Open-Source Repository

Machine Learning in Healthcare: Now for Everyone

Salt Lake City, UT (USA), February 2017 - Machine learning is a part of everyday life for most Americans, from navigation apps to Amazon's omniscient purchase recommendations. But in healthcare, the use of machine learning has so far been limited to niche science projects in large and academic health systems - those able to afford the highly skilled data scientists and dedicated teams required to turn their data into meaningful performance improvements. 

Health Catalyst is on a mission to change this by embedding the value of machine learning throughout healthcare. Last month, the company launched healthcare.ai™ to help make machine learning routine, pervasive, and actionable for healthcare organizations of all sizes. The collaborative, open-source repository of machine-learning tools and expertise, including topical blog content and weekly live hands-on machine-learning educational broadcasts, makes it easy to deploy machine learning in any environment.

Now, to bring the life-saving technology to hospitals and patients everywhere, Health Catalyst  is embedding machine learning as a core capability across the company's entire product line in an initiative called catalyst.ai™.  With optimized machine-learning models built into every Health Catalyst application, organizations can leverage the technology for predictions such as identifying patients who are most likely to acquire deadly infections; finding those who may have trouble paying their medical bills; spotting possible canceled appointments before they happen; or launching proactive medical interventions for patients who are at risk for dangerous complications.

Together, healthcare.ai and catalyst.ai represent the next generation of healthcare analytics. With these machine learning innovations now readily available to organizations large and small, American healthcare will be equipped to exchange today's limited, retrospective analysis for a new era of powerful, predictive analytics driving an orders-of-magnitude improvement in outcomes. 

"Predictive analytics powered by machine learning has truly vast potential in healthcare, but we lag other industries by several years largely because early efforts were extremely expensive one-off models requiring an army of data scientists to write and test the algorithms behind the technology," said Health Catalyst Executive Vice President of Product Development Dale Sanders. "catalyst.ai solves that problem by lowering the bar for entry and enabling data architects and analysts to become 'citizen data scientists.'

"Another factor limiting the usefulness of machine learning is that healthcare data is far more complex than data in other industries and difficult to aggregate. Machine-learning algorithms are only as good as the volume and quality of data that feeds them. We've invested tens of millions of dollars over the last few years to create high-volume, high-quality data content that these algorithms thrive on, and we are embedding the results in the workflow of clinicians in every department across a hospital or health system."

Indiana University Health uses machine learning to reduce hospital-acquired infections

Health Catalyst machine learning has already been deployed at multiple client sites with promising outcomes. Indiana University Health (IU Health), a 17-hospital nationally recognized healthcare system, engaged Health Catalyst machine learning in an attempt to reduce healthcare-associated infections (HAI) and achieve zero central line-associated blood stream infections (CLABSIs). Of the 41,000 patients who develop a CLABSI in US hospitals each year, one in four die, according to a Centers for Disease Control study. IU Health wanted to be able to predict which patients are most likely to develop a CLABSI so clinicians could proactively undertake prevention activities 100 percent of the time.

IU Health used Health Catalyst's enterprise data warehouse (EDW) and catalyst.ai-driven analytics to bridge information gaps in its EMR data and to paint a complete picture of patients' CLABSI risk.  Models on the data were developed and tested by Health Catalyst using machine-learning algorithms such as logistic regression and random forest - the workhorses of the machine-learning world. This led to the development of a CLABSI risk-prediction model that is built into a unit-level dashboard used by nursing staff to identify patient-level care gaps. In addition to the risk score, the top three risk factors for each high-risk patient are displayed, providing immediate insight into specific actions that can reduce the CLABSI risk for these patients.

As a result, today IU Health can predict which patients with a central line will develop a CLABSI with an estimated 87 percent accuracy and a false positive rate of 0.16.

"We have quite a few initiatives centered on hospital-acquired harm events that are supported with analytics," said Kristen Kelley, Director of Infection Prevention at IUH, in a recent Scottsdale Institute report on the project. "We are on the cutting edge. In the past six months we've experienced a 20 percent decrease in CLABSI and a 30 percent drop in harm events overall. So we are headed in the right direction."

Machine learning models available today

In addition to its CLABSI risk prediction model, Health Catalyst has leveraged catalyst.ai to deploy numerous predictive models at launch that are useful for clinical, financial, and operational decision support, including clinical decision support models for

  • central line-associated bloodstream infection future risk (CLABSI)
  • chronic obstructive pulmonary disease (COPD) readmission risk
  • pre-surgical risk for bowel surgery
  • diabetes future risk
  • financial and operational decision-support models, such as propensity to pay and appointment no-show risk

In addition to predictive models, Health Catalyst is currently developing algorithms that use machine learning to identify treatment patterns that lead to better outcomes. These algorithms will be used to drive treatment decisions on patients by showing how "patients like this" were treated and what the outcomes for a cohort of similar patients were.