With EHR, Two Heads Are Not Better Than One

Until narrative and data are united, EHRs will require too much effort, create too much risk, and provide too little benefit to justify imposing them by fiat.

Today, my e-mail offers me a white paper: "Brigham and Women's Hospital saves over $9 million, improves [medical transcription] productivity 123%" with Dragon Dictate. My immediate reaction is: The feds' EHR push is aimed squarely at data collection. Why is a technology leader like Brigham and Women's spending time and money on dictation, something that is the antithesis of collecting data for meaningful use? Could it be that data collection is not all it's cracked up to be? Could it be that you can't really create a meaningful medical record by treating it as a data collection problem?

Not only is data required for meaningful use, we are told that it is essential to achieving interoperability. The words dictation and interoperability (of structured data) are polar opposites. Each frames the discussion about EHR differently. Structured data makes EHR a data collection issue; dictation makes EHR a word-processing issue. George Lakoff, a cognitive/linguistic scientist at Berkeley, introduced the concept that the choice of words "frames" a discussion. Although he introduced the idea in the context of politics, where the issues are moral and the words are emotionally charged, the concept is applicable to EHR. There are two camps. Neither one's needs are satisfied by the other. As a result, the typical EHR resembles dicephalic parapagus twins - conjoined twins with separate heads but one body.

The dictation "head" produces free text. Quantitative or semi-quantitative facts become sequences of words swimming in a sea of other free text. While people can read the texts and infer the meaning, computer applications cannot do so reliably. On the other hand, free text can capture the unexpected, describe the context and can, by the careful choice of words, convey nuances of meaning and intent to the reader. Done well, dictation can be highly informative. Done poorly it can be uninformative or misleading. In either case, it is not a good source of data.

The data collection "head" records the answers to discrete questions. This can be effective if each question is carefully planned to answer a specific question and if sufficient qualifiers are included to allow an intelligent selection of which records are to be included in analyses or reports. Planning and intent are the keys to good data collection. One cannot simply "collect all the data" and figure out what to do with it later. It is impossible to determine which of the millions of possible data elements would be the "right" ones to collect and, without knowing the reason for asking each question, it is impossible to determine what qualifiers might be needed to permit meaningful analysis.

When working with data or information, one of the prime directives is to record each fact (bit of information) only once, store it in only one place and make it available to any authorized process that needs it.

Today's EHRs collect many informational elements twice, once in the narrative and again as "data." Until narrative and data are united, using an EHR will require too much effort, create too much risk, and provide too little benefit to justify imposing them on medicine by fiat.

I believe that that computer technology, if used correctly, does hold great promise to improve healthcare. Unfortunately, as people have debated EHR over the years, the discussion has been framed as if the benefits of EHR are real, not potential. Irrational exuberance on the part of the EHR policy wonks and government officials not only spread this notion but lend credence to it. The reality is that EHRs have yet to deliver most of the promised benefit. Perhaps, giving more weight to the hype than to reality explains why poor decisions about EHR are so common.