ing the data that appear on the all-important Centers for Medicare and Medicaid Services Nursing Home Compare website,
where facilities are rated by the number of stars they receive.
Taylor says that’s the most important metric right now, although there’s long been a debate around whether that metric
actually does a good job of reflecting the quality of a facility.
Predictive analytics have become good at predicting readmission rates for patients with joint replacements, while factors
surrounding patients with kidney injuries and dialysis remain
It’s in this setting, he says, where the race is on to develop tools
that can be used at a patient’s bedside. Taylor uses data analyt-
ics to develop screening tools that help predict a patient’s fall
risk, how likely a patient is to be readmitted to the hospital, or
how likely they are to return to skilled nursing after being dis-
charged. His goal is to create tools that integrate seamlessly
into the routine tasks a clinician is already performing, such as
summarizing patient status at shift change meetings, rounding
sessions, or when a patient is admitted to the facility. In Sym-
phony’s long-term care facilities, most charting is done on iPads
and iPhones, so Taylor and his team try to develop alerts that,
for example, warn the provider that the patient is at a higher fall
risk due to a combination of medications that they are entering
into the chart.
Taylor says he accounts for providers’ tendency toward alert
fatigue, so the tools are designed around the concept of “parsi-
mony,” which means, “What are the fewest number of variables
I can provide that give the most amount of information?” Taylor
says. Then, when nurses gather for shift change meetings, the
tool prompts them to share the most pressing details about each
patient to the nurses on the next shift.
These types of tools only scratch the surface of what’s possi-
ble—or may be possible in the future, Taylor says. For instance,
there’s research that suggests machine learning could help ana-
lyze photos of pressure ulcers and classify the severity and the
risks that accompany these wounds. “We just don’t have the ca-
pacity to analyze that. By no means are we close to analyzing
everything we could do [with analytics],” Taylor says.
‘Intelligent Dashboard’ Helps Long-Term Care Providers Assess Patient Population
Data in a Glance
THIS TABLE IS what Nathan Patrick Taylor, MS, MPH, CHDA, CPHIMS, director of data science and analytics at Symphony
Health, refers to as an “intelligent dashboard.” It reflects each patient that is “in house” at the Symphony Post Acute Network
facility in Joliet, IL. It contains demographic details, room numbers, primary provider, and a few standardized assessment
scores. It also displays readmission risk probability, and a readmission risk category. The nurses use this report in their shift
change meetings every morning discussing the current status of each patient.
The readmission risk is determined using a machine learning algorithm. The algorithm takes into consideration numerous
factors based on information collected upon admission and throughout a patient’s stay. Those factors include demographic
and social data, and clinical data, such as diagnosis codes, procedure codes, laboratory results, and medication lists. The
machine learning algorithm also parses free text from progress notes. After the machine learning algorithm is trained, it is then
deployed to a data warehouse and updates the predictions every night. The readmission report is then distributed every morning through a secure electronic interface.
The average patient at one of Symphony’s facilities is typically there for seven to 14 days, receiving rehabilitation services
for joint replacement surgery and other rehabilitation. There is also space to provide dialysis for severe kidney failure patients
and severe mental illness, although those are a very small portion of the patient population.