To assess the effect of weight change on the relationship between coffee and tea consumption and diabetes risk.
Prospective cohort study, using data from the First National Health and Nutrition Examination Survey Epidemiologic Follow Up Study. Survival analyses were conducted using 301 selfreported cases of diabetes and eight documented diabetes deaths during an 8.4-y follow-up.
A total of 7006 subjects aged 32–88 y with no reported history of diabetes were included in the study.
For all subjects combined, increases in consumption of ground-caffeinated coffee and caffeine at baseline were followed by decreases in diabetes risk during follow-up. There were significant statistical interactions between age and consumption of caffeine (P=0.02) and ground-caffeinated coffee (P=0.03). Age-stratified analysis showed that the decrease in diabetes risk only applied to ≤60-y-old subjects, for whom the decrease in diabetes risk also obtained for ground-decaffeinated coffee and regular tea. The multivariate hazard ratio (HR) and 95% confidence interval for a 2 cups/day increment in the intake of ground-caffeinated coffee, ground-decaffeinated coffee and regular tea was 0.86 (0.75–0.99), 0.58 (0.34–0.99) and 0.77 (0.59–1.00), respectively. The diabetes risk was negatively related to the consumption in a dose–response manner. There were strong statistical interactions between prior weight change and beverage consumption for ≤60-y-old subjects. Further analysis revealed that the decrease in diabetes risk only applied to those who had lost weight, and that there was a positive dose–response relationship between diabetes risk and weight change. For example, the multivariate HR and 95% confidence interval for >0 vs 0 cups/day of ground-decaffeinated coffee was 0.17 (0.04–0.74), 0.52 (0.19–1.42), 0.77 (0.30–1.96) and 0.91 (0.39–2.14) for subgroups with weight change of ≤0, 0–10, 10–20 and >20 lbs, respectively. There was no significant association between diabetes risk and consumption of instant-caffeinated coffee, instant-decaffeinated coffee or herbal tea. Caffeine intake appeared to explain some, but not all, of the diabetes-risk reduction and weight change.
The negative relationship between diabetes risk and consumption of ground coffee and regular tea, observed for all NHEFS subjects, actually only applied to nonelderly adults who had previously lost weight.
Coffee and tea consumption are widespread in the United States,1, 2 with more than 50% of the population using these beverages. The prevalence of type 2 diabetes mellitus (diabetes) is growing rapidly, and has been estimated to afflict about 12% of Americans.3 In light of recent findings that coffee and tea consumption are negatively associated with the risk of diabetes, there is promise that these beverages, or some of their constituents, may prove useful in the development of techniques for the prevention or treatment of diabetes.
Some,4, 5, 6, 7, 8 but not all9, 10 previous studies have found a protective effect from coffee or tea. However, the causal mechanism has not been identified. Confirmation of their observations and elucidation of the causal mechanisms are needed before their findings can be put to therapeutic use.
Several of the studies4, 6, 7 that found coffee or tea consumption to have a protective effect also found that higher levels of consumption were positively associated with body mass index (BMI), a measure of adiposity. However, one study found a negative association.5 It appears unlikely, therefore, that coffee or tea consumption reduces diabetes risk via an association with BMI, long recognized as an important risk factor for diabetes.
Caffeine11, 12, 13, 14 and tea11 have been shown to increase energy expenditure in humans, and weight loss has reduced risk factors for diabetes in clinical trials.15 Consequently, it seems possible that both coffee and tea consumption may decrease diabetes risk by helping individuals control their body weight.
The purpose of the present study was therefore to analyze the effect of caffeine and body-weight change on the relationship between consumption of coffee and tea and diabetes risk, using the National Health and Nutrition Examination Survey (NHANES-1) Epidemiologic Follow Up Study (NHEFS). This database contains data well suited for this purpose, including consumption levels of a wide variety of beverages, and sequential body-weight measurements.
NHEFS is a longitudinal follow-up study of the persons aged 25–74 y (N=14 407) examined in NHANES I, a probability-sample survey of the civilian United-States noninstitutionalized population. NHANES-1 was conducted in 1971–1975,16 and NHEFS contains four follow-up surveys, in 1982–1984, 1986, 1987, and 1992. The baseline data used in the present study were obtained by means of medical histories and examinations at the NHANES-1 survey, and at the first follow-up survey in 1982–1984.17
The 1982–1984 survey was used as a follow-up baseline in all survival analyses. Subjects with any missing data and those with a reported history of diabetes in 1982–1984 were excluded from the analyses. After exclusions, there were 7006 subjects between 32 and 87 y of age in the 1982–1984 survey, and 309 diabetes events that occurred during the subsequent follow-up, which lasted an average of 8.4 y for censored subjects who did not develop or die from diabetes during the follow-up. The 309 diabetes events were made up of 301 cases and eight deaths. Diabetes incidence data were based on responses to two questions in the 1986, 1987, and 1992 follow-up surveys. The first asked whether a doctor said the subject had diabetes, and the second asked which year the subjects were first told about diabetes. The time, in years, between this year and the year of the 1982–1984 interview was estimated as the time to event for the subjects who reported being told they had diabetes during the follow-up period. Diabetes mortality data were obtained from tracing activities conducted in the 1986, 1987 and 1992 follow-up surveys. Each death was confirmed either by death certificate or proxy interview. Death certificates were obtained for >95% of deceased subjects.18 For the subjects who died of diabetes during follow-up, time to event was calculated as time between the day of death and the day of the 1982–1984 interview. Subjects were considered to have a history of diabetes if in the 1982–1984 survey they responded in the affirmative to a question asking whether a doctor had said the subject had diabetes.
Body weight was measured in the 1971–1975 and 1982–1984 surveys. Weight change was defined as the measured weight in 1982–1984 minus the measured weight in 1971–1975. BMI in 1982–1984 was calculated as body weight measured in 1982–1984, in kilograms, divided by the square of height measured in 1971–1975, in meters.
Beverage and chocolate snack intake was determined from two food-frequency questions asked in 1982–1984. The first was the number of cups of the beverage or snacks of chocolate usually consumed, and the second was the time period in which they were consumed—day, week, month or year. Responses to these questions were used to derive the number of cups of ground-caffeinated coffee, ground-decaffeinated coffee, regular tea, instant-caffeinated coffee, instant-decaffeinated coffee, and herbal tea, and the number of chocolate snacks, consumed per day. Frequency of cola-soda consumption was based on two questions that were similar to those for coffee and tea, and a third question as to whether the cola soda was consumed seasonally. In calculating the average daily cola-soda consumption, a season was assumed to last 6 months. A variable representing an American style diet was constructed from similar food-frequency questions. This diet, which is based on the work of Van Dam et al,19 was assigned a numeric value equal to the total number of servings per week of TV dinners, fast food, meat and cheese dishes, cheese and white sauces, french fries, hot dogs, ice cream, beef, cheese, sausages, sweets and deserts, pork, bacon, white bread and whole milk.
In order to estimate daily caffeine intake, we derived estimates of the caffeine content of servings of beverages and chocolate snacks. We used data in the US Department of Agriculture Handbook20 and a literature review21 covering the time period of the NHEFS. Our estimates were: ground-caffeinated coffee—159 mg; instant-caffeinated coffee—83 mg; cola sodas—42 mg; regular tea—36 mg; and chocolate snacks—6 mg. These caffeine contents were then multiplied by the daily number of servings of each beverage and of chocolate snacks, and then totaled for each subject.
Cox's proportional-hazards regression22 was used to assess the multivariate hazard ratios (HRs) for diabetes risk due to increments in intake of each beverage and caffeine. Cox's regression was used in order to incorporate the effect of time to event in the analysis. Cumulative hazard and Log-minus-log plots provided no evidence against proportionality assumptions.23 Bivariate logistic regression was used to assess the relationship between beverage intake in 1982–1984 and the likelihood of being in the upper quintile (vs the lower four quintiles) of weight loss, and to adjust the relationship for the study's covariates.24 The −2 Log Likelihood (−2LL) test was used to compare the fit of alternative regression models.
In the single-predictor survival-analysis models, we found that the pattern of associations between beverage consumption and diabetes risk was essentially the same whether or not HRs were adjusted for covariates. Therefore, only the adjusted HRs were presented here. The same was found to be true of the bivariate logistic regression analyses of the associations between beverage consumption and weight gain.
Many individuals drank more than one beverage (Table 1). Consequently, in order to separate the independent effects of each beverage, survival analyses were conducted using models containing all beverages with significant single-predictor associations with diabetes risk. The same procedure was used in the logistic-regression analyses.
The following covariates were used in all survival and logistic-regression analyses: gender; race; educational level; per-capita income; smoking (four categories); alcohol consumption (four categories); physical activity (five categories); and age (10 5-year categories); American style diet (quintiles), and BMI in kg/m2 (six categories).
There were strong interactions between age and the three beverages for which there were significant HRs for all NHEFS subjects (Table 2)—ground-caffeinated coffee (P=0.03), caffeine (P=0.02) and cola sodas (P=0.09). The age of 60 y was selected as the age cutpoint because it was found to effectively illustrate the age effects. Table 1 shows that for both ≤60 and >60-y-old subjects, drinkers of ground-caffeinated coffee were younger, better educated, had lower BMI, and were less likely to have a history of diabetes and to drink ground-decaffeinated coffee or instant coffee than nondrinkers. They were also more likely to be male, to be drinkers of alcohol and cola sodas, and consumers of an American style diet. Only the >60-y-old subjects tended to be smokers and nondrinkers of herbal tea. And only the ≤60-y-old subjects had higher income and were more likely to drink herbal and regular tea, and were less likely to have gained weight between 1971–1975 and 1982–1984.
Age and caffeine intake
The single-predictor survival analyses showed that after adjusting for all covariates, increasing consumption of caffeine and ground-caffeinated coffee were followed by a significant decrease in diabetes risk for all subjects, and for those ≤60 y of age (Table 2). For instance, the HR and 95% confidence interval (95% CI) for a 300 mg/day increment in caffeine consumption was 0.86 (0.77–0.96) and 0.85 (0.74–0.97), respectively, for all subjects and those ≤60 y of age. In addition, in the ≤60 y age group, increasing ground-decaffeinated coffee and regular-tea consumption was followed by a significant decrease in diabetes risk, with the HR (95% CI) for a two cups/day increment in consumption being 0.58 (0.34–0.99) and 0.77 (0.59–1.00), respectively.
None of the beverages were associated with a decreased diabetes risk for subjects >60 y of age. On the contrary, increasing cola-soda consumption was followed by an increase in diabetes risk for subjects >60 y of age, and for all subjects, with an HR (95% CI) of 1.38 (1.03–1.84) and 1.21 (1.01–1.45), respectively. Similarly, increasing ground-decaffeinated coffee consumption was followed by an increase in diabetes risk in the >60 y age group.
In the >60 y-old group, both ground-decaffeinated coffee and cola sodas continued to be significantly positively associated with diabetes risk when both were simultaneously put in the model that adjusted for all covariates. The HR (95% CI, P-value) for a two cups/day increment in consumption was 1.46 (1.10–1.93, P=0.009) and 1.81 (1.23–2.66, P=0.003), for ground-decaffeinated coffee and cola sodas, respectively.
For ≤60-y-old subjects, categorical versions of ground-caffeinated and ground-decaffeinated coffee and regular tea were used in an attempt to define the consumption levels associated with diabetes reduction (Table 3). The model containing these three beverages and all covariates showed that each of the three beverages was significantly and independently associated with a reduction in diabetes risk, in a dose–response manner. For instance, the HRs (95% CI) for ground-caffeinated coffee consumption of 0, <2, 2–4 and ≥4 cups/day were 1.00 (referent), 0.82 (0.55–1.23), 0.75 (0.50–1.13) and 0.37 (0.22–0.64), with P for trend=0.007.
When we included caffeine in the model, the negative association with diabetes risk persisted for ground-decaffeinated coffee and regular tea (Table 3). However, the significant association for ground-caffeinated coffee disappeared, showing that the association between this beverage and diabetes risk could be explained by the effects of caffeine, after the effects of the other two beverages were considered. The −2LL tests confirmed this finding in that the model containing caffeine and the three beverages was not significantly better than the model containing the three beverages (−2LL=3.06, P=0.548); conversely, the model containing caffeine and the three beverages was significantly better than the model containing caffeine (−2LL=17.49, P=0.025).
Weight change and beverage consumption
Among ≤60-y-old subjects, the logistic-regression analyses showed that as single predictors, ground-caffeinated coffee (P=0.10) and ground-decaffeinated coffee (P=0.02) were the only beverages strongly associated with weight change, after adjustment for all covariates. The associations were negative, that is, as beverage consumption increased, the likelihood of gaining weight decreased. When these beverages were simultaneously entered in the multivariate model, the negative associations persisted. For ground-caffeinated coffee and ground-decaffeinated coffee, when compared with subjects who drank <2 cups/day, those who drank ≥2 exhibited an HR (95% CI, P-value) for being in the highest quintile of weight change of 0.83 (0.70–0.99, P=0.035) and 0.63 (0.45–0.88, P=0.007), respectively. When caffeine was added to the model, the HR for ground-caffeinated coffee and caffeine were not significant, but the HR for ground-decaffeinated coffee was essentially unchanged, at 0.64 (0.46–0.90, P=0.009).
For >60-y-old subjects, ground-decaffeinated coffee was the only beverage that was significantly and positively associated with weight change after adjustment for all covariates. Compared with subjects who drank <2 cups/day, those who drank ≥2 exhibited an adjusted HR (95% CI, P-value) for being in the highest quintile of weight change of 1.83 (1.21–2.78, P=0.004).
Weight change, beverage consumption and diabetes risk for ≤60-y-old subjects
Tests of statistical interaction were conducted for ≤60-y-old subjects in order to further investigate the role of weight change in the negative relationship between beverage consumption and diabetes risk. There were strong interactions between weight change and consumption of caffeine (0.04), ground-caffeinated (P=0.08), ground-decaffeinated coffee (P=0.08), and regular tea (P=0.09). Table 4 shows that compared to subjects who had gained weight, those who had lost weight were older, more likely to be female patients, smoke cigarettes and consume alcohol regularly. They were more physically active, less likely to eat an American style diet, and they weighed less. They were also more likely to have a history of diabetes, and they consumed more caffeine.
The ≤60-y-old subjects were then stratified into four different levels of weight change, each containing roughly the same number of diabetes events. The negative relationship between beverage consumption and diabetes risk for all ≤60-y-old subjects were found to be significant only in the subgroup containing subjects with a weight loss (weight change ≤0 lbs), and which had an average weight loss of −10.1 lb (Table 5). The same was true for caffeine. The HR (95% CI) for all subjects, and subgroups with weight change >20, 10–20, 0–10 and ≤0 lbs was 0.71 (0.52–0.99), 1.35 (0.74–2.47), 0.49 (0.24–1.02), 0.95 (0.45–2.02) and 0.39 (0.19–0.80), respectively.
There was a positive, dose–response relationship between diabetes risk and weight change. Adding subjects with increasing levels of weight gain to those with only weight loss (≤0 lbs) led to increasing HR levels. This finding was true for ground-caffeinated coffee, ground-decaffeinated coffee and regular tea in the multivariate regression model containing all three of these beverages. For instance, the HR (95% CI) for >0 vs ≤0 cups/day of ground-decaffeinated coffee was 0.17 (0.04–0.74), 0.38 (0.17–0.84), 0.54(0.32–0.89) and 0.56 (0.34–0.91) in subgroups with an average (range) weight change of −10.1 (≤0), −1.7 (≤10), 7.2 (≤40), and 8.6 (≤80) lbs, respectively. The same was found for caffeine: the HR (95% CI) for ≥300 vs <300 mg/day of caffeine was 0.39 (0.19–0.80), 0.60 (0.37–0.98), 0.63 (0.45–0.89) and 0.71 (0.51–0.98), respectively, in each of these weight-change subgroups, in the multivariate model containing only caffeine.
Separate analyses were conducted for ≤60 and >60-y-old subjects for our significant results concerning beverage intake and diabetes risk in Table 2, for: nonobese (BMI≤30) and obese (BMI>30) subjects; male and female subjects; sedentary (physical activity ≤4) and active (physical activity >4); nonsmokers and current smokers; subjects above and below the median score for an American style diet. This was also performed for the significant results concerning beverage intake and weight change reported in the section entitled ‘Weight change and beverage consumption.’ Unfortunately the dose–response relationships in Table 3, and the results in Table 5 for different weight-change levels could not be tested, due to small numbers of events. In the analyses that were conducted, the 95% CIs were wider and not all HRs were significant, but the patterns exhibited by the HRs were consonant with those reported above. For instance, for all ≤60 y-old women, the HR (95% CI) for diabetes risk for a two cups/day increment was 0.72 (0.58–0.89) for ground-caffeinated coffee, 0.35 (0.14–0.86) for ground-decaffeinated coffee, and 0.61 (0.39–0.95) for regular tea—a pattern of HRs that is not too different from the pattern for all ≤60-y-old subjects in Table 2.
To test whether our reported significant survival-analysis relationships between beverage intake and diabetes risk for ≤60 subjects in Table 3 were an artefact caused by subjects with unduly high risk for diabetes at follow-up baseline, the survival analyses were repeated after excluding the first 3 y of follow-up. This was found to widen 95% CIs and reduce levels of significance, but not to change the pattern of results. For instance, for ≤60-y-old subjects, after excluding events in the first 3 y of follow-up, the HRs (P for trend) in the model without caffeine in Table 3 became 1.00, 0.64, 0.73, 0.29 (0.012) for ground-caffeinated coffee, 1.00. 0.79, 0.44 (0.050) for ground-decaffeinated coffee, and 1.00, 0.75, 0.76 and 0.21 (0.013) for regular tea.
Analyses with significant results for ≤60 and >60-y-old subjects in Table 2 were repeated to assess whether they may have been affected by the addition of milk or cream to coffee or tea. This was performed by including this factor as a covariate in the survival analyses, in the form of a dichotomous variable, based on a food-frequency question in the 1982–1984 survey. It had no appreciable effect.
The main finding in the present analysis is that the negative relationship between beverage consumption and diabetes risk, observed for all NHEFS subjects, actually only applied to subjects ≤60 y of age who had previously lost weight. This is a new and potentially important addition to the already published finding that coffee consumption is negatively associated with diabetes risk.4, 5, 6, 7, 8 It is based on the following evidence: first, there was a significant statistical interaction between age and beverage consumption; and only subjects ≤60 y of age exhibited a significant negative relationship between beverage consumption and diabetes risk. The negative relationship was found for ground-caffeinated coffee, ground-decaffeinated coffee, regular tea and caffeine itself. The negative relationship was a dose–response relationship, a finding that has been reported by several previous investigators for coffee.4, 5, 7 Second, for subjects ≤60 y of age, there was a statistical interaction between previous weight change and beverage consumption; and only those subjects with prior weight loss exhibited a significant negative relationship between beverage consumption and diabetes risk. Also, diabetes risk and weight change were positively related, and in a dose–response manner. Third, consumption of ground-caffeinated and ground-decaffeinated coffee were found to be negatively associated with weight gain for subjects ≤60 y of age. Fourth, for >60-y-old subjects there were no negative associations between beverage consumption and diabetes risk and no negative associations between beverage consumption and weight gain. Finally, the associations between beverage consumption and diabetes risk and weight gain for ≤60-y-old subjects were found to be true for: both males and females; nonobese and obese subjects; sedentary and active subjects; smokers and nonsmokers; and whether or not subjects ate an American style diet.
Our finding that for ≤60-y-old subjects, the risk of diabetes was significantly, negatively and independently associated with the consumption of ground-caffeinated coffee, ground-decaffeinated coffee and regular tea offers clues to the underlying causes of the reduction in diabetes risk. It suggests that each of these beverages contains constituents that are involved in reducing diabetes risk, and that are not contained, or not active, in the other two beverages. Caffeine appears to have been involved in the reduction of diabetes risk, possibly by inducing weight loss, because: (1) caffeine accounted for the significance of the negative association between diabetes risk and ground-caffeinated coffee, after the effects of ground-decaffeinated coffee and regular tea had been considered; (2) caffeine also accounted for the significance of the negative association between weight gain and ground-caffeinated coffee, after the effects of ground-decaffeinated coffee had been considered; and (3) the negative association between caffeine intake and diabetes risk was only significant for subjects with prior weight loss. These findings are consistent with the finding in several previously-published studies that caffeine induces thermogenesis11, 12, 13 and stimulates lipid oxidation11, 13, 14 in humans. Caffeine might thereby help individuals control their weight, and hence decrease diabetes-risk.
We also found evidence that factors other than caffeine may have played a role in decreasing diabetes risk and controlling weight. First, caffeine was not involved in the independent negative association between ground-decaffeinated coffee and diabetes risk for ≤60-y-old subjects; nor in the finding that this association was only significant for subjects with weight loss. Second, caffeine was not involved in the independent negative association between ground-decaffeinated coffee and weight gain for ≤60-y-old subjects; Third, instant-caffeinated coffee was not negatively associated with diabetes risk and weight gain for ≤60-y-old subjects. These results support the possibility that noncaffeine constituents in ground coffee are involved in the reduction of diabetes risk and weight control. There is evidence, for example, that one of these constituents, chlorogenic acid, can attenuate glucose absorption in the digestive tract (Johnston et al25) which could help control weight and/or preserve normal glucose tolerance. It is also possible that the lack of any effect for instant-caffeinated coffee is due to the removal of such constituents in the process of manufacturing instant-caffeinated coffee. In addition, it might be that instant-caffeinated coffee contains noncaffeine constituents that inhibit the reduction of both diabetes risk and weight control.
We found regular tea intake to be independently and negatively associated with diabetes risk among ≤60-y-old subjects, in a dose–response manner; and this association did not appear to be due to caffeine, because it remained significant after the effects of caffeine were taken into account. We also found that the negative association between regular tea intake and diabetes risk only applied to ≤60 y subjects who had experienced weight loss. On the other hand, we found that tea consumption was not negatively associated with weight gain. The most likely explanation for this apparent contradiction seems to be that tea may have weak weight-control effects, and also contains constituents that directly decrease diabetes risk. This explanation is consonant with the finding by Dulloo et al,11 that green tea promotes thermogenesis and fat oxidation in excess of that attributable to its caffeine content.
We found no significant negative association between any beverage or caffeine and diabetes risk or weight gain for >60-y-old subjects. This result is not inconsistent with our finding that weight loss is involved in reducing diabetes risk for ≤60-y-old subjects. First, body weight has been found to increase with age for persons younger than 55 y, and to decrease with age for persons older than 55 y.26 Therefore weight loss, or the attenuation of weight gain, may not be as important to diabetes risk reduction among elderly as among nonelderly persons. In support of this idea, the rate of obesity (BMI>30 kg/m2) in 1982–1984 among NHEFS subjects who were not obese in 1971–1975, was 10.3% and 5.2% for ≤60 and >60-y-old subjects, respectively. Second, there is evidence that caffeine ingestion increases energy expenditure more for nonelderly than elderly persons.12 This is consistent with our finding that caffeine and ground-caffeinated coffee were negatively associated with weight gain for ≤60 but not >60 y-old NHEFS subjects.
For >60-y-old subjects cola sodas and ground-decaffeinated coffee were independently and positively associated with diabetes risk. The effect of cola sodas could be due to the high glycemic load, or the effect of its high fructose corn-syrup content on blood lipids. Ground-decaffeinated coffee was positively associated with weight gain as well as diabetes risk for >60-y-old subjects. This latter finding is consistent with the idea that coffee's effect on diabetes is mediated by weight change. This finding suggests that the constituents in ground-decaffeinated coffee that help control weight and reduce diabetes risk in non-elderly persons, are not effective in elderly persons.
This study has several limitations. First, it is possible that subjects with weight loss, and hence lower diabetes risk, were also, coincidentally, prone to drinking ground coffee and regular tea, and that the prior weight loss caused the negative association between these beverages and diabetes risk. However, given that previous studies have yielded evidence supporting the weight-reduction potential of both regular tea 11 and caffeine,11, 12, 13, 14 it seems more likely that the beverages themselves promoted the weight loss and hence the reduction in diabetes risk. Second, beverage consumption may be associated with some health behaviors or biological factors not represented by any of the covariates in our survival analyses, and which caused the observed reduction in diabetes risk. This may be particularly true of dietary behavior such as meal patterns and food choices. Third, our analysis was not designed to assess which biological aspects of glucose-metabolism may be involved in reducing diabetes risk. Two recent studies 27, 28 have found that long-term coffee consumption was associated with enhanced insulin sensitivity. Their findings suggest that coffee consumption may act by preserving insulin-dependent glucose clearance, or by protecting pancreatic beta-cell function. Fourth, our assessment of associations between beverage intake and weight change were based on cross-sectional 1982–1984 data. This type of analysis cannot account for the fact that subjects who died before 1982–1984 and after 1971–1975, when body weight was initially measured, may have had different levels of weight change than those who survived to 1982–1984. Fifth, our beverage consumption data were selfreported, and no validation study has been conducted. However, Salazar-Martinez et al,5 who found relationships between beverage intake and diabetes risk similar to ours, did conduct a validation study, and found good reliability for selfreported coffee and tea intake. Similarly, our estimates of time to event for diabetes incidence were crude for two reasons. They were based on selfreported occurrence of diabetes, and the date of occurrence was only reported to the nearest year.
There are also several strong points in the present study. The NHEFS cohort is a sample of the US population, so that the present results have wide applicability if confirmed. In addition, the NHEFS data contain a wider range of covariates than used in most previous analyses,4, 6, 7, 8, 9, 10 as well as measured body weights suitable for deriving accurate measures of weight change. Also, the NHEFS contains data on a wide variety of beverages, allowing us to make finer distinctions between different types of coffee and tea than have been made previously. For example, we were able to distinguish the effects of regular tea from those of herbal tea; and the effects of ground coffee from those of instant coffee.
In conclusion, the negative relationship between diabetes risk and consumption of ground coffee and regular tea consumption, observed for all NHEFS subjects, actually only applied to nonelderly adults who had previously lost weight.
We thank Kenneth Axen, Department of Health and Nutrition Sciences at Brooklyn College of the City University of New York, for his helpful insightful comments during the preparation of the manuscript. The original source of the NHEFS data is the National Center for Health Statistics (NCHS). The Inter-University Consortium on Political and Social Research (ICPSR) provided the data. Neither NCHSR nor ICPSR are responsible for this report, which is the work of the author, who appreciates being able to obtain and work with the data.
About this article
European Journal of Nutrition (2016)