J.Konstapel,Leiden, 3-11-2026.

Het probleem: Veel geneesmiddelen falen in een late fase van klinisch onderzoek, vaak met als reden “onvoldoende werkzaamheid”. Dat label verbergt echter dat een medicijn sommige patiënten kan helpen en anderen juist kan schaden, waardoor het gemiddelde effect nul is en niemand leert waarom.

De kern van de oplossing: Het artikel bepleit een meting vóór de start van de behandeling, nog voordat patiënten worden gerandomiseerd. In veel ziektes zou dit geen enkelvoudige waarde moeten zijn, maar een “herstelmaat” (return measure).

Het idee van “closing”: Ziekte wordt gezien als een pad dat “sluit” wanneer het lichaam terugkeert naar balans. Dit kun je alleen zien met ten minste twee metingen: eentje om te kijken wat er direct na een verstoring gebeurt (de “transient”) en eentje om te zien wat er blijvend veranderd is (het “record”).

De meetbare getallen: Deze aanpak levert twee concrete getallen op: de herstelfractie (hoeveel van de verstoring is opgelost tegen de tweede meting) en de hersteltijd (hoe lang het herstel duurde). Een enkel momentopname kan dit niet vaststellen.

Waarom dit werkt als stratificatiemiddel: Een geneesmiddel kan een enkele waarde direct omhoog of omlaag duwen, maar het kan niet rechtstreeks bepalen of een systeem wel of niet herstelt. Daarom is een herstelmaat een stratificatiemiddel dat een medicijn niet kan “manipuleren”.

Toepassing op bestaande medicijnen (repurposing): Een diagnose-onafhankelijke herstelmaat maakt zogenaamde mandtrials mogelijk, waarin patiënten met verschillende ziektes worden samengevoegd. Dit is essentieel voor het testen van oudere, niet-patent-geneesmiddelen, waarvoor vaak geen geld is voor aparte trials per ziekte.

Jump to the article with DOI: 10.13140/RG.2.2.20726.56648 here

Most drugs fail late, and the most common stated reason is “inadequate efficacy”.

But that label hides a problem.

If a drug helps one group of patients and harms another, the average result is zero.

The trial reads that as failure. Nobody learns why.

This essay argues that the fix already exists in outline.

We need a patient measurement taken before treatment starts.

And in many diseases, that measurement should not be a single level.

It should be a return — does the patient’s displaced biology bounce back, and how fast?

Two ideas that fit together

Idea 1: disease. A path of illness “closes” when the body returns to balance. Picture a fever. The acute response rises, then subsides. What remains afterwards is the immune memory. Closure is not a state. It is a path, and a path needs at least two measurements to see.

Idea 2: drug programmes. Programmes ask their questions backwards. They pick the molecule first. They pick the patients and the endpoint last. A six-gate procedure reverses that order. Its third gate is the sticking point: name, in advance, the measurable quantity that separates patients who benefit from patients who are harmed.

Programmes have nothing to put in that gate. That is not carelessness. It is a real gap in the field. The essay’s claim is narrow: the missing quantity is, in many diseases, a return measure — and it can be taken before anyone is randomised.

The measure, plainly

Take a disturbance at time zero. Watch two things.

1. The transient. What rises on contact and should subside. Activation markers, acute inflammation, effector cells.
2. The record. What stays changed afterwards. Memory cells, antibodies, retrained innate responses.

The path has closed when the transient returns to normal within a characteristic time, and the record keeps a stable trace. It has failed to close when the transient never settles.

Two numbers follow:

• Recovered fraction — how much of the disturbance has resolved by the second measurement.
• Return time — how long the recovery took.

Neither can be read from a single snapshot. That is the whole point. A drug can push a single level up or down directly. A drug cannot directly set whether a circuit returns. So a return measure is a stratifier a molecule cannot game.

Axis and direction

Gate 3 needs two things, from two places.

• The axis comes from the measure. Divide patients into “returns” and “does not return”. Do it before allocation.
• The direction comes from the drug’s mode of action. A drug that suppresses a stuck inflammation helps the non-returning side. A drug that supports a process still running its course helps the returning side — and may harm the other side by adding to a disturbance that was already failing to resolve.

This explains a puzzle. The same stratification can point in opposite directions in two diseases. The axis belongs to the patient. The sign belongs to the agent. Both are needed.

Five trials, re-read

Four of the five were divisions on whether a displaced quantity returns. None was designed that way.

• CANTOS (canakinumab after heart attack). Entry was by a single hsCRP level. The strong mortality signal appeared only when patients were divided by hsCRP under treatment — a second measurement against the first. A two-point measure carried the signal.
• HARP-2 (simvastatin in ARDS). Filed as negative. Reanalysis found a survival benefit in the hyperinflammatory group. Sustained, non-resolving inflammation is non-closure. But the division used only one timepoint.
• AYAME (bardoxolone in diabetic kidney disease). Its endpoint was the quantity the drug moves directly. The trial measured its own mechanism. A return measure is immune to that failure by design.
• SAVE-MORE (anakinra in COVID-19). Patients were selected by plasma suPAR before randomisation. It worked (odds ratio 0.36). One timepoint only. It is the closest thing to the proposed design.
• BEACON (bardoxolone in stage 4 kidney disease). Stopped for harm. Baseline predictors of harm existed in its own data. Reading them as “no reserve left to return a load” is interpretation, not demonstration.

The gap in current practice

Two fields each hold half of the proposal.

• ARDS research has a bedside assay that classifies patients into inflammatory subtypes before treatment. It reads one timepoint.
• Oncology uses ctDNA kinetics — return-like measures with strong discrimination (AUC 0.92–0.96, in simulated cohorts). But only after treatment has started.

Nobody uses a two-point return measure before randomisation as the stratifier. That empty position is the essay’s contribution — and its vulnerability.

Why it matters for repurposed drugs

Fewer than a quarter of recognised diseases have an approved treatment. Many are served, if at all, by off-label drugs. The best repurposed drugs act on junctions shared by many diseases — PI3K-AKT-mTOR, RAS-MAPK, JAK-STAT. Sirolimus alone is being tried in nine conditions. Each population is too small for its own trial. No sponsor profits from an off-patent drug.

A return measure is diagnosis-independent. It asks does this patient’s biology return?, not what disease is this?. So a basket trial can pool several diagnoses, divide them on one axis, and spend one statistical allowance. That is the shape a junction drug needs and cannot currently get.

What would refute it

1. The two-point measure adds nothing. Compare it against the published one-point classification on the same HARP-2 data. If it does not separate treatment effects better, the proposal is redundant. This test needs no new patients.
2. The axis does not transfer across diagnoses. Then the basket argument collapses.
3. Someone already occupies the position. Two fields were checked. Many were not.

Honest limits

• No return times or recovered fractions have actual numbers yet. The quantities are defined, not measured.
• The BEACON reading is interpretation.
• The oncology discrimination figures come from simulated cohorts.
• Only two fields were searched for prior art.
• The repurposing candidates are presented by their submitters as candidates, not proven treatments.

The proposal changes how trials are designed. It does not claim any particular drug works. A stratified trial that fails still fails — but it fails legibly. The knowledge survives. That is what “inadequate efficacy” failures never gave us.

---

Annotated reference list

1. Konstapel, J., The Right Order of Questions in a Drug R&D Programme, constable.blog, 30 October 2026.
   Source of the six-gate procedure and all attrition figures: 7.9% of phase 1 entrants reach approval; 28.9% phase-2-to-3 success; 344 of 640 pivotal entrants failed; 40.1% of failures published; phase 3 costs about $255.4M; preselection markers used in 6.03% of transitions, with 15.9% versus 7.6% approval. Annotation: the backbone of the programme line. The author himself notes the figures are registry associations and post-hoc analyses, not demonstrations.
2. Reddy et al., Development and validation of a point-of-care assay to identify ARDS subphenotypes.
   A bedside plasma assay (IL-6, soluble TNF receptor-1, vasopressor use) classifying ARDS patients in 36 minutes from one sample. Annotation: proves the first half of the gap — pre-treatment stratification is technically ready, but reads one timepoint. A development-and-validation report, not an outcome trial; prospective subphenotyping has not yet been done.
3. Early ctDNA kinetics as a dynamic biomarker of cancer treatment response, bioRxiv.
   Six samples per patient around treatment start; reported AUC 0.92 (targeted therapy) and 0.96 (radiotherapy). Annotation: proves the second half of the gap — kinetic return measures perform well, but only prognostically, after treatment begins. The discrimination figures come from simulated cohorts, the weakest evidence point in the essay.
4. Lancet Respiratory Medicine, ARDS subphenotypes and differential response to simvastatin: secondary analysis of HARP-2.
   The reanalysis finding survival benefit (P=0.008) in the hyperinflammatory class of a trial filed as negative. Annotation: the benchmark for the proposed zero-cost test in Appendix C of the source. A secondary analysis, with the usual post-hoc caveats.
5. Every Cure ROADMAP, Organizations repurposing Sirolimus.
   Nine conditions named across organisations, up to seventeen organisations involved. Annotation: a survey of what is being attempted, not what has been shown to work. Source of the sirolimus table in the essay’s Appendix B.
6. Every Cure, We Responded to the FDA’s Request for Information on Drug Repurposing, 2026.
   Fewer than 25% of recognised diseases have an approved treatment, against 22,701 diseases scored for unmet need. Annotation: an advocacy submission. Its scale figures are usable; its drug-disease pairs are explicitly candidates, not demonstrated results. Frames the regulatory-barrier argument.
7. EMBO Molecular Medicine 2025, Vascular malformations: from genetics to therapeutics.
   Evidence tiers for sirolimus, alpelisib, trametinib and miransertib against PI3K and RAS-MAPK pathways. Annotation: source of the junction table. It does not state approval status, which is why that column is absent.
8. Pediatric Rheumatology 2023, Efficacy and safety of baricitinib in Japanese patients with autoinflammatory type I interferonopathies.
   Nine patients; diary scores fell in NNS/CANDLE and SAVI, rose slightly in AGS; steroid use fell; all nine had adverse events. Annotation: the one repurposing entry with a published outcome set. Small, open-label, no comparator — the authors themselves call for further study.
9. Konstapel, J., prior programme papers: Protection Without a Component (Oct 2026); Closure in the Pair (Nov 2026); Closure, Carry and the Ladder; Where Illness Comes From (8th ed.); Curves in the Vacuum.Net; and the Vacuum.Net Theory Foundational Paper (9th ed.).
   Annotation: the disease line — the operational definition of closure and the positional vocabulary. The essay states that its argument does not require accepting this framework; only one constraint (closure is a path, not a point) does the work.
10. Named but not read: a Journal of Virology 2009 study on H5N1 pathogenicity in chickens, and an Avian Pathology 2023 study on a clade 2.3.4.4b H5N1 virus in mallards. Annotation: listed by the author to make the gap visible rather than hidden — they bear on the disease line and could not be retrieved.

Article