Clinical Computing: The Computer as Clinician Assistant: Assessment Made Simple

Across all of medicine, the assessment triad comprises screening, diagnosis, and monitoring. We screen to identify individuals who may have risk factors or a disorder that would benefit from treatment. We diagnose disorders to confirm a need for treatment and monitoring, or we conduct monitoring alone if treatment does not have a favorable cost-benefit ratio.

Screening, diagnosis, and monitoring each depend on data gathered from the patient through interviews and physical and laboratory examinations. Clinical interviews provide most of the data, physical examinations add a little, and laboratory tests add not much more. Leon Eisenberg (1) has noted that while some physicians become marvelous diagnosticians and experts at monitoring treatment, even the best can overlook important information or fail to weight properly the data they gather.

A study compared diagnoses made by two clinicians experienced in using the paper version of the PRIME-MD structured diagnostic interview (2) and by a computer-administered version of the same instrument (3). Clinicians made diagnostic errors 10 percent of the time due to inaccurate applications of the diagnostic algorithm, while the computer processed the algorithm perfectly each time. Other studies of the accuracy of psychiatric diagnoses have found error rates up to 37 percent because of clinicians' errors in applying diagnostic rules (4-6).

Monitoring the severity of and changes in a patient's condition is best done by tracking not only target complaints, which often change most in response to treatment, but also changes in the frequency and intensity of the common symptoms and signs of the disorder. These data are very valuable for assessing severity and change and also anchor the patient's problem in the larger context of patients with the disorder and those without it. Clinicians monitor chief or target complaints much better than specific symptoms because the chief complaint has a heavy loading for specificity and severity that is easily remembered from visit to visit.

However, consistently following up on the same set of symptoms and signs and remembering their intensity is not easily done by humans and is seldom done systematically at all. Even when clinicians steel themselves to repeated reviews of symptom intensity and frequency, such as by use of the Hamilton Depression Rating Scale, they do not administer the instruments the same way each time, thereby producing intrarater unreliability. And poor interrater reliability in multicenter trials can doom the separation of an effective treatment from a placebo control (7). Even if clinicians could overcome these limitations, collecting such data is tedious and not a good use of their abilities and time.

Computer interviews can help clinicians improve their screening, diagnostic, and monitoring functions. Computer interviews can gather information directly from patients either at a personal computer in the physician's office or on a touch-tone telephone connected to an interactive voice response (IVR) program on a central computer. With desktop and laptop personal computers, questions and information are presented on the screen. Patients read and answer the questions by using the keyboard. With an IVR program, patients listen to the questions and respond either by pressing keys on the keypad or by saying simple words, such as yes or no or the numbers from zero to ten. Computer terminals are available whenever the office is open. An IVR program allows interviews to be completed at any time and from any touch-tone telephone.

For screening, the capacity of computer interviewing is staggering. On National Depression Screening Day in 1995, a screening version of the Zung Depression Rating Scale was administered by computer via IVR to 21,037 persons. The figures were 32,300 persons in 1996 and 116,374 in 1997. Although these numbers are impressively large, they could be scaled upward exponentially on very short notice and with little cost.

For diagnosis, the accuracy of computer programs is comparable to that of excellent clinicians (3,8). But it is not in competition with the clinician that the computer enters the diagnostic process. The thoroughness of a structured computer interview complements and supplements the more intuitive diagnostic process of clinicians. Two heads are usually better than one, even when one is a computer

Clinicians can delegate monitoring to computers, just as they delegate laboratory tests to laboratory technicians. Monitoring is a repetitive task in which previous and present status are compared on important dimensions. The comparisons should be independent of one another and not influenced by extraneous information—both difficult processes for humans to carry out flawlessly. Perfect reliability is the ideal in monitoring. Computers can do repetitive tasks better than humans.

We have long accepted that automated laboratory assessments are more reliable and accurate, as well as less costly, than those performed by laboratory technicians. Relying on computers for monitoring is comparable. Clinicians and patients can use accurate and reliable severity and change data from the computer to make better overall decisions about treatment.

Dr. Greist, who is editor of this column, is distinguished senior scientist at the Dean Foundation for Health, Research, and Education, 2711 Allen Boulevard, Middleton, Wisconsin 53562, and clinical professor of psychiatry at the University of Wisconsin Medical School in Madison.