Pharmacia & Upjohn had invested hundreds of millions of dollars investigating a new potential drug to treat a mental ailment, and they had zeroed in on a compound that showed promise. To be approved, the compound had to be significantly more effective than a placebo. But after months and $10 million invested in a double-blind trial with 1,000 patients, the compound was not passing the statistical tests the FDA would use. The company needed to determine if they would move forward with the multi-year FDA application process, another billion dollar investment.
The time-critical business need was to make a rational investment decision on whether or not to go forward, not to pass the specific FDA tests. So, we focused on visualizing the effects of treatment, rather than on statistical significances. Previous analyses had measured patient outcome on an absolute scale. However, distributions seemed to vary significantly by several factors, especially the country of the patient and doctor, so we focused instead on relative improvement (or worsening) after treatment. The 500 patients who received the placebo are summarized in a three-dimensional density plot (Figure 1) we custom-designed for this project. Patients that moved toward the upper-right corner got better and those that moved toward the lower-left got worse. For the drug (Figure 2) the movement toward getting better (upper-right) is very strong. Clearly, the compound is having a very positive effect.
When shown the plots and the research behind them, Pharmacia & Upjohn moved forward confidently and partnered with Pfizer (who later acquired them) to make the $1 billion investment. Years of further work proved the drug’s efficacy to the regulatory agency; it was approved, adopted, and a blockbuster success. Such an outcome is so rare that it was one of only three drugs the firm created in that decade. The data science triumphs were discovering a real effect where experts in the domain had been stymied, and communicating decisive results to decision makers with a novel visualization.