This BBC article reports that Dr. Jon Sudbo of Oslo's Norwegian Radium Hospital completely fabricated his patient data for an article in the medical journal The Lancet titled Non-steroidal anti-inflammatory drugs and the risk of oral cancer.
From the article:
Stein Vaaler, director of external relations at the hospital, added: "He published an article in The Lancet in October last year whose data is totally false, actually totally fabricated.
"His database had been completely fabricated on his computer."
Norwegian daily newsaper Dagbladet reported that of the 908 people in Sudbo's study, 250 shared the same birthday.
The editor of The Lancet reiterated the point that peer review only targets bad science, not fraud. Notably, in many cases of scientific fraud, there's a certain amount of sloppiness; Sudbo repeating birthdates, Hwang's group reusing different parts of the exact same image, Jan Hendrik Schoen reusing the same graph to represent completely different experiments. Does this mean that people who commit fraud are foolish or pathological enough to not worry about such oversights, or that we're just not catching the clever ones? I'm inclined to believe the former, because there's the fundamental problem in science that if your work is of any significance at all, someone will attempt to build on it -- so the discovery of fraud is guaranteed.
I've personally experienced this in a slightly different scenario, where it was not fraud but an experimental artifact (that is, an effect that appeared under a certain set of experimental conditions, but which did not indicate what the researcher thought it did). Several labs, including ours, tried to build on a result reported by another lab; all the labs found that the reported effect did not have the consequences it should have. Taking a step back, they tried to directly replicate the result and found they couldn't. They had a talk with the reporting lab, who went back, did more experiments, and discovered their error. If your work means anything, it's just a matter of time before someone else tests it.