automated patient history in Arbor Medicus

As physicians, we are trained in the art of conversing with patients, which allows us to establish a rapport, assess the ability of the patient to give us reliable information and observe the patient's behaviors as we collect the history. The interaction between doctor and patient cannot be replaced by a machine.

The problem is, we have limited amounts of time to converse with our patients...shortcuts are taken and data is missed.

In medicine, the errors of omission are often worse than the errors of comission. It is in the nature of the physician to try to do a good job and pressures placed upon these physicians through insurance companies, employers and other sources, cause these shortcuts to be taken.

The patient is not always aware of what facts may be relevant and unless we skillfully and patiently probe for specific information, as well as screening information, early detection of many diseases can be missed.

We commonly consider two types of patient information gathering; one is the acquisition of structured information from structured questions. The other is free conversational or "open-ended" questioning.

Clinicians have tried to gather structured information through paper-based checklists, administered by the patient themselves or through an assistant. A common problem is the lack of completeness or the inability of the patient to understand the terminology. Certainly, paper-based systems are unable to probe deeper into positive responses and do not explain inconsistencies. In addition, we overestimate the literacy as well as health literacy of many patients.

With the advent of computing, electronic methods have evolved which allow patient interrogation through the use of computers. My group has been developing and using such methods since the mid-1980s, and these methods have become more sophisticated over time.

In those days, automated questionnaires worked better than simple paper questionnaires but a weakness was still apparent in those which asked simple questions or had simple branching.

I became very interested in the process of asking questions, as a clinician, and as a teacher. It became very apparent to me that an artificial intelligence method was needed. My colleagues and I subsequently developed the FloBase® language for the purpose of simulating expert interrogation. With such a system, questions could be asked according to specific rules, as developed by clinical experts. The program selects questions based upon previous answers, patient demographics, existing symptoms, or even as the program begins to form hypotheses along a certain line of inquiry.

Our next step was to develop a more personal interaction between patient and computer. A strength of the computer interview is the patient's perception of anonymity and the absence of any judgemental facial expressions since there was no live interviewer. A weakness was the impersonal text interface.

Over the years, we found it very useful for the patient to interact with a human face who is able to delve more deeply into specific sensitive areas. In addition, when the patient expresses difficulty understanding the question, the video nurse can clarify the question and ask it in different terms, to elicit a more accurate response from the patient. In addition, many people are more likely to understand auditory than text questions.

Likewise, when there is inconsistency in patient responses, for example when there is a screening question where the patient states they are not sleepy but later tells us that they consume large amounts of caffeine because of sleepiness. In this situation, the system asks the patient to clarify the inconsistency in stores the response accordingly.

In this way, an intelligent history can be acquired which gathers key information and activates important "red flags" for the clinician. This is a rules-based system and the rules are made by experts. In order to express this information in a meaningful manner, we found it necessary to expand the FloBase language to also generate reports.

Our system can apply a second set of rules to the data which was just acquired. For example, a patient might describe chest pressure and elsewhere in the review of systems might mention hoarseness and in another section they might report insomnia. Although each item might be reported within each of the respective sections of the Review of Systems, they may also be combined in a discussion which combines chest discomfort, insomnia and heartburn, presenting it in such a way that GERD needs to be considered. The same is true for shortness of breath which may be correlated with cardiac symptoms in the review of systems but also be correlated with respiratory symptoms or anxiety in other sections.

Such a rules-based system can be applied to existing guidelines which lead to specific diagnoses can be found in sources such as the DSM-IV, for example.

A further extension of our automated history is to ask pertinent and relevant questions across subsequent visits. The data points which are acquired in each patient encounter with the program, such as symptoms or diagnoses, can be followed as a thread from visit to visit.

The road to a practical patient computer interview... description of a long experience... by Edmund Messina MD
As physicians, we are trained in the art of conversing with patients, which allows us to establish a rapport, assess the ability of the patient to give us reliable information and observe the patient's behaviors as we collect the history. The interaction between doctor and patient cannot be replaced by a machine. The problem is, we have limited amounts of time to converse with our patients...shortcuts are taken and data is missed. In medicine, the errors of omission are often worse than the errors of comission. It is in the nature of the physician to try to do a good job and pressures placed upon these physicians through insurance companies, employers and other sources, cause these shortcuts to be taken. The patient is not always aware of what facts may be relevant and unless we skillfully and patiently probe for specific information, as well as screening information, early detection of many diseases can be missed. We commonly consider two types of patient information gathering; one is the acquisition of structured information from structured questions. The other is free conversational or "open-ended" questioning. Clinicians have tried to gather structured information through paper-based checklists, administered by the patient themselves or through an assistant. A common problem is the lack of completeness or the inability of the patient to understand the terminology. Certainly, paper-based systems are unable to probe deeper into positive responses and do not explain inconsistencies. In addition, we overestimate the literacy as well as health literacy of many patients. With the advent of computing, electronic methods have evolved which allow patient interrogation through the use of computers. My group has been developing and using such methods since the mid-1980s, and these methods have become more sophisticated over time. In those days, automated questionnaires worked better than simple paper questionnaires but a weakness was still apparent in those which asked simple questions or had simple branching. I became very interested in the process of asking questions, as a clinician, and as a teacher. It became very apparent to me that an artificial intelligence method was needed. My colleagues and I subsequently developed the FloBase® language for the purpose of simulating expert interrogation. With such a system, questions could be asked according to specific rules, as developed by clinical experts. The program selects questions based upon previous answers, patient demographics, existing symptoms, or even as the program begins to form hypotheses along a certain line of inquiry. Our next step was to develop a more personal interaction between patient and computer. A strength of the computer interview is the patient's perception of anonymity and the absence of any judgemental facial expressions since there was no live interviewer. A weakness was the impersonal text interface. Over the years, we found it very useful for the patient to interact with a human face who is able to delve more deeply into specific sensitive areas. In addition, when the patient expresses difficulty understanding the question, the video nurse can clarify the question and ask it in different terms, to elicit a more accurate response from the patient. In addition, many people are more likely to understand auditory than text questions. Likewise, when there is inconsistency in patient responses, for example when there is a screening question where the patient states they are not sleepy but later tells us that they consume large amounts of caffeine because of sleepiness. In this situation, the system asks the patient to clarify the inconsistency in stores the response accordingly. In this way, an intelligent history can be acquired which gathers key information and activates important "red flags" for the clinician. This is a rules-based system and the rules are made by experts. In order to express this information in a meaningful manner, we found it necessary to expand the FloBase language to also generate reports. Our system can apply a second set of rules to the data which was just acquired. For example, a patient might describe chest pressure and elsewhere in the review of systems might mention hoarseness and in another section they might report insomnia. Although each item might be reported within each of the respective sections of the Review of Systems, they may also be combined in a discussion which combines chest discomfort, insomnia and heartburn, presenting it in such a way that GERD needs to be considered. The same is true for shortness of breath which may be correlated with cardiac symptoms in the review of systems but also be correlated with respiratory symptoms or anxiety in other sections. Such a rules-based system can be applied to existing guidelines which lead to specific diagnoses can be found in sources such as the DSM-IV, for example. A further extension of our automated history is to ask pertinent and relevant questions across subsequent visits. The data points which are acquired in each patient encounter with the program, such as symptoms or diagnoses, can be followed as a thread from visit to visit.		The automated return visit allows symptoms to be followed over time, especially in situations where quantitation as possible. For example, the history quantitates number of cigarettes, BMI, drowsiness scale, depression scale and other measurable symptoms, and trends can be pointed out to the patient and to the clinician. Correlations between increasing symptoms of depression and increasing BMI or ethanol intake become obvious. A number of years ago, an excellent review was written by Dr. Bachman in the Mayo Clinic proceedings (The Patient-Computer Interview: A Neglected Tool That Can Aid the Clinician, JOHN W. BACHMAN, MD, Mayo Clin Proc. 2003;78:67-78). Dr. Bachman pointed out that traditional history taking is often incomplete and time-consuming for collection of information and also for documenting it. Over the years, I've discovered that patients are quite inaccurate with paper questionnaires because of their own health literacy or their actual literacy or understanding of the English language. Many of these deficiencies can be compensated through the use of interactive video, as we have found in our own experience. Dr. Bachman's article clearly illustrates the comparison between the different history taking techniques, showing the obvious advantages of interactive computer histories. There are many problems with conventional paper questionnaires: 1) Questionnaires are not complete or exhaustive, for obvious reasons. They also may contain irrelevant questions, such as asking males about their periods. 2) Patients have a tendency to skip quickly through lists of questions 3) Updates and maintenance of paper questionnaires is very difficult since specific versions would have to be defined 4) Questionnaires are too plentiful and primary care physicians are overwhelmed with stacks of questionnaires which deal with various issues, risk factors, diagnoses, etc. 5) Most physicians do not bother to use questionnaires because they're impractical. Most common place for their use is on a first visit our first encounter. 6) Questionnaires do not clarify symptoms or resolve inconsistencies By comparison, automated interactive computerized histories are extremely structured, are self-explanatory and self clarifying, and serve as excellent documentation into the medical record. They save physician time because the patient can answer questions by themselves or, in special-needs situations, with an assistant or family member. Data points identified in the patient's history, especially where inconsistencies arise, can serve as triggers for future patient education. It is been shown that computerized histories can document more information than a conventional physician interview. Although a significant time commitment is required on the part of the patient, the patient interface with the Arbor Medicus system can be broken down into multiple sessions. Since personal computers and Internet connections are plentiful and inexpensive, patients can access this automated system from home, public libraries or in designated rooms in the clinic. In addition to the empowerment the patient will experience on clearly expressing their problems, the automated history can also provide guidance or "speaking points" to better prepare the patient for their next medical visit. The patient history can be physically or electronically pasted into the physicians records, saving dictation costs. The accuracy of the patient computer interview is further enhanced by its nonjudgmental properties. The video nurse's nonthreatening and apathetic. She is not judgmental, reveals no disapproval with facial expressions nor does she appear to be morally judgmental. The interactive video componant breaks down barriers caused by culture, language and literacy. Today's physician need not feel threatened by artificial intelligence computerized interviews. So far, at least, computers are not capable of detecting nonverbal behaviors from the patient being interviewed. Although the computer is quite impersonal, the patient interacting with the video, will have a warmer experience and, of course, understand that this is simply data gathering and does not take the place of a face-to-face interview with a health professional. The actual interview of the patient, following computerized data acquisition, is much more personalized because the structured questioning is out of the way. The physician is better able to use open-ended questions or pursue deeper inquiry about areas exposed by the electronic interview. Electronic history taking does replace a physician but it is a valuable tool which can save considerable time. EM






Copyright 2009, Edmund Messina MD . All rights reserved. No part of this program may be reproduced in any manner whatsoever without written permission from the publisher, except as indicated in the instructions herein.
webdesign by *miya bruno*

The road to a practical patient computer interview... description of a long experience...

by Edmund Messina MD