Data Science Can Predict Patient Outcomes to Optimize Care

Article

Data science can help to predict patient outcomes for areas such as medication utilization and readmission rates to hospitals.

Data science can be effectively used to predict patient outcomes for areas such as medication utilization and readmission rates to hospitals, according to a session on applying data science at the DHX Virtual conference. These predictions can help the users of data science tools to support the delivery of optimal patient care.

Data science uses the scientific method, processes, algorithms, and systems to extract applicable knowledge from structural and unstructured data, explained speaker Dalton Fabian, PharmD, RPh, a data science analyst at UnityPoint Health, during the session.

The field is comprised of different disciplines in order to tap into their skillsets, including math and statistics, business knowledge, and computer science, according to Fabian. Each of these disciplines then fits in uniquely to the function of data science in pharmacy.

For the use of math and statistics, data science analysts apply these skills when using data from the past to predict the future, such as the likelihood of whether someone will be readmitted to the hospital within 30 days of discharge.

Business knowledge, on the other hand, is used by data science analysts to communicate with nurses regarding the tools that the analysts have made. This communication helps analysts confirm that the nurses have a full understanding of the tools they are using, according to Fabian.

“As pharmacists, we understand the intricacies of health care, like how to read lab descriptions and how to interpret lab values using those types of skills in our data work,” Fabian explained.

Additionally, Fabian noted that data scientists use computer science skills when writing sequel queries to obtain or analyze data implementing models and programming languages, such as R and Python, as well as visualizing data with platforms such as tableau.

David Berkowitz, PharmD, an Omnicell One field strategist and director of the Omnicell fellowships in data analytics and automation research, said during the session that obtaining data is always the first step to commencing work as a data science analyst.

“First, you need data,” Berkowitz said. “The standard nomenclature is ETL—extract, transform, and load. And that’s getting data from your source system, whether it would be an electronic health record, laboratory system, IOT device, blood pressure cuff, etc.”

The next step is to transform these data into a format that is usable, such as Epic, which is written using Massachusetts General Hospital Utility Multi-Programming System (MUMPS). MUMPS, a programming code that was developed in the late 1960s, is a language that is effective for fast operations, which allows users to click a lab result and find the necessary information quickly.

“That’s why when you build, say, a workbench report, it’ll sort of crap out after looking back 14 days, because that type of search is not meant for querying, doing research, and doing data analysis—it’s meant for operations,” Berkowitz said.

By moving the data and encoding it into a different format, which is usually structured query language (SQL) in health care, data science analysts are able to store the data. The databases available to for storage are generally SQL, No-SQL, and cloud storage; however, not all providers use these databases when storing data.

“There are still a lot of folks in health care, probably actually a majority, that store their data on premises, or you’ll hear ‘on prem,’ which can make it a little harder to do data science because you have to sort of, when doing a specific analysis, enrich data with different sources,” Berkowitz said. “It’s harder to do that on prem in real time because of connections. So, cloud first is my motto.”

At Omnicell, Berkowitz noted that his team uses almost exclusively SQL. However, other employees at Omnicell will code using Python, R, or Scala, although the language used is often dependent on the specific problem the analyst is trying to solve.

For data visualization, analysts commonly use Qlik, Power-BI, Tableau, or R-Studio in order to communicate data more effectively and support business intelligence and predictions, Berkowitz explained.

At UnityPoint Health, Fabian noted that they have used these different elements of ETL to create a tool called Population Health, which is primarily used by care managers to enroll patients in care management services. This tool can help to lower costs for high-risk patients by decreasing utilization and increasing the frequency of patients’ meetings with care managers. This helps to catch issues before they become critical and require a hospital visit.

“What we did, as a data science team, is develop a machine learning model that predicts the likelihood that a patient is going to go to the hospital or the ED in the next 12 months, and that’s how we determine high risk,” Fabian said.

Fabian noted that this tool helps to allow care managers to spend more time taking care of patients rather than needing to focus on administrative tasks that they were doing previously to figure out whom to enroll.

Another tool created by UnityPoint Health is the Readmission Heat Map, which is used to decrease 30-day admission rates. This tool allows for the prediction of whether patients will readmit to the hospital within 30 days of discharge, down to the day.

“In this tool, we’re taking all of the data we have access to in Epic, like disease states, medication, and the fact that people are in the hospital, and pairing that with a predictive model to help those nurses make better decisions about the care that a patient is going to need once they get out of the hospital,” Fabian said. “Anything over a 20% likelihood, and nurses will take heed and intervene on behalf of those patients.”

Berkowitz noted that by applying these same types of data science tools to medication utilization, analysts are able to help pharmacists visualize data from their pharmacy in a chart. For example, this could be applied to visualizing the effects of the pandemic on medication utilization at a pharmacy.

“I took a look at vaccination rates since January 2020 to see how COVID-19 has affected vaccination rates,” Berkowitz said. “Basically, it went down to nothing. This kind of tells me that either folks weren’t stepping on nails or they weren’t getting care and going to their pharmacies.”

REFERENCE

Fabian D, Berkowitz D. Applying Data Science. Presented at: DHX Virtual 2020; October 2, 2020; virtual. cpha.com/event/dhx-virtual/. Accessed October 2, 2020.

Recent Videos
Related Content
© 2024 MJH Life Sciences

All rights reserved.