One of the most important improvements in revenue cycle management (RCM) in the past 20 years has been the development of rules-based engines.
This technology has been a boon to hospital and health system revenue because it replaces manual, “every Nth” claim inspection with an automated system that can check every claim to ensure that everything that should be charged is being charged.
That said, rules-based engines have a severe limitation: they can only check for charges that are already known.
The problem with this approach is that healthcare is a very dynamic field. Unlike industries such as manufacturing or retail, healthcare has a tremendous amount variation – between hospitals, between providers, and between patients. Despite efforts at standardization of care, it will never be exactly the same from instance to instance.
There will always be X factors – things that vary between episodes of care – which means it is simply impossible to create enough rules to cover every possibility. And even if you could do so at the start, ongoing changes such as a health system acquiring a new hospital, shifts in a hospital’s demographics, or even the addition of a new catheter lab, would soon make the old rules obsolete.
The result is a great deal of revenue slips through the cracks as you fall victim to the problem of “you don’t know what you don’t know.”
Predictive modeling solves for X
This is the issue predictive modeling has been created to answer. Rather than requiring you to know all the potential issues up-front, it uses machine learning algorithms to parse through historical big data from multiple sources (clinical, claims, financial, operational, etc.) to extract patterns and trends.
Once you identify these patterns, you can build highly accurate models that predict the charges that should be associated with a particular encounter with a specific physician. When the model is compared to the actual charges, any discrepancies can be identified immediately and escalated for resolution.
For example, in orthopedics you’ll often see that Dr. Jones tends to use a certain set of devices. When examining an episode of care involving Dr. Jones the model will already have predicted the probability that those devices were used. If they are not included, the technology will flag it for review.
This same technology can also be applied to other, non-clinical areas, such as predicting the likelihood that a self-pay patient, or a patient with a high deductible, will pay his or her portion of the bill. Using claims, financial, socioeconomic, and other forms of data, you can identify those at the greatest risk of non-payment so arrangements can be made at the time of care in order to reduce bad debt and other issues.
Constant, automatic refinement
One of the most important aspects to predictive modeling is how quickly you can incorporate it into your financial operations. Unlike rules-based technology, where everything must be decided up-front, predictive modeling is designed to learn and add to its knowledge base as it is used, adjusting for changes in previously-identified patterns as well as adding new models as patterns are identified. In other words, if Dr. Jones stops using one device and starts using another, the machine learning technology will recognize it and make the appropriate adjustments to the models that are affected. If Dr. Jones starts doing something new, the algorithm will identify that as well as this information becomes available in the datasets that are modeled.
Machine learning can help make refinements to the financial models based on factors such as where an encounter occurred. Charge patterns will differ whether the encounter was an outpatient visit to a laboratory, acute care center, or primary care physician versus an emergency department event or an in-patient hospital stay. It will also constantly review model-building factors such as admit and discharge dates, diagnosis and procedure coding, admit sources, and the charges and reimbursement costs to deliver continuous improvement to the models to ensure the organization is always capturing all the revenue that is due.
One other way machine learning helps is by normalizing the data or similarly, optimizing the groupings of similar data points, which helps with varying products that address the same clinical issue. When the algorithms are built, they may not recognize that ibuprofen and acetaminophen are both over-the-counter anti-inflammatories. If the patient is given ibuprofen but the model calls for acetaminophen, it could say that acetaminophen is missing from the charge document.
By optimizing these groupings, however, this issue is avoided, and the model “learns” that the presence of either is acceptable for this type of encounter.
Time to break the rules
Now more than ever it’s important for hospitals and health systems to capture all the revenue they’re due. Making that happen, however, requires the ability to build better models even when all the variables aren’t known. Predictive modeling helps healthcare organizations do a better job of ensuring that X=revenue.
Author Bio: Paul Bradley, PhD, is the Chief Data Scientist at ZirMed, a recognized leader in cloud-based revenue cycle software and predictive analytics. He is an in-demand expert in the field of predictive analytics. In his role at ZirMed, Dr. Bradley oversees the research and development of new processes and technologies, including predictive analytics and data mining.
The views, opinions and positions expressed within these guest posts are those of the author alone and do not represent those of Becker's Hospital Review/Becker's Healthcare. The accuracy, completeness and validity of any statements made within this article are not guaranteed. We accept no liability for any errors, omissions or representations. The copyright of this content belongs to the author and any liability with regards to infringement of intellectual property rights remains with them.