Those are very good points, and certainly elucidate us on one of the various issues with conducting studies. I guess my personal prejudice against small studies is apparent. Lack of elimination of other factors is a bias that flaws many studies, but certainly there are many great examples of meta studies that were of extreme value. But I still believe that those studies are only as good as the FACTS and conclusions drawn in the included studies, and can be flawed by poor prior scientific methodology. Meta analysis must take care not to introduce its own bias in the author's selection of those prior studies that support the conclusions that they would like to come to. Case in point are many of the studies done by the tobacco industry which reviewed only those prior papers (many funded by them) that reached a conclusion that they had a preconceived notion that they wished to reach. Bottom line is no single study is worthy until the conclusions can be repeated by other independent investigators, and all bias removed from the study itself. Since Nellie is someone who understands this more deeply than I, would you please post a reply that explains bias and some of the forms that it takes? I think it would help everyone, especially those that routinely review the studies that we publish in the news section so that that can make intelligent decisions about what they are reading.

The evidence was only recently found at Johns Hopkins and the information surprised everyone since it was completely unexpected and was a by-product of them looking for something else. The article is to be published in the next few months, and I will put it in the news section of the site when it comes out. I had a chance to review a preliminary copy from Dr. Gilllison one of the co-authors.

In my opinion (which isn't always worth that much) I have seen so many small studies, that were poorly funded, and following that, poorly populated, and following that, had poor elimination of other factors which may have induced the same outcome that were not measured or even considered, I find many of them suspect. The bleaching study that I mentioned above, didn't even ask the two people who had developed oral cancers (with the only apparent commonality between them that they used tooth whiteners) if they had been tested for HPV, an already established causative agent that they would have no way of knowing they were positive for as people in their 20's who never smoked or drank to excess. This was by one of the most prominent otolaryngology professors in the US publishing!! The conclusion that whitening agents were the causative factor when one of the KNOWN causes was not even explored, was a mistake of huge proportion. But now the myth of whiteners and oral cancers is fully embedded in the public's mind and I get an email about it every week.....

The inability to eliminate other causative factors makes it easy to publish a paper which reaches a conclusion that something "needs further investigation", but by the time that last sentence is read, the masses have been exposed to the faulty studies sensational claim. That was exactly the case with the study - every major paper in the US picked up the story and ran with it...but guess how many looked at it carefully and thought about the lack of looking for alternative conclusions (HPV) or used the sentence that it was a preliminary finding and needed further looking into...none. Meta-analytical techniques are of great value when the need to establish repeatability is considered. A review of 12 different hospitals experience might be proof that even with different surgeons, different facilities, but the same population characteristics of patients would all have a similar outcome, validating the procedure, even when used in different environments by different doctors of varying skills.

I like this technique when you look at something historically over a protracted period. A retrospective look at cervical cancer death rates from a period of 1948 to 1958 would show a dramatic almost 70% reduction in death rates. That data would be partly assembled from a variety of sources since the SEER system was not fully implemented at the time. The FACT that the death rate dropped would be established from numerous sources, but the causative agent was left for another publication and study