Caloric restriction reduces age‐related but not all‐cause mortality

Recently, investigators conducting a large study of caloric restriction(definition) (CR) in non-human primates (macaca mulatta) presented their 25-year interim report on lifespan with positive results for both all-cause mortality and age-related mortality (Colman et al. 2014). This was in contrast to the other ongoing study that presented their latest interim analysis in 2012 with no significant differences for either all-cause or age-related mortality (Mattison et al. 2012). CR has long been suggested to prolong life span and is now widely reported as a well-validated method of prolonging the average life expectancy with important implications for humans (Colman et al. 2014). However, it should be noted that only three studies of CR have been reported in non-human primates, and only one of those show a significant effect on mortality (Colman et al. 2014). In contrast, the other two studies did not find significant effects on mortality (Bodkin et al. 2003, Mattison et al. 2012). There are marked differences between the three studies in terms of animal provenance, age of inclusion as well as in the protocols for caloric restriction as has been reviewed at length (Kemnitz 2011). However, if CR is to be a credible intervention in humans, the effect has to be robust across both genetic and environmental variations. Therefore, an interim meta-analysis of the results of CR in non-human primates to date is both timely and interesting, as indeed was suggested by Colman and co-workers: ‘It is highly likely that factors important in ageing that are not evident in either study independently will emerge from the analysis of the aggregate data’. Mortality data were collected from the latest publications of the three long-term studies of caloric restriction in rhesus monkeys both using all-cause mortality and age-related mortality as defined in the studies (Table 1). The studies by Mattison et al. (2012) and Colman et al. (2014) report both all-cause and age-related mortality. The study by Bodkin et al. (2003) only had three events in the CR group and did not define age-related deaths. However, they did report the cause of death in detail for the three mortalities in the CR group, and only one died of age-related disease as defined in the other studies. For the control group, age-related death was defined as cardiac- or diabetes-related death (n = 29), as neoplasia was not reported. In total, 324 animals were included whereof 103 were treated with CR. The meta-analysis using the Mantel–Haenszel statistic in the RMeta package (version 2.16 under R version 3.0.2; Auckland, New Zealand) did not reach significance for all-cause mortality (Table 1 and Fig. 1a), but there was a significant effect on age-related mortality (Table 1 and Fig. 1b). Power analysis using the pwr package (version 1.1.1; Lyon, France) showed 91% power to detect age-related mortality for the effect estimated in the meta-analysis. However, the low estimated effect for all-cause mortality results in only 11% power, or a suggested sample size of 2800 animals per group for a power of 85%. Thus, meta-analysis shows that CR does not decrease all-cause mortality in non-human primates based on the published data. While the meta-analysis for age-related mortality was significant, the definition of age-related mortality is problematic. Firstly, as the study by Bodkin and co-workers did not report it in the way reported by others (Bodkin et al. 2003), the age-related mortality rate in their control group has been estimated from morbidity data. The calculation used should provide a conservative estimate as neoplasia, and multi-organ dysfunction has been excluded, but solid data using the same definition would be better. However, even if we disregard this difference between the studies, there is a clear and well-documented increase in the risk of dying of any given disease or complication with increasing age (Jakobson et al. 2014). In addition, both all-cause mortality and complication-related mortality are known to be higher with lower body mass index (Flegal et al. 2014), which may tend to increase mortality of non-age-related causes in these studies. That is, while CR reduces age-related disease, it results in increased frailty that almost completely removes the total benefit in terms of mortality. Thus, the age-related mortality carries a significant risk of bias and should be regarded with care. It should be noted that all three studies included in this analysis have shown that CR has marked potentially beneficial effects on metabolism and morbidity in a large number of interim studies, as has been recently reviewed (Kemnitz 2011). While this seems promising, any final conclusion should be based on the total life span of the animals including all-causes of death, which carries the lowest risk of false-positive results. There are several ways of performing meta-analysis from re-analysis of all included individuals as a single study to the present way of studywise comparison. Of the methods of studywise comparison, the present analysis based on the Mantel–Haenszel statistic is considered the best when there are included studies with a low number of events such as the Bodkin study. However, it may be noted that this meta-analysis is an interim study of mortality, while the original studies were designed to evaluate life span. The narrow confidence intervals indicate that the estimated effect sizes are reasonable. Furthermore, this means that while age-related mortality appears robustly decreased by CR, the effect on total mortality is of negligible importance as an intervention in reducing the mortality in limited populations. The power analysis indicates that present study designs and sizes are much too small to robustly detect an effect on total mortality. A meta-analysis based on individual-level data on life span performed when all studies are finished will have greater statistical power to resolve small effects of CR not evident in the present analysi