Automated Care Pathway Modeling Using Agentic and Knowledge-Aware LLMs
Bibliographic record
Abstract
Clinical pathways (CPWs) translate evidence-based guidance into stepwise care but are often disseminated as free text, obscuring control-flow semantics needed for clarity and computability. Formalizing CPWs as process models - e.g., in the Business Process Model and Notation (BPMN) - improves comprehensibility and enables downstream automation. This thesis designs, implements, and evaluates LLM4CPW, a pipeline for automatic guideline-to-BPMN modeling using large language models (LLMs). We compare two contemporary frameworks - MAO (agentic, multi-role orchestration) and ProMoAI (single-agent with self-refinement loop) - under controlled execution with standardized evaluation. Automated metrics combine node-level and structural similarity (after Dijkman et al.) with graph-edit distance; a clinician provides fidelity ratings with qualitative annotations. We then investigate knowledge-aware modeling via a categorization of recurrent errors and curated clinical statements, testing prompt-only injections versus a dedicated Knowledge Advisor agent placed at different stages of the workflow. Across four Ontario stroke Quality-Based Procedures (QBPs; n = 15 runs per framework), MAO attains higher node similarity (≈ 0.782 vs. ≈ 0.696) and structural similarity (≈ 0.630 vs. ≈ 0.585) than ProMoAI with large effects and p < 0.001, and exhibits markedly lower run-to-run variability; expert ratings also favor MAO. Knowledge-aware variants of MAO yield measurable gains: introducing a Knowledge Advisor after semantic review phase improves node similarity by > 4 points and reduces Graph Edit Distance by ∼ 13 (to∼ 97), with statistically comparable outcomes when placed before review; expert deltas likewise favor the advisor-based designs. Refining statement wording improves medians and interpretability without shifting means. Contributions. This thesis contributes (i) an auditable LLM4CPW pipeline and evaluation protocol; (ii) empirical evidence that agentic orchestration improves fidelity and stability; and (iii) a principled, deployable strategy for knowledge-enhanced modeling via a specialized advisory phase. Collectively, the findings demonstrate the feasibility of reliable, automatically extracted BPMN models from concise clinical guidelines and chart a path toward broader, clinically grounded automation.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.000 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.001 | 0.001 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.002 |
| Open science | 0.000 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one teacher head, not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".