What the Controlled-Vocabulary Layer Cannot Say About Where a Depression Biomarker Was Measured: A Meta Research Analysis of MeSH Sampling-Compartment Coding, January 2024–June 2026

Abstract

Background: Peripheral biomarkers for major depressive disorder are accessible but correspond poorly to central nervous system processes, a point the evidence base can address only if its record registers the compartment in which each marker was measured. To determine, by meta research rather than evidence synthesis, whether the indexed record of recent depression biomarker research registersthe peripheral-central distinction, separating omission by indexers from absence mandated by the controlled vocabulary. That peripheral markers are imperfect proxies for central processes is established prior knowledge; the question here concerns the bibliographic infrastructure through which that literature is retrieved.

Methods: PubMed records published between 1 January 2024 and 30 June 2026 matching a title/abstract query for depression and biomarker terms were retrieved (Tier 1, n = 1166); the MEDLINE-indexed subset formed the analysable corpus (Tier 2, n = 825). We extracted marker-level descriptors and site qualifiers, determined from the 2026 MeSH descriptor file whether /blood and /cerebrospinal fluid are allowable per descriptor, and stratified by publication type.

Results: Of 1166 retrieved records, 341 (29.2%) carried no MeSH descriptor; in the closed year 2024, 70 of 317 (22.1%) remained unindexed, so roughly one article in five is permanently absent from the structured record. Within the 825 indexed records, 494 (59.9%) carried no markerlevel descriptor and 658 (79.8%) no site qualifier. Of 1211 distinct descriptors, 681 (56.2%) do not permit /blood at all, most of them terms for which a sampling compartment would carry no meaning; restricted to the 243 markerlevel and class-level descriptors, 30 (12.3%) are barred, C-reactive protein among them, the most frequent marker descriptor in the corpus (26 records), though the same descriptors can receive /cerebrospinal fluid. Primary studies recorded a compartment three times more often than reviews (38.8% vs 14.8%; χ ² = 52.1, p < 0.001), but recording tracked qualifier eligibility more closely than study design, rising from 5.0% to 59.4% with the number of /bloodeligible descriptors a record carried.

Conclusions: The controlled-vocabulary layer cannot reliably recover the sampling compartment of depression biomarker research, for two structural reasons: the peripheral compartment is unrecordable for an entire marker class, and recording depends on how many of a record’s descriptors the vocabulary permits to carry the qualifier at all. A distinction that cannot be recorded cannot be aggregated, audited or contested. The inflammatory endotype literature of difficult-to-treat depression is thereby constrained by a blind spot at the level of retrieval rather than biology.