Clinicians propose biology-modeled database architecture

Electronic medical records are based on input-oriented architectures separated into laboratory, pharmacy, radiology, and other hospital divisions, but if given a choice, clinicians would reorganize the records around physiological systems.

Analysis suggests this architecture is superior, an artificial intelligence expert told a Clinical Systems audience Monday afternoon at HIMSS. Herbert Doller, Ph.D., explained how "black box" analysis can be applied to modeling physiological systems and how this analysis produces an architecture superior to the current input-oriented model.

"All pure scientists know that quality applications begin with good architecture and that bad architecture will show up at the worst possible time - the time when users are maddest and demanding you to make immediate changes," said Doller, chief of the artificial intelligence laboratory at the VA North Texas Health Care System and an assistant professor of pathology at UT-Southwestern Medical Center in Dallas. "Bad architecture, however, often stands in the way of those changes. Good architecture, on the other hand, anticipates the needs of users and allows the application to evolve."

Current database design architecture in healthcare is leading us to develop sophisticated applications on top of a large global database, according to Doller. Such an architecture leads to applications of exponential design, and that the need of these applications for computer resources also grows exponentially, eventually reaching an infinite demand. Such designs are not solvable by large databases.

"We're not talking about them being difficult to solve - they're impossible to solve once they've reached a certain size," he said. "Current data repositories are already too large for this design. This architecture is already limiting your ability to create good applications in healthcare."

In order to design sophisticated medical applications, a partitioning scheme is required. Such a design would lead to polynomial growth instead of exponential. Although this growth can be substantial, it does not lead to infinite need for computer resources, as do current designs, Doller said. What he proposed is a partitioning scheme in which the design actually models biology.

Two camps are interested in biological data: bioinformatics and medical informatics. Using genetic sequences, bioinformatics wants to say something about the health of the organism through a mechanism that will allow a change in genetic sequence to propagate throughout the organism, where it manifests itself in different symptoms, Dollar said.

At the other end of the spectrum are the physicians with their clinical data wanting to use this data to correlate with the genetic information.

"What interested us is that they're all looking at same pathway from opposite directions, and the architecture that we're proposing will actually support a single pathway working in both directions," he said.