Machine Learning Driven Clinical Data Insights

Transcript

1. Machine Learning Driven Clinical Data Insights Chandi Kodthiwada Product Manager, Insights FASTER ACTING R&D
2. www.accenture.com /alsc | Have you been on an Uber ride today? Or Have you unlocked your phone by looking at it?
3. Copyright © 2012 - 2017 Accenture. All rights reserved. 3 WHY NOW? Reference: Mary Meeker – Internet Trends 2018
4. Copyright © 2012 -2017 Accenture. All rights reserved. 4 SIMPLIFIED AI LANDSCAPE SOPHISTICATION MASS ADOPTION Deep AI (continually learning/aware) Narrow AI (basic/routine tasks) Chat Bots Natural Language Processing Personal Assistants (Siri, Alexa) Natural Language Processing Speech Processing Machine Learning Tay by Microsoft Natural Language Processing Machine learning Automated Insights Machine Learning Natural Language Processing (with structured data) Autopilot by Tesla Machine Learning (with Unstructured data, Computer Vision/Situational awareness) Alpha Go (Neural Network) Deep Dream Machine Learning (Neural Network) Einstein (with Structured data) Watson Machine Learning + Speech Processing (with structured & unstructured data) Reference
5. www.accenture.com /alsc | Copyright © 2012 -2018 Accenture. All rights reserved. 5 “THE USUAL” L A B C L IN IC A L O P E R A T IO N S G E N O M IC E H R S E N S O R S & D E V IC E D R U G S A F E T Y E D C 3 RD P A R T Y Staging Processed “CDR” Analytics & Reporting Sources + (1 to 7 days) + (7 to 14 days) + (1 to 7 days)
6. www.accenture.com /alsc | Copyright © 2012 -2018 Accenture. All rights reserved. 6 WHEN, HOW? - THE UX Information Highways Apply Machine Learning to Data Transformat ion Apply Machine Learning to Insight generation Deliver Insights Feedback Loop Enable Insights to Actions Framework
7. www.accenture.com /alsc | Copyright © 2012 -2018 Accenture. All rights reserved. 7 HOW DO WE GENERATE INSIGHTS ? (CONVENTIONALLY) Source Data Report Specification Code Development Validated Code Reports/Logs
8. www.accenture.com /alsc | Copyright © 2012 -2018 Accenture. All rights reserved. 8 HOW DO WE GENERATE INSIGHTS REPORTS? (CONVENTIONALLY) Source Data Report Specification Validated Code Reports/Logs Code Development <Macro Library> <Copy/Paste Code>
9. www.accenture.com /alsc | Copyright © 2012 -2018 Accenture. All rights reserved. 9 APPLY CLASSIFICATION Source Data Feature Extraction Classifier DM.AGE or Not AGE x Variables
10. www.accenture.com /alsc | Copyright © 2012 -2018 Accenture. All rights reserved. 10 NOW, WHAT? Apply Classification knowledge to your Macro Library Macro call: Identify_Change (Study, from_var, Table_name, Visit_Name) Purpose: Parameterized Macro to return a “Change from X” value of the current row where X = baseline, previous visit etc. Invocation: For Study1: Identify_Change(Study1,”previous visit”, LB, VSTNUM) For Study2: Identify_Change(Study2,”previous visit”,LB2, VISIT) Run Parametrized Standard Algorithms Collect Patient Level Insights Burst Push Notifications Learn & Repeat
11. www.accenture.com /alsc | 11 Study Setup Connect Data Source CDIP Study Setup Continuous Surveillance Map data to Pharma-Std v1.0 Delta Detection Study Source Setup ODM Adapter Import Data Rules/Tasks Setup Continuous Surveillance Disable Continuous Surveillance Continuous Surveillance Tasks/Collaboration Notifications System generated Tasks Study Inactivation Study Status: Inactive/Completed /Discontinued USER EXPERIENCE Rules/Tasks Setup Study Data Review Task Setup Missing Data Tracker Tasks (H2H & M2H) Copyright © 2012 -2018 Accenture. All rights reserved.
12. www.accenture.com /alsc | Copyright © 2012 -2018 Accenture. All rights reserved. 12 USER EXPERIENCE End User Notified Insight to Task Collaboration
13. www.accenture.com /alsc | Copyright © 2012 -2018 Accenture. All rights reserved. 13 HOW WOULD THAT PLATFORM EVOLVE? • I can setu p m y stu d y • I can co nnect d ata so u rces & im p o rt d ata fo r the stu d y • D ata S cientist W o rkb ench • I can have Platfo rm m ap m y d ata to a d esired stand ard • I can keep track o f w hat I review ed so far, o nly review w hat’s new and w hat’s u p d ated Horizon 1 Horizon 2 Horizon 3 • Platfo rm tells m e ab o u t interesting trend s and insig hts – I either ig no re o r co nvert to a task • Platfo rm tells m e Insig hts relevant to Trial O p eratio ns fo r so m e alg o rithm d riven scenario s • Platfo rm p o w ers m y insig ht g eneratio n and lets m e act, d eleg ate and track to reso lu tio n o f Insig hts related to C linical Trial d ata as w ell as O p eratio nal health o f a S tu d y • I am ab le to share learning s fro m M ap p ing & Insig ht g eneratio n w ith a larg er co alitio n o f Pharm a Clinical Data Manager Medical Monitor Data Scientist Biostatistician Clinical Trial Ops Manager
14. THANK YOU Patient Inspired. Outcomes Driven.