Wolters Kluwer Improves Patient Outcomes with GigaSpaces XAP
Download as PPTX, PDF1 like750 views
The document discusses the development of a big data solution aimed at improving patient outcomes through real-time surveillance in healthcare, highlighting the importance of monitoring and intervention. It emphasizes the scalability challenges in processing vast amounts of data from hospitals while aiming for rapid delivery of insights to clinicians. The integration of technologies like Gigaspaces and Drools is presented as a solution to enhance system performance and responsiveness in clinical decision support.
Wolters Kluwer Improves Patient Outcomes with GigaSpaces XAP
- 1. Improving Patient Outcomes through Big Data Real Time Surveillance Paul Kudrle Lead Developer – Wolters Kluwers Clinical Solutions
- 2. 1 • Introduction • Background • Big data solution • Challenges • Future Agenda
- 3. 2 • Master’s Degree is Computer Science from Cal-Poly Pomona • NASA • Systems Engineer – deep space missions • Consultant (big data on the small scale) • Nano-tech – atom probe microscope • Bio-tech - DNA analysis • Lead Engineer and Team Lead for Pharmacy One Source – Wolters Kluwer My background
- 4. 3 Pharmacy One Source Nine SAAS healthcare applications • 1 in 3 hospitals in the U.S. utilize one or more of our applications • Active community of > 44,000 of pharmacy professionals and hospital clinicians • Purchased by Wolters Kluwers in 2010-11 and now part of Wolters Kluwers Clinicial Services
- 6. What do we mean by… Surveillance: Actively monitoring patients to identify opportunities for intervention Intervention: changing patient care to improve outcomes Outcomes: Quality & Safety Efficacy Efficiency Cost effectiveness 5
- 7. SaaS Model • We receive data from hospitals via HL7 standard message format and support 500+ hospitals • End users access through browser based client apps and mobile devices • Clinicians – Infection Prevention • Pharmacists • We manage our own hardware and data centers • Move to cloud potentially in the future • HIPPA 6
- 8. Electronic Medical Records (EMR) 7 Visit Surgeries EMR Observations Patient Clinician Lab technician Medications Diagnoses Vitals
- 9. Request for data from EMR • Clinician/Pharmacists wants to find patients… • Too many drugs or wrong combination of drugs • Infection not responding to drug for a specific patient • IT – create report • Overworked and understaffed • Batch systems • EMR • Long delivery times • Batch systems • Speed is of the essence for results 8
- 10. Enhancing the EMR 9 EMR Clinician •Activating data •Intelligent Decisions
- 11. Sentri7 surveillance rule builder 10
- 12. 11 - Pull based user centric rule evaluation - End user required to hit a web page to find results. - If they don’t log in then results are not changed - SQL based query executions - SQL stored procedures - Expensive big box machines - Lots of processors - Database locking - Data storage/persistence layer bottlenecks - Difficult to scale individual sites to meet their demands - A lot of smaller sites impacted by one larger hospital Legacy Rule Engine
- 13. Perception of responsiveness 1. Speed is everything—this is what information system users value most. 2. Anticipate needs and deliver in real time—deliver information when needed. -“Ten Commandments for Effective Clinical Decision Support” American Medical Informatics Association, 2003 12
- 14. Impact of Sepsis on Patient Health 13 Angus DC et al. Crit Care Med 2001;29(7):1303-1310. Patient Lives • > 200,000 deaths per year in the US • Mortality from severe sepsis is between 28-50% (with standard care) Hospital Costs • $16.7 billion in total annual U.S. costs • Average cost = $22,100 - $29,900 • Average LOS = 19.6 - 23.3 days
- 15. At very high risk for severe sepsis with shock 14 Group A -Pulse > 90 -Resp < 20 -PaCO2 < 32 mmHg -Temp not in the range 96.8 F to 100.4 F Group B -WBC Count not in the range 4 thou/uL to 12 thou/uL -Band > 10 % Group C -Lactic Acid >= 3.5 mmol/L -Serum Creatinine > 2 mg/dL -Estimated Creatinine Clearance < 50 ml/min -Platelet count < 80 thou/uL -Active Drug Order where Drug Name is not in the list ("WARFARIN SODIUM", "ENOXAPARIN SODIUM", "HEPARIN SODIUM, PORCINE") AND has one of the following: -PT (INR) > 1.5 -aPTT > 60 sec -BILIRUBIN TOTAL > 2 mg/dL AND ALT (SGPT) >114 U/L -VENOUS O2 SATURATION (VO2HB) < 70% -pH < 7.30 -PaO2 < 80 mmHg -pH < 7.35 AND PaCO2 < 50 mmHg Group D -SYSTOLIC BP < 90 -SYSTOLIC BP has decreased 28% Show all patients that have at least one match from each of the following groups in the last 24 hours.
- 16. Meet the scalability challenge to support real-time results • Every day we process over 8 million messages supporting over 1,000,000 active patients • Run 35,000 rules in real-time • Drug orders • Vitals data • Lab data • Textual observations • Surgery data • Combinations of any of these • Our architecture needs to be able to scale beyond 15 million HL7 messages a day • Enable our rules to process the messages and generate results back to the client in less than 10-secs, regardless of complexity Real-time results vs big data challenges 15
- 17. Focus on ambitious goals aligned with additional business needs Be cost-effective: • Increase production capacity but save Capex • Drive down our cost per hospital • Create a bridge to the cloud to get us out of the hardware business • We could not overhaul all of our code base to a completely new paradigm • Need something that could sit on top of existing systems 16Confidential
- 18. • Hadoop / MapReduce is a framework for processing huge datasets on certain kinds of distributable problems using a large number of computers (nodes) • Pros • Designed to process massive amounts of data • Allows for distributed processing of map and reduce operations • Mature • Low cost • Growing and growing business and development community • Cons • Not real-time • Batch processing Hadoop 17
- 19. • Could not find a single bullet that could solve all our needs and solutions • A combination of technologies was needed • Poll/Pull based solutions can be difficult to scale especially over shared resources • SQL Server based solutions are not ideal as we want to remove load off the main servers and have them be repositories and move toward read only replicas for client applications Solution(s) 18
- 20. GigaSpaces XAP – Elastic Application Platform 19 DATA PROCESSING EVENTS & MESSAGING WEB APPLICATION SUPPORT MANAGEMENT & MONITORING HIGH AVAILABILITY ELASTICITY CLOUD READINESS
- 21. GigaSpaces – a Commercial Java Spaces Implementation • GigaSpaces extends Java Spaces adding processing units concept: deployable code units & execution. • Grid = data + code • Data in the grid • Rules in the grid • Elastic grid, redundancy, real- time self-healing, load balancing 20Confidential
- 22. GigaSpaces • No centralized point of failure • Fast in-memory data grid for collocating data and processing • Self-healing grid • No emergency calls on Saturday evening while 4 hours away from a computer 21
- 23. Drools • Embedded rule engine based in Java • Business rules for insurance, finance, etc • Expert System • Allows for “Hot” deployable code – add/remove code in real-time • Utilizes Rete Algorithm • Shared processing path between similar rules • Promotes speed at the expense of memory • Lots of companies build their own rule engine • All suffer performance issues that Drools and Rete algorithm has solved • Drools developers are domain experts in rule systems and continually evolving the platform • Typically they are brute force • Temporal Reasoning • Time based rules 22
- 24. GigaSpaces + Drools 23 Performance Collocate rules with the data allowing extreme low latency rules execution Rules Management Dynamically load/unload rules leveraging GigaSpaces & Drools APIs Ultra Scalable Parallel rules execution across the different data-grid partitions Elasticity Scale- up/down, in/out system and leverage extra resources on-demand within private cloud or public cloud. High-Availability Continuous availability running rules 24X7 without any downtime
- 26. Massively Parallel Expert System 25 GigaSpaces Grid + Drools Surveillance Results
- 27. Results support our goals • On a single feed our processing jumped from 30 mps to >130 mps • Hospital cannot send the data fast enough • 95% of our clients have moved onto the platform in 6 months • Average response time for client surveillance rules to display in the user interface reduced by 98% 26
- 28. Results (average time to run rules) 27 0 2000 4000 6000 8000 10000 12000 6/23/13 6/25/13 6/27/13 6/29/13 7/1/13 7/3/13 7/5/13 7/7/13 7/9/13 7/11/13 7/13/13 7/15/13 7/17/13 7/19/13 7/21/13 7/23/13 7/25/13 7/27/13 7/29/13 7/31/13 8/2/13 8/4/13 8/6/13 8/8/13 8/10/13 Timeinmilliseconds Hospital A Hospital B Hospital C Hospital D Hospital E
- 29. Real-time impact • We receive an HL7 message from a Hospital we are able to run that specific data change against 300+ rules for a hospital • 300 rules processing in under 30 milliseconds • Some rules never worked under old system (timed out) • Pharmacists and Clinicians know the exact time a patient qualified for a rule. • We can tell them why! • All rules are constantly running for all patients • Average 1-2K patients per hospital • Growing to 9-10K patients for larger sites • Expect growth • Some patients will qualify over time without receiving any updates (lab not ordered) 28
- 30. New product horizons • Notifications • Patient qualifies for a rule -> pager, SMS, email notification sent • Pharmacist notified of a patient on too many drugs or the wrong combination • Sepsis benefits • Speed matters • <1 hour response time needed • Deliver decision support to the point of care • Integrate solutions to our other Wolters Kluwers Clinicial Services solutions • Up to Date supplies content • Health Language supplies industry standardization • Better benefits of integration • Everyone wants in 29
Editor's Notes
- #3: Nano-tech succeeded, Bio-tech failed
- #4: Clinicial Services division – Health Language – Standards mapping , Up to Date – expert recommendations on
- #5: Flagship application used for Clinical Decision Support.Surveillance -> monitor patients clinician care aboutIntervention -> a change to patient care based on surveillance -> too many drugsDocumentation -> document interventions to demonstrate cost savings, ROIReporting -> NIH reporting, Drug Resistant Organisms
- #7: Clinical Decision SupportHistorical Needs – Hippa data
- #8: Hospitals systems are often disparate and with no data integration or data integration is expensivePharmacists and Clinicians have a difficult time trying to find the data that will help them monitor their patients effectively and focus on improving their care
- #9: Talk about the lifecycle, how long before they clinician could see there result of a report. How often could it get generated etc.EPIC is in our back yard and we work with a lot of sites that have EPIC as their EMR. EMRs always promise there EMR can deliver this value. They can’t do it as fast, they are more expensive, they can’t do it well for the people that need the information
- #10: The Sentri7 application fills the gap for the clinicians and the EMR. We take that data and make it accessible to the Clinicians and help them make intelligent decisionsDescribe the data we get – secure connection from the EMR receiving HL7 Messages.
- #11: We give the pharmacists and clinicians the ability to write the rules through a catalog of rules we provide or through writing rules themselvesA rule builder interface so that pharmacists and clinicians can find the patients that need interventions.New drug being utilized on the market and the pharmacists needs to track the patients that are on it. They can’t wait for updates from IT, developers, EMR in order to handle this scenario. They need it on Monday.A new organism has been identified as being resistant to a specific drug from the CDC. The clinician needs to add a rule to monitor those organisms and drug resistance. They can’t wait for updates from service providers. They need to monitor now. Our software has found its niche because we put the tools to find this data in the hands of the clinicians
- #12: Start up company utilizing technology to deliver customers value. Realized after a couple years that we needed to scale horizontally. There are other challenges out there impacting the design
- #13: A paper studying the impact of clinical decision support systems with recommendations of how to make evidence based medicine a reality.Business requirementOur old system was not up to the taskThis was a direction the business felt we needed to head in. Speed Saves Lives!
- #14: Here is one example of how “Speed saves lives”Get to the patient before Sepsis escalates, early prevention is key!Speed is of the essence in Sepsis
- #15: Sepsis – severe body infection that can lead to organ dysfunction.Data was very complex for our query engine to reason over. As data volume grew this type of rule would get more difficult to run.Time is of the essenceOften rules of this complexity would not be able to return data in time > 90 seconds. Imagine a user sitting there watching a ball spin waiting for data to be returned in over 90 seconds.
- #16: Rule that was timing out before cannot time out!
- #20: In Memory data grid, scalable solution, add machines horizontally to scale solution, speed!! Fast processing of data
- #21: GigaSpaces extends JavaSpaces adds processing units concept: deployable code units & execution. Grid = data + codeAdding Drools to GigaSpaces: Embeddable rule engine, Hot deployable code in DRL, Drools knowledgebase allows declarative rules to be pre-compiled for fast executionJava Spaces: an implementation of the associative memory paradigm for parallel/distributed computingGigaSpaces better for no single point of failureGigaSpaces support for .NET, Java, C++ and other language platformsGigaSpaces a leader and innovator in the Elastic Caching platforms
- #25: Explain PartitionsExplain Drools, Working Memory, and mention Stateful SessionHydra is message processing + Rules evaluation (using Drools)
- #26: We have put the data in the hands of the pharmacists and clinicians who need it. If they have a specific drug they need to monitor or there is some specific uniqueness to their hospital that they need to monitor they can.