Trustworthy Clinical Decision Support Using Meta-Predicates and Domain-Specific Languages

Dmitry Etin; Eugenia Lvova; Michael Bouzinier; Michael Chumack; Sergey Trifonov

arxiv: 2604.21263 · v1 · submitted 2026-04-23 · 💻 cs.AI · cs.PL· cs.SE· q-bio.QM

Trustworthy Clinical Decision Support Using Meta-Predicates and Domain-Specific Languages

Michael Bouzinier , Sergey Trifonov , Michael Chumack , Eugenia Lvova , Dmitry Etin This is my paper

Pith reviewed 2026-05-09 21:56 UTC · model grok-4.3

classification 💻 cs.AI cs.PLcs.SEq-bio.QM

keywords meta-predicatesepistemological validationclinical decision supportdomain-specific languagesAI auditabilitygenomics variant interpretationdecision rules

0 comments

The pith

Meta-predicates validate the epistemological appropriateness of evidence in clinical decision rules before deployment.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper introduces meta-predicates as checks on predicates in a domain-specific language for clinical logic. These checks draw on a four-dimensional type system that classifies evidence by purpose, knowledge domain, scale, and method of acquisition. The validation identifies rules that rely on unsuitable evidence types, whether the rules originate from humans or from AI generators. It preserves the readability of the original rules and supplies per-decision audit trails. The method is shown on a genomics variant curation platform and positioned as a complement to post-deployment explanation techniques.

Core claim

Decision rules expressed in a clinical DSL can be annotated with meta-predicates that assert which evidence types from the four-dimensional epistemological classification are permissible; enforcing these constraints catches inappropriate evidence use prior to deployment and yields traceable decision paths, as demonstrated by reformulating variant interpretation trees as unate cascades on millions of genomic variants.

What carries the argument

Meta-predicates, which are predicates about predicates that enforce constraints from a four-dimensional epistemological type system (purpose, knowledge domain, scale, method of acquisition) on evidence used inside a domain-specific language for clinical rules.

If this is right

Rules can be audited for evidence compliance before any patient data is processed.
The same checks apply equally to human-authored and AI-generated decision logic.
Reformulation of decision trees as unate cascades produces explicit per-variant audit trails.
The pre-deployment constraints complement post-hoc explanation methods such as LIME and SHAP.
The framework supports regulatory requirements for auditable clinical AI across domains.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

Extending the type system with domain-specific dimensions could adapt the approach to non-clinical fields that require evidence-based audits.
Embedding meta-predicate checks inside automated rule generators might produce only epistemologically valid rules by construction.
The method could reduce reliance on post-deployment monitoring if pre-deployment validation proves sufficient for regulatory sign-off.
Combining the meta-predicates with formal verification tools might yield stronger guarantees than either technique alone.

Load-bearing premise

The four-dimensional epistemological type system accurately and exhaustively classifies all permissible evidence for any clinical decision rule.

What would settle it

A clinical rule rejected by the meta-predicates but later shown by independent expert review or regulatory acceptance to rest on appropriate evidence would disprove the claim that the system reliably catches epistemological errors.

read the original abstract

\textbf{Background:} Regulatory frameworks for AI in healthcare, including the EU AI Act and FDA guidance on AI/ML-based medical devices, require clinical decision support to demonstrate not only accuracy but auditability. Existing formal languages for clinical logic validate syntactic and structural correctness but not whether decision rules use epistemologically appropriate evidence. \textbf{Methods:} Drawing on design-by-contract principles, we introduce meta-predicates -- predicates about predicates -- for asserting epistemological constraints on clinical decision rules expressed in a DSL. An epistemological type system classifies annotations along four dimensions: purpose, knowledge domain, scale, and method of acquisition. Meta-predicates assert which evidence types are permissible in any given rule. The framework is instantiated in AnFiSA, an open-source platform for genetic variant curation, and demonstrated using the Brigham Genomics Medicine protocol on 5.6 million variants from the Genome in a Bottle benchmark. \textbf{Results:} Decision trees used in variant interpretation can be reformulated as unate cascades, enabling per-variant audit trails that identify which rule classified each variant and why. Meta-predicate validation catches epistemological errors before deployment, whether rules are human-written or AI-generated. The approach complements post-hoc methods such as LIME and SHAP: where explanation reveals what evidence was used after the fact, meta-predicates constrain what evidence may be used before deployment, while preserving human readability. \textbf{Conclusions:} Meta-predicate validation is a step toward demonstrating not only that decisions are accurate but that they rest on appropriate evidence in ways that can be independently audited. While demonstrated in genomics, the approach generalises to any domain requiring auditable decision logic.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The paper introduces meta-predicates and a four-dimension type system to enforce evidence appropriateness in clinical DSL rules before deployment, with a genomics demo via unate cascades, but offers no quantitative validation or tests of the taxonomy's completeness.

read the letter

The main thing here is a framework that uses meta-predicates to check if clinical decision rules draw on the right kind of evidence, classified along four dimensions: purpose, knowledge domain, scale, and method of acquisition. They instantiate this in AnFiSA and apply it to variant interpretation rules from the Brigham protocol on 5.6 million Genome in a Bottle variants, while reformulating decision trees as unate cascades to produce clear per-variant audit trails. The claim is that this catches epistemological mismatches early, for both human and AI-generated rules, and complements tools like SHAP by adding upfront constraints while keeping rules readable. This targets a genuine gap in regulatory needs for auditable clinical AI under frameworks like the EU AI Act and FDA guidance. What stands out as new is the specific layering of meta-predicates over an epistemological type system in a clinical DSL, rather than just syntactic validation as in prior formal languages for medical logic. The unate cascade reformulation is a concrete, practical move that directly supports traceability. The design-by-contract foundation fits the problem well and the genomics application shows they have built something that runs at scale. The soft spots are clear and central. The text describes the setup and the demonstration but reports no results—no counts of flagged rules, no error rates, no false-positive or false-negative figures for the checks. There is also no test of the four dimensions against a corpus of real clinical guidelines to check if they are complete or overly restrictive. Without that, the pre-deployment guarantee rests on an assumption about the taxonomy that is not shown to hold. This paper is for researchers and developers working on rule-based clinical decision support, especially in genomics or other areas needing regulatory audit trails. Readers focused on formal methods for trustworthy AI would get value from the architecture and the cascade trick. It shows coherent thinking on the regulatory and formal side and deserves a serious referee to examine the implementation details and require the missing validation experiments. I would send it to peer review.

Referee Report

3 major / 2 minor

Summary. The paper introduces meta-predicates—predicates about predicates—embedded in a domain-specific language (DSL) for clinical decision rules, paired with a four-dimensional epistemological type system (purpose, knowledge domain, scale, method of acquisition) to enforce pre-deployment constraints on evidence types. Drawing on design-by-contract principles, the framework is instantiated in the open-source AnFiSA platform for genetic variant curation; decision trees from the Brigham Genomics Medicine protocol are reformulated as unate cascades to produce per-variant audit trails. The approach is demonstrated on 5.6 million variants from the Genome in a Bottle benchmark and is claimed to catch epistemological errors in both human-written and AI-generated rules while preserving readability and complementing post-hoc explainers such as LIME and SHAP.

Significance. If the meta-predicate mechanism and type system can be shown to be complete and validated, the work supplies a concrete pre-deployment guardrail for epistemological appropriateness that is currently missing from most clinical AI pipelines. The reformulation of decision trees into unate cascades for explicit audit trails is a practical engineering contribution, and the open-source release of AnFiSA enables reproducibility. The framework generalizes beyond genomics to any domain requiring auditable logic under regulatory regimes such as the EU AI Act.

major comments (3)

Methods: The four-dimensional epistemological type system is asserted to classify permissible evidence for any clinical decision rule, yet the manuscript provides no systematic validation—neither expert review of the taxonomy nor testing against a corpus of published clinical guidelines (valid and invalid)—leaving the central guarantee that meta-predicates catch epistemological errors without demonstrated support.
Results: The claim that meta-predicate validation catches errors before deployment on the Brigham protocol is stated without quantitative metrics (false-positive rejection rate, missed-invalid-rule rate, or inter-rater agreement on rule classification), so the empirical demonstration on 5.6 M variants remains descriptive rather than evaluative.
Results: While unate-cascade reformulation is said to enable per-variant audit trails, the manuscript does not report how many rules were actually rejected or modified by the meta-predicates, nor does it compare audit-trail completeness against existing syntactic DSL validators.

minor comments (2)

Abstract and Methods: The term 'unate cascades' is introduced without a brief definition or reference to its prior use in logic synthesis, which may hinder readers unfamiliar with the concept.
Conclusions: The generalization statement to 'any domain requiring auditable decision logic' would benefit from a short discussion of how the four dimensions would be re-instantiated outside genomics.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for their constructive comments and recognition of the work's potential. We respond to each major comment below and will incorporate revisions to strengthen the manuscript.

read point-by-point responses

Referee: Methods: The four-dimensional epistemological type system is asserted to classify permissible evidence for any clinical decision rule, yet the manuscript provides no systematic validation—neither expert review of the taxonomy nor testing against a corpus of published clinical guidelines (valid and invalid)—leaving the central guarantee that meta-predicates catch epistemological errors without demonstrated support.

Authors: We agree that the manuscript would benefit from more explicit validation of the type system. The four dimensions are drawn from established epistemological principles in medicine. In the revised manuscript, we will add a justification subsection in Methods explaining the choice of each dimension with supporting literature. We will also include a limited validation example by classifying evidence types from a selection of published clinical guidelines and showing alignment with the proposed taxonomy. revision: yes
Referee: Results: The claim that meta-predicate validation catches errors before deployment on the Brigham protocol is stated without quantitative metrics (false-positive rejection rate, missed-invalid-rule rate, or inter-rater agreement on rule classification), so the empirical demonstration on 5.6 M variants remains descriptive rather than evaluative.

Authors: The current results focus on the practical demonstration at scale. We acknowledge the value of quantitative metrics for the error-catching claim. In the revision, we will add specific metrics by reporting the number of rules validated, any rejections or modifications made, and results from testing on intentionally flawed rules to estimate false-positive and missed-invalid rates. This will make the evaluation more evaluative. revision: yes
Referee: Results: While unate-cascade reformulation is said to enable per-variant audit trails, the manuscript does not report how many rules were actually rejected or modified by the meta-predicates, nor does it compare audit-trail completeness against existing syntactic DSL validators.

Authors: We will revise the Results section to report the number of rules rejected or modified by the meta-predicates during the demonstration on the Brigham protocol. We will also add a comparison to existing syntactic validators, explaining how the meta-predicate approach provides additional epistemological validation and more complete audit trails beyond syntactic checks. revision: yes

Circularity Check

0 steps flagged

No significant circularity; framework proposes independent type system and meta-predicates

full rationale

The paper introduces meta-predicates and a four-dimensional epistemological type system (purpose, knowledge domain, scale, method of acquisition) drawn from external design-by-contract principles to constrain clinical decision rules in a DSL. It instantiates the framework in AnFiSA and applies it to reformulate decision trees as unate cascades on the Brigham protocol using 5.6M Genome in a Bottle variants, but the type system is not fitted to or derived from this data, nor are claims reduced to self-definitions or self-citation chains. The pre-deployment validation is presented as complementary to post-hoc methods like LIME/SHAP without the central result depending on its own outputs by construction. The derivation remains self-contained against external benchmarks and principles.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 2 invented entities

The central claim depends on the assumption that evidence can be reliably typed along the four dimensions and that meta-predicates can enforce constraints without false negatives or over-restriction; no free parameters are mentioned.

axioms (1)

domain assumption Design-by-contract principles can be extended to epistemological constraints on clinical decision rules
Invoked to introduce meta-predicates as predicates about predicates.

invented entities (2)

meta-predicates no independent evidence
purpose: Assert epistemological constraints on clinical decision rules
New construct introduced in this work to enable pre-deployment validation.
epistemological type system no independent evidence
purpose: Classify annotations along purpose, knowledge domain, scale, and method of acquisition
Invented classification scheme to support meta-predicate enforcement.

pith-pipeline@v0.9.0 · 5618 in / 1349 out tokens · 39882 ms · 2026-05-09T21:56:22.460640+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

3 extracted references · 3 canonical work pages

[1]

AnFiSA: An open-source computational platform for the analysis of sequencing data for rare genetic disease,

M. Bouzinier, D. Etin, N. Khoshnevis, M. Shad, and S. Yockel, Research Data that Can be Trusted , 1st ed. in SpringerBriefs in Computer Science. Cham: Springer, 2026. [Online]. Available: https://link.springer.com/book/9783032210319 [2] M. A. Bouzinier et al. , “AnFiSA: An open-source computational platform for the analysis of sequencing data for rare gen...

work page doi:10.1016/j.jbi.2022.104174 2026
[2]

CQL Specification - Clinical Quality Language Specification v1.5.3

“CQL Specification - Clinical Quality Language Specification v1.5.3.” Accessed: Apr. 02, 2026. [Online]. Available: https://cql.hl7.org/ [15] A. A. Boxwala et al. , “GLIF3: a representation format for sharable computer-interpretable clinical practice guidelines,” J. Biomed. Inform. , vol. 37, no. 3, pp. 147–161, Jun. 2004, doi: 10.1016/j.jbi.2004.04.002. ...

work page doi:10.1016/j.jbi.2004.04.002 2026
[3]

The halo effect: Evidence for unconscious alteration of judgments,

R. E. Nisbett and T. D. Wilson, “The halo effect: Evidence for unconscious alteration of judgments,” J. Pers. Soc. Psychol. , vol. 35, no. 4, pp. 250–256, 1977, doi: 10.1037/0022-3514.35.4.250. [32] J.-M. Jézéquel and B. Meyer, “Design by Contract: The Lessons of Ariane,” Computer , vol. 30, no. 1, pp. 129–130, Jan. 1997, doi: 10.1109/2.562936. [33] L. de...

work page doi:10.1037/0022-3514.35.4.250 1977

[1] [1]

AnFiSA: An open-source computational platform for the analysis of sequencing data for rare genetic disease,

M. Bouzinier, D. Etin, N. Khoshnevis, M. Shad, and S. Yockel, Research Data that Can be Trusted , 1st ed. in SpringerBriefs in Computer Science. Cham: Springer, 2026. [Online]. Available: https://link.springer.com/book/9783032210319 [2] M. A. Bouzinier et al. , “AnFiSA: An open-source computational platform for the analysis of sequencing data for rare gen...

work page doi:10.1016/j.jbi.2022.104174 2026

[2] [2]

CQL Specification - Clinical Quality Language Specification v1.5.3

“CQL Specification - Clinical Quality Language Specification v1.5.3.” Accessed: Apr. 02, 2026. [Online]. Available: https://cql.hl7.org/ [15] A. A. Boxwala et al. , “GLIF3: a representation format for sharable computer-interpretable clinical practice guidelines,” J. Biomed. Inform. , vol. 37, no. 3, pp. 147–161, Jun. 2004, doi: 10.1016/j.jbi.2004.04.002. ...

work page doi:10.1016/j.jbi.2004.04.002 2026

[3] [3]

The halo effect: Evidence for unconscious alteration of judgments,

R. E. Nisbett and T. D. Wilson, “The halo effect: Evidence for unconscious alteration of judgments,” J. Pers. Soc. Psychol. , vol. 35, no. 4, pp. 250–256, 1977, doi: 10.1037/0022-3514.35.4.250. [32] J.-M. Jézéquel and B. Meyer, “Design by Contract: The Lessons of Ariane,” Computer , vol. 30, no. 1, pp. 129–130, Jan. 1997, doi: 10.1109/2.562936. [33] L. de...

work page doi:10.1037/0022-3514.35.4.250 1977