As I discussed in my last post, Claude Shannon defined information as the number of bits needed to communicate an arrangement or state of a system. The fundamental problem of communication is not to make oneself understood, but to reproduce “at one point either exactly or approximately a message selected at another point.” The destination of a message could be separated by space or time but the “semantic aspects of communication are irrelevant to the engineering problem.” In his model, the information source is the person or machine generating the message. The transmitter “operates on the message in some way” to produce a suitable signal. The channel is the medium used to transmit the signal. The receiver then inverts the operation of the transmitter and reconstructs the signal. The destination is the “person (or thing)” at the other end. Finally, the “noise source” is everything that corrupts the signal either predictably or unpredictably.
If you were to replicate the above model in terms of the patient-physician encounter, the patient is the information source and the transmitter of the signal in the form of symptoms. The physician plays the role of the receiver, and the electronic health record (EHR) is the destination of the message. The noise source that I want to focus on is represented by the physician entering the message into the EHR in the form of a physician note. Ideally, this note should be an accurate representation of the message transmitted by the patient. However, physicians are not trained to represent these encounters in a digital format. The bulk of medical training is spent on the process of receiving the message, interpreting the message, and delivering a recommendation based on that message. This is followed by a short handwritten note in a standardized format in the patient file. However, in the current paradigm of EHRs, this paradigm has been upended. Physicians now spend nearly 50% of their time with EHRs and clerical tasks and only 25% of their time interacting with patients. We have been reduced to a role of data enterers. This era of ‘desktop medicine’ is having pernicious consequences at that point of care in the form of a poor physician-patient rapport and downstream consequences of physician burnout and poor note quality.
Unfortunately, the problem is not just constrained to the physician level but also has potential downstream unanticipated consequences. As we look to move into the era of precision medicine, big data, and artificial intelligence the quality and fidelity of that message stored in the EHR will be paramount. Data quality is one of the cornerstones for predictive and prescriptive analytics. Algorithms are only as accurate as the training sets that they learn from, therefore, the quality of data represented in these training sets must be an accurate representation of the patient encounter. Poor data quality can lead to meaningless, inaccurate predictions and algorithmic biases. In my experience as a clinician and an informatician, I would conjecture that haphazard, hurried, and generically templated noted that do not accurately reflect the patient presentation are the rule rather than the exception. In an attempt to curb the data entry workload of physicians, we have designed workflow hacks such as smart phrases, templated review of systems and physical exams, copy and paste capabilities, and dictation software that not only decrease the burden of data entry but also dramatically decrease the quality of the physician note. Review of systems and physical exams are mostly templated and often generic and incorrect with providers often forgetting to change the templated elements for the particular patient, free text entry for history and medical decision making is often entered via dictation software fraught with error, copy and paste contributes to bloated notes with inaccurate information.
Physicians neither have the time nor the training to be concerned about the potential downstream ramifications of poor data quality. Their immediate concern is to the see the patient, make a diagnosis, make recommendations and document the encounter in harried, fifteen-minute blocks. The purpose of the HITECH Act was to “promote the adoption and meaningful use of health information technology” with the following five goals:
- Improving quality, safety, and efficiency
- Engage patients in their care; increase coordination of care
- Improve the health status of the population
- Ensure privacy and security
We have been largely successful in the adoption of Health IT as the adoption rate is more than 66% of all providers, however, we are struggling to reach any of the other four goals. The real promise of EHRs is in the secondary use of data. Utilizing this data will enable predictive and prescriptive algorithms that have the potential to revolutionize medicine. The central paradigm of classical information theory was the solution of the “engineering problem of the transmission of information over a noisy channel.” Likewise, a similar paradigm is needed to overcome the engineering problem of entering accurate data into the EHR.