Platform technologies

Widespread use of national terminologies by clinical systems provides considerable interoperability benefits and supports meaningful use of patient data for better health outcomes. However, rich and powerful clinical terminologies, such as SNOMED CT, are complex in nature.

This complexity makes implementation difficult and often costly, presenting a challenge to adoption. To address this challenge, we are developing new technologies that enable the advanced use of clinical terminologies such as SNOMED CT, LOINC and any FHIR-based CodeSystems.

Ontoserver

Ontoserver is the world-leading clinical terminology server implementing FHIR terminology services and supporting syndication–based content distribution.

Over the last year Ontoserver has continued to receive many new updates including:

• Only 3rd party terminology server to be validated for use with HL7 IG Publisher and Validation infrastructure
• Significant enhancements to code validation and translation support
• Further pre-adoption of FHIR R5 features where no backward compatibility issues exist
• Extended support for new SNOMED CT expression constraint language features
• FHIR R5 support

Ontocloak

Ontocloak is an authorisation server for managing access to Ontoserver and other related services.

Atomio

Atomio is a syndication service for managing distribution content.

Snapper

Snapper: Author is a web browser-based app for authoring FHIR terminology resources and publishing them to a FHIR terminology server.

Snapper: Map is a web browser-based app that enables authoring maps from legacy terminology to standards-based terminologies. Together, these tools support migration to and use of standard terminologies, and the adoption of the national approach to interoperable digital health information.

Snap2Snomed/Snapagogo

Snap2Snomed is an open source tool built and operated for SNOMED International to support collaborative mapping of term lists and local vocabularies to SNOMED CT. It builds on expertise developed with Snapper and leverages the automapping capabilities of Ontoserver to provide collaborative mapping to an international audience including SNOMED member countries and vendors such as Babylon Health.

Snapagogo extends the capabilities of Snap2Snomed by supporting additional target code systems hosted by Ontoserver such as LOINC, RxNorm, and ICD 10. Snapagogo is being made available to the Australian research community through a collaboration with the Australian Research Data Commons (ARDC).

SnoMAP

SnoMAP is a suite of SNOMED CT to ICD10-AM mapping products that enables diagnoses to be recorded using SNOMED CT-AU and mapped to ICD10-AM codes. We have developed two products:

• SnoMAP Starter: a simple SNOMED CT-AU diagnosis to ICD-10AM Codes FHIR ConceptMap, to support the use and reuse of SNOMED CT for analytics and research activities.
• SnoMAP ED: a mapping service for emergency department non-admitted patient reporting purposes, thus supporting the use and re-use of the standard clinical terminology for ED funding activities. This has been revamped to support mapping directly to the IHACPA ICD10-AM shortlist.

Snorocket

Snorocket is our classifier, which for the first time enabled semi-real-time authoring of very-large-scale clinical ontologies like SNOMED CT. Snorocket is available under an Apache 2.0 open source licence and as a Protégé plugin. It has also been licensed to SNOMED International and the Australian Digital Health Agency for their ongoing maintenance of SNOMED CT.

Shrimp

Shrimp is a widely used tool for browsing SNOMED CT, LOINC and other FHIR CodeSystems, powered by Ontoserver.

Supporting users around the world

Our clinical terminology and FHIR® enabled products are in use globally to support the advanced use of SNOMED CT, management of ValueSets and ConceptMaps and syndication of clinical terminologies. Shrimp and our public testbed are used worldwide. Ontoserver is also licensed commercially by users in Australia, New Zealand, Switzerland, Germany, England, Wales, Scotland, Estonia, Sweden, France, Belgium, Estonia, Indonesia, Canada, and the United States, with evaluation licences used across the United States, ASEAN region and South America. There are also several managed-service and embedded-use reseller licences in place with vendors in Nth American and Europe.

Pathling

Pathling simplifies the use of HL7® FHIR® and clinical terminology within data analytics. It is built on Apache Spark, and includes language libraries and a server implementation.

Pathling was designed to assist with these primary use cases:

1. Exploratory data analysis – Exploration of hypotheses, assessment of assumptions, and selection of appropriate statistical tools and techniques.
2. Patient cohort selection – Selection and retrieval of patient records based on complex inclusion and exclusion criteria.
3. Data preparation – Processing and re-shaping data in preparation for use with statistical and machine learning tools.

Pathling uses FHIRPath expressions for the aggregation and transformation of data, along with powerful and expressive search queries. This makes it easier to select and transform FHIR data as compared to a generalised query language such as SQL, and it also allows us to extend the functionality of the FHIR API to make it more capable for analytic use cases.

Pathling also integrates with the FHIR Terminology Services API to enable advanced terminology functionality within queries, at query time and at scale. This allows users to access terminological information and join it to clinical data in arbitrary ways, including advanced support for SNOMED CT and its expression constraint language.

Language libraries are available in the Python, Java and Scala languages, allowing for deep integration into existing applications and data science workflows. The server implementation provides a standard FHIR interface to analytic query operations, and is suitable for the delivery of web and mobile applications.

Smart Forms

Smart Forms is an open-source web browser-based app for rendering FHIR Questionnaires. It supports clinical integration with healthcare systems for capturing standards-based health information.

Based on FHIR’s Structured Data Capture specification, it notably provides these additional capabilities for forms:

• Data pre-population – Allows re-usability of data by pre-filling data existing in the healthcare system.
• Conditional rendering – Rendering of questions based on user decisions or pre-filled data.
• Calculations – Dynamic calculation of quantitative results based on answers existing in the form.
• Terminology support – Allows use of standardized medical terminologies to represent medical concepts and data in forms.

Smart Forms provides an open-source library for its React-based rendering engine which is recently adopted by a few Australian and New Zealand-based organisations in the digital health space. Other supporting libraries for data pre-population and modular assembly of sub-questionnaires are also provided.

The development of Smart Forms has been funded by the Department of Health and Aged Care.

Despite the increasing adoption of electronic medical records (EMRs) and the shift towards more formalised and structured content, clinical records will invariably contain sections of narrative or free-text information. These narrative sections often hold rich, valuable insights that are important for patient care. To fully utilise this information, it must be effectively queried, analysed, and reported. This presents an ongoing challenge and opportunity in healthcare, as extracting meaningful data from free-text requires advanced techniques in natural language processing (NLP), deep learning, and large language models (LLMs).

Medtex

Medtex is a semantic medical text analysis software that analyses free-text clinical documents for informing clinical decision making.

Medtex learns what statements to look for and uses SNOMED CT, the internationally defined set of clinical terms, to unify and reason with the language across information sources. It incorporates domain knowledge to bridge the gap between natural language and the use of clinical terminology semantics for automatic medical text inference and reasoning.

Analysis engines using the Medtex technology have been developed to:

• Standardise the free text by identifying medical concepts, abbreviations and acronyms, shorthand terms, dimensions and relevant legacy codes
• Relate key medical concepts, terms and codes using contextual information and report substructure; and
• Use formal semantics to reason with the clinical concepts, inferring complex clinical notions relevant to a health application.

Medtex scales to large amounts of unstructured data and has been integrated within a highly distributed computational framework. It turns the medical narrative into structured data that can be easily stored, queried or rendered by most systems for use in their health application. Medtex has been utilised to deliver solutions to healthcare practitioners, including cancer registries, and hospital radiology and emergency medicine departments. These solutions encompass:

• Analysing pathology and radiology reports, and death certificates, to extract cancer characteristics important for assessing cancer incidence and associated mortality rates.
• Analysing pathology and radiology test results, as well as discharge summaries, to streamline test result review management workflows in emergency departments, and to identify patients with misdiagnoses and/or those undergoing inappropriate antibiotic treatments due to drug-resistant infections.
• Analysing medical reports to provide the capability for medical record searching and advanced analytics.

Search engines for health data

With a rapid increase in health data—in all its myriad of forms—the need to effectively search this data rises. Simultaneously, much of this data is unstructured, making it difficult to search using methods tailored to structured data. Search engine technology was designed specifically for large amounts of unstructured data, making it well suited to the health domain.

We developed a suite of solutions for searching health data. Nowadays most of our work involves the training of AI models, specifically neural network models for ranking and natural language processing. These models are adept at understanding the meaning behind a user’s query and the relevant information they are looking for, making them much better at finding that relevant information.

The development of our search technology is driven by the idea that people are looking for information to make important health decisions. As such, we develop solutions that support the decision making, empowering users with the information they need rather than ceding control to a black box system.

Key technologies we have developed include:

• Evidence-based search systems capable of ingesting all of PubMed and all current clinical trials and suggesting relevant evidence to support clinical decision making
• Automated the matching of patients to clinical trials (or visa versa)
• Targeted cancer treatment recommendation for children with specific genetic findings
• Systematic analysis of human search behaviour in the health space to inform the development of better search engines

Chatbots bolster engagement in human-computer interaction. Fortunately, healthcare provides a plethora of opportunities for chatbots to support patients, carers and clinicians. A chatbot enables interactions between a knowledge base and a user in speech or text. Each chatbot is powered by a “brain” which needs to be developed and trained to support engaging dialogues. We’ve developed a range of chatbots for clinical and social settings. Recent examples include:

• The “What Matters 2 Kids” project uses a chatbot to engage with First Nation children that asks them to draw what matters to them. The chatbot then inquires about the drawing in order to capture the well-being of the child.
• “Dolores” a chatbot to discuss all things related to chronic pain with language suitable for the age of the user. Dolores has been piloted at pain clinics at the Royal Brisbane & Women’s Hospital and Melbourne Children’s Hospital.
• “Quin” a smoking cessation chatbot built from thematic analyses of Quitline counselling sessions. Quin is being designed for long-term use and support for a user wishing to cease smoking.
• “Aurora” a chatbot that administers a sleep-dependent memory test (developed at the University of Sydney) for people living with mild cognitive impairment. Aurora handles the testing within a critical time window and reschedules if the user misses the test.

AEHRC chatbots can function on mobile devices without requiring internet access and have support for:

• Smart on FHIR
• Incorporate LLMs or run entirely on a mobile device for privacy and security
• Custom user interface widgets
• Voice logging and processing

Example dialogues of Quin the smoking cessation chatbot. Here Quin is asking if anyone else lives with the user who smokes to anticipate triggers the user might be exposed to. The user may respond verbally, text or by drawing.