Conjoint Effects of Serum Calcium and Phosphate on Risk of Total, Cardiovascular, and Noncardiovascular Mortality in the Community
Objective— Hyperphosphatemia is a cardiovascular risk factor in patients with chronic kidney disease. Relations of circulating calcium (Ca) and phosphorus (Pi) to long-term mortality risk in the community require further investigation.
Methods and Results— Associations of serum Ca and Pi to mortality were evaluated in a community-based cohort of 2176 men (mean age, 50.1 years). During follow-up (median, 29.8 years), 1009 men died, and 466 of these deaths resulted from cardiovascular causes. In Cox proportional hazards models, serum Pi and [Ca×Pi] were independent predictors of total mortality (hazard ratio per SD, 1.06; 95% CI, 1.01–1.12; P=0.03; 1.07; 95% CI, 1.01–1.12; P=0.01) and cardiovascular mortality (1.10; 95% CI, 1.02–1.18; P=0.01; 1.10; 95% CI, 1.03–1.19; P=0.008). Serum Ca was associated with risk of total mortality (1.08; 95% CI, 1.01–1.16; P=0.02) and noncardiovascular mortality (1.10; 95% CI, 1.01–1.21; P=0.04). Results were consistent after multivariate adjustments in subsamples of individuals with estimated glomerular filtration rate >90 mL/min and low-to-normal serum Ca and Pi.
Conclusion— Circulating Ca and Pi levels are associated with risks of total, cardiovascular, and noncardiovascular mortality in the community, and their conjoint effects are additive. Additional studies are warranted to evaluate whether Ca and Pi are modifiable risk factors in the general population.
Maintenance of normal serum calcium (Ca) and inorganic phosphorous (Pi) levels is a prerequisite for multiple physiological processes, including bone formation, vascular function, several metabolic pathways, and intracellular signaling. Accordingly, abnormalities in Ca and Pi homeostasis have been implicated as direct or indirect culprits in a variety of skeletal, endocrine, and cardiovascular disorders.1–3 In the community, abnormalities in Ca and Pi metabolism are most commonly found in individuals with impaired renal function. Major consequences of hyperphosphatemia and an elevated Ca×Pi product ([Ca×Pi]) in patients with chronic kidney disease (CKD) are vascular calcification and increased risk of cardiovascular morbidity and mortality.4,5 Recent studies also support that a higher serum Pi levels, even within the normal range, are associated with abnormal vascular phenotypes such as increased carotid intima-media thickness6 and arterial stiffness.7
Data on the prospective associations of circulating Ca, Pi, and [Ca×Pi] to mortality in the community are scarce. In a recent community-based study, serum Pi, but not Ca, levels were related to risk of cardiovascular events.8 In contrast, another study showed that higher serum Ca predicted myocardial infarction in middle-aged men.9 It is largely unexplored whether the risks associated with higher Ca and Pi are sustained over longer time periods. More importantly, given the diverse functions of Ca and Pi in human physiology, it remains to be investigated whether circulating Ca and Pi levels in healthy individuals are related to noncardiovascular mortality. A recent study by Onufrak et al10 identified an association between serum Pi level and risk of all-cause mortality.
Herein, we investigated the separate and conjoint relations of serum Ca and Pi to total, cardiovascular, and noncardiovascular mortality in a large, prospective, community-based cohort of middle-aged men, with 2 prespecified subgroup analyses: (1) individuals with estimated glomerular filtration rate (eGFR) >90 mL/min/1.73 m2; and (2) individuals with normal or low serum Ca and Pi levels.
Subjects and Methods
The Uppsala Longitudinal Study of Adult Men (www.pubcare.uu.se/ULSAM) is an ongoing population-based study aiming to identify risk factors for cardiovascular disease. Between 1970 and 1973, all men born between 1920 and 1924 and residents in the municipality of Uppsala, Sweden, were invited to participate in a health survey. In total, 2322 of the invited men participated (82% of the targeted population). For the present study, we excluded men lacking data on creatinine, Ca or Pi measurements (n=139), or who had a Cockcroft Gault eGFRCG ≤60 mL/min/1.73m2 (n=7). This left 2176 men for the main analyses. We also investigated a subsample of 1777 men with eGFRCG >90 mL/min/1.73m2, and 1 subsample of 2155 men with serum calcium <2.6 mmol/L and serum Pi <1.45 mmol/L (the defined normal upper limits for healthy individuals). All men gave informed consent and the Ethics Committee at Uppsala University approved the study.
At age 50, the participants completed a questionnaire about their smoking habits and medical history and underwent a physical examination. The investigations have been described previously.11 Systolic and diastolic blood pressures were measured using a mercury sphygmomanometer to the nearest 5 mm Hg in the supine position after 10 minutes of rest. All blood samples were drawn in the morning after an overnight fast. Blood glucose was measured by spectrophotometry using the glucose oxidase method. Diabetes was defined as the use of insulin or oral hypoglycemic agents or fasting plasma glucose ≥7 mmol/L. Determinations of serum cholesterol and triglycerides were performed using enzymatic techniques on a Technicon Auto Analyzer type II. Serum albumin was measured using spectrophotometry with bromine cresol green with a coefficient of variation of 5.2% at the 3.2 g/100 mL level. Serum creatinine was measured using Jaffe reaction in spectrophotometry using an autoanalyzer from Technicon, with a coefficient of variation of 6.1% at the 87 μmol/L level. The eGFRCG as measure of renal function was calculated using the Cockcroft Gault equation: eGFRCG= [(140−age)×weight]/72×(serum creatinine). The choice of the Cockcroft Gault formula was motivated in the studied white population because the formula has been shown to be superior to the Modification of Diet in Renal Disease (MDRD) study equation in younger persons with preserved kidney function.12
Serum Ca was measured by flame photometry, before October 25, 1971, using an instrument from Eppendorf; thereafter, an IL 343 (Instrumentation Laboratory) was used, with a coefficient of variation of 1.3% at the 2.3 mmol/L level. To exclude the possibility that the change of instruments influenced our results, we analyzed mean Ca levels monthly at the time of shifting assays and measured the mean Ca level before and after implementation of the new measurement technique. Importantly, we found no evidence indicating alterations in Ca levels attributable to change of instruments. Serum Pi was measured by spectrophotometry using a complexometric method with ammoniummolybden on an LKB 2071 photometer (LKB), with a CV of 2.9% at the 2.2 mmol/L level. The [Ca×Pi] variable is the calculated product of serum Ca and Pi.
The men were followed through December 31, 2002. The end-point cardiovascular death (ICD-8 and ICD-9, codes 390 to 459; ICD-10 codes I00-I99) was established using the Swedish national cause-of-death register. Median follow-up time was 29.8 years (range, 0.04 to 32.2), rendering a total of 56 534 person-years at risk.
Initially, the distributional properties of all baseline variables were examined. We investigated relations of the independent variables Ca, Pi, and [Ca×Pi] to total, cardiovascular, and noncardiovascular mortality in separate models using multivariable-adjusted Cox proportional hazards analyses. Linear relations were investigated using continuous independent variables (per SD for ease of comparison between independent variables); nonlinear relations were investigated using tertiles of independent variables, and using multivariable regression spline models with up to 4 degrees of freedom allowed for the main independent variables Ca, Pi, and [Ca×Pi]. Because spline models are heavily influenced by extreme values, the spline models were performed in the subsample with serum Ca and Pi in the low to normal range. The same 2 sets of hierarchical models were fitted for the end points total, cardiovascular, and noncardiovascular mortality. Model A adjusted for albumin, eGFRCG, and age. Model B adjusted for albumin, eGFRCG, diabetes, use of antihypertensive medication, systolic and diastolic blood pressures, total cholesterol, triglycerides, age, body mass index, and smoking. Proportional hazard assumptions were confirmed by inspecting Schoenfeld residuals, and linearity assumptions were confirmed by inspecting Martingale residuals. The relative excess risk attributable to interaction between Ca and Pi was investigated13 using the extreme tertiles of both.
To examine the clinical utility of Ca and Pi measurements for prediction of subsequent mortality, we used the method described by Pencina et al.14 The integrated discrimination improvement (IDI) calculated is the mean of increments and decrements in the estimated probabilities of subsequent mortality for cases and noncases, respectively, comparing a fully adjusted model with the Ca, Pi, or [Ca×Pi] variable (our model B) to a model with the same covariates but without the Ca, Pi, or [Ca×Pi] variable. The corresponding probability value for test of the null hypothesis of no discrimination improvement for the larger model is also presented. The IDI measure has been demonstrated to be similar to the difference between C-statistics for the compared models.14 The IDI is a relatively new measure, and it is currently recommended as a key measure to report when researching new risk factors.15 Stata 10.1 (StataCorp) was used for all analyses.
Baseline characteristics of the study sample are presented in Table 1. During follow-up, 1009 men died; 466 of these men died of cardiovascular disease and 543 died of noncardiovascular causes. Of those who died of noncardiovascular causes, 362 died of neoplasms, 46 died of injuries or other external causes, 43 died of respiratory diseases, 18 died of psychiatric or nervous system diseases, 17 died of endocrine disorders, 12 died of liver diseases, 10 died of ventricular ulcers, and 35 died of other causes. The incidence rate of total mortality was 17.8/1000 person-years at risk (95% CI, 16.8–19.0), incidence rate of cardiovascular mortality was 8.2/1000 (95% CI, 7.5–9.0), and incidence rate of noncardiovascular mortality was 9.6/1000 (95% CI, 8.8–10.4).
Relations of Ca to Total, Cardiovascular, and Noncardiovascular Mortality
A 1-SD-higher serum Ca was associated with an 8% to 12% higher risk of total mortality in models A (adjusting for albumin, eGFRCG, and age) and B (adjusting for albumin, eGFRCG, diabetes, use of antihypertensive medication, systolic and diastolic blood pressures, total cholesterol, triglycerides, age, body mass index, and smoking; Table 2). A 1-SD-higher serum Ca was also associated with an 11% higher risk of cardiovascular mortality in model A, but an 8% statistically nonsignificant higher risk in model B (Table 3). In addition, 1-SD-higher Ca was associated with a 10% to 12% higher risk of noncardiovascular mortality in models A and B (Table I). The association of Ca to mortality was somewhat nonlinear, with the most marked risk increase in the highest tertile (Tables 2 and 3⇓; Table I); this was also apparent in a spline model (Figure 1).
When investigating the discriminatory value of serum Ca for subsequent mortality, it was found that adding a continuous Ca variable to the covariates in model B improved discrimination significantly for subsequent total mortality (IDI, 0.002; P=0.05) and improved discrimination borderline significantly for noncardiovascular mortality (IDI, 0.002; P=0.06); however, this was not so for cardiovascular mortality (IDI, 0.0004; P=0.41). Relations of serum Ca to total and cardiovascular mortality in the subsample with eGFRCG >90 mL/min/1.73m2 and the subsample with normal or low serum Pi and Ca were very similar to the relations in the total sample (Tables 2 and 3⇑). In the same subsamples, associations between serum Ca and noncardiovascular mortality were nearly identical to the total sample, except in model B in the subsample with normal renal function (Table 3).
Relations of Pi to Total, Cardiovascular, and Noncardiovascular Mortality
A 1-SD-higher serum Pi was associated with a 6% to 7% higher risk of total mortality and a 7% to 10% higher risk for cardiovascular mortality in models A and B (Tables 2 and 3⇑). A 1-SD-higher serum Pi was also associated with a 7% higher risk of noncardiovascular mortality in model A but a 2% statistically nonsignificant higher risk in model B (Table I). In tertile models, the associations appeared essentially linear (Table 2), as also was apparent in a spline model (Figure 2). The addition of a continuous Pi variable to the covariates in model B improved discrimination of subsequent cardiovascular mortality significantly (IDI, 0.002; P=0.05) but did not improve discrimination significantly for total mortality (IDI, 0.001; P=0.11) or noncardiovascular mortality (IDI, 0.000; P=0.99).
Relations of serum Pi to total and cardiovascular mortality were nearly identical in the subsample with eGFRCG >90 mL/min/1.73m2 (Tables 2 and 3⇑) and the total sample. In the subsample with normal or low serum Pi and Ca, some of these relations were attenuated and rendered statistically nonsignificant (Tables 2 and 3⇑). Associations of Pi to noncardiovascular mortality were nonsignificant in subsample analysis, except in model A in the subsample with low to normal serum Pi and Ca (Table I).
Conjoint Effects of Ca and Pi on Total, Cardiovascular, and Noncardiovascular Mortality
The relations of Ca and Pi to mortality were identical in models including both Ca and Pi variables and in models with only 1 of the Ca or Pi variables (Tables 2 and 3⇑). We explored the possibility of biological interaction between Ca and Pi on mortality risk. No deviation from an additive effect of Ca and Pi on total mortality risk was observed (relative excess risk attributable to interaction, 0.16; 95% CI, −0.21–0.53). This was confirmed graphically (Figure 3).
Finally, we modeled the [Ca×Pi] product. A 1-SD-higher [Ca×Pi] was associated with a 7% to 8% higher risk of total mortality in models A and B (Table 2) and an 8% to 10% higher risk of cardiovascular mortality in models A and B (Table 3). Further, a 1-SD-higher [Ca×Pi] was associated with a 8% higher risk of noncardiovascular mortality in model A but with a 3% nonsignificant increase in model B (Table I). In tertile models (Table 2) and a spline model (data not shown), these associations appeared essentially linear. The [Ca×Pi] product did not predict mortality significantly in models also including the Ca and Pi variables (Tables 2 and 3⇑).
The addition of a continuous [Ca×Pi] variable to the covariates in model B improved discrimination of cardiovascular mortality significantly (IDI, 0.002; P=0.04) and discrimination of total mortality borderline significantly (IDI, 0.002; P=0.06), but not discrimination of noncardiovascular mortality (IDI, 0.000; P=0.80).
In the subsample with eGFRCG >90 mL/min/1.73m2 and the subsample with normal or low serum Pi and Ca, relations of the [Ca×Pi] product to total, cardiovascular, and noncardiovascular mortality were similar to the relations in the total sample (Tables 2 and 3⇑; Table I).
In this long-term, longitudinal study extending nearly 30 years, we provide evidence that higher serum Ca, Pi, and [Ca×Pi] are associated with higher total mortality risk in middle-aged men. A higher serum Ca was mainly related to noncardiovascular mortality, and higher serum Pi and [Ca×Pi] were related to cardiovascular mortality. Although the risks were moderate, our observations for total and cardiovascular mortality were consistent in patients with normal renal function (eGFR >90 mL/min/1.73m2) and in a subsample with normal or low Ca and Pi. This suggests that measurement of Ca and Pi provide prognostic information in the general population, even in the absence of mild renal dysfunction.
Our finding that serum Ca is a predictor of total and noncardiovascular mortality in the community is novel. These associations were most pronounced for the highest tertile of Ca, indicating the possibility of a slightly nonlinear relation, as also is apparent in spline models. In agreement with our findings, Leifsson et al16 previously reported that a higher, unadjusted serum Ca value in the community is associated with higher mortality risk related to cardiovascular events and malignancies during a follow-up period of 10 years. In contrast, Onufrak et al10 did not find an association between serum Ca level and mortality or cardiovascular events. Summarizing, it remains uncertain whether a higher serum Ca is associated with higher mortality risk, although the observations in the present study supports such an association. Further, it is well-established that other states of abnormal Ca metabolism, such as primary hyperparathyroidism, are associated with an increased mortality risk.17 Importantly, the predominating cause of noncardiovascular death was neoplasms, implying that a higher Ca level could either predispose to or be a marker of an increased cancer risk.
Current knowledge about the relation of serum Pi to total mortality in the general population is scarce. To our knowledge, only one previous study has shown a relation between higher serum Pi and increased total mortality in persons with normal renal function.10 The biological significance of this relation remains unclear, and although we observed a similar trend, this was largely driven by cardiovascular mortality risk.
In recent years the concept of a cardio-renal syndrome has progressively emerged, characterized among others by Pi-related cardiovascular toxicity, even when serum Pi levels are within the normal range.18 In this regard, Dhingra et al8 demonstrated a graded independent relation between serum Pi and incidence of cardiovascular disease in the community. A higher normal serum Pi level has also been attributed to higher risk of cardiovascular disease in patients without CKD but with a previous myocardial infarction.19 Similarly, it was recently shown in predialysis CKD patients that higher serum Pi, within the normal range, were associated with the degree of vascular calcification, arterial stiffness,7,20 and increased mortality risk.21 Our long-term, prospective study provides additional support that higher serum Pi is associated with higher risk of cardiovascular mortality.
As a word of caution, it needs to be emphasized that our study is observational and it remains to be proven whether serum Pi is a modifiable risk factor. Although previous randomized, clinical trials targeting serum Pi levels have failed to reduce the overall mortality in dialysis patients,22 recent data support that the use of Pi binders may confer improved survival in incident hemodialysis patients.23 Reported future randomized clinical trials exploring the effect of lowering Pi and the [Ca×Pi] on patient survival in earlier CKD will be of great interest.24
Another observation that needs to be pointed out is that the effects of Pi and Ca on mortality were independent of each other, because the relation of Ca and Pi to mortality was nearly identical in models including both Ca and Pi compared to models just including one of these variables. Further, their relations to mortality risk were additive, implying that it is appropriate to use measurements of Ca or Pi either alone or in combination for clinical risk prediction.
In the current study, we only examined men; thus, it remains unclear whether our findings are applicable in women. Serum Pi levels have been reported to be higher in healthy women than in men, whereas the association between higher serum Pi level and overall mortality was present only in men.10 This may be explained by effect modification by gender, and further studies are warranted to elucidate possible gender differences with regard to serum Pi and cardiovascular risk.
Mechanistically, there are abundant data supporting the relation between higher Pi and vascular pathology. Elevated extracellular Pi leads to intracellular Pi influx and subsequent calcification in vascular smooth muscle cells in vitro.25,26 High intracellular Pi is also involved in the loss of smooth muscle lineage markers in the vasculature, and a simultaneous gain of osteogenic markers,27,28 which are important pathogenic mediators of vascular calcification and atherosclerosis. Similarly, a reduction of serum Pi decreases aortic calcification in vivo through decreased expression of the osteogenic markers Runx2, Msx2, and osterix.29
The current study provides several strengths: the inclusion of a large number of men; adjustments for relevant cardiovascular risk factors; exclusion of preexisting CKD; and the long follow-up period extending over 30 years, which is nearly twice as long as in previous reports and provides significant power in the long-term cardiovascular mortality risk estimation. Some limitations should be acknowledged. We were unable to adjust for circulating parathyroid hormone and vitamin D levels, and some residual confounding from dietary Pi and Ca intake, physical activity level, and seasonal variations in vitamin D production may exist. However, we did, in part, compensate for the lack of parathyroid hormone and vitamin D levels by analyzing subsamples with normal kidney function (eGFR >90 mL/min/1.73 m2) and normal to low serum Ca and Pi. These analyses excluded participants with overt hypercalcemic primary hyperparathyroidism and secondary hyperparathyroidism related to renal dysfunction (the most common forms of primary and secondary hyperparathyroidism, respectively). Nevertheless, we cannot rule out the possibility of secondary hyperparathyroidism caused by vitamin D insufficiency or idiopathic hypoparathyroidism, although these conditions are presumably very rare in this study population and are unlikely to affect the results. Finally, it remains unclear how well a single measurement of Ca and Pi at baseline reflects variations in these parameters over time; however, such variations are most likely to conservatively bias our risk estimates.
In conclusion, higher circulating Ca and Pi levels are associated with higher risk of total, cardiovascular, and noncardiovascular mortality in the community. Additional studies are warranted to evaluate the clinical implications of lowering serum Pi and the [Ca×Pi]; however, such studies should not be restricted to patients with CKD and should extend to the general population.
Sources of Funding
The Swedish Research Council, the Novo Nordisk Foundation, the Swedish Kidney Foundation, and the Swedish Society of Medicine supported this work.
Received August 28, 2009; revision accepted November 11, 2009.
Block GA, Klassen PS, Lazarus JM, Ofsthun N, Lowrie EG, Chertow GM. Mineral metabolism, mortality, and morbidity in maintenance hemodialysis. J Am Soc Nephrol. 2004; 15: 2208–2218.
Ix JH, De Boer IH, Peralta CA, Adeney KL, Duprez DA, Jenny NS, Siscovick DS, Kestenbaum BR. Serum phosphorus concentrations and arterial stiffness among individuals with normal kidney function to moderate kidney disease in MESA. Clin J Am Soc Nephrol. 2009; 4: 609–615.
Onufrak SJ, Bellasi A, Cardarelli F, Vaccarino V, Muntner P, Shaw LJ, Raggi P. Investigation of gender heterogeneity in the associations of serum phosphorus with incident coronary artery disease and all-cause mortality. Am J Epidemiol. 2009; 169: 67–77.
Pencina M, D‘Agostino RS, RB RJDA, Vasan R. Evaluating the added predictive ability of a new marker: from area under the ROC curve to reclassification and beyond. Stat med. 2008; 30: 207–212.
Hlatky MA, Greenland P, Arnett DK, Ballantyne CM, Criqui MH, Elkind MS, Go AS, Harrell FE Jr, Hong Y, Howard BV, Howard VJ, Hsue PY, Kramer CM, McConnell JP, Normand SL, O'Donnell CJ, Smith SC Jr, Wilson PW. Criteria for evaluation of novel markers of cardiovascular risk: a scientific statement from the American Heart Association. Circulation. 2009; 119: 2408–2416.
Tonelli M, Sacks F, Pfeffer M, Gao Z, Curhan G. Relation between serum phosphate level and cardiovascular event rate in people with coronary disease. Circulation. 2005; 112: 2627–2633.
Adeney KL, Siscovick DS, Ix JH, Seliger SL, Shlipak MG, Jenny NS, Kestenbaum BR. Association of serum phosphate with vascular and valvular calcification in moderate CKD. J Am Soc Nephrol. 2009; 20: 381–387.
Isakova T, Gutierrez OM, Chang Y, Shah A, Tamez H, Smith K, Thadhani R, Wolf M. Phosphorus binders and survival on hemodialysis. J Am Soc Nephrol. 2009; 20: 388–396.
Li X, Yang HY, Giachelli CM. Role of the sodium-dependent phosphate cotransporter, Pit-1, in vascular smooth muscle cell calcification. Circ Res. 2006; 98: 905–912.
Reynolds JL, Joannides AJ, Skepper JN, McNair R, Schurgers LJ, Proudfoot D, Jahnen-Dechent W, Weissberg PL, Shanahan CM. Human vascular smooth muscle cells undergo vesicle-mediated calcification in response to changes in extracellular calcium and phosphate concentrations: a potential mechanism for accelerated vascular calcification in ESRD. J Am Soc Nephrol. 2004; 15: 2857–2867.
Aikawa E, Nahrendorf M, Figueiredo JL, Swirski FK, Shtatland T, Kohler RH, Jaffer FA, Aikawa M, Weissleder R. Osteogenesis associates with inflammation in early-stage atherosclerosis evaluated by molecular imaging in vivo. Circulation. 2007; 116: 2841–2850.
Steitz SA, Speer MY, Curinga G, Yang HY, Haynes P, Aebersold R, Schinke T, Karsenty G, Giachelli CM. Smooth muscle cell phenotypic transition associated with calcification: upregulation of Cbfa1 and downregulation of smooth muscle lineage markers. Circ Res. 2001; 89: 1147–1154.
Mathew S, Tustison KS, Sugatani T, Chaudhary LR, Rifas L, Hruska KA. The mechanism of phosphorus as a cardiovascular risk factor in CKD. J Am Soc Nephrol. 2008; 19: 1092–1105.