Using Analytics to Determine the Cost of Care

By Jon Mattes, MSDS

Understanding the cost of care is one of the most difficult challenges in maintaining a healthy practice. Having a detailed understanding of the variable costs involved in treating your patients helps with everything from strategic planning [1] to a transition to value-based payment systems [2], but this knowledge doesn’t come easy. Because of disparate data systems and the heavy administrative costs involved in detailed expense tracking, many practices end up having a less-than-ideal grasp of the expenses involved in different patient encounters.
The most common methods for determining cost often involve things like Cost-to-Charge ratios or RVU-based approximations [3]. While at a high-level these methods can provide some insight into the health of individual departments or service lines, the details are entirely nonexistent. This can lead to frustrations when trying to use them as a basis for business decisions made within a practice. To become a truly data-driven organization, this information is needed.
 

COSTING THROUGH ANALYTICS

Many industries have similar difficulties in determining the costs of their goods and services. By combining the improvements in analytical technologies over the past 10 years, other organizations are able to get closer to this information.
Machine Learning models have recently helped Boeing understand their detailed engineering costs [4]. By feeding a Random Forest [5] algorithm large quantities of training data, they were able to create a model that could predict future unknown costs. This model works by creating hundreds, or even thousands, of simple decision-trees that have been randomly split based on your training data. Individually they aren’t worth much, but together they can help find hidden patterns in the data.
In this example, Boeing researchers used this method to develop a model which could be fed 14 factors that potentially influence the cost of an engineering project (e.g., Team Experience, Project Duration, Estimated Complexity, etc.) and receive an output predicting the cost of that project.
Another example of utilizing modern analytics in cost analytics comes from construction management. In that industry, understanding how much a development will cost before throwing significant resources into it is paramount in maintaining the solvency of the firm. Several researchers in this industry have created Artificial Neural Network-based models aimed at determining accurate cost estimations [6].
Neural Networks are the driving force behind many of the improvements in Artificial Intelligence over the past few years. They work by connecting a series of nodes housing functions that send an activation signal to other nodes depending on the value that is sent to it. For those familiar with Neurology, this is somewhat similar to the series of neurons in the brain that send signals via dendrites, synapses, and axons. These days, everything from the facial recognition used to unlock your iPhone [7] to the syntax checking performed by Grammarly [8] is powered using these models.
Matel et al. describe their use in analyzing the project costs associated with engineering firms in construction management. They developed a network similar to the one used by Boeing in that it can take a series of inputs potentially related to the cost and then build a model that can predict future unknown costs. The model “learns” what the most important combination of inputs leading to changes in the final cost estimate are, then uses this information to help make better business decisions.
 

USING MODERN COSTING ANALYTICS IN HEALTHCARE

Translating engineering- and project management-based costing to healthcare is not as straightforward as it might seem. Obviously, the goal is slightly different—creating estimates for individual patient encounters vs complex singular projects—but the data itself is also significantly more complicated. Providers know that each patient is a unique case that requires different levels of physician care, administrative work, supplies, etc. To truly understand the cost of providing care to a patient, you need to have at least some data on each of these and combine them in a way that accurately represents the structure of the business.
However, there are benefits of using healthcare data for costing analyses not seen in other industries—mainly, that there is a lot of it. Between Electronic Health Record, Practice Management, Accounting, Customer Relationship Management, and Scheduling systems, each practice has millions of data points to help build these models. Additionally, these data are available for many different practices dealing with similar processes. If you are able to marry the different systems and practices together effectively, the accuracy of Machine Learning and Artificial Intelligence models built on them can increase dramatically.
 

WHAT WE’RE WORKING ON

DataDx is working to bring detailed cost-of-care analyses that work for providers of all sizes. With a foundation in both practice management excellence and state-of-the-art analytics, we provide unique insight to our clients on how to run a data-driven practice. By combining your disparate data systems into one cohesive reporting architecture, DataDx can provide immediate benefits to practices at any level of analytical maturity [9].
 
 

Jon Mattes is the Director of Analytics and Development for DataDx. Jon has a passion for putting valuable data in the hands of decision-makers. He has spent nearly a decade using statistics and artificial intelligence to improve health outcomes and operational effectiveness across provider organizations of all sizes. He works with both clients and development partners to ensure DataDx is always at the cutting-edge of analytics and can provide each practice with the unique insights it needs to succeed. Jon received his Master of Science in Data Science from Northwestern University.
Contact Jon at: jmattes@datadx.com

Ready to take
your practice
to the next level?

Speak with a DataDx expert today.

Comments are closed.