Some healthcare provider organizations are using machine learning and other forms of artificial intelligence to provide clinicians with the best evidence-based care pathways.
A group’s aim could be to improve a patient’s care plan based on personalized analytics. Another goal could be the further merging of evidence-based care paths with historical utilization and outcomes in order to offer optimal patient care. Provider organizations might be using social determinants of health combined with machine learning to offer clinically meaningful services.
Healthcare IT News talked over these ideas with Niall O’Connor, chief technology officer at Cohere Health, a vendor of artificial intelligence technology and services designed to improve the provider, patient and payer experiences.
Q: How is machine learning being used to comprehensively enhance a patient’s entire care plan based on personalized analytics? And how is machine learning being used to combine evidence-based care paths – with real-world historical utilization, outcomes and the latest literature – to provide first-rate patient care?
A: Evidence-based guidelines are an important component of an intelligent care path solution. In fact, they are the starting point for our models. We would never want to relearn the complexity that has been elucidated in clinical guidelines.
At the same time, guidelines were written for the average patient and can’t possibly accommodate all the comorbidity permutations that exist for patients of high acuity. This is where machine learning can help. For patients that don’t perfectly fit existing evidence-based care paths, we can employ machine learning models to infer what has been the most efficacious path for diagnostically identical patients from real world historical data.
Q: How is machine learning being used to use social determinants of health and patient lifestyle to provide precise and clinically meaningful care?
A: Data regarding social determinants of health (SDOH) and patient lifestyle are not typically captured in standard electronic health records, but diligent physicians typically refer to this type of data in their clinical notes.
We can also supplement models with SDOH data – such as the U.S. Census – that can point to access or other patient challenges and incorporate patient-reported data, whether on lifestyle or health state.
This presents a challenge for typical analysis, so we employ natural language processing to help isolate and interpret references to things like lifestyle impacts or resumption of employment following surgery. Although detection of these phrases isn’t comprehensive, when present they can help provide valuable outcome endpoints for us.
Q: How did machine learning first come to be seen as useful in these areas?
A: Clinical data analysis isn’t a big data problem; it’s a messy data problem and is plagued by the fact that much of the valuable information isn’t readily available in structured form.
Volume also plays a big part in why we use machine learning; when we end up with thousands of attributes, we need machine learning to identify the variables that are driving the model. For SDOH in particular, machine learning will be crucial for variable selection and essential to refining some signal from the noise of operational clinical data.
Share:
About the Author: Cohere Health
