As a consultant, I am fortunate enough to spend time in a variety of different industries, I have spent time in the healthcare industry and in that period, I was tasked with creating an application that would assess patient care based on CMS (Centers for Medicare and Medicaid Services) ACO (Accountable Care Organizations) guidelines. The tool that was built pulled all patient information together from all the electronic health records of a large healthcare organization. It provided a clear view of that patient’s health history in relation to the attainment of several ACO quality measures. The tool was created not only for the healthcare team to use as an auditing system, but more importantly as reference to quickly visualize passing and failing measures for their respective patients. This allows them to take steps immediately to rectify any care gaps. One measure of the success of the project was the quick adoption and positive acceptance of the tool by providers. This preventative tool not only works to improve patient outcomes, but it generates a lot of additional revenue for the healthcare organization in the form of performance based payouts from CMS. It also eliminated the need to pay a large amount of money to a second party company to compile less complete data in a retrospective manner that did not involve the provider in any direct way.
While the product of that effort was fantastic, it still has untapped potential. I mean it was great that at a quick glance a physician could quickly identify areas of concern that needed to be addressed before even stepping foot into the patient’s room. It was also great that this tool allowed for the organization to maximize their incentive based payout by improving their performance in these measures.
What I mean by it still had untapped potential is this. Having all this new data presented to them in this fashion still only allowed for reactive patient care. What if we could take this a step further and provide proactive care by predicting what the future may hold for a patient?
I was at the SQL Pass Summit back in 2015 when I was first introduced how this dream could become a reality. The keynote was about a pilot program called ImagineCare by an academic health system on the east coast Dartmouth-Hitchcock. This program was a new and exciting way to address population health management. The approach they took was to collect a whole bunch of data from home healthcare devices and store it in the azure cloud where they could run machine learning algorithms to try and predict events in real time. This data was then used to alert the patient via text message or a phone call from a Registered Nurse to intervene.
Fast forward to today and we now have SQL Server 2016 which has taken a giant step forward in allowing for predictive models to be built on premises without having to move any data into the azure cloud. What they have done is introduce SQL Server R Services. R is an open source programming language for statistical computing. In SQL 2017 they have also introduced the ability to use machine learning R packages in SQL Server itself.
When I think about the application that was built and how predictive analytics could enhance it I am drawn to one measure. There is a measure that tracks all patients that were hospitalized and readmitted to a hospital within 30 days of discharge. Readmissions are a significant concern for hospitals, as they will receive cuts to their incentive payments for high readmission rates.
Readmission rates could be reduced by properly identifying who is most at risk and providing those individuals with extra attention. We already know that readmission rates follow certain trends. One study found that readmission rates were particularly high for certain admission diagnoses such as heart failure (24.8%), acute myocardial infarction (19.9%), and pneumonia (18.3%). With the power of R and SQL Server we can now build predictive models that will provide organizations with an individualized assessment of readmission risk per patient based on their own unique trends. These models could easily be expanded to look beyond hospital diagnoses. They could look for correlation based on any number of data points that may be statistically relevant. Think of things such as age, socioeconomic status, attending provider, health literacy status, or even pain level at discharge, and how these unique variables may impact the readmission rate. This information can become very valuable in determining patient follow up plan, prioritizing educational needs, and ensuring a safety net is in place for these high-risk patients. Having predictive analytics technology in place not only saves money for healthcare organizations, but also has the real potential to improve quality of life and reduce mortality related to avoidable complications.
Another way that predictive analytics can be beneficial to health care is to aid in reducing costs.
It is estimated that in the U.S. alone $210 billion dollars a year is wasted on unnecessary care and testing. There is a great article about this issue in the New Yorker. We can use analytics to offer recommendations on the appropriate tests or care to provide. Many lab tests can be very expensive and may only apply to certain demographics or research shows them to be very inefficient. We can build models to predict the cost/benefit of these tests.
It is estimated that in the U.S. alone $55 billion dollars a year is wasted due to missed prevention opportunities. The example above of readmissions would help drive down this high number, being able to predict readmission possibility and altering the care plan may prevent a future emergency room visit before being readmitted to the hospital.
It is estimated that in the U.S. alone $130 billion dollars a year is wasted because services were delivered inefficiently. CMS has a measure for screening a patient for fall risks and normally this is done by the nurse answering questions about the patient to give them a score, this score may alter the patients care plan. Now we can build models to accurately predict what the chances are that a patient may fall while in the hospital based on certain criteria such as age, procedure, current medications, etc… This is great because nurses are currently over worked and can free them up for more pressing patient care.
I am very excited for where this technology is going and how it can improve the future of healthcare.