1) Predictive analytics in healthcare often provides risk scores and predictions but lacks actionable insights on how to prevent outcomes.
2) The right methodology is needed to transform raw data like claims, prescriptions and medical records into meaningful predictions using machine learning algorithms.
3) Accurate predictions require measuring precision down to the individual level while accounting for both patient and provider factors that influence health outcomes.
Powerful Google developer tools for immediate impact! (2023-24 C)
“ High Precision Analytics for Healthcare: Promises and Challenges” by Sriram Vishwanath
1. High Precision Analytics for Healthcare:
Promises and Challenges
Sriram Vishwanath
Professor, UT Austin
Cofounder, Accordion Health
President, Brilliant.MD
2. Problems with Predictive
Analytics
Where Are My Actionable Insights?
“… Software X is a black box. I put my data,
and it gives me some sort of risk scores. I
know that high risk scores are bad. So, what
should I do next? …”
“… I purchased Software Y, and it gives me
a report that there have been thirty
preventable readmissions in the last month.
But I want to know what to do to prevent
them in the future … “
4. A Good Approach
• Population Health Personalized Health
• Identify High Risk Patients Predict Change of Risk
• I can Predict it all Based on Measured Precision
Key Insight
Provider is as critical as patient in determining outcomes
5. The Importance of Right Methodology
Claims
Rx
Labs
EHR
transform
into
tensors
feature
extraction
apply algorithms
(ML and traditional)
model
Input
Actionable
Insight
Intervention
GLM
kNN
RF
*courtesy Accordion Health
7. Example – Joe S.
• 69 y/o man with COPD & h/o acute
exacerbations
• Tend to occur annually with seasonal triggers
• Also has DM, HTN which are relatively poorly-
controlled
• He does not always take his COPD meds
• PCP: Dr. Alvarez (and other members of
healthcare ecosystem)
• Risk score: Medium
8. Example – Joe S.
Joe had a COPD
exacerbation
last spring…
So, it’s not surprising
that he will likely have
another exacerbation
next spring
Difficulty in Prediction : Easy
Associated Costs: High
Intervention: Medication Reminder Intervention:
Home-visit
Efficacy: Low Efficacy: High
9. Example – Linda R.
• 76 y/o woman with h/o well-controlled
Hypertension
• Family h/o of CVD
• Recently seen for palpitations, but
otherwise asymptomatic
• Mostly adherent to medication
• PCP: Dr. Lin
• Risk score: Low
10. Example – Linda R.
Although palpitations are
asymptomatic
We predict severe cardiac
dysrhythmia, like atrial fibrillation And the likelihood
of a stroke is high
Difficulty in Prediction : Hard
Associated Costs: Extremely High
Intervention: PCP-visit, additional medication prescribed
Efficacy: High
12. Predicted Superutilizers
Alice S.
Bob W.
Cindy N.
Doug D.
Eve A.
Frank L.
George B.
Hank T.
Ivana M.
Jack K.
Alice S.
Cindy N.
Keith L.
Larry L.
Mary W.
Nancy S.
Olivia Z.
Patrick W.
Quincy A.
Robert S.
14. BUNDLING: POST-ACUTE RISK PREDICTION
Post Acute Pathways
Discharge Date
Day 0
CJR Period
Day 90
Home Health
SNF
Inpatient
Good Decision:
Patient A (blue)
placed in a Skilled
Nursing Facility
(SNF), then goes
home.
Bad Decision:
Patient B (red)
placed in (HHA)
after discharge,
resulting in
readmission due to
surgical
complications.
Patient A
Patient B
*courtesy Accordion Health