On 7th May 2020, the OpenSAFELY Collaborative pre-printed the world’s largest ever study into factors associated with death from COVID-19, based on an analysis running across the full pseudonymised health records of 40% of the English population. This was an unprecedented scale of data. Since then, our team in Oxford has been busy: growing the OpenSAFELY platform to cover 58 million patients’ NHS GP records. That was - once again - an unprecedented scale of data - covering the whole population of England. This was only made possible by our team developing innovative new methods to protect privacy, and delivering outputs through highly secure federated analytics of patient data at source.

Since then we’ve been delivering important research for the NHS and policymakers to help inform the national response to the COVID-19 pandemic; and growing our team to become the Bennett Institute for Applied Data Science. In the coming weeks we will be sharing more about the teams that make up the Bennett Institute in Oxford, how we work and examples of what we have delivered.

In the first in this series, Brian MacKenna - Director of NHS Service Analytics at the Bennett Institute - shares how OpenSAFELY can be used for NHS service analytics, and gives an overview of some key examples from our work in this area.

What is OpenSAFELY?

OpenSAFELY is a productive, highly secure, transparent, non-commercial, fully open-source software platform specifically developed for analysis of NHS electronic health records data. It can be deployed to provide secure access to any existing database inside any existing data centre to create a Trusted Research Environment with unprecedented security, transparency, and efficiency. With these privacy protections the OpenSAFELY team, comprised of software developers, data scientists, researchers, and clinicians, has earned public trust to implement OpenSAFELY across an unprecedented volume of NHS patient data.

The timeliness of the data in OpenSAFELY is particularly critical here: by working closely with EHR vendors we have been able to deliver analytics outputs with data that is refreshed weekly, and can be re-built in just a couple of days; and all our analysis pathways are automated, end-to-end, so they can run and re-run very quickly. Because of this, results can be just two to nine days behind clinical events.

OpenSAFELY isn’t just for academic research: it has been used to deliver a large amount of NHS service analytics work where we run analyses examining, for example, the direct and indirect impacts of the pandemic to support response to COVID-19. This blog post sets out some key examples of our NHS Service Analytics work. If you’re interested - or would like to get involved in our work yourself - then more can be found at www.opensafely.org and dashboards at reports.opensafely.org.

COVID-19 Vaccine Coverage in Detailed Demographic and Clinical Subgroups

Vaccine coverage dashboards were deployed immediately in Dec 2020, with detailed coverage data in fine grained clinical, as well as demographic subgroups. OpenSAFELY was the first to raise the need to address lower vaccine coverage in: ethnic minorities; learning disabilities; severe mental illness; carehome residents; and more. No other data service can provide such detailed characterisation of the clinical background of vaccine recipients at such a huge population scale. We produced outputs in academic papers and more importantly, live updating dashboards which received substantial coverage prompting action across the NHS.

Coverage and uptake of antivirals and neutralising monoclonal antibodies for the treatment of non-hospitalised patients with COVID-19

Using our vaccine coverage framework, we deployed dashboards at the outset of the antiviral roll-out. OpenSAFELY highlighted the need to understand and to address lower coverage in: different NHS regions, socioeconomically deprived areas, ethnic minorities and care homes. Similar to our vaccine work we have produced outputs in paper and dashboard and the underpinning analytic code has been re-used to facilitate swift delivery of papers assessing the effectiveness of these medicines.

Inappropriate anticoagulation of patients with a mechanical heart valve

National guidance was issued during the COVID-19 pandemic to switch patients on warfarin to direct oral anticoagulants (DOACs) where appropriate as these require less frequent blood testing. DOACs are not recommended for patients with mechanical heart valves. Using OpenSAFELY we rapidly identified that potentially 768 people had been switched inappropriately. This analysis was only possible because OpenSAFELY uniquely contains detailed and timely data on the full clinical background and complete primary care treatment history of all patients. NHS England issued a national patient safety alert as a result and we rapidly deployed a dashboard to inform implementation. Furthermore, EMIS and TPP were able to send alerts to GP practices affected using our analytic code that we had shared to rapidly identify people for review.

NHS Service Restoration Observatory

We have developed and deployed code that monitors volume of coded activity and clinical outcomes for any aspect of GP service during COVID to identify outliers. We are able to provide detailed information on all coded activity during COVID-19 to describe disruption, inform response and prioritise recovery. To date, we have found that most activities exhibited significant reductions during pandemic wave 1 with recovery by September 2020 however many important aspects of care - especially those of a more time-critical nature - were maintained throughout the pandemic.

PINCER - Medication Safety

PINCER is a national patient safety programme, collecting data on compliance with complex clinical pathways related to patient safety. It was previously delivered through AHSNs by running searches in hundreds of practices to identify patients where there may be safety concerns or where treatment may not be in line with recommendations. These lists of patients are then reviewed by pharmacists in practices. In collaboration with PRIMIS and the University of Nottingham, we have re-implemented all of PINCER in OpenSAFELY to deliver all PINCER metrics for all practices in a single command, and provide dashboards across the whole nation, that can be updated with a single click. This is an unprecedented scale and depth of data on compliance with complex clinical pathways in the NHS. We are keen to hear from others doing this type of work to inform future project on the response to the COVID-19 pandemic.

Assessing implementation of NICE guidance for Pancreatic cancer

NICE pancreatic cancer quality standards recommend that all patients with pancreatic cancer are offered pancreatic enzyme replacement therapy (PERT). Using OpenSAFELY we assessed the impact of COVID-19 on prescribing PERT across England. Briefly, we found no reduction in pancreatic enzyme replacement therapy during the COVID-19 pandemic. However we also found this treatment is not being prescribed to all patients with pancreatic cancer with substantial regional variation. Previously, practical and privacy challenges around accessing GP data meant that the largest study to date in this area, RICOCHET, relied on manual audits by many local teams at a single point in time. This manual approach imposes a substantial resource burden on local teams collecting data as well as being hard, or indeed impossible, to reproduce on an ongoing basis. We are now developing features in OpenSAFELY to make our tools available to a broader range of teams to make it easier to deploy analysis like these across a broad range of clinical areas that may have been affected by the pandemic.

Long COVID

The ONS estimates that approximately 3.3% of the population experiences long-COVID based on self reports by individuals. Using OpenSAFELY to analyse the records of 58 million patients securely, we quickly identified that long COVID was only recorded in GP systems is 23,000 times by May 2021. This may reflect patients not presenting; clinicians and patients holding different diagnostic thresholds; or challenges with the design and communication of diagnostic codes. We alerted the system to this; EMIS and TPP made changes to their user interface to make relevant codes more accessible and NICE incorporated our work into a review of their rapid guidance on managing the long-term effects of COVID-19. Our dashboard now shows increased recording of Long COVID, although recorded diagnoses are still dramatically lower than self-reported survey reports.

Recreating the National Early Inflammatory Arthritis Audit using OpenSAFELY

The National Early Inflammatory Arthritis Audit (NEIAA) is the largest audit of its kind globally, reporting annually on care delivered across NHS rheumatology services. Mandatory data collection in NEIAA was paused during the pandemic, preventing comparisons of pre-, during, and post-pandemic care. Working with colleagues from King’s College London using OpenSAFELY, we were able to recreate key aspects of the NEIAA to benchmark the quality of care for people with IA. During the early COVID-19 pandemic, there was a 40% reduction in recorded IA diagnoses. Although we saw some impact on service delivery, this was less marked than might have been anticipated, and evidence of recovery was swift. Our findings closely reflect those reported in the national audit of IA care in England, without the need for manual data entry by clinicians.

Gathering the threads together!

It is common to hear people say that “Data Saves Lives”. But often, examples come from the world of pure academic research. In the Bennett Institute we think that practical service analytics - monitoring the activity and outcomes in the health service, and finding opportunities to optimise the delivery of care - is just as important and interesting as pure research. It is also, in many cases, very similar work. It uses the same raw data as academic research, the same skills, the same tools, and the same platforms. It benefits from all the methods and tools we have built into OpenSAFELY: like open and reproducible code that can run quickly, and securely, in automated pipelines, across huge populations. Lastly, service analytics have the same over-arching objectives: using data, thoughtfully and creatively, to help patients lead better and longer lives.

We are always keen to hear from people working on NHS service analytics, especially if they are interested in modern, open, automated approaches to working with health data. If you are living in that field, and would like to work with us: get in touch via bennett@phc.ox.ac.uk!