Improving patient outcomes with artificial intelligence

Improving patient outcomes with artificial intelligence

Photo composite: Freepick/Microsoft and Pangaea 

Pangaea’s Intelligence Extraction and Summarization uses AI to diagnose patients, much like a clinician does manually using all relevant data from a patient’s record  

Microsoft’s Sally Frank discusses some of the innovative life sciences start-ups that are using AI and machine learning models to better identify illnesses, manage patient care and free up clinician capacity 

Guest contributor |

Regardless of how you experience healthcare – as a provider, staff member, patient, or family member of a patient – all of us see inefficiencies that other, mostly non-regulated industries have overcome.  

Trying to transform an industry focused on people’s health is more difficult and has more at stake for individuals, populations, providers, life sciences and regulators. In healthcare, we struggle with data interoperability, consistency in care, resource management and effective innovation. Fortunately, there are trailblazers using artificial intelligence and other technologies to overcome these seemingly insurmountable challenges in healthcare. 

Signal 1 is a health AI start-up with the mission to transform patient care through responsibly deployed AI. The company provides hospitals with an end-to-end solution for integrating AI-driven insights into existing hospital workflows. Its most popular application is a clinically validated, real-time automated patient discharge predictor that has been developed and deployed at Grand River Hospital in Waterloo, Canada. This solution helps hospitals to improve quality and flow while reducing stress on frontline care providers. 

Discharge planning for admitted hospital patients involves many multi-disciplinary teams, including physicians, nurses and social workers. These teams have limited resources and are difficult to co-locate to coordinate discharge planning for each patient. This frequently leads to communication gaps, misaligned priorities and delays in resolving each patient’s discharge barriers which results in delays that extend a patient’s stay. Staffing challenges have exacerbated this issue, creating a situation where stretched resources are working in a very reactive environment while hospitals struggle to free up the necessary capacity.  

Many of the challenges in coordinating these teams relate to understanding which patients are approaching discharge readiness ahead of time, so that resources can be allocated appropriately to remove barriers in parallel with this approaching date. This is where the power of a machine learning-based risk predictor comes in. By studying historical datasets, the ML algorithm can uncover subtle relationships between a patient’s physiological data over time and their timeline until reaching clinical stability. The evidence is mounting that ML is superior to alternative approaches when it comes to addressing this problem. 

Signal 1 provides an ML discharge predictor that enables hospital clinicians to be more proactive and coordinated about discharge planning. By studying historical datasets, Signal 1 tailors their algorithm to each hospital’s local population to provide reliable predictions on which patients will be ready for discharge in the next 48 hours. Signal 1 integrates these notifications into a hospital’s existing workflows so that high priority patients are surfaced to multi-disciplinary teams at the right time. This provides hospital frontline teams with better visibility on which patients they should prioritise for discharge activities and what is required to enable that. 

However, before you can determine if a patient is discharged, you have to properly diagnose them. Clinicians spend more than 35 per cent of their time capturing patient data, symptoms, family histories, lab results and other key information in the form of digitised patient records. Despite this, more than 60 per cent of patients with rare and hard-to-diagnose conditions are not diagnosed in a timely manner. This is due to a lack of scalable and speedy characterisation of patients, which is based on clinically actionable intelligence from such records, and helps clinicians map patient journeys and disease trajectories. 

And this is where another business shines. Founded in 2018, Pangaea Data is a life sciences technology firm that provides Pangaea’s Intelligence Extraction and Summarization (PIES). The solution is driven by novel, unsupervised AI to characterise patients across 4-5,000 hard-to-diagnose conditions in a scalable and privacy-preserving manner, much like a clinician does manually using all relevant data from a patient’s record.  

Pangaea’s unsupervised AI reduces the bias observed in supervised natural language processing and text mining approaches since its does not require a preempted list of features to extract from a limited set of textual notes. It requires significantly less data to start with, since it already has a library to characterise patients across multiple conditions, combined with a medical knowledge base. PIES has a proven track record of success across different disease areas such as oncology, respiratory, cardiovascular, auto-immune, neurodegenerative and mental health. For example, in a dataset of 8,000 cancer patients, 51 had previously been identified as having a condition called cachexia, which causes unintentional weight loss, based on their patient records. PIES correctly identified the initial 51 patients and then found an additional 316 who also had the condition but were undiagnosed, misdiagnosed or miscoded. Following this study, PIES was deployed on a larger dataset of around 29,000 patients where it found 1052 per cent more cancer patients with cachexia, who were hidden in plain sight. These results were validated with help from clinicians. 

Both Signal 1 and Pangaea Data are helping improve patient outcomes by, using data and making it more readily available for analysis, generating analyses that result in more consistent care, and enabling clinicians to use data insights that drive effective care pathways.  

Sally Ann Frank is worldwide lead for health and life sciences at Microsoft for Startups 

This article was originally published in the Spring 2023 issue of Technology Record. To get future issues delivered directly to your inbox, sign up for a free subscription

Subscribe to the Technology Record newsletter

  • ©2024 Tudor Rose. All Rights Reserved. Technology Record is published by Tudor Rose with the support and guidance of Microsoft.