Each December we look at the year’s progress towards telling patients how data about them is used (from 2014 … 2020).
Good TREs Work, and good Trusted Research Environments are working. The remaining hold-outs are those whose ideology requires data to be copied in the shadows, avoiding both transparency and accountability.
All dissemination of linked patient-level data is unsafe, but some is self-evidently more dangerous than others – such as organisations receiving the same data for the same people for the same month, one set with opt-outs applied and one without opt-outs applied. We have ‘red flagged’ such organisations on TheySoldItAnyway.com, and they should be required to use the TRE for all future projects that need data each month.
Data recipients like these pose an unjustifiably high, systemic risk – especially when it has been shown that Good TREs Work.
2021, and what’s next?
GP Data for Planning and Research (GPDPR) collection is paused until NHS Digital’s TRE is working for all GP data, which it is not yet in a position to deliver. While Good TREs Work, NHS Digital has not yet delivered a TRE which is as good for everyone as the Secretary of State has committed them to doing – and as NHS England is actively undermining it from doing.
GP data cannot be collected until access is TRE-only, and there’s still quite some way to go on that. We would include details of how long that is likely to be, but NHS Digital does not publish the data that would allow that figure to be worked out.
The Goldacre Review should have been out by now – it is still due ‘soon’ – and in data terms it is expected to be largely uncontroversial. The handling practices of both GP and Hospital data have been dangerous for decades, but they can and must be reformed.
Hopefully NHSX, NHSD and NHSE will finally recognise that danger, and what the 2021 ICO Code of Practice on Data Sharing, (UK) GDPR and the 2018 Data Protection Act all say – that dissemination of highly detailed, sensitive personal data on the entire population of England can result in re-identification, including through the event dates that are entirely unprotected in the datasets. Continued denial of this danger will result in NHS patients being identified, as happened in Australia.
For 2022, we have updated the data usage report mock-up we first drafted in 2014.
The most noticeable changes in the report are the organisational names and logos, but the principles of consensual, safe, and transparent data handling remain. All data handling by NHS bodies could be transparent, and the data uses register format NHS Digital moved to in 2021 is an improvement – but bodies like NHS England, for example, still choose not to say how its COVID-19 Data Store was helpful in the pandemic.
With only twenty projects admitted to by NHSE, the panoply of missteps that occurred in the pandemic seems less surprising – if no less shocking – and there’s no published evidence of any value in Palantir Foundry at all. (We go into more detail on this in the available next steps to Data Usage Reports (2021).)
AI and data governance
As AI moves out of DHSC and the civil service into the “real” NHS, it will have to justify the budget and resources it has been given. Though there is a point in time in the history of everything that works when it didn’t work, there is never a point at which those things that don’t work did, no matter how much money was spent.
The AI strategy will re-emerge at some point, and NHS England will get to reconsider it. Our straightforward advice is this: one third the length, one third the budget, and three times the vision. Under NHSX, things have gone in the opposite direction…
DHSC and corporate interests are not the same as doctor’s interests or patient interests. Not even close. Recognising the AI advisory and former No10 Chief of Staff’s view of international agreements – that subterfuge and double-dealing are legitimate between parties – every supplier to the NHS should be required to provide a “datasheet for datasets” for every dataset it was trained on (and to check all the IG) so as to stop ‘data shortcuts’ being profitable.
Public and professional unease around both genomic data and AI is not limited to data governance. That Genomics England handles data safely does not eliminate concerns around how it may be used, e.g. for newborn baby screening. Just because something can be done with data, and can even be done safely, does not mean that it should be done at all.
COPI renewal – choosing a better timing cycle
COPI remains in force, and it is unlikely that DHSC will be able to make a good decision on renewal in any February or August. The March and September dates are simply a legacy of when the pandemic started and a 6-month renewal. This being the case, if HMG believes the pandemic really is ending, the next COPI extension should be for just three months – which would also put things onto a more reasonable cycle of making decisions before and after winter, not in the middle of it, should another variant emerge.
Will they ever learn?
In the context of the Government’s “new direction” on data that will make it harder for people to understand what is being done with their data, and easier for companies and authorities to use it beyond people’s expectations, NHS England’s hostile takeover of NHS Digital means all of these risks and responsibilities will become theirs.
The public may have been generally confused by who and what NHS Digital is – a symptom of what the Wade-Gery Review referred to as ‘split responsibilities’ and a ‘fragmented’ landscape – but everyone can understand that the institution Directed by Government that is NHS England is neither your doctor’s friend, nor yours.
In 2014, NHS England blamed the Health and Social Care Information Centre for its own care.data debacle, requiring HSCIC to cede more control, have an NHS England Board chair, and an organisational rename to NHS Digital – which is precisely who NHS England and NHSX (i.e. NHSE + DHSC) now blame for the collapse of GPDPR in the summer of 2021, a programme over which they had final say. Clearly no-one learned the lesson the architect of care.data was forced to, seven years ago: “We do not subscribe to artificial deadlines here – we will roll it out nationally only when we are sure the process is right.”
DHSC’s commitment to TRE-only can be delivered, and NHS Digital has started to deliver it – with 125+ organisations using some form of NHS Digital “system access”, according to its release register.
Trust requires transparency
The risks and issues around Hospital data and those of GP data are by and large the same. And in medConfidential’s dealings with NHS Digital / HSCIC over the past decade it has always been clear that hoped-for improvement was not just possible but eminently feasible. Yet, despite this, progress towards a TRE for all secondary uses of patients’ data and personalised data usage reports for each patient has been minimal, at best in only minor increments.
Lack of progress has in large part been due to DHSC disinterest, lack of adequate resourcing, and the outright intransigence and active kneecapping of positive intentions by NHS England. The evasions and lack of transparency around NHS England’s COVID-19 Data Store only highlights its culture of secrecy and contempt, suggesting a new corporate attitude and approach will be absolutely essential should NHS England come to govern all patient data in the English NHS.
NHS England’s fear of transparency and accountability are not necessarily irrational, however. Senior officials know what they and their NHS England colleagues already do with data, and clearly believe it would not stand up to public scrutiny – issues that Baroness Harding will know from experience have very real consequences. Is it really true that during the entire pandemic, with the unprecedented amount of health data it hoovered up under extraordinary powers, only 20 projects used the tens of millions of NHS patients’ data in NHS England’s COVID-19 Data Store?
If it persists in the absence of good governance that characterises its handling of our data, and with its favoured scapegoat no longer available to blame, NHS England may in the next act be exposed as the true cause for data despair.
Enc:
ONS analogies with NHS Digital datasets