The Computerization of the Dictionary: What EHRs Can Learn

Whether you want to become a great ballplayer, win a Nobel Prize, or develop a great EHR, it is important to choose the right exemplar - the right role model.

Paper medical records have been around for a long time as have other techniques for managing information: ledgers, filing cabinets, card files, McBee cards, encyclopedias, and dictionaries to name a few. Of these, the McBee card is the easiest to computerize - each card has a title, some narrative typed on it and notches around the edge that codify or classify the narrative.

It is hard to know the minds of the pioneers of biomedical computing but the obstacles they faced are clear. They could not afford the large mainframe computers used by industry nor was the nightly processing of batches of punch-cards appropriate in a clinical setting. They had little  choice but to use a new generation of relatively inexpensive computers (four or five, not six-figure prices) that became available in 1960. They were small and slow and couldn't store much or process large programs and there were no operating systems or programming language other than machine code and assembler. They were, to these investigators, what a stone or a block of clay is to a sculptor - a chance to create something out of nothing. They chose for their exemplar  STRINGCOMP, "one of three variants of JOSS II… that were developed by BBN. [STRINGCOMP] had extended string handling capabilities to augment JOSS's mathematical focus. It was a strong influence in the development of the programming language MUMPS."

Did they start with the goal of creating a computerized medical record? I think not, although I'm sure the thought had occurred to some. In retrospect, it would have been better if they had thought more about medical records at the beginning because what they devised was elegant for its purpose. It was so good in fact, that it was widely copied and is still in use today by many of the major players in EHR. Thus, many of today's EHRs have inherited both the strengths and weaknesses of those early design concepts. They were then, and are still, poorly suited to creating computerized medical records.

If the goal is to create a computerized medical record, there are two important steps: understand the objective and choose an exemplar and thus, two fundamental questions that should have been asked and answered before any attempt was made to computerize a medical record.

First: What are the most unique characteristics of a medical record?
• Many authors make entries containing a wide variety of both quantitative and qualitative (narrative) information.
• Most authors, and many others, need to be able to read and understand all the entries.
• Many of the entries have relationships to other entries and those linkages should be captured and made available when reading the record.
• The records not only have medico-legal significance, they are also the property of the patient and the author as well as the healthcare organization. The patient has a right to expect that the records will persist, unaltered, and be accessible during their lifetime regardless of how often the computer systems fail, become obsolete or are replaced.

Second: what other fields that deal with large volumes of information that share these characteristics and learning and how have they computerized their information?

Several come to mind:
• Court records including legal decisions and supporting documents
• Encyclopedias
• Dictionaries such as the Oxford English Dictionary (OED)

Had the computerization of the OED begun earlier, it would have been an excellent exemplar. Its information is like that in a medical chart in many respects. There is a great deal of narrative along with embedded discrete data and links (both internal and external) to other related entries.

Whereas most EHR developers have modeled the medical record as a table or tree of discrete data elements, the people at the OED spent years considering the most effective way to represent their information in an electronic form. The concept that they arrived at was to represent each dictionary entry was a separate “document” or page. They then worked with a community of authors and publishers to define and refine a set of standards designed to provide a consistent structure that could be applied to each of the thousands of documents that make up the dictionary. The structure surrounds the information, and data-like elements within it, with "mark up" that identifies and delimits the individual components. This allows the dictionary to be read like a book, sorted by different criteria, mined for data, and navigated like the Internet.

This is precisely what one needs to be able to do with a computerized medical record. This approach represents a new paradigm. Few know about it or understand it at the present time, but for those that do, it changes everything.