Cookies on this website
We use cookies to ensure that we give you the best experience on our website. If you click 'Continue' we'll assume that you are happy to receive all cookies and you won't see this message again. Click 'Find out more' for information on how to change your cookie settings.
  • 1 October 2015 to 30 September 2017
  • Project No: 327

This programme aims to develop a range of tools for novel or better use of live data in healthcare for immediate clinical impact. is a live data service that provides real time insights into NHS prescribing data, launched in prototype in December 2015. It allows any user to explore time trends and identify outliers for wasteful, riskier, or less evidence based prescribing.

We are developing and implementing a series of features within the site, including 100 standard measures for value, safety, effectiveness and innovation, such as “how much could I save if I moved to the 50th centile, 10th centile”; “send me an email if I slip across 15 centiles for X/Y or any of your standard measures.” We will be evaluating these features in a series of efficient low cost RCTs, and with interrupted time series analysis, to develop a series of insights around best practice for presenting low cost live data feedback to change behaviour, and outlier detection.

The team have also begun work on intelligent outlier detection using novel analytic methods, and automated detection of “interesting patterns in clinical decision making”, such as variation in care, using both complex and simple methods (such as kurtosis, skew*SD, and novel applications of principal component analysis). We intend applying this model of “live data analytics for actionable insights” to clinicians’ own electronic health record datasets. Some examples of early outputs from this work (currently in submission for publication) include: use of an extremely obscure antipsychotic (pericyazine) that is only prescribed in Norfolk; practices prescribing highly unusual products such as powdered opium; and identification of drugs which are reciprocally substituted. Where we can detect widespread and wide arbitrary variation in prescription of treatments that are currently believed to be in equipoise, this may be of value in justifying low cost pragmatic trials comparing those two treatments. We are also able to calculate in real time the local and national cost to the NHS of inefficient prescribing.

In addition we are conducting a series of conventional observational epidemiology analyses using features of practices as exposure variables, and prescribing behaviour as outcome variables, to explore factors associated with specific patterns of wasteful or high risk prescribing.

Lastly we are collaborating on a series of tools around intelligent use of time course data to better identify outliers for problematic prescribing, for example by identifying those practices who have been slow or fast to respond to previous changes in cost or evidence, using tools to automatically identify discontinuities in prescribing behaviour across the entire dataset of all prescribing. This work is also valuable in formally mapping the diffusion of knowledge throughout the medical community, and the variation in how quickly different doctors change their practice, which is in turn useful for informing policy choices on how to better ensure clinicians are kept up to date.

Future Work:

With additional resources we will accelerate development work on the live services, conduct more analyses faster, and speed up publications on the work set out above.

We are already negotiating access to smaller specific sources of individual patient data which are currently under-used. For example, we are currently negotiating access to a national individual patient prescriptions dataset that includes dose, patient ID and prescriber ID, which will allow us to build further “sentinel” services to identify outliers, but also “never events”. For example, we are currently working on a project to identify all situations nationally in which any individual has received a methadone prescription from more than one prescriber, to a total weekly dose that exceeds their previous weekly doses, suggesting that they have been able to solicit inappropriate prescribing. In addition, we are negotiating a communications pathway to “close the loop” by feeding back to CCGs, practices and clinicians where this has happened, which will enable us to assess whether such feedback reduces the risk of further incidents in future (whether by RCT or interrupted time series analysis). Following the launch of we have received various offers of access to commercial and public sources of individual patient data that we are keen to follow up on.

Our next step is to start applying the model of “live data analytics for actionable insights” to clinicians’ own electronic health record datasets ie expand our live data analysis paradigm into individual patient data. This area is considered a ‘holy grail’ by the NHS – utilizing the enormous ‘Big Data’ contained within the electronic health record for monitoring, educating, evolving, or other health gain purposes. With additional resource we will accelerate a scoping exercise to identify all amenable datasets locally and nationally, and the technical and information governance issues around reconciling and using them. Key to our way of working is “agile development” (rapid product iteration) of the live data tools, and close engagement of users. With additional resource we will be able to do more and faster user engagement using team members who are familiar with the opportunities from live data analysis, working with clinicians, patients, commissioners of health services, and others to establish what practical live data tools would be most beneficial to them, thus ensuring that what we build is consistently of high impact and high clinical relevance. Candidate projects here include: live low cost tracking of the impact of a new intervention, whether clinical or policy; and live outlier detection for unusual patterns of ordering investigations or imaging. At all stages we will formally assess impact using conventional qualitative and quantitative methods, and share code openly to maximise impact both on clinical practice and the culture of better uses of data in healthcare.

This programme offers a rapidly deliverable application of NHS Big Data, with academic data analysis to real world datasets with the aim of immediate clinical impact. The delivery of the OpenPrescribing demonstrator website shows we have a proven track record of delivering working live data analysis tools efficiently, addressing clinical need, and generating novel insights.

Projects by themes

We have grouped projects under the five SPCR themes in this document

Evidence synthesis working group

The collaboration will be conducting 18 high impact systematic reviews, under four workstreams.