Beset by Lawsuits, IBM Blocks a Study That Used Its Data
Science, May 14, 2004
Workers who make the chips that form the brains and memories of computers and electronic devices use an array of nasty chemicals, some of which are known to cause cancer or are suspected of doing so. Whether on-the-job exposure has actually caused disease among chip workers is, however, a hotly contested issue. Good data are scarce, but some occupational health experts believe valuable clues may lie in a trove of records maintained by the computer giant IBM, which is now at the center of a legal dispute.Two researchers who gained access to IBM's records have produced an epidemiological study, paid for by attorneys suing IBM, concluding that former workers at IBM's computer-chip factories faced increased risks of dying from brain, kidney, blood, and skin cancers. But IBM recently moved to block publication of these findings, arguing that it had turned over the data only for use in a lawsuit and that a judge had ruled that the analysis could not be introduced as evidence in a trial because it was irrelevant and could be confusing to jurors. The company itself has since hired its own expert to conduct a new analysis of the data.
The contested study--by epidemiologists Richard Clapp of Boston University and Rebecca Johnson of Epicenter in Circle Pines, Minnesota--analyzes a large set of mortality records on people who worked for IBM over a period of more than 3 decades. It was scheduled to appear in an upcoming special issue of Medical Clinics of North America, says guest editor Joseph LaDou, director of the University of California, San Francisco's International Center for Occupational Medicine. Four peer reviewers had read and approved the study, Ladou says. Then, on 30 March, Clapp withdrew it. IBM attorney Robert Weber confirms that an IBM attorney warned Clapp he shouldn't publish. LaDou says, "I wanted this in the journal because it's the most definitive cancer study" to date on this industry. IBM's action was a "serious disappointment to our scientific and academic freedom."
But Weber argues that academic freedom is not at issue. Clapp has no legal right to publish the mortality analysis because he is bound by the court's protective order that mandates he keep the data confidential, Weber says.
A judge originally ordered IBM to open its files to Amanda Hawes and Richard Alexander of Alexander, Hawes and Audet in San Jose, California. They are attorneys for families suing the company over cancers in California and New York. The plaintiffs' attorneys then paid Clapp and Johnson to analyze the data. IBM attorneys say the court ruled that the data were available for use only in litigation and that Clapp had signed on to the protective order, which legally binds him to maintain confidentiality.
Figure 1 Clean room? Workers employed in computer manufacturing in the 1970s and 1980s had an elevated risk of dying from specific cancers, according to a hotly contested study.
CREDIT: ED KASHI/CORBIS
Clapp declined to comment on the legal dispute, as did Hawes. Alexander and Johnson did not respond to telephone messages.
The IBM lawsuits grew out of long-standing allegations that workers in semiconductor manufacturing were poorly protected from hazardous solvents and other chemicals that cause cancers and birth defects. IBM and other companies have denied this. But the health risks are worth investigating, says cancer epidemiologist John Bailar, a professor emeritus at the University of Chicago and current scholar-in-residence at the National Academy of Sciences in Washington, D.C. "Semiconductor workers in the past and at present suffer exposure to potent organic chemicals, including some known carcinogens," says Bailar, who has no connection with the litigation. A handful of earlier studies had provided hints of elevated cancer risks among semiconductor workers, based on limited data.
In 1997, former workers and their survivors began suing IBM. In discovery proceedings for one of the first of the approximately 200 ongoing lawsuits, a New York judge ordered IBM to turn over a database called the corporate mortality file, which tabulates the cause of death of more than 33,000 former IBM workers, as well as a database of work history on more than 18,000 deceased former workers. As an expert for the plaintiffs, Clapp had access to these files.
In their study, which Science has obtained, Clapp and Johnson first conducted a preliminary analysis called a proportional mortality ratio. It spotlighted eight types of cancer in men and five in women that seemed to kill former IBM workers more frequently than would be expected in a group of average Americans. Then, they compared the percentage of cancer deaths from each type of cancer with the corresponding percentages in the general population--a measure called proportional cancer mortality ratio (PCMR). The PCMR analysis provided several statistically significant results: The 7697 IBM men who died of cancer were between 23% and 62% more likely to have died from cancers of the kidney, brain, blood, and skin. And the 1667 IBM women analyzed were 20% more likely to have died from kidney cancer. When the researchers focused on a subgroup more likely to have been exposed to chemicals--employees who worked at least a month at one of IBM's chip-manufacturing plants--PCMR results indicated that male workers were 62% to 79% more likely to have died from kidney, skin, or brain cancer, and female workers were 112% more likely to have died from kidney cancer.
Figure 2 Retreat. Under legal pressure, Clapp withdrew an epidemiological study from publication.
CREDIT: ANDREW BRILLIANT/WWW.BRILLIANTPICTURES.COM
Bailar calls some of the numbers "elevated enough to make me worry." The Clapp and Johnson analysis--the biggest cancer study in the electronics industry so far--is consistent with preliminary evidence from a small 2001 British-government study on a Scottish semiconductor factory that showed elevated cancer risk, he says. But the new results fall "short of proof." Occupational epidemiologist Harris Pastides of the University of South Carolina, Columbia, concurs. Although the authors "reported very objectively," he says, the study amounts to "an early part of the detective work to try to go down that road to causality."
IBM's lawyers reject the findings: "This is one of the clearest examples of what has been characterized as junk science," Weber maintains. "It's a litigation-produced study in which lawyers supplied key data and gave direction on how the study was to be done." Clapp says that he "totally disagrees" with that depiction. The corporate mortality file itself came from the plaintiffs' lawyers, he says, but he and Johnson did "everything after that," including designing the study, choosing the statistical software, and analyzing the data.
IBM's attorneys fought successfully to keep the analysis out of the first cancer lawsuit to come to trial, a closely watched case brought by two former IBM workers in San Jose, California. On 9 October, just before the trial started, Judge Robert Baines of California Superior Court barred jurors from seeing the Clapp and Johnson study because he said it did not document a link between the mortality data and workplace exposure to chemicals, making it "simply irrelevant and ... highly prejudicial." IBM won that case on 27 February.
Bailar and other epidemiologists suggest that IBM could clear up lingering questions about whether semiconductor manufacturing work raises cancer risks by giving outside scientists access to records for the entire workforce rather than just for those who have died, which would allow scientists to look at overall rates of death. Researchers also would like to see IBM's data on workers' assignments and chemical exposures.
In March, the Semiconductor Industry Association, a San Jose, California, trade group representing 95 of the largest semiconductor manufacturers, commissioned a retrospective study to see whether chip- manufacturing workers faced higher cancer risks. IBM's Weber says that the company has commissioned a separate study on possible workplace health hazards at IBM, led by Elizabeth Delzell of the University of Alabama, Birmingham. Delzell declined to discuss the details of the study but says she hopes to complete it this year, and she plans to publish it in a peer-reviewed journal.