Reducing Patient Costs through Machine Learning and Comparable Supply Categorization

Pexels – CC0 License


Written by Megan Bultema, Chief Data Scientist and Jackie Kokx, Data Scientist

Until recently, clinical practitioners focused almost exclusively on providing individual patients with safe and accurate treatment. Physicians favoured methods that yielded the best outcome for patients solely on a case-by-case basis.


The needs of the medical sector, however, are changing. An increase in demand has highlighted the variation that exists between patient treatments. Approaching surgeries on a case-by-case basis inherently adds an element of inconsistency between the cases for both patient outcomes and patient costs.  To tackle this rise in demand and address the variable patient outcomes, there is a need for doctors to shift their focus to a view that includes both accuracy and clinical variability.


The difference in patient expenses for the same procedures, even though their needs and circumstances are identical, can be sizable. Unfortunately, traditional analytics do not allow medical establishments to get a handle on these variations, contributing to the estimated $750 billion of wasted healthcare spending every year.


Different Supplies for the Same Clinical Purpose

One of the driving factors in surgical variation and waste is the use of different supplies for the same clinical purpose – some with substantial cost and outcomes variances. Significant progress can be achieved by identifying and reducing disparate supply utilization within comparable supply categories. Comparable supply categories, custom-built by Empiric Health clinical experts, group supply items that surgeons can substitute for another. For example, there are currently 199 items identified in the Empiric Health supply category for ‘BIOPSY FORCEPS’. Under traditional UNSPSC commodity classification these items would be fragmented across ten different categories (Endoscopic specimen retrieval forceps, Endoscopic retrievers or sets, Surgical clamps or clips or forceps, etc.) rendering direct comparison impossible. The categorization process aims to provide the Empiric Health team with an opportunity to identify where variation exists and find lower-cost supply options. In layman’s terms, it identifies ways to reduce the cost of a procedure without changing the technique of the procedure.


The Complexity of Supply Item Masters

The system currently has five hierarchical layers and has more than 5,800 distinct categories covering more than 1.6 million items. The supply use variation is viewed relative to specific procedures and surgical cohorts, allowing surgeons to see exactly where cost savings are possible. Along with the comparable supply items, practitioners can also analyze outcome and efficiency measures within the cohort, producing a more comprehensive picture of the opportunity. The number of supply items on the market is enormous, with the average hospital using more than 200,000 distinct supplies within the surgical space. Given this large number, manual categorization would be incredibly cumbersome. However, Empiric Health has developed an automated system of layered machine learning models to predict the placement of supply items into categories based on an item’s name, manufacturer information, and cost.


Machine Learning to Categorize Supplies

Empiric Health’s system uses a “model factory”- or series of independently trained Machine Learning models – with binary classification methods for each supply category. The results from all models are then combined and evaluated to determine an item’s best category placement. Using many binary machine learning models, as opposed to a single multi-class model, has been demonstrated to be more accurate and trustworthy in the automated process. This ability to quickly categorize supplies facilitates the discovery of supply variation and opportunities within relevant cohorts.



Physicians will always need to focus on quality and the best outcomes for their patients. There is a pressing need, however, to expand physician perspective to understand and reduce unnecessary variation leading to waste. Positioning this data with the right clinically relevant factors, such as comparable supply categories, is important. Providers must take the time to dig into their data and apply new tools that allow them to gain insight into the variation and act. Machine Learning models, such as the ones deployed by Empiric Health, are enabling clinically driven insights that can change the way physicians deliver care. Tackling clinical variability with a clinically driven approach is the first step in closing the gap in a health care system’s significant wasted spending.



For the latest insights and to connect with us directly, follow us on: