Since the invention of the electronic computer there have been 2 divergent opinions as to how this device should be used. One group saw the computer's principal use as a number manipulator, an extensive, ultrafast, and accurate calculator. Another group envisioned the computer as a symbol manipulator that might be taught to use logic and make decisions in a human fashion. The symbol-manipulation group eventually founded the discipline called artificial intelligence (AI). As time passed, artificial intelligence spawned many subdisciplines such as robotics, natural language recognition, and expert systems. The driving force behind expert-system development was the attempt to program a computer so it could "think" and make human-like decisions. A generally-accepted definition of an expert system is one that performs at or near the level of a human expert in a particular field of endeavor [6].

The first large expert system to perform at the level of a human expert was developed for use in the health care field [7]. This system, called MYCIN, was developed at Stanford University in the mid 1970s. The MYCIN program was designed to provide physicians with advice about bacteremia and meningitis; it emulated the expertise from a physician specializing in these infections. MYCIN asks the physician, who is requesting its advice, several questions. Additionally, it requires personal data about the patient and will ask for results of laboratory tests when it needs this information to aid in its analysis process. After analyzing patient information and data, MYCIN makes a diagnosis as well as a drug-therapy recommendation. MYCIN also has the ability to explain how it arrived at its diagnosis [8].

The architecture of the MYCIN program consists of 2 major, distinct elements: (1) a knowledge base and (2) an inference engine. The knowledge base contains the facts and heuristics necessary to derive medical analyses of bacteremia and meningitis infections. The inference engine contains the inference mechanisms and control strategies employed by MYCIN to derive the diagnosis and treatment recommendation. It soon became clear that this architecture was not just specific to MYCIN but could be used as a template or "shell" for encoding and applying expertise in any field.

One of the most interesting applications of expert system technology is for education. One goal among some artificial-intelligence researchers has been to develop a computerized tutor that performs at the level of an excellent human tutor. The resulting computer program that achieves this objective would be called an intelligent tutoring system (ITS). The effort to create truly-intelligent computer-based tutors has been underway for many decades [9]. Ultimately intelligent tutoring systems attempt to simulate the behavior of an intelligent human tutor in addition to acting as a domain expert. The characteristics of an intelligent tutoring system include the ability [10,11] to:

Although the development of expert systems for training has been moderately successful and other kinds of expert systems, such as income-tax preparation assistants, have been extremely successful commercially, intelligent-tutoring-system development did not advance significantly for several years after its initial efforts. It is within the last decade that prospects for intelligent tutoring systems have shown promise. Advancements in hardware, operating systems, and interactive-multimedia development tools have created a software environment that makes real intelligent tutoring systems a possibility [12- 14].

Interactive multimedia software is already playing a key and unique role in the educational process. It is used as a stand-alone educational module that is intended to enable an interested individual to learn about a particular topic or subject [15]. It has been used as a supplement to classroom presentations [16,17] and laboratories [18,19], and it is used as a dynamic textbook [20]. Recently, interactive multimedia have been integrated with expert-system technology, producing highly-interactive intelligent tutoring systems [21- 25]. These intelligent tutoring systems "learn" about the individual student and can tailor the material to meet the needs and the learning styles appropriate to the individual student.

One of the first successful large-scale expert systems was developed for diagnosis and treatment of certain infections [26]. The principles of this approach were subsequently applied to include the application of artificial intelligence techniques to medical diagnosis (for example, [27- 32]). Artificial intelligence research within medicine is very active today although attempts to develop physician-like diagnostic systems have waned. Instead, these systems are now geared to assist students and physicians to develop and hone diagnostic skills.

Because enthusiasm for medical diagnosis systems that would "assist" physicians and might even dispute their diagnoses was tepid, the development of systems designed for practitioner education has been advanced. Consequently, considerable effort has been made to produce software systems that could simulate medical situations and include a computerized mentor that would assist learners in their decision-making process. The rapid development of computer hardware has eliminated many of the early problems encountered in developing medical-training expert systems. Today's graphical user interfaces coupled with developments in various video and audio digital circuit boards have encouraged the development and use of multimedia software. Multimedia is usually defined as "the use of a computer to present and combine text, graphics, audio and video with links and tools that allow the user to navigate, interact, create, and communicate" [26]. Multimedia programs have been written to help educate medical practitioners in almost every aspect of health delivery. Systems exist to train and educate individuals in the principles of: laparoscopic procedures and epilepsy diagnosis and treatment [33], ureterorenoscopy [34], hysteroscopy [35], dermatology [36], diabetes management [37], communicable and rare diseases [11], red cell antibody identification [38], cervical cytology [39], psychotherapy [40], and many other health-related areas.

Besides educating practitioners, progress has been made recently in developing interactive multimedia software products for patient use and education. Agre et al, reporting on a project that develops a CD-ROM for cancer-related patient education [41] notes that CD-ROMs are more valuable for learning than booklets or videotapes. The study finds that CD-ROM technology allows for greater depth in content and has the ability to satisfy a broad range of educational needs. Guendelman et al [42] have used a computer-based self-management program called Health Buddy to enable children to assess and monitor their asthma symptoms and quality of life as well as to transmit this information to health care providers. Using a randomized control trial, 66 individuals were placed in the intervention group while 68 were in the control group. The study found that compared with an asthma diary, monitoring asthma symptoms and functional status with the Health Buddy increases self-management skills and improves asthma outcomes. Another recently-developed CD-ROM focused on the education of children with leukemia (ages 4-11) and their families about the disease and its treatment [43].

Multimedia has also been used, with some success, for educational intervention for low-education, low-income Latinas [44]. At the conclusion of the study, it was found that about 40% of the women that had scheduled or received mammography had attributed their decision to the intervention. Interactive multimedia has been used to deliver information: about self-medication [45], for patient colonoscopy education [46], for genital herpes education [47], for personal care of diabetes [48], for skin cancer prevention [49], for nutrition screening and counseling [50], for hypertensive patients [51], and for numerous other diseases or treatments.

Gustafson et al have developed a computer-based health information and support system [52,53]. This Comprehensive Health Enhancement Support System (CHESS) is a Web-based support system that includes modular programs on breast cancer, AIDS/HIV infection, sexual assault, alcoholism, and academic crisis. For example, the breast cancer module provides disease information, a treatment decision aid, an opportunity to contact health care providers via e-mail, testimonials from patients, and a patient forum to exchange information and to solicit social support. Evaluation studies have confirmed that the CHESS system is user-friendly and well accepted by patients of diverse demographic background [54]. The use of the CHESS system has also been linked to improvements of quality of life, and increased participation in health care among HIV-positive patients [54]. Despite its sophistication, the CHESS system lacks the capacity to present information in a targeted and tailored manner based on specific patient characteristics.

A decade ago researchers [10] noted that computer software has the potential to elicit information from patients and to use this information to tailor patient education. Because the literature is not always consistent in the use of the terms targeted and tailored with respect to health communications, a recent article has suggested clarification and standardization in terminology [55]. The authors suggested that health communications be labeled as targeted if they are directed towards a particular subgroup of the population, and to use the term tailored if the information is designed to address individual characteristics of a person. Both targeting and tailoring depend on the assessment of group or individual characteristics that have been derived from prior data collections [55].

The goal of targeting and tailoring health communications is to increase the personal relevance of the message to the individual. Research has demonstrated that messages that are relevant to the person are better understood and better remembered [56,57]. In addition, personally-relevant messages are also more likely to lead to behavior change [58]. Thus, it is generally expected that an individual will react to tailored messages with increased attention, improved understanding, and an increased tendency for behavior change [54,59].

The prostate interactive education system (PIES) is an interactive multimedia expert system that is being developed to help patients who have been diagnosed with early-stage prostate cancer. PIES provides the patient with a multitude of treatment information and encourages patients to obtain the kind of information that they desire about the disease. PIES is envisioned as an important step in the treatment-decision process. The software will enable a patient to follow a logical decision process and obtain the information he needs at his current state of decision-making. The software will prepare a factual report for a psychologist who will meet with the patient following his interaction with the software. The temporal placement of the software is shown in Figure 1. Once the patient is diagnosed with prostate cancer, he is offered the PIES software package to assist him in making a treatment decision. PIES provides additional information and does not preclude a second opinion, as shown in Figure 1. In fact, the patient may receive a second opinion before using PIES.

The purpose of the PIES software is threefold:

PIES is not a diagnosis tool. Nor does it give advice in the form of "this is what you should do." Instead, it allows patients to ask questions and get information by interacting with a multimedia computer program. The multimedia program allows a patient to visit a library to get up-to-date information about prostate cancer and its treatment. The patient will be able to interact with virtual physicians who can answer specific questions that the patient may have. Various physicians (eg, surgeons, radiologists) can be queried by the patient. The physicians "consult" with the patient through digital video sequences as well as through interactive multimedia question-and-answer sessions. The patient will also be able to "visit" a virtual support room where he will have the chance to listen to other patients talking about the disease and how they determined their own course of treatment. Support group members also discuss dealing with the side effects of their treatment. The patient will also be able to consult with a virtual sex therapist who will discuss methods and assistive technologies that will enable them to continue sexual activities. Finally, each patient will be provided with a decision aid that will assist him in treatment decision making.

An overview of the PIES architecture is shown in Figure 2. The patient interacts with the CD-ROM based system on a personal computer. The system is CD-ROM based but could be made available over the Internet. A CD-ROM delivery system was chosen so that those patients without a high-speed connection to the Internet would not experience delays caused by downloading videos. The information is developed and presented using Macromedia's Authorware [60]. Authorware is able to communicate with other software and can transmit information in real time or near real time. Authorware will allow the patient to have access to other programs such as a notebook to record information, and various applicable medical-applications software.

The patient model, described later, is a computer characterization of the patient; it is changed dynamically by the expert system. In turn, the expert system, also described later, analyzes how the patient interacts with the software and reports this in the results folder. The results folder contains information that the expert system will use to generate a report to the psychologist.

The paradigm used in PIES is that of a patient entering a health center. When the patient enters PIES, he is greeted by an information specialist. The information specialist welcomes the patient and shows him around the PIES Health Center. The Health Center consists of a reception area, a library, and a group meeting room on the first floor, and physician offices on the second and third floors, connected through an elevator. Each room is interactive and the patient is shown how to use the room's facilities when he first chooses to enter the room. After showing the patient the Health Center layout, the information specialist asks the patient to complete a questionnaire. The questionnaire requests data that the expert system will need to tailor the Health Center for his needs.

After completing the questionnaire, the patient is free to visit any room of his choosing.

Improved communication between physicians and patients and improved information provision about disease and its treatment are important factors in aiding patients to make treatment decisions and to cope with their disease [21,66,67]. Although, patients are generally interested in obtaining information about their disease, not all patients desire the same level of information [68,69] and thus an individualized approach to information provision is more appropriate to meet a patients information needs [70,71].

Research has identified 2 main profiles of information processing: high monitoring (information seeking) and low monitoring (information distracting). When confronted with medical information, high monitoring individuals typically process information very attentively and amplify its threatening aspects. High monitors prefer more information about their condition and its treatment, are consequently better informed, yet express greater dissatisfaction about the amount of information they receive [68]. Overall, they demand more support and reassurance [68,72].

In contrast, low monitors demonstrate an opposite pattern; they are more likely to ignore, avoid, and minimize threatening cancer-related information that diverges from their belief that they will be fine [41,73- 76]. Consequently, low monitors prefer and are satisfied with less information about their condition, are less knowledgeable about their disease status, and tend to underestimate the risk from the disease threat.

The expert system determines if the patient is a low monitor (ie, low information seeker), one who wants only the important points of a treatment, or a high monitor (ie, high information seeker), one who wants considerable detail about a treatment therapy. The expert system directs the interactive multimedia program to provide the amount of information that the patient desires. The expert system realizes that a patient might want to be a high monitor in some areas and a low monitor in others. For example, a patient may be very interested in knowing all about the side effects of external beam radiation while wanting only the essentials about the procedure itself. To our knowledge, PIES is unique in that it allows dynamic tailoring. Consequently, as patients proceed through PIES, its expert system determines how much information patients desire about a topic and provides it.

Before developing PIES we conducted a survey among the target audience (ie, men diagnosed with localized prostate cancer) to establish familiarity with computers. We mailed a brief questionnaire to 675 participants of an ongoing longitudinal study of prostate-cancer treatment decision making and quality of life [5].

We developed a demonstration CD of PIES to showcase its features to several focus groups. The first 3 focus groups consisted of 18 prostate cancer survivors. Men were recruited from the same ongoing longitudinal study mentioned above [5].

Of the 675 survey questionnaires mailed, 473 were returned (70% return rate). Demographic characteristics are listed in Table 1.

The majority (65%) of participants reported they have a PC at home, and an additional 7% indicated they had access to a PC at a public place, such as at work or at the library. Access to a computer came with access to the Internet for 70% of the respondents. When asked to rate their experience in using a computer, 80% of patients indicated they are familiar with computers. They rated their expertise from slightly experienced (39%), moderately experienced (18%), very experienced (11%), to expert (2%). Three quarters (76%) further indicated that they had access to a family member or a friend to help them with potential computer problems. When asked about any physical problems that might prevent them from using a computer, 89% of patients reported not having any eye problems that might interfere with using a computer, and 83% indicated they have no difficulty at all in operating a pointing device such as a mouse (10% indicated having "a little problem" with a mouse). Patients who have a computer at home spend an average of 3 hours (SD = 1.5 hours) per week using it. About 20% spend more than 2 hours per day on the computer. Thus, our target audience for PIES is experienced in using a computer.

Of the 18 men, 7 (39%) were African American. We oversampled African American men into the focus groups, in part because, prostate cancer is particular prevalent among African Americans and thus the use of the program might be specifically relevant to this group. Men were on average 67 years old and had at least a high-school education (33%), a large majority was married (83%), and had completed treatment (83%). External beam radiation was chosen by 72%, 22% chose surgery, and 6% chose brachytherapy. The remaining 2 focus groups were held with spouses of prostate cancer survivors (N = 15). The women were on average 60 years old and 50% had a college or postgraduate degree. All focus group participants received $40 as a token of appreciation.

Overall feedback on the demonstration CD was extremely positive. All ratings were performed on a 5-point scale, with higher scores indicating higher levels of interest. Patients and spouses uniformly stated that they were "very much" interested in the software (mean = 4.71; SD = 0.59; range, 3-5), and that it was "very" useful (mean = 4.71; SD = 0.47; range, 4-5). When asked whether they would prefer the software in lieu of traditional information sources, such as books or booklets, 83% stated a clear preference for the software program over print material, and 100% would prefer the program in addition to print material. On average, patients indicated that they would spend between 1 and 2 hours with the program and were willing to pay an average of $50 for it, if it were commercially available. A large majority of patients (82%) indicated that they had access to a personal computer and that they frequently use it (41% everyday, 29% 3-4 times a week, 18% 1-2 times a week); 94% had access to the Internet. In general, patients use their computer for e-mail (71%), and to obtain general (77%) and health (71%) information. Almost 60% have used the Internet to obtain prostate cancer information.

In summary, both men and spouses were enthusiastic about the program. They expressed great interest and regretted not having had such a tool when they had to make a treatment decision. When given the choice between traditional information sources such as books or pamphlets almost every participant indicated a preference for the CD-ROM. Of particular interest to patients were the physician's office and the support-group room, which offered patients expert as well as lay perspectives of prostate cancer treatments. Although men expressed proficiency with computers in general, it became clear that additional navigational and organizational help would enhance the usability of PIES.

Overall, the focus group members evaluated PIES very positively and provided important suggestions about potential changes and enhancements to the program. Based on these comments and in accordance with health-behavior and communication theories [7,77] we are planning to enhance the next version of the PIES software by including a virtual information specialist and interactive tutor (VISIT). This information specialist and tutor will be designed to further enhance the interactivity between user and software. Specifically, VISIT will proactively query the user about what kind of information he desires, inform him where he can obtain the information, and will offer to retrieve it for him. VISIT will further guide the user through the program by offering relevant information that had not been accessed, thus anticipating the user's informational needs. Furthermore, while offering treatment information that could be upsetting to some patients the program will also offer emotionally-reassuring and normalizing messages.

VISIT will be represented as a health care professional (see Figure 7) through video and still photos. Once the patient enters the health center, she will introduce herself and will give an overview of the center. She will offer suggestions on how to best use the program, depending on the user's needs. For example, if a patient is interested in hearing a physician's opinion, VISIT will offer to show him the way to the doctor's office. In contrast, if the patient is interested in learning about different treatment modalities in great detail, VISIT will accompany him into the library. Once there, VISIT will query the user about which treatment options he might want to learn about first and which aspects he is most interested. The corresponding information will then be presented.

Our preliminary evaluation has some limitations. First, focus group members evaluated a demonstration CD, designed to elicit input about the metaphor, its presentation, and possible content. These individuals had successfully completed treatment and thus their information needs were not as immediate as those of just-diagnosed patients. Yet, a number of survivors and spouses mentioned that they wished they had PIES while they were learning about prostate cancer. It should also be pointed out that PIES is intended to serve as an additional patient-education tool. It is not designed to substitute for personal consultations with a physician. Indeed numerous statements will be made to encourage the patient to discuss specific individual medical issues with a health care provider.

The next step in the evaluation of PIES will involve extensive usability testing with a patient sample. We will use the methodology and procedures set forth by the National Cancer Institute's own usability program (Usability.gov, [78]). Patients will be asked to talk aloud while exploring an alpha version of PIES, detailing what type of information they are looking for, where they think they can find it, and their reactions to it. Based on current usability standards we estimate that a small number of users (ie, approximately 5 patients) is sufficient to detect the majority of bugs. All suggestions from the usability version will be incorporated into a beta version of the program. This version will then be evaluated for efficacy in a randomized controlled trial, in which PIES will be evaluated against standard care (ie, information provided through brochures and pamphlets).

In this paper we have introduced a novel interactive multimedia system for prostate cancer education and treatment decision making. PIES is based on current psychological theory incorporating tailoring to information-seeking preferences with the latest software technology. The program is highly interactive, combining user input with animation, video, text, and still photos. Preliminary evaluation through focus groups has established the validity of our approach and has pointed to new directions to further enhance the user interface through the development of a virtual information specialist and interactive tutor.