Is the Patient Cold or Does the Patient Have a Cold?
A man walks into the doctor’s office and says, “Doc, you gotta help me, I’m cold.” The doctor says, “Take two aspirin and call me in the morning.” This might be good advice if the patient had a touch of acute rhinitis–the common cold–but it’s unsound medical advice (not to mention a weak punch line) if the patient is suffering from hypothermia or experiencing symptoms of lupus.
The truth is, in the doctor’s office, a physician can make the contextual distinction between having a cold and being cold, as can the medical transcriptionist and the coder. But as the health care industry moves toward the computerized patient record (CPR), the question is, can the computer?
The Data: Codes, Vocabulary and Images
In ADVANCE for Health Information Executives® February 1998 article “Clini-cal Vocabularies: The Missing Link in CPR Systems,” Robert B. Bruegel, PhD, and David J. Rothwell, MD, of The Health Language Center, discuss the shortfalls of current coding classification systems. “International Classification of Diseases (ICD) and Current Procedural Termino-logy (CPT) have relatively limited coverage of clinical concepts. These vocabularies are designed for–and primarily are limited to–capturing diagnoses and procedures.” Consequently, “They and other current clinical classifications have difficulty handling degrees of severity, trends and outcomes, as well as numerous other important assessments and measures.” Drs. Bruegel and Rothwell recognized the “design assumptions” behind these systems–assumptions that were “highly relevant in an era of fee-for-service medicine,” but fall short in the contemporary climate of managed care.
Medical informaticians and other health information experts seem to agree that the Systemized Nomenclature of Medicine (SNOMED), developed by the College of American Pathologists, is the most comprehensive of clinical terminologies, containing more than 180,000 terms and codes in multiple hierarchies with mapping to both ICD and CPT codes (see ADVANCE’s Aug. 7, 2000, article “HIM Professionals Help Solve the Vocabulary Standardization Puzzle”). In addition, it was recently announced that Read Codes, developed by the British National Health Service (now in Version 3) signed an agreement to merge with SNOMED.
“If you studied the differences in the systems, the United Kingdom put a lot of emphasis on preventive medicine in primary care,” something lacking in our own system, according to Susan Fenton, MBA, RHIA, a professor at the College of St. Scholastica in Duluth, MN. Merging two unique systems can create what she called “a formidable force” in clinical data.
But data are more than codes and vocabularies. “When we think about electronic data, we tend to think about text. We don’t think about the images,” observed Fenton. An important question she throws out to health information management (HIM) professionals is this: “Humans can look at two X-rays and say they’ve either changed or they haven’t, but how does a computer do that?” And even if it can, Fenton asked, “How does it index and classify that image?”
Formalization and the Focus on Content
“Data formalization is taking data and giving it a structure so computers can manipulate it,” explained Fenton. “When you talk about standardization, it’s more along the lines of fields for the record or structure that the electronic message will take.” On the other hand, Fenton clarified, “When we’re talking about data formalization, we’re talking about the content of the record and the message.” The standards organization Health Level Seven (HL7) is one group concerned with what Fenton referred to as “the criteria for content.” Essentially, HL7 is working toward a universally accepted protocol for the transmission of patient data, which may include XML or a comparable Internet-friendly language. In an HL7 data exchange, Fenton explained, “You can use almost any data, but how you interpret the content of that message is what data formalization is all about.”
Defining the CPR
The 1997 revision of the Institute of Medicine’s (IOM) study, “The Computer-Based Patient Record: An Essential Technology for Health Care,” defined the CPR as “electronically stored information about an individual’s lifetime health status and health care.” But a practical application of this description is another story.
“If your idea of a CPR is, ‘we have no more paper because we scan everything,’ that CPR is only as useful as a paper record.” On the other hand, Fenton pointed out, “If you’re utilizing the CPR for decision support, alerts, reminders, etc., you are truly utilizing the data.” She added, “We have to tell the computer how to interpret ALL of the data–in order to do that, it all has to be formalized.” Of course the data must be captured, first.
NLP: Capturing the Data through Voice
A man goes to see his psychiatrist. He tells the doctor, “I had this dream last night. First I’m a teepee, then I’m a wigwam. I’m a teepee, I’m a wigwam…” The doctor interrupts him, telling him, “Relax, relax, you’re just two tents!”
Voice recognition technology maps the phonetic sequence of the speaker’s voice into words, so that when this psychiatrist dictates that his patient was, “too tense,” his natural language processor may search its vocabulary and come up with “two tents.” And yet, many feel natural language processing (NLP) may be the key to transitioning providers into a more formalized health care environment.
In 1999, The Journal of the American Medical Informatics Association conducted a study titled, “Continuous Speech Recognition for Clinicians.” In the study, 50 discharge summaries were dictated into three speech recognition systems. Homonyms such as to/too/two and tense/tents were, according to the study, inevitable, but the study found that if users corrected errors as they occurred, accuracy improved by at least 5 percent within a two-week period, and a 98 percent accuracy rate was achievable. In fact, the study concluded that “Voice recognition technology has the potential to overcome one of the most significant barriers to implementing a fully computerized medical record, namely, direct capture of physician notes.”
Fenton agreed, commenting, “We must have the ability to identify and manipulate all of the patient data, and this solution might be found in voice recognition.”
No matter how the data are captured, the goal of a workable CPR is coherent health data, and key components of this coherence are comprehensive vocabulary and coding schemes. Fenton’s only fear is the lack of input from HIM professionals who have much to contribute.
“In HIM, we have to get ourselves out of this paper-record mentality,” she said. To those skeptical of a computer’s ability to capture all patient data, she asked rhetorically, “If we can never do 100 percent, does that mean we shouldn’t try to do 90 percent?”
Linda Gross is an editorial assistant at ADVANCE.