Predictive Analytics Help Achieve the Triple AimHolli White | BMHI MPS Student INLS 770 | University of North Carolina at Chapel Hill
Agenda
What is the Triple Aim? Financial/utilization trends Types of predictive analytics
Disease progression Hospitalization and/or death prediction Expenditure
Tying it all together
What is the Triple Aim?
2. Triple Aim Goals
A very small percentage of the population is responsible for half of the dollars spent.
1. Expenses for Costly Medical Conditions
3. NIHCM Foundation Data Brief July 2012
Predictive models can identify patients with whom to take a proactive approach.
Types of predictive analytics Disease progression Hospitalization and/or death prediction Expenditure
Progression of CKD to Kidney Failure Many biometric/laboratory variables High accuracy rate
8. Predictive for Progression of CKD
Predictive models can identify patients with whom to take a proactive approach.
Types of predictive analytics Disease progression Hospitalization and/or death prediction Expenditure
HF predictive models - uses risk markers and CHA2DS2-VASc scores, current treatment therapies and response, EF values
Models incorporating clinical data are more accurate than claims
4. Factors influencing HF predictive models
Predictive models can identify patients with whom to take a proactive approach.
Types of predictive analytics Disease progression Hospitalization and/or death prediction Expenditure
COPD study Successful risk model put together based on scoring of various
measurements
7. Pred. Model of Hosp/Death in COPD
Types of predictive analytics Disease progression Hospitalization and/or death prediction Expenditure
H
7. Pred. Model of Hosp/Death in COPD
Predictive models can identify patients with whom to take a proactive approach.
Types of predictive analytics Disease progression Hospitalization and/or death prediction Expenditure
Diabetes predictive model estimates cost of complications over 1-, 3-, and 5-year timeframes
Allows organizations to identify opportunities around network leakage or overutilization of services
9. Estimating the Cost of Incident Diabetes Complications
Predictive models can identify patients with whom to take a proactive approach.
Types of predictive analytics Disease progression Hospitalization and/or death prediction Expenditure
Most Costly Diseases Annual Expected Cost 5-Year Expected CostCHF $7,320,287 $50,697,865ESRD $4,225,384 $13,211,204Gangrene $2,844,381 $17,200,417
Most costly complications in a sample of 10,000 diabetic adults
9. Estimating the Cost of Incident Diabetes Complications
How do we tie this together with patient care?
Community based outreach for children with asthma had positive results Homes based education Quality of life increased Reduction in hospitalizations 93% of patients had management plan vs. 31% at baseline
6. Eval. of Community Based Outreach
How do we tie this together with patient care?
Aurora Healthcare in Milwaukee Utilized Optum One’s CHF predictive models Reduced all cause readmissions by 30% 60% reduction in HF admissions Reduced ER utilization
5. Predictive Analytics for Population Health Management
In summary, predictive modeling can identify high risk patients for intervention, aiding in achieving the Triple Aim.
Reduce spend Increase patient satisfaction Improve the health of the population
References
1. Cohen, S. (October 2014). The Concentration of Health Care Expenditures and Related Expenses for Costly Medical Conditions, 2012. Medical Expenditure Panel Survey, Statistical Brief #455.
2. Maine Health Management Coalition. (2016). Triple Aim Goals. Retrieved from: http://www.mehmc.org/about-us/what-we-do/triple-aim-goals/
3. NIHCM Foundation. (July 2012). The Concentration of Health Care Spending. NIHCM Foundation Data Brief July 2012.
4. Ouwerkerk, W., Voors, A., & Zwinderman, A. (2014). Factors Influencing the Predictive Power of Models for Predicting Mortality and/or Heart Failure Hospitalization in Patients With Heart Failure. The American College of Cardiology Foundation, Vol 2, 429-436. http://dx.doi.org/10.1016/j.jchf.2014.4.006
5. Paulus, J. & Salfity, K. 2016 SOA Health Meeting. Session 131 PD, Predictive Analytics for Population Health Management. June 15-17, 2016. Philadelphia, PA.
6. Primomo, J., Johnston, S., DiBiase, F., Nodolf, J., & Noren, L. (2006). Evaluation of a Community-Based Outreach WorkerProgram for Children With Asthma. Public Health Nursing, Vol 23, 234-241. doi: 10.1111/j.1525-1446.2006.230306.x
7. Schembri, S., Anderson, W., Morant, S., Winter, P., Pettitt, D., MacDonald, T., & Winter, J. (November 23, 2008). A predictivemodel of hospitalisation and death from chronic obstructive pulmonary disease. Respiratory Medicine, 103, 1461-1467.
8. Tangri, N., Stevens, L., Griffith, J., Tighiouart, H., Djurdjev, O., Naimark, D., Levin, A., & Levey, A. (April 20, 2011). A PredictiveModel for Progression of Chronic Kidney Disease to Kidney Failure. JAMA, Vol 305, 1553-1559.
9. Zhu, J., Kahn, P., Knudsen, J., Mehta, S., & Gabbay, R. (2016). Predictive Model for Estimating the Cost of Incident DiabetesComplications. Diabetes Technology & Therapeutics, Vol 18, 625-634. doi: 10.1089/dia.2016.0132