 24) ~14–21% Correlation 0.80 in between ~52 to 58 ~28 to 32 ~23 to 28 (−2 to 53) ~28–39% Correlation 0.50 largely independent ~61 to 68 ~20 to 24 ~40 to 47 (5 to 83) ~44–56% Modeled mean true effect vs placebo (mL), sensitivity analysis

Read with care. This is our model, not ALOFT data. It assumes an exchangeable prior across doses (Ph2 treatment-policy shrinkage, mean ~45 mL, SD ~39 mL), effective SD 275 mL and a FIBRONEER-type estimand. The dose correlation is our assumption and the topline does not resolve it: 0.95 = doses nearly alike; 0.50 = largely independent.

The Jascayd lines are a cross-trial reference, not a validated bar: FIBRONEER’s estimand, population and background mix differ from ALOFT’s. Dose gap column: range of mean gaps across the five procedures, with the 80% range of the gap in brackets. P column: modeled probability that the true 60 mg effect is at least 64 mL.

Conditioning on the 60 mg win and the 120 mg miss, our illustrative 60 mg estimate is ~44 to 68 mL across correlation settings and procedures; at 0.80 it is ~52 to 58 mL. Both estimates move with it: the 60 mg estimate rises from ~44 to 50 mL to ~61 to 68 mL, and the dose gap grows from ~8 to 10 mL when the doses are assumed nearly alike to ~40 to 47 mL when they are assumed largely independent. The topline does not tell us which assumption is right. The modeled probability that the true 60 mg effect is at least 64 mL rises from ~14 to 21% to ~44 to 56% on the same assumption.

View data
Dose correlation60 mg mean (mL)120 mg mean (mL)Gap, mean (80% range)P(60 mg ≥ 64 mL)
0.95~44 to 50~36 to 40~8 to 10 (−7 to 24)~14–21%
0.80~52 to 58~28 to 32~23 to 28 (−2 to 53)~28–39%
0.50~61 to 68~20 to 24~40 to 47 (5 to 83)~44–56%

Ranges span five testing procedures. Deterministic grid; no simulation.

Source: Clinaptis calculations (sensitivity analysis); Jascayd USPI (9 mg +48 mL, 95% CI 10 to 86; 18 mg +64 mL, 95% CI 25 to 102; placebo −170 mL); Ph2 treatment-policy analysis. Not disclosed results.

Three different bars

These numbers answer three different questions. None is a validated minimum clinically important between-group effect.

  1. Statistical bar. Clearing ALOFT’s test needs an observed 60 mg effect of at least ~38–50 mL, depending on the procedure above. That is evidence against no effect, not clinical value.
  2. Patient-level scale. At a median baseline FVC of ~2,710 mL and ~79% predicted, 1% point of predicted FVC is ~34 mL. A 40–60 mL between-group average is therefore ~1.2–1.7% points, and our illustrative 44–68 mL range is ~1.3–2.0% points. Both are model ranges, not disclosed results. An older IPF study estimated that a 2–6% point decline in predicted FVC within a patient can be clinically important. A smaller average difference could still matter if it persists and prevents large individual declines, but ALOFT’s progression and longer-term results must test that.
  3. Cross-trial reference. The Jascayd label shows +48 mL for 9 mg (95% CI 10–86) and +64 mL for 18 mg (95% CI 25–102) against a −170 mL placebo decline. Neither is a validated minimum for differentiation, and FIBRONEER’s estimand, population and background mix differ from ALOFT’s. Our illustrative 60 mg estimates (~44–68 mL) overlap that range and do not clearly exceed it.

Practical usefulness depends on what ALOFT has not disclosed: the effect in the 64% on background antifibrotics, tolerability against GI-limited standard of care, the progression composite and the liver profile. A modest FVC effect with clean tolerability as an add-on could still earn a role; a larger effect with a hepatic signal might not.

What the 120 mg miss does and does not suggest

Statistical noise around the threshold is the simplest explanation, because Exhibits 1 and 2 show it fits the topline without any biology. Fit is not a ranking against a true dose difference, which the arm-level estimates must settle. Winner’s-curse inflation at 60 mg and attenuation at 120 mg are the same mechanism viewed arm by arm. A miss at 120 mg means it did not clear its assigned alpha; it does not exclude a positive true effect.

Deaths, dose reductions and missing data are separate mechanisms. A FIBRONEER-style estimand penalizes deaths, which lowers a dose’s modeled FVC directly. Dose reductions change delivered exposure, and missing data depend on the estimand and imputation. Each stays open until the arm-level tables, and any of them would make the equal-dose noise explanation incomplete.

The Oct 12 release gave no liver data, so the arm-level liver table is the safety item to watch. BMY said on Oct 1 that a limited number of Ph3 liver events were mostly mild, transient and linked to other conditions or concomitant medications. BMY’s D/C statement does not address deaths, Week 52 FVC completeness or how the estimand handled them.

We find no prospective efficacy evidence for non-monotonic pharmacology in the previously disclosed data: Ph2 never tested 120 mg for efficacy, 30 mg tracked 60 mg through Week 16, and the only dose-related signal was 120 mg’s dedicated syncope cohort. The extension protocol states that BMY will select one “most favorable dose” after the primary analysis. 60 mg is the likely candidate. For a 60 mg filing, the 120 mg read matters mainly for what it implies about the 60 mg estimate and the liver table.

The read-through to progressive pulmonary fibrosis (PPF) is partial. ALOFT-PPF is fully enrolled (~1,057 patients), uses the same 60 mg and 120 mg BID arms and the same Week 52 FVC primary endpoint, and BMY expects topline in early 2027. We infer the dose-split and multiplicity questions carry over, but IPF efficacy does not transfer automatically to a more heterogeneous population, and the PPF testing procedure is also undisclosed.

Read-through to CTNM

Observed. Contineum (CTNM), a sub-$400M market cap company developing the LPA1 antagonist PIPE-791, closed at $8.42/sh on Oct 13, ~37% below its $13.38/sh close on Oct 9, before the disclosure. Possible interpretation. We infer the market read the 120 mg miss as a dose-response or class concern; a small-cap price move does not establish that reading.

Scientific read-through. ALOFT is the first positive Ph3 for LPA1 antagonism, after autotaxin inhibition failed in ISABELA, which supports the mechanism. If the 60 mg effect lands in our illustrative range, the efficacy bar for PIPE-791’s Ph2 PROPEL-IPF rises, a conclusion that inherits all the uncertainty in Exhibit 3. ALOFT says nothing directly about PIPE-791’s efficacy, dose response or liver safety, given its different scaffold and 10 mg QD dose.

What would change our view

  1. 60 mg point estimate and CI. Our 44–68 mL range covers modeled mean true effects. The reported number will also carry winner’s-curse inflation of ~8–15 mL. A reported estimate of ~64 mL or more, the Jascayd 18 mg label effect (cross-trial), would exceed the ~53–60 mL that equal 45 mL doses are expected to publish after selection (Exhibit 2), and would move us toward a larger true effect. An estimate of ~50–60 mL is compatible with equal doses and with a real 60 mg advantage, and cannot separate them without the 120 mg estimate. An estimate of ≤45 mL is possible only if ALOFT’s threshold sat at the low end of our 38–50 mL range, and would put it at or below the bottom of our modeled range. The case would then rest on progression, tolerability and add-on data, not FVC magnitude.
  2. 120 mg point estimate and the 60-vs-120 gap. A gap inside the ~22–31 mL that equal doses are expected to produce is compatible with equal doses, but it still adds some evidence for a real difference, and more as it grows. A gap above ~40–46 mL, the top of the equal-dose 80% ranges, or a 120 mg estimate near zero, is hard to reconcile with equal 45 mL doses and moves us toward a real dose difference, which raises the weight on 120 mg tolerability and the liver table. These are model ranges, not tests.
  3. The testing graph and model SE. They collapse Exhibit 1 to one row.
  4. Deaths, D/C, Week 52 completeness and the liver table, by arm. These test the missingness question, inform the filing dose and shape the label.
  5. Placebo decline, background-therapy strata and the progression composite. These determine clinical relevance as an add-on.

Hold list: ALOFT SAP, estimand and model SE; Corte 2025 adjusted Week 26 placebo change (−104 mL, unverified). All ALOFT effect figures in this note are Clinaptis model estimates, not disclosed results.

Sources