pith. sign in

arxiv: 2602.05407 · v3 · submitted 2026-02-05 · 💻 cs.AI · cs.CL

H-AdminSim: A Multi-Agent Simulator for Realistic Hospital Administrative Workflows with FHIR Integration

Jun-Min Lee , Meong Hi Son , Edward Choi This is my paper

Pith reviewed 2026-05-16 07:32 UTC · model grok-4.3

classification 💻 cs.AI cs.CL
keywords hospital administrationmulti-agent simulationFHIR integrationLLM automationworkflow evaluationhealthcare simulationadministrative taskssimulation framework
0
0 comments X

The pith

H-AdminSim provides a multi-agent simulator with FHIR integration as a standardized testbed for evaluating LLM automation of hospital administrative workflows.

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

Hospital administration involves thousands of daily requests, but prior LLM studies have examined only isolated tasks or patient-facing interactions. This paper introduces H-AdminSim to address the gap by generating realistic data and running multi-agent simulations of full administrative workflows. FHIR integration makes the environment interoperable across different hospital systems. Quantitative rubrics then measure how well various LLMs handle the tasks. The result is a unified testbed for judging the feasibility of LLM-driven automation in complex, high-volume settings.

Core claim

H-AdminSim combines realistic data generation with multi-agent-based simulation of hospital administrative workflows and FHIR integration to create a unified, interoperable environment for testing these workflows across heterogeneous hospital settings and assessing the feasibility and performance of LLM-driven administrative automation.

What carries the argument

H-AdminSim, a multi-agent simulator that generates realistic hospital data, models administrative workflows via agent interactions, and uses FHIR standards to ensure interoperability for LLM evaluation with rubrics.

If this is right

  • LLMs can be compared systematically on complete, multi-step administrative processes instead of isolated subtasks.
  • Testing becomes possible across varied hospital settings through a single FHIR-based interface.
  • Quantitative rubric scores provide concrete metrics for judging automation feasibility at scale.
  • Edge cases in high-volume daily request handling can be explored in a controlled environment before real deployment.

Where Pith is reading between the lines

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

  • If the simulator proves faithful, it could serve as a low-risk sandbox for iterating on LLM tools before any live hospital integration.
  • The framework might support future extensions that incorporate real-time hospital data feeds for ongoing validation.
  • Standardized benchmarks for administrative AI could emerge, allowing consistent progress tracking across research groups.

Load-bearing premise

The combination of generated data and multi-agent interactions in H-AdminSim sufficiently captures the complexity, variability, and edge cases of actual hospital administrative workflows.

What would settle it

Direct comparison of LLM-generated workflow outcomes, error patterns, and decision sequences inside H-AdminSim against anonymized logs from real hospital administrative systems would show whether the simulation matches observed behavior.

Figures

Figures reproduced from arXiv: 2602.05407 by Edward Choi, Jun-Min Lee, Meong Hi Son.

Figure 1
Figure 1. Figure 1: Diagram of the hospital administration simulation. Synthesized hospital data populate the hospital [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Comparison of intake task success rates across models under varying prior-diagnosis settings. Performance improves with higher prior-diagnosis rates and longer patient-staff dialogues (hatched bars). invoked an incorrect tool nor triggered the fallback mechanism for new appointment requests, resulting in the most stable performance in the Scheduling (T). For open-source models, most failures stem from in￾c… view at source ↗
Figure 3
Figure 3. Figure 3: Department assignment errors under six conditions (three prior-diagnosis settings × two conversation-round settings). Errors decrease with more patients having prior diagnoses and with longer conversations. 2.5 Flash. We also investigated the effect of conversa￾tion length by comparing simulations with maximum limits of five rounds (default) and eight rounds [PITH_FULL_IMAGE:figures/full_fig_p011_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Overview of the hierarchical structure of the data synthesis procedure. The left panel presents [PITH_FULL_IMAGE:figures/full_fig_p017_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: System and user prompts used by the LLM for post-processing crawled data and extracting [PITH_FULL_IMAGE:figures/full_fig_p019_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Example of synthesized data used in the outpatient administration simulation, illustrating hospital [PITH_FULL_IMAGE:figures/full_fig_p021_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Example FHIR resource instances used in the simulation. [PITH_FULL_IMAGE:figures/full_fig_p022_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Procedure diagram of the evaluation rubrics. [PITH_FULL_IMAGE:figures/full_fig_p023_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Human evaluation interface used for the intake-task dialogue assessment. Evaluators were presented [PITH_FULL_IMAGE:figures/full_fig_p025_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Figure [PITH_FULL_IMAGE:figures/full_fig_p026_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: System prompt template for role-playing the patient in the intake simulation. Braced elements [PITH_FULL_IMAGE:figures/full_fig_p031_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: The prompt represents the general behavioral guideline for a patient during the intake simulation. [PITH_FULL_IMAGE:figures/full_fig_p032_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: System prompt template for role-playing the administrative staff in the intake simulation. Braced [PITH_FULL_IMAGE:figures/full_fig_p033_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Prompt used by the administrative staff agent for post-dialogue processing in the intake task. The [PITH_FULL_IMAGE:figures/full_fig_p034_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: Patient agent system prompt template for new appointment scheduling. [PITH_FULL_IMAGE:figures/full_fig_p035_15.png] view at source ↗
Figure 16
Figure 16. Figure 16: Patient agent system prompt template for rejecting the initial schedule proposal. [PITH_FULL_IMAGE:figures/full_fig_p036_16.png] view at source ↗
Figure 17
Figure 17. Figure 17: Patient agent system prompt template for appointment rescheduling. [PITH_FULL_IMAGE:figures/full_fig_p037_17.png] view at source ↗
Figure 18
Figure 18. Figure 18: Patient agent system prompt template for appointment cancellation. [PITH_FULL_IMAGE:figures/full_fig_p038_18.png] view at source ↗
Figure 19
Figure 19. Figure 19: Tool-calling prompt for the administrative staff agent. [PITH_FULL_IMAGE:figures/full_fig_p039_19.png] view at source ↗
Figure 20
Figure 20. Figure 20: System and user prompts for the administrative staff agent in the appointment scheduling task. [PITH_FULL_IMAGE:figures/full_fig_p040_20.png] view at source ↗
Figure 21
Figure 21. Figure 21: The prompt represents the scheduling rules for the administrative staff agent during the appointment [PITH_FULL_IMAGE:figures/full_fig_p041_21.png] view at source ↗
Figure 22
Figure 22. Figure 22: Example dialogues from the hospital administration simulation. [PITH_FULL_IMAGE:figures/full_fig_p042_22.png] view at source ↗
read the original abstract

Hospital administration departments handle a wide range of operational tasks and, in large hospitals, process over 10,000 requests per day, driving growing interest in LLM-based automation. However, prior work has focused primarily on patient-physician interactions or isolated administrative subtasks, failing to capture the complexity of real administrative workflows. To address this gap, we propose H-AdminSim, a comprehensive simulation framework that combines realistic data generation with multi-agent-based simulation of hospital administrative workflows. These tasks are quantitatively evaluated using detailed rubrics, enabling systematic comparison of LLMs. Through FHIR integration, H-AdminSim provides a unified and interoperable environment for testing administrative workflows across heterogeneous hospital settings, serving as a standardized testbed for assessing the feasibility and performance of LLM-driven administrative automation.

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.

Referee Report

2 major / 1 minor

Summary. The manuscript proposes H-AdminSim, a multi-agent simulation framework that generates realistic hospital administrative data and models workflows via agent interactions, with FHIR integration to enable interoperability across heterogeneous hospital settings; it further claims to support quantitative LLM evaluation through detailed rubrics and to serve as a standardized testbed for assessing LLM-driven administrative automation.

Significance. If the simulation's fidelity to real workflows can be established, H-AdminSim would address a clear gap in prior LLM healthcare work by providing an interoperable environment for testing complex, multi-step administrative tasks at scale, potentially enabling reproducible comparisons that isolated-subtask benchmarks cannot.

major comments (2)
  1. [Abstract and §3] Abstract and §3 (framework description): the central claim that generated data plus multi-agent interactions plus FHIR integration produce workflows whose complexity, variability, and edge cases match real hospital administration is load-bearing yet unsupported; no quantitative fidelity metrics, no comparison to anonymized hospital logs, and no expert validation study are reported.
  2. [§4] §4 (evaluation): the rubric-based quantitative assessment of LLMs is described but no ablation is shown isolating the contribution of the multi-agent or FHIR components versus synthetic data alone, leaving the interoperability and realism assertions untested.
minor comments (1)
  1. [§3] Notation for agent roles and FHIR resource mappings could be clarified with a small table or diagram in §3 to aid reproducibility.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We are grateful to the referee for the thoughtful and constructive review of our manuscript. The comments have helped us identify areas where the presentation of our contributions can be strengthened. We address each major comment below and describe the revisions we intend to make.

read point-by-point responses

  1. Referee: [Abstract and §3] Abstract and §3 (framework description): the central claim that generated data plus multi-agent interactions plus FHIR integration produce workflows whose complexity, variability, and edge cases match real hospital administration is load-bearing yet unsupported; no quantitative fidelity metrics, no comparison to anonymized hospital logs, and no expert validation study are reported.

    Authors: We acknowledge that the central claims regarding workflow realism and complexity are currently grounded in the design of the data generation process and agent interaction rules, which draw from publicly documented hospital administrative procedures and FHIR resource specifications, rather than from quantitative fidelity metrics or direct comparisons to real hospital logs. No expert validation study is reported in the current manuscript. In the revised version we will expand §3 with a detailed rationale for the modeling choices, add qualitative examples of captured edge cases and variability, and insert an explicit limitations subsection that states the absence of quantitative validation metrics and outlines plans for future expert review and log-based comparisons. These changes will moderate the strength of the claims while preserving the framework as a proposed standardized testbed. revision: yes

  2. Referee: [§4] §4 (evaluation): the rubric-based quantitative assessment of LLMs is described but no ablation is shown isolating the contribution of the multi-agent or FHIR components versus synthetic data alone, leaving the interoperability and realism assertions untested.

    Authors: We agree that ablation experiments would strengthen the evaluation by isolating the contributions of the multi-agent simulation and FHIR integration. The present §4 reports end-to-end rubric scores on the full workflows but does not include such controls. For the revision we will add ablation results comparing LLM performance on synthetic data alone versus the complete multi-agent environment, together with a discussion of how FHIR resources enable cross-setting interoperability. These additional experiments will be reported in the revised §4. revision: yes

Circularity Check

0 steps flagged

No circularity: framework proposal without derivations or fitted reductions

full rationale

The manuscript proposes H-AdminSim as a simulation framework that combines synthetic data generation, multi-agent workflow modeling, and FHIR integration. No equations, parameter-fitting steps, or derivation chains appear in the abstract or described content. Claims of realism and interoperability are presented as design features of the proposed system rather than results obtained by reducing outputs to prior fitted inputs or self-citation chains. The work is therefore self-contained as a framework description and exhibits no circularity of the enumerated kinds.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

The central claim rests on the domain assumption that multi-agent systems plus generated data can faithfully model hospital administrative complexity; the simulator itself is the primary invented entity with no independent evidence supplied.

axioms (1)
  • domain assumption Hospital administrative workflows can be realistically simulated using multi-agent systems and generated data.
    Invoked in the proposal of H-AdminSim as a comprehensive framework.
invented entities (1)
  • H-AdminSim simulator no independent evidence
    purpose: To provide a standardized, FHIR-integrated testbed for LLM performance on hospital administrative workflows
    Newly introduced framework without external validation or falsifiable handles described.

pith-pipeline@v0.9.0 · 5430 in / 1264 out tokens · 36613 ms · 2026-05-16T07:32:23.702162+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

61 extracted references · 61 canonical work pages

  1. [1]

    doi: 10.2196/21929

    ISSN 2291-9694. doi: 10.2196/21929. URL https://medinform.jmir.org/2021/7/e21929. Zhijie Bao, Qingyun Liu, Ying Guo, Zhengqiang Ye, Jun Shen, Shirong Xie, Jiajie Peng, Xuanjing Huang, and Zhongyu Wei. Piors: Personalized intel- ligent outpatient reception based on large language model with multi-agents medical scenario simula- tion, 2024. URL https://arxi...

    work page doi:10.2196/21929 2021

  2. [2]

    BMC Nursing 20(1), 158 (2021) https://doi.org/10.1186/s12912-021-00684-2

    URL https://www.keiseruniversity.edu /primary-secondary-tertiary-and-quaternar y-understanding-levels-of-patient-care/. Yubin Kim, Chanwoo Park, Hyewon Jeong, Yik S Chan, Xuhai Xu, Daniel McDuff, Hyeonhoon Lee, Marzyeh Ghassemi, Cynthia Breazeal, and Hae W Park. Mdagents: An adaptive collaboration of llms for medical decision-making.Advances in Neural Inf...

    work page doi:10.1186/s12912-021-00684-2 2024

  3. [3]

    Junkai Li, Yunghwei Lai, Weitao Li, Jingyi Ren, Meng Zhang, Xinhui Kang, Siyu Wang, Peng Li, Ya-Qin Zhang, Weizhi Ma, et al

    URL https://arxiv.org/abs/2505.17818. Junkai Li, Yunghwei Lai, Weitao Li, Jingyi Ren, Meng Zhang, Xinhui Kang, Siyu Wang, Peng Li, Ya-Qin Zhang, Weizhi Ma, et al. Agent hospital: A sim- ulacrum of hospital with evolvable medical agents. arXiv preprint arXiv:2405.02957, 2024a. Yanzeng Li, Cheng Zeng, Jialun Zhong, Ruoyu Zhang, Minhao Zhang, and Lei Zou. Le...

    work page doi:10.1016/j.compbiomed.2025.109745 2024

  4. [4]

    American College of Osteopathic Internists

    work page

  5. [5]

    General Medical Council

    work page

  6. [6]

    Medical Board of Australia Association (CMA) 8 : gastroenterology, cardiology, pulmonology, endocrinology/metabolism, nephrology, hematology/oncology, allergy, infectious diseases, and rheumatology. A.1.2. Physician Data After hospital and department information are gener- ated, physician data are synthesized according to the predefined number of physicia...

    work page

  7. [7]

    Physician 1Physician k...Physician 1Physician l...Physician 1Physician m

    Canadian Medical Association 16 H-AdminSim Synthesize Hospital 1 Department 1Department 2Department n... Physician 1Physician k...Physician 1Physician l...Physician 1Physician m... Simulation Data #1 Start Hour End Hour Schedules Physicians Departments Hospital Data synthesizing order freefree free appn. 2freeappn. 3 appn. 1 appn. 1 free busybusy appn. 2 ...

    work page 2005

  8. [8]

    NHS Inform Health Encyclopedia Scotland

    work page

  9. [9]

    Summary statistics and representative examples are provided in Table 8

    Seoul National University Hospital Medicine Information clopedia,11 and the Severance Hospital Disease Ency- clopedia.12 Each disease was mapped to one or more of the nine internal medicine specialties. Summary statistics and representative examples are provided in Table 8. The total disease count reported in Table 8 exceeds 194 because diseases treatable...

    work page 2013

  10. [10]

    Asan Medical Center Disease Encyclopedia

    work page

  11. [11]

    Severance Hospital Disease Encyclopedia

    work page

  12. [12]

    metadata

    The official website of HL7 International Practitioner. ThePractitionerresource represents individuals who provide healthcare or related services and contains demographic attributes such as name, gender, birth date, and contact information. In the simulation, this resource is used to represent each physician’s demographic profile. PractitionerRole. ThePra...

    work page 2025

  13. [13]

    Patient agents were configured to reflect typical outpatient characteristics

    supports four configurable personality-related dimensions: (1) personality, (2) language proficiency, (3) confusion level, and (4) medical history recall level. Patient agents were configured to reflect typical outpatient characteristics. Among the six available personality types (neutral, distrustful, impatient, over- anxious, overly positive, and verbos...

    work page 2001

  14. [14]

    You must:

    Respond in a neutral tone without any noticeable emotion or personality.Language ProficiencyIntermediate (CEFR B) Act as a patient with intermediate English proficiency (CEFR B). You must:

    work page

  15. [15]

    Discuss familiar topics confidently but struggle with abstract or technical subjects

    Speaking: Use common vocabulary and form connected, coherent sentences with occasional minor grammar errors. Discuss familiar topics confidently but struggle with abstract or technical subjects. Avoid highly specialized or abstract words

    work page

  16. [16]

    Need clarification or simpler explanations for abstract, technical, or complex information

    Understanding: Can understand the main ideas of everyday conversations. Need clarification or simpler explanations for abstract, technical, or complex information. Words within your level: {understandwords}. Words beyond your level:{misunderstandwords}

    work page

  17. [17]

    Cannot use or understand advanced or specialized medical terms and require these to be explained in simple language

    Medical Terms: Use and understand common medical terms related to general health. Cannot use or understand advanced or specialized medical terms and require these to be explained in simple language. Below are examples of words within and beyond your level. You cannot understand words more complex than the examples provided within your level. Words within ...

    work page

  18. [18]

    No chronic conditions, regular medications, or relevant family medical history are reported.High 1.Accurately remember all health-related information, including past conditions, current medica- tions, and other documented details

    work page

  19. [19]

    Do not forget or confuse medical information

    work page

  20. [20]

    unknown” and “first hospital visit for this symp- tom,

    Consistently ensure that recalled details match documented records. D.1.2. Patient Prompt The patient agent assumes it is interacting with the hospital’s administrative staff, as specified in the in- take prompt. As shown in Figure 11, the prompt consists of three main components: (1) patient in- formation, (2) persona, and (3) behavioral guide- lines. Th...

    work page

  21. [22]

    Ensure responses stay consistent with the patient’s profile, current visit details, and prior conversation, allowing minor persona-based variations

    work page

  22. [23]

    Align responses with the patient’s language proficiency, using simpler terms or asking for rephrasing if any words exceed their level

    work page

  23. [25]

    Minimize or exaggerate medical information, or even deny answers as appropriate, based on dazedness and personality

    work page

  24. [26]

    Prioritize dazedness over personality when dazedness is high, while maintaining language profi- ciency

    work page

  25. [27]

    Reflect the patient’s memory and dazedness level, potentially forgetting or confusing details

    work page

  26. [30]

    11.Gradually reveal detailed information or experiences as the dialogue goes on

    Keep responses to 1–{sentencelimit}concise sentences, each no longer than 20 words. 11.Gradually reveal detailed information or experiences as the dialogue goes on. Avoid sharing all possible information without being asked. 12.Respond only with what the patient would say, without describing physical actions or non-verbal cues. 13.Do not directly reveal d...

    work page

  27. [31]

    The final decision will be made entirely by the administration office based on the symptoms you report

    Do not directly ask which department you should visit. The final decision will be made entirely by the administration office based on the symptoms you report. Figure 12: The prompt represents the general behavioral guideline for a patient during the intake simulation. This guideline is inserted into the{behavioral guideline}placeholder in Figure 11. 32 H-...

    work page

  28. [32]

    First decide if the patient should go to Internal Medicine or Surgery

    work page

  29. [33]

    You may ask up to{totalidx}questions before making your final decision

    Then guide the patient to the most suitable detailed department within that category. You may ask up to{totalidx}questions before making your final decision. Conversation guidelines:

    work page

  30. [34]

    Try to ask for all of the above information at once naturally, rather than separately

    You **must** ask about demographic information to the patient: Name, gender, phone number, personal ID, and address. Try to ask for all of the above information at once naturally, rather than separately

    work page

  31. [35]

    After obtaining the patient’s demographic information, you **must** ask the patient about any previously diagnosed diseases

    work page

  32. [36]

    Ask about: • Main symptom: when it started, how it feels, how long it lasts, and what makes it better or worse (use simple, everyday words), etc

    Focus on the patient’s main problem. Ask about: • Main symptom: when it started, how it feels, how long it lasts, and what makes it better or worse (use simple, everyday words), etc. • Medical history: If the patient has diagnostic records or a diagnosis from a previous hospital, you should make the final decision on the department based on this informati...

    work page

  33. [37]

    Even if the patient does not have medical records or a diagnostic history from a previous hospital, you must not make a medical diagnosis yourself. Your purpose is to assign the most appropriate department for treatment, based on the previous hospital’s diagnosis if available, or on the patient’s symptoms if no such records exist

    work page

  34. [38]

    yellowing of eyes

    Avoid medical jargon. Use everyday words (e.g., say “yellowing of eyes” instead of “icterus”)

    work page

  35. [39]

    If unclear, gently rephrase

    Adjust your questions based on the patient’s answers. If unclear, gently rephrase

    work page

  36. [40]

    I understand that must be uncomfortable

    Show empathy and reassurance (e.g., “I understand that must be uncomfortable.”)

    work page

  37. [41]

    Ask only one short and clear question at a time and keep your answers short (1–2 sentences per turn)

    work page

  38. [42]

    Three examples of the answer format: •‘Answer: 1

    Whenever you are able to determine the patient’s department, you **must** use the following answer format, including the corresponding number from the options below. Three examples of the answer format: •‘Answer: 1. orthopedics‘ •‘Answer: 4. neurology‘ •‘Answer: 3. oncology‘ Current department options in the hospital: {department} This is round{curridx}, ...

    work page

  39. [50]

    You are now the patient

    Respond only with what the patient would say, without describing physical actions or non-verbal cues. You are now the patient. Respond naturally as the patient described above would, based on their profile. Respond in one concise sentence only, with a maximum length of 20 words. Figure 15: Patient agent system prompt template for new appointment schedulin...

    work page

  40. [52]

    Ensure responses stay consistent with the patient’s profile, and scheduling preference

    work page

  41. [57]

    Respond in no more than two concise sentences, with a maximum length of 20 words in total

    work page

  42. [58]

    You are now the patient

    Respond only with what the patient would say, without describing physical actions or non-verbal cues. You are now the patient. Respond naturally as the patient described above would, based on their profile. Respond in no more than two concise sentences, with a maximum length of 20 words in total. Figure 16: Patient agent system prompt template for rejecti...

    work page

  43. [60]

    Ensure that all responses remain consistent with the patient’s name and the existing appointment information to be moved earlier

    work page

  44. [66]

    You are now the patient

    Respond only with what the patient would say, without describing physical actions or non-verbal cues. You are now the patient. Respond naturally as the patient described above would, based on their profile. Respond in one concise sentence only, with a maximum length of 20 words. Figure 17: Patient agent system prompt template for appointment rescheduling....

    work page

  45. [67]

    Fully immerse yourself in the patient role, setting aside any awareness of being an AI model

    work page

  46. [68]

    Ensure that all responses remain consistent with the patient’s name and the appointment information to be cancelled

    work page

  47. [69]

    Align responses with the patient’s language proficiency

    work page

  48. [70]

    Do not explicitly mention the personality

    Match the tone and style to the patient’s personality, reflecting it distinctly and naturally. Do not explicitly mention the personality

    work page

  49. [71]

    Avoid mechanical repetition and a robotic or exaggerated tone

    Keep responses realistic and natural. Avoid mechanical repetition and a robotic or exaggerated tone

    work page

  50. [72]

    Use informal, everyday language

    work page

  51. [73]

    Respond in one concise sentence only, with a maximum length of 20 words

    work page

  52. [74]

    You are now the patient

    Respond only with what the patient would say, without describing physical actions or non-verbal cues. You are now the patient. Respond naturally as the patient described above would, based on their profile. Respond in one concise sentence only, with a maximum length of 20 words. Figure 18: Patient agent system prompt template for appointment cancellation....

    work page 2025

  53. [75]

    If the patient has a preferred doctor, the appointment must be scheduled with that doctor

    work page

  54. [76]

    If the patient wants the earliest possible appointment, compare the available times of the doctors in the patient’s department and schedule the appointment with the doctor who can see the patient the soonest

    work page

  55. [77]

    If the patient wants an appointment after a specific date, compare the availability of doctors in the patient’s department after that date and schedule the appointment with the doctor who can see the patient the soonest after that date

    work page

  56. [78]

    Appointment times must be later than the ”current time” (ISO format) provided in the ”Hospital time information” above

    work page

  57. [79]

    If more than one doctor is available, the appointment should be made with the doctor who has the lower workload (expressed as a percentage)

    work page

  58. [80]

    For example, one doctor’s consultation time may be 0.25 hours, while another’s may be 0.5 hours

    Once the doctor for the appointment is determined, you must schedule according to that doctor’s outpatient consultation duration. For example, one doctor’s consultation time may be 0.25 hours, while another’s may be 0.5 hours

    work page

  59. [81]

    Output the patient’s scheduled appointment as the value of the ’schedule’ key in the JSON format shown below

    work page

  60. [82]

    Schedule appointments between the patient and the doctor while satisfying the above conditions, following the basic principle of booking sequentially from the earliest available date and time

    work page

  61. [83]

    Since there may be gaps in the schedule, carefully check the doctor’s schedule when assigning

    If a patient requests rescheduling due to a previous patient’s appointment cancellation, you **must** find and assign the earliest available date and time slot. Since there may be gaps in the schedule, carefully check the doctor’s schedule when assigning. In this case, appending a time slot may not be needed, and the earliest available time slot should be...

    work page 2025