For more than half a century, US regulatory policy on chemical and radiation risks has rested on a deeply flawed scientific foundation. Researchers such as Dr. Ed Calabrese and whistleblower Dr. Paul Selby have spent decades documenting how a pivotal scientific misrepresentation at Oak Ridge National Laboratory in the 1950s misled regulatory agencies and distorted risk assessment for generations. The ripple effect of this error persists, influencing regulations, shaping policy, and framing cancer risk in ways that may not accurately reflect reality.
Will EPA and other agencies finally take action to correct fraudulent science that has affected policy, regulations, and the public for over 50 years?
This is not a new issue. As I have written, Dr. Ed Calabrese has spent the last 20 years researching the historical foundations of cancer risk assessment. In 1996, Dr. Paul Selby, a scientific researcher in genetics, became a whistleblower when he contacted the Department of Energy (DOE) about potential scientific fraud that had occurred 40 years earlier at Oak Ridge National Laboratory. In 2022, Calabrese and Selby teamed up to publish the inside story of the scientific fraud. In 2025, they published a more detailed account of the scientific fraud, explaining how it had directly led to regulatory agencies using an incorrect model to assess cancer risk for the last 50 years.
This August, the journal Science published a letter from Calabrese and Selby stating that a major scientific fraud that began in 1956 was not only never corrected but also perpetuated by the scientific establishment. This led to the use of a scientifically flawed model that numerous US agencies have adopted for assessing chemical and radiation risks.
The Ripple Impact
The use of the flawed linear non-threshold (LNT) model, rather than the threshold dose-response model [1], has resulted in greatly distorted cancer risk estimates that have rippled through regulatory and policy decisions.
An example is the controversy surrounding the chemical ethylene oxide, which is used to sterilize medical equipment. As I previously discussed, the EPA finalized regulations for ethylene oxide in 2024 based on a type of LNT model that showed that a concentration of 0.1 parts per trillion ethylene oxide would result in a one-in-a-million lifetime cancer risk – the level that government agencies often use to represent a significant cancer risk. This level is 19,000 times lower than the amounts found naturally in the human body and 1,000 to 2,000 times lower than those found in typical urban air. The practical impact is that sterilization facilities that produce ethylene oxide for use in medical devices such as pacemakers and surgical equipment could be forced to shut down. If EPA had used the threshold dose-response model instead of the LNT model, the cancer risk estimates would have been dramatically reduced.
Another real-life example concerns CT scans. A recent article in JAMA Internal Medicine states that CT scans will cause one million excess cases of cancer in the coming decade. As discussed in an ACSH article, this prediction is the result of using the LNT model. It does not correspond to the data, which show that while CT scans have markedly increased over the past two decades, cancer rates have gone steadily downward. If the threshold dose-response model had been used, the number of predicted cancer cases would have been significantly lower, more in line with the real-world situation.
How Did We Get Here?
Before 1946, the US government used the threshold dose-response model to assess cancer risk from radiation and chemicals. This model assumes an exposure level below which there is no increased risk of cancer.
In 1951, Dr. William Russell at the Oak Ridge National Laboratory conducted research with millions of mice, examining mutations in the offspring of mice exposed to X-rays. He reported that the mice were 15-20 times more susceptible to the induction of transgenerational mutations than those previously reported in fruit flies.
In 1956, this data was used by the Biological Effects of Atomic Radiation (BEAR) I Genetics Panel to conclude that the data supported switching from the threshold dose-response model to the LNT model. The use of the LNT model became the standard methodology for most US government agencies, a policy that remains in place today.
Scientific Fraud
In 1995, Dr. Paul Selby, a long-time associate of William Russell, discovered that Russell had excluded clusters of spontaneous mutations seen in the control groups of mice from his experiments, with no mention of these mutations in his reports or publications.
Selby found that the actual results from the experiment revealed additional mutations in the control group, demonstrating a mutation rate in the controls that was almost identical to that in the irradiated mice.
“Data used by the Panel as supplied by William Russell were incorrect, thereby leading to massive overestimation of the hereditary damage resulting from exposure to ionizing radiation.”
Selby contacted the DOE leadership in June 1995, which led to a formal evaluation of Russell’s results by the DOE, including four external genotoxicity experts. In 1996, Russell admitted to suppressing the mutation rate in the control group by 120% which falsely elevated radiation-induced mutation risks. The DOE corrected their report, using the corrected mutation rate in the controls as supplied by Russell. This should have led to a policy rewrite nearly 30 years ago.
However, the incorrect findings and interpretations of Russell’s data were never highlighted or corrected in the peer-reviewed scientific literature. The DOE, the National Academy of Sciences (NAS), and the EPA did not follow up or change their policy based on these corrections, which more properly framed the risk of cancer.
Calabrese and Selby do not mince words:
“The ancient saying is relevant here: The fish rots from the head, that is, this problem started at the top of the scientific establishment, with an NAS Panel of distinguished geneticists with two Nobel Prize winners (Muller and Beadle). It is long past time to correct flawed regulatory policies and practices that are presently based on the unreliable information in the BEAR I Report.”
If the current Administration is serious about its commitment to “gold-standard science,” it must confront this legacy, acknowledge the errors, and correct the policies built on them. Half a century is long enough—correcting the record is overdue. Will the current Administration, rise to the occasion? Will the Administration assemble a panel of experts in cancer risk to unwind the error? The public deserves a risk framework grounded in accurate science, not outdated or manipulated data.
[1] The linear non-threshold model assumes that every increment of a chemical or radiation dose, no matter how small, constitutes an increased cancer risk. In contrast, the threshold dose-response model assumes that there is no increased cancer risk below a certain exposure level.
