In conversation with… Navid Toosi Saidy, PhD, CEO of Propel Health AI. 

In this edition we profile one of our industry partners, Propel Health. Spun out of Max Kelsen – which this month announced that its consulting and managed services divisions has been acquired by the global consultancy firm Bain & Company – Propel Health AI has the mission to make real-world patient data more accessible and actionable.    

1. Max Kelsen was in AI before AI was even cool. Tell us about your growth story and where you are today? 

Over the last 8 years, Max Kelsen has been developing and deploying bespoke machine learning and AI systems across a wide range of sectors, from financial services, retail and mining to healthcare with the core purpose of supporting companies to turn data into real business value.  With a keen interest in healthcare, we established track record in developing and translating AI-enabled clinical data platforms, Software as a Medical Device and Digital Therapeutics in partnership with Australian and global healthcare providers, researchers, and medical device manufacturers.  

While working with our healthcare and life sciences customers, we identified significant challenges in the utilization of patient data collected over the course of their care pathway. We saw a huge opportunity to flip the chronic underutilization of data on its head and for the past couple of years have been developing an end-to-end data and AI platform named Propel Health AI” designed to overcome some of the persistent challenges in real-world evidence accessibility, data governance and infrastructure for accessing and deploying machine learning systems in healthcare.  

Just this month, it was announced that Max Kelsen’s consulting and managed services divisions has been acquired by the global consultancy firm Bain & Company, with a core focus of bringing enhanced machine learning and AI services to clients globally. Accelerated by this acquisition, Propel Health AI has been spun out of Max Kelsen with the mission to make real-world patient data more accessible and actionable.  

Propel Health AI enables healthcare and life sciences companies to ingest, curate and harmonize (FHIR) their multimodal patient data while providing the governance capability including de-identification and consent management that supports data licensing and advanced analytics. It also provides the infrastructure for responsible development, deployment, and monitoring of AI-enabled clinical & operational decision support systems.  

2. How did the DHCRC project set up Propel Health AI on a pathway to success? 

DHCRC has supercharged our partnerships with end-users across the healthcare and life sciences industry. Supported by DHCRC funding, we have strategically partnered with the Peter MacCallum Cancer Centre, Australia’s leading Oncology academic tertiary care centre and Swinburne University to implement and validate our platform. This project aims to enable data aggregation of data from various sources across Peter Mac and support the presentation of curated, harmonised, de-identified and rich multi-modal datasets to researchers, data scientists and clinicians for data-driven R&D, advanced analytics and implementation of regulated digital health solutions.  

Through this project, we have gained direct insight into their data management systems, day-to-day operations, key analytics and R&D use cases and organizational structure as both healthcare providers and a research facility. This have enabled us to tailor our platform to their needs and make it translatable to the Australian healthcare data and research ecosystem. We’ve also gained insights into where the opportunities lie to boost utilization of patient data to improve operational efficiency and the quality of patient care. 

3. There is a lot of talk about AI in healthcare. Is all the hype warranted? 

It absolutely is. AI has the potential to fundamentally accelerate and personalise the way that we deliver healthcare, including over a range of clinical and operational tasks, contingent on addressing a range of quality, regulatory, privacy and security challenges. For example, AI can be used to diagnose cancer, identify patient suitability and eligibility for clinical trials, or predict the success of therapies for an individual. AI has the potential for removing the risk for medical error, which is the third biggest killer in the US and addressing the variability in patient care as we transition to a new era in personalised medicine. On the operational side, AI can be used to streamline administrative tasks, such as billing, optimize supply chain management by predicting the demand for medical supplies and medications, and performing risk predictions to reduce unplanned hospital readmission rates which for example is estimated to cost Australia $1.5B each year. These are not the sexiest sounding applications, but they are vital for improving the operational efficiency of our healthcare systems, saving money, and allowing us to redirect resources to improving patient care.  

4. What challenges does AI pose for those working in health and the healthcare sector as a whole? 

Navigating the integration of AI systems in healthcare is a complex journey with a range of nuanced regulatory, quality, data privacy and security challenges. An immediate challenge is with regulatory frameworks, as software and AI-models that have the intended purpose of diagnosing, treating, or managing diseases, are classified as ‘Software as a Medical Device’ and are rightfully subject to strict scrutiny. Regulatory frameworks also currently require ‘locked’ algorithms, putting them at odds with the nature of AI models that need to continuously learn and adapt to avoid performance degradation caused by data drift.  

 For the purposes of using AI systems for decision making, the ‘black box’ conundrum arises: the opacity of AI decision-making processes. The life-impacting nature of healthcare requires creative solutions to enhance AI’s explainability, transparency and interpretability. The issue of data bias is also a considerable and pressing challenge. Biases in training data can be perpetuated in the output from AI models, and can inadvertently disadvantage certain patient groups if left unaddressed. It is worth noting that, in most cases, AI models are only tested with retrospective data collected from a single site and limited demographic subgroup. While promising for retrospective analyses, we need to perform validation in prospective settings, across multiple sites and demographic subgroups to ensure their generalizability, reliability and scalability. 

5. Where do you see AI in healthcare heading in coming years? 

We are only just scratching the surface of what can be achieved with AI in healthcare. Applications of AI in healthcare are already being overhauled through the advent of new foundation models, like ChatGPT, that can accelerate our ability to scale the implementation of these technologies business- and industry-wide. 

Using these emerging technologies, we are going to see a massive increase in the availability of fully integrated general AI systems capable of multimodal data ingestion and analysis that will more closely reflect the inherent multimodal nature of medicine. This means AI models capable of ingesting data in a range of formats, including structured and unstructured electronic health records, radiological imaging data, genomics sequencing data and scientific domain knowledge and performing a diverse set of tasks. These generalized AI models will be able to use very little or no data that had to be purposefully annotated, and instead be extremely adaptable to a range of tasks, and interactable via text input so that non-technical users can meaningfully engage with the tools. It is important to highlight that navigating the use of foundation models for clinical decision-making in healthcare poses intricate challenges including potential biases in data, ethical issues around patient privacy and hallucination which is a term used to refer to these models providing outputs that are not grounded in factual information. However, they could be progressively integrated into low-risk administrative tasks, like documentation and report generation, to enhance efficiency while we build confidence and address outstanding issues.  

Using these new foundation models, we are harnessing a generalized approach to the development of AI tools that is unlocking an entirely new era in AI applied to healthcare. We’re thrilled to be working at the forefront of this field, delivering cutting edge solutions to our customers. 

Categories: News,


Follow us on Twitter

Increasing efficiencies in our healthcare systems.