European Journal of Nutrition

, Volume 53, Issue 1, pp 25–38

Coffee and caffeine intake and incidence of type 2 diabetes mellitus: a meta-analysis of prospective studies

Authors

  • Xiubo Jiang
    • Department of Epidemiology and Health Statistics, Medical CollegeQingdao University
    • Department of Epidemiology and Health Statistics, Medical CollegeQingdao University
  • Wenjie Jiang
    • Department of Epidemiology and Health Statistics, Medical CollegeQingdao University
Review

DOI: 10.1007/s00394-013-0603-x

Cite this article as:
Jiang, X., Zhang, D. & Jiang, W. Eur J Nutr (2014) 53: 25. doi:10.1007/s00394-013-0603-x

Abstract

Purpose

Coffee and caffeine have been linked to type 2 diabetes mellitus (T2DM). A dose–response meta-analysis of prospective studies was conducted to assess the association between coffee and caffeine intake and T2DM incidence.

Methods

Pertinent studies were identified by a search of PubMed and EMBASE. The fixed- or random-effect pooled measure was selected based on between-study heterogeneity. Dose–response relationship was assessed by restricted cubic spline.

Results

Compared with the lowest level, the pooled relative risk (95 % CI) of T2DM was 0.71 (0.67–0.76) for the highest level of coffee intake (26 articles involving 50,595 T2DM cases and 1,096,647 participants), 0.79 (0.69–0.91) for the highest level of decaffeinated coffee intake (10 articles involving 29,165 T2DM cases and 491,485 participants) and 0.70 (0.65–0.75) for the highest level of caffeine intake (6 articles involving 9,302 T2DM cases and 321,960 participants). The association of coffee, decaffeinated coffee and caffeine intake with T2DM incidence was stronger for women than that for men. A stronger association of coffee intake with T2DM incidence was found for non-smokers and subjects with body mass index <25 kg/m2. Dose–response analysis suggested that incidence of T2DM decreased by 12 % [0.88 (0.86–0.90)] for every 2 cups/day increment in coffee intake, 11 % [0.89 (0.82–0.98)] for every 2 cups/day increment in decaffeinated coffee intake and 14 % [0.86 (0.82–0.91)] for every 200 mg/day increment in caffeine intake.

Conclusions

Coffee and caffeine intake might significantly reduce the incidence of T2DM.

Keywords

CoffeeCaffeineType 2 diabetes mellitusDose–response meta-analysis

Introduction

The global age-standardized adult diabetes prevalence increased from 8.3 % in 1980 to 9.8 % in 2008 in men and from 7.5 to 9.2 % in women, and the number of people with diabetes increased from 153 million in 1980 to 347 million in 2008 [1]. Globally, 12 % of the health expenditures are estimated spent on diabetes in 2010 [2]. Given the high burden of diabetes and the associated costs, prevention through weight control, physical activity and improved diet quality is crucial [1]. Coffee is one of the most popular beverages in the world, and the latest coffee trade statistics estimated that world coffee export amounted to about 6.76 billion kg in 2011/2012 [3]. A number of biologic roles have been identified by which coffee components might affect diabetes development, and an increasing number of studies were conducted to assess the association of coffee intake with diabetes [4]. Results from the recent population-based prospective studies have not been summarized [513], and the dose–response relationship as well as the potential threshold effect of coffee intake on type 2 diabetes mellitus (T2DM) incidence is also unknown. In addition, the effect of caffeine intake on T2DM incidence is also not summarized. Although evidence from long-term randomized trials is ideal, these studies are difficult to implement on a practical basis, especially for an exposure such as coffee consumption. Considering prospective cohort studies do not suffer from recall bias and are anticipated to be less likely to have selection bias relative to case–control studies and are also believed to provide better evidence for causality in which coffee consumption precedes T2DM incidence compared with case–control studies, cross-sectional and ecologic studies. Therefore, we conducted a meta-analysis of prospective cohort studies to (1) first assess the T2DM incidence for the highest versus lowest categories of coffee, decaffeinated coffee and caffeine intake; (2) then evaluate the possible dose–response relationship of coffee, decaffeinated coffee and caffeine intake with T2DM incidence; (3) evaluate the modification of key covariates to the association of coffee, decaffeinated coffee and caffeine intake with T2DM incidence; (4) and assess the heterogeneity among studies and publication bias.

Methods

Literature search and selection

We performed a literature search up to August 2013 using the databases of PubMed and EMBASE, using the following search terms: coffee or caffeine and diabetes mellitus without restrictions. Moreover, we also reviewed the reference lists from retrieved articles to search for further relevant studies.

Two investigators independently reviewed all identified studies, and studies were included if they met the following criteria: (1) a prospective design; (2) the exposure of interest was coffee or caffeine; (3) the outcome of interest was T2DM; (4) multivariate-adjusted relative risk (RR) with 95 % confidence interval (CI) was provided; (5) for dose–response analysis, the number of cases and participants or person-years for each category of coffee, decaffeinated coffee or caffeine intake must be also provided (or data available to calculate them). If data were duplicated in more than one study, we included the study with the largest number of cases.

Data extraction

The following data were extracted from each study by two investigators: the first author’s last name, publication year, sample size and number of cases, location where the study was performed, variables adjusted for in the analysis, RR estimates with corresponding 95 % CI for the highest versus lowest categories of coffee, decaffeinated coffee and caffeine intake, respectively. For dose–response analysis, the number of cases and participants (person-years), and RR (95 % CI) for each category of coffee, decaffeinated coffee and caffeine intake were also extracted. The median or mean level of coffee, decaffeinated coffee and caffeine intake for each category was assigned to the corresponding RR for every study. If the upper boundary of the highest category was not provided, we assumed that the boundary had the same amplitude as the adjacent category [14]. We extracted the RRs that reflected the greatest degree of control for potential confounders.

Statistical analysis

Pooled measure was calculated as the inverse variance-weighted mean of the logarithm of RR with 95 % CI to assess the strength of association between coffee, decaffeinated coffee and caffeine intake and T2DM incidence. The I2 was used to assess heterogeneity, and I2 values of 0, 25, 50 and 75 % represent no, low, moderate and high heterogeneity, respectively [15]. The fixed-effect model (FEM) was used as the pooling method if moderate or lower heterogeneity (I2 < 50 %) was found; otherwise (I2 > 50 %), the random-effect model (REM) was adopted that considers both within-study and between-study variation. A sensitivity analysis was performed with one study removed at a time to assess whether the results could have been affected markedly by a single study. Publication bias was evaluated using the Egger regression asymmetry test. Meta-regression with restricted maximum likelihood estimation was performed to assess the potentially important covariates that might exert substantial impact on between-study heterogeneity. Study quality was assessed using the 9-star Newcastle-Ottawa Scale (http://www.ohri.ca/programs/clinical_epidemiology/oxford.asp, accessed 9/5/2013).

For dose–response analysis, a two-stage random-effects dose–response meta-analysis [16] was performed to compute the trend from the correlated log RR estimates across levels of coffee, decaffeinated coffee and caffeine intake, respectively, taking into account the between-study heterogeneity. In the first stage, a restricted cubic spline model with three knots at the 25th, 50th, and 75th percentiles of the levels of coffee, decaffeinated coffee and caffeine intake, was estimated using generalized least square regression taking into account the correlation within each set of published RRs [17]. Then the study-specific estimates were combined using the restricted maximum likelihood method in a multivariate random-effects meta-analysis [18]. A P value for nonlinearity was calculated by testing the null hypothesis that the coefficient of the second spline is equal to 0. All statistical analyses were performed with STATA version 12.0 (Stata Corporation, College Station, TX, USA). All reported probabilities (P values) were two-sided with P ≤ 0.05 considered statistically significant.

Results

Literature search and study characteristics

The detailed literature search was shown in Supplementary Figure 1. Four articles [1922] were excluded because of duplicate reports from the same study population [11, 12, 23]. One nested case–control study [24] was further excluded. Among the 27 articles [513, 23, 2541] included in this meta-analysis, the article by Salazar-Martinez et al. [30] assessing the association of coffee, decaffeinated coffee and caffeine intake with T2DM incidence included the same cohort population with that of the article by Bhupathiraju et al. [12] assessing the association of coffee and decaffeinated coffee intake with T2DM incidence; thus, the result for caffeine intake from Salazar-Martinez et al. and the result for coffee and decaffeinated coffee intake from Bhupathiraju et al. were included in this meta-analysis. Dietary intake of coffee and caffeine was assessed by food frequency questionnaire in all included studies except for one study [40], in which a 24-h dietary recall was used. Among the 27 articles, 12 were conducted in United States, 11 in Europe and 4 in Asia. The duration of follow-up ranged from 2.6 to 24 years, with 10 articles following less than 10 years and 17 articles more than 10 years. The number of subjects ranged from 910 to 362,045, with 10 articles involving less than 30,000 subjects and 17 more than 30,000 subjects. All included studies met the quality score of 6–7 stars. All studies included in this meta-analysis adjusted for body mass index (BMI). Detailed information of the included studies in this meta-analysis is summarized in Table 1.
Table 1

Characteristics of studies included in this meta-analysis on coffee and caffeine intake and incidence of type 2 diabetes mellitus

Source

Study name

Age range (mean) (years)

No. of participant (cases)

Follow-up (years)

Assessment of T2DM

Sex

Exposure

RR (95 % CI)

Adjustment for covariates

van Dam et al. [25], Netherlands

None

30–60 (44)

17,111(306)

7

SR

M and W

Coffee

≥7 versus ≤2

0.50 (0.35–0.72)

Age, sex, town, BMI, lifestyle, CVD, HT, hypercholesterolemia

Saremi et al. [27], USA

Pima Indians Study

≥15 (27)

2,680 (824)

11

OGTT

M and W

Coffee

Occasionally

heavy versus none

0.81 (0.55–1.03)

Age, sex, BMI

Reunanen et al. [26], Finland

MCHES

20–98 (45)

19,518 (855)

16

NR

M and W

Coffee

≥7 versus ≤2

0.92 (0.73–1.16)

Age, sex, BMI, smoking, leisure time PA

Salazar-Martinez et al. [30], USA

HPFS

40–75

(53)

41,934 (1,333)

12

CSR

M

Caffeine

>417 versus <37

0.80 (0.66–0.97)

Age, BMI, PA, total caloric intake, FHDM, alcohol, smoking, intakes of glycemic load, trans-fat, polyunsaturated fatty acid, cereal fiber, magnesium and menopausal status and postmenopausal hormone use (for women)

NHS

30–55

(46)

84,276 (4,085)

 

CSR

W

Caffeine

>534 versus <140

0.70 (0.63–0.79)

Rosengren et al. [29], Sweden

BEDA study

40–75 (49)

1,361 (74)

18

SR and NR

W

Coffee

>6 versus ≤2

0.57 (0.26–1.29)

Age, smoking, low PA, education, BMI, serum cholesterol, triglycerides

van Dam et al. [31], Netherlands

Hoorn study

50–74 (61)

1,312 (128)

6

OGTT

M and W

Coffee

≥7 versus ≤2

0.69 (0.31–1.51)

Age, sex, BMI, WHR, PA, alcohol, smoking, history of CVD, use of antihypertensive medication, intake of fiber, total energy, saturated fat, polyunsaturated fat

Carlsson et al. [28], Finland

Finnish Twin Cohort

30–60 (45)

10,652 (408)

20

NR

M and W

Coffee

≥7 versus ≤2

0.65 (0.44–0.96)

Age, sex, BMI, education, leisure time PA, alcohol, smoking

Greenberg et al. [32], USA

NHANES-1

32–88 (57)

7,006 (309)

8.4

SR

M and W

Coffee

Decaf

Caffeine

≥4 versus none

0.37 (0.22–0.64)

≥2 versus none

0.43 (0.20–0.93)

>600 versus ≤150

0.65 (0.24–1.77)

Per capita income, education level, race, sex, PA, smoking, alcohol, BMI, age, type of diet

van Dam et al. [37], USA

Nurses’ Health Study II

26–46 (36)

88,259 (1,263)

10

CSR

W

Coffee

Decaf

Caffeine

≥4 versus none

0.53 (0.41–0.68)

≥2 versus none

0.52 (0.36–0.74)

528 versus 22

0.55 (0.45–0.67)

Age, BMI, PA, smoking, alcohol, use of hormone therapy, oral contraceptives, FHDM, history of HT, cereal fiber intake, history of hypercholesterolemia, sugar-sweetened soft drinks, punch, quintiles of processed meat, polyunsaturated to saturated fat intake ratio, total energy intake, glycemic index, cereal fiber intake, mutual adjustment for caffeinated and decaffeinated coffee

Iso et al. [33], Japan

JACC Study

40–65 (53)

17,413 (444)

5

CSR

M and W

Coffee

Caffeine

≥3 versus <1 cup/week

0.58 (0.37–0.90)

416 versus 57

0.67 (0.47–0.95)

Age, BMI, FHDM, smoking, alcohol, magnesium, PA, consumption of other beverages

Hu et al. [23], Finland

None

35–74 (50)

21,385 (964)

13.4

NR

M and W

Coffee

≥7 versus <2

0.61 (0.49–0.76)

Age, sex, study year, education, SBP, bread, vegetable, fruit, sausage, tea, alcohol, smoking, PA, BMI, sex

Pereira et al. [35], USA

Iowa Women’s Health Study

55–69 (61)

28,812 (1,418)

11

CSR

W

Coffee

Decaf

≥6 versus none

0.79 (0.61–1.02)

0.68 (0.43–1.09)

Age, education, baseline HT, alcohol, smoking, BMI, WHR, PA, energy intake, total fat, keys score, cereal fiber, tea, soda consumption, magnesium, phytate

Paynter et al. [34], USA

ARIC study

45–64 (54)

5,414 (718)

6,790 (719)

12

FBG/NFBG/

SR

M

W

Coffee

≥4 versus none

0.77 (0.61–0.99)

0.91 (0.70–1.18)

Age, race, education, FHDM, BMI, WHR, total caloric intake, dietary fiber, smoking, alcohol, leisure PA, HT, serum magnesium

Smith et al. [36], USA

Rancho Bernardo Study

≥50 (66)

910 (84)

8.3

OGTT

M and W

Coffee

≥5 versus none

0.60 (0.26–1.40)

Age, sex, PA, BMI, smoking, alcohol, HT, FPG

Hamer et al. [38], UK

Whitehall II study

35–55 (49)

5,823 (387)

11.7

SR/OGTT

M and W

Coffee

Decaf

>3 versus none

0.80 (0.54–1.18)

0.65 (0.36–1.16)

Age, sex, ethnicity, employment grade, BMI, WHR, smoking, sex-specific alcohol intake, PA, FHDM, HT, cholesterol, total energy intake, diet pattern, mutual adjustment for all beverage types

Fuhrman et al. [40], Puerto Rico

PRHHP

35–79 (56)

4,685 (519)

2.6

SR/FBG

M

Coffee

≥4 versus none

0.75 (0.58–0.97)

Age, BMI, smoking, FHDM, education, alcohol, PA, milk and sugar intakes

Odegaard et al. [39], Singapore

Singapore Chinese Health Study

45–74 (55)

36,908 (1,889)

5.7

CSR

M and W

Coffee

≥4 versus non-daily

0.70 (0.53–0.93)

Age, year of interview, sex, dialect, education, HT, smoking, alcohol, BMI, PA, dietary variables and magnesium

Kato et al. [41], Japan

JPHC

Study Cohort

40–69 (51)

24,826 (1,601)

31,000 (1,093)

10

CSR

M

W

Coffee

Coffee

≥5 versus none

0.82 (0.60–1.11)

0.40 (0.20–0.78)

Age, BMI, smoking, alcohol, FHDM, PA, HT, mental stress, levels of Type A behavior and hours of sleep

van Dieren et al. [5], Netherlands

EPIC-NL cohort

20–79 (49)

40,011 (918)

10

CSR

M and W

Coffee

>6 versus <1

0.84 (0.65–1.08)

Cohort, sex, age, BMI, highest education, PA, FHDM, smoking, alcohol intake, energy intake, energy-adjusted intake of saturated fat, fiber and vitamin C, hypercholesterolemia, hypertension, tea

Boggs et al. [6], USA

BWHS

30–69 (38)

46,906 (3,671)

12

CSR

W

Coffee

Decaf

≥4 versus none

0.83 (0.69–1.01)

≥4 versus none

1.10 (0.81–1.49)

Age, questionnaire cycle, energy intake, education, FHDM, vigorous activity, smoking, alcohol, glycemic index, cereal fiber, sugar-sweetened soft drinks, BMI, history of hypertension, history of high cholesterol, tea, mutual adjustment for caffeinated and decaffeinated coffee

Oba et al. [7], Japan

Takayama study

≥35 (52)

5,897 (278)

7,643 (175)

11

SR

M

W

M

W

M

W

Coffee

Coffee

Decaf

Decaf

Caffeine

Caffeine

0.69 (0.49–0.98)

0.70 (0.44–1.12)

1.09 (0.73–1.61)

0.66 (0.36–1.23)

0.95 (0.69–1.30)

0.95 (0.63–1.43)

Age, smoking status, BMI, PA, length of education in years, alcohol, total energy intake, fat intake and women’s menopausal status

Sartorelli et al. [8], France

E3N/EPIC

41–72 (53)

69,532 (1,415)

11

CSR

W

Coffee

Decaf

Caffeine

≥3 versus none

0.76 (0.59–0.98)

>1.1 versus 0 cups/meal

0.67 (0.47–0.95)

397 versus 48

0.67 (0.58–0.78)

Age, FHDM, PA, alcohol, educational level, hypercholesterolemia, hypertension, smoking, energy-adjusted fiber and saturated fat, total energy without alcohol, menopausal status, hormone replacement therapy, use of oral contraceptives at baseline, BMI, magnesium (for coffee)

Hjellvik et al. [9], Norway

Health surveys

(52)

362,045 (9,886)

4

NR

M and W

Coffee

>9 versus <1

0.63 (0.58–0.69)

Year of birth, gender, BMI, smoking, education, PA

Zhang et al. [10], USA

The Strong Heart Study

45–74 (55)

1,141

(188)

7.6

OGGT/FPG/

use of an oral hypoglycemic agent or insulin

M and W

Coffee

8–11 versus 1–2

0.78 (0.44–1.37)

Age, gender, smoking, alcohol, FHDM, PA, BMI

Floegel et al. [11], Germany

EPIC–Germany study

35–65 (49.7)

42,659 (1,432)

8.9

CSR

M and W

Coffee

Decaf

≥4 versus <1

0.77 (0.63–0.94)

0.70 (0.46–1.06)

Age at recruitment, center (Potsdam/Heidelberg), sex, smoking, alcohol, PA, education, employment, vitamin and mineral supplement use during past 4 weeks, total energy intake, tea intake, decaffeinated coffee (for coffee), caffeinated coffee (for decaf) and BMI, waist-to-hip ratio, and prevalent hypertension (yes or no)

Bhupathiraju et al. [12], USA

NHS

HPFS

30–55 (50)

40–75

(53)

74,749 (7,370)

39,059 (2,865)

24

22

CSR

CSR

W

M

W

M

Coffee

Decaf

>5 versus none

0.65 (0.58–0.73)

0.65 (0.49–0.85)

>3 versus none

0.73 (0.64–0.84)

0.84 (0.68–1.03)

Age, time interval, smoking status, alcohol, postmenopausal hormone use, PA, FHDM, AHEI, consumption of caffeinated tea, fruit punch, sugar-sweetened beverage, or artificially sweetened beverage, hypertension, hypercholesterolemia, adherence to a low-calorie diet, reported weight change, total energy intake and BMI

Doo et al. [13], USA

Multiethnic Cohort (Hawaii study)

45–75

(61)

(58)

36,120 (4,541)

39,020 (4,041)

14

CSR

M

W

M

W

Coffee

Decaf

≥3 versus <1

0.89 (0.80–0.99)

0.66 (0.58–0.77)

1.07 (0.93–1.23)

0.85 (0.72–1.01)

Age, ethnicity, BMI, physical activity, education, history of hypertension, energy, alcohol, smoking status, sugared sodas, dietary fiber per 4,184 kJ and processed meat per 4,184 kJ

AHEI alternate healthy eating index, BMI body mass index, CSR confirmed self-report, CVD cardiovascular disease, decaf decaffeinated, FHDM family history of diabetes mellitus, FPG fasting plasma glucose level, GGT gamma-glutamyltransferase, HT hypertension, M men, NR national register, OGTT oral glucose tolerance test, PA physical activity, SBP systolic blood pressure, SR self-report, WHR waist-to-hip ratio, W women

Quantitative synthesis

Results of pooled analysis are summarized in detail in Table 2.
Table 2

Pooled measures on the association of coffee, decaffeinated coffee and caffeine with incidence of T2DM

Study

No. (results)

No. (cases)

No. (subjects)

Risk estimate (95 % CI)

I2 (%)

Pa

FEM

REM

Coffee

Overall

31

50,595

1,096,647

0.71 (0.69–0.74)

0.71 (0.67–0.76)

54.2

 

 Sex

      

0.29

  Women

13

25,868

606,586

0.67 (0.63–0.71)

0.67 (0.60–0.74)

50.3

 

  Men

9

17,187

304,330

0.73 (0.69–0.77)

0.73 (0.65–0.81)

63.8

 

 Mean age (years)b

     

0.74

 

  ≤50

12

18,472

371,114

0.70 (0.66–0.75)

0.71 (0.64–0.79)

54.2

 

  >50

19

32,123

725,533

0.72 (0.68–0.75)

0.72 (0.65–0.79)

56.3

 

 Quality scores (stars)

     

0.75

 

  6

12

16,701

292,254

0.77 (0.72–0.82)

0.71 (0.63–0.81)

60.6

 

  7

19

33,894

804,393

0.68 (0.65–0.72)

0.71 (0.65–0.76)

39.1

 

 Location where the study was conductedc

  Europe

11

16,773

591,409

0.68 (0.64–0.72)

0.71 (0.63–0.79)

47.5

 

  United States

14

28,342

381,551

0.74 (0.70–0.78)

0.72 (0.65–0.80)

65.1

0.74

  Asia

6

5,480

123,687

0.69 (0.59–0.80)

0.69 (0.59–0.80)

0.00

0.71

 Follow-up duration (years)b

   

0.75

   

  >10

21

35,598

605,457

0.75 (0.71–0.78)

0.74 (0.68–0.80)

53.5

 

  <10

10

14,997

491,190

0.65 (0.60–0.69)

0.65 (0.59–0.73)

23.2

 

 Adjustment for total energy intake

   

0.16

   

  Yes

16

31,339

538,006

0.75 (0.71–0.79)

0.74 (0.69–0.81)

53.0

 

  No

15

19,256

558,641

0.66 (0.62–0.70)

0.67 (0.60–0.74)

41.4

 

 Adjustment for 3 behavioral factors (smoking, alcohol and physical activity)

0.72

      

  Yes

26

40,196

693,932

0.73 (0.70–0.77)

0.72 (0.67–0.77)

46.8

 

  No

5

11,945

402,715

0.66 (0.61–0.71)

0.69 (0.56–0.86)

69.3

 

 Adjustment for 3 or more dietary factors

   

0.92

   

  Yes

14

31,331

518,436

0.73 (0.69–0.77)

0.72 (0.65–0.79)

61.1

 

  No

17

19,264

578,211

0.69 (0.65–0.73)

0.71 (0.64–0.79)

46.3

 

 Adjustment for hypercholesterolemia

  

0.49

    

  Yes

9

18,269

382,811

0.69 (0.64–0.74)

0.69 (0.61–0.77)

48.9

 

  No

22

32,326

713,836

0.72 (0.69–0.76)

0.73 (0.67–0.79)

56.5

 

 Adjustment for family history of diabetes mellitus

  

0.74

    

  Yes

14

22,592

452,235

0.71 (0.67–0.76)

0.72 (0.66–0.80)

38.3

 

  No

17

28,003

644,412

0.71 (0.68–0.75)

0.70 (0.64–0.78)

64.0

 

 Adjustment for magnesium

  

0.33

    

  Yes

6

6,603

164,869

0.77 (0.69–0.86)

0.77 (0.69–0.86)

0.00

 

  No

25

43,992

931,778

0.71 (0.68–0.74)

0.70 (0.65–0.76)

59.8

 

 Body mass index (BMI) (kg/m2)

   

0.40

   

  <25

5

0.60 (0.42–0.85)

0.53 (0.32–0.86)

30.6

 

  >25

8

0.78 (0.70–0.86)

0.71 (0.59–0.85)

43.3

 

 Smoking

   

0.18

   

  No

5

0.66 (0.56–0.78)

0.49 (0.32–0.75)

59.4

 

  Yes

8

0.80 (0.70–0.91)

0.76 (0.64–0.91)

25.4

 

 Assessment of T2DMc

  SR

18

35,029

652,931

0.73 (0.69–0.76)

0.70 (0.64–0.77)

62.1

 

  NR

4

12,113

413,600

0.65 (0.61–0.71)

0.69 (0.57–0.83)

68.2

0.83

  Serum

4

1,036

6,043

0.77 (0.60–0.99)

0.77 (0.60–0.99)

0.00

0.58

  Mixed

5

2,417

24,073

0.80 (0.70–0.91)

0.80 (0.70–0.91)

0.00

0.26

Decaffeinated coffee

Overall

13

29,165

491,485

0.84 (0.78–0.90)

0.79 (0.69–0.91)

65.2

 

 Sex

      

0.04

  Women

7

19,353

354,921

0.77 (0.70–0.84)

0.75 (0.64–0.88)

53.2

 

  Men

3

7,684

81,076

1.07 (0.94–1.22)

1.07 (0.94–1.22)

0.00

 

 Mean age (years)b

     

0.44

 

  ≤50

5

14,123

258,396

0.74 (0.66–0.83)

0.73 (0.58–0.93)

61.8

 

  >50

8

15,042

233,089

0.91 (0.83–0.99)

0.84 (0.72–0.99)

56.9

 

 Quality scores (stars)

     

0.07

 

  6

7

14,430

212,124

0.95 (0.86–1.04)

0.89 (0.74–1.07)

60.0

 

  7

6

14,735

279,361

0.73 (0.66–0.81)

0.73 (0.65–0.81)

8.10

 

Caffeine

Overall

8

9,302

321,960

0.70 (0.65–0.75)

0.71 (0.64–0.80)

48.6

 

 Sex

      

0.09

  Women

5

7,151

260,396

0.67 (0.62–0.72)

0.66 (0.58–0.75)

51.2

 

  Men

3

1,842

54,558

0.84 (0.72–0.98)

0.84 (0.72–0.98)

0.00

 

 Mean age (years)b

     

0.10

 

  ≤50

2

5,348

172,535

0.66 (0.60–0.73)

0.63 (0.50–0.80)

76.2

 

  >50

6

3,954

149,425

0.75 (0.67–0.82)

0.76 (0.67–0.87)

23.4

 

 Quality scores (stars)

     

0.44

 

  6

5

2,621

107,491

0.73 (0.64–0.82)

0.76 (0.64–0.90)

31.0

 

  7

3

6,681

214,469

0.69 (0.63–0.75)

0.68 (0.57–0.82)

72.6

 

FEM fixed-effect model, REM random-effect model

aP value for meta-regression

bMean age (years) and duration of follow-up duration (years) were included as continuous variables in meta-regression

cLocation where the study was conducted (Europe as the reference group) and assessment of T2DM (SR as the reference group) were included as dummy variables in meta-regression

Coffee intake and T2DM incidence

Overall, 26 articles with 31 separate results were included involving 50,595 T2DM cases and 1,096,647 participants. The pooled RR (95 % CI) of T2DM for the highest versus lowest categories of coffee intake was 0.71 (0.67–0.76, Fig. 1), and moderate between-study heterogeneity was found (I2 = 54.2 %).
https://static-content.springer.com/image/art%3A10.1007%2Fs00394-013-0603-x/MediaObjects/394_2013_603_Fig1_HTML.gif
Fig. 1

The multivariate-adjusted risk of type 2 diabetes mellitus for the highest versus lowest categories of coffee intake. The size of gray box is positively proportional to the weight assigned to each study, which is inversely proportional to the standard error of the RR, and horizontal lines represent the 95 % confidence intervals. D + L denotes random-effect model (REM), I–V denotes fixed-effect model (FEM)

In order to explore the observed between-study heterogeneity, subgroup analysis and univariate meta-regression were conducted. Results from subgroup analysis suggested that the association of coffee intake with T2DM incidence was stronger for women [0.67 (0.63–0.74), I2 = 50.3 %] than that for men [0.73 (0.65–0.81), I2 = 63.8 %]. A stronger association of coffee intake with T2DM incidence was found for non-smokers [0.49 (0.32–0.75), I2 = 59.4 %] and subjects with BMI <25 kg/m2 [0.60 (0.42–0.85), I2 = 30.6 %]. The association of coffee intake with T2DM incidence was attenuated but still significant after adjusting for total energy intake [0.74 (0.69–0.81), I2 = 53.0 %], and an attenuated association was also found in studies with the follow-up duration of more than 10 years [0.74 (0.68–0.80), I2 = 53.5 %]. The association of coffee intake with T2DM incidence was also not influenced substantially by mean age (≤50 and >50 years), quality scores (6 and 7 stars), location, where the study was conducted (Europe, United States and Asia), and adjustment (yes or no) of 3 behavioral factors (smoking, alcohol and physical activity), 3 or more dietary factors, hypercholesterolemia, family history of diabetes mellitus and different methods in assessing T2DM cases. Univariate meta-regression analysis showed that no covariates had a significant impact on between-study heterogeneity (all P values > 0.05).

For dose–response analysis (Fig. 2), data from 20 articles [5, 6, 8, 9, 1113, 23, 25, 29, 3139, 41] including 46,722 cases were used overall. There was some evidence of a nonlinear association between coffee intake and T2DM incidence overall (P < 0.01), and the incidence of T2DM declined faster before 7 cups/day of coffee intake. The RR (95 % CI) of T2DM was 0.93 (0.92–0.94), 0.87 (0.85–0.89), 0.82 (0.80–0.84), 0.78 (0.76–0.80), 0.75 (0.72–0.77), 0.71 (0.69–0.73), 0.68 (0.66–0.71), 0.65 (0.62–0.68), 0.63 (0.60–0.66) and 0.61 (0.57–0.64) for 1, 2, 3, 4, 5, 6, 7, 8, 9 and 10 cups/day of coffee intake, respectively. The incidence of T2DM decreased by 12 % [0.88 (0.86–0.90), P < 0.00] for every 2 cups/day increment in coffee intake in random-effect dose–response model assuming linearity.
https://static-content.springer.com/image/art%3A10.1007%2Fs00394-013-0603-x/MediaObjects/394_2013_603_Fig2_HTML.gif
Fig. 2

The dose–response analysis between coffee intake and incidence of type 2 diabetes mellitus with restricted cubic splines in a multivariate random-effects dose–response model. The solid line and the long dash line represent the estimated relative risk and its 95 % confidence interval. Short dash line represents the linear relationship

For dose–response analysis in women, data from 12 articles [6, 8, 9, 12, 13, 23, 29, 3335, 37, 41] including 21,754 cases were used. There was some evidence of a nonlinear association between coffee intake and T2DM incidence (P = 0.02), and the incidence of T2DM declined faster before 6 cups/day of coffee intake. The RR (95 % CI) of T2DM was 0.92 (0.90–0.93), 0.84 (0.82–0.87), 0.79 (0.76–0.82), 0.74 (0.72–0.77), 0.70 (0.67–0.73), 0.66 (0.63–0.69), 0.62 (0.59–0.65), 0.59 (0.55–0.63), 0.56 (0.51–0.60) and 0.53 (0.48–0.58) for 1, 2, 3, 4, 5, 6, 7, 8, 9 and 10 cups/day of coffee intake, respectively. The incidence of T2DM decreased by 15 % [0.85 (0.82–0.88), P < 0.00] for every 2 cups/day increment in coffee intake in random-effect dose–response model assuming linearity.

For dose–response analysis in men, data from 7 articles [9, 12, 13, 23, 33, 34, 41] including 14,806 cases were used. Linear relationship was found (P = 0.72), and incidence of T2DM decreased by 8 % [0.92 (0.91–0.93), P < 0.00] for every 2 cups/day increment in coffee intake in random-effect dose–response model. The RR (95 % CI) of T2DM was 0.96 (0.93–0.98), 0.91 (0.88–0.95), 0.87 (0.84–0.91), 0.84 (0.81–0.88), 0.81 (0.77–0.85), 0.78 (0.74–0.82), 0.75 (0.70–0.78), 0.72 (0.68–0.76), 0.68 (0.64–0.73) and 0.66 (0.61–0.71) for 1, 2, 3, 4, 5, 6, 7, 8, 9 and 10 cups/day of coffee intake, respectively.

Decaffeinated coffee intake and T2DM incidence

Ten articles [68, 1113, 32, 35, 37, 38] with 13 separate results were included involving 29,165 T2DM cases and 491,485 participants. The pooled RR (95 % CI) of T2DM for the highest versus lowest categories of decaffeinated coffee intake was 0.79 (0.69–0.91), and moderate to high between-study heterogeneity was found (I2 = 65.2 %). Meta-regression and subgroup analysis were conducted to explore the observed between-study heterogeneity. Univariate meta-regression suggested that sex significantly (P = 0.04) contributes to the heterogeneity, and quality score might also contribute to the heterogeneity (P = 0.07). In subgroup analysis, significant association was found for women and for studies with 7 stars, but not for men and for studies with 6 stars. And the association was stronger for subjects ≤50 years than subjects >50 years.

For dose–response analysis, data from 7 articles [6, 1113, 32, 37, 38] including 23,781 cases were used overall. Linear dose–response relationship was found (P = 0.72), and incidence of T2DM decreased by 11 % [0.89 (0.82–0.98), P = 0.02] for every 2 cups/day increment in decaffeinated coffee intake in random-effect dose–response model. The RR (95 % CI) of T2DM was 0.96 (0.92–0.99), 0.90 (0.86–0.94), 0.86 (0.81–0.91) and 0.81 (0.74–0.89) for 1, 2, 3 and 4 cups/day of decaffeinated coffee intake, respectively.

For dose–response analysis in women, data from 4 articles [6, 12, 13, 37] including 14,386 cases were used. Linear relationship was found (P = 0.32), and incidence of T2DM decreased by 11 % [0.89 (0.78–1.02), P = 0.08] for every 2 cups/day increment in decaffeinated coffee intake in random-effect dose–response model. The RR (95 % CI) of T2DM was 0.96 (0.92–1.01), 0.90 (0.86–0.95), 0.83 (0.77–0.88) and 0.77 (0.70–0.85) for 1, 2, 3 and 4 cups/day of decaffeinated coffee intake, respectively. Only 2 articles provided the data for dose–response analysis in men; thus, dose–response analysis in men was not performed in this meta-analysis.

Caffeine intake and T2DM incidence

Six articles [7, 8, 30, 32, 33, 37] with 8 separate results were included involving 9,302 T2DM cases and 321,960 participants. The pooled RR (95 % CI) of T2DM for the highest versus lowest categories of caffeine intake was 0.70 (0.65–0.75), and moderate between-study heterogeneity was found (I2 = 48.6 %). Meta-regression and subgroup analysis were conducted to explore the observed between-study heterogeneity. Univariate meta-regression suggested that sex, mean age and quality score did not significantly contribute to the heterogeneity. In subgroup analysis, stronger association was found for women, subjects ≤50 years and studies with 7 stars than men, subjects >50 years and studies with 6 stars, respectively.

For dose–response analysis (Fig. 3), data from 5 articles [8, 30, 32, 33, 37] including 8,711 cases were used overall. Linear relationship was found (P = 0.08), and incidence of T2DM decreased by 14 % [0.86 (0.82–0.91), P < 0.00] for every 200 mg/day increment in caffeine intake in random-effect dose–response model. The RR (95 % CI) of T2DM was 0.96 (0.94–0.99), 0.92 (0.87–0.97), 0.86 (0.81–0.91), 0.80 (0.76–0.85), 0.73 (0.69–0.79), 0.68 (0.65–0.74) and 0.63 (0.58–0.68) for 100, 200, 300, 400, 500, 600 and 700 mg/day of caffeine intake, respectively.
https://static-content.springer.com/image/art%3A10.1007%2Fs00394-013-0603-x/MediaObjects/394_2013_603_Fig3_HTML.gif
Fig. 3

The dose–response analysis between caffeine intake and incidence of type 2 diabetes mellitus with restricted cubic splines in a multivariate random-effects dose–response model. The solid line and the long dash line represent the estimated relative risk and its 95 % confidence interval. Short dash line represents the linear relationship

For dose–response analysis in women, data from 4 articles [8, 30, 33, 37] including 7,819 cases were used. Linear relationship was found (P = 0.20), and incidence of T2DM decreased by 17 % [0.83 (0.77–0.90), P < 0.00] for every 200 mg/day increment in caffeine intake in random-effect dose–response model. The RR (95 % CI) of T2DM was 0.95 (0.92–0.98), 0.91 (0.86–0.96), 0.85 (0.79–0.90), 0.78 (0.74–0.83), 0.72 (0.67–0.77), 0.66 (0.61–0.71) and 0.61 (0.56–0.67) for 100, 200, 300, 400, 500, 600 and 700 mg/day of caffeine intake, respectively. Only 2 articles provided the data for dose–response analysis in men; thus, dose–response analysis in men was not performed in this meta-analysis.

Sensitivity analysis and publication bias

Sensitivity analysis showed that no individual study had excessive influence on the above-mentioned pooled effect. Egger test showed no evidence of significant publication bias for the analysis between T2DM incidence and coffee intake (P = 0.63, Supplementary Figure 2), decaffeinated coffee intake (P = 0.16, Supplementary Figure 3) and caffeine intake (P = 0.51, Supplementary Figure 4), respectively.

Discussion

Findings from this meta-analysis of prospective studies indicated that coffee and caffeine intake might be significantly inversely associated with T2DM incidence, and the association might be stronger for women than that for men. And stronger association of coffee intake with T2DM incidence was also found for non-smokers and subjects with BMI <25 kg/m2. The association of coffee intake with T2DM incidence was not substantially modified by adjustment of behavioral and dietary factors. Dose–response relationship was also found between coffee, decaffeinated coffee and caffeine intake and T2DM incidence.

Results from a randomized controlled trial suggested that improvements in adipocyte and liver function, indicated by changes in adiponectin and fetuin-A concentrations, may contribute to the beneficial metabolic effects of long-term caffeinated and decaffeinated coffee consumption [42]. Caffeinated coffee was found positively related to insulin sensitivity, and decaffeinated coffee was found favorably related to measures of beta cell function in a cross-sectional study [43]. Sex hormone-binding globulin may also account for the inverse association between caffeinated coffee consumption and T2DM risk among postmenopausal women [24], and a dose-dependent association was found between regular coffee consumption and increased adiponectin levels in men [44]. Regular coffee consumption appears to have beneficial effects on subclinical inflammation and HDL cholesterol in a clinical trial [45], and during a 2-h oral glucose tolerance test, decaffeinated coffee yielded lower glucose and a higher insulin sensitivity index than caffeine [46]. Intakes of caffeinated and decaffeinated coffee and caffeine were found each inversely associated with C-peptide concentration, a marker of insulin secretion in the Nurses’ Health Study [47]. In addition, the antioxidants such as chlorogenic acid, polyphenols and lignan as well as magnesium, which are rich in coffee, might also contribute to the observed protection of regular coffee and decaffeinated coffee on T2DM risk [4, 48, 49]. Furthermore, caffeine might also protect against T2DM incidence through increasing metabolic rate and thermogenesis, stimulating fat oxidation and free fatty acid release from peripheral tissues and mobilizing glycogen in muscles [4, 32, 50].

As shown in Fig. 1, although inverse association was found in all 31 results, significant association was found in 10 of the 12 results involving more than 30,000 subjects, while significant association was found only in 4 of the 19 results involving less than 30,000 subjects. And strong association was found in the three results with the most number of subjects involved: RR (95 %) = 0.63 (0.58–0.69) [9] and 0.53 (0.41–0.68) [37] and 0.65 (0.58–0.73) [12]. For caffeine, only 1 result [33] reported a significant association among the 4 results involving less than 30,000 subjects, while all of the 4 results involving more than 30,000 subjects reported a significant association. For decaffeinated coffee, 5 of the 8 results involving more than 30,000 subjects reported a significant or marginally significant association, while only 1 significant association [32] was found among the other 5 results involving less than 30,000 subjects. And strong association was also found in the three results with the most number of subjects on decaffeinated coffee and T2DM risk: 0.67 (0.47–0.95) [8], 0.73 (0.64–0.84) [12] and 0.52 (0.36 = 0.74) [37]. Among the 8 studies providing subgroup analysis by sex for coffee, the magnitude of association was stronger for women than that for men in 5 studies, while similar results were found in the other 3 studies. For caffeine, 2 studies reported a stronger magnitude of association for women than that for men, while similar result was found in 1 among the 3 studies providing subgroup analysis by sex. For decaffeinated coffee, significant or marginally significant association was found for women, while the association was not significant for men among the 3 studies providing subgroup analysis by sex.

Smoking, the established risk factor for T2DM [51], was also found positively associated with coffee consumption [8, 1012]; thus, the adverse effects of smoking may cancel out the potential benefits of coffee consumption on T2DM incidence. The association between coffee and T2DM incidence was slightly attenuated after adjusting for smoking as well as alcohol and physical activity. And stronger association was found among non-smokers both in the included original studies [6, 19, 29, 30] and this meta-analysis, which was also consistent with the experiment findings that smokers eliminate caffeine faster [52]. BMI was found an important predictor of T2DM incidence [53]; thus, the benefits of coffee consumption on T2D incidence might also be canceled out by the adverse effects of higher BMI. Subgroup analysis by BMI (<25 and >25 kg/m2) also suggested that BMI might be a modifier on the association of coffee with T2DM risk, and subgroup analysis suggested that age might also be a modifier. Coffee is rather rich in magnesium, which is also significantly inversely associated with T2DM incidence [49], and the association of coffee with T2DM incidence was slightly attenuated after adjusting for magnesium. Adjustment for total energy intake, which is positively associated with coffee consumption [5, 8, 39], also slightly attenuated the association. Coffee consumption was found to increase serum concentrations of cholesterol in a clinical trial [45], and a possible mediating effect by serum lipids on the association of coffee with T2DM incidence was indicated [29]. However, the strength of the association between coffee and T2DM incidence increased slightly after adjusting for hypercholesterolemia in this meta-analysis.

An earlier review [54] suggested that every 1 cup/day increment in coffee intake was associated with a 7 % reduction in the excess risk of (T2DM). Since the review [54] was published, nine population-based prospective studies [513] (36,592 T2DM cases and 764,782 subjects) were also included in this meta-analysis, making it possible to describe the relationship of coffee intake with T2DM incidence in subgroup, assess the potential threshold effect of coffee intake on T2DM incidence and assess whether risk factors influence the association of coffee intake with T2DM incidence. Strength of this meta-analysis is the large number of participants and T2DM cases included from prospective studies, providing high statistical power with which to quantitatively assess the relationship of coffee and caffeine intake with T2DM incidence and to conduct subgroup analyses. However, as a meta-analysis of published observational studies, several potential limitations should also be considered.

First, misclassification of coffee consumption could be of concern, and some participants might change their coffee habits during follow-up. However, results from dietary validation studies suggested that the self-reported coffee consumption with a food frequency questionnaire is highly reproducible and agrees well with assessments using diet records [55, 56], and coffee use tends to be a constant and well-reported habit [25]. Second, although adjustment for the selected key covariates did not substantially influence the association, stronger association of coffee with T2DM incidence was found among lean people (BMI < 25 kg/m2) and non-smokers. Thus, residual confounding owing to measurement error in the assessment of confounding factors or unmeasured confounding should be considered. Third, similar results were found for filtered and instant coffee in Nurses’ Health Study II [37] and for boiled and other types of coffee in Norway health survey [9], and the results were also similar for filtered and boiled coffee when analyzing the two subpopulations separately [19]. However, filtered coffee, but not instant coffee, was found significantly associated with a reduced risk of diabetes in E3N/EPIC study [8], and the limited information precluded a more robust assessment of the effect of different types of coffee on T2DM incidence. Fourth, between-study heterogeneity was found, but it was not completely explained by meta-regression and subgroup analysis; thus, other factors might also account for the observed between-study heterogeneity. In addition, misclassification of T2DM status might also lead to an underestimation of the true magnitude of the associations [39]. Finally, in a meta-analysis of published studies, publication bias could be of concern because small studies with null results tend not to be published. However, no evidence of significant publication bias was found in this meta-analysis.

In conclusion, coffee and caffeine are significantly associated reduced risk of T2DM incidence, and long-term randomized trials are needed to confirm the findings.

Conflict of interest

None.

Supplementary material

394_2013_603_MOESM1_ESM.pdf (86 kb)
Supplementary material 1 (PDF 85 kb)

Copyright information

© Springer-Verlag Berlin Heidelberg 2013