Two countries united by natural language processing?

Source: DigitalHealth. George Bernard Shaw famously quipped that the US and the UK were two countries separated by a common language. But in natural language processing, State-side hospitals are finding a means of better assesing the appropriateness of specific tests or procedures – and Dan Kazzaz argues the NHS could valuably follow suit.

Comparing the US healthcare system to that of the UK is often great sport. Points of comparison are frequently the cost of care, quality of care and waiting times. Less obvious is that the US and UK systems actually have a great deal in common, primarily how healthcare is funded.

The largest US healthcare insurer is Medicare, which is comparable to the NHS. Although Medicare is primarily for those over 65 years of age and does not cover all items covered by the NHS, both are highly dependent on the working population to support the retired population through taxes.

The US and UK have other major points in common, such as excellent health systems and low birth rates. Excellence in healthcare translates into people living longer which increases the demand for healthcare services. Low birth rates create an obvious challenge – fewer working age people to pay for the complicated healthcare of an expanding population of seniors.

The march of prior authorisation

When it comes to best practice, the US and UK frequently emulate each other, although from time to time they are loathe to admit it. Recently, the US public and private payers have been shifting the “risk” to hospitals by paying them a fixed amount per person per year.

This may help somewhat, but more is needed. For its part, the UK is now beginning to model one of the US structures – “prior authorisation”. This involves verifying if specific services or procedures are warranted before performing them, by asking questions such as whether the patient needs to be hospitalised or could be cared for at home, or which type of scan might be needed.

In the US, hospitals and physicians decry the current prior authorisation methodologies. Why? It simply takes too long and requires too much effort. Seeking prior authorisation is quite labour intensive, especially for government and insurance entities. US hospitals also find it difficult to accept risk, especially as they see rising expenses due in part to individual physicians who, more often than not, are free to order treatments or tests based on their clinical judgement, not on cost.

Enter NLP…

The bright spot on the horizon for healthcare is the use of computer science technology applied to the evaluation of clinical orders. Previously, customary wisdom was to assign numerical values to diagnoses and treatments. The enumerations of patient conditions can be found in SNOMED and others.

The encapsulation of this information, as well as physician notes and other clinical information, is found in the Continuity of Care Document standards. However, these enumerations are insufficient to the immediate task at hand, which is to describe the current state of patients. The task is made more difficult by physicians using free hand text notes to describe the state of patients.

Computer science is addressing this challenge through the use of Natural Language Processing (NLP). For the uninitiated, NLP is one technology behind internet search engines. When a question is typed on the web, such as “When was the queen of England born?”, the NLP engine parses the question into a logical query on the databases of all web pages.

Organisations specialising in NLP for clinical settings – such as Linguamatics, a UK company, and more recently, Amazon, with its Comprehend Medical platform – are offering solutions designed to extract information from unstructured medical text accurately and quickly.

So how to set it up?

Applying NLP to prior authorisation or clinical decision support systems can radically reduce the time and effort required to process these costly orders. The challenge: how to set it up?

I’d suggest there are four key areas to consider. First, queries against the clinical information must be developed. Fortunately, the guidance for these is well defined, which is why machine learning can process the guidelines and produce the query for each treatment.

Second, secure communication protocol standards must be in place to to move the clinical information to the decision spot. In the US, certified electronic medical record systems (EMRs) all have the protocol needed. They use it for point-to-point transfers of clinical information today. This protocol could, and perhaps should, be used in the UK.

Third, medical images and the existing data must be converted into free form text for the NLP engine.

Finally, regardless of whether the query is delivered outside the hospital or is an internal process, the results must be evaluated by a trained clinician and, if denied, perhaps appealed.

By reducing the amount of labour involved in processing the growing number of prior authorisation requests, substantial savings are achieved. The US Medicare system is pushing hard to have computer assisted review of documentation. Perhaps the NHS might find this to be a promising area of focus on as well.

Dan Kazzaz is chief executive of Secure Exchange Solutions

Click here for the full article on Digital Health.

About the Author :