The answer appears in a newly published paper:
“Is it dangerous or beneficial to drink coffee? Reflections on a meta-analysis on risk at birth and a population study on risk in late life,” Ingmar Skoog, European Journal of Epidemiology, epub October 11, 2014.
Professor Ingmar Smoog (pictured here) is head of the Neuropsychiatric Epidemiology research group at the Centre for Health and Aging AGECAP at the Institute of Neuroscience and Physiology at the University of Gothenburg in Gothenburg, Sweden.
Professor Smoog appears (to the eyes of this non-specialist) to explain that he reaches no firm conclusion.
The final paragraph of Professor Smoog’s paper clarifies (or does not clarify) the situation, in wording that is (or is not) clear:
When using approaches such as meta-analyses, systematic reviews or pooling data, or doing any other evaluation of results from different studies, it needs to be emphasized that characteristics that may influence associations, i.e. factors which stimulate or inhibit the effect of another factor, including genetic or social heterogeneity, may differ considerably between different study locations, even within the same nations or even the same geographical area. Thus, an association may be true, even if it is not confirmed in all studies, and even if results are conflicting. This may be especially true regarding caffeine or coffee consumption, which differs between countries and socioeconomic groups. This is also important to keep in mind in a time when the research community is asking for ‘big data’, i.e. pooling of heterogeneous data sets from many different sources of different quality and with different methodologies and selection of study populations.