The study was conducted between May 2006 and September 2008. Participants were volunteers recruited from the community and were deemed eligible to participate if they were regular coffee consumers (≥ 2 cups/day), nonsmokers, aged 18 years or older, and overweight (body mass index (BMI) 25-35 kg/m2), but otherwise healthy according to a detailed screening that included a medical history, physical examination, electrocardiogram, and laboratory tests. Participants were excluded if they had diabetes mellitus (fasting blood glucose ≥ 7 mmol/L), heart disease, stroke, hypertension (blood pressure > 140 mmHg/90 mmHg or antihypertensive medications), alcoholism, drug abuse, abnormal hepatic function (transaminases > twice the upper limit of normal), abnormal renal function (creatinine > 97.2 μmol/L), malabsorption syndromes, gastro-esophageal reflux disease or a history of ulcer. We also excluded individuals on medications with potential interactions with caffeine or that might impact study results, and women who were breastfeeding or planning a pregnancy.
Of the 65 individuals that participated in the screening, 11 did not qualify to participate and 9 withdrew consent prior to randomization. Of the remaining 45 individuals, three participants discontinued between the baseline and 4-week visits, and one participant dropped out after the 4-week visit (See Additional file 1, figure s1). The study was approved by the institutional review boards of the Beth Israel Deaconess Medical Center and the Harvard School of Public Health and all participants provided written informed consent to participate. The clinical trial registration number is NCT00305097.
Participants were counseled by a research dietitian prior to starting the study and asked to abstain from coffee and caffeine-containing foods for the duration of the study (except for the coffee provided as part of the treatment regimen) and received a list of caffeine containing foods and beverages to avoid. After two weeks of caffeine washout and after an overnight fast (minimum 12 hours), participants attended the clinical research center for baseline assessment. Physical examination and baseline blood draw were performed followed by an oral glucose tolerance test (OGTT).
At this visit, participants were randomly assigned by a blinded statistician using a randomization schedule generated via the PROC PLAN procedure in the Statistical Analysis System 9.1 (SAS Institute Inc., NC, US). The treatment allocation was 1:1:1 (caffeinated coffee: decaffeinated coffee: no coffee) in block sizes of six with stratification by sex. A surrogate randomization list was generated and validated by the randomization statistician. Participants, investigators, and laboratory staff were blinded regarding the treatment assignment for the caffeinated and decaffeinated coffee arms.
Participants were asked to maintain stable exercise and dietary habits throughout the study and were asked to keep 3-day food diaries (1 weekend day and 2 weekdays) prior to each visit. Participants returned for repeat assessments after four weeks and 8 weeks. Participants were assessed for changes in medications or health status, for side-effects of coffee and for adherence to the intervention by questionnaires. Participants had an additional non-fasting visit at 6 weeks (typically scheduled between 12-2 pm) for a blood draw to measure serum caffeine concentrations for compliance assessment.
Participants who were assigned to one of the two coffee arms of the study were provided with five pre-weighed two-gram portions of instant coffee per day (caffeinated or decaffeinated Nestlé's Taster's Choice®). They were instructed to prepare the coffee with 6 ounces (177 mL) of boiling water and to consume coffee five times per day with every meal, and at mid-morning and mid-afternoon. Participants who reported using sugar in their coffee were provided with a non-caloric sweetener and participants had the choice of using a non-dairy creamer, which was also provided. Participants who were randomized to the control (no coffee) intervention were instructed to consume five 6-ounce glasses of water also with every meal, and at mid-morning and mid-afternoon.
Caffeine content of the coffee was measured at Tufts University School of Medicine (Boston, MA, US). Chlorogenic acid was measured at the RIKILT Institute of Food Safety (Wageningen, The Netherlands) as previously described . Trigonelline content was measured in quadruplicate at the Canterbury Health Laboratories (Christchurch, New Zealand) using methods described by Lever et al. . The caffeine concentration in the prepared regular coffee was 344 μg/mL. Chlorogenic acid and trigonelline were measured both in samples kept at room temperature during the study and samples that were stored in a freezer at the beginning of the study. Chlorogenic acid concentrations were 34 mg/g (frozen: 33 mg/g) for regular and 25 mg/g (frozen: 23 mg/g) for decaffeinated coffee and trigonelline concentrations were 8.7 mg/g (frozen: 8.7 mg/g) for regular and 7.4 mg/g (frozen: 7.1 mg/g) for decaffeinated coffee. Therefore, the 5 daily cups of caffeinated coffee provided 345 mg caffeine, 302 mg chlorogenic acid, and 78 mg trigonelline. Five cups of decaffeinated coffee provided 216 mg chlorogenic acid and 65 mg trigonelline.
At the baseline, 4-week and 8-week visits, participants wearing light clothing and no shoes had weight, height and waist circumference measured by a trained investigator. Waist circumference was measured half way between the iliac crest and the lower border of the ribcage. Body composition was measured using Tanita (model Quantum II, Lean Body software, RJL Systems, Clinton Township, MI, US) single frequency bioelectrical impedance analysis. Blood pressure was measured under fasting conditions using a validated oscillometric device with appropriately sized cuffs (Dinamap Pro100, Critikon, Tampa, FL). At each assessment, 3 blood pressure readings, taken in intervals of five minutes, were averaged.
An intravenous cannula was placed in the antecubital fossa and blood samples were collected after fasting for at least 12 hours. Subsequently, participants consumed 75 grams of glucose (7.5 ounces of glucola) and blood samples were taken at 30, 60 and 120 minutes. Glucose was measured by glucose hexokinase method and lipids using an autoanalyzer in the central clinical laboratory at the Beth Israel Deaconess Medical Center. All other samples were stored in a liquid nitrogen freezer at < -130°C until assayed in duplicate with all samples for the same participant included in the same batch. Double-antibody radioimmunoassay (Immulite® Chemiluminescence, Siemens Co., New York, NY, US) was used to measure insulin [intra-assay coefficient of variation (CV) 2.1%], C-reactive protein (CRP) (CV 3.6%), and IL-6 (CV 4.7%). Adiponectin (CV 3.9%) and fetuin-A (CV 2.4%) levels were measured by enzyme linked immunosorbent assay kits (Millipore Corporation, Billerica, MA, US and Alpco, Salem, NH, US, respectively).
At the 6-week visit, a non-fasting blood sample was taken for measurement of serum caffeine levels. Plasma concentrations of caffeine and metabolites (i.e., theobromine, paraxanthine, theophylline) were determined by high-performance liquid chromatography at the Tufts University School of Medicine with within- and between-assay CVs of 9% or lower.
We compared the differences in change in biomarkers from baseline to the 8-week visit between the treatment groups using an analysis of covariance model. The model included the change from baseline as the dependent variable with treatment group as a main effect and baseline values of the dependent variable as an additional covariate. Weight change from baseline was also included as a covariate. Differences between treatment groups were based on linear contrasts for (1) caffeinated versus no coffee, (2) decaffeinated versus no coffee, and (3) caffeinated versus decaffeinated coffee. The adjusted mean with standard error was reported by treatment, and 95% confidence intervals (CIs) were computed. Statistical significance was evaluated at an alpha level of 0.05.
Glucose and insulin area under the curve (AUC) values during the OGTT were calculated using the trapezoidal rule . The composite insulin sensitivity index (CISI) was calculated using the formula: [10,000/the square root of (fasting glucose × fasting insulin) × (mean glucose × mean insulin)] . The homeostasis model assessment for insulin resistance (HOMA-IR) was calculated as [(fasting glucose × fasting insulin)/22.5] . The ratio between the 30 minute increment in insulin to 30 minute increment in glucose concentration (insulinogenic index) was used as a measure of early insulin secretion. Variables with obvious departures from normality were log-transformed, then converted back to their original scale to yield geometric means and associated 95% CIs. Normality was achieved for these variables after log-transformation. For log-transformed variables, mean differences between groups yield a ratio given the principles of logged numbers and differences for log-transformed variables are therefore presented as percentages.
Statistical analysis was performed using the last-observation-carried-forward method for imputation of missing values. One observation was carried forward from the Week 4 visit, except for fetuin-A which was not measured at week four. The final Week 8 analysis was based on 41 participants as three participants discontinued after baseline and one participant had a missing value for change from baseline in weight. For glucose homeostasis measurements, data were based on 40 participants because one participant additionally had a missing value for fasting glucose. Outliers for the outcome variables were identified using a criterion of ± three times the interquartile range. We performed sensitivity analyses to test differences between the treatment groups in changes in risk factors after outliers were removed. Because results remained essentially the same as for the main analysis and statistically significant differences remained significant, we only present results without removal of outliers. Spearman correlation coefficients between adiponectin, IL-6, and fetuin-A and markers of glucose metabolism were calculated at baseline, as well as correlating changes over the 8-week intervention period. All statistical analyses were performed using SAS.