The field of data science is not a monolith–it’s nuanced. There are many different technologies and applications. We polled and interviewed our data science team to ask them what they thought were the most exciting uses of data science in healthcare today. Here’s what a few of them said and why:
Natural language processing (NLP) for care teams
Coordinating care for a member population with diverse needs and varying accessibility to resources is nothing short of a Herculean task. To do this well, many organizations rely heavily on training, a high degree of oversight, and manual performance reviews. But what if it didn’t have to be this way?
Using NLP, organizations can automatically process unstructured free text (e.g., from a chat) or audio from a phone call to facilitate scalable care coordination on multiple fronts. For example, care coordinators can leverage NLP as a virtual assistant to automatically suggest relevant actions to take on behalf of a member without extensive training. Additionally, organizations can easily analyze all member interactions, rather than a small sample, to truly measure and ensure quality of service.
– Derek Macklin
Natural language processing (NLP) for patients
On the flip-side of the coin, one of the greatest untapped data science potentials in healthcare is the use of natural language processing (NLP) to establish the patient voice.
Gaining trust and restoring agency to the patient depends foremost on improving accessibility of information. NLP can empower patients through automated text summarization, information extraction, and knowledge graphs (which map out concepts and their relationships). For example, clinical guidance from a doctor visit can be automatically distilled and linked to existing medications, treatments, and episodes to provide a complete contextual picture, while enabling patients to absorb information at a pace and setting that suits them.
Additionally, rapid adoption of conversational user interfaces have opened new avenues to build products that deliver a more holistic and integrated healthcare experience that feels more natural. Studies show that healthcare solutions with conversational agents (often virtual coaches) substantially improve management of chronic diseases including depression, diabetes, weight loss, and smoking cessation.
– Eric Carlson
Clinical decision support
Healthcare providers are finding themselves with little time to devote to each patient and investigate the patient’s medical history. This leads to lower quality care and higher costs for the patient.
How can data science provide some relief? Data scientists are actively developing tools to assist doctors here. The truly exciting work is using machine learning to identify patterns in medical data: surfacing important medical events to the provider, alerting doctors if a diagnosis or treatment seems unusual given the patient’s history, and even recommending diagnoses and treatment paths. These tools are still in their infancy, and have not been widely adopted. But I’m optimistic about the clinician superpowers we’ll see developed in the coming decades.
– Steve Martin
Deep learning for modeling disease progression
Healthcare is very personal, and one approach for personalizing care delivery is via machine learning models that understand the healthcare trajectories of individual patients.
One exciting result in this area comes from researchers who used an advanced machine learning technique, known as deep learning, to predict unplanned readmissions by taking into account the irregular timing of prior healthcare events and interventions.
In their research, this approach yielded a 7.6% improvement in readmissions predictions, compared to the best traditional approaches. Used in a hospital setting, this would help identify higher risk patients who might need additional care and coaching to keep them out of the hospital. That would be a big win for the patients and the healthcare system.
– Peyton Rose
Episodes of Care
Healthcare providers and organizations, including Grand Rounds, must truly understand members and their needs to properly navigate them to appropriate care. Insights needed for this task are often derived from member and provider encounters claims data.
To best understand needs, claims data needs to be segmented into episodes of care, such as a surgery and its associated services as a single episode of care. This creates homogeneous and clinically segmented views that provide insight into individual patient journeys, provider and practice performance, and population health, as episodes of care enable “apples to apples” comparisons between groups.
Episodes have promising future use cases, such as surfacing more holistic member medical views to care coordinators and identifying members with impactable healthcare events.
– Jack Sullivan
Social determinants of health and health care equality
Health care inequality is a huge problem, causing families and individuals to suffer unnecessarily. Adverse health outcomes are consistently linked to social determinants of health, such as poor access to care, racial segregation & discrimination, low income, and low education. The cost is also enormous; disparities are estimated to cost $93 billion in excess care and $42 billion in lost productivity per year.
Data science can be used to address these disparities by reducing or eliminating biases in algorithms and data, surfacing insights & recommendations for care teams (e.g. encouraging vaccination among members from communities with low vaccination rates), and supplementing targeted outreach efforts.
– Daniel Austin
Mark your calendars! Register for our June 10th webinar: How Data Science Makes Healthcare & Health Benefits Smarter.