How it decides

Published rules, applied in this order. The model only finds the note's own words; it never decides.

1. eGFR, CKD-EPI 2009 creatinine equation (drift check only; race coefficient omitted)

   The older 2009 equation, shown only to see whether the answer depends on the equation.

   ckd_epi_2009@1.0 · Levey et al., Ann Intern Med 2009;150:604-12 (2009 eGFR equation, drift check) · 6/6 boundary tests pass · Draft · pending review encoded from the published source; not yet signed off by a clinical reviewer

2. eGFR, CKD-EPI 2021 creatinine equation (race-free)

   Estimate the kidney filtering rate from creatinine, age and sex (2021 equation, no race term).

   ckd_epi_2021@1.0 · Inker et al., N Engl J Med 2021;385:1737-49 (2021 race-free eGFR equation) · 9/9 boundary tests pass · Draft · pending review encoded from the published source; not yet signed off by a clinical reviewer

3. Creatinine-change criteria and staging, with four selectable baselines

   A creatinine rise during a stay: 0.3 within 48 hours, or 1.5 times an earlier value; four ways to pick that value.

   kdigo_aki_2012@1.1 · KDIGO 2012 AKI guideline, Rec 2.1.1-2.1.2 and Table 2 · 5/5 boundary tests pass · Draft · pending review encoded from the published source; not yet signed off by a clinical reviewer

4. GFR and albuminuria categories, and the 3-month persistence rule

   Low filtering rate (below 60) on two results at least 90 days apart; otherwise say what is missing.

   kdigo_ckd_2024@1.1 · KDIGO 2024 CKD guideline (G/A categories, more than 3 months) · 9/9 boundary tests pass · Draft · pending review encoded from the published source; not yet signed off by a clinical reviewer

5. Two-sided comparison of documented kidney statements with the computed criteria

   Compare with what the note says, in both directions; say 'not assessable' when the data cannot decide.

   kidney_compare@1.1 · this tool's comparison rules · 7/7 boundary tests pass · Draft · pending review encoded from the published source; not yet signed off by a clinical reviewer

6. Exclusion gate: conditions in which creatinine-based estimates or creatinine-change criteria are unreliable

   First, check the record for conditions where creatinine misleads (dialysis, kidney transplant, pregnancy, amputation).

   kidney_exclusions@1.0 · KDIGO 2024 CKD guideline (eGFR limitations); local list · 4/4 boundary tests pass · Draft · pending review encoded from the published source; not yet signed off by a clinical reviewer

7. Staff view only: ICD-10-CM ladder and combination-code linkage

   Staff view only: the coding ladder. Never shown to the clinician.

   kidney_staff@1.0 · ICD-10-CM FY2026 guidelines I.A.15, I.C.9.a.2; CMS FY2026 Tables 6I/6J (staff view only) · no boundary tests · Draft · pending review encoded from the published source; not yet signed off by a clinical reviewer

8. Units accepted for creatinine and urine albumin-to-creatinine ratio, and their conversion

   Put every creatinine result in the same unit, and set aside impossible values.

   lab_units@1.1 · KDIGO 2012 and 2024 guidelines, unit conversion (creatinine 1 mg/dL = 88.4 umol/L; ACR 1 mg/mmol = 8.84 mg/g) · 4/4 boundary tests pass · Draft · pending review encoded from the published source; not yet signed off by a clinical reviewer

Earlier versions (4)

• kdigo_aki_2012@1.0 · Creatinine-change criteria and staging, with four selectable baselines
• kdigo_ckd_2024@1.0 · GFR and albuminuria categories, and the 3-month persistence rule
• kidney_compare@1.0 · Two-sided comparison of documented kidney statements with the computed criteria
• lab_units@1.0 · Units accepted for creatinine and urine albumin-to-creatinine ratio, and their conversion

Logic: kidney.py (kidney-engine 2.0). A rule file changed without a new version is refused, and the check refuses to run if any boundary test fails.