In the study by Holmes et al. it is concluded that “the reduction of alcohol consumption, even for light to moderate drinkers, is beneficial for cardiovascular health” .
However in Fig. 2 there are no statistically significant effects for the groups of "light drinkers", "moderate drinkers" and "heavy drinkers" when these groups are taken separately (the 95% confidence intervals cross the "no effect" line).
A statistically significant effect is reported for the broader group of "drinkers only". What confuses me is that the average protective effect for the "drinkers only" group is larger than the average effect of any of the three following groups: "light drinkers", "moderate drinkers" and "heavy drinkers". This should not be so.
From Fig. 2 we see the "light drinkers", "moderate drinkers" and "heavy drinkers" groups can be combined to a total of 9827 cases and 97243 individuals. In the "drinkers only" group there are 10130 cases and 107478 individuals. This leaves us with some unknown group (let’s call it X).
Group X consists of very strange people who are classified as "drinkers only" but not as "light drinkers", "moderate drinkers" or "heavy drinkers" and in this group there are 303 cases and 10235 individuals.
The proportion of cases to the number of individuals is as follows:
Non-drinkers: 5883/43029 ~ 0,137
Light: 4686/47246 ~ 0,099
Moderate: 3222/33772 ~ 0,095
Heavy: 1919/16225 ~ 0,118
Group X: 303/10235 ~ 0,03
This means that this group X has 3-4 times less cases per individual than any other mentioned group. When this strange group is mixed with the “light”, “moderate” and “heavy” drinkers the “protective effect” appears to increase so dramatically for the group of “drinkers only” that the effect becomes statistically significant and stronger (by average effect size) than in any of the three groups (of "light", "moderate" and "heavy" drinkers).
This suggests the presence of a discrepancy in the data that could have led to an erroneous conclusion.
1. Holmes MV, Dale CE, Zuccolo L, et al. Association between alcohol and cardiovascular disease: Mendelian randomisation analysis based on individual participant data. Bmj 2014;349:g4164 doi: 10.1136/bmj.g4164 (Published 10 July 2014)
Competing interests: No competing interests