S A V S N E T. Mining data from veterinary records Real-5me veterinary surveillance in companion animals. Phil Jones

Transcription

1 S A V S N E T The Small Animal Veterinary Surveillance Network The Small Animal Veterinary Surveillance Network 1 1 Mining data from veterinary records Real-5me veterinary surveillance in companion animals Phil Jones University of Liverpool, Department of Epidemiology and Population Health, Institute of Infection and Global Health, Leahurst Campus, Neston, Cheshire, CH64 7TE. United Kingdom.

2 The Brief How SAVSNET surveillance system works. What data is collected. How it is collected, especially the extraccon from praccce and laboratory informacon. How the data can be used and its value. Applying to the pig industry. LimitaCons and how they have been addressed for the following challenges: Data ConfidenCality, StandardisaCon, Good coverage, Cost, Real-Cme reporcng in/out.

3 S A V S N E T The Small Animal Veterinary Surveillance Network Collec5ng data How SAVSNET surveillance system works. What data is collected. How it is collected, especially the extraccon from praccce and laboratory informacon.

4 SAVSNET structure Project 1 Commercial laboratory diagnoscc data Project 2 Real-Cme electronic health record data from veterinary praccces Data colleccon, collacon and analysis Research-ready data made available through Data Access and PublicaCons Panel (DAPP) (see for more details)

5 Veterinary praccce data Project 2 Real-time electronic health record data from veterinary practices Every participating veterinary practice has been actively recruited Signed agreement

6 Ethical & professional approval Research Ethics Committee : ALL University research projects (including undergraduate projects) must have ethical approval. Aims supported by :

7 Veterinary praccce data In the waicng room Assumes owners have an opportunity to view poster. See later Consent by optout Approved by the University of Liverpool s Research Ethics Committee. Supported by RCVS.

8 Veterinary praccce data In the consultacon room

9 Veterinary praccce data At the end of the consultacon 1. Please indicate the clinical signs present* 100% 10% Diarrhoea Small intestinal diarrhoea Large intestinal diarrhoea/colitis Haemorrhagic diarrhoea Melaena Vomiting without blood Vomiting with blood Weight loss/failure to gain weight Poor appetite Other 2. Please indicate disease severity Mild illness i.e. normal apart from GI disease Moderately ill Severely ill/debilitated 3. How does this consultation relate to this episode of illness First presentation Revisit/check-up Don t know 4. How long approximately has the pet had this episode of illness? Average time to complete Less than 1 week = 6-7 seconds Between 1 week less than 1 month 1 month and over Don t know 5. What diagnostic options will be used today for this episode of illness? * None Faecal parasitology/bacteriology Faecal virology Virus serology Diagnostic Imaging Haematology/biochemistry Serum B12/Folate and/or serum TLI Canine/feline specific pancreatic lipase Urinalysis Other The SAVSNET window 6. What advice did you give today? * Change of diet Fasting Admit patient for treatment Refer patient Check-up in near future Other *indicates multiple options may be chosen

10 Veterinary praccce data The two-way informacon highway Content within SAVSNET window is not hard-coded. PMS provides modal empty iframe at end of consultation SAVSNET works closely with practice management system (PMS) providers: SAVSNET servers provide content

11 Distribution of SAVSNET practices (14 Sept 2016) Where is SAVSNET?

12 Survey alloca5on Survey allocacon

13 How much data?

14 Veterinary praccce surveillance What data do we collect? Financial data Veterinary surgeon ID Owner name and address IdenCficaCon Geography Signalment Historical Miscellaneous Treatments Freetext The SAVSNET window Date of consultacon PracCce ID Animal ID Owner s full postcode Species Breed Date of birth Gender Neutering status VaccinaCon history Weight history Insurance status Microchip status and number (if present) Prescribed drugs and other treatments Clinical notes Q1 QuesConnaire (if administered)

15 Laboratory diagnoscc data Project 1 Commercial laboratory diagnostic data Laboratories are recruited as data providers SAVSNET accepts data in most convenient format for the laboratory

16 Laboratory diagnoscc surveillance Overview Minimal imposicon Currently submi?ng data? What data do we request? Sample ID Species Breed Full postcode of veterinary praccce Sample type Assay type and method Result InterpretaCon (if needed) PTDS POWELL TORRANCE DIAGNOSTIC SERVICES PRACTICE SUP THE CHALLENGE different terms used by different labs LAB 1: / canine / wes5e / CH / faeces / parvo PCR / nega5ve LAB 2: A342 / dog / WHWT / CH / faecal sample / PCR CPV / - VetXML schema somecmes! Unified database with consistent terms

17 Recent figures Lab-based project Total number of enrolled laboratories 11 (signed-up) Number of labs contribucng data 8 Number of test results 42M + Number of samples 4.5M + Prac5ce-based project Total number of contribucng veterinary praccces 223 Number of individual premises / clinics 489 Total number of consultacons 1.5M + Current consultacons per day (Mon Fri) 5,500 6,000

18 S A V S N E T The Small Animal Veterinary Surveillance Network How do we use the data? How the data can be used and its value.

19 Using the data 1. Data summaries and benchmarking in context for data contributors (laboratories and veterinary praccces) We don t pay for data but we benchmark for free Secure portal login 2. Quarterly surveillance reports in veterinary literature 3. Research-ready data available to researchers through an easy applicacon process. 4. General public online access to data summaries

20 Benchmarking for providers CompleCng the circle(s) SAVSNET Providing value Data quality Engagement Inform clinical decisions labs SAVSNET Impact vets owners benchmarking www

21 We don t pay for data but Secure real-cme anonymised praccce portal PRACTICE VALUE BOX & WHISKER OF ANONYMISED PEERS Maps showing consult location Coloured areas are published deprivation MADE UP DATA INTERACTIVE CHARTS TO ANALYSE SPECIES, BREED, GENDER, AGE, NEUTER, INSURANCE AND MICROCHIP DAILY BREAKDOWN OF EACH SYNDROME / CONSULTATION TYPE

22 Laboratory diagnoscc surveillance InteracCve benchmarking report Where samples come from Where TESTS come from Breakdown by species Dogs Cats Test result by day negative day positive What tests are done in combination toxoplasma entire cat male Eg proportion of toxoplasma negative results from male neutered cats toxoplasma

23 A resource for research Data access and publicacon panel (DAPP) BSAVA / UoL / Independent Secure data transfer tools Agreements DAPP SAVSNET Research SAVSNET OUTPUTS Academia UoL, UoB Surveillance Veterinary Record Commercial Pharma Government VMD EXTERNAL DATABASES labs Geographical Census Financial Climatic Agricultural vets owners benchmarking www Fees for data access used for sustainability and further development only.

24 Data access Data access available for all researchers Data Access and PublicaCons Panel BSAVA University of Liverpool External reviewers Simple applicacon SAVSNET Reference Number Application for the use of SAVSNET data Please refer to guidance notes in completing this form. We recommend that you have an informal discussion with one of the SAVSNET team before making your application. Details of applicant(s) Lead applicant s name Names of other people who will have access to SAVSNET data Affiliation (Company / University / Organization or other) Purpose for requesting data Purpose for requesting data Question being asked process Access period Data required Type of data Number of consultations / laboratory reports Date range for data Species Breed Age range Any other requirements Page 1

25 S A V S N E T The Small Animal Veterinary Surveillance Network Dealing with challenges Data ConfidenCality, StandardisaCon, Good coverage, Cost, Real-Cme reporcng in/out.

26 Challenges 1. Data confiden5ality (need to inspire confidence) Ethical approval & support from RCVS Owner consent Signed agreements (e.g. commercially-sensicve data not disclosed) UoL secure servers De-idenCficaCon sooware DAPP and data chaperone Secure portals for data providers Avoiding combinacons of idencfiers ( the 19yo widow in Manchester ) Independence of University Close links with Farr InsCtute of Health InformaCcs Research

27 Challenges 2. Standardisa5on Development of API (applicacon programming interface) Integrate data from mulcple PMS sooware 3. Good coverage IniCally worked with market-leading PMS providers User experience of SAVSNET window paramount Intense recruitment drives Work with mulcple PMS providers

28 Challenges 4. Cost Majority of inical funding from large grants AddiConal funding from commercial / academic researchers All funds used for sustainability We don t pay for data but we benchmark for free 5. Real-5me repor5ng in/out Team Academic staff 3.5 FTE core team (inc. sooware developer and epidemiologist)

29 S A V S N E T The Small Animal Veterinary Surveillance Network Beyond the cute and cuddly Poten5al to apply SAVSNET technology to other veterinary sectors Applying to the pig industry.

30 Value and availability of data SAVSNET has shown that: Lab data can be collected Vet data accessible through sooware ProducCon data could provide important insights Data already exists but under-uclised Enhance current surveillance Economical Technically feasible Quality of data Labs Vets Producers Volume of data Pyramid of scrutiny (Meah and Lewis, 1999)

31 Building on SAVSNET experience ApplicaCon for ethical approval Contracts and agreements UoL legal RecruiCng veterinary praccces Pre-exisCng relaconships with PMS providers Laboratory data colleccon Specific rules need to be developed but technical principles established Real-Cme benchmarking for data contributors Secure data-handling and data confidencality DAPP applicacon process for data access StaCsCcal analyses Natural language processing of EHRs Financial models for long-term sustainability

32 Challenges 1. Ethics and owner consent Consent-by-opt-out Assumes owners have opportunity to read poster before consultacon Geographical markers (e.g. postcode) may be unique idencfiers in rural seqng Agricultural data may be more sensicve Therefore, consent-by-opt-out is NOT appropriate for farm vets Need to develop new model for consent Workable and meets REC approval Recruit veterinary praccces? Recruit farmers (consent-by-opt-in)

33 Challenges 2. Mobile electronic health records Pilot studies at UoL suggest a change in veterinary culture may be required to obtain farm-animal EHRs Collaborate with commercial third party BUT many commercial companies are developing mobile apps Mobile device Veterinary practice PMS SAVSNET ethos regarding not paying for data, free benchmarking and accessibility to data must be observed UoL secure servers Dealing with groups rather than individuals

34 Challenges 3. Owner data In companion animals, EHRs held almost exclusively by veterinary praccces BUT that is likely to change! Vast amounts of produccon data is roucnely collected for management purposes Can owner data be integrated?

35 EA01 EA02 EA03 EA04 EA05 EA06 EA07 EA08 EA Number of piglets Farrowing date Farm EA10 (Moving averages) Piglets born live per litter on 10 farms Piglets born live per litter on 10 farms Number of piglets Number of piglets Number of piglets Farm EA Farrowing date Farm EA Farrowing date Farm EA Farrowing date Number of piglets Number of piglets Number of piglets Farm EA (Moving averages) 2009 Farrowing date Farm EA Farrowing date Farm EA Farrowing date Preliminary Number of piglets Number of piglets Farm EA Farrowing date Farm EA Farrowing date farm-based data Harriet Bell, (undergraduate vet student) Number of piglets Number of piglets Farm EA Farrowing date Farm EA Farrowing date Number of piglets mean of liveborn ea ea ea03 Challenges 3. Owner data of piglets Farrowing date Moving average for all farms Number of live piglets born per litter Number of live piglets born per litter Moving average for all farms 15 mean liveborn ea01 ea02 ea03 ea04 ea05 ea06 ea07 ea08 ea09 ea10 0Mean 510 of piglets born of live per litter by parity category on each farm to 6 7 & over Parity number Mean number of piglets born live per litter by parity category on each farm ea ea05 4 to 6 ea Farrowing date Parity number ea07 2 to 3 ea08 7 & over 2010 ea ea10

36 Challenges 4. IniCal funding during development phase CompeCCon for funding is intense! Team of people can be expensive Database developer Sooware developer Epidemiologist Project co-ordinator Admin support BUT can pick the barles and grow incrementally

37 Challenges 5. Trust and compeccon for data Commercially-sensiCve data Need to inspire confidence in process If data has a value, why give it to us? Track record Transferrable technologies available Independent data curacon BUT Full value of data is only realised when combined with data from across the whole sector

38 Conclusions Data mining of veterinary ERH for surveillance and research is feasible. Many of the methods, experiences and technologies developed for SAVSNET are transferrable to other veterinary sectors. BUT collacng farm data (veterinary praccce / producer data) presents some unique (but not insurmountable) challenges.

39 S A V S N E T The Small Animal Veterinary Surveillance Network Who keeps SAVSNET running? Meet the core team

40 The SAVSNET Team Core team Ros Gaskell Susan Dawson Maya Wardeh Tarek Menacere The software developer Bethany Heayns Project co-ordinator SAVSNET Board of Directors Fernando Sanchez-Vizcaino Veterinary epidemiologist Susan Bolan Administrator Post-graduate research students Michael Day Sally Everitt Katie McConnell Alan Radford P-J Noble Phil Jones Jenny Newman David Singleton

41 S A V S N E T The Small Animal Veterinary Surveillance Network CAPITAL