Persons aged 65 years or older are the fastest growing portion of the population in the United States. However, there are disproportionately few specialists in geriatric medicine to provide the majority of these individuals with high quality medical care. One solution to this problem is through the development and use of computer-based medical expert systems. Based on techniques of artificial intelligence, such systems have the ability to assist non-specialist clinicians or allied health professionals in the preliminary evaluation and diagnosis of common disorders in the elderly. Depression and dementia are the two most prevalent of such disorders, yet these conditions frequently are misdiagnosed by non-specialist clinicians. Failure to detect or distinguish the two disorders leads to missed treatment opportunities and preventable disability. The proposed project will compare the feasibility, statistical accuracy, and user acceptance of two clinical decision support systems -- a rule-based expert system and a neural network computational classifier - to be used in the evaluation of depression and dementia in the elderly. The user interface for the two systems will be designed and tested to meet usability criteria desired by practicing clinicians, including voice-recognition input, computer- speech output, interactive graphic environments, ranked differential diagnoses, and the ability to explain the system's reasoning, conclusions, and recommendations. The system will include an automatic foreign language translator initially programmed for Spanish, but customizable to other languages. The project will be conducted in four phases: 1) Knowledge Acquisition, 2) Prototype Development, 3) Usability Testing, and 4) System Validation. Specific goals include: 1) efficient implementation of diagnostic rules and criteria obtained from clinician interview and literature review, 2) successful development of a prototype system to meet formalized usability standards, 3) preliminary validation of the system to classify with 90 percent accuracy a large set of existing clinical data obtained from referrals to a medical school geriatric psychiatry clinic, and 4) statistical comparison of the classification accuracy of the prototype expert system with the neural network classifier. The completed expert system will form the foundation for development of future clinical modules applicable to other geriatric disorders. Ultimately, systems developed from this prototype will provide diagnostic and treatment recommendations in medical settings without ready access to clinical specialists, such as in rural health care or under-served urban areas.
