How machine learning speeds clinical insights

Understanding and managing unwarranted clinical variation is a significant and costly challenge in today’s value-based health economy.

Every patient is unique, so variation is a natural element in most healthcare delivery. But improving patient outcomes, minimizing medical errors and reducing costs is difficult when hospitals are unable to draw hidden insights from their own data.

Data is the catalyst for eliminating unwarranted clinical variation and is essential to care models based on value. However, the complexity and exponential growth of patient data can be overwhelming to even the most advanced organizations.

Machine learning is helping to overcome these barriers. These applications, which combine algorithms from computational biology and other disciplines, find patterns within billions of data points and help organizations uncover evidence-based insights improving the quality and cost of healthcare.

Since the 1990s, provider organizations have attempted to curb unwarranted variation by developing clinical pathways. A multi-disciplinary team of providers use peer-reviewed literature and patient population data to develop and validate best-practice protocols and guidance for specific conditions, treatments and outcomes.

However, the process is burdened by significant limitations. Pathways often require months or years to research, build and validate. Additionally, today’s clinical pathways are typically one-size-fits-all. Health systems that have the resources to do so often employ their own experts, who review research, pull data, run tables and come to a consensus on the ideal clinical pathway, but are still constrained by the experts’ inability to make sense of billions of data points.

Additionally, after the clinical pathway has been established, hospitals have few resources for tracking the care team’s adherence to the agreed-upon protocol. This alone is enough to derail years of efforts to reduce unwarranted variation.

Machine learning is the evolutionary leap in clinical pathway development and adherence. High-performance machines and algorithms can examine complex continuously growing data elements far faster and capture insights more comprehensively than traditional or homegrown analytics tools—imagine reducing the development of a clinical pathway from months or years to weeks or days.

But the true value of machine learning is enabling provider organizations to leverage patient population data from their own systems of record to develop clinical pathways that are customized to the organization’s processes, demographics and clinicians.

Additionally, machine learning applications empower organizations to precisely track care team adherence, improving communication and organization effectiveness. By guiding clinicians to follow best practices through each step of care delivery, clinical pathways that are rooted in machine learning ensure that all patients receive the same level of high-quality care at the lowest possible cost.

St. Louis-based Mercy, one of the most innovative health systems in the world, used a machine-learning application to recreate and improve upon a clinical pathway for total knee replacement surgery.

Drawing from Mercy’s integrated electronic medical record (EMR), the application grouped data from a highly complex series of events related to the procedure and segmented it. It was then possible to adapt other methods from biology and signals processing to the problem of determining optimal way to perform the procedure—which drugs, tests, implants and other processes contribute to that optimal outcome. It also was possible to link predictive machine learning methods like regression or classification to perform real-time pathway editing.

The application revealed that Mercy’s patients naturally divided into clusters or groups with similar outcomes. The primary metric of interest to Mercy as an indicator of high quality was length of stay (LOS). The system highlighted clusters of patients with the shortest LOS and quickly discerned what distinguished this cluster from patients with the longest LOS.

What this analysis revealed was an unforeseen and groundbreaking care pathway for high-quality total knee replacement. The common denominator between all patients with the shortest LOS and best outcomes was administration of pregabalin—a drug generally prescribed for shingles. A group of four physicians had seen something in the medical literature that led them to believe that administering the drug prior to surgery would inhibit postoperative pain, reduce opiate usage and produce faster ambulation. It did.

This innovation was happening in Mercy’s own backyard, and it was undeniably a best practice—the data revealed that each of the best outcomes included administration of this drug.

Using traditional approaches, it is highly unlikely that Mercy would have asked the question, “What if we use a shingles drug to improve total knee replacement?” The superior outcomes of four physicians would have remained hidden in a sea of complex data.

This single procedure was worth more than $1 million per year for Mercy in direct cost savings.

What Mercy’s experience demonstrates is that the most difficult, persistent and complex problems in healthcare can be resolved through data. The key lies in having the right tools to navigate that data’s complexity.

The ability to determine at a glance what differentiates good outcomes from bad outcomes is incredibly powerful—and will transform care delivery.