Researchers led by Koichiro Yuji and Osamu Wada-Hiraike from the University of Tokyo, Kitasato Institute Hospital, and Nippon Medical School Hospital have proposed HERA (Health Engagement and Reflective Agent architecture), a conceptual, hypothesis-generating digital health framework designed to support engagement, behavior change, and equitable self-management in menstrual and menopausal health across the female life course. Outlined in PLOS Digital Health, the multi-agent proposal represents a design reference and conceptual framework rather than an implemented, functioning, or clinically validated system.
The authors contextualize the proposal by highlighting that menstrual-cycle-related conditions and menopausal symptoms are highly prevalent across the life course, imposing an under-recognized burden on quality of life and occupational productivity. Citing evidence from a nationwide survey of more than 32,000 women, the researchers note an estimated annual productivity loss of roughly nine days per woman due to presenteeism related to menstrual symptoms. While smartphone period-tracking applications are widely used and wearable devices continue to emerge alongside public health policies promoting personal health record infrastructure and inclusion, existing self-management tools remain fragmented, struggle to sustain long-term engagement, exhibit variable content quality, and risk widening disparities in digital access and literacy.
Agentic artificial intelligence, defined in the paper as autonomous software capable of planning toward specific goals across multiple steps under human oversight and governance, has previously been explored for clinical triage and laboratory medicine. However, the authors emphasize that existing healthcare agent models are primarily clinician-facing and focused on diagnostic evaluation. In contrast, the HERA architecture targets a patient-facing domain characterized by the need for sustained self-monitoring, continuous behavior change, equitable engagement, and ongoing follow-through across varied everyday settings.
Under continuous human governance, the proposed HERA framework combines two adaptive agents: an interpretation agent and an orchestration agent. The interpretation agent is tasked with selecting among established behavior change techniques, such as goal-setting strategies or behavioral prompts, or choosing to defer interventions based on recent user adherence patterns and signal confidence metrics. The orchestration agent coordinates overall component outputs, manages user consent, and handles escalation pathways when designated clinical thresholds are exceeded. If internal system components disagree, the orchestration agent logs the conflict and defaults to the more conservative output.
Operating in coordination with the two adaptive agents are three rule-based modules dedicated to self-monitoring and data integration, guidance and content governance, and equity and engagement monitoring. The data integration module processes information that users voluntarily choose to share, including wearable device signals, symptom and menstrual cycle logs, patient-reported outcomes, and consented health records. The content governance module maintains curated, version-controlled guidance under human oversight, ensuring all module communications pass through the orchestration layer. In situations marked by high uncertainty, automated guidance is suppressed and escalated for human review. All system outputs remain explicitly advisory and plainly disclose uncertainty, ensuring that individuals and their clinicians retain final decision-making authority.
To delineate where digital self-management ends and formal clinical care begins, the authors provide a worked clinical scenario. In this example, a user who logs increasingly heavy menstrual bleeding combined with deteriorating daily functioning transitions from routine self-care to targeted adjustments with tailored support. When predefined red-flag clinical thresholds are crossed, the orchestration agent withholds further self-management advice and directs the user to seek professional clinical assessment. The framework also notes that while workplace occupational health services can assist with job accommodations, they are not an appropriate channel for evaluating abnormal bleeding. Because menstrual and menopausal health involve distinct biological trajectories and risks, operational thresholds must be established separately for each use case.
A primary architectural priority of HERA is treating sustained user engagement and health equity as first-class design objectives. Recognizing that many users abandon digital health applications prematurely, the framework incorporates an equity module that audits reach and engagement across demographic subgroups, alerting human operators when disparities arise. Human administrators can then introduce assisted pathways, low-technology alternatives, language adaptations, or targeted community outreach rather than relying on automated corrections. The authors also highlight the need for capability-sensitive design and participatory co-design with intended users to avoid assumptions that exclude marginalized populations, noting that in lower-resourced environments, the architecture could prioritize low-technology channels such as messaging services or community health worker mediation.
The paper explicitly distinguishes technical features that are feasible today from aspirational capabilities that require substantial future research, validation, and governance. Currently feasible elements include the integration of wearable and mobile application data with patient-reported outcomes, interpretable feedback delivery, human-governed version-controlled guidance, and subgroup engagement audits, although scaling individualized behavioral support remains challenging. Conversely, continuous population-level health monitoring, fully dynamic real-time guidance retrieval, and automated conflict resolution without human oversight remain aspirational goals.
Because the proposed system would handle sensitive reproductive, physiological, clinical, and employment-related information, the authors state that user consent mechanisms alone are insufficient. Robust data governance must define retention policies, restrict secondary data uses, control algorithmic inferences, and enforce role-based access permissions. Crucially, the architecture requires strict operational separation between occupational health coordination and employer visibility to safeguard user privacy. Furthermore, because its escalation functions mirror clinical decision support, real-world implementations would fall under medical device and health data regulations, demanding transparent logging of AI-influenced actions, defined update protocols, and explicit accountability for advisory suggestions and overrides.
To evaluate the framework rigorously, the authors propose a staged evaluation plan tracking equitable reach, behavioral and functional outcomes such as symptom scores and work productivity, and safety measures including escalation accuracy, over-referral rates, human override frequencies, and missed red flags. As an illustrative threshold, an equity benchmark could require twelve-week retention in the least-reached subgroup to achieve at least 0.8 times that of the best-reached subgroup alongside a 50 percent overall retention target. Evaluation would advance through retrospective baselining, usability testing, pilot studies, and controlled trials comparing HERA against standard care and non-agentic digital controls. The authors stress that HERA currently lacks real-world usability, clinical efficacy, and cost-effectiveness data, offering the architecture openly as a conceptual baseline for future empirical evaluation.
In evaluating the surrounding technological landscape, the authors contrast HERA with current digital solutions and prior artificial intelligence frameworks. Most existing tools for menstrual and menopausal health remain confined to basic cycle logging and symptom tracking, offering fragmented observations rather than continuous support. Meanwhile, agentic artificial intelligence across healthcare continues to be predominantly exploratory, with remarkably few studies directly involving patient participants. Prior structural models also reveal notable gaps: general health-agent frameworks remain domain-agnostic without reproductive specialization, while equity-focused chatbot roadmaps outline conceptual design stages rather than concrete operational architectures. Most critically, none of these earlier approaches maintain an enduring longitudinal state connecting adolescence through the menopause transition, and none integrates an equity signal directly into automated routing choices or pairs self-monitoring with governed guidance and multidisciplinary oversight.
Beyond individual symptom management and behavioral support, the researchers suggest that an integrated life-course framework could yield valuable physiological insights. Systematically collecting unified longitudinal data from menstrual logs, wearable sensor readings, and standardized patient-reported outcomes could help clarify how cyclical menstrual dynamics interact with and influence other physiological systems across the body. However, the researchers emphasize that using agentic artificial intelligence to accomplish these goals remains entirely unproven. The field currently lacks empirical implementation studies, real-world usability benchmarks, clinical validation, health outcome data, and formal cost-effectiveness analyses. Consequently, any clinical or operational benefit depends heavily on robust human governance, strict equity protections, and reliable supervisory oversight rather than unvalidated technological autonomy.
To ensure practical applicability and mitigate potential biases, the proposal outlines deliberate methodological safeguards for system development. The authors stress that designs implicitly presuming a cisgender, heterosexual user risk alienating or excluding marginalized populations, necessitating capability-sensitive design principles and continuous participatory co-design. While one of the study authors contributed directly to the conceptual architecture as an intended user herself, the team notes that broader, inclusive co-design with varied communities remains essential. Furthermore, although the conceptual model is informed by Japanese health policies emphasizing personal health records and life-course perspectives, the architecture is designed to be adaptable rather than transplanted intact into disparate environments. In lower-resourced global contexts, the equity monitoring structure would prioritize accessible, low-technology alternatives, such as text messaging entry points or integration with community health workers, supported by locally governed clinical guidance and contextual validation.
To isolate the specific therapeutic value of automated agent coordination, the authors propose a structured multi-phase experimental sequence before any clinical deployment. The evaluation pathway begins with retrospective baselining to establish standard reference metrics, moving subsequently through formal usability assessments, exploratory pilot studies, and rigorous controlled comparisons. Crucially, these planned comparative trials would test HERA against routine standard care to assess overall system efficacy, alongside a dismantled control group featuring identical educational resources, user features, and human professional support but lacking the autonomous multi-agent coordination layer. This comparative design aims to confirm whether agentic orchestration itself delivers measurable clinical improvements beyond standard digital interfaces. Supported by a Health and Labour Sciences Research Grant from the Ministry of Health, Labour and Welfare of Japan, the authors provide the conceptual architecture openly to invite independent technical adaptation, external scrutiny, and empirical validation across diverse digital health settings.
Subject of Research: A multi-agent digital health framework for menstrual and menopausal self-management
Article Title: Supporting engagement, behavior change, and equitable self-management in menstrual and menopausal health: A multi-agent digital-health framework
Article References: Yuji, K., Yuji, W., Nariai, M., Enomoto, Y., & Wada-Hiraike, O. (2026). Supporting engagement, behavior change, and equitable self-management in menstrual and menopausal health: A multi-agent digital-health framework. PLOS Digital Health, 5(11), e0001821. https://doi.org/10.1371/journal.pdig.0001821
Image Credits: AI Generated
DOI: 10.1371/journal.pdig.0001821
Keywords: digital health, artificial intelligence, menstrual health, menopause, multi-agent systems, health equity, self-management, behavior change, Supporting, engagement, behavior, change
News Source: Ophelia Keating. (November 9, 2026). Multi-Agent AI Framework Proposed for Menstrual and Menopausal Self-Management. Scienmag.
