How PathDigest selects, tags, and summarizes the literature.
Transparency matters when automated tools touch clinical literature. This is exactly what PathDigest does, and where its methods have limits.
Literature sources
All article metadata comes from the U.S. National Library of Medicine's PubMed via the E-utilities API. PathDigest fetches new records on a recurring daily schedule, retrieves abstracts and publication types, and detects free full text through PubMed Central and PubMed LinkOut. PathDigest does not scrape publisher websites and does not host article full text.
How journals are selected
PathDigest monitors a curated set of 67 journals rather than all of PubMed. The list is pathology-first: core general and surgical pathology titles that most subspecialists read, plus subspecialty-specific journals and a small number of cross-disciplinary oncology titles where practice-changing pathology work regularly appears. The goal is high signal — journals a working pathologist would actually want to track — not exhaustive coverage. The complete list is on the Journal Coverage page.
Subspecialty and organ-site assignment
Each article is assigned to one or more of 12 subspecialties and tagged by organ site. Assignment uses two complementary signals: the journal's subspecialty scope, and text analysis of the title and abstract against curated term lists (with exclusion terms to avoid common false positives — for example, not tagging a tumour's liver metastasis as primary liver pathology). Articles pulled from broad journals are shown in a subspecialty feed only when their classification actually assigns that subspecialty.
How AI summaries are produced
Summaries and a four-way category label (diagnostic, translational, clinical, basic science) are generated server-side by a large language model from the published abstract, on a batch schedule. Summaries are stored and displayed read-only; nothing is generated live in your browser. Articles without an abstract are classified from the title alone and are marked accordingly.
Open access detection
PathDigest identifies free full text two ways: PubMed Central identifiers extracted from the article record, and free full-text links reported by PubMed LinkOut. When a free version is found, a direct link is surfaced on the article card. Absence of a flag means no free version was detected — not necessarily that none exists.
Limitations of automated classification and summarization
Automated tagging and summarization are imperfect. Organ-site and subspecialty tags can be wrong or incomplete, particularly for multi-organ studies, unusual entities, or abstracts with atypical wording. AI summaries can omit nuance, miss caveats, or contain errors. Category labels are heuristic. None of this is a substitute for reading the paper.
The original publication is always authoritative. PathDigest is a triage aid for deciding what to read — not a source of clinical guidance, and not a replacement for the primary literature. If you spot a misclassification or a summary error, please tell us on the Corrections page.