Friday Lunchtime Lecture: Why Anonymity Fails - With Professor Ross Anderson
Uploaded by
Open Data InstituteFriday Lunchtime Lecture: Why Anonymity Fails - With Professor Ross Anderson
Uploaded by
Open Data InstituteWhy anonymity fails
Ross Anderson Cambridge University
Open Data Institute, 4/4/2014
Synopsis
Health data are moving to the cloud, causing serious tussles over safety and privacy The extension of the open data idea to healthcare is now a slow-motion train wreck Everyone from drug companies to insurers want access to masses of personal data Yesterday: we learn that HSCIC gave hospital episode statistics data to over 1000 firms Patients can often be easily identified
Open Data Institute, 4/4/2014
Do patients want access to records?
Google Health discontinued in 2011 after four years trying to build a health platform Microsoft equivalent turned into a business platform for hospitals etc Healthspace, a project to provide patient access to summary records, had only a handful of users The penny drops: most people are healthy 95% of the time and not interested in looking at record When sick they mostly have other priorities
Open Data Institute, 4/4/2014
Big Pharma certainly wants access
In 1998 a startup (DeCODE) offered Icelands health service free IT systems in return for access to records for research Funding was from Swiss drug company Roche Records to be de-identified by encrypting the social security number, but would be linked to genetic, family data Icelandic Medical Association got 11% of citizens to opt out Eventually the supreme court ruled the system should be opt-in, and the scheme collapsed
Open Data Institute, 4/4/2014
European case law
European law based on s8 ECHR right to privacy, clarified in the I v Finland case Ms I was a nurse in Helsinki, and was HIV+ Her hospitals systems let all clinicians see all patients records So her colleagues noticed her status and hounded her out of her job Finnish courts wouldnt give her compensation but Strasbourg overruled them Now: we have the right to restrict our personal health information to the clinicians caring for us
Open Data Institute, 4/4/2014
Recent UK history
Tony Blair ordered a National Programme for IT in the NHS in 2002 Idea: replace all IT systems with standard ones, giving a single electronic health record with access for everyone with a need to know This became the biggest public-sector IT disaster in British history Billions wasted, suppliers dropped out, huge lawsuits, and the flagship software didnt work
Open Data Institute, 4/4/2014
Scope Creep
Weve had big tussles over shared care E.g. giving social workers access to GP records in Oxford has made young mums there reluctant to discuss post-natal depression Lobbying win: after the 2010 election, we killed the childrens databases designed to share data between health, school, probation and social work (Database State, Munro review) The NHS Information Centre now wants to revive the idea, but under its control
Open Data Institute, 4/4/2014
Public Opinion
2,231 adults asked October 2006 on central records database with no opt out: strong support 12% tend to support 15% neither 14% tend to oppose 17% strongly oppose 36% dont know 6% Several surveys since say the same: most dont want wide sharing, or research use without consent And theres the Catholic Bishops Conference
Open Data Institute, 4/4/2014
Secondary Uses
Cameron policy announced January 2011: make anonymised data available to researchers, both academic and commercial, but with opt-out Wed already had a laptop stolen in London with 8.63m peoples anonymised records on it In September 2012, CPRD went live a gateway for making anonymised data available from (mostly) secondary care (now online in the USA!) From this year, GPES hoovering up GP stuff So: how easy is it to anonymise health records?
Open Data Institute, 4/4/2014
Advocating anonymisation
Open Data Institute, 4/4/2014
Inference Control
Also known as statistical security or statistical disclosure control Started about 1980 with US census Before then only totals & samples had been published, e.g. population and income per ward, plus one record out of 1000 with identifiers removed manually Move to online database system changed the game Dorothy Denning bet her boss at the US census that she could work out his salary and won!
Open Data Institute, 4/4/2014
Inference Control (2)
Query set size controls are very common. E.g. in New Zealand a medical-records query must be answered from at least six records Problem: tracker attacks. Find a set of queries that reveal the target. E.g for our female profs salary
Average salary professors Average salary male professors
Or even these figures for all non-professors! On reasonable assumptions, trackers exist for almost all sensitive statistics
Open Data Institute, 4/4/2014
Inference Control (3)
Contextual knowledge is really hard to deal with! For example in the key UK law case, Source Informatics (sanitised prescribing data):
Week 1 Week 2 Week 3 Week 4 Doctor 1 Doctor 2
Doctor 3
Open Data Institute, 4/4/2014
17 20
18
21 14
17
15 3
26
19 25
17
Inference Control (4)
Perturbation add random noise (e.g. to mask small values) Trimming to remove outliers (the one HIV positive patient in Chichester in 1995) We can also use different scales: practice figures for coronary artery disease, national figures for liver transplants Random sampling answer each query with respect to a subset of records, maybe chosen by hashing the query with a secret key
Open Data Institute, 4/4/2014
Inference Control (5)
Modern theory: differential privacy (pessimistic) Practical problem in medical databases: context Show me all 42-yo women with 9-yo daughters where both have psoriasis If you link episodes into longitudinal records, most patients can be re-identified Add demographic, family data: worse still Active attacks: worse still (Iceland example) Social-network stuff: worse still Paul Ohms paper: Broken Promises of Privacy
Open Data Institute, 4/4/2014
CPRD
The clinical practice research datalink, run by the MHRA, makes some data available to researchers (even to guys like me :-) Freedom of information request for the anonymisation mechanisms Answer: sorry, that would undermine security Never heard of Kerckhoffs? Search for me, cprd on whatdotheyknow.com
Open Data Institute, 4/4/2014
Next problem care.data
The PM promised in 2011 our records would be anonymised, and wed have an opt out The Secretary of State for Health, Jeremy Hunt, assured us in March 2013 that existing opt-outs would be respected In July this was reversed by the NHS England CIO NHS opt-outs are like Facebooks: the defaults are wrong, the privacy mechanisms are obscure, and they get changed whenever too many people learn to use them
Open Data Institute, 4/4/2014
The row over HES
Hospital Episode Statistics (HES) has a record of every finished consultant episode going back 15 years (about a billion in total) Mar 13: formal complaint to ICO that PA put HES data in Google cloud despite many rules on moving identifable NHS data offshore Apr 3: HSCIC reveals that HES data sold to 1200 universities, firms and others since 2013
Open Data Institute, 4/4/2014
The row over HES
Some HES records have name, address, Some have only postcode, dob, Some have this removed but still have HESID which usually contains postcode, dob Even if the HESID were encrypted, what about cardioversion, Hammersmith, Mar 19 2003? Yet the DoH describes pseudonymised HES data as non-sensitive!
Open Data Institute, 4/4/2014
HES data bought by
4042, 4647, 6266, 9598, 159162 : selling data outside the NHS 1912, 321, 329, 331, 362 drug marketing 329, 336, medical device marketing 408: Imperial College with HESID, date of birth, home address, GP practice: still marked "non sensitive Many: market anaysis, benchmarking, efficiency
Open Data Institute, 4/4/2014
The big tussle in Europe
Data Protection Regulation currently making its way through the Europarl Attempt to exempt medical data (art 81, 83) Youll be deemed to consent to secondary use and forbidden to opt out retrospectively, or even claim that consent was coerced Most lobbied ever law in Europe with 3000+ amendments from big pharma, researchers Looks like it will be stalled till after election But I v Finland is still case law
Open Data Institute, 4/4/2014
Now add DNA
The UK Department of Health is launching a 100,000 genomes project to use genetic analysis in both direct care and research All sequence data centralised; if you dont consent to unlimited research use (including sharing with 23andme) then no treatment The FDA just stopped 23andme from offering health advice to new customers In the UK, a Nuffield Bioethics Council inquiry
Open Data Institute, 4/4/2014
Take-away
Think safety and privacy, not security Scale matters! A national system with 50m records is too big a target (even 5m) Governance failure has real safety costs Privacy failings limit access to healthcare, especially for the vulnerable Similar debates in the USA, Norway, Austria Above all we need honesty we need to stop pretending that pseudonyms protect privacy
Open Data Institute, 4/4/2014
Snowden?
When you discover that a paraplegic Canadian woman was denied entry to the USA after a border-guard accessed a database that revealed she'd once been suicidally depressed, it's easy to see how you or someone you love might suffer far-reaching consequences even from accurate data used for the purpose it for which it was intended. Cory Doctorow, Guardian
Open Data Institute, 4/4/2014
