Methods · version 0.3

How the index is searched, matched, and staged.

This page makes the beta reproducible enough to critique. It is a curated evidence index—not a systematic review, clinical guideline, or product comparison.

Current scope: spineSearch cutoff: Aug 24, 2026Planned re-verification: quarterly

01 · Scope

What qualifies for the staged index.

Regulatory pathway

Current entries are FDA 510(k)-cleared platform families. “Cleared” is not used as a synonym for approved, clinically superior, or proven beneficial.

Spine relevance

The intended use must include a spine-specific diagnostic, imaging, measurement, planning, navigation, or monitoring function used in a clinical pathway.

AI identification

The FDA record must explicitly identify AI or machine learning, or describe a trained algorithm under an applicable software classification. Commercial claims alone are insufficient.

Exclusions

Bench-only concepts without a relevant clearance, nonclinical tools, general workflow software, and records with unresolved AI ambiguity are excluded or placed on the watchlist.

02 · Search protocol

Primary records first.

Regulatory and safety sources

FDA 510(k) database and decision summaries; FDA AI-enabled medical-device list; FDA recall database; predicate and reference submissions named in the latest included summary.

Clinical and registry sources

PubMed and publisher full text when available; ClinicalTrials.gov for registered studies and current registry status. Manufacturer pages may identify leads but do not establish inclusion or stage.

Search construction

Trade name, manufacturer, submission number, predecessor names, and core function are combined with terms such as clinical, validation, prospective, reader, workflow, outcome, trial, recall, and software version.

Screening window

Database inception through August 24, 2026. Searches are English-language and public-record focused. Up to three key studies are displayed when several papers address the same evidence level.

The FDA calls its AI-enabled device list noncomprehensive. At the cutoff used here, the displayed table ended with decision dates through March 30, 2026; absence of a later record from that list is therefore not treated as evidence that the record lacks AI.

03 · Family and version matching

Evidence is not assumed to transfer silently.

Platform family

Iterative submissions sharing a trade name and core function are grouped so readers can follow the current record, selected predicates, older studies, and family-level safety signals.

Latest included record

The submission shown in the main table is the latest included FDA record located for that platform family as of the cutoff. Earlier records remain linked when they contain the relevant validation.

Direct evidence match

The cited regulatory or clinical evidence evaluates the named platform and materially matching function. Direct does not guarantee that a paper reports the exact cleared software version.

Partial evidence match

A device-family, software-version, workflow, or function mismatch limits transferability. The mismatch is described in the evidence snapshot and must not raise the stage beyond what the matched evidence supports.

04 · Evidence maturity

A vertical ladder, not a product score.

0Regulatory record onlyClearance plus bench or non-human testing; no human clinical validation located.1Human performance validationHuman data appear in FDA materials; no peer-reviewed study of the exact or materially matching function located.2Clinical validityPeer-reviewed human evidence evaluates the exact or a materially matching function.3Clinical utilityEvidence evaluates effects on clinician performance, workflow, or decision-making.4Comparative patient outcomesProspective comparative evidence evaluates meaningful patient outcomes.

The displayed stage is the highest evidence type located for the platform family. It does not summarize study quality, comparative effectiveness, safety, or overall product value.

05 · Independence labels

Who produced the evidence matters.

Independent

No manufacturer authorship, direct study funding, or disclosed commercial support was located in the displayed evidence.

Mixed

The record includes both external clinical investigators and a manufacturer relationship, author, funding source, software provision, or institutional collaboration.

Manufacturer-linked

The displayed peer-reviewed evidence has manufacturer authorship, funding, or material support without a clearly independent exact-function study.

FDA-only

The stage rests on sponsor-submitted performance information summarized by FDA rather than a peer-reviewed exact- or materially matched clinical paper.

06 · Trials and safety

Status is reported, not implied.

Registered studies

Each displayed ClinicalTrials.gov record includes its registry status and the date checked. Registered, recruiting, and ongoing are not treated as interchangeable terms.

Recall search

Searches use the exact submission, trade and family names, manufacturer, and relevant predecessor records. Exact-submission and family-level findings are labeled separately.

No recall located

This means no FDA recall linked to the displayed submission was found in the documented search as of the cutoff. It does not mean that no adverse events or family-level safety issues exist.

MAUDE boundary

MAUDE reports may identify surveillance signals but cannot establish incidence, causation, or comparative safety and do not determine the evidence stage.

07 · Review and limitations

Structured checks, with limits stated plainly.

Each entry receives regulatory, clinical-literature, and editorial consistency checks against linked primary records. The current beta has not undergone independent dual-human systematic-review screening or journal peer review.

Potential omissions remain possible because device names change, publications may not name commercial versions, FDA summaries vary in detail, and paywalled records may restrict full-text inspection. Uncertainty is preserved as a mismatch or watchlist entry rather than filled by inference.

Substantive corrections are dated in the public changelog. Suspected errors should include the device, statement at issue, and a PMID, DOI, FDA submission, recall, or registry record when possible.

Submit a correction →