SaskiaMiddeldorp
commented
4 years ago Dear Armand Happy NY to you as well. I have thought about the paper but did not make notes (yet). I think even if the findings are "negative" it is worth reporting, even though it is not as exciting. This is my first thought ;-). With regard to migraine: migraine clusters with OC because migraine with aura increases the risk of stroke and may be aggravated by OC use - I am not surprised by the green bars here. I will think about it in the next days and discuss with Eva, and get back to you!
Armand1
Dear Eva and Saskia,
Happy New Year!
I have been working, albeit a bit sporadically, on the remaining section of the paper.
In particular, I have been looking at the best way to represent the way in which the side-effects of oral contraception and hormonal therapy have been studied in our papers. Now, I have tried many things. For example, I spent days looking at various clustering methods (HCA, network communities etc). The hope was that OC, HT venous thrombosis (VT), breast-cancer etc topics might “cluster” together because they are found in the same papers — that one could just “ let the data speak” and say, “here is a natural group of topics”. The bottom line is that they do, sort of, sometimes, depending on how you subset the data (into groups of years, or papers just about women and so on) but the signal is clearly very weak. I mean that there are loads of papers about each of these topics that are not about the others.
So, I took a different approach. I just said: let’s pull out all the OC and HT papers: 1606 of them, by my definition (cutoff topic probability >0.05). Then, I said: let’s just look at the 146 topics that are clearly about “medical conditions”, that is, identify potential side effects (e.g.. “venous thrombosis”). I then asked, for of my 1606 papers if they were also about each medical condition, estimated the frequency of each for three eras, restricted my attention to the top 10 (combining the rest as "other") and found this:
So, it shows that VT is actually not THAT important. It's found in about 2-5% of OC+HT papers. Perhaps you can add to that two other generally topics: "cardiac risk factors" and "hypertension". They're comparably important. But so is migraine (weirdly --- who would have guessed --- why?) and then breast cancer etc.
These frequencies change a bit over time (looking by year rather than era shows more fluctuations of course, but the number of papers per year is smaller, so the estimates are less reliable). Osteoporosis becomes more popular in the 2000s as the emphasis switches from OC to HRT papers. But they're not easy to tell apart since an HRT paper will usually reference OC and hence be 'about" it too. I mean -- one could, but is it worth the effort? I have just chosen to consider them together.
What do you make of this? We need to write. But I worry that this just isn't a very exciting result. Yes, VT and allied conditions are important, but they hardly drive the study of OC & HRT and, yes, it changes in frequency a bit (perhaps statistically significantly), but not very much. Thoughts?
Here is my analysis
side-effects-of-oral-contraceptives-and-hormonal-therapy.pdf