Predictive Analytics

New Gartner research forecasts that by 2016, 70 percent of the world’s most profitable companies will manage business processes by using real-time predictive analytics. Healthcare executives must integrate data mining now to be among the 70 percent.

Gartner isn’t the only organization boasting about the benefits of predictive modeling. Venerable medical organizations like the Journal of American Medicine, Journal of Epidemiology and American Heart Association have all reported that predictive modeling is a valuable clinical decision support tool.

Why Now?
Until recently, healthcare’s application of predictive modeling technologies was mostly limited to the payers’ side. Health insurance companies have been leveraging the value of their data for some time now to better anticipate cost and design, and control and avoidance measures.

Until 2010, healthcare providers were reluctant to delve into data like their payer counterparts because it would require extensive implementation of security protocols and processes to ensure patients’ privacy. Three words have deemed the additional protocols worth the effort: Affordable Care Act (ACA).

The ACA has enacted a pay-for-performance healthcare system, which rewards improved care, efficiency and technological integration. Several provisions within the act demonstrate this conclusion:

  • Hospital readmission provision — Under ACA guidelines, Medicare may withhold payments if/when a patient is readmitted to the hospital within 30 days of release. If hospitals don’t curb their readmissions, billions in revenue will be lost.

  • Meaningful Use standards — The Center for Medicare and Medicaid Services (CMS), by way of the ACA, have designed a set of incentive payments to providers based on the organization’s stage of electronic health record (EHR) integration. The belief is that EHR implementation will give physicians, patients and their families better access to health information to enable faster and more reliable diagnosis. Further, if by 2015 your practice had not met meaningful use standards, you will face drastic Medicare reimbursement withholdings.

  • Bundled payment structures — The ACA shifts from the traditional payment system where payers separately reimburse labs, hospitals and physicians for a patient’s care to a payment structure based on joint settlements to all agencies involved in the patient’s care plan per episode. This will require providers to more accurately predict their care costs to ensure proper and fair payment.

Increased reliance on technology and analytics will aid providers in their adherence of these three ACA adjustments. Predictive modeling technology combines both technology and analytics.

Predictive Modeling
Predictive modeling is a statistical methodology used to predict future outcomes. In marketing, for example, consumers’ gender, age and purchase history are predictors of their likelihood to buy a product in the future. In healthcare, a patient’s clinical history, geographic location and clinical setting are leading indicators of the patient’s likely reaction to treatment, and the probability that the patient will soon require additional treatment.

A predictive modeling toolset can be deployed in a healthcare agency to mine existing patient data, determine trends and predict outcomes the human eye may miss. Here are three examples of predictive modeling solutions that can assist providers in delivering better care, improving their bottom lines and adapting to reforms:

  • Readmission reduction — This predictive modeling solution is specific to hospitals’ post-acute partners. Since hospitals can be financially penalized when a patient a readmitted, it’s important home care and skilled nursing facilities also take advantage of predictive modeling. It combs through existing patient data to determine the top 10 percent of patients most at risk for readmission, alerts clinicians and informs them what risk factors lead to the patient’s at-risk classification. Medalogix’s readmission reduction solution has a telephonic function that further automates the process by calling at risk patients to check in on them and identify their state. Alternate Solutions HomeCare deployed this solution in 11 of their agencies and reduced their average rate of 30-day readmissions by nearly 36 percent. Their hospital partners have taken notice.

  • Patient transfer notification — Guesswork traditionally determines when a patient should be transferred from one care phase to another. That won’t cut it in an ACA healthcare landscape. For instance, prematurely transferring a patient from hospital to post-acute care could mean readmission, which translates to improper patient care and withheld reimbursements. Predictive modeling can mine your data to determine the statistically appropriate time to graduate patients to the next level of care. Other transfer models that may be appropriate are palliative and recovery.

  • Bundled payment financial risk management — In the bundled care model, if the cost of care is less than the bundled payment, providers retain the difference. Understanding exactly how much care costs will arm providers in the new bundled payment structure. An effective predictive model can determine clinical and financial risk to protect the quality of care while aggressively monitoring the financial impact. According to a study conducted by Singletrak Analytics and DataGen, when structured well, bundled payment models can be effective and profitable for hospitals that can reduce the cost of each episode of care while not compromising the quality of care. This list is in no way comprehensive. Predictive modeling technologies can be leveraged with a high degree of accuracy on most datasets and in almost any healthcare setting. The technology is versatile and scalable. There are a few caveats in navigating available predictive modeling providers. Not all predictive modeling services are created equally:

  • Specific versus broad — With predictive modeling all the buzz, the term is being thrown around loosely lately. Many software providers claim to be predictive but what they’re actually offering is simple state and nationwide benchmark comparisons that are not agency or patient specific. This is an extremely important consideration. Let’s say your agency typically has excellent outcomes when treating patients with congenital heart failure (CHF). CHF is typically indicative of high readmission risk. If you deploy a “predictive modeling” toolset that measures against statewide or national data, a CHF patient will undoubtedly qualify in the high-risk category for readmission. While that is certainly true, this simplistic qualification neglects an extremely important element – specificity. Your CHF patient may not be at as high a risk of readmission because your agency has a long track record of success with this condition. Your predictive model should account for this and accurately identify that, unlike other agencies, a CHF patient at your agency is less at risk of readmission than a pneumonia patient, for example. When making the extremely important resource deployment decisions that healthcare providers have to make everyday, it is critical to know not just that a CHF patient is a high risk patient, but how high risk as compared to the rest of the patients under your care. It is this better targeting of resources and interventions that is the key to delivering better care more efficiently.

  • Analysis paralysis — Although data is key in providing optimal care, too much data can be worthless. Some predictive modeling providers boast hundreds of predictions and reports. Clinicians don’t have time to sort through it all to make the best decisions. The best predictive modeling offerings arrange predictions in a straightforward and tangible format. Quality healthcare will always be rooted in the hands of highly skilled caregivers, but moving forward, caregivers’ decisions will be more reliant on data analysis. Whether deploying independently or accessing these tools through your EHR, predictive modeling is a preventive rather than a reactive solution to patient care, which is crucial in America’s evolving healthcare system.

Dan Hogan is the founder and CEO of Medalogix, a Nashville-based health care technology company offering data mining solutions to predict future outcomes in patients to better target clinical resources. A former home health owner, Hogan understands the unique challenges facing healthcare.

About The Author