Introduction

Cadmium (Cd) is a naturally occurring element in the earth’s crust and a widespread contaminant in soil from natural sources, cadmium-containing fertilisers, or deposition from industrial emissions. It accumulates in growing plants and is present in e.g. rice, cereals, potatoes and other food items consumed by humans [1]. Smoking is the second most common way to accumulate Cd in the body. Today, occupational exposure (e.g. at production of batteries, soldering, or cadmium plating) is rare.

Cadmium enters the body mainly through the lungs (in smokers) or via the gastrointestinal tract. It is accumulated in the kidneys and is eliminated at a very slow rate, with a half time of decades. Cadmium levels in blood and urine are used for monitoring exposure and body burden. The most well-known adverse effect is kidney damage, but cadmium has also been associated with detrimental impact on bone structure and an increased risk of osteoporosis and fracture [2–4]. A dose-response relation is still uncertain, and there are also studies observing no skeletal effects by increasing levels of cadmium [4]. Several previous studies in Sweden observed an increased risk of osteoporosis and/or fracture in individuals exposed to Cd environmentally [5–8]. One study from the southern part of Sweden found decreased bone mineral density (BMD) even by low levels of cadmium [9]. Cadmium is also a human carcinogen (International Agency for Research on Cancer (IARC) group 1) and has also been associated with an increased risk of cardiovascular disease [10].

The Malmö Diet and Cancer Study (MDCS) was primarily initiated to study the association between diet and risk of cancer. In the cardiovascular sub-cohort, several studies were performed on associations between Cd and atherosclerosis and cardiovascular disease [10–12]. In the present study, this sub-cohort was used to investigate the relationship between levels of Cd in blood at baseline and the risk of the first incident fracture during follow-up.

Subjects and methods

Study population

The MDCS was initiated in 1991 and was previously described in detail [13]. In short, all women and men born 1923–1945 and residing in the city of Malmö were invited to participate in the study. In addition, a random selection of subjects (n = 6103) was invited in 1992–1994 to participate in the cardiovascular cohort (MDC-CV) to further include ultrasonography of the carotid artery and fasting blood sampling [14]. In the present analysis of the MDC-CV cohort, all women were included (n = 2920) that had results from the baseline questionnaire, physical examination, and blood cadmium (B-Cd) levels. Follow-up regarding vital status was available for participants until the 31st of December 2013 and includes vital status at endpoint, i.e. alive, dead, or emigrated.

All participants provided written informed consent and the study was approved by the Lund University Ethics Committee (MDC LU 51-90) and for the cadmium sub-study (2009/633). The current study was also approved by the steering committee of the MDC study (2015-011). Data retrieval from the separate outpatient data registry covering the years 1992–2001 were granted by the Deputy Chief Health Officer (Region Skåne; decision number 236-16).

Blood cadmium assessments

Cadmium levels were analysed in erythrocytes and whole blood concentration was calculated using haematocrit (erythrocyte-Cd × haematocrit/100). An inductively coupled plasma mass spectrometry (Agilent 7700× ICP-MS, Agilent Technologies, Santa Clara, CA, USA) was used for cadmium erythrocyte concentrations [12]. Samples were analysed in three different rounds including two external quality control (QC) samples. The results were similar for all three rounds [12]. The limit of detection (LOD) was 0.02 μg/L and no sample was below that limit. Blood Cd concentrations were categorised into quartiles (Cd-quartile 1 <0.18 μg/L, Cd-quartile 2 0.18–0.28 μg/L, Cd-quartile 3 0.28–0.51 μg/L, and Cd-quartile 4 >0.51 μg/L).

Measurement of covariates

C-reactive protein (CRP) was analysed by a high-sensitive method from frozen plasma, collected at the baseline examination, with a Tina-quant CRP latex high-sensitivity assay (Roche Diagnostics, Basel, Switzerland) [15]. HbA1c, insulin, and whole blood glucose were analysed with standard procedures at the Department of Clinical Chemistry, Skåne University Hospital, Malmö [16]. HbA1c was measured by ion exchange chromatography and insulin was measured by radioimmunoassay in milli-international units per litre [16].

Smoking was self-reported and categorised as “regular smoker”, “occasional smoker”, “previous smoker”, or “never smoker”. In these analyses, regular smokers and occasional smokers were grouped as “current smokers”. Alcohol intake was self-reported and grouped according to no alcohol intake during the last 30 days or more, current intake of <15 g/day, and current intake of >15 g/day of alcohol.

Fracture outcome data

Data for fracture and fracture code, according to the International Classification of Disease (ICD) 9 and 10, for first incident fractures during follow-up were collected using every individual’s unique personal identification number up until the 31st of December 2013 from the National Patient Registry (Patientregistret) kept by the National Board of Health and Welfare (Socialstyrelsen). Only fractures occurring after individual inclusion in the study were included in the analyses. Fractures requiring hospitalisation are included for the entire nation since 1987, and fractures treated as outpatient in policlinic care are included since 2001. Outpatient data regarding fractures occurring between 1992 and 2001 were retrieved from the regional diagnosis registry.

Typically, osteoporotic fractures of the hip, distal radius, forearm, pelvis, distal femur, and humerus were also grouped into non-vertebral osteoporotic fracture for further analysis [17] excluding vertebral fractures as not all vertebral fractures are diagnosed [18]. In order not to underestimate non-vertebral osteoporotic fractures, all first non-vertebral osteoporotic fractures were included even if occurring after a non-osteoporotic first incident fracture.

The following ICD 9 codes were included in the search algorithm: 805, 805A, 805C, 805E, 805W, 807, 807A, 808, 808W, 810, 811, 812A, 812C, 812E, 813, 813C, 813E, 814, 815, 816, 820, 820A, 820C, 821, 821A, 821C, 822, 823, 823A, 823C, 824, 825, 825A, 826, and 827.

The following ICD 10 codes were included in the search algorithm: M485, M843, S2200, S2220, S2230, S2240, S2290, S320, S3210, S3230, S3240, S3250, S3280, S420, S4210, S4220, S4230, S4240, S4270, S5200, S5210, S530, S5250, S5260, S5261, S5280, S6200, S6210, S6220, S6230, S6250, S6260, S6280, S7200, S7210, S7220, S7230, S8200, S8210, S8220, S8230, S8240, S8250, S8260, S8280, S8281, S8290, S9200, S9210, S9220, S9230, S9240, and S9250.

Statistical analyses

The distribution of continuous data was controlled by the Kolmogorov-Smirnov test. All parametric data are presented as means (standard deviation (SD)) and non-parametric data as medians (interquartile range (25th–75th quartile)). The Students’ t test, the Mann-Whitney U test, and the independent Kruskal-Wallis test were used for comparing groups. Categorical data were analysed using the chi-square test. Cox proportional hazards regression was used to calculate hazard ratios (HR) and 95% confidence intervals (CI).

Cox regression analyses were performed as univariate crude and multivariate adjusted for BMI, age, smoking status (never/previous/current smoker), self-reported diabetes mellitus, and gastric ulcer (verified by X-ray or gastroscopy).

All analyses were carried out using the IBM SPSS version 22.0 (IBM Corporation) apart from the competing risk analysis that was performed using Stata Statistical Software (STATA) ver. 12.1. (StataCorp). A p value of <0.05 was considered statistically significant.

Results

In total, 998 women sustained a first incident fracture during follow-up. The number of non-vertebral osteoporotic fractures were 659 occurring either as a first incident non-vertebral osteoporotic fracture (n = 590) or subsequent to a previous non-osteoporotic first incident fracture (n = 69). The most common fractures were distal radius fracture (n = 294 (29.5%)), hip fracture (n = 199 (19.9%)), and upper humerus fracture (n = 89 (8.9%)) occurring either as a first incident fracture or as a non-vertebral osteoporotic fracture. Time to first fracture was 11.4 (6.0–16.5) years for all women with a fracture, and no difference was noticed for women in Cd-Q4 compared to Cd-Q1 11.2 (6.9–16.8) vs. 11.6 (6.2–16.3) years (p = 0.49). The number of any first incident fractures were not different for women in Cd-Q4 compared to Cd-Q1 243 (33.3%) vs. 248 (34.0%); p = 0.78, but more women in Cd-Q4 had died at the end of follow-up compared to Cd-Q1 153 (21.0%) vs. 82 (11.2%); p < 0.001). Median total follow-up time in the study from baseline inclusion to censored endpoint was 20.2 years (12.5–21.2 years).

Women in Cd-Q4 had a lower weight and BMI and a slightly lower age at menopause than women in Cd-Q1 (Table 1). Fewer women in Cd-Q4 were married or cohabiting, and more of them were current smokers and high consumers of alcohol (daily intake >15 g/day) (Table 1). For current smokers, the number of smoked cigarettes/day (n = 686) correlated with B-Cd: Spearman’s rank correlation coefficient r = 0.49 (p < 0.001). The median level of B-Cd (25th and 75th percentile) according to smoking status was 0.21 (0.16–0.29) μg/L for never smokers, 0.25 (0.18–0.35) μ/L for previous smokers, and 0.92 (0.56–1.41) μg/L for current smokers (p < 0.001).

Table 1 Descriptive data at baseline for all women and for women in quartiles 1–4
Full size table

The level of CRP was not different when comparing women in Cd-Q4 to Cd-Q1, but women in Cd-Q4 had higher CRP level than women in the combined group of Cd-Q1–3 1.5 (0.7–3.4) mg/L vs. 1.3 (0.7–2.6) mg/L (p = 0.002). Women in Cd-Q4 had higher level of HbA1c (%) than women in Cd-Q1 5.0 (4.6–5.3) vs. 4.8 (4.5–5.1) (p < 0.001), but no difference was observed for fasting blood glucose (mmol/L) 4.8 (4.5–5.2) vs. 4.8 (4.5–5.2) (p = 0.92) nor for plasma insulin levels 6.0 (4.0–9.0) vs. 6.0 (4.0–8.0) (p = 0.41).

No significant differences were found in history of previous chronic diseases or treatment of stroke, hypertension, goitre, cancer, inflammatory bowel disease, rheumatoid arthritis, asthma/chronic bronchitis, or kidney stones comparing women with fracture or non-vertebral osteoporotic fracture to those without fracture. More women with non-vertebral osteoporotic fractures had self-reported diabetes (4.0%) compared to women without fracture (2.2%) (p = 0.017). More women with non-vertebral osteoporotic fracture (10.9%) (p = 0.028) and any fracture (9.6%) (p = 0.044) had been treated for gastric ulcer (found by X-ray or gastroscopy) compared to women without fracture (7.2%). More women with non-vertebral osteoporotic fracture (3.3%) (p = 0.003) and any fracture (2.4%) (p = 0.038) had been treated for myocardial infarction compared to women without fracture (1.2%). Self-reported diabetes and treatment of gastric ulcer were further evaluated using univariate Cox regression analysis (Table 2). Treatment for myocardial infarction was not included in the multivariate analysis as previous studies of this cohort have demonstrated an increase in cardiovascular disease with higher B-Cd levels [10].

Table 2 Uni- and multivariate Cox regression analysis for fractures and death
Full size table

In the univariate crude Cox regression, no significant association with fracture risk was observed with increasing levels of B-Cd but increasing age, self-reported diabetes, and gastric ulcer (Table 2). In the multivariate-adjusted Cox regression, independent predictors for increased fracture risk were gastric ulcer and age whereas increasing BMI decreased fracture risk (Table 2). The risk of first non-vertebral osteoporotic fractures was similarly not affected by increasing levels of B-Cd at baseline: HR 1.02 (95% CI 0.82–1.27) when comparing Cd-Q4 vs. Cd-Q1. Analysing never and ever smokers separately regarding any fracture did not affect the results: HR 1.09 (95% CI 0.68–1.73) for never smokers and HR 1.08 (95% CI 0.83–1.39) for ever smokers comparing Cd-Q4 vs. Cd-Q1.

In total, 403 women died during follow-up. More women in Q4 died compared to Q1, 21.0 vs. 11.2% (p = <0.001). In the multivariate-adjusted Cox regression model, increasing age and BMI, current smoking, and diabetes were independent factors for increased mortality (Table 2). When analysing never and ever smokers separately regarding overall mortality, a HR of 1.10 (95% CI 0.50–2.42) for never smokers and a HR 2.23 (95% CI 1.44–3.44) for ever smokers were observed comparing Cd-Q4 vs. Cd-Q1.

Some women n = 95 (3.3%) had had prevalent fractures, i.e. fractures before inclusion in the study. There was no difference in the number of women with prevalent fracture in Cd-Q1 vs. Cd-Q4 (n = 28 vs. n = 24; p = 0.57). Cox regression analyses for risk of fracture Cd-Q4 vs. Cd-Q1 during follow-up excluding women with prevalent fractures were similar: HR 1.07 (95% CI 0.90–1.29) compared to analysis for all women. Hence, analyses were performed including women having had a fracture before inclusion in the study.

Use of hormone replacement therapy (HRT) occurred at baseline in 500 women out of 2536 women responding to the question. No difference in HRT users was observed for Cd-Q1 (18.9%) vs. Cd-Q4 (16.2%); (p = 0.08). The use of HRT decreased fracture risk during follow-up in the univariate analyses, HR 0.70 (95% CI 0.58–0.84), as well as mortality, HR 0.75 (95% CI 0.56–0.99). When added to the multivariate analysis, it did not affect the risk of B-Cd on fracture risk for Cd-Q4 compared to Cd-Q1, HR 1.04 (95% CI 0.79–1.37), but remained significant itself, HR 0.75 (95% CI 0.61–0.92). Regarding mortality in the multivariate analysis, the use of HRT was not associated to decreased risk of mortality during follow-up, HR 0.89 (95% CI 0.64–1.22), but B-Cd in Cd-Q4 compared to Cd-Q1 was no longer significantly associated to mortality, HR 1.43 (0.93–2.21).

Kaplan-Meier curves for fracture (Fig. 1a) and all-cause mortality (Fig. 1b) are shown (Fig. 1). Since high Cd was associated with increased mortality, which potentially could reduce the relationship between Cd and fractures at older age, a competing risk analysis, with mortality from any cause as the competing event, was performed. The competing risk analysis did not observe an increased fracture risk for B-Cd as a univariate variable (Cd-Q1 as HR 1.0) sub-distribution hazard ratio (SHR) for Cd-Q4 0.97 (0.81–1.16).

Fig. 1
figure 1

a Kaplan-Meier graph for Q1–4 and cumulative percentage of women without fracture and time to fracture endpoint (ticks are censored cases). b Kaplan-Meier graph for Q1–4 and cumulative percentage of women alive and time to total endpoint (ticks are censored cases)

Full size image

Discussion

This study found no increased fracture risk associated with increasing levels of B-Cd but increased risk of overall mortality. Most studies on cadmium exposure, BMD, and fracture risk considered the relationship between urinary cadmium (U-Cd) and not blood cadmium as analysed in this study. The concentration of B-Cd reflects mostly recent exposure but also long-term exposure whereas U-Cd is a measure of total body burden of Cd reflecting the long-term exposure [1]. The follow-up time for fracture in this study was long, approximately 20 years, and it is possible that B-Cd at baseline did not reflect long-term cadmium exposure.

Few previous studies have investigated B-Cd levels and fracture risk. In a case-control study, Sommar et al. observed no increased risk of hip fracture risk associated with increasing erythrocyte-CD adjusting for smoking, BMI, and height [19]. However, in a separate analysis for women, there was an increased risk of fracture (adjusted for smoking, BMI, height, and hormone replacement therapy) with OR 3.33 (95% CI 1.29–8.56) [19]. Alfvén et al. found an inverse association between BMD and B-Cd levels in participants over 60 years [5]. Subjects with high cadmium levels had three times increased risk of low BMD (OR 2.9 (95% CI 1.4–5.8)) compared to those with low cadmium levels [5].

Several studies found an association between increasing Cd levels and decreasing BMD and increased fracture risk [4]. A recent meta-analysis revealed an increased pooled relative risk of any fracture with increasing cadmium levels, relative risk 1.30 (95% CI 1.13–1.49), but also concluded that the result should be interpreted cautiously due to heterogeneity of studies and risk of publication bias [20].

The impact of cadmium on bone metabolism has been debated and three main theories exist: [1] inhibited renal activation of vitamin D, [2] decreased absorption of calcium in the intestines due to competing action of cadmium, and [3] direct effects of Cd on collagen metabolism affecting bone structure [1]. There is also a more recent study indicating a direct effect of cadmium on bone cells [21].

In a review, Åkesson et al. found support for the hypothesis that cadmium is associated with decreased BMD and an increased fracture risk would occur at U-Cd as low as 0.5–2 μg/g creatinine which is equivalent to B-Cd levels of approximately 0.5–2 μg/L [4]. The median level of B-Cd in all women in this study was 0.28 (0.18–0.51) μg/L and in the highest quartile (Cd-Q4) the level was 1.00 (0.69–1.49) μg/L; hence, women in Cd-Q4 are well in the B-Cd range where cadmium could be associated with BMD and fracture risk. BMD was not measured in this study, but in another study based on the MDC cohort, BMD was measured and an inverse correlation between BMD and B-Cd was found. However, the results did not remain significant when adjusted for smoking [22].

This study investigates first incident fractures and in order to avoid underestimation of osteoporotic fracture risk, the risk of the first non-vertebral osteoporotic fracture has also been investigated if occurring after a non-osteoporotic first fracture. Engström et al. did not observe an increased overall first fracture risk in all women HR 1.16 (95% CI 0.89–1.50) when comparing higher levels of U-Cd with lower levels of U-Cd, but when including only never smokers, the risk was increased HR 2.03 (95% CI 1.33–3.09) [7]. In the present study, no difference was observed comparing never smokers with ever smokers in regard to fracture risk associated with increased B-Cd levels.

Increased mortality was observed in women with high B-Cd levels at baseline even after adjustment for smoking which is in agreement with previous reports [23, 24]. Hence, if more women in Cd-Q4 died during follow-up, this could result in fewer fractures in Cd-Q4, but the survival analysis using Cox regression considered this, since those who died became censored. The competing risk analysis also verifies the results. A recent review of all-cause mortality found that levels of U-Cd was associated with increased mortality HR 1.44 (95% CI 1.25–1.64) when comparing the highest category with the lowest [25]. A study based on B-Cd found a HR 1.50 (95% CI 1.07–1.20) [23] for mortality with increasing B-Cd levels. An association between cadmium exposure and cardiovascular mortality has been investigated in this cohort previously where increasing levels of cadmium was associated with development of atherosclerotic plaques in the carotid artery [12] and cardiovascular disease [10].

Diabetes has been studied previously in regard to fracture risk and both type 1 diabetes mellitus [26] and type 2 diabetes mellitus [27] are associated with increased fracture risk. Our results agree with this, as women with diabetes showed an increased risk of fracture. In the present study, women with a history of gastric ulcer had an increased fracture risk HR 1.33 (95% CI 1.02–1.75) in the multivariate analysis. These findings strengthen our recent results in another study of postmenopausal women aged 50–64 years where we found an increased risk of fracture in women using proton pump inhibitors (PPI) with OR 2.53 (95% CI 1.28–4.99) [28]. In line with the vast majority of observational studies on HRT, statistically significant reductions by HRT were found on overall mortality and fracture incidence [29, 30].

The results of this study are substantiated by a long follow-up period and fracture data from national registries. The relatively large number of included women is also a strength of the study. Limitations include the fact that smoking habits and history of diseases other than fractures are self-reported based on questionnaires. The study has investigated the association between B-Cd and fracture in women only, and hence, the results are limited to women only. Even though women in Cd-Q4 consumed more alcohol daily, there was no observed association with increased fracture risk in this group and in this aspect, it would have been interesting to study the number of falls in this group of women. The number of falls is a well-known risk factor for fracture. Unfortunately, there is no data regarding falls available.

In conclusion, high blood cadmium levels did not increase fracture risk, but doubled the overall mortality rate in middle-aged women. Gastric ulcer disease was associated with increased fracture risk after adjustment for other risk factors in this group of women.