Computer-Interpretable Guideline¶
A clinical-guideline knowledge artifact that formally links patient conditions and decisions to recommended actions or plans for computational interpretation.
Core Idea¶
A computer-interpretable guideline (CIG) is clinical guidance encoded so that software can interpret represented patient conditions, choices and recommended actions or plans for a case. The clinical recommendation is the source content; the formal step or plan structure is the artifact. The artifact can exist before a local electronic health record (EHR) is connected, a real patient is processed or a clinical benefit is shown.[ref-ae9dc38afeea][ref-ff2c94c4f96d][^ref-d28daad00fe8]
A CIG does not mean that every choice is automatic. GLIF3 distinguishes decisions whose criteria can be evaluated from patient data from choices left to a clinician when safety or unspecified knowledge matters. The authors classify one chronic-cough encoding as intermediate in computability.[ref-9b7d5da4835c][ref-ae9dc38afeea]
Scope of Application¶
The InterMed group's GLIF3 chronic-cough example supplies a formal criterion for cough persisting over three weeks: latest_cough_end_time >= now and latest_cough_start_time < (now – 3 weeks). Get_Data actions define the two timestamp variables by linking them to patient-data concepts in a domain ontology. The same original paper represents a conditional historical recommendation: when cough is productive, order four-view sinus radiographs before beginning postnasal drip syndrome (PNDS) therapy. This documents an encoded source recommendation, not current medical advice. The paper does not show that this particular PNDS branch was automatically evaluated on a patient.[^ref-9b7d5da4835c]
The unlike Asbru newborn-hyperbilirubinemia case represents diagnosis and treatment as a hierarchy of time-oriented plans. Plan filters express eligibility, diagnosis can appear as treatment conditions or explicit diagnostic plans, and plan conditions govern selection, aborting and completion. The accessible original abstract and publisher previews support this plan-level mapping, without an exact bilirubin threshold, individual patient run or outcome finding.[^ref-d28daad00fe8]
Clarity¶
Look for five linked parts: identifiable clinical guidance, a formal patient-state or eligibility condition, a decision or plan relation, a recommended action, and a convention that lets software interpret the relation for a case. The cough criterion and PNDS recommendation exhibit these roles in GLIF3; Asbru uses filters, selection and plans instead. Neither syntax is necessary to every CIG.[ref-9b7d5da4835c][ref-ff2c94c4f96d][^ref-d28daad00fe8]
Boxwala and colleagues distinguish conceptual, computable and institution-specific implementable specifications within GLIF3. Those levels are not a universal pipeline for all guideline formalisms, and the last level is not evidence that the chronic-cough example was installed. The 2004 InterMed review calls the overall cough encoding intermediate despite the formal criterion documented in the earlier paper.[ref-ff2c94c4f96d][ref-ae9dc38afeea]
Manages Complexity¶
A formal representation separates patient facts from the rule or plan that uses them and from the clinical action that follows. GLIF3 links patient-state, decision and action steps; the InterMed review depicts a nested version of the chronic-cough flow with three action steps as subguidelines. Asbru organizes a different clinical case with timed plan hierarchies. These structures help specify what a reader or program is meant to interpret.[ref-ae9dc38afeea][ref-d28daad00fe8]
A diagram alone is not enough. The InterMed authors report that earlier GLIF2 step attributes were unparsed prose strings and therefore did not supply the automatic inference required for computer execution. Local EHR bindings, full evidence tracing and site adaptation may improve use or audit but are not constitutive of every encoded guideline.[^ref-ae9dc38afeea]
Abstract Reasoning¶
For the chronic-cough criterion, the expression uses the represented end and start timestamps to test whether the modeled cough period exceeds three weeks. The original shows the expression and Get_Data links, but no particular EHR return, missing-data policy or executed PNDS decision. GLIF3's general case-step versus clinician-choice distinction should not be projected onto that specific branch without evidence.[^ref-9b7d5da4835c]
In Asbru, eligibility filters and plan-choice relations play the corresponding structural roles even though the artifact is a time-oriented plan hierarchy, not a GLIF3 step network. Remove the formal patient-condition-to-recommendation relation and a list of clinical actions remains, but it is no longer a CIG in this sense.[^ref-d28daad00fe8]
Knowledge Transfer¶
To assess another proposed CIG, trace one source recommendation into an encoded patient condition, a choice or plan rule, and a linked action. State separately what the formalism permits, what the cited case actually encoded, and what was later deployed or evaluated. A formal eligibility example does not prove that every branch is executable or that the encoded medical advice is clinically valid.[ref-9b7d5da4835c][ref-ae9dc38afeea]
The common act of mapping guidance into a different interpretable medium is a strict instance of the live Representation Prime. The clinical patient-condition and recommended-action roles make the CIG narrower. A representation need not be clinical; a CIG need not already be an operating rule system.[ref-ff2c94c4f96d][ref-d28daad00fe8]
Example¶
GLIF3 chronic cough: the three-week expression is a concrete patient-state criterion. The historical productive-cough clause connects a condition to imaging before PNDS therapy. The authors' later Table 1 lists chronic cough as outpatient diagnosis plus management over multiple encounters and rates its computability intermediate. The case is therefore an encoded condition-and-action mapping, not proof of complete branch execution or EHR deployment.[ref-9b7d5da4835c][ref-ae9dc38afeea]
Asbru neonatal hyperbilirubinemia: an identified newborn guideline is encoded with plan filters, choice among treatment options and timed plans that can include completion or abort conditions. The original abstract and accessible publisher previews do not warrant numerical thresholds, a patient trace or clinical-effect claims.[^ref-d28daad00fe8]
Relationships to Other Abstractions¶
Current abstraction Computer-Interpretable Guideline Domain-specific
Parents (1) — more general patterns this builds on
-
Computer-Interpretable Guideline is a kind of Representation Prime
A computer-interpretable guideline represents clinical recommendations in a formal medium that supports patient-case reasoning.
Hierarchy path (1) — routes to 1 parentless root
- Computer-Interpretable Guideline → Representation → Abstraction
Neighborhood in Abstraction Space¶
Computer-Interpretable Guideline sits in a sparse region of the domain-specific corpus (97th percentile for distinctiveness): few abstractions share its structure, so a faithful description tends to retrieve it precisely.
Family — Unclustered & Miscellaneous (2551 abstractions)
Nearest neighbors
- Rule-Based System — 0.80
- Spectrum Bias — 0.78
- En ventre sa mere — 0.77
- Drug Repositioning — 0.76
- Blood Culture — 0.76
Computed from structural-signature embeddings · 2026-10-08
Not to Be Confused With¶
A narrative guideline, text-only flowchart, single alert with no demonstrated multistep guideline map, or engine with no encoded guideline knowledge does not by itself meet this identity. WHO level-two Digital Adaptation Kits are software-neutral structured requirements; an L2 kit alone is not proof of a formal case-condition-to-action artifact. Local record integration and evidence tracing are further features, not universal prerequisites.[ref-ae9dc38afeea][ref-6f338f6cfd28]
The PNDS imaging wording is a historical represented recommendation, not advice for care. The two source examples establish formal clinical-guidance artifacts at their stated level of detail. They do not establish all-branch automation, flawless source fidelity, patient safety, observed clinical outcomes or a universal scale of computability.[ref-9b7d5da4835c][ref-ae9dc38afeea][^ref-d28daad00fe8]
References¶
[^ref-ae9dc38afeea]: Mor Peleg, Aziz A. Boxwala, Samson Tu, Qing Zeng, Omolola Ogunyemi, Dongwen Wang, Vimla L. Patel, Robert A. Greenes, and Edward H. Shortliffe, The InterMed Approach to Sharable Computer-interpretable Guidelines, A Review, Journal of the American Medical Informatics Association 11(1) (2004), 1–10, DOI 10.1197/jamia.M1399. The printed title uses a colon before “A Review.” Original author full text, especially “A Shared Guideline Modeling Language,” Table 1, Figure 2, “The Relationship between GLIF and Other Guideline Formalisms,” and Conclusions.
[^ref-ff2c94c4f96d]: Aziz A. Boxwala et al., GLIF3, a representation format for sharable computer-interpretable clinical practice guidelines, Journal of Biomedical Informatics 37(3) (2004), 147–161, DOI 10.1016/j.jbi.2004.04.002. The printed title uses a colon after “GLIF3.” Original article abstract inspected; its three specification levels are not evidence of this chronic-cough case's completed local deployment.
[^ref-9b7d5da4835c]: Mor Peleg, Aziz A. Boxwala, Samson Tu, Robert A. Greenes, Edward H. Shortliffe, and Vimla L. Patel, Handling Expressiveness and Comprehensibility Requirements in GLIF3, MEDINFO 2001, Studies in Health Technology and Informatics 84(Pt 1) (2001), 241–245, DOI 10.3233/978-1-60750-928-8-241. Original author-uploaded full text inspected through indexed text; direct PDF open was unavailable. Printed pp. 242–243 “Definitions, recommendations, and algorithms” and Figure 2 provide the chronic-cough expression and represented PNDS recommendation; pp. 243–244 “Representing decision-support guideline tasks” distinguish case from choice steps. Original PubMed record confirms work identity.
[^ref-d28daad00fe8]: Andreas Seyfang, Silvia Miksch, and Mar Marcos, Combining diagnosis and treatment using ASBRU, International Journal of Medical Informatics 68(1–3) (2002), 49–57, DOI 10.1016/S1386-5056(02)00064-3. Original article abstract and publisher section previews inspected; full article pages and exact numerical thresholds were not independently checked.
[^ref-6f338f6cfd28]: World Health Organization, SMART Guidelines, Digital Adaptation Kits, Implementation research and technical support, official WHO project brief, 6 February 2022, Overview. The printed web title uses a dash and a colon; the linked title uses commas to give the citation binder a full, parseable title basis. The page describes level-two DAKs as software-neutral structured documentation, used here only as a boundary source.