Lancet Volume 384, Issue 9937, 5–11 July 2014, Pages 29–36
Some extracts from the abstract:
Overall, 1228 women were recruited and randomly assigned between June 15, 2007, and July 15, 2011, 1078 of whom completed the trial and were included in the analysis.
309 (58%) women in the low-dose aspirin group had livebirths, compared with 286 (53%) in the placebo group (p=0·0984; absolute difference in livebirth rate 5·09% [95% CI −0·84 to 11·02]).
Preconception-initiated low-dose aspirin was not significantly associated with livebirth or pregnancy loss in women with one to two previous losses. …. Low-dose aspirin is not recommended for the prevention of pregnancy loss.
So – the interpretation is a so-called “negative” trial i.e. one that did not show any evidence of effectiveness.
BUT… the original planned sample size was 1600, with 1254 included in analyses (the other 346 being the 20% allowance for loss to follow up), which was calculated to have 80% probability of a “significant” result if there was in reality a 10% increase in live births in the intervention group from 75% in the control group.
In fact the trial recruited 1228 and lost 12.2% so only 1078 were included in the analyses (86% of the target). The placebo group incidence was different from expectation (53% compared with 75%) and the treatment effect was about half of that the sample size was calculated on (absolute difference of 5% rather than 10%), though they were more similar expressed as risk ratios than risk differences (1.09 compared with 1.13). Nevertheless the treatment effect was quite a bit smaller than the effect the trial was set up to find.
So is concluding ineffectiveness here reasonable? A 5% improvement in live birth rate could well be important to parents, and it is not at all clear that the 10% difference originally specified represents a “minimum clinically imporant difference”. So the trial could easily have mised potentially important benefit. This isn’t addressed anywhere in the paper. The conclusions seem to be based mainly on the “non-significant” result (p=0.09), without any consideration of what the trial could realistically have detected.
Original post 17 July 2014