AKI Evidence Matrix — Operationalisable Risk Prediction

	A	B	C	D	E	F	G	H
1	AKI Evidence Matrix
2	Operationalisable Clinical Risk Prediction · AKI demonstrator

3
4	Authority	HOLD	Target / features / model	No global accepted/proposed contract	Reviewed references	71	Last checked	3 Oct 2026
5	Programme	Operationalisable Clinical Risk Prediction	Subproject	AKI demonstrator	T1 comparators	15	Needs evidence	1
6	Repository	csmmei/prospective-medication-aware-aki	Matrix role	Evidence navigation + synthesis/project-design crosswalk — not scientific or execution authority.
7
8	Workbook map
9	G01–G03 Governance	Project crosswalk → research gate → evaluation gate	D04 / D21 Definitions	Operational/computable anchors → medication entity and terminology layer	E10–E15 / E22–E23 Evidence	Guidance → dynamic studies → phenotype/design/input/feature detail → medication claims and BNF interactions	C16–C17 / C24 Clinical synthesis	Phenotype catalogue + evidence-linked relationships → medication relation synthesis
10	P18–P20 / P25–P26 Project specification	Project inputs → feature–phenotype crosswalk → implementation/QC → medication observability and candidate review	M90–M99 Maintenance	Dynamic SR crosswalk → fulltext queue → UK guidance audit → Edinburgh AKI map → hidden medication lookups → references	Recommended path	G01 → G02 → G03 → D04/D21 → E10–E15/E22–E23 → C16/C17/C24 → P18–P20/P25–P26 → M90–M93 → M99	Current authority	HOLD — the complete target/features/model remain neither proposed nor accepted. Candidate.8 remains canonical non-executable; use CURRENT_AUTHORITY.yaml for scientific/execution authority.
11	Prefix legend	G governance · D definitions · E evidence · C clinical synthesis · P project specification · M maintenance. Tabs are ordered by layer (G→D→E→C→P→M), not by numeric suffix; medication tabs sit inside their matching layer.	Evidence boundary	E tabs describe source evidence; C tabs synthesise clinical/medication concepts; P tabs map those concepts into project representations without creating scientific or execution authority.	Key detail tabs	E14 study/model inputs · E15 literature feature detail · D21 medication entities · E22/E23 medication claims/interactions · C24 medication relations · P25/P26 medication observability/review · M93 Edinburgh map · M99 references	Visual conventions	SCr = blue · UO = orange · KRT = purple · NO/None = grey
12	Legacy MLP reuse	Structure/provenance ideas are adapted selectively; old imputation, thresholds and nephrotoxic labels are not inherited by default.	Authority rule	The matrix documents evidence and design state; it does not promote target/features/model.	Candidate.8	Current self-contained target-spec version: defines who is eligible, when predictions occur, what counts as first prospectively ascertainable KDIGO Stage ≥2, and how SCr/UO/KRT, missingness, timing, terminal events and revisions are handled. It is not a model, dataset or literature source.	Last checked	3 Oct 2026
13	Rules
14	Scope	Targeted adult hospital/ICU evidence map for dynamic/severe AKI prediction, phenotype/ascertainment, medication context and actionability, validation and translation. Not a registered systematic review.
15	Tiers	T1 = core comparator; T2 = supporting comparator; T3 = lineage / extended methods.
16	Evidence boundary	NR / not verified stays unresolved. Do not infer missing details or compare headline performance across different populations, targets, horizons or validation designs as if equivalent.
17	Source rule	Use the primary paper for paper-specific claims and inspect the official/paper-reported repository when implementation details matter.
18	Citekeys	External sources that may be cited in a paper or thesis use their exact PhD Research Better BibTeX citekey, including institution-authored guidelines and reports. Internal governance/project syntheses retain project source keys. Canonical GitHub main currently contains 631 active-parent Zotero manifest records at main b06d425; Zotero reconciliation branch commit 924a964 contains 664 active parents but is not yet canonical.
19	Column names	Shared concepts use canonical headers: Citekey, Source key, Evidence type, Review status, Setting / population, Model / method, Caution / limitation, Source URL and Notes. CE/CA/PE are written in full on overview tabs and abbreviated to CE/CA/PE on downstream comparison tabs for compact navigation. Domain values use shared sortable stage prefixes: 00 Governance; 10 Design; 20 Development; 30 Validation; 40 Implementation; 50 Actionability; 90 Historical. Rows in the same stage share the same prefix.
20	Code meaning
21	PR	Project relevance — how important the source is to current project decisions.
22	CE	Clinical evidence strength — evidence about care, action or outcomes.
23	CA	Clinical applicability — how closely the study maps to the intended AKI clinical task.
24	PE	Prediction / methodological evidence strength.
25	Codes & scoring legend
26	P / T / M / V	Domain: phenotype / temporal design / medication-context / validation.
27	NR	Not reported or not verified — leave unresolved.
28	PR1 / PR2 / PR3	Project relevance priority: PR1 = current decision relevance; PR2 = supporting evidence; PR3 = lineage / extended context. Not an evidence-strength rating.
29	CE1	Direct clinical evidence: interventional evidence or pooled trial evidence directly informing care, action or outcomes.
30	CE2	Strong supporting clinical evidence: prospective implementation, quality-improvement or strong clinical-association evidence.
31	CE3	Indirect clinical evidence: observational evidence or structured expert/consensus evidence.
32	CE4	Contextual clinical evidence: narrative, commentary or clinical framing rather than direct effect evidence.
33	CA1	Direct clinical applicability: direct or near-direct match to the intended AKI clinical task.
34	CA2	Close clinical applicability: broadly relevant, with material differences in population, outcome, setting or timing.
35	CA3	Indirect clinical applicability: specialised population/task or substantial mismatch with the intended use.
36	CA4	Low direct clinical applicability: mainly methodological or contextual for the intended AKI use.
37	PE1	Direct / strongest prediction-method evidence: external or prospective validation, or systematic synthesis directly addressing the prediction-method question.
38	PE2	Strong supporting prediction-method evidence: internal or temporal validation, multi-dataset evidence, or strong methodological evaluation.
39	PE3	Indirect prediction-method evidence: method proposal, limited validation, or structured methodological evidence.
40	PE4	Contextual methodological evidence: narrative or conceptual methodological framing rather than direct validation.
41	Strength blanks	The source is not scored on that dimension. For prediction papers, CE may appropriately be blank; use CA for clinical applicability and PE for prediction/method strength. In review-crosswalk rows, blank may also mean the primary paper has not yet been sufficiently verified. CE/CA/PE are navigation aids, not formal GRADE/PROBAST ratings or project authority.
42	Current framing
43	Programme	Clinically meaningful, knowledge-informed, actionability-aware and operationalisable clinical risk prediction.
44	AKI role	Current demonstrator. Generic AKI risk is prognostic; it is not automatically preventable or actionable.
45	Knowledge	External clinical and pharmacological knowledge can inform representation, constraints, interpretation and downstream action pathways. Potential value includes improved prediction, calibration, robustness, interpretability and actionability; these benefits should be tested empirically using prospectively groundable, evidence-bounded knowledge.
46	Medication	Core contextual/actionability workstream, not the whole programme.
47	Actionability	Prediction should be interpreted against the intended decision and actionable context. Keep prognostic risk, preventable risk, treatment-responsive risk and actionable risk distinct.
48	Evidence & provenance
49	Project notes	REV-PHENO-01; REV-TRANS-01; REV-MED-08; REV-INTUSE-07; REV-ASCERT-01.
50	Zotero	PhD Research is the citation authority for this workflow. Canonical GitHub main currently contains 631 active-parent citekey-manifest records at main b06d425. Zotero reconciliation branch commit 924a964 contains 664 active parents and remains non-canonical until merged.
51
52
53
54
55
56
57
58
59