AI for Quality & Risk Management

Transcription

1 AI for Quality & Risk Management Reducing Complexity to Deliver Improved Clinical Outcomes, Operational Efficiency, and Profitability Risk Adjustment Quality & Care Management Underwriting Ash Damle Founder & CEO

2 Healthcare s 21st Century Challenges 2

3 Business Leaders Ask 3 Key Data Questions To Align the Mission & Business of Healthcare What is the current and future health of the people we serve? What is the current and predicted future utilization, and associated cost, for each person based on what we know? What resources or products (and at what price) help make that utilization a practical reality? 3

4 ECOSYSTEM CHALLENGES The next 20 years 1. Supply vs Demand 2. Operations & Coordination 3. Reimbursement: Transitioning From Fee-For-Service to Value-Based 4. Data & Analytic 5. Regulatory 6. Access

5 So, What is AI? 5

6 { Is this Siri or Alexa? }

7 7

8 So: What Is All this Stuff? Artificial Intelligence Systems that learns to solve complex tasks that previously required humans to do Machine Learning How computers are able to learn without being explicitly programmed Deep Learning Machine learning mimicking how the human brain works based on our current understanding 8

9 How AI is Used Today It s Early Days, But AI is Being Applied in Healthcare... Clinical decision support at the point of care (IBM Watson) Reading and interpreting images (i.e., radiology) Understanding and interpreting audio (i.e., clinical notes transcription) Reading and translating handwriting and unstructured data Understanding and interpreting video...unfortunately, the results are not uniformly good or reliable.

10 AI s Challenge in Healthcare 10

11 AI Healthcare Challenges Spans 3 Dimensions α Data Transparency Packaging Healthcare is databased industry Why is just as important as What Realized Value = Value / (Ease of Use)

12 Data : Healthcare is the most complex industry Missing Data N-M Mappings Fuzzy Classifications Inconsistent Data Fragmented Records Sampling 3-Year Churn 1.1M Condition Variables 600k Procedure Variables Zetabytes/yr and High Variability 4.5M Medication Variables 2.5M Lab / Imaging Variables 200K Provider Variables 2.5M Other Variables 80% of time spent cleaning/preparing data

13 Transparency : Would you use Google Maps without the Maps?

14 Packaging: Making it super simple to use

15 WELCOME TO Lumiata Align the mission and business of healthcare Lumiata AI interprets and extends your data to drive improved health outcomes clinically, behaviorally, and financially 80M+ Patient Records 50M+ Knowledge Articles 40K+ Physician Hours

16 Create Better Data with Lumiata s AI that synthesizes the knowledge of a physician. empathy of a therapist. rigor of an actuary Clinical Behavioral Financial Accurate Diagnosis Targeted Impactability High-Fidelity Utilization Predict the current and future conditions with clinical rationale Identify the most impactable patients and providers to prioritize resources Understand the projected care and costs for individuals and groups

17 Benefit From The First Medical Self-Learning Modeling Platform Lumiata s Medical Model & ETL Interpretable Deep Learning Real-time Prescriptive Ranking + = Built on decades of research mapping medical conditions combined with 80M+ patient records to enrich raw data, build and train models Lumiata artificial intelligence technology combines our medical models with personal, social, and family medical history to predict medical futures Lumiata converts complex, disparate data into easy-to-use prioritized targeted population lists for health plans & providers to allocate resources optimally Learn More

18 Simplify Prescriptive Ranking Delivering An Individualized Population Matrix for more Impactful Lists Targeted Quality & Care Management Comprehensive Risk Adjustment Accurate Underwriting

19 One Platform, Multiple Benefits For Health Plans, ACOs, Health Systems Diagnosis and Detection Quality & Care Management Underwriting Risk Adjustment Acute diagnosis of medical condition based on symptoms, history, etc. for better triage, and treatment options to deliver best outcomes at lowest cost Identify members with impact-able conditions so that health plan can implement programs to increase clinical efficacy, lower lifetime costs, and optimize utilization Accurate prediction of future individual and group health trajectories and associated costs leads to better underwriting decision making Identifying undocumented and/or undiagnosed conditions along with clinical rationale for more accurate and justifiable payments

20 Packaged Prioritized Lists Because Healthcare Runs on Lists Health Plans Providers Patients risk adjustment chase lists quality management lists care management lists underwriting lists... daily patient list scheduling list high risk list high utilizers list... provider list care plan list... 20

21 Product: Prescriptive Lists for Health Plans Risk Utilization Quality & Care Risk Adjustment Management Utilization and Underwriting Management Quality and Care Management Prioritized lists that easily plug into current processes to help correctly diagnose all members, and ensure both quality care and accurate payments. Predicting future clinical states and mapping those to existing risk scores and actuarial tools Predicting risk of conditions & optimal interventions with associated clinical rationale 21

22 Powered by Explainable AI for Healthcare Predictive/Prescriptive Medicine With A Different ML Approach Trained By Data Scientists/Physicians Medical Knowledge Graphs & Maps 50M PubMed Articles 40K Physician Hours 40M Concept 70M+ Patient Records & Growing Network Effect 175M Patient Record Yrs 20M+ 3yrs+ data 31M 1 yrs+ data Interpretable Deep Learning Answering why not only drives credibility, but is key to determining the right care path. 22

23 What We Do : Prescriptive Lists for Health Plans Demographics Past Medical History Med/Pharmacy Claims Social + Behavioral Phys. Exam, Vitals, Labs Lumiata AI 60M Patient Records Medical Knowledge Explainable Deep Learning Clinical Conditions & Categories Per Dx Associated Providers Care Quality Risk Adj. Wearables, IoT 23

24 Rethinking Data Through AI RAW DATA AI for DATA PREP AI for DATA ANALYSES AI for DATA ACTION Aggregate multiple sources and types of data structured and unstructured Normalize and enrich data to create longitudinal member records Predict future health states and associated costs on an individual and population basis Organize and prioritize data with associated clinical rationale

25 Transparency Key to Inspiring Action

26 Precision Clinical ROI 2x-3x Better than Precision MEASURABLE Leveraging more than 60 million patient records, medical science, and deep learning, Lumiata AI delivers a provable 30% or better increase in accuracy and at least a 3x ROI Standard Market vs Lumiata AI 3x+ ROI Recall 26

27 CASE STUDY Topline Impacts (e.g., Risk Adjustment for Diabetes) Standard Regression Regression + Lumiata AI for Data Lumiata AI for Data + Deep Learning 1K List $540K $1.5M $1.5M 5K List $1.35M $3M $3.6M 10K List $1.2M $3.3M $3.9M e.g. Median Payout for a DM2 Chase List for Medicare Advantage 10 : 1 Ratio on a 100K population where cost = $300 and benefit = $3,000 27

28 CASE STUDY Topline Impacts (e.g., Cost Capture and Underwriting) Cost Capture as a Function of Member Rank 2017 Member Population of 3m 28

29 FINANCIAL CASE STUDY 2x Lift (e.g., Chase Lists for Risk Adjustment) MEASURABLE The utility of each process is predicated on the accuracy of list prioritization. Lumiata AI optimizes chases by simulating the possibilities and then delivers an order of operation designed to get the best results. List Capacity Precision 1, % 5, % 7, % 10, % 15, % 20, % 29

30 Product: Current & Potential of Lumiata s RA List Value lum_pred lum_max client_max client_pred ~$68M ~$40M ~$17M With a Potential Additional $28M in the Future Delta Value Generated by Lumiata. 30

31 CLEAR CUSTOMER ROI $24M+ Revenue & 10x Value Risk Adjustment Process Analysis of a commercial risk-adjusted population POPULATION PIPELINE RESULTS ESTIMATED ROI 967,891 Members 244 Potential Deep Learning Models 57 Select HCC Models Applied 58 Applied Models for Optimal Provider Per Member / HCC 27,198 Member Chase List ( 1 Resource Per Member & Provider Prediction) 44,187 Suspected HCCs After Confidence & Accuracy Thresholding $24+ Million Additional Revenue Estimated Value Capturable 31

32 E.G. CUSTOMER ROI It s Not Just Billing, It s Improved Quality Risk Adjustment Process Analysis of a commercial risk-adjusted population : Care Triggering Process POPULATION RESULTS BETTER CARE OPPORTUNITIES (The Uncoded/Undiagnosed & Uncared For) 967,891 Members 27,198 Member Chase List ( 1 Resource Per Member & Provider Prediction) 44,187 Suspected HCCs After Confidence & Accuracy Thresholding 9257 Diabetics 6157 Diabetics with Complications 5289 Asthmatics 2123 Tumors, and Other Cancers 1674 Specified Heart Arrhythmias 1662 Rheumatoid Arthritis 1247 Congestive Heart Failure 988 Inflammatory Bowel Disease s More + Predicted Associated Providers 27,169 Provider Matches found for Members Ensuring Care By Outreach To 2932 family medicine 2226 internal medicine 653 pediatrics 511 cardiovascular disease 498 obstetrics & gynecology 441 allergy & immunology 349 neurology 313 gastroenterology 312 endocrinology s More 32

33 E.G. VIZ With Clinical Rationale that Leads to Better, More Confident Decisions (Mock Viz) Moving the conversation from an administrative one to a clinical one, where action is taken on data insights up to percent of the time because each opportunity is backed by a clear clinical rationale 33

34 RISK MATRIX For Risk Adjustment Predict, identify, and rank HCCs based on all available member data for more accurate, better targeted chart-pulls and reviews, resulting in accurate coding and correct payments. Better identification of members: Moving from blunt scoring to precise, rank-ordered lists 1 1 Better optimization of opportunities: Moving from inefficiency to highly targeted chartpulls 34

35 RISK MATRIX Delivering Accuracy What Was Coded What Should Have Been Coded Condition ICD-10 Code HCC Risk Score Condition ICD-10 Code HCC Risk Score Diabetes Mellitus with Diabetic Nephropathy E Diabetes Mellitus with Diabetic Nephropathy E Peripheral Vascular Disease, Unspecified I Peripheral Vascular Disease, Unspecified I Chronic Obstructive Pulmonary Disease, Unspecified J Chronic Obstructive Pulmonary Disease, Unspecified J RAF Score: Total Payment: $10,130 Sick Sinus Syndrome I Chronic Viral Hepatitis C B BMI , Adult Z RAF Score: Total Payment: $19,240

36 UNDERWRITING MATRIX High Level Illustration Group X Renewal Use Current Data, Additional 3rd-Party Data New Get Characteristics, Additional 3rd-Party Data Generate Per Member FHIR Bundle (Data per patient transformed into FHIR, standardized, normalized, and temporally ordered) Apply AI for Prediction, Associated Utilization Group X Underwrite Renewal Accurate Pricing, Improved Profitability Underwrite New Business Competitive Pricing, Improved Profitability Apply to Pricing to Get to Street Rate Generate Lumiata Risk Assessment FHIR Resource 36

37 QUALITY & CARE MATRIX Overview Care Matrix predicts, and ranks the current and predicted future health state of each individual based on all available member data, together with predicted associated future utilization costs, to better identify candidates for modifiable risk programs. The Prediction The Risk Drivers Ranked and Segmented List Integrated in Current Workflow CKD CLINICAL RATIONALE 37

38 QUALITY & CARE MATRIX High Level Illustration Group X Known Use Current Data, Additional 3rd-Party Data New Get Characteristics, Additional 3rd-Party Data Generate Population Simulator Generate Per Member FHIR Bundle (Data per patient transformed into FHIR, standardized, normalized, and temporally ordered) Apply AI for Data, Prediction, and Engagement Rank-ordered lists of modifiable risk candidates by disease subgroup Apply AI for Clinical, Financial, and Interveneability Assessment Generate Lumiata Risk Assessment FHIR Resource Utilization Matrix + Risk Matrix + Clinical Rationale 38

39 Thank you lumiata.com 39