Thursday, December 1, 2016

When statistics are meaningless

Low-quality evidence renders statistics meaningless

John Ioannidis, noted canary in the coal mine of bad science, just put an article on ResearchGate that caught my attention. Late on November 30, 2016, he posted the latest article on which he was a co-author. The article is called “Diet, body size, physical activity, and the risk of prostate cancer.” Here’s the abstract, and here’s the full article. This article reviews the meta-analytic evidence regarding the risk factors for prostate cancer.

The findings summarized by the abstract? 176 out of 248 meta-analyses used continuous exposure assessment to measure the impact of each factor. Of those 176, none satisfied all of the authors’ pre-set criteria for using the best meta-analytic methods to provide strong evidence of the factors linked to prostate cancer. Not one.

The authors graded the strength of evidence in these meta-analyses according to the following categories: strong, highly suggestive, suggestive, and weak. The only strong, reliable risk factor for developing prostate cancer? Height.

For every 5 additional centimeters in height, the risk of developing prostate cancer increases by 4%.
  • Quick, somebody, feature the headline:
    Does Being Tall Give You Cancer? Shocking New Research Shows That Taller Men Are More Likely to Develop Prostate Cancer
How are my clickbait-headline-writing skills?

...Okay, the scientist in me demands that I present a more serious and evenhanded treatment of the topic. So, I’ll report that there is also some evidence to suggest that BMI, weight, amount of calcium in the diet, and alcohol intake are also factors that appear to impact prostate cancer development.

However, the authors did emphasize in the abstract that “...only the association of height with total prostate cancer incidence and mortality presented highly suggestive evidence...” The other factors I listed above are “supported by suggestive evidence.”

But, considering the reflections on the state of biomedical science that Ioannidis published in February of 2016, one wonders just how “suggestive” that evidence really is!

I think this represents a good applied example of why an understanding of stats is important in today’s world! I think it’s also a good example of how easily the competition for funding can corrupt proper scientific procedures! But, lest you think I’m trying to pick on biomedical research, here’s another example that hits frighteningly close to home.

My takeaway: No matter how advanced your statistical techniques or how powerful your software, statistics are meaningless when the evidence itself is biased...

No comments:

Post a Comment