The Potential for Predictive Analytics in a Pediatric Care Setting A Case Study

Transcription

1

2 The Potential for Predictive Analytics in a Pediatric Care Setting A Case Study March 11, 2015 Dr. John Showalter, CHIO for UMMC Ritesh Sharma, COO for Jvion

3 Learning Objectives By the end of today s session, you will be able to: Build a business case for predictive analytics Understand how predictive analytics can target hospital acquired conditions Describe how predictive analytics can improve overall patient safety

4 By a show of hands Raise your hand if you are currently using advanced predictive modeling technology within your organization. 4

5 The Healthcare Industry and Predictive Analytics Resistance to change Disparate data collection Underinvestment in IT Barriers to sharing data

6 The Healthcare Industry and Predictive Analytics Supply of data Demand for better data Government programs Technological advancements

7 The Healthcare Industry and Big Data

8 Predictive Analytic Sausage Making Identify and account for 1000s of variables Apply deep machine learning technologies Incorporate clinical intelligence Continuously learn

9 Predictive Analytic Sausage Making

10 Predictive Analytic Sausage Making The Data Monster Dimensionality vs. Granularity Causality Low Occurrence Events

11 What Makes A Predictive Analytic Solution Right for Healthcare? Practicality Accuracy Scalability Flexibility Affordability

12 By a show of hands Raise your hand if you would you consider using predictive modeling technology to calculate patient illness or deteriorating conditions? 12

13 Applying Predictive Analytics Infection Control Population Health Readmissions High-risk Patients Patient-level Disease Predictive Analytics Financial Losses

14 Applying Predictive Analytics Infection Control Population Health Readmissions High-risk Patients Patient-level Disease Predictive Analytics Financial Losses

15 Building the Business Case

16 Project Overview Setup (complete): during which we establish the file transfer, extract definitions, kickoff the project with key stakeholders, identify the user group, and demo the solution to physician champion(s) Machine Modeling (complete): during which we build the heuristic model, build the clinical model for the AMI use case, integrate the clinical and heuristic models, and validate/benchmark the tuning of the models Initial Results (complete): during which we present initial findings, review the process, and integrate the predictive outputs into the clinical workflow Steady State (ongoing): during which the predictive machine fires daily results in the background technical landscape, continue to tune processes and integration to maximize value, and users actively access the portal/receive support

17 Potential Impact 21% decrease in unreimbursed cost related to stage iii and iv pressure ulcers 28% drop in number of patients with stage iii and iv pressure ulcers 30% decrease in total number of length of stay days

18 By a show of hands Which of the following is the biggest barrier to predictive analytic adoption? A. Workflow disruption B. Mistrust of the outputs/data C. Perceived complexity D. Peer acceptance 18

19 Getting Our People Engaged Clinician adoption is critical to our success. Our solution had to: Easily fit into the clinician s workflow Demonstrate immediate accuracy and effectiveness Be proven and hold up to scrutiny Be easy to use and understand Be accepted by fellow clinicians

20 Improving the Patient Path Before Care Preventable Admissions Population Management ED Frequent Flier Prevention Elective Surgery Market Share During Care Hospital Acquired Conditions High Impact Disease Cost Control (Sepsis/CHF) LOS Optimization Case Mix Optimization After Care Readmission Prevention Aftercare and Rehab Cost Control Home/Healthy Risk Nursing Home Irregular Care ER Home Regular Care Outpatient Inpatient Emergency Sx Medical Mgmt. Elective Sx Adverse Event No Event Mortality High LOS Optimal LOS Nursing Home Home Mortality Readmission ER Outpatient

21 Lessons Learned The solution has to fit into the real world of the hospital Knowing when a model is wrong is just as important as know when it is right Predictive solutions should come pre-seeded with use cases Associations are powerful

22 Next Steps in the Predictive Analytic Journey Integrate new forms of data as they become available: Health Information Exchange Referring Physician Offices Patient and Family Inputs Expand visibility to care givers outside the walls of the health system Continually measure the effectiveness of our preventions Identify new target areas where we can apply low cost interventions to avoid potentially high cost and high patient suffering events Utilize data to transition to new value-based models of reimbursement and care coordination

23 The Potential for Predictive Analytics in a Pediatric Care Setting Questions?