MINESET'S DATA VISUALIZATION ENHANCES CLINICAL STUDIES AT PROCTER & GAMBLE
by Daniel Stevens

Daniel Stevens is a scientist for the Healthcare Research Center at Procter & Gamble. This is a report documenting his product review, which was originally published in DM Review Magazine. The hardware architecture considered here consists of Indigo, O2 and Octane workstations, buttressed by a 16-CPU (Central Processing Unit) Origin 2000 server on the back end. Procter & Gamble over-the-counter and pharmaceutical health care products are developed at the Healthcare Research Center outside Cincinnati, Ohio. Founded in 1837, Procter & Gamble is one of the world's largest consumer products manufacturers, with worldwide sales of $35.8 billion.

The problem P&G faced was the need to analyze a database derived from clinical drug trials. There was a need to understand the dynamics of drug efficacy and human study populations, while also identifying ways to improve the efficiency of future clinical drug trials.

On average, the total cost to bring a pharmaceutical drug to market is approximately $500 million. A large portion of that amount has already been invested in a drug's development by the time it is ready for clinical trial. Therefore, it is imperative that the trial runs effectively and that all data collected is verifiably accurate and fully assessed. MineSet's data visualization features condense information into readily accessible and easily understood data-rich graphics. When its thin learning curve and relatively short implementation time frame were factored in, MineSet presented itself as the optimum solution to this end user.

THE PRODUCT AT WORK: STRENGTHS & WEAKNESSES

The types of analyses necessary to comprehend clinical databases have historically been performed by statisticians in a cyclical manner. In brief, data presented at meetings is reviewed, and this generates a host of new questions and theories requiring multi-week analysis. This labor-intensive process repeats itself until all questions and issues have been indisputably resolved. What MineSet affords is similar data analysis with nearly instantaneous feedback (a few seconds versus a month by traditional means).

The core strength of Silicon Graphics' MineSet is its ability to facilitate data mining for the typical business user, thereby empowering a larger share of an organization. MineSet's unique data visualization features condense information into readily accessible and easily understood graphics. A chief benefit the end user sees with MineSet is increased accessibility to warehouse assets by non-statistician query. Furthermore, results feedback is speedier, allowing time for exploration of a greater number of theoretical scenarios.

However, MindsSet 2.5 is not Web-enabled like many other Silicon Graphics products. Also, the documentation needs to be matured -- still the concept of visualization data is so new that the documentation will only improve as time goes on.

The FDA requires the use of "clean" databases in clinical drug trials. MineSet's advantage here is the ability to transparently access existing corporate databases (e.g., Oracle, Sybase, Informix) without programming or major consulting. While the end user did evaluate other data mining tools, MineSet was selected because it rated convincingly as the most evolved and proven of the group.

In working with MineSet, this drug development team uses very large health care databases on how experiment drugs work on individuals. These databases contain a large number of critical variables and risk factors. In early drug development stages, teams create theories ("what-if" scenarios) on how risk factors (high blood pressure, etc.) may impact the disease state and how the therapeutic agent will impact the disease state. The MineSet data mining tool uses databases to confirm theories developed during the drug development process. The software finds associations within this data and makes a prediction through "what-if" scenarios.

Vendor support of this product is good to excellent. SGI quickly answered questions either by voice or e-mail. As an early user of MineSet, several of our comments and feedback to SGI have been incorporated into the product. The product is still in an early documentation stage. We expect that it will improve in the future.

For more information, see http://www.sgi.com