Introduction

Osteoporosis is a leading cause of disability in older persons and is associated with increased risk of fracture and subsequent disability and mortality, particularly among women [1].

High caffeine intake has been cited as a risk factor for osteoporosis. Although the mechanisms underlying this association are incompletely understood, caffeine consumption has been associated with displacement of milk from the diet [2, 3], inhibition of intestinal calcium absorption [4–7], and increased urinary calcium excretion [4–7]. Some observational studies have suggested that the impact of heavy caffeine consumption on calcium metabolism and bone mass or fracture risk may be most detrimental for individuals with very low calcium intake [8–12]. Other studies have suggested that caffeine-induced increases in urinary calcium excretion are not associated with impaired calcium absorption and/or bone density among younger women [4, 13], and a small placebo-controlled randomized crossover study demonstrated no significant changes in the calcium economy associated with 400 mg/d of caffeine administered for 19 days to 16 women aged 26–35 [14].

Although numerous epidemiological studies have attempted to evaluate the impact of caffeine intake on bone mass or related outcomes such as fracture or calcium balance, reported findings are inconclusive. The majority of studies of this association have been conducted among postmenopausal women [8–12, 15–24], a population at particularly increased risk for osteoporosis. Approximately half report caffeine to be weakly, but significantly, associated with osteoporosis [8–12, 15–17].

Few studies have examined this association among adolescent and young adult women [25–29], which may also be an important group since they are in the process of attaining peak bone mass. None has detected a statistically significant relationship after adjusting for relevant confounders, but these studies of younger subjects may have been limited in their ability to detect an association due to their cross-sectional design [25, 27], relatively small sample sizes [25, 26, 28], homogeneity in the caffeine exposure status of study population (ranging only from 4–146 mg/d, as calculated from six years of prospective diet records [26]), or misclassification of caffeine exposure status (by computing a composite score representing the number of cups of coffee, tea, or cola consumed per day, without taking into account the actual caffeine content of the beverages [29]). Furthermore, most previous investigations have focused on BMD and have not included evaluations of BMC. Evaluation of this variable is additionally informative since BMC may better reflect bone strength, and therefore, fracture risk [30, 31]. Analyses of changes in BMC are also useful among populations such as adolescent and young adult women who are experiencing appreciable changes in bone size over time [30].

The purpose of the current population-based prospective study was, therefore, to determine whether habitual caffeine intake was associated with either bone mass or changes in bone mass in a large cohort of 625 women, aged 14 to 40. The relatively large sample size, availability of longitudinal measures, detailed information on relevant covariates, and young age distribution of study subjects provided a unique opportunity to examine this relationship in an as yet understudied population. We hypothesized that heavy consumption of caffeine would be associated with lower BMC and BMD and a decreased rate of bone mass accrual (or an increased rate of bone loss) over time. We also hypothesized that this inverse association would be attenuated among individuals with adequate calcium intake.

Materials and methods

Study population

The study population consisted of 625 women, aged 14 to 40 years at baseline, who were enrollees of Group Health Cooperative, a U.S. health maintenance organization with approximately 500,000 subscribers in the states of Washington and Idaho. The study subjects were recruited to participate in either of two 3-year longitudinal studies examining the effects of use and discontinuation of depot medroxyprogesterone acetate (DMPA) injectable contraception (Depo-Provera; Pfizer Pharmaceutical Group, New York, NY), on bone mineral density. The study population has been described in detail elsewhere [32–35]. Briefly, the first study (November 1994-April 1996) enrolled 457 women aged 18 to 40 and followed participants for 36 months. The second cohort, (June 1999-December 2000) consisted of 174 women aged 14 to 18 who were followed for a minimum of 24 months and a maximum of 36 months. For each study, DMPA users and a random sample of age-similar non-users were identified using electronic databases. Women with conditions or using other medications known to affect bone mass were excluded from participation and all adolescents had attained menarche. Of 631 women enrolled in these studies, bone mass measurement and data regarding self-reported habitual caffeine intake at baseline were available for 625 (99%) study subjects. All study procedures were reviewed and approved by the Group Health Cooperative Human Subjects Committee and written informed consent was obtained.

Data collection

Study subjects were asked to visit the clinic every six months throughout the 24- to 36-month follow-up period. At baseline and each follow-up visit, participants completed questionnaires that solicited information about race/ethnicity, education, employment, marital status, mood, general health, physical activity, diet, smoking, alcohol use, menstrual history, hormonal contraceptive use, pregnancy, and family history of fracture. Height and weight were measured in the clinic and were used to compute body mass index (BMI).

Exposure assessment

Caffeine consumption was estimated for individuals who answered “Yes” to the following: “In the past 6 months, have you drunk [regular coffee containing caffeine, lattes, colas that contain caffeine, or regular tea or iced tea with caffeine] every day or almost every day?” and who reported current caffeinated beverage consumption. If the response was “No”, we assumed no caffeine consumption. Self-reported quantities of caffeinated beverage intake were then converted to mg caffeine per day. While the caffeine content of beverages varies (influenced by source and method of preparation), the following values were used to calculate daily caffeine intake based on estimates for the caffeine content of beverages derived from the scientific literature and U.S. FDA reports.: 136 mg per 8 oz. coffee, 48 mg per 8 oz. tea, 36 mg per 12 oz. caffeinated soft drink, and 40 mg per 1 oz. shot of espresso [36]. After computing the estimated caffeine intake based on cups or cans of caffeinated beverages consumed per day, as reported on the semi-annual questionnaires, we then classified women into one of three caffeine exposure categories: none (0 mg/d), low (≤200 mg/d), and high (>200 mg/d). These cutpoints roughly approximate tertiles for caffeine exposure reported by women in our cohort: at baseline, there were 190, 260, and 175 subjects in each of the exposure groups, respectively. These categories are also useful for separating women who habitually consumed at least two cups of coffee per day from those with no or lower habitual caffeine exposure.

Outcome measurement

BMC and BMD of the total hip, the lumbar spine (L1-L4), and the whole body were measured in the clinic at baseline and each follow-up visit by dual-energy x-ray absorptiometry (DEXA) using Hologic bone densitometers (Hologic, Inc., Bedford, MA). Standard techniques were used by densitometrists who were trained by the manufacturer. Spinal BMC and BMD measurements collected for the younger cohort were adjusted for presence of navel jewelry, which was common among these study subjects.

Statistical analyses

Baseline categorical characteristics of study participants were compared using Pearson’s χ 2 tests to assess the statistical significance of overall differences among caffeine intake groups, and multiple partial F-tests (using 0 mg/d as the referent group) were used to evaluate differences in continuous covariates.

Cross-sectional and repeated measures data analysis methods were used to assess associations between caffeine intake and BMC and BMD, and change in BMC and BMD. Specifically, mean baseline BMC and BMD measurements for each baseline caffeine exposure group were estimated in g and g/cm2 using crude and multivariable least-squares linear regression models. Identical methods were used to evaluate the mean percentage changes and absolute changes in BMC and BMD measurements from baseline to 6, 12, 18, and 24 months for each of the three baseline caffeine consumption groups. Marginal generalized estimating equations (GEE) were used to examine crude and adjusted mean BMC and BMD by baseline caffeine exposure group. This approach allowed us to analyze longitudinal measures while accounting for the within-subject correlation of the data. An independent working correlation matrix was utilized for the GEE analysis, since there were both time-dependent covariates (i.e., age, BMI, physical activity score, dietary calcium and alcohol intake, smoking, DMPA use, and prior pregnancy) and fixed covariates (i.e., race/ethnicity and family history of fracture at any point prior to or during follow-up) in the models, and robust standard errors were computed.

For all models, individual Wald’s tests were used to evaluate between-group differences (using 0 mg/d as the referent group) and linear trends (using a grouped linear term for the categorized caffeine consumption variable). A priori rationale and results from preliminary bivariate analyses were used to select the covariates included in multivariable models. Adjusted mean values for each caffeine exposure group were obtained after accounting for the observed distribution of relevant covariates within the study population, with the exception of DMPA use, which was over-represented in the study population as part of the original study design. Thus, DMPA parameter weights were manually inputted to reflect DMPA use among the general population: 3.3% for DMPA use in the analyses of baseline data, and 1.2% for current DMPA use and 2.1% for previous DMPA use in the analyses using longitudinal data [37].

To evaluate whether adequate calcium intake modifies the association between caffeine and bone mass, we tested the significance of interaction terms between calcium intake (dichotomized as <800 mg/d and ≥800 mg/d) and caffeine consumption category, added to our multivariable models. Additionally, to determine if caffeine exerts differential effects on bone among women of different ages, we tested the significance of interaction terms between age (<30 years and ≥30 years, the age at which peak bone mass is typically attained), and caffeine intake category in our multivariate models. Analyses restricting the study population to DMPA users or DMPA non-users were also performed to determine whether associations between caffeine intake and BMC or BMD differed by DMPA status.

The statistical analyses were performed using STATA statistical software (Version 9.2; StataCorp, College Station, Texas). Two-sided tests were used, and the level of significance was set at p = 0.05.

Results

Thirty women in the dataset completed only the baseline visit and, thus, are included in the analyses of BMC and BMD at baseline, but not in the analyses of change in bone mass during study follow-up. The number of women who completed the 6-, 12-, 18-, 24-, 30-, and 36-month follow-up visits was 529, 507, 491, 469, 427, and 374, respectively. Data from these follow-up visits were included in the repeated measures and percentage change and absolute change analyses. Effort was made to retain all women in the original older cohort (n = 457) for the full 36-month duration of the study; however, maximum available follow-up for the adolescents (n = 174) ranged from 24 to 36 months, dependent upon date of enrollment. Attempting to become or becoming pregnant, and moving outside of the study area were the most common reasons for loss to follow-up. Length of follow-up time did not differ significantly by baseline caffeine exposure status (p = 0.51).

Reported caffeine intake within subjects varied little over time. Approximately 60% of women who reported no caffeinated beverage intake at baseline reported no intake on all follow-up questionnaires. More than 70% of women with high (>200 mg/d) caffeine exposure at baseline maintained this status at follow-up. Only 27 women (4%) reporting no habitual caffeinated beverage consumption at baseline ever reported high caffeine intake during follow-up, and just 17 women (3%) reporting >200 mg/d caffeine at baseline later reported 0 mg/d caffeine.

Table 1 summarizes the distribution of demographic and other characteristics of study subjects as well as unadjusted mean site-specific BMC by baseline caffeinated beverage intake. Older women were more likely to report higher levels of habitual caffeine intake and, compared to individuals with no or very low caffeine intake, women who consumed >200 mg/d caffeine had menstruated longer, had higher educational attainment, and were more likely to report ever having been pregnant. Women regularly exposed to high levels of caffeine also consumed less dietary calcium, consumed more alcohol, and were more likely to report being a current or former smoker than those who reported no regular caffeinated beverage intake. Women reporting high caffeine intake also reported more weight-bearing physical activity, yet had significantly higher BMI compared to the referent group. Caffeine intake was not strongly associated with race/ethnicity, contraceptive use, or family history of bone fracture in this study population. Although mean hip BMD at study enrollment was significantly lower among women reporting the most caffeine intake compared to the referent group, there were no significant between-group differences after adjusting for relevant covariates (BMD data not shown). The associations between caffeine and BMC and BMD did not differ significantly by calcium intake status (<800 mg/d vs. ≥800 mg/d) or age (<30 years vs. ≥30 years) (data not shown).

Table 1 Baseline study population characteristics by self-reported daily caffeine intake
Full size table

Adjusted mean absolute change in BMC from baseline for selected follow-up intervals also was not significantly associated with caffeine intake (Fig. 1). Results were similar with respect to absolute change in BMD, as well as percentage change in BMC and BMD at all anatomical sites (data not shown).

Fig. 1
figure 1

Mean absolute change in bone mineral content (BMC) from baseline to 6, 12, 18, and 24 months by self-reported daily caffeine intake at baseline. aAdjusted for baseline BMC, race/ethnicity (indicator variables for White/Caucasian, Black/African-American, and other), family history of fracture at any point prior to or during follow-up (binary, Y/N), and time-dependent covariates: baseline age (continuous), body mass index (BMI) (continuous), physical activity (continuous), dietary calcium intake (continuous), alcohol (continuous), smoking (binary, current Y/N), depot medroxyprogesterone acetate (DMPA) use (indicator variables for Never User, Current User, or Discontinuer [used at some point during the study, but no longer using]), and prior pregnancy (binary, Y/N). DMPA parameter weights were manually inputted to reflect the prevalence of DMPA use among the general population (1.2% for current use and 2.1% for previous use). bError bars represent 95% confidence intervals for group mean estimates

Full size image

Repeated measures multivariable analyses using GEE revealed non-significant inverse associations between self-reported caffeine intake at baseline and mean anatomical site-specific BMC measured during follow-up (Fig. 2). There were no associations between caffeine and repeated BMD measurements (data not shown).

Fig. 2
figure 2

Generalized estimating equations (GEE) analysis of mean bone mineral content (BMC) at any visit, by self-reported daily caffeine intake at baseline. aAdjusted for race/ethnicity (indicator variables for: White/Caucasian, Black/African-American, and other), family history of fracture at any point prior to or during follow-up (binary, Y/N), and time-dependent covariates: current age (continuous), body mass index (BMI) (continuous), physical activity (continuous), dietary calcium intake (continuous), alcohol (continuous), smoking (binary, current Y/N), depot medroxyprogesterone acetate (DMPA) use (indicator variables for: Never User, Current User, or Discontinuer [used at some point during the study, but no longer using]), and prior pregnancy (binary, Y/N), and using GEE to account for repeated measures within subjects, independent working correlation, and robust standard errors. DMPA parameter weights were manually inputted to reflect the prevalence of DMPA use among the general population (1.2% for current use and 2.1% for previous use). bError bars represent 95% confidence intervals for group mean estimates

Full size image

Although interaction terms between DMPA use and caffeine exposure group were not significant, subgroup analyses of baseline data restricted to DMPA users revealed modest inverse associations between caffeine intake and bone mass (Table 2). Among DMPA users, women who regularly consumed >200 mg of caffeine per day had lower adjusted hip, spine and whole body BMC and BMD compared to women reporting no caffeine intake. Findings were significant for spine BMC (p < 0.01), and marginally significant for whole body BMC (p = 0.07) (BMD data not shown). When caffeine intake was modeled as a continuous variable (mg/d), inverse associations with hip (p = 0.05), spine (p = 0.07), and whole body (p = 0.05) BMC were detected among DMPA users, while inverse associations with hip, spine, and whole body BMD were non-significant (p = 0.37, 0.17, and 0.32, respectively). With respect to absolute or percentage change in bone mass, there were no significant findings associated with caffeine consumption when we restricted analyses to women reporting consistent DMPA use over a 6-month to 12-month period. Caffeine intake was also not associated with either BMC or BMD at baseline, or change in BMC or BMD over time among DMPA non-users (p > 0.5 for all models).

Table 2 Mean baseline bone mineral content (BMC) by self-reported daily caffeine intake among DMPA users and DMPA non-usersa
Full size table

In sensitivity analyses, varying the cutpoints for the caffeine exposure categories to evaluate more highly exposed women (>300 mg/d), did not measurably impact our findings. Our results did not differ when we restricted analyses to women who were classified in the same caffeine exposure category at both baseline and 24-months (n = 282), or to those with the same caffeine exposure status at every visit between baseline and 24-months (n = 138). In this investigation, associations between habitual caffeine use and bone mass were evaluated, basing caffeine exposure on self-reported amounts of caffeine consumed “every day or almost everyday”. Thus, 65 women who reported sporadic caffeine use, but did not report consumption of caffeinated beverages “every day or almost everyday”, were categorized as non-users, potentially biasing our findings toward the null. When these women were excluded from analyses, the mean site-specific bone mass estimates for the non-user group were higher; however, no significant associations between caffeine and BMC or BMD were detected.

Discussion

The current population-based study of 625 adolescent and premenopausal adult women provided an opportunity to prospectively evaluate the association between habitual caffeine intake and bone mass. Although a number of studies have reported an inverse association between caffeine and BMD among postmenopausal women [8–12, 15–17], and have also suggested that this association might be stronger in those with inadequate calcium intake [8–11], in our study of premenopausal women we saw no associations among women in general, noting only modest inverse associations between caffeine and BMC among women using DMPA injectable contraception. Based on our cross-sectional and repeated measures analyses, recent and current habitual caffeinated beverage intake, even heavy intake, does not appear to adversely impact attainment or maintenance of BMC or BMD among DMPA non-users. Calcium sufficiency and age did not appear to modify the associations. We reported only a subset of our BMD findings, as BMC is the most appropriate measure for the hypotheses we were evaluating. However, we conducted parallel analyses using BMD as the outcome measure in order to facilitate comparison to previous studies. We note that the analyses of BMD added nothing to the data presented, and, in fact, failed to detect the difference in spine bone mass among the DMPA users. [BMD results are included in Tables 1 and 2 as an on-line appendix for this issue of the journal.]

Other investigations of the association between caffeine and bone mass among teen-aged or young adult subjects report similar findings [25–29]. In a cross-sectional investigation of 677 healthy Caucasian young women aged 18–35, Rubin et al. observed a marginally significant inverse association between number of servings of caffeinated beverages consumed per day and BMD at the hip (p = 0.06), but not the lumbar spine, in unadjusted analyses; the association was not significant at either anatomical site in multivariable analyses [29]. Caffeinated cola consumption was not associated with BMD at the dominant heel and non-dominant distal radius among 1,335 boys and girls aged 12 and 15 in a multivariable cross-sectional analysis by McGartland et al. [27]. In a multivariable cross-sectional analysis among 177 Caucasian women aged 19–26, Conlisk et al. detected an inverse, but non-significant, association between average caffeine intake over the previous 12 months and hip and lumbar spine BMD. There was no evidence to suggest that the association differed for women who consumed more or less than the median 836 mg calcium intake per day [25]. In unadjusted analyses of data collected over a 6-year period among a cohort of 81 Caucasian girls enrolled at age 12, Lloyd and colleagues reported that mean daily caffeine intake from soda (<1, 1–2, or >2 12 oz. cans/day) was not significantly associated with either whole body BMC gain over the follow-up period, or with hip BMD at age 18 [26]. Similarly, in multivariable longitudinal analyses of caffeine intake and bone mass among 145 women in their early 20s, Packard and Recker also found no significant associations between caffeine and percentage change in spine BMC or BMD per decade [28]. Associations between caffeine and BMC or BMD among DMPA users have not been previously assessed.

Compared to prior studies of this association, strengths of the current study include the large sample size and wide age range, the prospective design and availability of repeated measures, the wide range of caffeine exposure in the study population (0 to 1,760 mg/d), and the collection of detailed covariate information. The prospective data collection allowed us to examine mean BMC and BMD as well as absolute and relative changes in these variables over time. The over-sampling of DMPA users in the original study design also allowed us to evaluate these associations separately among DMPA users in this cohort, and our findings suggest that caffeine may exert a greater impact on bone in this subset of young women, compared to the general population.

Our data also had limitations. Our study sample, though large, was fairly homogenous with respect to sociodemographic characteristics such as race. Additionally, FFQ estimates suggested that our study population was for the most part calcium sufficient, which may have limited our ability to detect effect modification by inadequate calcium intake. Our ability to fully specify exposure status may have been limited by the reliance on self-reported data. We were unable to verify the volume or caffeine content of the servings reported. We also did not collect data on other sources of caffeine exposure such as over-the-counter drugs, coffee-flavored foods, or chocolate. Lastly, women were asked to report quantities of caffeinated beverages consumed over the past 6 months, not lifetime use.

Our cross-sectional and longitudinal results from a large population-based cohort provide evidence that that even fairly high levels of habitual caffeine exposure do not negatively impact bone mass in young women in general, but may be modestly inversely associated with bone mass in young women who also use DMPA.