The Border That Moves With the Question

I.The First Known Thing — the line that is not there

thermodynamics · the supercritical fluid

Below its critical point a fluid can be liquid or gas, and the line between them is real: cross it and the density jumps, latent heat is taken up, and a meniscus can be seen. The line ends at the critical point. Beyond it, at higher temperature and pressure, there is no coexistence, no jump and no meniscus; a fluid can be carried from a state that behaves like a liquid to one that behaves like a gas without passing through any discontinuity at all. This is textbook thermodynamics and it has been known since Andrews's experiments on carbon dioxide in the 1860s. For most of the time since, the supercritical region was treated as a single undifferentiated phase — one fluid, varying smoothly.

yet something still happens there

It does not behave uniformly. Along any path through the supercritical region the response functions — the isobaric heat capacity, the isothermal compressibility, the coefficient of thermal expansion, the correlation length of density fluctuations — rise to a maximum and fall again. Near the critical point these maxima lie almost on top of one another, and the line traced by the maximum of the correlation length, which continues the coexistence line into the region where there is nothing to coexist, was given a name in 2005 by the group around H. Eugene Stanley: the Widom line, after Benjamin Widom. In 2010 Simeoni and colleagues, using inelastic X-ray scattering on supercritical argon, showed that the Widom line separates regions in which sound propagates in a liquid-like and a gas-like manner, which made the line look like a boundary after all — a phase in all but name.

and then the line comes apart

The part that matters here came next. Brazhkin and co-workers, working through the Lennard-Jones fluid in 2011, followed the maxima of the separate response functions away from the critical point and found that they do not stay together. The line of heat-capacity maxima, the line of compressibility maxima and the line of thermal-expansion maxima fan out as the distance from the critical point grows, and further out the maxima flatten and disappear. A single Widom line exists only in a narrow neighbourhood of the critical point; beyond it there is a family of lines, one per observable, and then no lines. Later work, including a 2024 analysis of thermodynamic crossovers across many fluids, has treated the region as a set of crossovers rather than a boundary. The discipline's operational habit follows: whether a supercritical boundary is meaningful is judged by whether the observables agree on where it is. Where they agree, a transition is near and the line may be used; where they disagree, the line is a convention of whichever quantity was chosen, and the size of the disagreement says how far the system is from anything that deserves to be called a change of state.

II.The Second Known Thing — the disease that is a number

drug discovery · disease definition, indication choice, endpoint selection

Walk to a hall that decides what is to be treated. A large share of the indications a drug can be developed for are not defined by a lesion or an organism but by a threshold on a continuous measurement: type 2 diabetes by fasting glucose or glycated haemoglobin, hypertension by blood pressure, chronic kidney disease by estimated glomerular filtration rate and albuminuria, osteoporosis by a bone-density T-score, hypercholesterolaemia and obesity by a lipid level and a body-mass index. Every one of these indications is a line drawn across a distribution with no gap in it. The line decides who is enrolled in a trial, what the label says, and what the market is.

where the lines were drawn, and by what

The lines were drawn by particular outcomes. The glucose thresholds for diabetes adopted in 1997 and 1999 were fixed where the prevalence of diabetic retinopathy begins to climb in the Pima, Egyptian and American survey data, and the International Expert Committee's recommendation of a glycated haemoglobin of 6.5 per cent in 2009 was argued on the same ground. Retinopathy has a bend; the curve relating glucose to cardiovascular risk, measured in the same populations, has no comparable bend at that level and rises continuously through it. Blood pressure is the extreme case. The Prospective Studies Collaboration's analysis of a million adults in 2002 found vascular mortality log-linear in blood pressure down to at least 115/75 mmHg, with no threshold anywhere in the range studied, and the line that defines hypertension has been moved several times since without any change in the underlying curve. In kidney disease, plotted separately for death, cardiovascular death and kidney failure, the hazard curves across filtration rate do not bend at the same place.

and the thing the threshold is taken to mean

None of this is unknown, and clinicians routinely say that the thresholds are conventions. But the conventions are used as if they were classes. A trial enrols above the line, powers on an endpoint, and reports an effect in the disease; a program is described as being in an indication; and when an intervention reduces one outcome and not another — as intensive glucose lowering in the large diabetes trials reduced microvascular endpoints more convincingly than macrovascular ones — the discrepancy is attributed to the drug, the duration or the patients. What the field does not have is a quantity answering the question a physicist would ask first of a line drawn across a smooth region: do the outcomes that matter agree on where it is? — nor any suggestion that the answer, if measured, might be different for different indications and might predict how their trials behave.

III.The Bridge

no diagnostic-criteria paper cites the Widom line

The halls share nothing. The physics of supercritical fluids and the setting of diagnostic thresholds have no common journals, no shared citations and no review that spans them; where physics has entered medicine through the idea of transitions, it has done so through tipping points and early-warning signals in the course of an illness, which concern a system moving through time towards a genuine change of state — the opposite of the situation here. The nearest kin on the medical side is taxometrics, the family of methods from Paul Meehl's school that asks whether a psychiatric construct is a category or a dimension, and it is named below as one of the things this candidate may turn out to be. What the two halls have in common is a situation. Both have a continuous state and a practical need for a boundary across it. Both place the boundary where some response is steepest. One of the two halls has learned that the place depends on which response is chosen, and uses the disagreement as a measurement.

X: the disagreement between boundaries is the measurement

Call it crossover spread. Where a state varies continuously without a discontinuity, any boundary drawn through it is the locus of maximal response of some chosen observable, and different observables place that locus in different positions; their agreement is the signature of a genuine transition nearby, and their disagreement grows with the distance from one. So the spread of the boundaries that different observables draw is itself a measurement of whether a category exists, and a threshold fixed by the bend in one observable carries no claim at all about the others. Two consequences are worth stating exactly, because they are not the ordinary caution that thresholds are arbitrary. First, arbitrariness is not uniform: on this account some threshold-defined indications sit close to something like a transition, where the outcomes agree and the line is nearly a class, and others sit far from anything, where the line is purely the property of the outcome that happened to set it; and the difference is computable from cohort data before a trial is designed. Second, the endpoint and the indication are not independent choices. If the indication's line was drawn by the bend of one outcome, a trial enrolled at that line and read on a different outcome is enrolling on one observable's boundary and measuring another's, and in a wide-spread indication the effect it reports will depend on that pairing more than on the drug.

IV.The Question That Becomes Possible

neither discipline would ask this alone

The clinical hall asks where a threshold should be and debates each one separately, outcome by outcome and committee by committee. The physical hall has never looked at a patient. Neither asks whether the agreement between outcomes on where a disease begins is a quantity that varies from one indication to the next, and whether that quantity predicts what happens when a drug is tested in the indication.

the crossover-spread hypothesis

Does a threshold-defined indication have a measurable crossover spread — the dispersion, along its defining biomarker, of the points at which each clinically relevant outcome's hazard bends most sharply — such that indications with a narrow spread behave in trials as classes, with treatment effects that step at the threshold and agree across endpoints, while indications with a wide or undefined spread behave as crossovers, with effects that change smoothly with the enrolment cut-off and depend more on which endpoint was chosen than on where the line was drawn?

why it would matter

Because indication choice and endpoint choice are among the most consequential decisions a program makes and are currently made as though every threshold-defined indication were equally a disease. If the spread is real and predictive, a narrow-spread indication is one in which enrichment at the threshold buys something and a composite of several outcomes is coherent; a wide-spread indication is one in which the enrolment cut-off is a free parameter to be optimised for the chosen endpoint, a composite endpoint mixes outcomes whose boundaries lie in different places, and a claim of efficacy in the disease is a claim about one observable's version of it. It would also give a principled reading of the familiar discordances — the drug that moves the outcome that set the line and not the outcome that matters commercially — as predictable from the geometry of the indication rather than as surprises about the drug. And it bears on where new lines are drawn: when an indication is being defined for a novel biomarker, the spread can be estimated first, and a biomarker whose outcomes disagree widely is one on which no single threshold will define a population that behaves as one.

V.How It Could Be Proven — or Broken

a candidate must be breakable, or it is merely poetry

An intercognix candidate earns nothing until it names its own test. This one needs no new experiment for its first pass: the cohort consortia and the trial record already hold both halves of what it makes a claim about.

the honest status

Note the narrowness. The claim is about indications defined by a threshold on one continuous biomarker, and about one statistic of their outcome curves; it says nothing about diseases defined by pathology, by an organism or by a syndrome, and it does not say that any threshold is wrong. It proposes no new assay and no revision of any guideline: only that, for each such indication, the outcome curves be laid side by side and the dispersion of their bends be computed and tested against the trial record. It is also worth saying what the import does not carry. A fluid is a system of identical particles in equilibrium, whose response functions are exact derivatives of one free energy; a population of patients is none of those things, and what transfers is a practice — reading the agreement of observables as the test of a boundary — and not the thermodynamics that explains why the practice works in a fluid. The whole thing rests on an assumption that is checkable and may fail at once: that outcome curves in real cohorts bend enough, and are estimated well enough, for their bends to have locations at all, in which case there is nothing to spread and the question does not arise. Until the curves are fitted and the trials are matched, this remains exactly what its stamp says: a candidate. A structural echo between two halls of the library, waiting for someone with the tools of both to walk the corridor.