meta-analyses – testing relationships

One of the nice things about working at a university is that there is almost always an interesting talk to go to (supposing you have the time…). Yesterday I managed to go to a fascinating discussion of the use of meta-analyses by a Waikato graduate, Shinichi Nakagawa. (I suspect that Grant knows much more about this technique than I do, but Shinichi’s talk was very post-provoking.)

Shinichi began his studies here a year or so after I joined the University staff. He was invited to do a BSc(Hons), something reserved for the really able students, & it’s a sign of the quality of his research project (on zebra finches) that he’s since published five papers from it. After gaining his PhD at a UK university, Shinichi’s come back to New Zealand to work at the Universit of Otago. Over the last few years he’s begun to use meta-analyses more & more to identify relationships (or the lack of them) between data sets, & this particular research tool & its applications were the subject of his enthusiastic presentation. (He recommended Morton Hunt’s 1997 book How science takes stock: the story of metanalysis as an excellent introduction to the area, written by a journalist for a lay audience.)

I’ve referred to meta-analyses myself, from time to time. The Cochrane Collaboration, for example, makes use of them in examining the (claims for) efficacy of various medical treatments. Essentially the technique involves combining the data from a range of studies (allowing for sample sizes & so on), producing summary results that may allow recognition of a statistically significant pattern in the data. (It may equally show that there’s no real relationship between a set of factors, as Shinichi noted. Alternative treatment modalities such as acupuncture & homeopathy, for example, have been shown to perform no better than placebo as an outcome of  meta-analyses.) The larger sample size afforded by a meta-analysis allows greater confidence about the results.

The technique does have its disadvantages: you could be accused of comparing apples with oranges, for example, although this could be overcome by proper selection of the studies for inclusion in the analysis. There’s perhaps a greater problem (Shinichi described it as the ‘file drawer’ problem) whereby studies with a non-significant result are less likely to be published, thus biasing the pool of studies available for inclusion in a meta-analysis. Shinichi described how he’d had difficulty publishing a paper on sparrow parental care because it had a negative result. Now, there’s nothing wrong with this, & a fairly large proportion of experiments would come up with such an outcome, but unfortunately this doesn’t make good headlines:-) And the editorial attitude described here both skews publications in a particular direction & also skews public perceptions of how science is done.

We then heard about a few research areas where meta-analysis had been applied to a large body of data to test prevailing views. Should we take antioxidant supplements, for example? Not according to a 2007 study, that found an increased mortality rate associated with their use. While this particular study had its critics, a paper published last year in Nature found evidence that antioxidants can actually help cancerous cells to grow. (Mind you, there is a need for caution in interpreting studies like this last one, given that it was done on cancer cells in vitro – there needs to be a fair bit of follow-up work to see if this holds true in the body.)

The final example in Shinichi’s talk looked at the widespread view that a restricted calorie intake can prolong life. (Obviously there would be limits to this one & it’s not a case of living longer & longer on less & less. Eventually there’d be a point in which the lifespan was shortened rather abruptly. And terminally.  Rather like the work of Famine in the wonderful Good Omens by Terry Pratchett & Neil Gaiman.)

Anyway, individual studies of rats & monkeys, fruit flies & nematodes, & even yeast, seem to bear out this idea – it looks like a general biological phenomenon. Our speaker seems to enjoy doing meta-analyses – he commented that a preliminary review f the literature, & an introductory analysis, shows no conclusive evidence that calorie restriction has any positive effect on the length of life, not as a general principle. He found there was no consistency in the data for different species. In the discussion at the end, someone pointed out that almost all the work in the area of dietary restriction’s been done in lab-bred animals, and might not reflect what happens in ‘wild-type’ individuals. And it’s more important to look at the carbohydrate/protein balance in the diet, rather than the overall reduction in calories.

And, of course, the sting in the tail – from an evolutionary point of view, if you don’t produce fertile offspring & thus pass on your genes, the length of life is actually irrelevant. If you lived to 120, but left no fertile children, you’d be an evolutionary non-event…

One thought on “meta-analyses – testing relationships”

  • Jim Thomerson says:

    Your last paragraph is overstated. You are only considering simple Darwinian fitness, a measure of how many of your offspring reach sexual maturity. However, perhaps more important is inclusive fitness, a measure of how many copies of your genes make it into the next generation. Even though you might have 0 simple Darwinian fitness, you could help relatives, who share a percentage of your genes, to increase their simple Darwinian fitness. Doing so would retain you as an evolutionary event. As in some sports, where players get recognition for assists as well as goals.

Leave a Reply

Your email address will not be published. Required fields are marked *