This implementation report describes the development and deployment of a health informatics system (HIS) in the Central Adelaide Local Health Network (CALHN) in South Australia (SA), which includes the Royal Adelaide Hospital and The Queen Elizabeth Hospital and services approximately 40% of SA public patients. The report is in accordance with the iCHECK-DH (Guidelines and Checklist for the Reporting on Digital Health Implementations) guidelines and checklist (Checklist 1) [1]. The implementation took place in the context of Australia’s National Digital Health Strategy (2023‐2028), which prioritizes interoperability and secondary use of clinical data, and SA Health’s digital transformation agenda, which consolidated electronic medical record (EMR) infrastructure; however, complementary analytics and decision-support capabilities remained underdeveloped.

In 2011, SA public hospitals began migrating from paper-based records to the Sunrise (Altera Digital Health) EMR system. The implementation was a major public digital infrastructure project and was successful in the context of its resource and time constraints. Tools enabling back-end data extraction and enriched clinical information displays were not prioritized. Clinical staff subsequently reported frustration with the usability of clinical displays, and operational and research teams reported difficulty extracting timely data for reporting and analysis (World Health Organization Classification of Digital Health Interventions, items 2.1.1, 2.3.1, 2.5, and 2.10) [2].

Three sequential projects were funded and implemented within CALHN, each delivering a digital tool for a specific user group: critical care clinicians (real-time decision support at the bedside), clinical coders (coding accuracy and hospital funding allocation), and operations managers (bed availability and patient flow).

The 3 projects are at deliberately different stages of maturity, and this distinction is maintained throughout this report: the CALHN Critical Care Informatics System (CCCIS) has been implemented and evaluated; CODEXA, the clinical coding application, is in operational validation; and the Patient Flow application is in co-design. Throughout this report, “HIS” refers to the platform; CCCIS, CODEXA, and Patient Flow are the applications built upon the platform; and HeartAI Pty Ltd is the company involved in the platform’s development and commercialization.

Most large health networks in high-income countries have completed EMR migrations and are at varying stages of maturity in the secondary use of clinical data [3,4]. This implementation was primarily inspired by the Ambient Warning and Response Evaluation (AWARE) intensive care unit (ICU) information system at the Mayo Clinic, a clinician-facing system that aggregates EMR data into structured dashboards to support bedside decision-making, which has been evaluated in randomized crossover trials [5,6]. The HIS shares AWARE’s approach, which restructures raw EMR data into clinician-centered displays, but differs in architecture (cloud-native, open-standard technologies rather than proprietary infrastructure) and scope (extending beyond critical care to coding and operational flow). An extended comparison with other international and commercial systems is provided in Multimedia Appendix 1.

The primary aim of this project was to develop and deploy a contemporary HIS in SA public hospitals, built with technology capable of extension across SA Health. Table 1 summarizes the platform’s objectives, their measurable indicators, the corresponding governance records, and their status.

The HIS comprises 6 architectural layers (Figure 1). In the cloud layer, Microsoft Azure provides computing, storage, and networking managed through a version-controlled, infrastructure-as-code approach [7]. The operational layer uses Red Hat OpenShift (IBM) for container orchestration [8]. The network layer enforces access controls that keep clinical data within defined security boundaries. The service layer manages interapplication communication and near–real-time processing of EMR data, which is essential for clinical displays that must reflect current patient status. The application layer hosts CCCIS, CODEXA, and Patient Flow, and the analytical layer supports AI model training within restricted, graphics processing unit–enabled environments. Extended technical details, including the deployment tooling and software patterns used in each layer, are provided in Multimedia Appendix 1.

Data access is controlled through an identity management framework that assigns users to groups and roles on a minimum-necessary basis, with encryption at rest and in transit, auditing of modifications, and detection of unusual access patterns. All patient data remain within the SA Health network at all times. The HIS accesses EMR data under the governance and security frameworks of SA Health and CALHN, and no patient data are held by or transferred to HeartAI Pty Ltd or any external entity. Data sovereignty therefore remains with the public health system regardless of the commercial structures supporting the platform’s development.

In SA, EMRs are owned and governed by the SA government through the Department for Health and Wellbeing, SA Health, and Digital Health SA (DHSA), which manages information technology across the health system. Integration of the HIS with the Sunrise EMR system (achieved May 2022) required formal approvals across 4 governance domains, described in brief here and in full in Multimedia Appendix 1.

Technical governance comprised a security impact assessment (initiated in 2021 and ongoing), an information asset classification (completed in 2021), and an independent external cybersecurity review (completed in 2022). On May 6, 2026, the DHSA Technical Design Review Committee (TDRC) unconditionally endorsed the HIS platform deployment. TDRC endorsement is the highest formal step in SA Health technical governance, and this concluded a 5-year process from initial security review to design baseline approval.

Clinical governance is provided by the Clinical Governance Committee, which has met quarterly since May 2022. Membership includes the CALHN chief executive officer (or delegate), the executive lead for critical care and perioperative medicine, the director of digital strategy, a consumer representative, and senior critical care clinicians, with dedicated project subcommittees reporting to the Clinical Governance Committee.

Corporate governance was established through the incorporation of HeartAI Pty Ltd in February 2022, enabling formal, legally binding partnerships with SA Health, DHSA, CALHN, and AusHealth. Ethics approval for CCCIS was granted by the CALHN human research ethics committee (HREC) in 2023 (reference 18079), with amendments extending coverage to CODEXA (reference 17287) and the Patient Flow project (pending review). This approval governs the research activities conducted using the HIS; the platform’s technical development preceded the research program and was governed by the technical processes above.

The involvement of a private company in the development and commercialization of a platform serving a public health system creates actual and perceived conflicts of interest. All authors except JB hold equity in HeartAI Pty Ltd, and these conflicts have been declared through the appropriate institutional pathways at CALHN. Four mechanisms manage them: independent oversight by the Clinical Governance Committee, whose membership includes non–equity holders; joint ownership of the products developed under the CCCIS agreement by CALHN and AusHealth rather than by HeartAI alone; commercialization through AusHealth, a self-funded medical research charity that reinvests revenue into health innovation; and research priorities set by publicly funded grant agreements rather than by HeartAI’s commercial interests.

Early platform builds, implementation frameworks, and project protocols were completed on personal time before incorporation. Four grant-funded agreements followed between December 2022 and April 2026, totaling approximately Aus $1.84 million across CCCIS, CODEXA, and Patient Flow (Table 2). Private equity from 11 registered shareholders has supported core platform development and corporate operations, comprising a minority of total funding; its value is not disclosed for commercial confidentiality reasons.

Under commercial agreements, CCCIS and Patient Flow are jointly owned by CALHN and AusHealth, so the South Australian public remains a direct beneficiary, while AusHealth retains ownership of CODEXA with a share of revenue directed to HeartAI Pty Ltd. Sustainability planning couples this hybrid funding with governance informed by published frameworks for AI governance and ethics in health care [9-12], a transition path from research-only to operational funding through CALHN’s digital health budget, and progressive Fast Healthcare Interoperability Resources adoption. Milestone-based grant funding has at times constrained project continuity and resourcing, and the concentration of funding across a small number of grant bodies creates dependency risk. Extended funding and sustainability details are provided in Multimedia Appendix 1.

The implementation followed 5 overlapping phases across approximately 5 years (Figure 2): platform development and technical governance (2021‐2022), EMR integration and data validation (2022), clinical application development and pilot deployment (2022‐2024), formal usability evaluation and iterative refinement (2024‐2025), and expansion to additional use cases and user groups (2024-ongoing). Key decisions included a cloud-native rather than on-premises architecture, co-design with end users throughout development, establishment of governance structures before clinical deployment, and phased introduction of applications beginning with the highest-need user group. Decision-making followed a tiered structure, with technical decisions made within parameters set by DHSA, clinical decisions escalated to the Clinical Governance Committee, and strategic decisions requiring approval from CALHN leadership and AusHealth (details in Multimedia Appendix 1).

At the time of writing, the HIS supports 3 applications at explicitly different stages of maturity. CCCIS has been implemented and evaluated. It has been deployed across the 2 CALHN ICUs since 2022 and is in daily clinical use, and a formal usability evaluation has been completed. CODEXA is in operational validation. It is functionally complete and undergoing structured validation with the CALHN clinical coding team, with quantitative model performance reported below; it is not yet in operational use. The Patient Flow application is in co-design with CALHN’s Network Operations Centre and has no deployed components. Table 3 summarizes the status, adoption indicators, and available evaluation metrics for each application.

Ten governance milestones were achieved between 2021 and 2026, from information asset classification (2021) through EMR integration approval and Clinical Governance Committee establishment (May 2022) to TDRC endorsement of the technical design baseline (May 2026) (Figure 2 and Table 1). The sequence reflects the prerequisite relationship between technical governance (platform readiness), corporate governance (formal partnerships), and ethical governance (research activity).

CCCIS extracts live data from the Sunrise EMR back end and presents these data through a structured tabular interface modeled on the A0 paper chart that was used in ICUs for over 25 years before the EMR rollout, displaying clinical notes, vital signs, pathology, fluid balance, and ventilation parameters (Figure 3). It is deployed across CALHN’s 2 ICUs (54 beds, approximately 5000 admissions per year) and operates as a research-only, read-only system alongside EMRs, avoiding disruption to existing clinical workflows.

Three indicators of adoption and workflow integration are available. First, clinicians can access CCCIS on clinical workstations during ward rounds, handovers, and ad hoc patient reviews in both ICUs (evaluations ongoing). Second, the system automates real-time submission of Australian and New Zealand Intensive Care Society Centre for Outcome and Resource Evaluation (ANZICS CORE) registry data, covering 115 variables across approximately 5000 ICU admissions annually, replacing a previously manual process [13]. Third, ICU data registry nurses came to use the CCCIS displays as a primary source of clinical information during registry data collection. This reliance on a research-phase system required governance attention, described in the Discussion section, and is direct evidence of integration into daily clinical work.

In a formal usability evaluation, 8 senior CALHN intensivists interacted with CCCIS while it was connected to real patient data, using cognitive walkthrough and think-aloud methods with standardized instruments [14,15]. The mean System Usability Scale score was 76/100 (range 65‐95), in the “good” band on the Bangor adjective scale. Overall NASA Task Load Index workload was low (21/100), with a mental demand score of 38/100 and a frustration score of 18/100. Participants rated the interface at 7-9/10 for clinical relevance, timeliness, and accuracy on a bespoke questionnaire, and the qualitative analysis identified 3 recurrent design principles: structured tabular display, consistency with ICU workflow, and rapid synthesis of physiological, intervention, and trajectory data. Full methods and qualitative findings are reported in a companion publication.

CODEXA supports clinical coders who translate patient data into Australian Refined Diagnosis-Related Group classifications for activity-based funding [16]. It extracts EMR data and uses Longformer-based models [17] to associate ICD-10-AM (International Statistical Classification of Diseases and Related Health Problems, Tenth Revision, Australian Modification) and Australian Classification of Health Interventions (ACHI) codes with a patient encounter through an interactive interface that provides weighted summaries and automated cost determinations via national weighted activity unit calculators (Figure 4).

Quantitative validation metrics are available from an internal validation report (August 2026). Successive models were trained on 25,000 ICU episodes, 200,000 episodes, and 500,000 episodes encompassing all admitted services across CALHN, with performance assessed against a held-out test set of 50,000 episodes that were unseen by any model. The current production candidate achieved a pooled F₁-score of 71% across the full hospital case mix. Excluding high-volume, autocoded renal dialysis episodes, which inflate aggregate performance, the model achieved an F₁-score of 57.86% across approximately 32,000 complex acute episodes, with a precision of 64.72%, a recall of 55.53%, an exact primary diagnosis accuracy of 52.17%, and an exact primary procedure accuracy of 53.23%. Performance was strongest in structured, high-volume settings (F₁-scores of 99.4%‐99.7% in dialysis units and 78.7%‐86.9% in focused specialties such as neurology) and weakest in psychiatric and complex multimorbid care (F₁-score of 24%‐47%).

For context, purpose-built coding models, such as pretrained language models for international classification of diseases and multiple synonyms matching networks, achieve F₁-scores of 58% to 61% when evaluated across the full code set on the Medical Information Mart for Intensive Care (MIMIC) benchmarks MIMIC-III and MIMIC-IV [18-20]. Direct comparison should be treated cautiously: the MIMIC benchmarks use curated research datasets and US coding systems, whereas CODEXA was evaluated on uncurated operational EMR data using the Australian ICD-10-AM/ACHI system. With that qualification, these results suggest that performance is broadly in line with published models under harder, real-world conditions. These internal metrics have not yet been externally peer reviewed; a full evaluation will be reported in a dedicated publication.

The Patient Flow application will give CALHN’s Network Operations Centre near–real-time visibility of patient movement, capacity, and service pressures across the continuum from ambulance arrival to discharge, drawing live EMR data through the HIS. The project is in co-design with Network Operations Centre leadership; information-finding interviews and user interface prototype evaluations with operational staff were completed in July 2026, and no components have been deployed.

Definitive clinical impact metrics, including length of stay, mortality, coding accuracy uplift in production use, and patient flow gains, require multiyear deployment data and are not yet available. Available adoption indicators and evaluation metrics are summarized in Table 3, staged according to the SALIENT framework [21].

This report describes the development, governance, and deployment of a cloud-native, locally developed informatics platform integrated with a statewide EMR system in a complex public health environment. The implementation required concurrent development of technical infrastructure, multilayered governance, a public-private commercial model, and sustained co-design with clinical end users. Its products sit at 3 deliberately distinct stages of maturity: an implemented and evaluated critical care application in daily use, a coding application in operational validation with quantitative performance now benchmarked against published models, and an operational flow application in co-design.

The HIS shares conceptual foundations with AWARE [5,6], although AWARE was developed in a single, well-resourced academic center and has been evaluated in randomized crossover trials, whereas as of September 2026, evaluation of the SA HIS has addressed usability and model validation rather than clinical outcomes. The usability problems reported in a Finnish national survey of ICU clinical information systems [22] parallel those that this implementation set out to address, suggesting that these are systemic issues across health systems rather than unique to this context.

Implementation models for clinical informatics in public hospitals fall into 3 categories: in-house development, procurement from a commercial vendor, and dedicated public-private partnership. In-house development offers maximum institutional control but typically struggles with technical depth, senior engineering recruitment, and the iterative agility that contemporary AI development requires. Vendor procurement provides mature, supported products with clear contractual boundaries but constrains customization for local workflows, introduces ongoing licensing costs, and embeds vendor priorities that may not match public-sector goals.

The model adopted here produced specific realized benefits: rapid iteration through colocated clinical and technical teams, retention of intellectual property in state-led ownership structures, sustained consistency between commercial viability and clinical relevance, and risk-tolerant capital flow through targeted innovation grants under state technical governance. Realized risks include the conflict-of-interest management described in the Methods section, key-person dependency requiring deliberate succession planning, and the governance overhead of operating across 3 institutional bodies. Replication elsewhere depends on the maturity of public-sector innovation funding in the jurisdiction, regulatory clarity around clinician equity in publicly funded health services, and the availability of cofounders able to operate credibly across the clinical, research, operational, and technical domains. Where any of these preconditions are absent, vendor procurement or in-house development may be the more pragmatic route.

Enablers included the practical support of the DHSA technical team, strong executive sponsorship from CALHN leadership, and early collaboration with frontline clinicians, which grounded the mapping of clinical data entry processes to extractable database structures. Barriers included political resistance from stakeholders outside the immediate project team, managed through open communication with data governance authorities and regular demonstrations of the system’s performance and safety; the significant technical difficulty of secondary extraction from an EMR system whose data structures were not designed for it; and milestone-based grant funding that at times reduced capacity and forced difficult prioritization.

One unintended consequence warrants emphasis. Although the HIS remained in a research posture with safeguards against the exposure of clinical workflows to system failures, ICU data registry nurses developed a de facto operational dependency on the CCCIS displays because the system met information needs that the native EMR interface did not. The dependency was identified through governance monitoring and raised concerns about validation status, infrastructure redundancy, and liability. It was managed in 2025 through targeted adjustments to the clinical, ethical, and operational governance of the HIS within the ICU environment. Successful research tools may create operational dependencies before formal transition to operational status, and implementers should plan for this possibility.

This report has 5 limitations. First, it describes a single health network in a single-state health system, limiting direct generalizability to contexts with different governance structures, EMR systems, or organizational cultures. Second, evidence of clinical impact is limited to usability and acceptability data for CCCIS and internal validation metrics for CODEXA; there are no controlled comparisons, clinical outcome data, or health economic analyses, and the CODEXA metrics have not yet been externally peer reviewed. Third, all authors except one hold equity in HeartAI Pty Ltd, creating potential for bias in the reporting of implementation successes and challenges despite the governance mechanisms described. Fourth, the implementation has depended heavily on a small number of key individuals, particularly for technical development, which creates sustainability risk. Fifth, the extraction and integration approaches described are specific to the Sunrise EMR system and may not transfer to other platforms without significant adaptation.

Three transferable lessons stand out. The first is to establish governance structures early and let them mature with the workload; governance here grew from an early special interest group into the Clinical Governance Committee as deployment expanded. The second is to begin with a constrained task built on a highly validated reference dataset; in this case, the ANZICS CORE registry was used. The third is to plan upstream education for clinical data-entry staff, because errors at the point of entry propagate through every subsequent layer of analysis.

Two structural choices would change with hindsight. A cocommercial lead arrangement retained by HeartAI alongside the funding partner would have made engagement with third parties more efficient as the products matured. Incorporation was delayed until 2022 while public ownership models were exhausted; the process was challenging but ultimately legitimized the decision to incorporate. More broadly, EMR deployment alone is insufficient to realize the value of digitized health data, and complementary platforms that extract, restructure, and present EMR data for specific clinical, operational, and research purposes appear to be a necessary next step in the digital health maturity of large health systems.

Future work has 3 priorities: completing CODEXA’s operational validation and evaluating its effect on coding accuracy and hospital funding in production use, progressing the Patient Flow application from co-design to deployment and evaluation, and transitioning CCCIS from research to operational status, with the infrastructure and governance changes required by the transition, ahead of wider scaling across SA Health.