Proceedings. AMIA Symposium最新文献

Too often, online searches for health information are time consuming and produce results that are not sufficiently precise to answer clinicians' or patients' questions. The PERSIVAL project is designed to circumvent this problem by personalizing and tailoring searches and presentation to the demands of the user and the particular clinical context. This paper focuses on a cognitive evaluation of one component of this project, a Query User Interface (QUI). The study examines the system's ability to allow users to easily and intuitively express their information needs. We performed several analyses including a cognitive walkthrough of the interface and quantitative estimations of cognitive load. The paper also presents a preliminary analysis of usability testing. The analyses suggest that there are features in the QUI that contribute to a greater cognitive load and result in greater effort on the part of the subject. The results of usability testing are consistent with these findings. However, subjects found it to be relatively easy and intuitive to generate well-formed queries using the interface. This study contributed to the iterative design of the interface and to the next generation of the PERSIVAL system.

This paper reports the methods and results of a computer-based algorithm that takes as input the expression levels of a set of genes as given by DNA microarray data, and then searches for causal pathways that represent how the genes regulate each other. The algorithm uses local heuristic search and a Bayesian scoring metric. We applied the algorithm to induce causal networks from a mixture of observational and experimental gene-expression data on genes involved in galactose metabolism in the yeast Saccharomyces cerevisiae. The observational data consisted of gene-expression levels obtained from unmanipulated inverted exclamation mark degrees wild-type inverted exclamation mark +/- cells. The experimental data were produced by deleting ( inverted exclamation mark degrees knocking out inverted exclamation mark +/-) genes and measuring the expression levels of other genes. We used this data to evaluate several variations of the local search method. In each evaluation, causal relationships were predicted for all 36 pairwise combinations of nine key galactose-related genes. These predictions were then compared to the known causal relationships among these genes.

Genes and proteins are often associated with multiple names, and more names are added as new functional or structural information is discovered. Because authors often alternate between these synonyms, information retrieval and extraction benefits from identifying these synonymous names. We have developed a method to extract automatically synonymous gene and protein names from MEDLINE and journal articles. We first identified patterns authors use to list synonymous gene and protein names. We developed SGPE (for synonym extraction of gene and protein names), a software program that recognizes the patterns and extracts from MEDLINE abstracts and full-text journal articles candidate synonymous terms. SGPE then applies a sequence of filters that automatically screen out those terms that are not gene and protein names. We evaluated our method to have an overall precision of 71% on both MEDLINE and journal articles, and 90% precision on the more suitable full-text articles alone

All clinical simulation designers face the problem of identifying the plausible diagnostic and management options to include in their simulation models. This study explores the number of plausible diagnoses that exist for a given case, and how many subjects must work up a case before all plausible diagnoses are identified. Data derive from 144 residents and faculty physicians from 3 medical centers, each of whom worked 9 diagnostically challenging cases selected from a set of 36. Each subject generated up to 6 diagnostic hypotheses for each case, and each hypothesis was rated for plausibility by a clinician panel. Of the 2091 diagnoses generated, 399 (19.1%), an average of 11 per case, were considered plausible by study criteria. The distribution of plausibility ratings was found to be statistically case dependent. Averaged across cases, the final plausible diagnosis was generated by the 28th clinician (sd = 8) who worked the case. The results illustrate the richness and diversity of human cognition and the challenges these pose for creation of realistic simulations in biomedical domains.

The combination of a) our changing understanding of genotypic and phenotypic classification of diseases and b) the rapid growth and expansion of the number of entries in two databases targeted toward clinicians resulted in the need to develop a flexible dynamic hierarchical classification system for genetic disorders. The two databases making use of this classification schemas are the GeneClinics (GC) database - www.geneclinics.org and the GeneTests (GT) database - www.genetests.org The GC and GT databases serve respsectively as the users manual and yellow pages of genetic testing. The GeneTests/GeneClinics (GT/GC) classification hierarchy is maintained as a simple set of parent/child relationships in a relational database. The hierarchy is generated in real time in response to a user request. It is not maintained as a set of members with relationships defined by characters that are parsed to determine the structure of the tree. The GT/GC classification hierarchy entries are handled as objects by the data maintenance and search tools and may have a number of attributes and associations that create a rich tool for defining and examining genetic disorders

Disseminating information from large biomedical databases can be crucial for research. Often this data will be patient-specific, and therefore require that the privacy of the patient be protected. In response to this requirement, HIPAA released regulations for the dissemination of patient data. In many cases, the regulations are so restrictive as to render data useless for many purposes. We propose in this paper a model for obfuscation of data when served to a client application, that will make it extremely unlikely that an individual will be identified. At Partners Healthcare Inc, with over 1.4 million patients and 400 research clinician users, we implemented this model. Based on the results, we believe that a web-client could be made generally available using the proposed data obfuscation scheme that could allow general usage of large biomedical databases of patient information without risk to patient privacy.

To address the growing problem of novices accessing incomplete and unreliable information on the Web, this paper begins by probing the nature of strategic knowledge known by expert healthcare searchers. The analysis reveals the existence of a strategy that is focused to a specific type of healthcare question. To investigate whether there exist other similar strategies, we conduct two pilot studies. The first leads to the development of a taxonomy for skin cancer questions based on real-world questions. This taxonomy has high inter-rater agreement when tested with new real-world questions. The second pilot helps to identify 6 strategies that are specific to question types in the taxonomy. We demonstrate how these strategies can be made available to a wide range of users through a prototypical system referred to as a Strategy Portal. In conclusion, we briefly describe our current and future research to test if such a system can make patients more effective and efficient in the retrieval of reliable and relevant healthcare information on the Web.

Current models for predicting outcomes are limited by biases inherent in a priori hypothesis generation. Knowledge discovery algorithms generate models directly from databases, minimizing such limitations. Our objective was to generate a mortality model from a PICU database utilizing knowledge discovery techniques. The database contained 5067 records with 192 clinically relevant variables. It was randomly split into training (75%) and validation (25%) groups. We used decision tree induction to generate a mortality model from the training data, and validated its performance on the validation data. The original PRISM algorithm was used for comparison. The decision tree model contained 25 variables and predicted 53/88 deaths; 29 correctly (Sens:33%, Spec:98%, PPV:54%). PRISM predicted 27/88 deaths correctly (Sens:30%, Spec:98%, PPV:51%). Performance difference between models was not significant. We conclude that knowledge discovery algorithms can generate a mortality model from a PICU database, helping establish validity of such tools in the clinical medical domain.

Episode creation, the task of classifying medical events and related clinical data to a high-level concept, such as a disease, illness or care, has been primarily an interest of healthcare payers for purposes of cost outcomes analysis. Traditional challenges in episode creation have included: inconsistencies in defining episodes; lack of sufficient information to infer episodes; and differences in methods for diagnosing and resolving episodes. However, with the advent of the electronic medical record, which contains multiple sources of patient-related information, data is now accessible to construct more accurate and refined episodes. This work presents a context-sensitive episode creation methodology that utilizes features extracted from different medical repositories (e.g., claims records, structured medical reports) to associate the data with their respective motivating episodes. The combinatorial approach used to find the optimal clustering of patient-related data into episode groups and the measure used to evaluate candidate episode sets are described.

Clinicians encounter many medical questions while providing outpatient medical care. A significant number of these questions can be answered using MEDLINE; however it has proven to be difficult to incorporate MEDLINE into routine clinical workflow and for clinicians to generate well constructed MEDLINE queries. This study however hypothesized that that well-constructed MEDLINE queries could be semi-automatically generated by an application named LitButton which was incorporated into the TeleWatch telemedicine system. The LitButton application was then prospectively evaluated in a pilot study by four nurse case managers (NCM) who monitored sixty-eight outpatients for three weeks. During this period the NCMs used the LitButton application sixteen times, and they subjectively reported in real-time that they obtained an answer in 56% of the cases, but that none of the successful information retrieval events resulted in a change in a patient's clinical management. The small number of LitButton events and lack of clinical impact was likely due to the fact that the LitButton function was designed to search MEDLINE for treatment related information; however the NCMs had limited medical decision making responsibilities. As a result there was a mismatch between the user's information needs and the system capabilities.