Together with several co-authors, with first author AN in the lead, we did a meta analyses on the role of Lp(a) as a risk factor of stroke. Bottomline, Lp(a) seems to be a risk factor for stroke, which was most prominently seen in the young.
The results are not the only reason why I am so enthusiastic by this article. It is also about the epidemiological problem that AN encountered and we ended up discussing over coffee. The problem: the different studies use different categorisations (tertiles, quartiles, quintiles). How to use these data and pool them in a way to get a valid and precise answer to the research question? In the end we ended up using the technique proposed used by D Danesh et al. JAMA. 1998;279(18):1477-1482 that uses the normal distribution and the distances in SD. A neat technique, even though it assumes a couple of things about the uniformity of the effect over the range of the exposure. An IPD would be better, as we would be free to investigate the dose relationship and we would be able to keep adjustment for confounding uniform, but hey… this is cool in itself!