Why medicine needs a shared language

Key points

• Two clinicians can look at the same image, agree on every pixel, and still disagree about whether a patient has "progressed." The disagreement is about what the word names, not about what the image shows. Separating the term from the concept it names is the first act of a shared language.
• The story of Babel preserves an old intuition about communication: a common undertaking does not need to be destroyed to fail. It only needs its words to stop meeting. The story is a narrative, not evidence.
• The first Greek philosophers turned that intuition into questions that are still open: what is held in common, what it means to say that something is, and whether names fit things by nature or by agreement. Aristotle answered with tools: categories, and definition by genus and difference.
• Ontology asks what kinds of things there are and how they relate. Epistemology asks what warrants a claim to know. Medicine confuses them constantly: "what is this finding?" and "how sure are we?" are different questions that one word, or one confidence score, cannot answer at once.
• Back in the reading room, the remedy is not that one reader wins. It is that the report says which concept it means, under which criterion, and in what relation to the findings it rests on.

A disagreement about one word

The following scene is a synthetic illustration. The patient, the clinicians, and the images are invented, and nothing here is a treatment recommendation.

A tumor board is reviewing a man who has received two cycles of a radioligand therapy for a cancer that expresses the target the drug binds. His follow-up PET/CT has just come back. The medical oncologist, call her Dr. Amari, opens the discussion with a single sentence: "He has progressed." The nuclear medicine physician who read the study, call him Dr. Lindqvist, answers immediately: "He has not."

They are looking at the same screen. Neither disputes what is on it. The two largest sites of disease have shrunk, and their tracer uptake has fallen. Two small new foci of uptake have appeared in bone that were not visible on the baseline study. A serum marker, measured the week before, has risen modestly.

Dr. Amari means something precise by "progression." In the response framework she uses in trial reports, the appearance of a new lesion is, by definition, progressive disease, whatever happens to the lesions that were already there. Dr. Lindqvist means something equally precise and equally defensible. In the reading conventions of his discipline, the dominant disease has responded, small new foci early in a course of therapy may represent disease that was present but below detection at baseline, and whether they count as "new lesions" depends on a threshold his colleague never stated. A urologist at the table has a third meaning in mind: for him, progression is a clinical judgment combining symptoms, the marker, and the images, and the marker is what worries him.

The frustrating thing about this conversation is that it can go on for twenty minutes without anyone being wrong. Published response criteria exist precisely because the word is ambiguous; the anatomic criteria known as RECIST and the PET-based criteria known as PERCIST were each written to fix the meaning of "response" and "progression" for a stated purpose, and they do not fix it the same way (Eisenhauer et al., 2009; Wahl et al., 2009). Each clinician is using the word correctly inside a community that has settled on a meaning. The failure is that the communities are sitting at the same table.

Notice what would resolve it. Not better images. Not more data. A sentence such as: "By anatomic criteria, this is progression; by the PET-based criteria we agreed to use for this therapy, it is a partial response with two indeterminate new foci; clinically, the marker rise is concerning and we should repeat it." That sentence does not pick a winner. It separates one word into three concepts and names the criterion behind each.

A word is a term. What the word points at, within a community that has agreed how to use it, is a Concept. The term "progression" is one string of letters. The concepts it was carrying around that table were at least three. Most of what follows in this essay, and most of what NucLex is for, comes down to taking that distinction seriously.

Babel, read as a story about communication

The oldest account of this failure in the Western tradition is short enough to tell in a breath. In the eleventh chapter of Genesis, the whole earth has one language. People settle on a plain, learn to bake brick, and resolve to build a city with a tower reaching the sky. Then their speech is confused, each can no longer understand the other, the building stops, and they are scattered across the earth.

It is worth being clear about what kind of thing this is. It is a narrative, and this essay reads it as a narrative about communication. It is not evidence for any claim about the history of languages or anything else empirical, and it is not treated here as such.

Read as a story, the striking thing is the order of events. The tower does not fall. The language is confused, and the building simply ceases. Once the workers cannot agree on what "brick" or "higher" or "here" means, the project loses its ground. The scattering is not a punishment added to the confusion; it is the confusion, seen from outside.

Umberto Eco traced the long European preoccupation with recovering what Babel supposedly lost, from speculation about the language of Adam to the invented philosophical languages of the seventeenth century (Eco, 1995). What those projects share with the story that inspired them is a sense that a common language is not a given. It can be lost, and when it is lost, collective undertakings fail quietly.

Medicine is one of the most demanding of such undertakings. A physician cannot treat alone. The findings of one examiner must be legible to the next; a dose prescribed by one person is administered by a second and recorded for a third; a trial pools reports from sites that never meet. When words stop pointing reliably at the same thing, medicine does not collapse dramatically. Like the tower, it stops being built where the words no longer meet. The tumor board above was a small Babel: no drama, no scattering, just twenty minutes in which three competent people could not proceed.

The first questions

The Greek thinkers conventionally grouped as pre-Socratic inherited the intuition that shared speech is fragile and tried to think it through. What survives of them is fragmentary, often quoted by later authors with their own agendas, and any summary should be read with that caution; the standard scholarly collection is Kirk, Raven, and Schofield (1983).

Heraclitus of Ephesus, writing around the turn of the fifth century BCE, is remembered for the river that is never the same twice. The fragments are at least as concerned with something he called the logos: an account, a gathering, a saying that holds in common. One repeated complaint, in paraphrase, is that although the logos is common, most people live as if they had a private understanding. The three clinicians at the tumor board were each inside a private understanding that felt, from the inside, like the common one.

Parmenides of Elea went in another direction and arrived at the same door. His poem asks what it can even mean to say that something is, and treats the blurring of "is" and "is not" as the great error. Whatever one makes of his conclusions, the method is a discipline of reference: before you argue, know what your words are picking out.

Plato's Cratylus puts the question of names in its most explicit form. Are names correct by nature, so that a true name fits its thing the way a key fits a lock, or correct by agreement, because a community has settled on them? Socrates argues both sides and the dialogue ends without a verdict, which is part of its honesty. But it leaves behind a distinction every later theory of meaning has had to confront: the difference between a word that is merely used and a word that has been fixed, by nature or by decision, to a thing. "Progression" in the tumor board was a word merely used. RECIST and PERCIST are two different decisions to fix it.

Aristotle: categories and definition

Aristotle answered the question of names by building tools, and the tools are still in use.

Sometime in the fourth century BCE he set down a short treatise, the Categories, that would be read continuously for two thousand years. It opens with a distinction between things that share a name but not a definition (a man and a painted man are both called "animal" in name only) and things that share both. Then it proposes that whatever can be said of anything falls under a small number of heads: what it is (substance), how much, what sort, relative to what, where, when, position, state, doing, and being done to. Scholars have argued ever since about whether these are kinds of words, kinds of predication, or kinds of being (Ackrill's translation and notes, 1963, remain the standard English entry point).

Two ideas from this work matter here. The first is that substance is primary. This particular patient, this particular focus of uptake, is an individual; everything else is said of it or present in it. A size does not float free; it is the size of something. The second is that among the things said of an individual, some are said essentially. "Human" is said of Socrates in a way that "pale" is not: if Socrates stopped being human he would stop being Socrates, whereas he could tan. This is the seed of the idea that a definition states what a thing is rather than merely describing it.

The method of definition is worked out more fully in the Topics. A definition states the genus, the broader kind to which a thing belongs, and the differentia, the feature that marks it off from everything else in that genus: the human being as a rational animal. Aristotle also noticed, as any working classifier does, that real things strain against neat schemes. The method is a tool for clarity, not a promise of tidiness.

Six centuries later Porphyry's Isagoge, an introduction to the Categories for students, laid out a nested scheme that medieval readers turned into a diagram: substance divides into corporeal and incorporeal, body into living and non-living, and so on down to individuals (Barnes's translation, 2003). This Tree of Porphyry is, in modern terms, a single-inheritance hierarchy with a top node, and it is the ancestor of every Taxonomy in this publication's lexicon.

A caution is needed at exactly this point. It is tempting to say that Aristotle invented the knowledge graph, or that SNOMED CT is Porphyry's tree with a reasoner attached. The temptation should be resisted. Aristotle was asking what it is for a thing to be what it is. A modern formal ontology is a specification written so that software can check consequences, and its authors usually make deliberately modest commitments about what their classes are "really" like. The lineage of technique is real: essential predication, definition by genus and difference, nested kinds. The identity of purpose is not. This essay does not attribute computational definitions to ancient authors.

Two questions that medicine keeps confusing

The word "ontology" has two lives, and both are needed here. In philosophy, Ontology is the study of what there is: what kinds of things exist, what makes something the kind of thing it is, and how kinds relate (Smith, 2004). In informatics, an ontology is an explicit specification of a shared conceptualization, in a phrase from Gruber (1993) that has been refined and argued over ever since (Guarino, Oberle, and Staab, 2009). The two senses overlap but are not interchangeable, and the monograph on Ontology treats the distinction at length. For this essay the shared core is enough: ontology answers the question "what kinds of things are we talking about, and how do they relate?"

Epistemology asks a different question: what warrants a claim to know? The classical answer, that knowledge is justified true belief, is at least as old as Plato and was famously shown to be incomplete by a three-page paper (Gettier, 1963) that offered cases of justified true belief that no one would call knowledge. The monograph on Epistemology follows that argument into science and the clinic. What matters here is the shape of the question: not "what is it?" but "how do we know, and how well?"

Return to the tumor board. "Is this focus of uptake a lesion?" is partly an ontological question. It asks what kind of thing the finding is: a site of disease, an area of physiologic uptake, an artifact of reconstruction, or a finding whose kind is not yet determined. A good answer depends on having a vocabulary in which those kinds are distinct and their relations stated. "How sure are we that it is disease?" is an epistemological question. It asks what Evidence supports the claim: the intensity of uptake, correlation with anatomy, prior imaging, the behavior of the marker, and the known performance of the tracer. A good answer depends on being able to say what the evidence is and how strong it is.

Medicine collapses these constantly, and the collapse is not harmless. A report that says "probable metastasis" has fused an ontological claim (this is a metastasis) with an epistemic qualifier (probable) in a way that lets neither be examined. A structured field that asks for "lesion: yes/no" has deleted the epistemic question altogether. A machine-learning score of 0.83 attached to a finding is sometimes read as a definition of what the finding is and sometimes as a measure of how sure the model is, and it is neither; its meaning depends on how the model was trained and what its training labels meant. A confidence number is not philosophical justification, and a label is not a definition. Keeping Uncertainty separate from kind is one of the quiet disciplines that a shared language makes possible.

What a concept does to what we notice

There is a further point that the philosophical tradition makes and that clinical practice confirms daily: concepts do not just name what we see. They shape what we are able to see at all.

Heidegger's early work argued that we do not first encounter a neutral world of objects and then paste labels onto it. We encounter a world already laid out as a field of involvement: a hammer shows up as for hammering, an image as for reading (Heidegger, 1962). Things appear as what they are for, within a practice, and language is not a layer above that world but the medium in which it is articulated and held open for others. One need not accept the whole of that philosophy to take its practical lesson, and this essay does not attempt to reproduce its vocabulary. The lesson is that a reader trained to look for "new lesions" literally sees the scan differently from a reader trained to look for "response in the dominant disease." Each is attending to what his concepts make salient.

Wittgenstein reached a related conclusion from a very different starting point: the meaning of a word is, for a large class of cases, its use within a shared practice, and a language that only one person could understand would not be a language at all (Wittgenstein, 2009). Meaning is held in common or it is not held. The private understanding Heraclitus complained of is not merely a failure to listen; it is not, strictly, an understanding of the shared word at all.

For the clinic, this means that a shared vocabulary is not an administrative convenience added after the medicine is done. It is part of what makes a finding a finding. Cimino's desiderata for medical vocabularies (1998) read, from this angle, like a practical answer to the Cratylus: concepts should be non-vague, non-ambiguous, non-redundant, and permanent, and each concept should be distinct from the strings used to name it. Bowker and Star (1999) add the necessary corrective: every classification encodes choices about what matters and who is counted, and those choices are easy to forget once the categories feel natural. The alternative to a chosen vocabulary, though, is not freedom. It is the tumor board.

Term, concept, hierarchy, relation

It helps to see the pieces laid out against the opening scene. The table below is a synthetic example built from the invented case above; the items in it are not drawn from any terminology, and the identifiers are local to this page.

Figure 1. Four things that were tangled together in one word (text table, synthetic example). Organizing criterion: the role each element plays in making "progression" sayable and checkable.

Element	What it is	In the tumor board
Term	A string used by a community	"progression"
Concept	A unit of meaning, with a definition and a stated criterion	At least three: anatomic progressive disease by new lesion; PET-based progressive disease by uptake criteria; clinical progression by combined judgment
Hierarchy	Where the concept sits among broader and narrower kinds	Each of the three is a kind of "disease status assessment"; none is a kind of the others
Relation	A typed, directed connection to other concepts	Each assessment "is based on" particular findings, "uses" a named criterion, and "was made at" a particular time

Alt text: a three-column table listing term, concept, hierarchy, and relation, with a definition and a tumor board example for each.

The table makes one thing visible. The three clinicians did not need a different term. They needed three concepts, each with its criterion named, placed under a common parent so that their kinship was explicit, and each connected by stated relations to the findings it rested on. That is what an ontology, in the modest informatics sense, is for. The companion essay From classification to clinical understanding takes the same pieces and asks when a hierarchy helps and when it conceals, and the lexicon entry on Relationship gives the short definition.

Back to the reading room

Nuclear medicine needs a shared language more urgently than most fields, for reasons that have nothing to do with philosophy and everything to do with its structure.

A radiopharmaceutical is at once a chemical entity, a radioactive source, an administered quantity, a regulated product, and a clinical intervention. Its "dose" is absorbed energy to a dosimetrist and an injected activity to a nurse. Its "target" is a receptor to a biologist and a focus on an image to a radiologist. The practice of Theranostics deliberately pairs a diagnostic agent with a therapeutic one and asks imaging, chemistry, physics, and oncology to agree on what they have seen and what will follow. Each discipline brings a vocabulary that is precise within its own walls and treacherous at the border, and the border is where the patient is.

The field is also growing faster than its words. New agents, targets, dosimetry methods, and trial designs arrive every year, each with a provisional name. Whether those names settle into shared concepts or drift into private understandings will decide whether the data of this decade can be pooled with the data of the next.

So return, finally, to Dr. Amari and Dr. Lindqvist. The right ending to their disagreement is not a ruling. It is a report, and later a record, in which "progression" never appears alone. Each use is bound to a concept with a definition, the criterion that fixed it is named, the findings it rests on are linked to it by stated relations, and the confidence in each finding is recorded separately from the finding's kind. When that is done, the three clinicians can disagree about the patient, which is their job, instead of disagreeing about a word, which is nobody's.

NucLex is an attempt to build the shared language that makes such reports possible for nuclear medicine and theranostics. In this publication's first release that attempt is editorial: essays that explain why the distinctions matter, monographs that develop them, and a lexicon of draft concept entries that state them compactly. It proposes a terminology, and later a computable one, but proposes is the correct verb. What it has today is an argument and a vocabulary in draft. The argument is the one made here: a name begins a conversation, and only a shared concept lets us ask whether we mean the same thing.

Limitations

This is an AI-assisted draft that has not been source-checked or reviewed by a domain expert or an editor. Its claims should be read accordingly.

The opening scene is wholly synthetic. It is designed to show a conceptual difficulty, not to represent any real patient, clinician, or institution, and it makes no recommendation about how any response criterion should be applied. Readers who need the actual definitions in RECIST or PERCIST should consult those documents directly; this essay paraphrases their general approach and does not reproduce their rules.

The philosophical material is compressed. Heraclitus and Parmenides survive only in fragments and later testimony, and any paraphrase involves interpretive choices that specialists dispute. The treatment of Aristotle follows standard English editions and does not engage the scholarly literature on how the Categories should be read. The paragraphs on Heidegger and Wittgenstein borrow a practical lesson from each and do not claim to represent either philosophy in full. None of these thinkers is presented as an anticipation of modern informatics.

The distinction between ontology and epistemology is drawn sharply here for clarity. In practice the two questions interact: what we count as a kind depends partly on what we can know, and what we can know depends partly on the kinds we have. The monographs on Ontology and Epistemology take up that interaction.

Finally, the essay describes what NucLex proposes. It does not describe an implemented terminology, a live service, or a validated tool. No such capability exists in this release.

Sources and further reading

• Genesis 11:1-9. In: The Hebrew Bible. Any scholarly edition, for example the Jewish Publication Society Tanakh (1985) or the New Revised Standard Version (1989).
• Eco U. 1995. The Search for the Perfect Language. Translated by Fentress J. Blackwell.
• Kirk GS, Raven JE, Schofield M. 1983. The Presocratic Philosophers. 2nd ed. Cambridge University Press.
• Plato. Cratylus. Translated by Reeve CDC. In: Cooper JM, ed. 1997. Plato: Complete Works. Hackett.
• Aristotle. Categories and De Interpretatione. Translated with notes by Ackrill JL. Clarendon Press, 1963.
• Aristotle. Topics. Translated by Pickard-Cambridge WA. In: Barnes J, ed. 1984. The Complete Works of Aristotle. Princeton University Press.
• Porphyry. Introduction (Isagoge). Translated by Barnes J. Clarendon Press, 2003.
• Heidegger M. 1962. Being and Time. Translated by Macquarrie J, Robinson E. Harper and Row. Original: Sein und Zeit, 1927.
• Wittgenstein L. 2009. Philosophical Investigations. Translated by Anscombe GEM, Hacker PMS, Schulte J. 4th ed. Wiley-Blackwell. Original: 1953.
• Gettier EL. 1963. Is justified true belief knowledge? Analysis 23(6):121-123.
• Gruber TR. 1993. A translation approach to portable ontology specifications. Knowledge Acquisition 5(2):199-220.
• Guarino N, Oberle D, Staab S. 2009. What is an ontology? In: Staab S, Studer R, eds. Handbook on Ontologies. 2nd ed. Springer.
• Smith B. 2004. Ontology. In: Floridi L, ed. The Blackwell Guide to the Philosophy of Computing and Information. Blackwell.
• Cimino JJ. 1998. Desiderata for controlled medical vocabularies in the twenty-first century. Methods of Information in Medicine 37(4-5):394-403.
• Bowker GC, Star SL. 1999. Sorting Things Out: Classification and Its Consequences. MIT Press.
• Eisenhauer EA, Therasse P, Bogaerts J, et al. 2009. New response evaluation criteria in solid tumours: revised RECIST guideline (version 1.1). European Journal of Cancer 45(2):228-247.
• Wahl RL, Jacene H, Kasamon Y, Lodge MA. 2009. From RECIST to PERCIST: evolving considerations for PET response criteria in solid tumors. Journal of Nuclear Medicine 50(Suppl 1):122S-150S.