In a Hypothesis and Theory article published in Frontiers in Ophthalmology, oculoplastic surgeon Zvi Gur and clinician-investigators Itay Nitzan and Tehila Shlomov propose a framework they describe with a deliberately provocative label: mentorship as metadata. In data science, metadata describes data and makes it interpretable; without it, a field of values has no defined meaning, provenance, or permissible use. The authors argue that mentorship performs an analogous function in surgical education, teaching the context around the code—what the diagnosis included, what the operation intended, what the outcome meant, and what the dataset omitted. Crucially, they are careful to state that this is a labelled metaphor, not a claim of equivalence with formal metadata, data provenance, or research-governance documentation. The deliverable they propose is something more concrete: documented clinical interpretive annotation, a written, dated, co-signed record of the clinical limits of a dataset for a stated question, filed with the study protocol.

The framework rests on a three-layer model that classifies what any oculoplastic database question actually requires. The first layer is the code: diagnosis, procedure, medication, laboratory, imaging order, encounter, and date. The second is the phenotype: severity, laterality, anatomy, activity, function, imaging, histology, and patient-reported symptoms, which are only variably captured in routinely collected sources. The third is judgment: indication, expectation, operative plan, surgeon threshold, revision tolerance, and the interpretation of success or failure, which are rarely captured anywhere. Availability typically decreases across these layers, and the authors emphasize that the layers classify what a question requires rather than ranking what a source contains. If a study concerns utilization, the code may be the construct itself; if it concerns severity, the code is a proxy; if it concerns aesthetic success, operative quality, or the meaning of revision, the dataset must contain images, operative notes, structured measurements, patient-reported outcomes, or validation.

To ground the framework in evidence, the authors examined ten recently published oculoplastic database studies spanning federated electronic-health-record networks, national claims and Medicare data, a population cancer registry, and an audited surgical-quality registry. The pattern they found is consistent. A TriNetX study of cannabis use and thyroid eye disease can test a coded association but remains constrained by smoking behavior, thyroid control, and disease severity. A comparison of endoscopic and external dacryocystorhinostomy can evaluate trends and coded revision rates, but a revision code does not equal anatomical failure, persistent epiphora, canalicular disease, or dissatisfaction. A study of blepharoptosis and falls can test a population-level association without revealing visual-axis obstruction, chin-up posture, field loss, or frailty. Analyses of glucagon-like peptide-1 receptor agonists and oculofacial procedures detect utilization signals but cannot distinguish biologic tissue change from weight loss, aesthetic demand, or coding of cosmetic procedures. Medicare analyses of orbital decompression around teprotumumab approval demonstrate changing volume, approach, and spending but not inflammation, compressive optic neuropathy, or operative intent.

Two of the ten studies are retained deliberately as counterexamples, because the boundary in each falls late rather than early. An analysis of orbital floor fracture repair in the National Surgical Quality Improvement Program reports outcomes abstracted by trained reviewers against standardized definitions and subject to inter-rater audit, so its 30-day systemic endpoints are adjudicated rather than inferred from codes; the limitation concerns which construct was chosen, not how well it was measured. A SEER analysis of eyelid sebaceous gland carcinoma likewise reports incidence, stage distribution, and relative survival, which are constructs a population cancer registry is designed to measure and measures well. The authors use these counterexamples to make an important point: adequately captured phenotype is attainable within routinely collected data, and their framework is intended to identify where the boundary falls in a given study rather than to assume a fixed position for it.

At the center of the proposal is a construct the authors call the refusal point: a documented, pre-specified boundary at which a dataset should not be used to answer a given clinical question. The refusal point is produced before data extraction, co-signed by three parties—the oculoplastic mentor accountable for its clinical content, the trainee or lead analyst accountable for the operational definitions, and a methodologist or data scientist accountable for the identification conditions—and filed with the study protocol, where it can be audited against the final manuscript. The authors are explicit about what the refusal point is not. It is not a new statistical construct, and it does not identify a failure mode that existing methods cannot already name: a non-estimable estimand, a positivity violation, an unverifiable exchangeability assumption, and a construct the data do not measure are all established concepts. What is lacking, they argue, is a record of that assessment. Data fitness for purpose is at present usually informal; it is performed at a variable point in the workflow, by an unspecified person, and is reported, if at all, as a limitation written after the results exist.

The refusal point is designed to complement, not replace, established frameworks. STROBE requires transparent reporting of design, variables, and limitations in observational studies, but limitations are written after the analysis and no owner, timing, or storage is specified for the judgment about whether the data can support the intended inference. RECORD extends reporting to routinely collected health data—code lists, linkage, validation, cleaning, and population selection—but leaves unspecified whether a correctly reported code is an adequate surrogate for the clinical construct at issue. The ICH E9(R1) estimand framework requires precise specification of population, treatment, endpoint, and intercurrent events, but does not say what follows when the specified estimand is not estimable in the available data, who decides, or where the decision is recorded. Target trial emulation requires explicit specification of the hypothetical randomized trial that observational data approximate, but does not require disclosure when the data lack the clinical variables the emulation needs. The refusal point adds the missing documentary layer: it names the adjacent, more tempting question that the data cannot answer, before extraction, with dated signatories.

The authors advance a single falsifiable hypothesis: that pre-specified, co-signed refusal-point documentation changes research behaviour and output. They predict that oculoplastic real-world data projects completing a co-signed refusal point before data extraction will show a higher frequency of pre-analysis reframing of the research question, greater concordance between the estimand recorded in the protocol and the estimand the published conclusions actually support, and a lower frequency of causal or comparative-effectiveness claims unsupported by the available phenotype, as judged by blinded independent assessors. The hypothesis is refuted if projects using the documentation show no measurable difference relative to contemporaneous or historical controls, or if independent mentor–methodologist pairs given the same protocol cannot agree on where the refusal point falls, which would show that the construct is not reproducibly identifiable. To test it, they set out a prospective, two-centre implementation study of consecutive oculoplastic database projects over 24 months, with blinded outcome assessment, inter-rater agreement measured with Cohen’s kappa, and pre-specified refutation criteria including a kappa below 0.4 or a median completion time exceeding 60 minutes per project.

Worked examples show the discipline changing conclusions rather than merely qualifying them. In one, a trainee proposes to use a claims network to show that endoscopic dacryocystorhinostomy is superior to the external approach by a lower coded revision rate. The phenotype map forces three admissions: revision is a utilization code, not an adjudicated failure; the two approaches may be embedded in different specialty pathways; and obstruction site and symptom severity are unmeasured, producing confounding by indication of unknown direction. The refusal point is reached for the word superior, and the study is downgraded to a descriptive, hypothesis-generating account of trends. In another, a fellow asks whether teprotumumab approval changed orbital decompression practice. The causal question is unanswerable without Clinical Activity Score or optic-neuropathy data, so the refusal point is written against it; but naming it clarifies a legitimate, well-specified utilization estimand—the population-level rate, approach, and cost of coded decompression before and after a fixed approval date—converting a vulnerable comparative-effectiveness study into a defensible interrupted-time-series description. The discipline is therefore not only subtractive; it relocates a project onto the firmest construct the data can carry.

The authors applied their framework retrospectively to their own published study of reoperation for ptosis after intraocular surgery, deliberately using their own work to permit a fully candid appraisal. The exercise downgraded the defensible claim of that paper from recurrence to coded reoperation: laterality was unverifiable, so ipsilateral and contralateral procedures could not be distinguished; the contralateral Hering effect, planned staged surgery, cosmetic versus functional indication, and individual surgeon revision threshold were all absent from the source; and exchangeability was not verifiable because indication for the index repair was unmeasured. The framework remains a proposal, not a validated instrument, derived by three authors at a single institution without formal consensus methodology or external review, and the illustrative studies were selected non-systematically by clinicians who had already formed a view. Yet the central principle is one that resonates far beyond oculoplastics: real-world data remain valuable for utilization, trends, and rare associations, but they are not self-interpreting. Training surgeon-scientists to name, in writing and before analysis, the boundary between code and clinical reality may make database research more transparent, restrained, and reproducible. In oculoplastics, the authors conclude, mentorship is the clinical annotation that turns data into evidence.

Subject of Research: Mentorship as metadata: clinical annotation and the limits of real-world data in oculoplastic surgery

Article Title: Mentorship as metadata: clinical annotation and the limits of real-world data in oculoplastic surgery

Article References: Nitzan, I., Shlomov, T., & Gur, Z. (2026). Mentorship as metadata: clinical annotation and the limits of real-world data in oculoplastic surgery. Frontiers in Ophthalmology, 6, Article 1972892. https://doi.org/10.3389/fopht.2026.1972892

Image Credits: AI Generated

DOI: 10.3389/fopht.2026.1972892

Keywords: Mentorship, metadata, clinical, annotation, limits, real-world, data, oculoplastic, surgery, scientific research