Pfizer Developing New Methods For Detecting Hepatotoxicity Signals

Current methods for detecting hepatotoxicity have no mathematical basis, Pfizer says. The company plans to evaluate patient data for Rezulin and Trovan, which were the subject of liver toxicity concerns.

Pfizer is developing vector analysis methods for use in combination with stochastic differential equations to detect changes in liver enzyme levels.

"We're hoping that each hepatotoxicity diagnosis such as necrosis…will have a different path that it will follow and we'll be able to use [the vector analysis methods to detect] for diagnostic signals," Pfizer Senior Director and Global Head of Mathematical Medicine­ Donald Trost, MD/PhD, said at a conference on drug-induced hepatotoxicity in Philadelphia Sept. 20.

Current methods of detecting hepatotoxicity safety signals have no mathematical basis and lack objectivity and sensitivity, Trost maintained. He noted the skewed reference ranges at some labs that do clinical research for pharmaceutical companies.

"When you get a normal range from a lab and it's zero, that's because they got a negative number and they don't know what to do with it….That's the first sign that the lab doesn't have a clue about normal ranges," he said.

"Even if the lab does everything appropriately, they do log transformation, they calculate their statistics correctly, they have a sample size greater than 120, they're still probably going to give you the wrong reference range."

To achieve a confidence interval of plus or minus one-tenth of 95%, the sample sizes need to be somewhere between 500 and 1,000 patients, Trost said.

"There's no external validity here whatsoever in these reference ranges. I don't think these things are very well known or people would be talking about them."

In addition to the reference range, Trost states that another problem with current detection methods is that correlational analysis is not being used to compare univariate and multivariate abnormality.

When correlational analysis is not used to evaluate the strength of the relations between variables, taking the intersection of two normal distributions on a graph creates a box.

However, "the true distribution or the joint distribution of two gausians (normal distribution curves) is circular or elliptical," Trost said. "The box method is only slightly better than flipping a coin."

"You have a 60% chance of having a false positive if you do 20 lab tests on one blood sample" using the box method. "And the way we dealt with that in the pharmaceutical industry, and perhaps other places, is we've created this thing called three times upper limit of the normal."

The reason why hepatotoxicity is rarely picked up in clinical trials is "if you go to three times the upper limit of the normal, the box is 60 times larger than the ellipse," Trost said. At this size, there won't be any abnormalities remaining on the plotted chart, he added.

For the vector analysis project, Trost plans to evaluate patient data for two Pfizer products that have been the focus of hepatotoxicity concerns - Rezulin (troglitazone), which was pulled off the market and Trovan (trovafloxacin), which was severely restricted.

Trost will examine whether early hepatotoxic effects for the drugs could have been detected.

An FDA advisory committee noted that incidence of liver injury associated with AstraZeneca's Exanta suggests a higher risk of toxicity than Rezulin. The committee concluded Sept. 10 that Exanta's liver toxicity outweighs its benefit (¹ (Also see "AstraZeneca Exanta Liver Toxicity Outweighs Benefit – FDA Advisory Cmte." - Pink Sheet, 10 Sep, 2004.)).

­- Andrew Shelton