Imagine a future where medical professionals are alerted to a patient’s deteriorating condition well before physical signs can be seen. That’s a future CSIRO is working towards, leading a new study to develop such technology through machine learning.
The study, CSIRO explained, showed that early warning deterioration alerts can be set to monitor patients two to eight hours before they are triggered by current clinical criteria. This early warning detection isn’t through the use of sensors, but rather an algorithm that pulls data stored in electronic medical records.
Medical professionals could use the data contained in electronic medical records (EMRs) to predict when a patient’s vital signs such as blood pressure or temperature are likely to reach a danger zone, triggering patient decline. But that would require medical records to be kept up to date, and for people to actually use them. My Health Record has a pretty abysmal uptake and as someone who received one as part of the initial trial, there are chunks missing from my record as I moved state-to-state, bulk-billed late-night medical centre to medical centre. There are a number of EMRs out in the wild, not just My Health Record, but a problem CSIRO would run into is interoperability with other hospitals, for example, and the issue of the files not being complete.
“With the massive amount of data in the EMR comes the potential for better patient care,” CSIRO wrote in a press release.
“For example, the information from the data can be used to help medical staff make decisions that can prevent a patient’s deterioration from adverse events and acute illness. Up until recently, and still in some hospitals, patient data was not available electronically, restricting the capacity to develop digital tools to benefit from it.”
As CSIRO scientist Dr Sankalp Khanna added, until now, there hasn’t been a way to harness all the data in the EMR to predict patient health.
“This new tool has the potential to transform the day-to-day functioning of health systems”, Khanna said.
The alerts would warn medical staff when a patient is at risk of deterioration leading to possible death, cardiac arrest or unplanned admission to ICU, CSIRO said. The tool can also notify of the need for clinical intervention.
“When applied to a test cohort of 18,648 patient records, the tool achieved 100 per cent sensitivity for prediction windows two to eight hours in advance for patients that were identified at 95 per cent, 85 per cent and 70 per cent risk of deterioration,” Khanna said.
But there’s so much that needs to be ironed out with EMRs to make this machine learning tool a reality. Incomplete datasets would do more harm than good in this situation. Hopefully we get to a place where this tool can be utilised to its full potential. CSIRO scientists are now in discussion with partners for a clinical trial to explore how the alerts work and how they can be best implemented into clinical workflows.