By: Jeff Deal, Program Chair, Predictive Analytics World Healthcare
Sept 27-Oct 1, 2015, we asked Daniel Chertok, PhD, Sr. Data Scientist at NorthShore University HealthSystem, a few questions about incorporating predictive analytics into healthcare. Catch a glimpse of his presentation, Predictive Analytics Applications to Populations Health Management and Staffing Optimization, and see what’s in store for the second annual PAW Healthcare conference in Boston.
Q: In your work with predictive analytics, what area of healthcare are you focused on?
A: The objectives of my work include improving patient outcomes, increasing operational efficiency and managing costs. Examples of these efforts include assessing patient mortality, admission and readmission risks, optimizing nurse staffing in the Emergency Department (ED), and making cost containment recommendations based on comparing procedure costs by provider.
Q: What outcomes do your models predict?
A: We have successfully predicted relative probabilities of mortality, admission and readmission risks for patients with congestive heart failure (CHF) and chronic obstructive pulmonary disorder (COPD), admission risks for patients with diabetes mellitus (DM) and for the general patient population. By “relative probability” I mean patient ranking with respect to the corresponding risk: while the “absolute” probability may not be “correct,” patients appearing at the top of list are much more likely to have a negative outcome than those below them.
Q: How does predictive analytics deliver value at your organization? What is one specific way in which it actively drives decisions or impacts operations?
A: The Clinical Analytics team at NorthShore University HealthSystem is fortunate to be part of concerted effort by the top management to use the wealth of available clinical data for the mutual benefit of both the patients and the organization. Our mission is to streamline the efforts of clinicians, case managers and operational leaders by providing them with predictive and reporting tools for efficient decision-making. One recent example is MyPanel, a Tableau-based tool hosted on a server accessible from individual physicians’ offices, which aggregates patient data for the providers on a menu of visually rich dashboards; its current utilization rate exceeds 96%. Another example is the ED utilization dashboard showing the most recently available ED demand data superimposed over historical aggregates; it allows ED staff managers to make tactical decisions about resource deployment.
Q: Can you describe a successful result, such as the predictive lift of your model or the ROI of an analytics initiative?
A: A case management program developed by the Clinical Analytics team based on the Elixhauser comorbidity method yielded a lift in excess of 40x. To put it in perspective, a case manager reaching out to those NorthShore patients who are most at risk for hospitalization due to CHF, COPD, coronary arterial disease (CAD) and DM is able to reduce the number of patients in her cohort from 1000 to 25 while achieving the same result. While it is very hard estimate actual savings achieved as a result of using our models in the context of managed care, a very rough estimate would place that number potentially in the hundreds of thousands of dollars.
Q: What surprising discovery have you unearthed in your data?
A: In the course of analyzing ED demand during different times of the year, it turned out that the weekend following Labor Day had a higher utilization rate than the actual holiday weekend. I did not realize that a hangover could last that long…
Q: What areas of healthcare do you think have seen the greatest advances or ROI from the use of predictive analytics?
A: Speaking in monetary terms, healthcare organizations can undoubtedly benefit from applying machine learning techniques to billing. One of our team members is currently working on identifying billing errors using a rule-based algorithm and has already achieved a remarkably high true positive rate. From the care point of view, we started to think about collecting bedside sensor data in order to alert staff to potentially critical changes in the patient’s condition. We could even potentially pair up with aggregators of wearable device data with the view of monitoring our patient’s health in real time, but for now it is still in the distant future.
Q: Sneak preview: Please tell us a take-away that you will provide during your talk at Predictive Analytics World.
A: I will be covering our accomplishments in developing ED utilization and staffing model. My hope is to show the attendees that staffing optimizing in the ED department can be achieved through accurate data collection and relatively straightforward modeling tools. Improving ED patient outcomes and experience is a realistic high impact goal that can be accomplished internally with relatively modest resources. The main ingredient of success here is unwavering commitment on the part of the organization.
Don’t miss Daniel’s conference presentation, Predictive Analytics Applications to Populations Health Management and Staffing Optimization, on Tuesday, September 29, 2015 at 4:15 to 5:00 pm at Predictive Analytics World for Healthcare. Click here to register to attend.
By: Jeff Deal, Conference Chair, Predictive Analytics World Healthcare