SLIDE 1 NIH Collaboratory Grand Rounds
Improving Chronic Disease Management with Pieces: Overview of PCCI and Pieces™
Ruben Amarasingham, MD, MBA
Associate Professor, UT Southwestern Medical Center CTSA Director of Biomedical Informatics CEO, Parkland Center for Clinical Innovation (PCCI)
Friday, January 16, 2015 12:00-1:00 pm CT
SLIDE 2
- What is electronic health predictive analysis (e-HPA)?
- PCCI’s work in this area: Pieces™ software
- Application of Pieces™ in the CKD Pilot Study and ICD-Pieces
trial
- Pieces™ in Community Health Information Exchanges
Objectives
SLIDE 3 What Clinicians Do in Medicine: Prediction
- 1. What does this patient have?
- 2. What will this patient develop?
- 3. What will be the effect of a given therapy?
SLIDE 4 Prediction in the Context of Modern Medicine
30 90 45 1900 1920 1940 1960 1980 2000 2020 2040 2060 2080 2100 1 15 60 75 105 120 135 150
Year
Doubling Time of Medical Knowledge 1900: 150 years
SLIDE 5 Prediction in the Context of Modern Medicine
30 90 45 1900 1920 1940 1960 1980 2000 2020 2040 2060 2080 2100 1 15 60 75 105 120 135 150
Year
Doubling Time of Medical Knowledge We are here: 1 year
SLIDE 6 Prediction in the Context of Modern Medicine
30 90 45 1900 1920 1940 1960 1980 2000 2020 2040 2060 2080 2100 1 15 60 75 105 120 135 150
Year
Doubling Time of Medical Knowledge We are here: 1 year
- Staggering increase in medical information
- Increasing volume of decisions at multiple levels
- High fragmentation of care
- Increasing capacity for error
SLIDE 7 7
Some Definitions
Clinical risk prediction models: defined as models that “combine a number
- f characteristics (e.g. related to the patient the disease or treatment) to
predict a diagnostic or prognostic outcome” [Steyerberg 2009] Electronic health care predictive analytics (e-HPA): the technologies or software systems that can autonomously employ – and sometimes re- engineer, modify, or update – these models” [Amarasingham 2014]
SLIDE 8
PCCI Organizational Background
A 501c(3) non-profit research and development corporation specializing in the development of clinical prediction and surveillance software to help prevent adverse clinical events.
SLIDE 9
PCCI Scientific Funding for Predictive Modeling
SLIDE 10 Every Adverse Event has a Timeline
30 days 90 days Years Hours
Arrest
Complications
Diabetic Complications
Hospitalizations
hypertension, CKD Readmissions
SLIDE 11 Adm ission Discharge 3 0 Days 9 0 Days
24 hours
Every Adverse Event has a Timeline
SLIDE 12 2 1 3 5 5 4
ID Risk List Orders Inpatient Intervention Outpatient Intervention
Adm ission Discharge 3 0 Days 9 0 Days
24 hours 7 days
Pieces
6
Evaluation & Improvement
EMR
Preventing Heart Failure Readmissions
SLIDE 13
Identification of HF patients in Real-Time Using Natural Language Processing and Data Mining
SLIDE 14 Natural Language Processing
“68 yo WF presents with acute on chronic non ischemic systolic and diastolic chf, severely depressed ef and grade ii diastolic dysfunction.” Disease/ Symptom Time Attribute Acute Heart Failure current and primary
depression in ejection fraction;
grade 2
Chronic Heart Failure historic
SLIDE 15
System calculates risk for readmission
SLIDE 16 8.77 14.27 17.94 26.93 51.65 45.68 26.0 19.98 16.08 12.22
10 20 30 40 50 60 70 30-Day Readmission (%)
Very Low Low Intermediate High Very High
Predicted Readmission Risk Category Derivation Samples Validation Samples
Identifying High-Risk Patients in Real-Time
*
Amarasingham et al, Medical Care, 2010
SLIDE 17
Pieces provides list of targeted high risk patients
SLIDE 18
Activation of Clinical Pathways in the EMR
SLIDE 19
Pieces tracks interventions in the EMR
SLIDE 20
Pieces monitors outcomes
SLIDE 21
Complexities of Predictive Modeling in Healthcare
SLIDE 22
The Complexities of Predictive Modeling
Amarasingham et al, Health Affairs, 2014
1. Interventions for highest risk patients * 2. Considering clinical vs. social risk 3. Explanation vs. Prediction 4. Non-health care data sources * 5. Changing EMR data models 6. Changing clinical interventions 7. Changing populations
SLIDE 23
management efforts
- n ¼ of the patients
- 26% relative reduction
in odds of readmission
readmissions per 100 index admissions
Amarasingham et al, BMJ, 2013
SLIDE 24
A Different Hospital: Readmission Performance
SLIDE 25 NIH-Funded CKD Pilot Study
Screening Last follow-up visit Clinical Measurement % at Goal % at Goal n=107 n=107 P-value (McNemar’s test) Follow-up duration, month median [range] 11.2 [0.2 – 21.5] Systolic blood pressure 34.6% 44.0% 0.14 Diastolic blood pressure 57.9% 66.1% 0.17 ACEI/ARB 57.8% 87.2 <.0001 Statin 45.0% 79.8 <.0001
if positive test for proteinuria or albuminuria, then goal BP <130/80; Otherwise goal BP < 140/90).
SLIDE 26
ICD-Pieces Study Sites
SLIDE 27
Connecting the Community: WW Caruth
SLIDE 28 Connecting the Community: WW Caruth
prescriptive analytics on medical and social data to identify at risk individuals
SLIDE 29
Thank You!
SLIDE 30
Questions
Contact Information:
Ruben Amarasingham – ruben.amarasingham@phhs.org www.pccipieces.org
