This is all true, but the atricle says something quite different than what you are commenting on. The "original findings" which are being reversed were based on 4,000 pieces of peer-reviewed research.
After reviewing more than 4,000 studies, the [1997] authors were persuaded that green vegetables helped ward off lung and stomach cancer.
It's easy to dismiss a single instance as an arbitrary data point. When you can dismiss a large portfolio of such points--you have another problem. That problem is a core-weakness in methodology/approach/and institutional incentives. That seems to be the larger issue highlighted here.
Well, the sentence immediately preceding that one gets it right:
>The situation seemed very different in 1997, when the World Cancer Research Fund and the American Institute for Cancer Research published a report, thick as a phone book, concluding that diets loaded with fruits and vegetables might reduce the overall incidence of cancer by more than 20 percent.
With "might" being the most important word there. These 4,000 studies were all retrospective analyses, and retrospective analyses only find conclusions of the "might" sort, which is well understood by the scientists, but not so well communicated by journalists.
Still, a 20% "might" is a an absolutely enormous amount of potential upside in cancer prevention, even a "maybe" of this magnitude justifies great excitement. Which is why they embarked on the prospective study, which is the real story here.
It is absolutely ridiculous to conclude that there are methodology or misaligned institutional incentives based on this case, it's exactly how things should work. It's just another example of scientists being blamed by bad reporting that they have no control over.
That said, there are many other cases where methodology is wrong or institutional incentives are wrong.
After reviewing more than 4,000 studies, the [1997] authors were persuaded that green vegetables helped ward off lung and stomach cancer.
It's easy to dismiss a single instance as an arbitrary data point. When you can dismiss a large portfolio of such points--you have another problem. That problem is a core-weakness in methodology/approach/and institutional incentives. That seems to be the larger issue highlighted here.