Skip to main content



Todd Stewart, M.D., Vice President, Clinical Integrated Solutions





Business Challenge

Managing clinical variation is a challenge for any provider, but it’s especially difficult when managing multiple hospitals across a broad geography.

Mercy, a healthcare network of 40+ hospitals and 700+ physician practices and outpatient facilities headquartered near St. Louis, MO, set out to standardize processes, reduce variation and improve health outcomes by creating clinical pathways (best practice steps in caring for patients with specific conditions).

Previously, Mercy’s pathways were based solely on medical literature and a panel of expert providers’ experience. Mercy realized there were opportunities for improving the efficiency and effectiveness of the pathway process.

Analytics Solution

Today, Mercy is applying machine learning to help tackle clinical variation across its health system for a more effective approach to developing, deploying and managing clinical pathways.

In 2014, Mercy partnered with Ayasdi, a Silicon Valley machine intelligence pioneer, to gain the ability to explore vast amounts of data to discover critical patterns related to the best medical treatments.

Initially, Mercy chose three pathways (total knee replacement, laproscopy and lumbar fusion) to see if machine learning could validate or potentially improve earlier work. Using unsupervised (i.e., unbiased) machine learning, the team constructed “topological summaries” of thousands of EHR data points covering all possible medical treatments, as well as other variables that impact outcomes. With this approach, Mercy could clearly identify the treatments that led to the best results.


Mercy’s new approach helped contribute to a five percent cost reduction in total knee replacements while maintaining low rates of mortality. Overall, Mercy has demonstrated that using pathways reduces mortality by 30 percent or more and cost per case by $800+, resulting in $14 million saved in fiscal year 2016. While Mercy has 40+ pathways in use today, it plans to apply machine learning for as many as 80 in total to reduce variation, lower costs and ensure patients are receiving consistent, high-quality care.