The science community is now closing in on an example of scientific fraud at Duke University. The story sounds awfully familiar.
ANIL POTTI, Joseph Nevins and their colleagues at Duke University in Durham, North Carolina, garnered widespread attention in 2006. They reported in the New England Journal of Medicine that they could predict the course of a patient’s lung cancer using devices called expression arrays, which log the activity patterns of thousands of genes in a sample of tissue as a colourful picture. A few months later, they wrote in Nature Medicine that they had developed a similar technique which used gene expression in laboratory cultures of cancer cells, known as cell lines, to predict which chemotherapy would be most effective for an individual patient suffering from lung, breast or ovarian cancer.
At the time, this work looked like a tremendous advance for personalised medicine—the idea that understanding the molecular specifics of an individual’s illness will lead to a tailored treatment.
This would be an incredible step forward in chemotherapy. Sensitivity to anti-tumor drugs is the holy grail of chemotherapy.
Unbeknown to most people in the field, however, within a few weeks of the publication of the Nature Medicine paper a group of biostatisticians at the MD Anderson Cancer Centre in Houston, led by Keith Baggerly and Kevin Coombes, had begun to find serious flaws in the work.
Dr Baggerly and Dr Coombes had been trying to reproduce Dr Potti’s results at the request of clinical researchers at the Anderson centre who wished to use the new technique. When they first encountered problems, they followed normal procedures by asking Dr Potti, who had been in charge of the day-to-day research, and Dr Nevins, who was Dr Potti’s supervisor, for the raw data on which the published analysis was based—and also for further details about the team’s methods, so that they could try to replicate the original findings.
The raw data is always the place that any analysis of another’s work must begin.
Dr Potti and Dr Nevins answered the queries and publicly corrected several errors, but Dr Baggerly and Dr Coombes still found the methods’ predictions were little better than chance. Furthermore, the list of problems they uncovered continued to grow. For example, they saw that in one of their papers Dr Potti and his colleagues had mislabelled the cell lines they used to derive their chemotherapy prediction model, describing those that were sensitive as resistant, and vice versa. This meant that even if the predictive method the team at Duke were describing did work, which Dr Baggerly and Dr Coombes now seriously doubted, patients whose doctors relied on this paper would end up being given a drug they were less likely to benefit from instead of more likely.
In other words, the raw data was a mess. The results had to be random.
Duke had … chosen, within a few months of the papers’ publication (and at the time questions were being raised about the data’s quality) to launch three clinical trials based on their work. Dr Potti and his colleagues also planned to use their gene-expression data to guide therapeutic choices in a lung-cancer trial paid for by America’s National Cancer Institute (NCI). That led Lisa McShane, a biostatistician at the NCI who was already concerned about Dr Potti’s results, to try to replicate the work. She had no better luck than Dr Baggerly and Dr Coombes. The more questions she asked, the less concrete the Duke methods appeared.
Now we have two “skeptics” wondering about the data. Here it begins to acquire a marked similarity to another scientific controversy, although one with much larger world wide consequences.
In October 2009, officials from the university arranged for an external review of the work of Dr Potti and Dr Nevins, and temporarily halted the three trials. The review committee, however, had access only to material supplied by the researchers themselves, and was not presented with either the NCI’s exact concerns or the problems discovered by the team at the Anderson centre. The committee found no problems, and the three trials began enrolling patients again in February 2010.
Some of us call that a whitewash and it also bears some similarity to another such committee investigation in 2010.
Subsequently, the committee’s members interviewed Dr Baggerly about the problems he had encountered trying to sort the data. He noted that in addition to a lack of unfettered access to the computer code and consistent raw data on which the work was based, journals that had readily published Dr Potti’s papers were reluctant to publish his letters critical of the work. Nature Medicine published one letter, with a rebuttal from the team at Duke, but rejected further comments when problems continued.
Other journals that had carried subsequent high-profile papers from Dr Potti behaved in similar ways. (Dr Baggerly and Dr Coombes did not approach the New England Journal because, they say, they “never could sort that work enough to make critical comments to the journal”.) Eventually, the two researchers resorted to publishing their criticisms in a statistical journal, which would be unlikely to reach the same audience as a medical journal.
Now it is getting really familiar to some of us. Journals were blocking access to critics of a paper that was considered a major advance. Grants were involved and, potentially, lucrative commercial ties.
The university’s lapses and errors included being slow to deal with potential financial conflicts of interest declared by Dr Potti, Dr Nevins and other investigators, including involvement in Expression Analysis Inc and CancerGuide DX, two firms to which the university also had ties. Moreover, Dr Califf and other senior administrators acknowledged that once questions arose about the work, they gave too much weight to Dr Nevins and his judgment. That led them, for example, to withhold Dr Baggerly’s criticisms from the external-review committee in 2009. They also noted that the internal committees responsible for protecting patients and overseeing clinical trials lacked the expertise to review the complex, statistics-heavy methods and data produced by experiments involving gene expression.
Now, the commercial connections are all, according to the alarmists, on the global warming skeptic side but some of us doubt this. The more we have been involved in research, the more we doubt that version of the story. Trillions of dollars, and Euros, and Yen are at stake. The grants and the other goodies available to researchers who come to the correct conclusions are in a similar scale.
The difference here is that the dubious results of flawed or faked research can be tested against real world consequences. So far, the global warming models have not been tested and the honesty of the research workers is all we have. And we know what that is worth.