Copyright	
  ©	
  2015	
  Splunk	
  Inc.	
  
Big	
  Data	
  Analy<cs	
  and	
  Decision	
  Support	
  
	
  
Adrish	
  Sannyasi,	
  	
  
Healthcare	
  Solu<ons	
  Architect	
  
	
  
 	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  Agenda	
  
2	
  
Aim	
  of	
  Big	
  Data	
  Analy<cs	
  
Opera<onal	
  Decision	
  Support	
  
Clinical	
  Decision	
  Support	
  
Data	
  Analy<cs	
  Infrastructure	
  and	
  Methods	
  
3	
  
Triple	
  Aim:	
  Healthcare	
  Delivery	
  System	
  
VALUE	
  =	
  Outcome	
  (quality,	
  
safety,	
  experience)	
  /	
  Cost	
  
Aim	
  of	
  Big	
  Data	
  Analy<cs	
  
4	
  
Help	
  Make	
  Op<mal	
  Decisions	
  
Pa<ents	
  
Administrators	
  and	
  
Policy	
  Makers	
  
Providers	
  
Proac,ve	
   Precise	
   Predic,ve	
  
Moving	
  towards	
  precise	
  decision	
  making	
  
5	
  
VS	
  
Current	
  Vs.	
  Desired	
  Decision	
  Support	
  
6	
  
Current	
  Approach	
   Desired	
  Approach	
  
Rule	
  based	
  system	
  
High	
  rate	
  of	
  false	
  alarms	
  
Missed	
  opportuni<es	
  
Precise	
  and	
  Context	
  Sensi<ve	
  
Workflow	
  Interrup<on	
   Automated	
  Data	
  Collec<ons	
  
Mainly	
  structured	
  EMR	
  or	
  Claims	
  data	
   Structured	
  and	
  Unstructured	
  data	
  from	
  EMR,	
  
sensors,	
  wearable,	
  behavior,	
  and	
  
environmental	
  data,	
  and	
  condi<on	
  focused	
  
social	
  network	
  data.	
  
One	
  <me	
  measurements	
  of	
  physiological	
  
sta<s<cs	
  
Con<nuous	
  measurements	
  	
  and	
  pathway	
  
oriented	
  measurements	
  
Low	
  transparency	
  and	
  accountability	
   Transparent	
  and	
  Accountable	
  to	
  pa<ents	
  and	
  
care	
  team	
  
Building	
  a	
  “Learning	
  and	
  Improvement	
  Engine”	
  
7	
  
Sympathy-­‐
Man,	
  We	
  could	
  
do	
  be`er	
  
	
  	
  Empathy	
  –	
  I	
  feel	
  your	
  pain	
  
Compassion-­‐	
  
Let	
  me	
  help	
  you	
  
Source:	
  HCA	
  
Big	
  Data	
  Analy<cs	
  must	
  be:	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  
Valid:	
  hold	
  on	
  new	
  data	
  with	
  some	
  certainty	
  
	
  
Useful:	
  Should	
  be	
  ac<onable	
  
	
  
Unexpected:	
  non-­‐obvious	
  to	
  consumers	
  
	
  
Understandable:	
  humans	
  should	
  be	
  able	
  to	
  interpret	
  
	
  
Measurement	
  is	
  useful	
  if	
  it	
  facilitates	
  ac,on.	
  	
  
Measure	
  what	
  is	
  important	
  to	
  customer.	
  	
  
Examples:	
  Opera<onal	
  Decision	
  Support	
  
9	
  
•  Scheduling	
  and	
  Staffing	
  Assistance	
  
•  Predic<ng	
  and	
  alloca<ng	
  service	
  area	
  and	
  unit	
  capacity	
  
•  Predic<ng	
  bed/room	
  requests,	
  LOS	
  targets,	
  transport	
  services	
  	
  
•  Op<mizing	
  Asset/Inventory	
  U<liza<on	
  	
  
•  Reducing	
  claims	
  processing	
  cost,	
  error,	
  and	
  <me	
  
•  Reducing	
  Fraud,	
  Waste,	
  and	
  Abuse	
  
	
  
	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  
	
  
Opera<ons	
  Decision	
  Support	
  Center:	
  	
  
Air	
  Traffic	
  Control	
  System	
  for	
  Opera<onal	
  Decisions	
  	
  
•  Modeled	
  afer	
  your	
  security	
  opera<ons	
  center	
  and	
  IT	
  opera<ons	
  center	
  
•  Track	
  pa<ent	
  movements	
  and	
  oversee	
  opera<ons	
  and	
  throughput	
  
•  Proac<vely	
  an<cipate	
  needs	
  for	
  services	
  
•  Coordinate	
  staffing	
  and	
  scheduling	
  
•  Coordinate	
  admissions,	
  transfers,	
  discharge	
  planning	
  and	
  execu<on	
  
•  Reduce	
  cross	
  departmental	
  hand-­‐off	
  issues	
  
Informa<on	
  Flow	
  in	
  Care	
  Delivery:	
  Spagheh	
  	
  
h`p://www.ncbi.nlm.nih.gov/pmc/ar<cles/PMC3002133/	
  
From	
  Spagheh	
  To	
  Lasagna:	
  
Reduce	
  Unwarranted	
  Varia<ons	
  
§  Source:	
  processmining.org	
  
Finding	
  and	
  Troubleshoo<ng	
  Bo`lenecks	
  
h`p://convergingdata.com	
  
Healthcare	
  Service	
  Delivery	
  Overview	
  
h`p://convergingdata.com	
  
Detect	
  poten<al	
  hand-­‐off	
  issues	
  
Examples:	
  Clinical	
  Decision	
  Support	
  
16	
  
	
  
•  Be`er	
  decisions	
  using	
  con<nuous	
  physiological	
  streaming	
  data	
  
•  Op<mize	
  alarm	
  and	
  alert	
  sehngs	
  in	
  devices	
  and	
  applica<ons	
  
•  Care	
  Coordina<on	
  for	
  complex	
  co-­‐morbid	
  condi<ons	
  
•  Hyper-­‐personalized	
  engagement	
  	
  
•  Crea<ng	
  checklists	
  based	
  on	
  predic<on	
  of	
  cri<cal	
  events,	
  
	
  	
  	
  	
  	
  	
  early	
  warning	
  signs.	
  
	
  
Clinical	
  Decision	
  Support	
  Center:	
  	
  
Air	
  Traffic	
  Control	
  System	
  for	
  Clinical	
  Decisions	
  	
  
•  Think	
  of	
  this	
  is	
  like	
  a	
  department	
  like	
  Radiology	
  
•  Helps	
  with	
  near	
  real	
  <me	
  evidence	
  findings,	
  implementa<ons,	
  and	
  valida<ons	
  
•  Provide	
  data	
  driven	
  opinions	
  when	
  no	
  established	
  guidelines	
  exists	
  
•  Help	
  validate	
  output	
  of	
  analy<cs	
  with	
  exis<ng	
  guidelines	
  and	
  evidence	
  from	
  clinical	
  trials.	
  
Prac<ce	
  Based	
  Evidences	
  (source:	
  greenbu`on.stanford.edu)	
  
Prac<ce	
  Research	
  
Applying	
  Evidence	
  
Genera<ng	
  Evidence	
  
19	
  
Example	
  App:	
  Care	
  Coordina<on	
  Assistance-­‐	
  Find	
  gaps,	
  redundancies,	
  
conflicts,	
  and	
  interac<ons	
  and	
  predict	
  adverse	
  events	
  
20	
  
Example	
  App:	
  Find	
  Similar	
  Pa<ent	
  Pathway	
  for	
  
Evidence	
  Based	
  Interven<ons	
  
Virtual	
  
Physical	
  
Cloud	
  
21	
  
Healthcare	
  Data	
  Is	
  Time	
  Oriented	
  and	
  Diverse	
  
EHR	
  
Systems	
  
Web	
  
Services	
  
Developers	
  
App	
  
Support	
  Telecoms	
  
Networking	
  
Desktops	
  
Servers	
  
Security	
  
Devices	
  
Storage	
  
Messaging	
  
Pa<ent	
  
Surveys	
  
Clickstream	
  
HIE	
  
Pa<ent	
  
Networks	
  
Healthcare	
  Apps	
   IT	
  Systems	
  and	
  Med	
  Devices	
   Pa,ent-­‐Generated	
  Data	
  
Medical	
  
Devices	
  
CDR	
  
Mobile	
  
	
  	
  	
  	
  PHI	
  Access	
  
Audit	
  Logs	
  
HL7	
  
Messaging	
  
Sensors	
  
Departmental	
  
and	
  
Homegrown	
  
Applica<ons	
  	
  
Disrup,ve	
  Approach	
  to	
  Diverse	
  Data	
  
What	
  Happened?	
   What's	
  Happening?	
  
Structured	
  
RDBMS	
  
SQL/Cube	
  
Schema	
  at	
  Write	
  
ETL	
  
Search	
  
Schema	
  at	
  Read	
  
Universal	
  Indexing	
  
Unstructured	
  
Volume	
  |	
  Velocity	
  |	
  Variety	
  
22	
  
What	
  Might	
  Happen?	
  
Predict/Prescribe	
  
Opera,onalize	
  
Machine	
  Learning	
  
Data	
  Analy<cs	
  Infrastructure	
  
23	
  
DATA	
  SOURCES	
  
IOT	
  DATA	
  
IT	
  DATA	
  
Acquiring	
  
Enriching	
  
(real	
  <me)	
  
In	
  Mo<on	
  Data	
  Acquisi<on,	
  Analysis,	
  and	
  Engagement	
  
(security	
  and	
  privacy	
  monitoring	
  and	
  audit)	
  
Searching	
  
Analyzing	
  
(real	
  <me)	
  
Delivering	
  
Engaging	
  
(real	
  <me)	
  
At-­‐Rest	
  Data	
  Acquisi<on,	
  Analysis,	
  Compose,	
  and	
  Deploy	
  
(security	
  and	
  privacy	
  monitoring	
  and	
  audit)	
  
	
  
APPS	
  DATA	
  
Data	
  At-­‐Rest	
  
Historical	
  Data	
  Storage	
  
Data	
  Discovery,	
  Explora<on,	
  Modeling,	
  
Evalua<on	
  
(At	
  Rest)	
  
Compose	
  and	
  Deploy	
  
(DevOps)	
  
Streaming	
  Data	
  Storage	
  
Data	
  In	
  Mo<on	
  
80%	
  of	
  healthcare	
  data	
  
in	
  unstructured	
  text	
  
High	
  velocity	
  <me	
  
series	
  data	
  from	
  
devices-­‐	
  different	
  
<me	
  zones,	
  different	
  
<me	
  intervals	
  
Variety	
  of	
  
structured	
  formats	
  
for	
  the	
  same	
  object	
  
Unit	
  of	
  Measures	
  
do	
  not	
  match	
  
Data	
  Integra<on	
  and	
  Normaliza<on	
  
24	
  
Probabilis<c	
  Methods	
  to	
  validate	
  exis<ng	
  data	
  or	
  fill	
  in	
  missing	
  data	
  	
  
Data	
  Analy<cs	
  Knowledgebase	
  
25	
  
•  Computable	
  
Care	
  Plans	
  
•  Guidelines/
Rules	
  
•  Health	
  
System	
  
Workflow	
  
•  Data	
  
Models	
  
•  Ontologies	
  
•  Treatment-­‐
Outcome	
  
data	
  
Data	
  Analy<cs	
  Methods	
  
26	
  
	
  
	
  
	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  
	
  
•  What	
  you	
  feed	
  into	
  the	
  algorithm	
  differen<ates	
  winners	
  from	
  averages.	
  
•  Sophis<cated	
  techniques	
  are	
  generally	
  worse	
  than	
  simple	
  methods.	
  
Visualiza<on	
  Search/Explora<on	
   Sta<s<cs	
  and	
  
Machine	
  Learning	
  
Sofware	
  
Engineering	
  
27	
  
Data	
  Analy<cs	
  Driven	
  User	
  Engagement	
  
28	
  
	
  
	
  
	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  
	
  
Task	
  
Bo`lenecks,	
  
Issues	
  
Knowledge	
  
Integra<on	
  
User	
  
Incen<ves,	
  
Habits	
  
Impacts	
  of	
  new	
  
knowledge,	
  
Trust	
  
Detail	
  or	
  
summary	
  or	
  
Both	
  
Responsive	
  
Adap<ve	
  
Managed	
  
People	
  have	
  priori,es	
  beyond	
  just	
  
geSng	
  treated.	
  
Courtesy:	
  DJ	
  Pa,l	
  
Lastly,	
  do	
  not	
  forget	
  Sofware	
  Engineering	
  prac<ces	
  
29	
  
•  Tes<ng	
  
•  Privacy	
  and	
  Security	
  
•  Design	
  and	
  Refactoring	
  
•  Version	
  Control	
  and	
  Provenances	
  
•  Logs	
  and	
  Documenta<ons	
  	
  
•  Produc<on	
  Deployment	
  Review	
  
 	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  	
  Summary	
  
30	
  
Aim	
  of	
  Big	
  Data	
  Analy<cs	
  is	
  to	
  help	
  make	
  op<mal	
  
decisions-­‐	
  opera<onal	
  or	
  clinical.	
  	
  
Success	
  	
  in	
  analy<cs	
  requires	
  mul<-­‐disciplinary	
  skills.	
  
Personalize	
  the	
  analy<cs	
  output	
  to	
  alter	
  current	
  
behavior/habits.	
  
Copyright	
  ©	
  2015	
  Splunk	
  Inc.	
  
Thank	
  You!	
  
31	
  
Adrish	
  Sannyasi	
  
Splunk,	
  asannyasi@splunk.com	
  
	
  

Big Data Analytics for Healthcare Decision Support- Operational and Clinical

  • 1.
    Copyright  ©  2015  Splunk  Inc.   Big  Data  Analy<cs  and  Decision  Support     Adrish  Sannyasi,     Healthcare  Solu<ons  Architect    
  • 2.
                                                                     Agenda   2   Aim  of  Big  Data  Analy<cs   Opera<onal  Decision  Support   Clinical  Decision  Support   Data  Analy<cs  Infrastructure  and  Methods  
  • 3.
    3   Triple  Aim:  Healthcare  Delivery  System   VALUE  =  Outcome  (quality,   safety,  experience)  /  Cost  
  • 4.
    Aim  of  Big  Data  Analy<cs   4   Help  Make  Op<mal  Decisions   Pa<ents   Administrators  and   Policy  Makers   Providers   Proac,ve   Precise   Predic,ve  
  • 5.
    Moving  towards  precise  decision  making   5   VS  
  • 6.
    Current  Vs.  Desired  Decision  Support   6   Current  Approach   Desired  Approach   Rule  based  system   High  rate  of  false  alarms   Missed  opportuni<es   Precise  and  Context  Sensi<ve   Workflow  Interrup<on   Automated  Data  Collec<ons   Mainly  structured  EMR  or  Claims  data   Structured  and  Unstructured  data  from  EMR,   sensors,  wearable,  behavior,  and   environmental  data,  and  condi<on  focused   social  network  data.   One  <me  measurements  of  physiological   sta<s<cs   Con<nuous  measurements    and  pathway   oriented  measurements   Low  transparency  and  accountability   Transparent  and  Accountable  to  pa<ents  and   care  team  
  • 7.
    Building  a  “Learning  and  Improvement  Engine”   7   Sympathy-­‐ Man,  We  could   do  be`er      Empathy  –  I  feel  your  pain   Compassion-­‐   Let  me  help  you   Source:  HCA  
  • 8.
    Big  Data  Analy<cs  must  be:                                                                                                                                                                     Valid:  hold  on  new  data  with  some  certainty     Useful:  Should  be  ac<onable     Unexpected:  non-­‐obvious  to  consumers     Understandable:  humans  should  be  able  to  interpret     Measurement  is  useful  if  it  facilitates  ac,on.     Measure  what  is  important  to  customer.    
  • 9.
    Examples:  Opera<onal  Decision  Support   9   •  Scheduling  and  Staffing  Assistance   •  Predic<ng  and  alloca<ng  service  area  and  unit  capacity   •  Predic<ng  bed/room  requests,  LOS  targets,  transport  services     •  Op<mizing  Asset/Inventory  U<liza<on     •  Reducing  claims  processing  cost,  error,  and  <me   •  Reducing  Fraud,  Waste,  and  Abuse                                    
  • 10.
    Opera<ons  Decision  Support  Center:     Air  Traffic  Control  System  for  Opera<onal  Decisions     •  Modeled  afer  your  security  opera<ons  center  and  IT  opera<ons  center   •  Track  pa<ent  movements  and  oversee  opera<ons  and  throughput   •  Proac<vely  an<cipate  needs  for  services   •  Coordinate  staffing  and  scheduling   •  Coordinate  admissions,  transfers,  discharge  planning  and  execu<on   •  Reduce  cross  departmental  hand-­‐off  issues  
  • 11.
    Informa<on  Flow  in  Care  Delivery:  Spagheh     h`p://www.ncbi.nlm.nih.gov/pmc/ar<cles/PMC3002133/  
  • 12.
    From  Spagheh  To  Lasagna:   Reduce  Unwarranted  Varia<ons   §  Source:  processmining.org  
  • 13.
  • 14.
  • 15.
  • 16.
    Examples:  Clinical  Decision  Support   16     •  Be`er  decisions  using  con<nuous  physiological  streaming  data   •  Op<mize  alarm  and  alert  sehngs  in  devices  and  applica<ons   •  Care  Coordina<on  for  complex  co-­‐morbid  condi<ons   •  Hyper-­‐personalized  engagement     •  Crea<ng  checklists  based  on  predic<on  of  cri<cal  events,              early  warning  signs.    
  • 17.
    Clinical  Decision  Support  Center:     Air  Traffic  Control  System  for  Clinical  Decisions     •  Think  of  this  is  like  a  department  like  Radiology   •  Helps  with  near  real  <me  evidence  findings,  implementa<ons,  and  valida<ons   •  Provide  data  driven  opinions  when  no  established  guidelines  exists   •  Help  validate  output  of  analy<cs  with  exis<ng  guidelines  and  evidence  from  clinical  trials.  
  • 18.
    Prac<ce  Based  Evidences  (source:  greenbu`on.stanford.edu)   Prac<ce  Research   Applying  Evidence   Genera<ng  Evidence  
  • 19.
    19   Example  App:  Care  Coordina<on  Assistance-­‐  Find  gaps,  redundancies,   conflicts,  and  interac<ons  and  predict  adverse  events  
  • 20.
    20   Example  App:  Find  Similar  Pa<ent  Pathway  for   Evidence  Based  Interven<ons  
  • 21.
    Virtual   Physical   Cloud   21   Healthcare  Data  Is  Time  Oriented  and  Diverse   EHR   Systems   Web   Services   Developers   App   Support  Telecoms   Networking   Desktops   Servers   Security   Devices   Storage   Messaging   Pa<ent   Surveys   Clickstream   HIE   Pa<ent   Networks   Healthcare  Apps   IT  Systems  and  Med  Devices   Pa,ent-­‐Generated  Data   Medical   Devices   CDR   Mobile          PHI  Access   Audit  Logs   HL7   Messaging   Sensors   Departmental   and   Homegrown   Applica<ons    
  • 22.
    Disrup,ve  Approach  to  Diverse  Data   What  Happened?   What's  Happening?   Structured   RDBMS   SQL/Cube   Schema  at  Write   ETL   Search   Schema  at  Read   Universal  Indexing   Unstructured   Volume  |  Velocity  |  Variety   22   What  Might  Happen?   Predict/Prescribe   Opera,onalize   Machine  Learning  
  • 23.
    Data  Analy<cs  Infrastructure   23   DATA  SOURCES   IOT  DATA   IT  DATA   Acquiring   Enriching   (real  <me)   In  Mo<on  Data  Acquisi<on,  Analysis,  and  Engagement   (security  and  privacy  monitoring  and  audit)   Searching   Analyzing   (real  <me)   Delivering   Engaging   (real  <me)   At-­‐Rest  Data  Acquisi<on,  Analysis,  Compose,  and  Deploy   (security  and  privacy  monitoring  and  audit)     APPS  DATA   Data  At-­‐Rest   Historical  Data  Storage   Data  Discovery,  Explora<on,  Modeling,   Evalua<on   (At  Rest)   Compose  and  Deploy   (DevOps)   Streaming  Data  Storage   Data  In  Mo<on  
  • 24.
    80%  of  healthcare  data   in  unstructured  text   High  velocity  <me   series  data  from   devices-­‐  different   <me  zones,  different   <me  intervals   Variety  of   structured  formats   for  the  same  object   Unit  of  Measures   do  not  match   Data  Integra<on  and  Normaliza<on   24   Probabilis<c  Methods  to  validate  exis<ng  data  or  fill  in  missing  data    
  • 25.
    Data  Analy<cs  Knowledgebase   25   •  Computable   Care  Plans   •  Guidelines/ Rules   •  Health   System   Workflow   •  Data   Models   •  Ontologies   •  Treatment-­‐ Outcome   data  
  • 26.
    Data  Analy<cs  Methods   26                                       •  What  you  feed  into  the  algorithm  differen<ates  winners  from  averages.   •  Sophis<cated  techniques  are  generally  worse  than  simple  methods.   Visualiza<on  Search/Explora<on   Sta<s<cs  and   Machine  Learning   Sofware   Engineering  
  • 27.
  • 28.
    Data  Analy<cs  Driven  User  Engagement   28                                       Task   Bo`lenecks,   Issues   Knowledge   Integra<on   User   Incen<ves,   Habits   Impacts  of  new   knowledge,   Trust   Detail  or   summary  or   Both   Responsive   Adap<ve   Managed   People  have  priori,es  beyond  just   geSng  treated.   Courtesy:  DJ  Pa,l  
  • 29.
    Lastly,  do  not  forget  Sofware  Engineering  prac<ces   29   •  Tes<ng   •  Privacy  and  Security   •  Design  and  Refactoring   •  Version  Control  and  Provenances   •  Logs  and  Documenta<ons     •  Produc<on  Deployment  Review  
  • 30.
                                                                     Summary   30   Aim  of  Big  Data  Analy<cs  is  to  help  make  op<mal   decisions-­‐  opera<onal  or  clinical.     Success    in  analy<cs  requires  mul<-­‐disciplinary  skills.   Personalize  the  analy<cs  output  to  alter  current   behavior/habits.  
  • 31.
    Copyright  ©  2015  Splunk  Inc.   Thank  You!   31   Adrish  Sannyasi   Splunk,  asannyasi@splunk.com