The Class the Examples Draw

I.The First Known Thing — the words at the end of the list

statutory interpretation · ejusdem generis

Statutes are full of lists that end in a general phrase. A house, office, room or other place. Seamen, railroad employees, or any other class of workers. Theatres and other places of amusement. The drafter lists what was in mind, then adds words meant to catch what was not, and the court is left to decide how much the catching words catch. The common law's answer, old enough to have kept its Latin, is the canon called ejusdem generis — of the same kind. General words that follow a list of specific ones are read as reaching only things of the same kind as the listed. The reasoning is about the drafter rather than the words: a drafter who meant any place whatever would not have troubled to write house, office, room first, so the list is taken as the better evidence of what was meant, and the general phrase is cut down to the class the list reveals.

the reach belongs to the list

The consequence that matters here is where the reach of the general phrase comes from. It does not come from the phrase. It comes from what the enumerated items have in common, and the court's task under the canon is to find the narrowest class that contains all of them and read the general words as that class. In Powell v Kempton Park Racecourse, decided by the House of Lords in 1899, a betting statute forbade keeping a house, office, room or other place for betting, and the question was whether an outdoor enclosure at a racecourse was such a place; it was not, because the listed items were all enclosed premises and other place was read as another such. In Circuit City Stores v Adams, decided by the United States Supreme Court in 2001, a statute exempted the employment contracts of seamen, railroad employees, or any other class of workers engaged in foreign or interstate commerce; the general clause, taken alone, would have reached nearly every worker in the country, and the Court read it, on the strength of its two examples, as reaching transportation workers only. In each case the general phrase was the same width on the page. Its reach was set by the spread of the examples in front of it.

the two conditions under which the rule lets go

The canon is not a rule that lists always narrow. It has two well-known conditions under which it does not apply at all, and both concern the shape of the list rather than the general words. The first is that there must be a list. In Allen v Emmerson, decided in 1944, a licensing provision covered theatres and other places of amusement, and the question was whether a funfair was one. The court held that it was, because a single specific word cannot constitute a class: with one example there is no kind for the general words to be of the same kind as, and they take their full ordinary meaning. English authorities have since stated the requirement plainly — at least two specific terms before the general one, or the canon has nothing to work on. The second condition is that the examples must share a kind narrower than the general words themselves. Where the listed items have nothing in common except that each is an instance of the general term, the canon cannot cut the term down, since the only class the list reveals is the one the term already names; and where the listed items exhaust their class, the general words must mean something beyond it or be surplusage, which courts are reluctant to find. So the canon has a shape that the profession knows in its hands and does not often state in full. The examples set the reach. One example sets nothing. Examples that share nothing but the term leave the term at full width. And an express exclusion at the end of a list — other place, but not a racecourse — is an admission that the drafter already knew the general words would otherwise reach it.

where the reach is actually measured

One more feature of the canon is worth stating because it is the one that transfers least comfortably. The reach of the general words is never decided at the drafting. It is decided later, in litigation, by a case that was not on the list — the racecourse, the funfair, the shop clerk with an arbitration clause — and the class the examples draw is discovered only when something outside the examples arrives to test it. Until then the general phrase is as wide as it reads, and everyone acting under the statute acts on that width. The field regards this as ordinary. Statutes are read when they are applied, and the drafter's examples are evidence to be weighed when a hard case comes, not a boundary surveyed in advance.

II.The Second Known Thing — the indication that names no tumour

drug discovery · disease definition, indication choice, endpoint selection

For most of its history a cancer drug was approved for a cancer — a tumour named by the organ it arose in, and then, as the field learnt to subdivide, by organ and histology and stage and line of treatment. The indication was the disease, and the disease was the tissue. That changed in 2017, when the United States regulator approved pembrolizumab for any solid tumour, wherever it arose, that carried a particular defect in DNA mismatch repair. The evidence was a pooled analysis of a hundred and forty-nine patients drawn from five trials and spread over fifteen tumour types; ninety of them had colorectal cancer, and the rest were scattered across a long tail of others. The indication was written not as a list of cancers but as a marker followed by the words solid tumours, and the field acquired a name for the form — the tissue-agnostic indication — and a trial design to supply it, the basket trial, in which patients are enrolled by marker and sorted into cohorts by histology afterwards.

the list behind every agnostic label

There are now about ten such indications, and each was granted on a list. Larotrectinib was approved in 2018 for any solid tumour carrying a fusion of an NTRK gene, on fifty-five patients spread across seventeen tumour types, from salivary gland to sarcoma to lung; entrectinib followed a year later on similar ground, and repotrectinib in 2024. Pembrolizumab was extended in 2020 to tumours with a high mutational burden, on ten cohorts. Dostarlimab was approved for mismatch-repair-deficient tumours in 2021. Selpercatinib, first approved for lung and thyroid cancers carrying RET alterations, was extended in 2022 to any solid tumour with a RET fusion on a few dozen patients outside those two organs. Dabrafenib with trametinib was approved in 2022 for solid tumours carrying the BRAF V600E mutation. Trastuzumab deruxtecan was approved in 2024 for solid tumours expressing HER2 at the highest immunohistochemical level, on a hundred and ninety-two patients across three trials. In every case the label reads as a general phrase — solid tumours — and the evidence for it is an enumeration of the tumours actually treated, the enumeration being as long, as wide, and as evenly filled as the accident of accrual made it.

the field already knows the list is not the label

The field knows perfectly well that the marker does not always carry the drug across tissues, and the knowledge is written into the labels. The BRAF V600E indication excludes colorectal cancer by name, because the field had learnt, in the years after the mutation was found to drive melanoma, that colorectal tumours with the same mutation barely respond to the same drugs: response rates near half in melanoma fell to a few per cent in the bowel, for a reason that was eventually traced to a feedback loop through a growth-factor receptor that the colon expresses and the skin does not. The mismatch-repair label was granted on a set in which the largest cohort responded less well than the others. And the regulator's guidance on the form, issued in draft in 2022, asks for evidence across multiple tumour types and a strong biological rationale that the marker, and not the tissue, is what the drug acts on, without saying how many types are enough or how they should be spread. What the field has, then, is a general indication, an enumeration behind it, a known case in which the enumeration would have misled, a carve-out written to prevent it, and a rule of thumb — more tumour types is better — that counts the enumeration without asking what shape it has.

what nobody scores

What the field does not do is treat the enumeration as the thing that fixes the reach. It treats the marker as the thing that fixes the reach and the enumeration as evidence that the marker does so, and the two are not the same claim. A basket in which fifteen tumour types were treated but ninety of a hundred and forty-nine patients had one of them is counted as fifteen; a basket of fifty-five patients across seventeen types is counted as seventeen and doubted for its size; and no measure in use asks how far apart the tried tumours were — whether they share a lineage, an embryological origin, a microenvironment, a second driver — or predicts, from that spread, how the drug will do in the tumour types that were never enrolled. Those tumour types are, after all, the point of the label. The agnostic indication is granted precisely so that the drug may be given in histologies too rare or too scattered to have appeared in the trial; the label's value lies in the unenumerated cases; and the question of how the enumerated cases bear on them is answered, at present, by a count.

III.The Bridge

no treatise on construction cites a basket trial; no approval cites a canon

The halls do not meet. The canon belongs to the interpretation of statutes and contracts, and its literature is about drafters, courts and the weight of words; the tissue-agnostic indication belongs to oncology drug regulation, and its literature is about markers, baskets and the biology of response. No review spans them, and the one word they share — an indication, which in the legal hall means a hint of intent and in the regulatory hall means a licensed use — is a coincidence of English. Three neighbours should be set aside. It is not the demarcation by adjacency of an earlier specimen, in which a category defined as the complement of its neighbours inherits their revisions; there the category has no boundary of its own, whereas here it has one, drawn by its own examples. It is not the contrastive load of another, in which two categories are distinct only if some environment makes exchanging them matter; that specimen concerns whether two named things are one or two, whereas this one concerns how far an unnamed thing extends. And it is not the crossover spread of a third, in which the disagreement of observables about a boundary measures whether a category exists; here the boundary is not found by observables at all but by the spread of the instances that were offered as evidence for it. What the halls share is a rule for reading a general term that was granted on a list.

X: the examples draw the class, and the count of them draws nothing

Call it enumerated reach. Where a general term is granted on the strength of enumerated instances — the instances offered as the reason the general words should be allowed at all — the term's reach is not the width of its wording but the narrowest class that contains the instances. Five things follow. The reach is set by the span of the examples, how far apart they lie, and not by their number nor by how many cases fall under each; ten examples of one kind draw a class of one kind. A single example draws no class, so a general term granted on one instance has no reach that can be read from it and is either refused or left at its full wording, with nothing in between. Examples that share nothing except the general term itself cannot narrow it, and the term is left at full width, which is the one case in which the wording and the reach coincide. An express exclusion appended to the term is an admission that the grantor already knew the term would otherwise reach the excluded case, and so marks the point at which the examples and the wording were known to disagree. And the reach is not surveyed when the term is granted; it is measured only when a case outside the examples arrives to test it, so that everyone acting under the term before then acts on its wording. Read into the regulatory hall, each has a consequence the hall does not state. A tissue-agnostic label's reach into the tumours it was never tried in should be predicted by how far apart the tried tumours were, not by how many there were or how many patients each held, so that a basket dominated by one histology has, whatever its count, drawn a narrow class. A marker whose evidence rests on one tumour type has drawn no class and cannot found an agnostic claim, which is a rule the regulator applies without having stated the reason. A marker whose tried tumours share nothing but the marker — no lineage, no organ system, no common second driver — is the one whose label is safe at its full width, and that condition is checkable from the enumeration alone. The colorectal carve-out from the BRAF label is the point at which the examples and the wording were known to disagree, and every agnostic label without such a carve-out is one in which that disagreement has not yet been found rather than one in which it does not exist. And the label's true reach is measured only in the unenumerated histologies, after approval, by patients treated on the strength of its wording.

IV.The Question That Becomes Possible

neither discipline would ask this alone

The regulatory hall asks whether a marker predicts response across tissues and reads the enumeration behind an agnostic label as evidence for the marker, to be weighed by its size. The legal hall has no tumours, but it has established that a general term granted on a list reaches as far as the list's span, that one example sets nothing, that examples sharing only the term leave it whole, and that the reach is learnt from the case that was not listed. Neither hall asks whether the performance of an agnostic drug in the histologies its trial never enrolled is predicted by the spread of the histologies it did enrol.

the enumerated reach hypothesis

Across the tissue-agnostic indications granted to date, is a drug's response rate after approval in histologies absent from its pivotal enumeration predicted by the span of the enumerated histologies — the distance between them along a lineage classification of tumours — better than by their number, by the number of patients, or by the pooled response rate on which the label was granted; is the shortfall between enumerated and unenumerated response concentrated in the labels whose enumerated histologies share a lineage besides the marker; and does a label whose enumerated histologies share nothing but the marker show no shortfall at all?

why it would matter

Because the agnostic indication is the field's chosen form for the rare and the scattered, and its value lies entirely in the cases the trial did not contain. If the span rule holds, a regulator can read the reach of a proposed label from the shape of the basket before granting it, ask a sponsor to enrol for spread rather than for size, and write the label's width to match — a general phrase where the examples share nothing, a list where they share a lineage, a carve-out where the examples predict one. If the single-example rule holds, it gives a stated reason for a practice the regulator already follows and marks the boundary at which a second histology turns an indication into a class. And if the rule fails — if a drug's reach into untried tumours is set by the marker's biology and owes nothing to the shape of the list — then the count the field currently keeps is as good as any measure, the basket's job is only to be large, and the field can stop worrying about its spread.

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 trial; it needs the pivotal data behind each agnostic label, which are public, and the record of how the drugs have done since in the histologies those data did not contain, which exists in registries, expanded-access programmes, real-world databases and the long-term follow-up the sponsors themselves publish.

the honest status

Note the narrowness. The claim concerns the reach of an agnostic label into histologies its pivotal data did not contain, and says nothing about whether the drug works in the histologies it did; a marker may be an excellent predictor within the enumeration and the list may still have drawn a narrow class. It does not say that agnostic labels are wrong, or that baskets should be smaller; it says that the shape of the basket carries information the count discards, and that the information concerns the cases the label exists for. It does not say that the number of tumour types is worthless; it says the number is a proxy for the span and should be replaced by it if the two come apart. It is worth saying what does not transfer. A court reading a list has a drafter whose intent the list is evidence of, and the canon is a theory of that intent; a basket has no drafter, and its examples were supplied by prevalence, referral and chance. The legal hall's rule that one example founds no class is a rule about language, and its regulatory counterpart may hold for a different reason — that one histology cannot separate the marker from the tissue — which is a fact about confounding and not about kinds. And the legal hall measures reach in a single hard case decided once; the regulatory hall measures it in a scatter of uncontrolled reports, and the two may not be commensurable. The whole thing rests on one tabulation that a single analyst with the approval packages and a tumour classification could make in a season: whether the spread of the histologies behind each agnostic label predicts how the drug has done in the histologies it never saw. If it does not, this is an elaborate way of saying that more tumour types are better, and should be said so. Until that tabulation is made, 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.