There is no shortage of opinions, predictions and questions about the impact of data science and machine learning on health care. Excitement and speculation about the power of data and analytics to transform health care is centered on the premise that advances will allow computers to perform both mundane (e.g., documentation) and highly complex (e.g., predictive analytics) tasks so that clinicians have more time to spend with patients.
The potential for data to transform and improve care is indeed exciting, but by the same turn can become overwhelming without a plan for turning data into knowledge, and knowledge into action. Hospitals and health systems have an overabundance of data, with more than a decade of documentation and patient data housed in the electronic health record (EHR). For health systems to fully realize the potential and power of their EHR, and see a return on investment, leaders — from the C-Suite to physician end users — must not only derive intelligence from these large stores of data, but translate that intelligence into prescriptive action that can deliver better outcomes for patients, help manage the health of populations and improve results for their organizations.
At Children’s Hospital of Orange County (CHOC) we look to our data to keep children well, and to keep those that are ill as well as they can be. Hospital readmissions interfere with our ability to do this, not to mention create potentially disastrous patient outcomes and a staggering price tag of $41.3 billion in total hospital costs across the U.S. health care industry when patients are readmitted within 30 days, according to the Agency for Healthcare Research and Quality. At CHOC Children’s, we wanted to develop a way to understand which patients were at highest risk for readmission at seven and 30 days so we could target appropriate interventions. A key to this process has been engaging a data scientist to look at our historical, longitudinal data and develop an algorithm that can help predict which patients are most at risk for readmission. Using advanced data science tools, we’re able to use data from thousands of patients to predict an outcome for a single patient. As clinicians, we know what’s in the best interest of our patients. Working side-by-side with a data scientist, however, we’ve bridged the gap in turning the data in our system into actions that improve outcomes.
Utilizing data to augment clinical decision-making represents the real potential of data science and machine learning to transform health care. After years of dutifully documenting in the EHR, our clinicians are eager to see higher-level value emerge out of the system. Advances in medical research will continue to produce an explosion of data that will have the potential to inform care and will require tools and expertise to make it usable by clinicians. For health systems to keep pace and reap the benefits of this new wave of data, engaging the skills of a data science expert can help to incorporate new information that can inform better care protocols and outcomes.
Another consideration is how and where data will be stored. Cloud-based storage and computing allows us to have access to larger data sets and apply more sophisticated variables to finely tune our algorithms. When we first began applying intelligence to our data, the largest study had 98 variables and spanned 37,000 patients. Now, on our cloud-based model, we currently have a data set of 384 variables covering 13.9 million patients. With cloud-based resources, we utilize the elasticity in computing ability to compute faster and include more variables, leading to more accurate algorithms.
Incorporating a data science strategy into a hospital or health system is an investment, but one that can yield positive patient outcomes and financial returns, especially in a new era of value-based care. Our readmission algorithm, for example, has given us the ability to not only predict which patients may need intervention, but also has helped to reduce the number that do get readmitted, a key metric in new value-based reimbursement frameworks.
While we’re early in our implementation journey on readmissions, working in this super-charged environment will help us develop other algorithms much quicker and easier. For instance, we’re working to refine a patient deterioration algorithm that will alert us earlier to patients who may need to be transferred to the ICU. The earlier a patient gets transferred to the ICU, the less likely they are to need extra resources like ventilation and medications to support the cardiovascular system. The algorithm will allow us to intervene faster for patients that are deteriorating and need intensive care while making best use of our resources. Like the readmissions algorithm, this will help us provide better care while limiting costly procedures and longer stays.
The future of data science lies not in replacing clinicians — an algorithm can’t do a physical exam or talk to a human in an effective way or capture emotions or provide support — but in augmenting physicians with information and tools that allow them to make more informed decisions and provide an enhanced, refined level of care. Health care organizations can begin by organizing the right tools and resources to help make the most out of their data.