evidence-briefs

Instruction file imported from alexanderb95-jpg/OncologyWiki (.cursor/rules/evidence-briefs.mdc). Copyright stays with the author.

Evidence pages

• Path: pathways/{domain}/{disease}/{setting}/evidence.md (e.g. pathways/gu/prostate/mHSPC/)
• One disease + setting per page. Use short named sections, nested bullets, and tables only for genuinely tabular comparisons; do not merge a disease's settings into a long page.
• Purpose: scannable decision support, not a guideline dump.
• Required sections, in order: Who this applies to, Standard options, Landmark evidence, Biomarkers, Watch list, Toxicity, Guideline references, Sources, Changelog. Keep Toxicity near the bottom—after clinical and watch-list content, but before Guideline references and Sources—and render the paired Epic phrase before Guideline references, so guidelines and sources remain at the page bottom.
• Guidelines: keep a brief bottom reference to current NCCN, EAU, ESMO, and AUA/ASCO/SUO guidance as applicable to that disease + setting; include a published category/strength only when verified and never paste copyrighted guideline text.
• Standard options: use a risk group × option table with key inclusion and typical duration when relevant.
• Eligibility: state the pivotal trial's histology, risk groups, key timing window, and noneligible boundary in a dedicated table or crisp bullets. Do not blend guideline timing with trial eligibility.
• Landmark evidence: use a trial table with N, population, randomization, primary endpoint, DFS/RFS/OS HRs, and published landmark rates with years. For a practice-changing survival endpoint, add full source-recreated KM curves. Use reported KM data where available; when a published figure must be digitized, label the reconstruction as approximate and cite the source and data cutoff. Never reuse journal artwork or present interpolated values as reported data.
• Cross-trial comparison: whenever Landmark evidence compares two or more trials, place the standalone marker <!-- .cross_trial --> immediately before its markdown table. Use columns for Trial, Population, Intervention, Comparator, Primary endpoint, Clinically meaningful outcomes, and Generalizability-critical differences. Report hazard ratios and published landmark rates where available; do not add ARR, NNT, or other derived risk metrics.
• Watch list: include only named ongoing trials or high-impact recent papers that could change care; state whether evidence is peer-reviewed, congress-only, or pending. Keep generic research gaps and routine re-check reminders out of this section.
• Toxicity: cover every approved regimen discussed for the setting in one concise, trial-first table with grade ≥3 overall events, immune-mediated events where applicable, discontinuations, and clinically important named toxicities. For named toxicities, retain only trial- or label-reported events occurring in >5% of patients at any grade; prefer those with a grade ≥3 signal. Keep regimen-level safety outcomes even though they are not individual toxicities. Do not list or mirror rare (<1%) named events in counseling unless they are clinically critical. Include real-world evidence only when it adds a distinct, useful clinic point; label its setting and do not import non-comparable rates. Add short plain-language patient counseling on retained toxicity symptoms that require prompt contact and the potential for lasting toxicity, and mirror this counseling in the paired dot phrase.
• Formatting: use bold only for necessary document structure; prefer headings and plain prose over inline emphasis.
• Add a paired dotphrase.md in the same folder for decision counseling. Its first line is an Epic trigger beginning with #; update it when the evidence page's standard-of-care discussion changes.
• Do not use a Bottom line section.
• Header fields: Last reviewed, Next review (~90 days), Owner, Purpose, Status
• Never invent trial statistics; verify or leave qualitative. Replace checked-source hedging with a working source link; retain explicit uncertainty only when the source itself is immature, pending, or unreported.
• On every edit: set Last reviewed and Next review (YYYY-MM-DD) and append Changelog
• After evidence edits: python3 scripts/build_wiki.py so site/ stays in sync
• Sister EpiAI project stays at /Users/Alex/Documents/EpiAI