Proactive Knowledge Inquiry in Doctor-Patient Dialogue: Stateful Extraction, Belief Updating, and Path-Aware Action Planning

Zhenhai Pan; Yan Liu; Jia You

cs.AIMar 18, 2026

Proactive Knowledge Inquiry in Doctor-Patient Dialogue: Stateful Extraction, Belief Updating, and Path-Aware Action Planning

Authors: Zhenhai Pan, Yan Liu, Jia You

Organizations: The Hong Kong Polytechnic University, Hong Kong

Abstract

Most automated electronic medical record (EMR) pipelines remain output-oriented: they transcribe, extract, and summarize after the consultation, but they do not explicitly model what is already known, what is still missing, which uncertainty matters most, or what question or recommendation should come next. We formulate doctor-patient dialogue as a proactive knowledge-inquiry problem under partial observability. The proposed framework combines stateful extraction, sequential belief updating, gap-aware state modeling, hybrid retrieval over objectified medical knowledge, and a POMDP-lite action planner. Instead of treating the EMR as the only target artifact, the framework treats documentation as the structured projection of an ongoing inquiry loop. To make the formulation concrete, we report a controlled pilot evaluation on ten standardized multi-turn dialogues together with a 300-query retrieval benchmark aggregated across dialogues. On this pilot protocol, the full framework reaches 83.3% coverage, 80.0% risk recall, 81.4% structural completeness, and lower redundancy than the chunk-only and template-heavy interactive baselines. These pilot results do not establish clinical generalization; rather, they suggest that proactive inquiry may be methodologically interesting under tightly controlled conditions and can be viewed as a conceptually appealing formulation worth further investigation for dialogue-based EMR generation. This work should be read as a pilot concept demonstration under a controlled simulated setting rather than as evidence of clinical deployment readiness. No implication of clinical deployment readiness, clinical safety, or real-world clinical utility should be inferred from this pilot protocol.

Figures & tables

Panel of Figure 1 — Figure 1: Conventional linear EMR generation versus the proposed proactive inquiry loop. The main control loop updates dialogue state and feeds back into the ongoing interaction, while the EMR is generated as a structured output branch from the evolving state. The figure emphasizes that the proposed framework intervenes before note finalization by querying unresolved clinically relevant gaps.

State	Interpretation	Control implication
observed_result	Current observed finding or measured result	Strongly updates state and belief
historical_result	Previously known finding or past exam result	Supports history with lower immediacy
recommended	Suggested future action	Triggers planning; not completed evidence
pending_verification	Mentioned but unconfirmed	Gap source; often triggers verification
negated	Explicitly denied	Supports exclusion and contradiction handling

Table 1: Representative examples from the state taxonomy used in stateful extraction.

Panel of Figure 2 — Figure 2: Hybrid retrieval in the proposed framework: vector coarse retrieval narrows the candidate set, object-level reranking aligns candidates with state structure, and path-aware reasoning prioritizes clinically actionable support.

Step	Operation
1	Extract stateful events $E_{t}\leftarrow\mathrm{StatefulExtract}(u_{t})$ .
2	Update current state $C_{t}\leftarrow\Phi(C_{t-1},E_{t})$ .
3	Update belief $B_{t}\leftarrow\mathrm{BeliefUpdate}(B_{t-1},E_{t})$ .
4	Derive gaps $\Delta_{t}\leftarrow\Delta(C_{t},G_{t})$ .
5	Retrieve supportive objects and candidate paths from knowledge base $K$ .
6	Generate candidate actions from state, belief, gaps, and retrieval outputs.

Table 2: Turn-level control loop of the proactive knowledge-inquiry controller.

Method	Coverage (%)	Risk Recall (%)	Structural Completeness (%)	Redundancy (%)	$T_{\mathrm{goal}}$
Baseline A: Direct generation	55.6	30.0	59.3	N/A	N/A
Baseline B: Chunk-only RAG	69.4	60.0	71.4	24.8	6.8
Baseline C: Rule-template questioning	77.8	70.0	78.6	28.4	7.4
Ours	83.3	80.0	81.4	15.8	5.8

Table 3: Pilot evaluation under a controlled simulated setting. Higher is better for Coverage, Risk Recall, and Structural Completeness. Lower is better for Redundancy and

T_{\mathrm{goal}}

. Baseline A is non-interactive; therefore Redundancy and

T_{\mathrm{goal}}

are not applicable.

Metric	Chunk-only RAG	Hybrid Retrieval
Recall@5	0.770	0.870
MRR@5	0.560	0.710
nDCG@5	0.680	0.790
Object hit rate	0.71	0.83
Path hit rate	0.54	0.73

Table 4: Retrieval results on a 300-query benchmark aggregated across dialogues.

Appendix figures & tables2 assets

Supplementary material from the paper’s appendix.

Appendix

Field	Role in scoring	Example values	Used by metrics
section	Record region / discourse region	HPI, ROS, Plan	Coverage, Structural Completeness
slot	Canonical clinical variable	cough, chest_pain, chest_xray	Coverage, Structural Completeness
normalized_value	Normalized semantic value	present, absent, duration=3d	Coverage, state matching
status	Completion state	completed, recommended, not_done	Coverage, action logic
temporality	Time anchoring	present, recent_past, future	Coverage, risk interpretation
assertion	Polarity / certainty	positive, negative, proposed	Coverage, risk interpretation

Appendix

Table 5: Reference gold schema used to define the audit units in the controlled pilot protocol.

Query point type	Source state / gap	Example prompt	Risk-critical?
Allergy verification	Medication plan present but allergy status unresolved	“Before recommending antibiotics, has penicillin allergy been explicitly confirmed?”	Yes
Medication clarification	Drug mentioned without normalized dosage or frequency	“What dosage and frequency should be recorded for the current antihypertensive medication?”	No
Symptom duration follow-up	Chief complaint present but temporal detail missing	“How long has the chest discomfort been present, and is it worsening?”	Yes
Red-flag escalation check	Risk-sensitive symptoms partially observed but escalation status unresolved	“Given persistent upper-abdominal pain with vomiting, should urgent testing or referral be prompted?”	Yes
Exam / test completion status	Imaging or examination recommended but completion state unresolved	“Was the chest X-ray already completed, or is it only being recommended at this stage?”	No

Appendix

Table 6: Illustrative protocol-defined retrieval query points from the pilot audit. These examples are derived from unresolved state gaps in the gold schema rather than from naturally occurring user queries.

State Tracking

State tracking studies how AI systems maintain and update information about entities, tasks, or environments across sequential observations and actions.

See all 99 papers

A Proactive EMR Assistant for Doctor-Patient Dialogue: Streaming ASR, Belief Stabilization, and Preliminary Controlled Evaluation
Mar 18, 2026Zhenhai Pan, Yan Liu, Jia YouElectronic Health Records Medical Report Generation
The Hong Kong Polytechnic University, Hong Kong
PCQC: Privileged Counterfactual Question Credit for Multi-Turn Medical Dialogue
Sep 23, 2026Chenxuan Li, Jiayi Wan, Xinrong Chen +3Clinical Reasoning Medical LLMs
Peking University, Beijing, China · Department of Medical Bioinformatics, School of Basic Medical Sciences, Peking University, Beijing, China
Beyond Information Seeking: Severity-Aware Question Supervision for Proactive Medical Dialogue
Aug 25, 2026Chenxuan Li, Xinrong Chen, Luyan Zhang +5Clinical Decision-Making
Peking University, Beijing, China · Northeastern University, Boston, MA, USA · Sun Yat-sen University, Guangzhou, China +1

Mar 18, 2026J/K move · Enter open · S save

A Proactive EMR Assistant for Doctor-Patient Dialogue: Streaming ASR, Belief Stabilization, and Preliminary Controlled Evaluation

Zhenhai Pan, Yan Liu, Jia You

The Hong Kong Polytechnic University, Hong Kong

Electronic Health Records Medical Report Generation Clinical Speech Processing Clinical Decision Support

Open paper AskPDF arXiv

Most dialogue-based electronic medical record (EMR) systems still behave as passive pipelines: transcribe speech, extract information, and generate the final note after the consultation. That design improves documentation efficiency, but it is insufficient for proactive consultation support because it does not explicitly address streaming speech noise, missing punctuation, unstable diagnostic belief, objectification quality, or measurable next-action gains. We present an end-to-end proactive EMR assistant built around streaming speech recognition, punctuation restoration, stateful extraction, belief stabilization, objectified retrieval, action planning, and replayable report generation. The system is evaluated in a preliminary controlled setting using ten streamed doctor-patient dialogues and a 300-query retrieval benchmark aggregated across dialogues. The full system reaches state-event F1 of 0.84, retrieval Recall@5 of 0.87, and end-to-end pilot scores of 83.3% coverage, 81.4% structural completeness, and 80.0% risk recall. Ablations further suggest that punctuation restoration and belief stabilization may improve downstream extraction, retrieval, and action selection within this pilot. These results were obtained under a controlled simulated pilot setting rather than broad deployment claims, and they should not be read as evidence of clinical deployment readiness, clinical safety, or real-world clinical utility. Instead, they suggest that the proposed online architecture may be technically coherent and directionally supportive under tightly controlled pilot conditions. The present study should be read as a pilot concept demonstration under tightly controlled pilot conditions rather than as evidence of clinical deployment readiness or clinical generalizability.

Abstract

State Tracking

Explore similar work

A Proactive EMR Assistant for Doctor-Patient Dialogue: Streaming ASR, Belief Stabilization, and Preliminary Controlled Evaluation

PCQC: Privileged Counterfactual Question Credit for Multi-Turn Medical Dialogue

Beyond Information Seeking: Severity-Aware Question Supervision for Proactive Medical Dialogue