​

Skilled nursing is facing a seemingly insurmountable challenge. Just as the rapidly expanding aging population is beginning to need skilled care, many experienced caregivers are retiring, and a strong economy and tighter immigration controls are shrinking the available pool of candidates to fill their vacancies.

Human resources professionals are focused on better methods for recruiting and training caregivers, while business owners are considering how to increase wages and still maintain profitability. While useful, these solutions are not adequate for addressing the problem. It is a simple case of unbalanced economics: Tomorrow’s demand is greater than today’s supply.

What’s more—even though the “supply” is shrinking, the “demands” are increasing. As the Patient-Driven Payment Model (PDPM) looms on the horizon, skilled nursing centers will need to work even more efficiently to provide high-quality care and avoid readmission penalties.

But before declaring the situation hopeless, know that there is a way to help existing caregivers “work smarter” so they can achieve optimal outcomes despite staffing shortages. John Damgaard, president and chief executive officer of MatrixCare, explains how:

“By employing machine learning to assess the millions of data points that flow within an electronic health record (EHR) system, MatrixCare is developing software that dramatically increases ‘caregiver leverage’…that is, the number of patients or residents that a single caregiver can effectively manage without compromising high-quality care.”

State guidelines may dictate a minimum number of caregivers per a certain number of residents, depending on age and acuity. However, this is not particularly helpful for an organization trying to work more efficiently under value-based care models. Today, for example, the state-of-the-art in staff scheduling software leverages acuity-based staffing tools that allow prediction of how many caregivers with a certain level of experience are needed to deliver care to residents, based on the case mix of that community. 

These systems use historical data and best practices gathered over decades of care to guide staffing recommendations.

Although this approach can be useful, its usefulness is likely to decline as case mixes change under PDPM. In addition, Damgaard stresses that it does not fulfill the promise of what can be achieved with artificial intelligence (AI) and machine learning.

Computer science academics define machine learning as the scientific study of algorithms and statistical models that computer systems use to perform specific tasks without the need for explicit instructions, relying on models and inference instead. This process is considered a subset of AI.

“These tools can exponentially increase the value of technology-assisted scheduling by factoring in hundreds of additional demographic, clinical, and social data points,” he says. This includes everything from vital signs and movement derived from telehealth devices, to social interactions, diet/nutrition, isolation, and more—to provide a much more comprehensive picture of a patient’s overall health and wellness, he says. Here are a couple of notable examples.

Rather than, “Where can we hire more staff?” the more interesting question is: “When Caregiver Jones comes on shift, where is the first place she should be?” Most organizations schedule their caregivers according to geography (Nurse Garcia has the south wing, Nurse Brown has the east wing) or by function (Nurse Lee dispenses medications, certified nurse assistant [CNA] Schmidt helps with dressing).

Neither of these approaches takes into consideration patient need, which in the senior population can shift dramatically day-to-day, hour-to-hour, minute-to-minute. As Damgaard suggests, organizing a caregiver’s day based on a dynamic ranking of the patients’ immediate needs is a much better approach for ensuring that quality outcomes are achieved.

Consider Mrs. Monroe, a senior who has been identified at risk for falls. Machine learning can learn the patterns of a patient, so if she has a pattern of using the bathroom every day after breakfast—and requires assistance because she becomes unsteady—the software could alert the staff to assist her immediately after breakfast, thereby preventing a fall.

The algorithms must also include “aging criteria” so that less urgent needs surface to the top if they are unattended for too long. The goal of dynamically ranking the population in a “most-need-to-see” order is to improve outcomes across the entire population while maximizing operational efficiency.

This method can help increase satisfaction in both those who are being cared for as well as caregivers, because each interaction adds value. At any point in time and space, the software should put the staff member where he or she can do the most good.  

Taking advantage of the huge increases in the capability of top-tier cloud-computing environments to manage big data environments and the maturation of machine-learning toolsets, innovative leaders are beginning to realize the incredible potential of EHRs to vastly improve the efficiency and quality of care.

A range of new capabilities makes use of longitudinal personal health records (PHRs), augmented with Internet-of-Things (IoT) data from things like scales, continuous positive airway pressure (CPAP) machines, and glucometers, as well as personal device data. These data are inclusive of social, environmental, financial, behavioral, and genetic factors, Damgaard says. These advances can play an important part in creating additional “caregiver leverage.”

On this person-centered foundation, leading technology innovators are leveraging the targeted application of deep-machine learning on both structured and unstructured data in the comprehensive PHR to identify early indicators for changes in condition and opportunities for intervention.

“In one example of this targeted application of deep-machine learning, a neural network is trained with millions of free-text progress notes and subsequent incidents of falls to identify key phrases/combinations of phrases that predict fall events with a very high confidence level—90 percent or more,” says Damgaard. These neural networks operate like the human brain but with the added benefit of being able to quickly process vastly more information than even the most experienced clinician.

As subsequent progress notes are entered, this trained network (powered by the Microsoft Azure cloud) returns an updated fall-risk score each time.

Should the risk score exceed the documented threshold or change by a certain magnitude, the intelligent EHR can then use this information to pre-emptively suggest invoking the Johns Hopkins falls management protocols* to lower fall risk, says Damgaard.

“In this way, a single, less-experienced caregiver is ‘coached’ to provide care at a level that previously would have required a whole team of more experienced caregivers,” he says. “This allows the organization to continue to provide high-quality outcomes even as more experienced staff retire or are otherwise unavailable.”

Skilled nursing operators that want to succeed under PDPM—despite the imbalance between the demand for and the supply of labor—will need to look toward technology as a tool to achieve their goals.

“Innovations like this radically bend the cost curve, while preserving—or even enhancing—care outcomes,” says Damgaard. “And because every interaction is meaningful, they also have a material positive impact on patient and caregiver satisfaction.”