Mars' data archiving system poised for nationwide intro

University medical centers pose a unique marketing challenge for healthcare systems vendors because of the vast and varied resources available to large academic institutions. Their prestige makes many universities valuable partners for beta

University medical centers pose a unique marketing challenge for healthcare systems vendors because of the vast and varied resources available to large academic institutions. Their prestige makes many universities valuable partners for beta testing new products. Many are also fertile incubators for new technologies that serve as the impetus for high-tech start-ups, however, turning the vendors into customers.

Such is the case at the University of Pittsburgh, which has spawned a number of medical technology spin-offs, including Stentor and Mars (Medical Archival Systems Inc.). Mars was formed in 1997 to commercialize the University of Pittsburgh Medical Center's medical archival system (MARS). After evaluating a number of commercial products in the mid-1980s, UPMC decided to build its own information management system. It launched the first production system in 1990 and integrated financial and administrative data in 1991-1992.

"We came to the conclusion that we could build a better system, based on parallel processing, than anything that was available on the market," said Dr. John Vries, senior vice president of Mars. "Our system had the computational power to do full-text indexing."

The initial plan was to provide an integrated medical record to support clinical care at the point of service, enabling physicians to see all relevant information from any location, according to Vries. The next step was to do a commercial-grade rewrite of the software, a process that began in 1997 and took about two years to complete.

Today Mars is in the middle of a seven-year contract with UPMC to provide full archival service. The system serves 17 hospitals in the UPMC health system and contains around 150 million records on two million individuals indexed on 10 billion individual searchable items. The information in MARS comes from 25 major record types from different legacy systems and is fed into the database in real-time.

"We consider every kind of information as text, including structured data like laboratory, pharmaceutical, and financial information," Vries said. "We can literally drill down into every piece of individual information using Boolean operations (and, or, not). The system is very fast, and users can get to the information quickly. However, because the system is so information rich, it can be difficult to formulate a query."

The data feeds are mapped into an internal canonical format based on a proprietary markup language that is highly flexible. Once in this uniform format, data can be indexed and converted into other forms, such as XML and HL7. The purpose of MARS is to integrate data, according to Vries. The client chooses the preferred display method.

Given its success with UPMC, the firm Mars plans to begin marketing the system MARS outside of the university. It has formed a relationship with Internet Venture Works, which will help develop the business plan, provide key management and marketing personnel, and assist in raising capital, according to Vries. In exchange, Internet Venture Works is taking an equity stake in Mars.

In addition, Mars has garnered financial support from a NIST grant, which the company will use to fund development of an analytical system that will extract anonymous data from the core archival system. Vries sees potential in marketing anonymous data to third parties such as pharmaceutical companies that need verified information.

"We applied to NIST for a grant to support pilot development of a robust analytical system to extract data," he said. "We want to make the system more user friendly, without the user needing to be an information technology expert to get information out of the database."

According to Vries, the fundamental idea of the analytical system is to combine the Boolean search engine with a vector space engine (similar to those used by Internet search engines such as Google or Alta Vista) that does comparisons based on similarity. If a user with domain expertise finds an applicable record, that user can do further research using that record as the search.

"The Boolean search engine acts like a lens or focal engine," Vries said. "Cache knowledge is accumulated over time, which allows a domain expert to get the information quickly."

Mars has been doing feasibility studies on the analytical system for more than a year and will be hiring eight to 10 people under the grant, which officially starts April 1. The analytical system should be a fully mature product within two years, according to Vries. Mars will begin marketing the anonymous data in the middle of 2001 and will probably pursue patents on the technology.

The company offers its services on an ASP model, using a per-click transaction basis to determine pricing. Although the original MARS featured fully integrated image data, the developers eventually opted to focus solely on textual information sources. While the market for anonymous data is tremendous, it does not involve images to a substantial degree, according to Vries. The current system can integrate with any PACS, however.

"We came to the conclusion that (integrating images) was a bad idea, and that no one will make any money out of PACS because of the storage and bandwidth requirements," he said. "Most decision-making is based on analysis of textual and numerical data, except in certain specialties. There is a gee whiz factor with images, but unless you're in radiology, you're more interested in reports."