Umbrella Review on the Health Benefits of Coffee Consumption

Roasted ground coffee (Coffea spp., Rubiaceae) beans provide one of the most-consumed drinks worldwide. Studies have reported varying conclusions about its effects, depending on outcomes, design, coffee type, and other variables. Beans can be roasted to varying degrees and processed in different ways; these processing steps can highly alter the chemical composition. Roasted coffee can contain over 1000 different bioactive compounds, many of which have reported health benefits. The most promising compounds include caffeine, polyphenols, and diterpenes, although the antioxidant activity of polyphenols reportedly varies between subjects’ genotypes and gut microbiota. Existing reports assess links between coffee at different consumption rates and all-cause mortality; cancer; diseases of the cardiovascular (CV), metabolic, neurological, musculoskeletal, gastrointestinal (GI), and hepatic systems; and pregnancy outcomes.

Most research has been observational and summarized in systematic reviews (SRs) and meta-analyses (MAs). Since observational trials cannot show causality, randomized controlled trials (RCTs) are needed. RCTs may be informed by existing higher-level evidence to place the strength of associations across outcomes in perspective, prioritize research needs, and reduce any potential harms of increased intake. The authors searched electronic databases for SRs and/or MAs on all outcomes linked with a specific coffee exposure. Studies were included if coffee intake was all or part of the exposure of interest, or of a subgroup analysis. Eligible MAs were included in an umbrella review. Studies reporting on caffeine and not coffee were excluded, as were those on genetic polymorphisms in coffee metabolization. Databases were searched from inception through July 2017, with 1180 unique titles found, and 917 excluded by title and/or abstract, leaving 263 for screening. Of these, 126 did not meet eligibility criteria. Hand searches of references added four reports. Of 141 MAs, 135 were of observational studies reporting 67 outcomes (n=201); six, of RCTs measuring nine outcomes (n=17).

Data extraction was independently performed by two authors. Methodological quality of MAs was measured using AMSTAR (A MeaSurement Tool to Assess Systematic Reviews). The Grading of Recommendations Assessment, Development and Evaluation (GRADE) method was used to assess evidence for outcomes in the umbrella review. Results of 83 MAs were reanalyzed using the DerSimonian and Laird random-effects model. The 10 most harmful and most beneficial associations of three measures of coffee exposure (high vs. low, any vs. none, and intake of one more cup/day) are represented in forest plots of relative risks (RRs) across relevant MAs. For example, in MAs reporting on high vs. low intake, six of the 10 most harmful effects were pregnancy-related. A strong benefit was seen in hepatic outcomes, with cirrhosis, liver cancer, and chronic liver disease all showing reduced RRs at higher intake. In 27 cohort studies on type 2 diabetes (T2D), lower RRs for higher intake of coffee were found in MAs with some of the better AMSTAR scores. Coffee’s effects on some pregnancy outcomes were among the 10 most harmful in MAs of any vs. no consumption; liver benefits, among those for any vs. none and one-cup-more/day MAs; and T2D, also among the latter. Benefits of coffee against Parkinson’s disease were seen in MAs of all three measures of exposure; against Alzheimer’s disease, in those of the latter two. For high vs. low exposure, significance was reached for beneficial links with 19 outcomes; harmful ones, for six. For any vs. no intake and one-cup-more/day, significance was reached for beneficial links with 11 outcomes and harmful ones for three in each type of MA.

Coffee’s reported benefits in all-cause mortality, liver/GI outcomes, CV-related mortality, metabolic disease (including T2D), and neurological diseases are discussed, as are benefits and/or harms in cancer, musculoskeletal conditions (with a gender variation in risk of fracture), gynecological concerns, and pregnancy. In discussing CV benefits, the authors point out the paradox of unfavorable changes seen in lipid profiles, especially from unfiltered coffee’s diterpenes. Some MAs differentiated between decaffeinated coffee (DC) and natural coffee. Some reports of potential harms did not find harms for DC. While coffee is associated with reduced RRs of several cancers, there was a consistent association of higher RRs for lung cancer with high vs. low consumption. This was less in studies that corrected for tobacco (Nicotiana tabacum, Solanaceae) smoking, strongly associated with coffee drinking and the greatest single risk factor for lung cancer. An MA of two studies found high vs. low DC intake associated with lower RR of lung cancer. In other reports, DC did not impact lipid profiles as did natural coffee.

Of 83 MAs reanalyzed (83% of those for high vs. low intake; 79%, any vs. none; 0, one-cup-more), about 40% had significant heterogeneity; in 90% of those, I2>50%. Of 54 MAs not reanalyzed, 19% had significant heterogeneity. Egger’s test for publication bias could be performed only for 40% of reanalyzed MAs; the rest contained <10 studies. Of those reanalyzed, 20% had evidence of bias. The median AMSTAR of these MAs was 5 of a possible 11. Quality of evidence for each outcome, by GRADE classification, was poor, with 25% rated low; 75%, very low. Even MAs of RCTs had low-quality evidence. Few MAs included dose-response analyses. Where these were done and suggested non-linearity, the largest RR reduction was seen at 3-4 cups/day, with higher intake resulting not in harm but in less benefit. Effect sizes of >2 or <0.5 allow observational evidence to be upgraded in the GRADE system. This level was reached only for associations between high vs. low coffee intake and both liver cancer and chronic liver disease. Importantly, for future RCTs, the evidence suggests no convincing harms of coffee drinking except for pregnant women and possibly those at increased risk of fracture, who should be excluded from RCTs until more evidence is available. Pregnancy outcomes may be improved with DC since caffeine’s half-life is doubled in pregnant women. RCTs with coffee intake optimized at 3-4 cups/day would be unlikely to harm other adults.

One of the authors (Fallowfield) has received research grants from GlaxoSmithKline and Intercept Pharmaceuticals and personal fees from Novartis and Merck. Another author (Hayes) has received grants from Roche and personal fees from MSD, Gilead, AbbVie, Janssen, BMS, Pfizer, Novartis, and Roche.


Poole R, Kennedy OJ, Roderick P, Fallowfield JA, Hayes PC, Parkes J. Coffee consumption, and health: umbrella review of meta-analyses of multiple health outcomes. BMJ. November 22, 2017;359:j5024. doi: 10.1136/bmj.j5024.