By: Jeff Deal, Program Chair, Predictive Analytics World Healthcare
In anticipation of his upcoming keynote conference presentation at Predictive Analytics World for Healthcare Boston, Sept 27-Oct 1, 2015, we asked Dr. Michael Dulin, Chief Clinical Officer for Analytics and Outcomes Research at Carolinas Healthcare System, a few questions about incorporating predictive analytics into healthcare. Catch a glimpse of his keynote presentation, Turning Big Data into Better Care, 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: Mainly clinical outcomes and quality improvement. For example, we currently are using predictive models in the areas of readmission risk and length of stay. These models mainly serve two main purposes: To predict which patients have a high risk of readmission via model-based risk bands and to proffer interventions based on which variables are loading high in the model. We are also doing this with length of stay.
Q: What outcomes do your models predict?
A: In addition to the readmission and length of stay models, we just completed state-of-the-art dynamic time-to-event models for hospitalization and developing Type II diabetes. In the case of hospitalization, we construct patient level survival curves, which capture the amount of time to the event of first hospitalization and their associated probabilities. In the case of Type II diabetes, we construct survival curves at the patient level for developing diabetes. With these models, we are also able to see insulating factors in our patient population, which may offer ways of reducing a patient’s probability of hospitalization or developing Type II diabetes.
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: One way is through the objective lens of analytics. We recently completed a purely data driven segmentation of our patient population, where we found seven segments in our population. Along-side of this, we developed a classification model to score new patients into the segments with very high accuracy. This allows us to understand the population at a very deep level and optimize care to patients in each segment. With our classification model, we are able to see the segment migration and understand the variables that drive the migration, which offers possible patient interventions to stop migration to more unhealthy segments.
Q: Can you describe a successful result, such as the predictive lift of your model or the ROI of an analytics initiative?
A: Missing data is always with us. In an effort to model non-patient probability of having commercial coverage, missing data was a significant impediment to our effort. Our elite population health analytics team (the special forces of DAA) created and implemented missing data methods to increase the positive prediction of the model by about 35%.
Q: What surprising discovery have you unearthed in your data?
A: In patient discharge data, there was a long held belief in cyclical (seasonality) patterns longer than one week, such as summertime or wintertime effects. We found there is no statistical evidence of this belief, though the use of spectral decomposition and statistical smoothing.
Q: What areas of healthcare do you think have seen the greatest advances or ROI from the use of predictive analytics?
A: The application of predictive analytics in healthcare is just in its infancy. Digital healthcare data is doubling about every two years and the amount of semi-structured and unstructured data is increasing as well. Some of the greatest advances in the future will be in the efficient and meaningful delivery of relevant information for bettering patient care and outcomes, and reducing healthcare costs.
By: Jeff Deal, Conference Chair, Predictive Analytics World Healthcare