Remote image based retinopathy of prematurity diagnosis: a receiver operating characteristic analysis of accuracy.

BACKGROUND/AIMS: Telemedicine offers potential to improve the accessibility and quality of diagnosis of retinopathy of prematurity (ROP). The aim of this study was to measure accuracy of remote image based ROP diagnosis by three readers using receiver operating characteristic (ROC) analysis. METHODS: 64 hospitalised infants who met ROP examination criteria underwent two consecutive bedside procedures: dilated examination by an experienced paediatric ophthalmologist and digital retinal imaging with a commercially available wide angle camera. 410 images from 163 eyes were reviewed independently by three trained ophthalmologist readers, who classified each eye into one of four categories: no ROP, mild ROP, type 2 prethreshold ROP, or ROP requiring treatment. Sensitivity and specificity for detection of mild or worse ROP, type 2 prethreshold or worse ROP, and ROP requiring treatment were determined, compared to a reference standard of dilated ophthalmoscopy. ROC curves were generated by calculating values for each reader at three diagnostic cut-off levels: mild or worse ROP (that is, reader was asked whether image sets represented mild or worse ROP), type 2 prethreshold or worse ROP (that is, reader was asked whether image sets represented type 2 prethreshold or worse ROP), and ROP requiring treatment. RESULTS: Areas under ROC curves ranged from 0.747-0.896 for detection of mild or worse ROP, 0.905-0.946 for detection of type 2 prethreshold or worse ROP, and 0.941-0.968 for detection of ROP requiring treatment. CONCLUSIONS: Remote interpretation is highly accurate among multiple readers for the detection of ROP requiring treatment, but less so for detection of mild or worse ROP.

Making grandma's data secure: a security architecture for home telemedicine.

Home telemedicine presents special challenges for data security and privacy. Experience in the Informatics for Diabetes Education And Telemedicine (IDEATel) project has demonstrated that data security is not a one-size-fits-all problem. The IDEATel users include elderly patients in their homes, nurse case managers, physicians, and researchers. The project supports multiple computer systems that require a variety of user interactions, including: data entry, data review, patient education, videoconferencing, and electronic monitoring. To meet these various needs, a number of different of security solutions were utilized, including: UserID/Password, PKI certificates, time-based tokens, IP filtering, VPNs, symmetric and asymmetric encryption schemes, firewalls and dedicated connections. These were combined in different ways to meet the needs of each user groups.

When seconds are counted: tools for mobile, high-resolution time-motion studies.

Time-motion (TM) studies are often considered the gold-standard for measurements of the impact of computer systems on task flow and duration. However, in many clinical environments tasks occur too rapidly and have too short of a duration to be captured with conventional paper-based TM methods. Observers may also with to categorize caregiver activities along multiple axes simultaneously. This multi-axial characteristic of clinical activity has been modeled as multiple, parallel finite-state sets and implemented in three computerized data collection tools. Radiology reporting is a domain in which tasks can be characterized by multiple attributes. A radiologist may also switch among multiple tasks in a single minute. The use of these tools to measure the impact of an Automated Speech Recognition (ASR) system on Radiology reporting is presented.

A Web-based, secure, light weight clinical multimedia data capture and display system.

Computer-based patient records are traditionally composed of textual data. Integration of multimedia data has been historically slow. Multimedia data such as image, audio, and video have been traditionally more difficult to handle. An implementation of a clinical system for multimedia data is discussed. The system implementation uses Java, Secure Socket Layer (SSL), and Oracle 8i. The system is on top of the Internet so it is architectural independent, cross-platform, cross-vendor, and secure. Design and implementations issues are discussed.

An object-oriented taxonomy of medical data presentations.

A variety of methods have been proposed for presenting medical data visually on computers. Discussion of and comparison among these methods have been hindered by a lack of consistent terminology. A taxonomy of medical data presentations based on object-oriented user interface principles is presented. Presentations are divided into five major classes-list, table, graph, icon, and generated text. These are subdivided into eight subclasses with simple inheritance and four subclasses with multiple inheritance. The various subclasses are reviewed and examples are provided. Issues critical to the development and evaluation of presentations are also discussed.

Use of the Extensible Stylesheet Language (XSL) for medical data transformation.

Recently, the Extensible Markup Language (XML) has received growing attention as a simple but flexible mechanism to represent medical data. As XML-based markups become more common there will be an increasing need to transform data stored in one XML markup into another markup. The Extensible Stylesheet Language (XSL) is a stylesheet language for XML. Development of a new mammography reporting system created a need to convert XML output from the MEDLee natural language processing system into a format suitable for cross-patient reporting. This paper examines the capability of XSL as a rule specification language that supports the medical XML data transformation. A set of nine relevant transformations was identified: Filtering, Substitution, Specification, Aggregation, Merging, Splitting, Transposition, Push-down and Pull-up. XSL-based methods for implementing these transformations are presented. The strengths and limitations of XSL are discussed in the context of XML medical data transformation.

What do ER physicians really want? A method for elucidating ER information needs.

Prior discharge summaries are a critical source of information for treating emergency room patients. However, reading discharge summaries may occupy more time than emergency care clinicians can afford. It would be beneficial to present vital information in the reports to them so that they would be able to quickly extract and digest it. There are several possible ways to present the information without changing the structure or content of the report itself. As a prelude to an effective study concerning the efficiency of the various presentation approaches, it is first necessary to know which diagnoses would benefit from past history, and what kind of information is most important to present for each of the diagnoses. In this study, we present a method for elucidating emergency care information needs from clinicians. Analysis of the data obtained from clinicians resulted in generation of a list of important diagnoses and informational categories. For validation, the clinicians were shown sample reports and were asked to highlight critical information. Overall, predicted important items correlated with physicians highlighting (Pearson correlation coefficient of 0.650, significance level 0.01).

A Java speech implementation of the Mini Mental Status Exam.

The Folstein Mini Mental Status Exam (MMSE) is a simple, widely used, verbally administered test to assess cognitive function. The Java Speech Application Programming Interface (JSAPI) is a new, cross-platform interface for both speech recognition and speech synthesis in the Java environment. To evaluate the suitability of the JSAPI for interactive, patient interview applications, a JSAPI implementation of the MMSE was developed. The MMSE contains questions that vary in structure in order to assess different cognitive functions. This question variability provided an excellent test-bed to evaluate the strengths and weaknesses of JSAPI. The application is based on Java platform 2 and a JSAPI interface to the IBM ViaVoice recognition engine. Design and implementations issues are discussed. Preliminary usability studies demonstrate that an automated MMSE maybe a useful screening tool for cognitive disorders and changes.

Costs and benefits of connecting community physicians to a hospital WAN.

The Washington Heights-Inwood Community Health Management Information System (WHICHIS) at the Columbia-Presbyterian Medical Center (CPMC) provides 15 community physician practices with seamless networking to the CPMC Wide-Area Network. The costs and benefits of the project were evaluated. Installation costs, including hardware, office management software, cabling, network routers, ISDN connection and personnel time, averaged $22,902 per office. Maintenance and support costs averaged $6,293 per office per year. These costs represent a "best-case" scenario after a several year learning curve. Participating physicians were interviewed to assess the impact of the project. Access to the CPMC Clinical Information System (CIS) was used by 87%. Other resource usage was: non-CPMC Web-based resources, 80%; computer billing, 73%; Medline and drug information databases, 67%; and, electronic mail, 60%. The most valued feature of the system was access to the CPMC CIS. The second most important was the automatic connection provided by routed ISDN. Frequency of access to the CIS averaged 6.67 days/month. Physicians reported that the system had significantly improved their practice of medicine. We are currently exploring less expensive options to provide this functionality.

Practical lessons in remote connectivity.

Community Health Information Networks (CHINs) require the ability to provide computer network connections to many remote sites. During the implementation of the Washington Heights and Inwood Community Health Management Information System (WHICHIS) at the Columbia-Presbyterian Medical Center (CPMC), a number of remote connectivity issues have been encountered. Both technical and non-technical issues were significant during the installation. We developed a work-flow model for this process which may be helpful to any health care institution attempting to provide seamless remote connectivity. This model is presented and implementation lessons are discussed.

Expressiveness of the Breast Imaging Reporting and Database System (BI-RADS).

The Breast Imaging Reporting and Database System (BI-RADS) was developed by the American College of Radiology and is used by a number of computerized mammography tracking systems. The ability of BI-RADS to encode the data contained in 300 mammography reports at the Columbia-Presbyterian Medical Center was examined. BI-RADS was able to encode normal reports and "special masses" (such as lymph nodes) without difficulty. However, none of the general masses and only 17% of the calcifications could be encoded in BI-RADS. The implications of this for the design of mammography databases are discussed.

Notations for high efficiency data presentation in mammography.

As a result of improvements in Medical Language Processing, the availability of categorical information (such as diagnoses or radiology findings) is increasing rapidly. This increased availability has created a need for more efficient methods for computer presentation. One method for developing such presentations would be to adapt the hand-written notation systems already used in paper-based records. We have characterized one such notation system, the Mammography Notation Sublanguage(MNS). The MNS is a true medical sublanguage, with a definable lexicon and syntax. Compared with text reports, it represents a 37-fold size compression. A single "base", sublanguage pattern is identified for possible computer presentation of mammography findings. The issues involved in using such sublanguages for data presentation are discussed.

Using intermediate states to improve the ability of the Arden Syntax to implement care plans and reuse knowledge.

The Arden Syntax is one of a few knowledge representation languages currently in use for clinical decision support. While some of these languages are being used in active patient care settings, none have gained widespread acceptance as a clinical tool. Prior attempts to represent temporally complex care plans in the Arden Syntax have revealed difficulties in representing and tracking series of consecutive time-oriented events and recommendations, in sharing and reusing knowledge and in dealing with unobtainable data. In an attempt to improve Arden's ability to deal with these problems and demonstrate the importance of these factors, the clinical event monitor has been adapted to store coded data representing Intermediate States in the Columbia Presbyterian Medical Center (CPMC) central data repository. The Intermediate States define the current state of the patient as laid out in the care plan. Four care plans were constructed. The findings include an improved ability to track complex series of events and recommendations over long periods of time. The knowledge generated by the electronic care plans was able to be reused by the care plan that generated it, by other elements of the knowledge base and by non-decision support applications. Modular development, facilitated by the changes, simplified dealing with data not available to the central data repository by aiding the implementation of those parts of the care plan for which sufficient data is available.

Architectural requirements for a multipurpose natural language processor in the clinical environment.

A considerable amount of research has been concerned with the development of natural language systems to automate the encoding of clinical information that occurs in textual form. The task is very complex, and not many language processors are used routinely within clinical information systems. Those systems that are operational, have been implemented in narrow domains for particular applications. For a system to be truly useful, it should be designed so that it could be widely used within the clinical environment. This paper examines architectural requirements we have identified as being necessary for portability and describes the architecture of the system we developed. Our system was designed so that it could be used in different domains to serve a variety of applications. It has been integrated with the clinical information system at Columbia-Presbyterian Medical Center where it routinely encodes clinical information from radiological reports of patients.

Description generation of abnormal densities found in radiographs.

In this paper we present a system for describing renal stones found in radiographs. The system generates descriptions that adhere to those generated by radiologists. The descriptions are formulated by discovering the spatial relationships that exist between the major organs and the renal stones. The system consists of three major components. The first is the image processing component which is responsible for locating the stone. The second component is the inference network minimization component which determines which spatial relationships, of all those that exist between the stone and the organs, is the most descriptive. The third component is the natural language generation component which is responsible for translating the spatial relationships into appropriate medical terminology. We will illustrate all these components on several examples.

Encoding a post-operative coronary artery bypass surgery care plan in the Arden Syntax.

The Arden Syntax for medical logic modules (Arden) was used to test the feasibility of encoding large, complex care plans. The critical portions of an existing paper-based care plan for the management of patients following coronary artery bypass graft (CABG) surgery were encoded in Arden and an X-windows user-interface was developed. The Arden Syntax proved adequate for encoding all of the necessary functions of the care plan. The limitations of the current Arden Syntax and possible additions to Arden are discussed.

Comparison of three Knowledge Representation formalisms for encoding the NCEP Cholesterol Guidelines.

Although many Knowledge Representation (KR) formalisms have been used to encode care guidelines, there are few direct comparisons among different formalisms. In order to compare their suitability for encoding care guidelines, three different KR formalisms were used to encode the National Cholesterol Education Panel (NCEP) guideline. PROLOG, a First Order Logic system, CLASSIC, a frame-based representation system, and CLIPS, a production rule system, were used in the comparison. All three representations allowed accurate encoding of the guideline. PROLOG produced the most compact representation, but proved the most difficult to debug. The lack of arbitrary disjunction in CLASSIC greatly increased the complexity of the encoding. Overall, the CLIPS representation was the most intuitive and easiest to use.