Data Capture and Navigation

Ian Foster itf@mcs.anl.gov
Mark Frisse Mark_Frisse@msnotes.wustl.edu
Andy Grimshaw grimshaw@virginia.edu
Carl Kesselman carl@vlsi.cs.caltech.edu
Melanie Loots mloots@nlm.nih.gov
Peter Steenkiste peter.steenkiste@cs.cmu.edu
Joel Saltz saltz@cs.umd.edu

Introduction

This working group focused on the ways in which pervasive access to data could impact the care provided to individual patients. and the computer science issues that, if successfully addressed, can be expected to increase the utility and reduce the cost of this data access.

At present, there are many structural impediments to the effective use of computing technologies in health care, such as limited computerization of databases, a lack of standard database formats, limited networking, and a reluctance to share data. However, it appears likely that within a five to ten year timeframe, these structural impediments will be largely removed. In part because of a demand for closer accounting for health care costs, and in part because of the increasing trend to consolidation of hospitals into larger administrative units, we expect that the health care system of the future will be characterized by:

Widespread integration of databases
Use of universal patient identifiers to provide unambiguous access to patient information
Development of high speed ATM networks, and digitization of most health care data modalities (X-rays, MRI, etc.)

We also assume the widespread deployment of settop boxes in the home, allowing interactive consultations with remote physicians; it appears quite likely that these will form a significant part of a future health care infrastructure.

Health Care Application

Listed below are a few specific applications that have implications for computer science research:

Teleconsultation with patients in their homes: This class of applications will require sophisticated computer-aided diagnosis support to aid in diagnosis from limited data.
Identification of "similar" case histories.: This is a complex problem that requires the ability to encode "process" (care plans and diseases) and to detect similarities between process encodings
Software Integration: For example, the integration of both computer simulations and image processing modules into diagnosis systems requires developments in parallel algorithms and distributed systems
Expert assistants: The development of "expert assistants" for diagnosis, alerts for unusual events during treatment, and public health warnings requires significant developments in A.I.; it may also involve difficult

High Peformance Computing Issues

Making effective use of the database, networking, and computing infrastructure that will be deployed in hospitals in the next five to ten years is first and foremost a very complex distributed computing problem. To a large extent, the research challenges that must be addressed to make this infrastructure truly usable will be the same as those faced in commerce, manufacturing, and entertainment. However, health care also introduces some unique concerns. In the following, we examine a range of both generic and health care-specific issues.

The most fundamental challenge, in our view, is to develop system architectures will provide the flexibility, scalability, and extensibility required to support a wide range of health care applications. The danger is that the health care information systems developed in the next five years will become incompatible, hard-to-extend legacy systems of tomorrow. Computer scientists can help informaticients avoid this problem by working with them to develop appropriate system architectures. Particular challenges include:

Software Architectures: Developing architectures that support the interoperation of diverse information and computational resources. The integration of computational resources into a data-intensive distributed computing infrastructure.
Resource naming, discovery, and management: Applications must be able to locate patient records quickly in an emergency, from a large distributed database system. Expert assistants and image processing systems must be able to locate computational resources, or database resources containing (for example) representative case histories for comparative purposes. The soft real time character of many medical applications makes the efficiency of these mechanisms, and the algorithms used to schedule resources, particularly important.
Encoding process: One important research area which seems to underly many aspects of future health care systems is the development of techniques for encoding "processes," such as care plans and diseases. These techniques need to support the development of process representations, the automatic detection of processes from database records, and identification of "similar" process representations. The research problems here seem to be primarily concerned with knowledge representation and A-I., although HPCC issues may arise if the fundamental algorithms are computationally demanding.
Security: Problems particular to health care include maintaining patient privacy when sharing medical information, and guarding against denial of service. Privacy is problematic because of numerous database applications will require the sharing of data, yet it is remarkably difficult to "sanitize" patient records to prevent detection of a patient's identity. Denial of service issues arise because inevitably computational and network resources used for life-critical applications (such as image processing or expert assistance during surgery) will also be used for other applications.
Reliability: The health-care infrastructure must clearly be highly reliable and fault tolerant. As demands on this infrastructure can be expected to be particularly heavy following disasters, the infrastructure must be designed to cope significant damage. Tradeoffs between security, reliability, and performance must be carefully evaluated.
Parallel algorithms: Applications developed using the infrastructure are likely to incorporate computer simulations, impage processing modules, large-scale statistical analysis, and expert assistants. Many of these system components must execute in soft real time. Research in parallel algorithms will be required to allow their efficient execution.