The statistical analysis and reality
Michaëlsson et al. (2014) examined whether high milk consumption is associated with mortality and fractures in women and men. Based on data obtained via food frequency questionnaires, processed by multivariable survival models with the purpose to determine the association between milk consumption and time to mortality or fracture, the study found that high milk intake was associated with higher mortality in one cohort of women and in another cohort of men, and with higher fracture incidence in women.
My point is that the paper itself with data supplement hasnot shown in a satisfactory way that the results obtained are results that are valid for the population of humans from which the samples were drawn.
I suggest that more efforts should have been made in showing why the steps taken in the processing of data were motivated, that change the results of consuming milk from being positive for health to being negative. Combination of data in Table 1 and 2 (ibid.) gives the results in Figure 1.
Figure 1. Change of mortality among women measured in two ways in groups with different reported amounts of milk consumed in Michaëlsson et al. (2014), reference value 1.00 for the group with lowest consumption (mean 60 g per person and day), “Mortality” is the actual fraction that died per subsample, “Age adjusted HR” is one mortality measure in Michaëlsson et al.
One reason for the difference between measures is if the distribution among ages differed between the subsamples. The mean ages from the group with the lowest reported consumption of milk to the one with the highest were 53.2, 54.0, 54.1 and 52.8 years, i.e. almost identical. Given the substantial number of observations per subsample and assuming that the populations behind the subsamples had identical age-distributions, the probability is low that pure chance should affect the subsamples in such a way that the measure in Figure 1 shows such a smooth line of decreasing mortality with increasing reported milk intake measured by “Mortality” while the measure “Age adjusted HR” shows a similar smooth but increasing mortality with increasing reported milk intake.
Another possibility is that the method generating “Age adjusted HR” systematically distorts the analysis so that obtained results are a biased representation of the characteristics of the populations from which the subsamples were drawn. I do not say that this is a case. I suggest that this is a possibility.
Reported energy intake
Figure 2 shows reported average energy intake (ibid.) in the different subsamples in relation to official recommendations in Sweden (Nordic Council of Ministers 2014) for women at the age of 31-60 years, with average physical activity, and for men at the age of 60. The difference because the average age at entrance of the study differed between men and women.
Figure 2. Relation between reported average energy intake of recommended intake and mortality among women and men in four groups with different reported average intake of milk. Mortality measured as the fraction of subsamples that died.
For both cohorts, women and men, reported energy intake increased with reported amount of milk consumed. As averages, reported consumption of food covered energy recommendations for all subsamples in the male study, while it was lower than recommendation for all subsamples in the female study. The subsample of women with highest reported milk consumption had on average an intake of 94% of recommended while the subsample with the lowest reported intake reached 67% of recommendation. Interesting is that the mean body mass index and the mean length in both the female and the male study was similar among subsamples. Figure 2 points towards two possible sources of error regarding the female study: That
(i) There was an increasing deficit in nutrient (energy) supply with decreasing reported amount of milk consumed, and/or that
(ii) the share of food intake reported of total food intake decreased with decreasing reported amounts of milk consumed.
Aspect (i) could manifest itself in increasing mortality with lower intake of milk due to insufficient nutrient supply, aspect (ii) gives decreasing accurateness of data with decreasing reported milk intake, as the fraction of food actually consumed not captured increases.
The difference in reported energy intake per day for the female subsample with the lowest and highest reported intake of milk, respectively, was 553 kcal per day. That equals 2.31 MJ. Over one year, as an average, women in the subsample with the highest intake of milk consumed 845 MJ more energy than the group with the lowest intake, i.e. the amount of energy in 23 kg fat. The physical activity measured in metabolic equivalents was equal among subsamples. Although the conclusion of Michaëlsson et al. (2014) that a higher intake of milk/dairy products is associated with a higher total energy intake, without affecting the weight is supported by four references, the conclusion can be questioned. If this would be the case, then current way of estimating energy content in dairy products does not reflect their true value in the human energy metabolism. To my knowledge, dairy products given the context of Northern Europe are among the food items best examined. Rather, I suggest, the differences in reported energy intake reflects a weakness in the data gathering in the study affecting the reliability of the study.
Michaëlsson, K., Wolk, A., Langenskiöld, S., Basu, S., Warensjö Lemming, E, Melhus, H. & L. Byberg. 2014. Milk intake and risk of mortality and fractures in women and men: cohort studies. BMJ 2014;349:g6015.
Nordic Council of Ministers. 2014. Nordic Nutrition Recommendations 2012 Integrating nutrition and physical activity. Report Nord 2014:002.
Competing interests: No competing interests