“Data is not information, information is not knowledge, knowledge is not understanding, understanding is not wisdom”
Clifford Stoll, Gary Schubert

My previous post on this subject ended up as a bit of a cliffhanger – the feedback was that I’d identified issues on which we could agree, but not given any clarity on ways to mitigate the problem (Thanks Mike Rutherford for your comments).  The problem statement might look like “how do I, as the person responsible in the NHS trust, identify, gather and package all responsive documents for a legal case in all of the diverse formats and modalities in a timely and compliant fashion?”

Major steps are underway in Government and the NHS to adopt a “Digital by Design” strategy, that would in theory at least mean that all information comprising a patient medical record resided in an easy-to-query format.  While it is gratifying to see that ruggedised tablets are in increasing use in secondary (non-GP) care settings for some patient interactions, the patient’s journey still relies on paper files at the sharp edge of treatment in many cases: paper continues to be, if not prevalent, then in many cases preferred by clinicians for it’s ease of use and perceived authenticity.

Patient Administration Systems are increasingly able to intercommunicate, smoothing the interaction between primary and secondary healthcare sources.  In GP’s surgeries, digitised summaries of a patient’s history are intended to make the old Lloyd George folders obsolete. NHS England have nominated Capita to manage elements of the GP2GP patient record management scheme, designed to simplify the process of patient record transfer between GP practices.  However, this hasn’t been an entirely smooth process, with assertions of incomplete transcriptions, irretrievability of attachments, and delays in paper file transfer causing friction and delays.

A curated PDF binder compiled using search terms agreed by interested parties has become the accepted format for legal review, but the varying formats of medical data continue to present expensive and time consuming obstacles:  MRI and other medical output modalities generate huge and proprietary file formats, and paper documents can have enormous volumes, even for one specific patient episode.  However, snapshot overviews in PDF format can be made for medical imagery, so one remaining problem is how to identify relevant paper records and render them into PDF?

There are two approaches to paper medical record digitisation –  proactive and reactive. many NHS trusts have attempted the proactive bulk digitzation of cohorts of their paper records.  The problem has been that to correctly identify each document within a record, extract relevant information and match that to enhanced metadata relating to the patient and episode has been prohibitively expensive, given that perhaps only 1% of the digitised record may ever be subsequently viewed.  The alternative approach – little or no document classification and a minimally separated record renders the digital experience unusable to clinicians and can actually make the problem worse, since a poorly-classified and separated medical casenote PDF may be several thousand digitised pages that cannot be usefully navigated in a realistic timeframe by a clinician.  However, it would be reasonable to surmise that a legal team involved in representing a patient would painstakingly scour that file, searching for documentary evidence of poor or inadequate care…

The answer may lie in new and more powerful interpretations of AI-based document classification, that can correctly separate, classify and index unstructured and semi-structured digitised case notes.  One such technology – called CloudHub360, has been able to demonstrate and deliver a cloud-based Document Transformation engine capable of high-speed, high-volume and highly accurate bulk document processing – with no human intervention, once configured.  

The historic challenge has been to maintain the accuracy of automated document classification, while maintaining an acceptably high classification rate.  The test results shown below suggest that this new implementation of AI can make the trade off between classification accuracy and classification rate less of an issue..

Results in red show results using the Cloudhub 360 engine against a leading competitor, based on 800 documents spanning 8 classification types..

Further, since the characteristics of medical record types demonstrate many consistencies in content, if not appearance, the memory of learned documents can be transported from one NHS trust to another, vastly reducing setup times and making the notion of a universal transformation processing engine, cloud-accessible, a reality.  This engine has now been integrated with a browser-based viewer, and made usable within the NHS IT environment.

We welcome the opportunity to work with you to investigate the applicability of this technology to your use case, and are very excited by it’s possibilities.  Please get in touch with us to find out more…

image 1 courtesy of Newtown Graffiti – https://www.flickr.com/photos/newtown_grafitti/5525985630

images 2,3,4 courtesy of CloudHub360

image 5 courtesy of Weiss Paarz – https://www.flickr.com/photos/141290938@N03/26682786574

Leave a Reply