Title : Diseasequest: Multi-agent AI and reasoning analytics for infectious disease management in medical education
Abstract:
Background: Infectious-disease management requires clinicians to integrate host factors, epidemiology, examination, laboratory and imaging data, microbiology, treatment response, antimicrobial stewardship, prevention, and communication under uncertainty. Yet learners have limited opportunities to repeat these decisions safely while faculty observe the reasoning process. DiseaseQuest was developed as a gamified, multi-agent virtual-patient platform that converts clinical actions into structured, analyzable reasoning data.
Platform and Case Design: The working prototype organizes encounters through introduction, mentor preparation, patient interview, physical examination, diagnostic orders, treatment, differential and final diagnosis, and evaluation. Specialized Patient, Mentor/Tutor, Diagnostic, and Evaluator agents operate under progressive-disclosure rules that prevent premature release of diagnoses, results, and outcomes. An orchestrator governs stage transitions, role boundaries, and test-result availability. Learner questions, examinations, orders, interpretations, treatments, timing, communication, and reflections can be mapped to transparent 100-point rubrics and behaviorally anchored Q100 scoring hooks. This event-level architecture supports de-identified individual and class analytics rather than scoring only final-answer accuracy. At scale, pooled event streams could reveal diagnostic delay, premature closure, over-testing, delayed escalation, unnecessary broad-spectrum exposure, missed prevention steps, and variation across cases, learner levels, or institutions.
Infectious-Disease Portfolio: The Amelia prototype and two adult cases in development test generalizability across pathogens, ages, settings, and decision horizons. Amelia Thompson, a 9-month-old with pneumococcal meningitis, progresses through lumbar-puncture reasoning, empiric therapy, seizures, septic shock, pediatric intensive care, microbiologic confirmation, and antibiotic narrowing. John Whitman, an adult living with HIV with interrupted antiretroviral therapy and discontinued Pneumocystis prophylaxis, presents with severe pneumococcal community-acquired pneumonia, hypoxemia, altered mentation, and evolving sepsis. Learners must distinguish bacterial pneumonia from Pneumocystis pneumonia, viral pneumonia, tuberculosis, fungal pneumonia, and noninfectious alternatives; stabilize respiratory and hemodynamic failure; interpret cultures; de-escalate antimicrobials; and address HIV care, adherence, stigma, vaccination, and follow-up. James Elliott, a stable adult with jaundice and acute hepatitis B, shifts the platform from minute-to-minute resuscitation to longitudinal viral-disease management. Learners interpret hepatocellular injury and sequential HBV markers, assess severity, avoid unnecessary treatment, counsel on transmission, protect contacts through testing, vaccination, and post-exposure measures, preserve confidentiality, and confirm resolution or chronic infection over time.
Evaluation Plan and Significance: The prototype includes learner-facing case progression, feedback pathways, dashboards, and administrative functions. Early formative review described the experience as realistic, engaging, and visually appealing while identifying interface refinements. Planned studies will measure usability, faculty-AI scoring agreement, time to critical actions, diagnostic prioritization, test-ordering efficiency, antimicrobial de-escalation, prevention decisions, communication quality, and recurrent class-level errors. Versioned clinical content, faculty oversight, and future retrieval-augmented generation from curated biomedical sources will support safety, transparency, and updating. The immediate application is education rather than autonomous diagnosis; its contribution to infectious-disease management is workforce preparation and measurable decision science.
Conclusion: DiseaseQuest applies multi-agent AI and big-data-ready learning analytics to the human decisions that precede safe infectious-disease diagnosis and management. By making reasoning visible and measurable across acute bacterial, immunocompromised-host, and longitudinal viral cases, the platform may enable scalable deliberate practice, targeted remediation, and research on how clinicians learn to manage infections before errors reach real patients.
