In healthcare, actionable information provided as early as possible can dramatically improve patient outcomes. Leidos and the University of Miami Health System are working together on a predictive analytics infrastructure that delivers data to physicians where and when they need it, enabling them to make informed decisions at the point of care.
In 2014, Leidos began working with Dr. David M. Seo, MD, Chief Information Officer and Associate Vice President, University of Miami Health System and Miller School of Medicine, Chief Medical Informatics Officer, Chief Research Information Officer, Miller School of Medicine. Dr. Seo had long realized the value of a healthcare information system that could provide actionable information based on extensive data analysis.
“This information should not just be about the patient's past,” he said. “We need a data environment that can do complex statistical analysis to help us move away from reactive medicine and toward proactive medicine, in which we get to patients before they get sick and prevent the disease from occurring."
Leidos and the University of Miami began building a framework for the processing and delivery of predictive analytics insights for health professionals. Leidos designed the platform to support interchangeable predictive analytics models that will help grow the hospital's capabilities in various medical areas.
One thing that we realized early on was that busy physicians using this system would have to see the information when they needed it (during patient consultations) and where they needed it (in their medical records).
Leidos integrated the predictive analytics system directly into the hospital’s electronic health record system, (Epic) that physicians at the University of Miami use every day. This ensures that healthcare workers do not have to learn a new system to get the medical recommendations that they need. Instead, the system delivers its information as part of the physician’s workflow, as close to the point of care as possible.
Predictive analytics in action
In a typical medical interaction, a nurse reads and enters a patient's details into the electronic medical record software before the patient sees the doctor. The doctor will then review the data and issue an order for treatment into the same system. This, in turn, contacts the Leidos predictive analytics software with a request to run a risk score for the patient, which processes the data and sends a message back to the electronic medical record software. The doctor sees the system's recommendations on the same screen and can take immediate action during the consultation.
The predictive analytics system can also trigger on a time basis, enabling doctors to see recommendations each month based on a watch list of high-risk people. It can also update doctors based on events such as another specialist updating a patient's vital signs.
To deliver predictive analytics results directly at the point of care, we connected the electronic medical records system to open source integration software. This communicates with the various medical systems needed to support application delivery using Health Level Seven's HL7 messaging standard.
A platform for many models
Leidos built an infrastructure and data architecture that would allow physicians at the University of Miami to run a variety of statistical models when determining a patient's risk level. These models take various inputs from the patients' medical record and determine their correlation to an output variable called the risk quotient.
The team initially used the diabetes self-assessment scoring model developed by Dr. Heejung Bang, Ph.D. (the 'Bang model'). This model uses six variables to determine patient risk. This relatively small number made it a useful model when developing and refining the underlying data flows and architecture for the predictive analytics system.
The modeling process is complex, with combinations of different models used in concert to create highly targeted recommendations. While some models may directly address specific illnesses, others may cover other issues such as readmission risk. Different medical institutions may use models lending themselves to populations with a specific demographic, or to serve well-defined regulatory or financial goals.
Modeling positive medical outcomes
To help the Leidos team manipulate these models, the predictive analytics needed access to large volumes of data from a range of applications contributing to patient healthcare records, along with public data sets.
For model research and creation, the system de-identifies patient information for privacy purposes and feeds it into a data lake – a large collection of diverse data sets – that we can use to mine that information. We used a statistical modeling language called R to scan through that data, building clusters of information for use as the basis for risk models.
Once Leidos created these models, we can export them using a vendor-independent model interchange format and re-identify the underlying data with personally-identifiable information to use the models in patient consultations.
Leidos imports models into an open source engine that it integrated into its system. This creates a library of models called the OpenScoring Dashboard that physicians and medical researchers can call upon for insights into patient health risks.
While Leidos’ skilled research team is building an initial portfolio of models, the researchers at the University of Miami are learning to utilize the platform to develop revolutionary models for enhancing clinical care in precision medicine, minimizing readmissions, and improving patient outcomes.
Working closely with medical staff
The Leidos design and deployment team faced some challenges when creating a predictive analytics system for a busy hospital environment securing access to physicians' time. Doctors are constantly time-constrained, so we had to use alliances with key medical staff to help persuade more doctors to participate in the process and made efficient use of the time they made available.
The predictive analytics project is still in the pilot stage, and Leidos and the University of Miami are still assessing its results. The ideal outcome would see wide acceptance from physicians, and a more effective mechanism for detecting medical risks, along with improved quality of care. The original business case shows a potential cost saving of up to $12.5 million as pre-diabetic patients are successfully identified and put through diabetes prevention training.
Promising initial results from the pilot project are encouraging Leidos and the University of Miami to explore further opportunities for the predictive analytics system.
One of these opportunities involves introducing new models. The implementation team is already exploring congestive heart failure models and is considering modeling certain cancers, too.
The team is also hoping to process more record types in the predictive analytics system, using natural language processing to digest information from clinical notes and structured technical data including medical imaging and reports. It may also be able to include genomic data from medical laboratories in its predictive analytics process.
Finally, the team will refine the system to make medical care not only proactive but precise. Precision medicine uses genotype information to create treatments precisely tailored to suit individual patients, which drives up positive outcomes. As the system consumes more data, researchers will create predictive analytics models that target individual patient needs more accurately.
Bringing actionable insights to physicians at the point of care can help them to have conversations with patients that lead to real-world benefits and dramatically increase their long-term quality of life. Leidos and the University of Miami will continue to assess the results in this area and refine a platform that will ultimately enable physicians to interact with complex statistical models in meaningful ways.
Our quality health solutions in predictive analytics and big data, lifecycle management, business intelligence, investigation tracking, enterprise search and data management help improve performance and expand capabilities, thus maximizing value for its users. For more information, you can visit our website or contact us to learn how we’re advancing the next generation of care.