HONE: A learning health system platform for advancing early intervention in first episode psychosis.

HONE: A Learning Health System Platform for Advancing Early Intervention in First Episode Psychosis - PMC Skip to main content An official website of the United States government Here's how you know Here's how you know Search Log in Primary site navigation Search Log in Search PMC Full-Text Archive Search in PMC Journal List User Guide As a library, NLM provides access to scientific literature. Inclusion in an NLM database does not imply endorsement of, or agreement with, the contents by NLM or the National Institutes of Health. Learn more: PMC Disclaimer | PMC Copyright Notice Schizophr Res . Author manuscript; available in PMC: 2026 Jun 1. Published in final edited form as: Schizophr Res. 2025 Apr 9;280:1–9. doi: 10.1016/j.schres.2025.04.009 HONE: A Learning Health System Platform for Advancing Early Intervention in First Episode Psychosis Walter S Mathis a, * , Toni Gibbs-Dean a , John D Cahill a , Vinod H Srihari a Author information Article notes Copyright and License information a. Program for Specialized Treatment Early in Psychosis (STEP), Department of Psychiatry, Yale University School of Medicine & Connecticut Mental Health Center (CMHC), 34 Park St., New Haven, CT, 06519, USA. * Corresponding Author: walter.mathis@yale.edu Issue date 2025 Jun. PMC Copyright notice PMCID: PMC12052476  NIHMSID: NIHMS2073174  PMID: 40209525 The publisher's version of this article is available at Schizophr Res Abstract The emergence of psychosis in early adulthood necessitates rapid, specialized care to improve outcomes and reduce the duration of untreated psychosis. Early intervention services (EIS) for First Episode Psychosis (FEP) are exemplars of healthcare reform but face challenges in sustaining durable improvements across patients’ lifetime illnesses. Learning Health Systems (LHS) present a transformative framework, integrating care delivery with continuous quality improvement and research. In this paper, we detail the development and evolution of the Health Outcomes Network and Education (HONE), a novel informatics platform designed to support an LHS for FEP. HONE facilitates data harmonization, analytics, and visualization to drive quality improvement, leveraging user-centered design to engage clinicians and align with clinical workflows. The platform emphasizes evidence-based outcomes, integrating diverse data sources, including patient-reported outcomes, clinical assessments, and external datasets, within a harmonized repository. By employing a minimalist, question-based approach to data visualization, HONE delivers actionable insights that inform clinical practice while generating research questions to improve care. We discuss key lessons from HONE’s development, including the importance of user engagement, iterative refinement, and balancing standardization with site-specific flexibility. HONE has been successfully deployed across multi-site networks, demonstrating its scalability and adaptability to diverse healthcare settings. This paper highlights HONE’s contributions to bridging the gap between knowledge and care, fostering bi-directional knowledge translation, and advancing mental health outcomes. Future efforts will focus on expanding HONE’s reach, integrating predictive analytics, and evaluating its long-term impact on FEP care within the LHS framework. Keywords: First Episode Psychosis, Learning Health System, Health Informatics, Data Visualization, Early Intervention Services, Population Health 1. Problem statement: the need for Learning Health Systems The clinical diagnosis of a schizophrenia spectrum disorder likely indexes multiple diseases. Several distinct pathways that include interactions of genetic risk, environmental factors and chance derailments in neurodevelopmental processes likely contribute to the illness phenotypes that clinicians are trained to recognize ( Tandon et al., 2024 ). Despite this underlying heterogeneity, there are useful commonalities in diagnostic practice, prognosis and treatment that make these disorders a useful target around which to organize healthcare services. Early intervention for recent onset schizophrenia (or first-episode psychosis, FEP) is an increasingly successful and worldwide exemplar of such service reform ( Hastrup et al., 2013 ; Srihari et al., 2015 ; Correll et al., 2018 ; Srihari et al., 2022). The emergence of psychosis in early adulthood requires rapid professional assessment and initiation of care. Modern Early Intervention Services (EIS) for schizophrenia should aim to both shorten the duration of untreated psychosis and deliver comprehensive medical, psychosocial, and rehabilitative care for FEP. Research across international settings has demonstrated improved outcomes and delivered implementation guidance ( Edwards & McGorry, 2002 ; IEPAWG, 2005 ; NIMH, 2014 ). However, such efforts require initial capital investments, have been variably resourced, and often can only offer specialized care for a few years. The positive effects of extant EIS models may not be durable across the trajectories of these lifetime illnesses ( Hansen et al., 2023 ). As should be expected, advances in our evidence base have raised new and important targets for further innovation in service implementation and new questions for research. The presence of early intervention services also presents an opportunity for investigators who wish to better understand the causes and pathophysiology of the distress, disability, and premature mortality caused by these disorders. FEP patients are relatively less impacted by the effects of chronic illness and medication treatment, and as such offer access to biological processes that are less confounded by such factors and may also be more sensitive to intervention ( Keshavan & Narayan, 2010 ). This may be even more true of those in the prodromal phase of illness and the project to more accurately identify and treat individuals before the onset of psychosis is an active area of investigation ( McGlashan et al., 2003 ; Fusar-Poli et al., 2012 ). Early intervention services can thus contribute to two domains. First, EIS can disseminate existing knowledge about effective models of care. This involves the challenging work of adapting principles and practices derived from research demonstrations to the resources available in varied settings, while preserving and even improving the impact on patients and families. Second, EIS can play a pivotal role in helping patients and families to participate in research to improve the effectiveness of current best practices. In routine behavioral healthcare settings, there is poor integration of service delivery and knowledge production workflows. As a result, questions that arise during patient care often do not inform research agendas, and scientific discoveries about the causes or mechanisms of illness frequently fail to influence clinical practice. This disconnect hampers the development of effective treatment strategies and undermines efforts to improve patient outcomes. There is an urgent need to design services that resolve these barriers. The concept of a Learning Health System (LHS) can help reform existing EIS and engineer new services to participate in bi-directional knowledge translation ( Srihari & Keshavan, 2022 ). An LHS is one in which “science, informatics, incentives, and culture are aligned for continuous improvement and innovation, with best practices seamlessly embedded in the delivery process” ( Institute of Medicine, 2007 ). In such systems, patients and families are recipients of care, but also participants in activities that can improve delivery of existing knowledge (quality improvement) and research to address knowledge gaps. Data collection for all these purposes is expected within the care experience and should be embedded within a systematic approach to informatics. In this paper, we highlight the critical role of health informatics in enabling this approach and detail the development of a novel analytics and visualization platform designed to support an LHS for FEP. 2. Informatics: a core component of Learning Health Systems Health informatics involves the effective use of information and technology to improve individual health outcomes, enhance healthcare services, and advance public health and biomedical research ( Shortliffe & Cimino, 2006 ). Informatics can facilitate the collection and integration of diverse data types including administrative databases, community surveys, electronic health records, clinical assessments, and patient-reported outcomes. Advanced analytics can then generate actionable insights, which are fed back to the healthcare team to inform responses at the level of the individual patient, clinic, or guide regional policies – all within an ethos of continuous improvement. Informatics thus serves as the backbone of an LHS, enabling data-driven decision-making and efficient knowledge dissemination. Data integration and continuous learning are fundamental to a Learning Health System (LHS) ( Kohn et al., 2022 ). By leveraging data from all patients, LHS frameworks create cycles that transform data into actionable knowledge, apply it in practice, and use the results to inform future decisions, driving continuous improvement. An effective LHS requires a secure, interoperable informatics infrastructure that integrates seamlessly with clinical workflows. This socio-technical system supports automated data collection, analysis, and dissemination on a large scale, facilitating clinical decision support and personalized care through predictive analytics that offer tailored interventions based on diverse data sources. Achieving success in an LHS also relies on cultivating a culture of continuous improvement and data-driven decision-making. This shift involves training clinicians in data science and informatics, equipping them to work effectively with big data, AI, and predictive analytics. Equally important is the design of user-centered data visualizations – intuitive, clinically relevant tools that make data accessible in real time and directly linked to clinical outcomes. These data products foster clinician engagement and support the cultural transformation necessary for an effective LHS by creating buy-in and promoting meaningful relationships with the data. This paper will use the example of a purpose-built platform to illustrate the role of informatics within an LHS for FEP. We will share the design, evolution and ongoing implementation of HONE (Health Outcomes Network and Education). All software underlying the versions of HONE have been made available without cost to collaborators and were developed to optimize use across varied healthcare settings. While technical documentation is available upon request, this paper aims to bring the reader ‘behind the hood’ to explain our design approach and current implementation, and to discuss the lessons learned through several versions. By sharing our experiences, we hope to contribute to the broader effort to enhance patient outcomes in schizophrenia care and encourage stakeholders to apply these concepts within their own contexts. 3. Health Outcomes Network and Education (HONE): Design Principles 3.1. Population Health Outcomes as the basis for a System of care The program for Specialized Treatment Early in Psychosis (STEP) has organized its early intervention service in Connecticut ( Srihari et al., 2015 ) within a population health framework (Yoviene Sykes et al., in this issue). The premises of population health envision (i) multiple and non-medical determinants of health, (ii) the clinical service as an integrator of multiple stakeholders across a defined geographic catchment to improve both access and care quality and (iii) measurement and responsivity to health (and not merely healthcare) outcomes, as well as disparities in these outcomes. In 2016, we proposed a list of core outcomes to orient the work of such a population health based early intervention service for FEP ( Srihari et al., 2016 ). This was based on prior work involving an iterative sequence of steps that continues to inform our approach ( Fig. 1 ). First, the values of diverse stakeholders (e.g., family members, clinicians, payers) were used to derive a core set of ‘Objectives’ (e.g. rapid access, improved employment) that we then translated into measurable constructs (e.g., Duration of untreated psychosis, BRS labor classification) and workflows within the EIS to collect the necessary data elements. These population outcomes are benchmarked against aspirational standards as reported in the best performance of services around the world. The number of objectives is constrained by the availability and feasibility of measures to capture relevant data elements in an active EIS, and to those outcomes for which effective interventions are available and for which meaningful benchmarks can be derived. The initial set of outcomes ( Srihari et al., 2016 ) envisioned expansion (e.g., development of new and validated measures, addition of new objectives) but within these constraints (see Yoviene Sykes et al., in this issue). The goal remains to create a limited (and therefore manageable) list of outcomes that reflect the values of stakeholders and can thus be used to drive quality improvement led by the service (when outcomes lag behind expected results from published performance of best practice implementations) and research (when new knowledge is needed to improve outcomes). These parameters (objectives, measures, outcomes) are also legible to lay stakeholders who can more easily critique, improve, and advocate for necessary resources. All healthcare and social services, in the final analysis, are funded by the same stakeholders, and the transparency of population health-based outcomes merely reflects this reality and can inform sustainable reform ( Emanuel & Fuchs, 2008 ). Fig. 1. Open in a new tab The figure illustrates the STEP Learning Health System Informatics Workflow, outlining the process from stakeholder-driven values to defining objectives, selecting measures, and developing standards. Data from various sources are integrated through a workgroup and iterative testing of the HONE platform, ensuring alignment with quality improvement, research, and workforce development goals. EPINET: Early Psychosis Intervention Network ( Heinssen & Azrin, 2022 ). 3.2. User-centered design To promote buy-in from clinicians – key users of the platform for quality improvement – the design process for the software interface was centered on their needs. The project team continuously engaged with the multidisciplinary providers that are typical of EIS (social work, psychology, psychiatry, rehabilitation staff) to ensure that the system was not only relevant but also perceived as a tool that added value to clinical practice, rather than an administrative burden. This engagement ensured that the platforms addressed the real-world needs of clinicians, enhancing usability and relevance to clinical workflows. Other stakeholders within the healthcare system were included with attention to their roles. For example, the first version of HONE featured multi-level dashboards tailored to clinicians, clinical administrators, and network administrators and providing data reporting at various levels (patient, caseload, clinic, and network). A key design element is the use of regular feedback from clinicians to improve the utility of the visualizations. HONE presents near real-time population level data that can reveal trends over time within and across clinics, and disparities within patient subgroups. Clinicians are routinely exposed to this perspective and encouraged to reflect on these trends and draw inferences about clinic performance that can drive quality improvement initiatives. The ensuing QI projects generate demands for new analyses or visualizations, which are initially responded to with ad hoc tables or figures but also considered for inclusion in the default set of visualizations. Such proposals are processed in an informatics workgroup consisting of individuals with different roles in the program: clinicians, clinical administrators, researchers, statisticians and coding staff. Emerging visualizations are then ‘beta-tested’ within STEP ( Fig. 1 ) before broader release to other clinics. This process addresses the risk of burnout among clinicians, particularly when new software features are introduced, and proactively addresses this ‘in-house’ at STEP before offering the new visualizations to collaborating clinics. 3.3. Flexibility and modularity HONE was built to support networks of EIS that wish to measure and share their performance to drive QI and collaborate on research. In order to achieve this, HONE needed to be flexible and as such modular in its construction. This involved meeting two goals. The first goal requires a common set of measures that can permit rational comparison. The second may require specific measures related to specific studies that are not relevant or ready for inclusion in a common set of outcomes. Further, some clinics may be required (e.g. by funders, regulators) or wish to measure additional outcomes. Given the diversity of contexts that EIS must operate within, HONE is designed to be adaptable, allowing for customization to meet site-specific needs while maintaining consistency in core outcome measures. The development approach, consistent with the ethos of LHS, welcomes new questions and problems as opportunities to improve the visualizations and features available to all current and future clinics, while also permitting explorations of requests that may be idiosyncratic to one site, and may thus need to be separately resourced. Designing an informatics platform with a modular architecture enables the creation of a flexible and adaptable tool that accommodates evolving outcome measures while supporting customization for site-specific requirements and specialized research priorities. Moreover, HONE is designed to be agnostic to the electronic health record (EHR) systems used at different locations, ensuring compatibility across diverse technological ecosystems and reducing barriers to adoption. 4. Evolution of HONE: Lessons learned The HONE platform was designed to support individual and networks of clinics that seek to implement a n LHS for FEP. This effort was motivated by the awareness that existing clinical information systems were not fit for such a purpose, and almost a decade later this continues to be true. Despite transitions to electronic medical records across behavioral healthcare agencies, a persistent managerial preoccupation with coding and billing, pervasive use of boilerplate to satisfy regulators and payers, and lack of attention to key data elements make most clinical records embarrassingly devoid of useful information about the health of the populations being served. Most healthcare teams thus remain ‘proprioceptively blind’ to their impact on population health ( Wolfson, 1994 ). Creative clinical operators and managers, even when fully aligned with the mission of improving patient outcomes, lack a way to gather timely feedback from their own healthcare records to guide efforts to improve care processes. Instead, clinicians are subject to various top-down initiatives to improve performance. These are often crafted in response to regulatory pressure to address real or imagined gaps in care. 4.1. HONE 1.0 In 2015, a competitive research foundation grant resourced initial software development of HONE (initially called CSC-ONE, Coordinated Specialty Care Outcomes, Network, Education) as a platform to support an LHS for EIS. The project aimed to develop a unified approach to clinical data collection, evaluation, and knowledge generation that would be prototyped within STEP (New Haven, CT) ( Srihari et al., 2014 ). All program staff at STEP across multiple clinical and research disciplines helped beta-test features in the context of a busy EIS. STEP is based within a public sector agency with a variety of independent paper and digital systems for notes, prescriptions, medication orders, and labs. This informatics environment lacks a unified EHR, and over the subsequent decade added new and siloed approaches that progressively limited the ability to aggregate and audit health outcomes. This environment proved to be a formative influence. HONE was designed to be agnostic to any specific electronic medical record but also flexible in its ability to interface with a wide variety of clinical and research databases. From the beginning, the platform was designed to scale across multiple clinics. A clear and persistent ethos was developed to value and protect a focus on the use of population outcomes to improve care and separate this from required charting procedures to comply with billing and regulatory requirements. This meant however, that STEP clinicians would need to interact with one more electronic interface which would not be part of the official medical record. Therefore, emphasis was placed on user-centered design and the development of practice management tools to incentivize adoption of the system, including features such as SMS appointment reminders, a team calendar with integrated prompts for scheduled participant measurements, and assistance with creating clinical notes. The first functional prototype was completed in 2015, followed by beta testing and iterative development. The other elements of an LHS at STEP (Yoviene Sykes et al., in this issue) provided a hospitable environment for the early version of an informatics platform. Scrutiny of population level data became a part of the workflow in the clinic with one hour ‘population rounds’ held every month to review recently encoded versions and discuss potential new analyses. Quality improvement projects spawned from these discussions, including two that exemplified the use of data to drive improvements in wait-times ( Ferrara et al., 2022 ) and the effectiveness of care transfers after graduation from the early intervention service ( Gallagher et al., 2021 ). However, the project faced several challenges at the HONE 1.0 stage. First was the need for a larger initial capital investment to ensure the platform’s core features were adequately developed before entering the dev-ops phase. The resource constraints resulted in a prolonged period of bootstrapping, during which the team was forced to play catch-up to address critical functionality gaps. This phase was also marked by a heightened risk of scope creep, as the initial excitement and diverse stakeholder input created pressures to expand the platform’s feature set beyond its foundational goals. Additionally, while the practice management features added beyond the core informatics tools aimed to enhance user engagement by integrating clinical workflows, they inadvertently led to redundancy with existing EHR systems. Interfacing with third-party software also required additional resources for periodic debugging. While clinicians found some of the practice management tools valuable, the requirement to use scheduling tools that synchronized with other clinics in the agency necessitated double entry. Also, the communication of desired changes to remote software engineers who were unfamiliar with the clinical workflow required a significant investment of time. As a result, desired changes in visualization or requests for new analyses of data were delayed and decreased the momentum of ongoing QI projects. The development of HONE 1.0 highlighted the potential for a unified, scalable informatics platform to enhance population health outcomes in FEP care but also revealed significant challenges, including resource constraints, risks of scope creep, and redundancies with existing EHR systems. Despite these obstacles, the project established a foundation for iterative improvement and demonstrated the value of integrating user-centered design and actionable data to drive quality improvement initiatives. 4.2. HONE branches The concept and prototype for HONE responded to a common problem and as such several academic and community-based agencies expressed interest in using the platform. The co-inventors (JC, VHS) were invited to a meeting hosted by NIMH’s Division of Services and Intervention Research (DSIR) in September 2017 to present the platform and contribute to the development of a funding initiative ( NIMH, 2017 ). The ensuing EPINET initiative funded 8 networks of CSCs to build learning health systems ( Heinssen & Azrin, 2022 ; NIMH, n.d. ). Branches of the HONE code base were customized and configured as the informatics platform for two of the EPINET networks: AC-EPINET (Vohs et al, this issue) and ESPRITO (Robinson et al. this issue) - collectively referred to as ‘HONE-Academic’. A branch of HONE 1.0 was also created to assist a non-academic statewide initiative ( Breitborde et al., 2022 ). Expanding the informatics platform to research networks and a statewide project showcased its appeal but added software development challenges. For example, specific features were added to support multi-site research recruitment and tracking. This was analogous to the practice management features of 1.0 but oriented to the needs of research staff. These and other needs from the wider network resulted in significant adaptations that continue to evolve and will be the subject of future reports. These branches of the HONE 1.0 code base continue to support LHS initiatives for over 25 clinics (and growing) across 11 US states. Alongside the additional research element, STEP’s legacy battery of data elements was also expanded to integrate the Common Assessment Battery (CAB) (agreed upon by EPINET investigators). HONE-Academic additionally shed all the practice management features of HONE 1.0, as these would have been too expensive and inefficient to implement across diverse healthcare organizations. Regarding visualizations, pragmatic responses to data privacy regimes across different clinical sites (limiting data-sharing or Safe Harbor de-identified datasets) meant that dates of service were not available, limiting the usefulness of some of the analyses and figures. However, one network was able to successfully import HONE 1.0 visualizations. This network also elected to capture all data via the REDCap system ( Harris et al., 2009 ) rather than direct entry into HONE with consideration of established workflows at the sites and alleviating the need for HONE to receive identified data. REDCap (Research Electronic Data Capture) is a secure, web-based platform designed to support data capture for research studies, enabling customizable forms, real-time validation, and seamless data export for analysis. An Application Program Interface (API) was built to channel relevant data elements from REDCap into HONE for data visualization. To improve on this process new visualizations (see funnel plots in Vohs et al, in this issue) were developed that were more robust to direct comparison of population health metrics across sites. These HONE branches, implemented across other networks, have validated the original design principles. Future publications will explicate these adaptations, that we expect to be of great value to the LHS community. Several refinements and lessons from these collaborations were incorporated into the ongoing development towards HONE at STEP. These were formative in the evolution towards the next version (detailed below). 4.3. HONE 2.0 In February 2024, STEP launched a statewide LHS (Yoviene Sykes et al., in this issue). The STEP-Learning Collaborative (STEP-LC) includes 13 behavioral health agencies within a statewide (Connecticut) network. STEP serves as the hub and as such, screens and consents all eligible FEP patients. Patients are then routed to a behavioral health agency within their local area. HONE 1.0 was not initially designed to support significant data aggregation, integration, and harmonization from multiple sources. The evolution to HONE 2.0 marks a focused shift to better support Connecticut’s statewide LHS. The platform has been streamlined to emphasize core informatics functions – data harmonization, analytics, and visualization – while offloading extra tasks back to local EHRs or alternative data collection tools (see Fig. 1 ). These changes reflect a commitment to enhancing the LHS through improved data collection, advanced data management, innovative data visualization, and the integration of modern tools. Collectively, these advancements provide a more efficient, purpose-built platform to effectively support continuous learning and improvement in patient care for First Episode Psychosis. 5. Key Changes for HONE 2.0 5.1. Evolution of Data Collection In early iterations of HONE, data collection was a collaborative effort involving both trained research raters and clinicians who were not specifically trained in data collection methodologies. Clinicians would enter data into their own medical records (partial EHR) and HONE, while researchers collected data separately into REDCap. This dual-entry system required merging datasets from HONE and REDCap, leading to increased complexity in backend data management and elevating the risk of errors due to duplicate data entry. To streamline this process and enhance data integrity, we transitioned to a more centralized approach. Baseline data are now exclusively collected by trained raters during the intake process, ensuring consistency and reliability in the initial assessments. Subsequent data are gathered directly from participants through online REDCap surveys, administered every three months. Trained raters also act as curators and supervisors, following up with participants who do not respond to surveys to maintain high response rates. This shift not only simplifies the backend by consolidating all data within REDCap but also eliminates the administrative burden on clinicians, who no longer need to input data into multiple systems. We have deliberately shifted away from clinician-collected measures, both to address logistical challenges and to avoid the inherent conflict in clinicians serving as both caregivers and evaluators. Relying on clinicians for outcome data can introduce measurement biases ( Bilsker & Goldner, 2002 ). Examples include patients who may wish to please their clinician and thus minimize reported distress, or clinicians whose beliefs about the treatments offered can influence their ratings. Furthermore, clinicians may feel pressured by perverse incentives, especially the possibility that outcome data can influence program funding or job evaluations. To ensure greater objectivity, independent raters curate patient responses and collect information from family members and clinicians when necessary. However, there are instances when the data needed are too specialized or complex to be adequately captured by existing tools. A pertinent example is the intricate information related to “pathways to care” – the myriad interactions and experiences patients have between the onset of illness and the initiation of treatment. Understanding these pathways is essential for uncovering both experiential data, such as aversive encounters, and identifying sources of delay within the healthcare system ( Mathis et al., 2022 ). In time-sensitive conditions like FEP, where delays to treatment are critical metrics impacting patient outcomes ( Marshall & Rathbone, 2011 ), capturing this nuanced data becomes even more crucial. To efficiently collect and analyze such specialized information, we developed an internal tool and data format tailored to these specific needs ( Mathis et al., 2024 ). The evolution of our data collection methodology reflects a deliberate effort to simplify data management, reduce administrative burdens, and enhance the objectivity and reliability of our outcome measures. By centralizing data collection within REDCap and leveraging trained raters and participant-reported outcomes, we have created a more efficient and unbiased system that better serves the goals of our project. 5.2. Advancements in Data Management In the earlier stages of our project, data management was a fragmented and labor-intensive process. As mentioned, both trained raters (into REDCap) and clinicians (into HONE) were entering different data separately. Periodically, HONE would import data from REDCap into its own database, including measures not originally collected within HONE such as parental income and historical data from clinic branches that existed prior to HONE’s implementation. Further, as the study evolved, different clinic branches had variations in data collection methods – some measures weren’t collected uniformly, frequencies differed, and definitions of variables sometimes varied. Resolving these inconsistencies to create a coherent, interoperable dataset required a meticulous, fine-toothed understanding of each data branch. The complexity intensified when the program began considering incorporating data from external sources like census information or All-Payers Claims Data (APCD). Harmonization of these data if done within HONE would be code-intensive, computationally demanding, and prone to bugs, making the process overly complex and less reliable. To address these issues, we transitioned to a new data management strategy centered around a harmonized repository. This approach involves abstracting and harmonizing all data management and governance tasks into a centralized repository before any analysis or visualization takes place. Data cleaning, resolving interoperability challenges, and computing derived measures are performed upfront, with new columns created as necessary. Missing data – an important element for any real-world clinical study – can be transparently and consistently tracked and declared. External data sources, such as census data and APCD, are also imported and integrated into this unified system. By standardizing these processes, we can document, verify, and trust each step, ensuring the data’s integrity and consistency. This revamped approach provides a clean, reliable foundation for building analytics and visualizations. With a standardized and harmonized dataset, we can more effectively develop insights without the overhead of on-the-fly data processing. The harmonized repository serves as a single source of truth, simplifying data governance and reducing the potential for errors. This evolution in our data management practices not only streamlines our workflow but also enhances the reliability and validity of our analytical outcomes. 5.3. Innovations in Data Visualization In the early stages of HONE, off-the-shelf visualization tools were used, offering initial functionality but presenting challenges such as unnecessary clutter, performance issues, and cumbersome modifications to emphasize critical data points. To address these limitations, HONE 2.0 adopted a custom visualization approach, leveraging a harmonized repository for direct analysis and focusing on clarity, interpretability, and user-centered design. Each visualization was tied to specific core outcomes central to improving patient care in FEP, using a question-centered approach to present data in a clinically relevant and actionable manner – for example, “What percentage of our patients are employed?” This method aligns with the principles of a Learning Health System (LHS) by emphasizing evidence-based objectives, reducing inefficiencies, avoiding spurious correlations, and supporting iterative learning cycles that drive meaningful improvements in care. Drawing on Edward Tufte’s principles of effective data visualization ( Tufte, 1983 ), HONE 2.0 prioritizes simplicity and clarity by maximizing the “data-ink” ratio, eliminating unnecessary elements, and enhancing the interpretability of complex information. HONE 2.0 also introduced layered visualizations, which provide clear overviews upon page load while enabling users to explore deeper insights through interactive features. These visualizations allow for dynamic investigation of time points, data gaps, and subgroup comparisons, fostering a more comprehensive understanding of clinical trends. This approach encourages users to engage deeply with data, facilitating evidence-based decision-making and aligning with the iterative improvement ethos of an LHS. To accelerate development, HONE 2.0 leveraged AI-assisted code generation, enabling rapid prototyping and refinement of visualizations. This streamlined process allowed developers to incorporate clinician feedback quickly, producing iterations within hours rather than weeks, while maintaining strict data security by ensuring all processing occurred locally. Clinicians and researchers were actively involved in the development process, aligning visualizations with clinical needs and ensuring relevance without reliance on external coders. Evaluation played a critical role in refining the visualizations, ensuring they remained practical and impactful for end users. Structured evaluations, including usability testing, surveys, and the “think-aloud” method ( Van Someren et al., 1994 ), allowed for detailed assessment of how clinicians interacted with the dashboard and identified areas for improvement. Additionally, an integrated feedback button facilitated spontaneous input from users, and utilization data provided quantitative insights into dashboard performance and adoption. Together, these evaluation methods supported iterative refinement, aligning the visualizations with clinicians’ needs and enhancing their utility within the LHS framework ( Börner et al., 2019 ; Van Someren et al., 1994 ). Through these advancements, HONE 2.0 succeeded in creating visualizations that effectively communicate complex data, foster clinician engagement, and drive actionable insights, ultimately contributing to improved care in FEP and advancing the goals of a Learning Health System. 6. Discussion Our development and implementation of the HONE platform has addressed critical challenges in establishing a Learning Health System (LHS) for First Episode Psychosis (FEP). By focusing on specific, a priori outcome measures and employing a question-based framing of data, we have created tools that streamline data collection and enhance data interpretability for clinicians. The advancements in data visualization, guided by a minimalist design philosophy inspired by Tufte’s principles ( Tufte, 1983 ; Powsner & Tufte, 1997 ), have facilitated deeper user engagement and empowered clinicians to make evidence-based decisions. These innovations collectively contribute to improved patient care and exemplify the principles of an effective LHS. Our approach aligns with the broader movement towards LHS implementations utilizing informatics and visualization tools. The Institute of Medicine emphasized transforming healthcare into a learning system where knowledge generation is a natural byproduct of the care process ( IOM, 2007 ). Bates et al. (2014) proposed a framework highlighting the role of health information technology in enabling continuous learning and improvement within healthcare systems. Similarly, the University of Michigan’s LHS leverages EHR data to inform patient care across various conditions, integrating science and informatics for continuous improvement ( Friedman et al., 2017 ). The Veterans Health Administration’s Corporate Data Warehouse serves as a nationwide LHS by integrating vast amounts of patient data to enhance veteran care ( Brown et al., 2017 ). While these systems are comprehensive, HONE differentiates itself through its focus on the nuanced field of FEP care, employing a minimalist visualization approach and emphasizing user-centric design to facilitate clinician engagement. Effective data visualization is essential in healthcare for interpreting complex data and supporting decision-making ( Krum, 2013 ). Khorana et al. (2019) highlighted the growing importance of data visualization in health informatics through a bibliometric analysis Börner et al. (2019) emphasized the need for data visualization literacy to effectively interpret and utilize visual data representations. Our design philosophy aligns with these insights, prioritizing clarity and simplicity to enhance data comprehension among clinicians. The implications of our work for clinical practice and LHS are significant. By streamlining data collection and providing intuitive visualization tools, we reduce the administrative burden on clinicians, allowing them to focus more on patient care. The emphasis on specific outcome measures aligns with evidence-based practice, facilitating targeted and effective interventions. Our approach exemplifies the LHS cycle of continuous learning and improvement, as outlined by Börner et al. (2019) demonstrating how data-driven insights can directly translate into enhanced patient outcomes and care quality. Despite these advancements, several ongoing challenges will require attention. Incomplete data and patient disengagement will limit the utility of any informatics platform. Integration challenges with existing systems in different settings must also be considered. While HONE was deliberately designed to be EHR-agnostic to support diverse and often fragmented healthcare environments, we recognize the critical importance of interoperability with existing electronic health record systems. Seamless integration with EHRs can reduce redundant data entry, lower administrative burdens, and improve clinician adoption by embedding learning health system functions directly into clinical workflows. In our next phase of development, we are exploring standardized protocols such as scheduled EHR data extractions integrated into backend HONE data workflows to enable data exchange between HONE and commonly used EHR platforms. These efforts aim to allow near-real-time synchronization of patient data, promoting efficiency and scalability and enhancing the sustainability of the platform by aligning with broader institutional infrastructure and minimizing disruption to routine care delivery. Looking ahead, we see significant potential in expanding HONE’s capabilities through the integration of patient-facing digital tools, such as mobile applications and secure web portals. These tools could empower participants to engage more directly with their own health data. Incorporating mobile-friendly interfaces would support asynchronous, user-driven data entry, potentially improving both the completeness and accuracy of patient-reported outcomes. Furthermore, such platforms could foster sustained engagement, giving patients and families greater visibility into progress and helping them feel more connected to the care process. Ethical considerations regarding data privacy and security are paramount, particularly when dealing with sensitive mental health information ( Caine & Hanania, 2013 ). Addressing these limitations requires ongoing refinement and adaptation to diverse clinical environments. Protecting sensitive mental health data is central to HONE’s design. All data collection is governed by IRB-approved protocols, and personal identifiers are stored on HIPAA-compliant servers with restricted access. De-identified data are used for analysis and visualization, and role-based permissions ensure appropriate use across teams. HONE avoids storing raw clinical notes, focusing instead on structured, pre-defined outcome data. As we expand to patient-facing tools, we are exploring granular consent options and enhanced transparency features to strengthen trust and align with ethical best practices in digital health. Several important lessons emerged from the development and evolution of HONE. Aligning platform features with user needs is paramount. Engaging clinicians in the design process and being responsive to their feedback ensured that the platforms remained relevant and useful in real-world settings. This user-centered approach is essential for successful implementation and adoption. Balancing standardization with site-specific flexibility is also critical. While maintaining consistent core outcome measures is important for comparability and benchmarking, allowing customization enables the platform to meet the unique needs of different clinical settings. This balance promotes wider adoption and enhances the platform’s applicability across diverse environments. Finally, the value of iterative development and responsiveness to feedback cannot be overstated. The willingness to reevaluate and adjust the platforms based on user experiences and emerging challenges was key to their evolution and improvement. This iterative process allowed the platforms to remain dynamic and responsive to the changing needs of clinicians and patients, embodying the principles of a Learning Health System. Looking forward, we plan to expand the HONE platform to other regions and explore its applicability to additional health conditions. Enhancements in AI and analytics, such as predictive modeling and personalized care recommendations, offer promising avenues for future development. Further research is needed to evaluate the long-term impact of the platform on patient outcomes and to explore its scalability and adaptability in various healthcare contexts. While HONE was developed in the context of First Episode Psychosis, its core design principles – user-centered visualization, outcome-driven metrics, and modular, EHR-agnostic architecture – make it adaptable to other mental health conditions. Many of the platform’s features, including harmonized data management and real-time analytics, are directly applicable to conditions like mood disorders, substance use, and anxiety, where outcome monitoring and coordinated care are similarly vital. Future iterations may explore condition-specific modules to extend HONE’s impact across the broader landscape of behavioral health. 7. Conclusion The development of HONE represents a significant advancement in integrating informatics within an LHS for FEP. By focusing on specific evidence-based outcome measures and employing a minimalist, question-based approach to data visualization, we have created tools that enhance clinician engagement and facilitate evidence-based decision-making. The platform embodies the principles of continuous learning and improvement central to the LHS framework, facilitating bi-directional knowledge translation by enabling data-driven insights to inform clinical practice and by allowing clinical experiences to generate new research questions. Through this approach, we contribute to bridging the gap between empirical knowledge and routine care, ultimately enhancing patient outcomes in schizophrenia. Future efforts will focus on expanding the platform’s reach, integrating advanced analytics, and conducting further research to evaluate its long-term effectiveness. By fostering a collaborative and data-driven care environment, we aim to continue advancing the quality of FEP treatment and contribute to the global effort to improve mental health outcomes. Fig. 2. Open in a new tab Example visualization exemplifying design approach. By getting the visualization out of the way, we allow the data to become the focal point, enabling exploration and interrogation. This empowers clinicians and other stakeholders to engage with the information critically, fostering a deeper understanding and facilitating informed decisions. Acknowledgements: This work was funded in part by the State of Connecticut, Department of Mental Health and Addiction Services, but this publication does not express the views of the Department of Mental Health and Addiction Services or the State of Connecticut. The views and opinions expressed are those of the authors. Role of funding sources: This work was supported by The Patrick and Catherine Weldon Donaghue Medical Research Foundation (PIs: Srihari & Cahill); National Institutes of Health (R01MH103831, PI: Srihari) and the Gustavus and Louise Pfeiffer Research Foundation (PI: Srihari). The funding sources had no involvement in study design, collection, analysis and interpretation of data; in the writing of the report; or in the decision to submit this article for publication. Footnotes Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain. Declaration of Interest Statement: Drs. Srihari and Cahill report being a cofounders of STEP-Forward, L.L.C., which provides consultation for services development and workforce development. The other authors report no known competing financial interests or personal relationships that could have appeared to influence the work reports in this article. References Bates DW, Sheikh A, & Wright A (2014). A framework for learning health systems: Lessons from health care. Health Affairs, 33(7), 1159–1166. 10.1377/hlthaff.2014.0360 [ DOI ] [ Google Scholar ] Bilsker D, & Goldner EM (2002). Routine outcome measurement by mental healthcare providers: Is it worth doing? The Lancet, 360(9346), 1689–1690. 10.1016/S0140-6736(02)11610-2 [ DOI ] [ PubMed ] [ Google Scholar ] Börner K, Bueckle A, & Ginda M (2019). Data visualization literacy: Definitions, conceptual frameworks, exercises, and assessments. Proceedings of the National Academy of Sciences of the United States of America, 116(6), 1857–1864. https://www.jstor.org/stable/26663754 [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Breitborde NJK, Parris CJ, Stearns WH, et al. (2022). Promoting the success and sustainability of coordinated specialty care teams in Ohio. Psychiatric Services. Advance online publication. 10.1176/appi.ps.20220126 [ DOI ] [ PubMed ] [ Google Scholar ] Brown SH, et al. (2017). The VA’s Corporate Data Warehouse: Uses and implications for improving veteran care. Studies in Health Technology and Informatics, 245, 1256. 10.3233/978-1-61499-830-3-125629295341 [ DOI ] [ Google Scholar ] Caine K, & Hanania R (2013). Patients want granular privacy control over health information in electronic medical records. Journal of the American Medical Informatics Association, 20(1), 7–15. 10.1136/amiajnl-2012-001023 [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Correll CU, Galling B, Pawar A, Krivko A, Bonetto C, Ruggeri M, … & Kane JM (2018). Comparison of early intervention services vs treatment as usual for early-phase psychosis: a systematic review, meta-analysis, and meta-regression. JAMA psychiatry, 75(6), 555–565. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Edwards J, & McGorry PD (Eds.). (2002). Implementing early intervention in psychosis: A guide to establishing early psychotic services. London: Martin Dunitz. [ Google Scholar ] Emanuel EJ, & Fuchs VR (2008). Who really pays for health care? The myth of “shared responsibility”. JAMA: The Journal of the American Medical Association, 299(9), 1057–1059. 10.1001/jama.299.9.1057 [ DOI ] [ PubMed ] [ Google Scholar ] Ferrara M, Gallagher K, Yoviene Sykes LA et al. (2022). Reducing delay from referral to admission at a U.S. first-episode psychosis service: A quality improvement initiative. Psychiatric Services. Advance online publication. 10.1176/appi.ps.202100374 [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Friedman CP, Rubin JC, & Sullivan KJ (2017). Toward an information infrastructure for global health improvement. Yearbook of Medical Informatics, 26(1), 16–23. 10.15265/IY-2017-014 [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Fusar-Poli P, Bonoldi I, Yung AR, Borgwardt S, Kempton MJ, Valmaggia L, Barale F, Caverzasi E, & McGuire P (2012). Predicting psychosis: Meta-analysis of transition outcomes in individuals at high clinical risk. Archives of General Psychiatry, 69(3), 220–229. 10.1001/archgenpsychiatry.2011.1472 [ DOI ] [ PubMed ] [ Google Scholar ] Gallagher K, Ferrara M, Pollard J, et al. (2021). Taking the next step: Improving care transitions from a first-episode psychosis service. Early Intervention in Psychiatry. Advance online publication. 10.1111/eip.13127 [ DOI ] [ PubMed ] [ Google Scholar ] Hansen HG, Starzer M, Nilsson SF, Hjorthøj C, Albert N, & Nordentoft M (2023). Clinical recovery and long-term association of specialized early intervention services vs treatment as usual among individuals with first-episode schizophrenia spectrum disorder: 20-year follow-up of the opus trial. JAMA Psychiatry, 80(4), 371. 10.1001/jamapsychiatry.2022.5164 [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Harris PA, Taylor R, Thielke R, Payne J, Gonzalez N, & Conde JG (2009). Research Electronic Data Capture (REDCap) A metadata-driven methodology and workflow process for providing translational research informatics support. Journal of Biomedical Informatics, 42(2), 377–381. 10.1016/j.jbi.2008.08.010 [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Hastrup LH, Kronborg C, Bertelsen M, Jeppesen P, Jorgensen P, Petersen L, … & Nordentoft M (2013). Cost-effectiveness of early intervention in first-episode psychosis: economic evaluation of a randomised controlled trial (the OPUS study). The British Journal of Psychiatry, 202(1), 35–41. [ DOI ] [ PubMed ] [ Google Scholar ] Heinssen RK, & Azrin ST (2022). A national learning health experiment in early psychosis research and care. Psychiatric Services, 73(9), 962–964. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Institute of Medicine (US) Roundtable on Evidence-Based Medicine. The Learning Healthcare System: Workshop Summary. (Olsen L, Aisner D, McGinnis JM, eds.). National Academies Press; (US: ); 2007. [ PubMed ] [ Google Scholar ] International Early Psychosis Association Writing Group. (2005). International clinical practice guidelines for early psychosis. The British Journal of Psychiatry. Supplement, 48, s120–124. 10.1192/bjp.187.48.s120 [ DOI ] [ PubMed ] [ Google Scholar ] Keshavan MS, & Narayan SA (2010). Neurodevelopmental trajectory of psychosis: Implications for intervention. Schizophrenia Bulletin, 36(3), 513–521. 10.1093/schbul/sbq013 [ DOI ] [ Google Scholar ] Khorana S, Kumar S, Das T, et al. (2019). Data visualization in health care: A bibliometric analysis. JMIR Medical Informatics, 7(3), e11849. 10.2196/11849 [ DOI ] [ Google Scholar ] Kohn MS, Topaloglu U, Kirkendall ES, Dharod A, Wells BJ, & Gurcan M (2022). Creating learning health systems and the emerging role of biomedical informatics. Learning Health Systems, 6(1), e10259. 10.1002/lrh2.10259 [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Krum D (2013). Cool infographics: Effective communication with data visualization and design. Wiley. [ Google Scholar ] Marshall M, & Rathbone J (2011). Early intervention for psychosis. Cochrane Database of Systematic Reviews, 6(CD004718). 10.1002/14651858.CD004718.pub3 [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Mathis WS, Ferrara M, Burke S, Hyun E, Li F, Zhou B, Cahill J, Kline ER, Keshavan MS, & Srihari VH (2022). Granular analysis of pathways to care and durations of untreated psychosis: A marginal delay model. PLOS ONE, 17(12), e0270234. 10.1371/journal.pone.0270234 [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Mathis WS, Ferrara M, Cahill J, Karmani S, Tayfur SN, & Srihari V (2024). TimelinePTC: Development of a unified interface for pathways to care collection, visualization, and collaboration in first episode psychosis. PloS One, 19(7), e0302116. 10.1371/journal.pone.0302116 [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] McGlashan TH, Zipursky RB, Perkins D, Addington J, Miller TJ, Woods SW, Hawkins KA, Hoffman R, Lindborg S, Tohen M, & Breier A (2003). The PRIME North America randomized double-blind clinical trial of olanzapine versus placebo in patients at risk of being prodromally symptomatic for psychosis. I. Study rationale and design. Schizophrenia Research, 61(1), 7–18. 10.1016/s0920-9964(02)00439-5 [ DOI ] [ PubMed ] [ Google Scholar ] National Institute of Mental Health. (2014). Coordinated specialty care for first episode psychosis: Manual II – Implementation. National Institutes of Health. https://www.nimh.nih.gov/sites/default/files/documents/health/topics/schizophrenia/raise/csc-for-fep-manual-ii-implementation-manual.pdf [ Google Scholar ] National Institute of Mental Health. (2017). Harmonizing clinical data collection in community-based treatment programs for first episode psychosis. Retrieved November 24, 2024, from https://www.nimh.nih.gov/news/events/2017/harmonizing-clinical-data-collection-in-community-based-treatment-programs-for-first-episode-psychosis National Institute of Mental Health. (n.d.). Early Psychosis Intervention Network (EPINET). U.S. Department of Health and Human Services. Retrieved November 11, 2024, from https://www.nimh.nih.gov/research/research-funded-by-nimh/research-initiatives/early-psychosis-intervention-network-epinet [ Google Scholar ] Powsner SM, & Tufte ER (1997). Summarizing clinical psychiatric data. Psychiatric Services (Washington, D.C.), 48(11), 1458–1461. 10.1176/ps.48.11.1458 [ DOI ] [ PubMed ] [ Google Scholar ] Shortliffe EH, & Cimino JJ (Eds.). (2006). Biomedical informatics: Computer applications in health care and biomedicine (3rd ed.). Springer Science+Business Media. 10.1007/0-387-36278-9 [ DOI ] [ Google Scholar ] Srihari VH, Tek C, Pollard J, Zimmet S, Keat J, Cahill JD, Kucukgoncu S, Walsh BC, Li F, Gueorguieva R, Levine N, Mesholam-Gately RI, Friedman-Yakoobian M, Seidman LJ, Keshavan MS, McGlashan TH, & Woods SW (2014). Reducing the duration of untreated psychosis and its impact in the U.S.: The STEP-ED study. BMC Psychiatry, 14(1), 335. 10.1186/s12888-014-0335-3 [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Srihari VH, Tek C, Kucukgoncu S, Phutane VH, Breitborde NJ, Pollard J, … & Woods SW (2015). First-episode services for psychotic disorders in the US public sector: a pragmatic randomized controlled trial. Psychiatric Services, 66(7), 705–712. [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Srihari VH, Jani A, & Gray M (2016). Early intervention for psychotic disorders: Building population health systems. JAMA Psychiatry, 73(2), 101–102. 10.1001/jamapsychiatry.2015.2821 [ DOI ] [ PubMed ] [ Google Scholar ] Srihari VH, & Keshavan MS (2022). Early intervention services for schizophrenia: Looking back and looking ahead. Schizophrenia Bulletin, 48(3), 544–550. 10.1093/schbul/sbac024 [ DOI ] [ PMC free article ] [ PubMed ] [ Google Scholar ] Tandon R, Nasrallah H, Akbarian S, Carpenter WT, DeLisi LE, Gaebel W, Green MF, Gur RE, Heckers S, Kane JM, Malaspina D, Meyer-Lindenberg A, Murray R, Owen M, Smoller JW, Yassin W, & Keshavan M (2024). The schizophrenia syndrome, circa 2024: What we know and how that informs its nature. Schizophrenia Research, 264, 1–28. 10.1016/j.schres.2023.11.015 [ DOI ] [ PubMed ] [ Google Scholar ] Tufte ER. The Visual Display of Quantitative Information. Cheshire, CT: Graphics Press; 1983. [ Google Scholar ] Van Someren M, Barnard YF, & Sandberg J (1994). The think aloud method: a practical approach to modelling cognitive. London: AcademicPress, 11(6). [ Google Scholar ] Wolfson MC (1994). Social proprioception: Measurement, data, and information from a population health perspective. In Evans RG, Barer ML, & Marmor TR (Eds.), Why are some people healthy and others not? The determinants of health of populations (pp. 287–316). Aldine de Gruyter. [ Google Scholar ] ACTIONS View on publisher site PDF (798.0 KB) Cite Collections Permalink RESOURCES Similar articles Cited by other articles Links to NCBI Databases Follow NCBI NCBI on X (formerly known as Twitter) NCBI on Facebook NCBI on LinkedIn NCBI on GitHub NCBI RSS feed Connect with NLM NLM on X (formerly known as Twitter) NLM on Facebook NLM on YouTube National Library of Medicine 8600 Rockville Pike Bethesda, MD 20894 Web Policies FOIA HHS Vulnerability Disclosure Help Accessibility Careers NLM NIH HHS USA.gov Back to Top

Related documents

• Process Evaluation of Digital Mental Health Interventions for Psychosis: Scoping Review With Framework Synthesis.#439257
• Promoting Healthy Lifestyle Behaviors#892917
• Religiosity and Engagement in Early Intervention Programme for First-Episode Psychosis.#422835
• COllaborative Shared Care to IMprove Psychosis Outcome#808099
• Client Engagement With an Early Psychosis Program and Post-Program Health Service Use: A Retrospective Cohort Study.#319181