St. Charles Health System: Looking Forward with Predictive Analytics
The tradition of medicine and healthcare is, by nature, predictive. Patient assessment leads to a prescribed treatment and a prognosis for recovery. However, a web search on healthcare analytics yields dashboard projects that focus on what happened yesterday or last week and shows very little work being done with predictive analytics. This underscores the innovative course charted by Rapid Insight customer, St. Charles Health System (SCHS).
Today, all healthcare organizations are pursuing the triple aim: A better patient experience, stronger population health and cost containment. SCHS is raising the bar with the use of data and technology in pursuit of the triple aim by establishing a goal to use predictive modeling to better manage their organization to pave new pathways in healthcare.
St. Charles is the largest provider of medical care and the largest employer in Central Oregon, with 4200 caregivers at four hospitals, 350 active and 200 visiting staff members. Actionable and effective information is vital to their decision-making process. With four major capital projects weighing in at 80 million dollars, optimizing their resources is crucial.
Improve Healthcare With Predictive Analytics
To ensure this was happening, Dr. Michael Johnson came on board as the Analytics Specialist for Decision Support at SCHS. Several predictive models are now in place and the team is seeing exciting signs of positive change. Dr. Johnson had previous experience with predictive analytics and modeling. First, during his career in the army and more recently, as Director of Institutional Research at Dickinson College. It’s at Dickinson where he first began using Rapid Insight software and saw it as a natural fit to blending data and developing predictive models for SCHS.
“The Decision Support Team is a relatively new addition to the St. Charles IT department and there is some very forward thinking in the organization,” says Johnson. He’s excited by what they have been able to accomplish in a relatively short amount of time. The flexibility of Construct, Rapid Insight’s data blending software, allows the team to pull data from any source, including their electronic health record (EHR) and to incorporate pre-built SQL scripts provided by their data experts. The easy aggregation and transformation of this data lets Johnson and his team deliver predictive models very quickly and they have become essential to the hospital staff’s daily routine.
The goal of the first model, analyzing Length of Stay (LOS), is to improve patient care by optimizing LOS based on primary diagnosis. This particular model uses approximately 18 months of historical data to “score” current inpatients on their potential to exceed the recommended length of stay. Johnson has used Construct’s scheduling tools to automate this process so that each night the model runs, incorporates the latest data, and creates a report that is distributed to administrators and clinical managers at all four SCHS locations early each day. These emailed reports create a higher level of treatment and help caregivers to modify care plans, prioritize efforts and provide care attuned to potential needs of patients with higher LOS scores.
Watch this on-demand video featuring Dr. Michael Johnson, Analytics Specialist for Decision Support at St. Charles, as he shares the path they followed that allowed them to develop, implement, and verify the accuracy of several predictive models over a very short period of time.
The second model is a project to address the potential for patient readmission. The work proved remarkably effective. “It didn’t take much work to tweak the Length of Stay Model for the readmission project,” Johnson said. “The same Construct job could be used to pull the historical data.”
These models were created with the intent of being able to score both pre- and post-discharge patients. The goal of the former is to improve patient care by minimizing the number of current in-patients who have an unscheduled readmission within 30-days of discharge. Dr. Johnson and his team also found that the Rapid Insight readmission models that incorporated data from LACE assessments came to be significantly more accurate than the LACE scores on their own. The more accurate risk rating derived through predictive analytics is a key tool for caregivers in treatment and aftercare who work to reduce the occurrence of readmission… a process that is costly to the organization and an experience that is unpleasant for the patient.
SCHS is able to predict whether or not a pre-discharge patient will have an unscheduled readmission. All patients are scored every day like they are in the LOS model and automated reports are sent each morning to the teams. The process for the post-discharge model is similar, but it scores patients who were discharged in the last five days, and no longer at the facility. Caregivers and administrators get deeper insights into who could potentially walk back through the door, which allows them to better develop protocols for these sort of patients, as well as tune resource planning.
Get Cozy With Predictive Analytics
A concern in healthcare is always how staff adapt to the use of predictive analytics. Caregivers at SCHS are receptive to the new insights that can help them improve patient care. To some, the reporting model functions as a task list for patient follow-up. The reporting also assists with staff time management and helps to allocate limited resources to those who are most at risk. Additionally, the unit staff uses the pre-discharge readmission list to assign a case manager to each patient. This ensures that high-risk readmission patients are closely monitored and aftercare gets focused attention.
For St. Charles Health System, this is only the beginning and future plans for predictive analytics are evolving. Dr. Johnson shares that they will look to begin to build models that predict which individuals are prone to expensive high utilization of services, as well as a model to evaluate the potential for complication or improvement by primary diagnosis. Also in the line-up are two more readmission models. One model focuses on predicting readmit rates by physician. Specifically, identifying over and underperformers based on their patient’s acuity and the case mix index. The second is a readmit model for short-stay and observation patients. The Decision Support Team would also like to create an employee turnover model for human resources.
St. Charles has clearly rejected the status quo of using retrospective, descriptive metrics, realizing that management against what has already happened is simply not good enough. With the use of Rapid Insight’s Construct and Predict, SCHS has truly adopted a forward-looking analytic mindset that helps them change the course of costly outcomes and negative patient experience. Predictive modeling is driving exciting changes in their daily operations and inspiring an innovative future with predictive analytics.