The patient journey is one that we have all experienced. A complex and unique journey for each patient, it usually begins at our local medical centre. Manned by the gatekeepers of healthcare, our GPs, this crucial stage of the journey is usually the one that determines whether we enter secondary care. From here, a series of important decisions are made that will impact where, when, and how we receive care. One of these decisions is the triage of referrals, a task that can have huge implications on when we get seen, and ultimately treated.
Triaging of GP referrals is a task that warrants careful consideration, with each referral document laden with data about complex patients. However, this does mean that doctors may spend many hours a week triaging referrals; placing patients into various risk categories, or sending them directly for investigation. Although the electronic replacement for paper-based referrals has already significantly improved this process, the triaging process remains time-consuming and costly. Triaging of referrals has huge implications on both the utilisation of resources and patient outcomes – for instance, if one patient is seen sooner it impacts other patients and determines whether there will be delays in treatment. This PDH project aims to streamline this process.
Today’s world grants us opportunities to utilise large, diverse amounts of data. For precision medicine to become a reality, these opportunities demand data-driven health; the ability to harness the wealth of data available and turn it into meaningful information that can be used across many aspects of healthcare. Data-driven health in its current state faces the challenge of using datasets built for storage and reporting, not data science. As health data takes various shapes and forms, it can be difficult to navigate through, which often means data-driven health projects are usually done on an ad-hoc, hospital-specific basis.
Building on other PDH projects, this project aims to make the triaging of electronic referrals more efficient by using machine learning techniques, more specifically, deep learning. Machine learning is the use of algorithms to take existing data and learn from it to make predictions, and deep learning, a subfield of this, results in an artificial neural network that, much like our brain, can continually learn and make decisions. This means that existing data about referrals will be used to build a decision-support system that can predict triage codes.
This ambitious project also aims to change the approach to data-driven health in New Zealand by focusing on transfer learning – how hospital-specific knowledge can be applied to other hospitals and populations. Transfer learning leverages knowledge already gained by utilising a machine learning model trained for one purpose and building on it to use it for another purpose. In practice, this means that learnings from one hospital can be transferred and built upon for another hospital. This project will involve the application of a machine learning model in one hospital which can be compared to clinicians’ performance. The original hospital-specific model can then be applied to a second hospital with transfer learning.
Automation of tasks, such as triaging, using deep learning has potential to reduce the burden on our health system, freeing up time for clinicians and streamlining the patient journey, leading to consistent decisions, reduced costs and, ultimately, better health outcomes.