Source-linked AI summary

Benchmarking Clinical Decision Pathway Adherence in Large Language Models

Nuo Chen, Xinyang Jiang, Zilong Wang, Zhifei Zhang, Xiaoye Qu, Jiajun Deng, Yulan Guo, Cairong Zhao

arXiv:2608.26592v1cs.CL

TL;DR

Medical LLM benchmarks have largely emphasized final-answer accuracy, leaving guideline adherence and decision-process fidelity insufficiently evaluated. MEGA-CDP addresses this gap with a large guideline-derived benchmark, dual task settings, and CDP-oriented pathway evaluation. Across 16 LLMs, the experiments show that outcome accuracy and pathway consistency differ, both settings remain challenging, and medical specialization does not consistently improve adherence.

  • Problem

    Existing medical LLM benchmarks largely focus on final outcomes or general rationale quality rather than faithful adherence to guideline-defined clinical decision pathways.

  • Method

    MEGA-CDP uses an automated pipeline to derive CDPs and case vignettes from clinical guidelines, then evaluates generated pathways in single-turn and multi-turn settings using step-level scoring and DTW-based consistency.

  • Results

    Experiments on 16 LLMs show that outcome accuracy and pathway consistency capture different aspects of decision-making, both task settings remain substantially challenging, and medical models do not consistently outperform general-purpose models.

  • Takeaways & Limitations

    MEGA-CDP supports guideline-adherence evaluation that distinguishes correct outcomes from the consistency of the clinical decision pathway.

  • Takeaways & Limitations

    Evaluation was restricted to models supporting English and Chinese inputs with context windows of at least 40K tokens, using default configurations.

Abstract

from arXiv · show

Following clinical decision pathways (CDPs) defined by clinical practice guidelines is essential for safe and reliable medical decision-making. However, existing medical large language model (LLM) benchmarks mainly evaluate final-answer accuracy, providing limited evaluation of models' ability to adhere to guidelines. To address this gap, we introduce MEGA-CDP, a benchmark for evaluating whether medical LLMs can generate guideline-adherent CDPs using provided guidelines as references. MEGA-CDP is constructed from 2,274 English and Chinese clinical practice guidelines through a guideline-to-case pipeline, yielding 42,353 clinical cases with explicit reference CDPs. It supports both single-turn vignette and multi-turn interactive settings, and introduces a CDP-oriented evaluation framework for measuring pathway consistency. Experiments on 16 representative LLMs show that reliable clinical decision support remains challenging for current models, demonstrating the need for CDP-oriented evaluation and the value of MEGA-CDP for advancing guideline adherence in medical LLMs.

1 Introduction

MEGA-CDP addresses the need to evaluate whether medical LLMs follow guideline-defined clinical decision pathways, not merely reach correct final decisions. It introduces a scalable benchmark and evaluates pathway adherence in single-turn and interactive settings.

  • Clinical decision-making requires multi-step reasoning over heterogeneous patient evidence, so evaluation should assess decision processes alongside final answers.
  • Omitting or misordering intermediate CDP steps can produce clinically unsafe, non-explainable, or difficult-to-review decisions despite correct outcomes.
  • Existing studies largely overlook guideline variation and updates, motivating evaluation of models using provided guidelines as references for faithful CDP generation.
  • MEGA-CDP transforms 2,274 English and Chinese guidelines into over 42,353 pathway-aligned cases and reference CDPs through an automated guideline-to-case pipeline.
  • The benchmark evaluates models in single-turn vignette and multi-turn interactive settings, with a CDP-oriented framework for measuring pathway consistency.

2 Related Work

Medical benchmarks have expanded from exam-style question answering toward richer clinical reasoning and decision-making tasks. However, most existing frameworks do not explicitly encode guideline-based decision structures or systematically assess CDP adherence.

  • MedQA, MedMCQA, and PubMedQA evaluate medical question answering primarily through exam-style tasks, while emrQA and MedDialog add richer clinical contexts.
  • DR.BENCH, MedJourney, and LLMEval-Med extend evaluation toward diagnostic reasoning, progressive-information decision-making, and judged clinical reasoning.
  • MIMIC-CDM, CPGBench, and MedGUIDE evaluate guideline-consistent or CDP-adherent recommendations, including multi-turn and flowchart-based settings.
  • MEGA-CDP represents guidelines as CDPs linked to structured case vignettes and canonical reasoning sequences, enabling process assessment of guideline adherence.

3 MEGA-CDP

MEGA-CDP constructs guideline-derived cases and evaluates model-generated CDPs by aligning predicted and reference decision sequences. Its framework supports both vignette-based and interactive tasks while accounting for variable pathway length and ordering.

  • Data Curation Pipeline: The data pipeline extracts guideline-defined CDPs and synthesizes corresponding clinical case vignettes for downstream evaluation.
  • Framework: Figure 2 organizes MEGA-CDP into guideline collection, guideline-to-case construction, task settings, and CDP-oriented evaluation.
  • Task Settings: The single-turn task provides a guideline and case vignette, requiring a step-by-step predicted CDP and final decision outcome.
  • Task Settings: The multi-turn task has the model ask one question per turn, receive patient information from an environment, and continue until a final decision.
  • CDP-Oriented Evaluation: The evaluation first scores consistency between predicted and reference steps, then aggregates scores into pathway-level consistency using dynamic time warping.
  • CDP-Oriented Evaluation: DTW converts step-consistency scores into alignment costs and preserves decision order while aligning sequences of potentially different lengths.
  • CDP-Oriented Evaluation: The pathway consistency metric reflects order alignment, with lower PCC values indicating better alignment, and is compared with alternative metrics and human judgments.

4 Experiments

Experiments evaluate 16 LLMs on guideline-adherent clinical decision-making in single-turn and multi-turn settings, using outcome and pathway-level measures. Results show substantial variation across models and settings, with multi-turn interaction and pathway adherence remaining especially challenging.

  • Evaluation: PCC measures pathway consistency cost alongside outcome accuracy and success rate across the two evaluation settings.Lower PCC and higher Acc and SR are preferred.
  • Experimental setup: 7,458 cases from 440 English and Chinese guidelines are evaluated in both single-turn vignette and multi-turn interactive settings.The test set contains 210 Chinese and 230 English guidelines, with each instance including a guideline, vignette, outcome, and reference CDP.
  • Main results: Proprietary models generally perform strongest, while specialized medical models do not consistently outperform general models in guideline adherence.Baichuan-M3-235B’s PCC degrades under both settings despite clinical decision-making fine-tuning.
  • Multi-turn interactive setting: 26.98 percentage points higher PCC and 40.51 percentage points lower outcome accuracy occur on average in multi-turn versus single-turn evaluation.Claude-Sonnet-4.6 performs best in the multi-turn setting, with PCC of 55.92% and outcome accuracy of 45.40%.
  • Pathway deviations: Correct final decisions can coexist with process deviations such as omission, redundancy, skipping, and reversal.Figure 3 illustrates this pathway-level inconsistency in a model-generated CDP.
  • Human alignment: 88.5% and 91.5% agreement rates across two human-preference groups yield 90.0% overall agreement between PCC and human judgments.Agreement between the two human annotators is also 90.0%.
  • Metric and judge analysis: GPT-5.4 Mini and Qwen3-8B correlate strongly across metrics, while DTW and OT show higher pairwise consistency when the judge model is fixed.The analysis selects GPT-5.4 Mini as the default judge and DTW as the primary metric.

5 Conclusion

MEGA-CDP evaluates whether medical LLMs generate clinical decision pathways that adhere to guidelines across single-turn and multi-turn settings. Experiments show that pathway consistency and outcome accuracy capture different aspects of decision-making, while both settings remain challenging.

  • MEGA-CDP is a large-scale benchmark for evaluating guideline-adherent clinical decision pathways generated by medical LLMs.
  • The benchmark uses an automated guideline-to-case pipeline to extract guideline-defined pathways and synthesize pathway-aligned clinical cases.
  • MEGA-CDP supports both single-turn vignette and multi-turn interactive evaluation settings.
  • Outcome accuracy and pathway consistency capture different aspects of clinical decision-making.
  • Experiments on 16 representative LLMs show that both task settings remain substantially challenging and specialized medical models do not consistently outperform general-purpose models.

A Reproducibility Details

The experiments combine API-based inference for closed-source models with local GPU-based inference for open-source models. Random seeds were fixed to 42 for stochastic local experiments to support reproducibility.

  • Closed-source models were accessed through official APIs, while open-source models were deployed locally using GPU-based inference.Table 3 summarizes the hardware and software configurations for local inference and evaluation.
  • Random seeds were fixed to 42 for local experiments involving stochastic components.This configuration was used to ensure reproducibility.

B.1 OT-based Measurement

The OT-based measurement models pathway alignment as a global matching problem using step consistency, positional penalties, and flexible transport. PCCOT normalizes the optimal transport cost, with lower values indicating better alignment.

  • OT establishes global correspondences between reference and predicted pathway steps without enforcing strict sequential alignment.The method uses overall matching cost rather than strict sequence-by-sequence alignment.
  • The consistency-based distance is defined as 1 − C_i,j, converting step-level consistency scores into transport costs.
  • A relative positional penalty encourages order-aware alignment by comparing normalized positions of reference and predicted steps.
  • The unified transport cost combines consistency-based distance and positional alignment, with α = 1.0 and β set to 0.2 for GPT-5.4 Mini and Qwen3-8B and 0.1 for BERTScore.
  • The transport plan assigns mass between reference and predicted steps, using uniform marginals within an unbalanced KL-regularized OT formulation.Unbalanced OT permits partial mass variation while softly penalizing deviations from the marginal distributions.
  • PCCOT computes normalized total transport cost from the optimal plan, where lower values indicate better reference–prediction alignment.

B.2 LNDS-based Measurement

The LNDS-based measurement evaluates whether matched pathway steps preserve decision order. It normalizes the longest non-decreasing subsequence length by the number of reference steps, with higher PCSLNDS indicating better adherence.

  • LNDS-based Measurement: LNDS identifies the largest subset of matched steps that preserves decision order, unlike DTW and OT's more flexible matching.
  • LNDS-based Measurement: Candidate matches are selected by highest step-level consistency and retained only when C_i,j_i ≥ γ.Matches below the threshold are treated as unmatched before sequence analysis.
  • LNDS-based Measurement: The longest non-decreasing subsequence of matched predicted-step indices represents the largest order-preserving alignment with the reference pathway.
  • LNDS-based Measurement: PCSLNDS normalizes the subsequence length m′ by the number of reference steps m.
  • LNDS-based Measurement: Higher PCSLNDS indicates better adherence to the guideline-defined decision order.

C Statistics of CDP Lengths

Reference CDP lengths are characterized by pathway-level descriptive statistics and distributions across the full, test, and non-test sets. The test and non-test subsets have similar length statistics, indicating comparable structural characteristics.

  • CDP length counts the non-leaf nodes along a root-to-leaf pathway, representing clinical decision conditions before the final outcome.
  • Figure 5 shows reference CDP-length distributions for the full CDP set and its test and non-test subsets, with the long-tail region magnified.
  • The test and non-test subsets exhibit similar CDP length statistics, suggesting that the held-out test set preserves the overall CDP set’s structural characteristics.

D Additional Evaluation Results

Additional evaluations show that pathway-consistency conclusions are broadly stable across judge models and pathway-level metrics. Narrow bootstrap confidence intervals also indicate stable estimated scores.

  • Similar model rankings and relative performance trends are generally preserved across different judge models.The reported 95% confidence intervals are estimated from 10,000 bootstrap resampling iterations.
  • DTW-PCC, OT-PCC, and LNDS-PCS show consistent relative trends across models in both single-turn and multi-turn settings.These metrics provide broadly consistent assessments despite using different mathematical formulations.
  • Narrow confidence intervals indicate that the estimated pathway-consistency scores are stable.

E Prompt Templates

The prompt templates define MEGA-CDP’s guideline extraction, case-oriented inference, and pathway-consistency scoring procedures. They operationalize clinical pathways as structured decision or questioning sequences and require standardized outputs.

  • Additional Evaluation Outputs: Tables 5–7 define DTW-, OT-, and LNDS-based pathway metrics, with lower PCC values or higher PCS values indicating better pathway consistency.
  • Reasoning Tree Extraction: The reasoning-tree prompt converts guideline content into a JSON decision tree with yes/no branches and patient-specific clinical criteria.It separately records the guideline’s primary clinical focus without representing that focus as a decision node.
  • Reasoning Tree Extraction: Tree construction requires explicit, objective criteria, one independent variable or threshold per node, sequential representation of multiple criteria, and no external medical knowledge.Ambiguous guideline text should be preserved without speculation.
  • Reasoning Tree Extraction: Each decision-tree node receives a unique numeric index and exact supporting guideline sentences drawn from clinically meaningful recommendations or statements.
  • Multi-Turn Step-Level Consistency Scoring: The multi-turn templates represent clinician questioning pathways as ordered lists of questions for comparing reference and candidate information-gathering sequences.
Loading 2608.26592v1…