Over the weekend of the 15th and 16th of October, more than 200 participants worked on unique healthcare challenges across Australia, using real world de-identified healthcare datasets at the National Healthcare Datathon, organised by IntelliHQ and ANZICS – a first event of its kind.
In Brisbane, Melbourne, Perth and Sydney, clinicians teamed up with data scientists, technologists, and data & computer science students to develop solutions to healthcare problems using artificial intelligence and machine learning models.
The collaboration between these multi-disciplinary teams generated impressive data driven solutions, developed using Datarwe’s platform, the Clinical Data Nexus (CDN). The CDN platform allowed the teams to explore the healthcare data sets, visualise the data patterns, prototype and test ai-ml models in a secure manner.
2 teams in each region were selected at the end of weekend to compete and pitch at the National Healthcare Pitch Night on the 20th of October. Below are the solutions developed by the national and regional winning teams. Congratulations to all participants!
National Winning Team
After close deliberation, the first place went to ‘Sydney Team 9’ who developed an algorithm to combat the lack of universally recorded pressure injury incidence.
NLP classification algorithm
The Sydney’s Team 9, from NSW Agency for Clinical Innovation (ACI), developed a natural language processing (NLP) classification algorithm that can assist in reducing morbidity and mortality of patients suffering from pressure injury, and reduce the burden of this medical injury, which costs 820 million dollars of healthcare funding each year:
➡️ By determining if a patient has a pressure injury, or not, based on the documentation in clinical progress notes.
➡️ Which will help clinicians leverage a wealth of information locked behind text-based data and provide real time monitoring by detecting words associated with an increased incidence of adverse events like pressure injuries to provide immediate feedback to ICU units so they can mitigate risks, improve patient outcomes and increase resource allocation.
📈 Data sets used: they used the Kaggle open-source medical notes data and they visualised, built and tested their models securely in Datarwe’s Clinical Data Nexus platform to develop the bias-mitigated solution.
Whilst initially applied to pressure wounds, this NLP solution has the promise of scaling to other injury types. The value to clinical staff in time savings promises to scale across a variety of applications.
National Runners up
Sydney team ‘Hyperthermic Headbangers’ and the Melbourne team ‘There Won’t Be Blood’ are the dual runners up of the National Healthcare Datathon.
A model using the Trapezoidal rule and AUC for fever and mortality in patients admitted to ICU with isolated traumatic brain injury
They challenged the incidence of fever and mortality in patients admitted to the ICU with isolated traumatic brain injury (TBI) as this is associated with poor health for the condition. The team explored a model:
➡️ To predict and define temperatures over 37 degrees basing this on the Trapezoidal rule for temperatures and maximising the AUC (area under curve).
➡️ Which will help alert clinical staff of any physiological changes in their patients so they can reassess as required, improve health outcomes and enhanced operational efficiency.
📈 Data sets used: Interrogated the NSW Health ICU (eRIC) data (2017-2022) on Datarwe’s secure Clinical Data Nexus platform to explore their bias-mitigated solution.
With access to more cases, via linking the eRIC and ANZICS data on Datarwe’s platform, the model has the potential for further research and could have a lot of potential applications in clinical practice and clinical research.
Decision tool to reduce blood test in hospitalised patients
Tackling over incidence in blood draws amongst hospitalised patients’, the Melbourne team ‘There Won’t Be Blood’ worked together to develop a clinical decision tool:
➡️ To help ICU clinicians decide when blood work should be performed next on hospitalised patients based on the patients’ recent bloods.
➡️ Which will help avoid unnecessary pain for the patient, reduce waste and free clinician time that can be better spent elsewhere.
📈 Utilising data analytics and a random forest approach, the team securely explored real patient data from MIT Critical Data’s MIMIC IV dataset in Datarwe’s Clinical Data Nexus platform to develop the bias-mitigated solution.
Regional Winning Teams
Below are the solutions developed by the other regional winning teams. They did not make it to the National podium but the solutions developed are valuable assets to increase patient outcomes and hospital efficiency.
An antimicrobial resistance prediction model for intubated ICU patients
Coming together to tackle prevalence in hospital acquired infections (1 in 10 Australian patients) 10% of which are due to antimicrobial resistance, Brisbane’s multidisciplinary ‘Team 4’, workshopped a predictive model to identity antimicrobial resistance in intubated ICU patient:
➡️ To correctly detect infection so clinicians can suitably prescribe antibiotic treatments that are appropriate to the patient (20% of ICU-prescribed antibiotics are inappropriate) and are administered in a timely manner.
➡️ Which will reduce delayed treatment and improve overall patient outcome as well as enhance hospital resource management through reduced costs, freeing clinician time and ICU-bed availability.
📈 They interrogated MIT Critical Data ’s MIMIC IV dataset securely through Datarwe’s Clinical Data Nexus to explore this solution, and propose further development of an ai-powered clinical decision dashboard supported by an XR interface.
A model to identify differences of oxygen measurements for Indigenous Australians in ICU
Brisbane ‘Team 1’ probed disparity in oxygen measurements by pulse oximeters between Indigenous Australians and non-Indigenous Australians amid TGA’s safety warning in January of 2022 that these devices might not be as effective on darker skin tones:
📈 By examining a cohort of sepsis patients (who are at risk of hypoxemia – low blood O2 saturation) from the iMDsoft Metavision dataset through Datarwe’s Clinical Data Nexus platform, the team found that pulse oximeters overestimate the concentration of oxygen in blood of Torres Strait Islander people.
➡️ Which means this cohort is not receiving the oxygen they need and validated this with higher mortality rates for Indigenous Australians diagnosed with sepsis compared to non-Indigenous Australian sepsis patients.
➡️ Understanding pulse oximetry is commonly used despite known accuracy issues in people with darker skin color, the team are looking to explore whether this is true for Indigenous patient cohorts and are interested to continue exploring this on larger cohort datasets.
A predictive model of expected sepsis mortality for a given hospital
Melbourne team SPRINT worked on early detection of sepsis and optimisation of intensive care resources using machine learning. In aim of reducing sepsis mortality (11 million deaths globally, 8700 deaths in Australia) the team developed a predictive tool:
➡️ To define expected sepsis mortality for a given hospital by identifying contributing factors to sepsis outcomes and estimate the hospital performance sepsis response adjusted for patient, disease and resource factors.
➡️ Which will help clinicians detect sepsis earlier and define correct treatment plans to improve patient prognosis and outcome, reduce mortality and optimise resource allocation and research.
📈 Explored real patient data as part of the ANZICS registry available in Datarwe’s Clinical Data Nexus platform to develop the bias-mitigated solution.
Total of $30,000 worth of solution support offered by Datarwe
As an official sponsor and platform provider of the National Healthcare Datathon, Datarwe offered a total of $30,000 worth of solution development support for the winning teams of the National Healthcare Datathon.
- Access to Datarwe’s Clinical Data Nexus platform: platform, tools, and technology;
- Ongoing data engineering support, data science expertise and project support.
This prize gives the winning teams the opportunity to pursue the development of their healthcare solution after the Datathon.
The Clinical Data Nexus expands the capacity and capability of the research community to undertake collaborative translational health and medical research.
Datarwe’s clinical Data Nexus Platform: secure, data de-identification, enrichment, labelling and privacy services
Datarwe’s Clinical Data Nexus provides advanced data engineering services, which accelerates the research process. Typically, data centric research requires data de-identification, enrichment, labeling & privacy preservation, which take 80% of research project time. Datarwe provides all of these services through its Clinical Data Nexus data-as-a-service platform, allowing researchers to focus on applying their subject matter expertise to the research question.
The Clinical Data Nexus expands the capacity and capability of the research community to undertake collaborative translational health and medical research by facilitating the connections between hospitals, researchers, clinicians and product developers with research-ready, enriched real world clinical data.
To know more about our platform, reach out to us by emailing: firstname.lastname@example.org. Our team of data scientists will be happy to answer all your questions and provide you more information.