Why are cancer guidelines stuck in PDFs?
Cancer treatment guidelines are typically published as long, static PDFs, which many see as durable and universally accessible but poorly suited for automation, integration with electronic health records, or rapid updates as research evolves. Commenters debate whether these guidelines should also be released in structured, machine-readable formats (e.g., FHIR, CQL, DAGs) to power clinical decision support, while highlighting technical, legal, economic, and organizational barriers, including vendor incentives and regulatory risk. Others caution that oncology is complex, evidence is often incomplete, and over‑reliance on opaque tools or AI—without clear accountability—could harm patients or slow the field’s ability to adapt to new science.
Why guidelines stay in PDFs
- PDFs are seen as durable, portable, universally viewable, and stable across devices and decades; they “just work” and are easy to email and print.
- Many clinicians prefer a single, shared PDF over practice‑specific tools that may be brittle, locked behind logins, or go down.
- Critics note PDFs are poorly machine-readable, often lack semantic structure, and make downstream parsing expensive and error‑prone, yet are used as systems of record anyway.
- Some point out PDF can embed structured data (XML/JSON, structure trees), but authoring tools and workflows rarely exploit this.
Push for structured / computable guidelines
- Several argue guidelines are fundamentally decision trees/DAGs, and should be published in machine-interpretable formats alongside PDFs.
- Suggested benefits: EMRs could offer context-aware prompts, automatic test suggestions, consistency checks, and generate PDFs from a single source of truth.
- Existing standards and efforts mentioned: HL7 FHIR (PlanDefinition, CQL), CDS Connect, WHO SMART Guidelines, FHIR clinical reasoning specs, and earlier “computable clinical guidelines” and expert systems.
Complexity, skepticism, and limits
- Implementers report that encoding real guidelines is hard: ambiguous clinical concepts, varying local semantics, incomplete evidence, and frequent guideline changes.
- There is concern about guidelines becoming constrained by whatever data model or spaghetti code exists, drifting away from cutting‑edge clinical knowledge.
- Some argue guidelines are not true decision trees; many branches rest on weak or non‑differentiating evidence and require human judgment or patient‑specific tradeoffs.
- Standards like FHIR/CQL are powerful but perceived as complex and intimidating for small teams.
Incentives, industry, and access
- Commenters highlight business incentives: organizations charge for structured “template” data or EMR integration while offering only PDFs freely.
- Lab and EMR vendors are criticized for poor implementations and profit‑driven underinvestment in quality and interoperability; others attribute failures partly to inherent difficulty and lack of clear ROI.
- Licensing barriers (e.g., for structured cancer protocols or value sets) are seen as gatekeeping that hinders open tooling.
AI/ML and decision support
- Some envision AI/LLMs as key to extracting structure or even generating treatment plans; others warn about hallucinations, bias, and lack of proven outcome benefits.
- Debate over complex opaque models vs interpretable decision trees: advanced models may be more accurate in principle but are hard to explain, regulate, and defend in court.
Clinician behavior and role of guidelines
- There is disagreement over how diligently doctors, especially oncologists, keep up with rapidly evolving literature.
- Several emphasize guidelines as frameworks for informed teams, not strict algorithms for identical patients; structured rules should remain secondary to clinical expertise and ongoing research.