A ‘Google’ for Electronic Health Records

Ten minutes can be an eternity for medical professionals that need to make split-second decisions to save lives. If they miss a single piece of vital information it could prove critical – but equally, if they waste precious time searching for data, that too can be damaging to the patient’s health.
That’s the situation faced daily by Camilla Howard, an ICU specialist nurse working in the Critical Care Outreach team at Waitemata District Health Board. Before she can confidently assess and treat an urgent referral for a deteriorating patient, she needs to quickly analyse the patient’s recent and past discharge summaries and clinic reports. Ms Howard can spend over ten minutes manually reviewing each individual electronic clinical document to find both historical and current clinical information for any given patient.

An ideal solution for Ms Howard would be a kind of ‘Google for Electronic Health Records’, that would consist of an information retrieval application which supports the real-time searching of a patient’s record based on semantic concept grouping, such as “admission to ICU”, “procedure history”, or “respiratory function”.

Ms Howard is the clinical lead on the Smart Search project that is led by Principal Researchers Dr Edmond Zhang and Michael Hosking. Using techniques such as Natural Language Processing (NLP) and Machine Learning (ML), the project team is creating a proof of concept that can be applied to EHRs. The challenge lies in the myriad of ways that the data is presented, which includes clinical notes, clinic letters, discharge summaries and diagnostic reports. Complicating matters is that there is often no consistency when describing clinical concepts, for example ‘HTN’, ‘HBP’, High Blood Pressure’ and ‘Hypertension’ all mean the same thing.

Collaboration between clinicians and data scientists is critical to the project’s success and the initial stages of the research have focused understanding how search is currently done, what the current problems are and how NLP and ML can be incorporated into clinical products.

As part of this consultation the team has begun to look at adding an ‘exploration mechanism’. This is because many clinicians don’t immediately know what they need to be looking for. An example of this are patients that have a limited background about why they are in hospital and in that case the medical professional is able to navigate around the tool rather than typing in specific conditions at the outset.

The benefits that will arise from Smart Search are numerous, with the most obvious being time-saving and improved productivity. Access to faster and more incisive search results will mean clinicians are less likely to order duplicate and expensive tests, while automated clinical coding of discharge summaries could provide hospital administrators with the ability to identify resource gaps and help with future planning.

The application of this research is likely to extend beyond New Zealand and could provide significant commercial returns. A Transparency Market Research report published in early 2017 predicted that the global health care NLP market will be worth $4.3 billion within the next decade. It seems New Zealand clinicians are not the only ones in need of advanced search tools, and that an exciting opportunity may arise to export the software that is developed here to markets around the world.