Do Vitamin D Level and Dietary Calcium Intake Modify the Association Between Loop Diuretics and Bone Health?

This study was embedded in the first three cohorts of The Rotterdam Study (RS-I, RS-II and RS-III), an ongoing, population-based cohort study in Ommoord, a suburb of Rotterdam, the Netherlands [19]. Since January 1990, participants of 55 years and over were recruited for RS-I (N = 7983). In 2000, the study was extended to 3011 participants (RS-II). Later in 2006, the study was extended with a third cohort of participants of 45 years and older (RS-III). Overall response for all three cycles at baseline was 72% (14,926 of 20,744) [19]. Participants were interviewed at home by a trained research assistant, after which they were invited for a physical examination and dietary assessment at the research centre. More details on the main objectives, design and diagram of examination cycles of the Rotterdam Study (RS) have been published elsewhere [19]. Because of a possible persisting effect on bone, users of bisphosphonates were excluded from the study. For the current analysis, 6908 participants with data available for femoral neck (FN)-BMD, 6677 participants with lumbar spine (LS)-BMD and 6476 participants with LS-TBS were included from the fourth examination of the first cohort (RS-I-4, 2002–2004), the second examination of the second cohort (RS-II-2, 2004–2005) and the first examination of the third cohort (RS-III-1, 2006-2008) (Fig. 1).

Dietary data were collected at baseline (between 1989 and 1993 in RS-I-1, between 2000 and 2001 in RS-II-1 and between 2006 and 2008 in RS-III-1) using a validated semi-quantitative Food Frequency Questionnaire (FFQ) managed by a trained dietician, at the study centre [20, 21]. For RS-I-1 and RS-II-1, a two-stage 170-items FFQ was used (during first stage, participants mentioned on 170 food item, which foods they consumed at least twice a month in the preceding year, and in the second stage, dietician identified how often and in which amounts the foods were consumed). For RS-III-1, a one-stage 389-items FFQ was used. Dietary intake of nutrients (incl. total energy and dietary calcium intake) was determined using the Dutch Food Composition Tables (NEVO) from 1993, 2001 and 2006, using standardized portion sizes [20, 22]. Intake of calcium was adjusted for energy corresponding to the residual method [23]. Serum 25(OH)D was measured in the blood collected at the same time as dietary data, between 1990 and 2008 using electrochemiluminescence immunoassay (COBAS, Roche Diagnostics GmbH). The sensitivity of the test was 10 nmol/L, the range of serum 25(OH)D concentrations was from 7.5 nmol/L to 175 nmol/L, the within-run accuracy was less than 7.8%, and the intermediate precision accuracy was less than 13.1% [24, 25]. Vitamin D deficiency was defined as a serum 25(OH)D level ≤ 50 nmol/L according to the current recommendations for older adults aged > 70 years by the Institute of Medicine and the Dutch Health Council [26]. In addition, we used also a vitamin D deficiency threshold < 20 nmol/L in our analysis and showed the stratification analysis in categories of ≤ 20 nmol/L, between 20–50 nmol/L and ≥ 50 nmol/L.

As of 1st January 1991, pharmacy records of prescriptions were collected via all pharmacies in the Ommoord region with details on product name, ATC code, number of tablets/capsules in each prescription, and prescribed daily number [27]. LD use (ATC code C03C) was determined from baseline to the date of the DXA scan [19] and defined in three different categories: current-users, past-users and never-users. Current-users were defined when the participant had a drug exposure period between the date of the DXA scan and 120 days prior to the performed DXA scan. Past-users were defined when the participant had a drug exposure period more than 120 days prior to the performed DXA scan. In addition, if the participants had no drug exposure from baseline till the DXA scan, the participants were considered as never-users. The duration of LD use among current-users was categorized into 1–120 days, 120–365 days and more than 365 days. Never-use of LD was used as the reference category.

Co-variables related to lifestyle, body composition and socioeconomic status (SES) were included. Weight (kg) and height (cm) were measured at study entry. BMI was calculated as weight divided by height squared (kg/m²). During home interviews physical activity (PA) was assessed by means of the Zutphen Physical Activity Questionnaire [28]. Metabolic equivalent of tasks were calculated (MET hours/week) according to time spent in categories of light, moderate and vigorous activity [28]. Socioeconomic status variables (i.e. educational level and income level), smoking (yes/no), pack-years, use of alcohol, prevalence of coronary heart diseases (CHD), stroke and diabetes mellitus (DM) were assessed by home interview. Also the use of bisphosphonates was determined using pharmacy dispensing records in the same way as the use of LD. Serum calcium, magnesium and sodium were measured in blood samples by the Department of Clinical Chemistry of the Erasmus Medical Center using the Roche/Hitachi cobas c501 analyzer (Roche Diagnostics, Indianapolis, IN, USA).

Femoral neck and lumbar spine BMD was measured at RS-I-4, RS-II-2 and RS-III-1 (between 2002 and 2008) by dual-energy X-ray absorptiometry (DXA) using a ProdigyTM fan-beam densitometer (GE Lunar Corp, Madison, WI, USA for all participants [19]. The DXA-derived trabecular bone score (TBS), which is measured in the Lumbar Spine (LS-TBS), is a grey-level texture measurement that correlates with 3D parameters of bone microarchitecture, connectivity density, trabecular separation and trabecular number [29]. LS-TBS predicts future fractures (all type) independent of areal BMD [30]. Moreover, recent studies have shown that LS-TBS may be an applicable measure of trabecular bone integrity to study in regard to lifestyle factors that are adaptable, such as a dietary intake [31, 32]. LS-TBS was derived from the same lumbar DXA scans that BMD was obtained from and it was analysed using TBS iNsight software (version 1.9; Medimaps, Geneva, Switzerland) at the Bone Disease Unit of the University of Lausanne (Lausanne, Switzerland). LS-TBS was calculated for a subgroup of RS-I-4, RS-II-2 and RS-III-1 and represents the mean value of the individual vertebral measurements from L1 to L4. Subjects with a BMI higher than 37 kg/m² were excluded from the study since LS-TBS measurements in morbidly obese persons are not accurate. Furthermore, the LS-TBS was standardized according to sex using the residual method, due to technical differences between LS-TBS across sexes. The method of LS-TBS calculation has been described in detail elsewhere [32].

First, for all variables normal distribution was examined by visual check of histograms. When necessary, data were log transformed. Linear regression analysis (cross-sectional) was used to assess the association between LD use (past-, current-use and the duration of 1–120 days, 120–365 days and > 365 days) and FN-BMD, LS-BMD and LS-TBS. All the analyses were adjusted for age and sex and cohort (model 1). Thereafter, the co-variables were added additionally to model 2, based on literature relevance as well as principles of causal inference combined with the change-in-effect-criterion of ≥ 10% [33, 34]. Potential confounders were BMI, smoking (pack-years), alcohol intake (g/day), SES (education and job), total physical activity (MET hours/week) and comorbidities (prevalence of CHD, stroke and DM). For the analysis of the duration (continuous), the model was additionally adjusted for the past-users. An earlier study showed that in subjects with osteopenia/normal BMD levels, TBS is significantly associated with vertebral fractures [35]. Even though BMD and TBS are correlated, they present different aspects of bone health [35]. For that reason, as sensitivity analyses, we wanted to assess how the association between LD and BMD may depend on measures of bone architecture (i.e. TBS) and how a potential association between LD and TBS is dependent on BMD. So, the association between LD and FN-BMD and LS-BMD was additionally adjusted for LS-TBS and vice versa (model 3). In addition, in all analyses, sensitivity analyses were performed with additional adjustment for serum vitamin D, season of blood collection of vitamin D [winter (September till end of February) and summer (March till end of August)], serum calcium, magnesium and sodium concentrations.

To account for missing data in co-variables (varied from 1.6 to 66.8%), we used a multiple imputation approach (n = 10 imputations, 10 iterations). Briefly, as described by Sterne et al, multiple imputation is created on the correlation between each variable and missing values with other participant characteristics [36]. Linear regression analyses were then separately accomplished in each of the 10 datasets [36]. Betas were pooled by taking the average of the effect sizes of the 10 imputed datasets. The pooled standard errors and respective 95% Confidence Intervals (CI) were then calculated by using Rubin’s rules [37]. For details of the multiple imputation, see supplement Tables S1 and S2. The statistical software package of SPSS 24.0 was used for the statistical analyses (SPSS Inc., Chicago, Illinois, USA). For all the analyses except the effect-modification analysis, p values of < 0.05 were considered statistically significant.

Baseline characteristics are presented for the total population, LD users and non-users (Table 1) and baseline characteristics before and after multiple imputation are shown in Table S2. The median age, on the DXA scan date, was 65.0 years [57.0–99.0 IQR] for the total population (n = 6990). Of the total population, 7.8% (n = 543) ever used LD from baseline till the DXA scan with a mean duration of 51.2 days (325.1 SD). The median age of the 543 LD users was 77.0 years [52.0–99.0 IQR] and 64.0 years [51.0–97.0 IQR] for the LD non-users (Table 1).

The median serum 25(OH)D level of the LD users was 42.3 nmol/L [28.8–62.3 IQR] and for LD non-users, 55.4 nmol/l [37.2–76.7 IQR]. LD users showed a significantly lower serum 25(OH)D level than non-users (p < 0.001). The mean dietary calcium intake of the LD users was 1189 mg/day (394 SD) and 1121 mg/day (394 SD) for LD non-users. Results of the linear regression analysis of LD use and indices of bone health are shown in Table 2 and discussed below.

Compared to LD never-use, current-use of LD was only associated with LS-TBS in the crude model (β = − 0.018, 95% CI − 0.034; − 0.001) and past-use of LD was associated with lower LS-TBS, in model 1 (β = − 0.031, 95% CI − 0.044; − 0.017). Adjustment for covariates in model 2 attenuated the association somewhat whereby the effect size was 30% lower. The analysis of categories of the duration of LD use among users and LS-TBS showed no significant associations (Table 2).

Current-use of LD in 210 participants was associated with significantly higher LS-BMD compared to never-use of LD in the crude model (β = 0.037, 95% CI 0.010; 0.065). Past-use of LD was associated with higher LS-BMD compared to never-use of LD in the crude model, model 2 and fully adjusted model (model 3) (β = 0.048, 95% CI 0.026; 0.071, β = 0.029, 95% CI 0.006; 0.052 and β = 0.038, 95% CI 0.016; 0.060). In the analyses of the duration of LD use, current-use of LD between 121 and 365 days showed a significantly higher LS-BMD in fully adjusted model (β = 0.075, 95% CI 0.007; 0.143 Table 2).

LD current-users showed no significant association with FN-BMD compared to never-use of LD. However, past-use of LD showed a significantly higher FN-BMD only in the crude model (β = 0.016, 95% CI 0.001; 0.031, Table 2).

Additional adjustment for serum vitamin D, season of blood collection of vitamin D, serum calcium, magnesium and sodium did not change the results (data not shown).

Furthermore, we evaluated the association between LD and LS-TBS, LS-BMD and FN-BMD by sex and found no evidence that the association between LD and LS-TBS was significantly different according to sex (P-interaction = 0.83). However, the association between LD and LS-BMD and FN-BMD was significantly different according to sex (P-interaction for both = 0.04) (Supplemental table S3).

Effect-modification analyses of serum 25(OH)D level and dietary calcium intake in the association between LD use on FN-BMD and LS-TBS are shown in Tables 3 and 4. P value for interaction terms varied from 0.04 to 0.99. Serum 25(OH)D level was an effect modifier in the association between LD use and LS-TBS (P for interaction = 0.04). After stratification in model 2, the group of serum 25(OH)D level ≥ 50 nmol/L showed a significantly lower LS-TBS compared to serum 25(OH)D ≤ 20 nmol/L and between 20 and 50 nmol/L for LD past-use (β = − 0.036, 95% CI − 0.060; − 0.013 vs. β = − 0.012, 95% CI − 0.036; 0.013 and β = − 0.031, 95% CI − 0.096; 0.034, respectively, Table 3).

There was no effect modification by serum 25(OH)D level on the association between LD use and FN-BMD and LS-BMD (Table 3). After stratification, for the analysis of the association between LD and LS-BMD, participants with serum 25(OH)D level between 20 and 50 nmol/L had a significantly higher LS-BMD compared to participants with serum 25(OH)D level ≤ 20 and ≥ 50 nmol/l for LD current-use; however, there was no significant interaction (P for interaction = 0.30) (Table 3).

No significant effect modification by dietary calcium intake was observed on the association between use of LD and FN-BMD, LS-BMD and LS-TBS (Table 4). After stratification of dietary calcium intake in categories of intake ≤ 950, 950–1200 and ≥ 1200 mg/day, no significant associations were found with FN-BMD, LS-BMD and LS-TBS (Table 4).

A strength in our study is the availability of data on dietary intake of calcium, serum 25(OH)D level and LD use, and the different indices of bone health (i.e. FN-BMD as well as LS-BMD and LS-TBS). Our study also adds data to the quite unexplored field of nutrient–drug interactions in a population of community-dwelling older persons, who are at risk of musculoskeletal diseases, malnutrition and consequences of polypharmacy. However, some limitations need to be taken into account when interpreting our results. These include the observational study design and potential residual confounding (for example, underlying disease), which prevents us from drawing conclusions regarding causality and the direction of the effect size of the association. Furthermore, as all our analyses were hypothesis-based, we did not adjust for multiple comparisons. With regard to possible type I errors, stringent interpretation of p values should be made with caution. Likewise, potential misclassification of LD use using pharmacy records may occur because data regarding actual compliance was lacking. Also, dietary calcium intake was assessed using self-reported dietary assessment, which is subject to bias and measurement error. Also, dietary intake was measured only once (at baseline) and as a result we were unable to account for differences in dietary habits over time (e.g. due to disease or medication use). Moreover, the complete data on calcium or vitamin D supplement use was not available, which could lead to an underestimation of dietary intake. In addition, vitamin D and calcium prescriptions were not analysed as exposure variable because of confounding by indication. Also, we did not have a comprehensive assessment of calcium homeostasis. For example, the free calcium concentration in plasma is strongly controlled through a complicated physiological system including the interaction of calciotropic hormones such as PTH and 1,25(OH)D and only in extreme situations will the serum calcium concentration deflect from the normal range [14]. Since we did not have the availability of serum PTH, we could not investigate potential pathways between the use of LD and calcium intake and vitamin D level. Finally, our result of inconsistent effect modification by serum 25(OH)D and dietary calcium intake on the associations between LD and the parameters of bone health could be explained by managing the parathyroid hormones (PTH) and 1.25 (OH)D that maintain calcium homeostasis, especially when serum calcium is reduced [43].

In conclusion, this study does not support the hypothesis that the association between loop diuretic use and indices of bone health is modified by serum 25(OH)D level and calcium intake. However, because of polypharmacy effects and a higher risk of malnutrition in elderly, further research and replication is warranted on nutrient–drug interaction on bone health (considering the subjects with osteopenia and osteoporosis and people with malnutrition), using other biomarkers as PTH and bone turnover markers as well as long-term loop diuretic use.