Serum Calcium and Incident and Fatal Prostate Cancer in the National Health and Nutrition Examination Survey

Little is known conclusively about the etiology of prostate cancer, which is the second leading cause of cancer deaths among U.S. men (1). We recently proposed that high serum levels of parathyroid hormone increase the risk of advanced prostate cancer (2). Under normal physiologic conditions, serum parathyroid hormone acts to maintain serum calcium levels within a narrow range (∼9-10.5 mg/dL). However, prostate cancer cells express receptors for parathyroid hormone (parathyroid hormone–type I receptors) and for calcium (calcium-sensing receptors; ref. 3). In laboratory studies, parathyroid hormone and calcium each promotes the growth and metastasis of prostate cancer cells (4, 5). These observations led us to ask whether men with high levels of serum calcium are at increased risk for prostate cancer.

We conducted a prospective cohort study of serum calcium and the risk for incident and fatal prostate cancer using the first National Health and Nutrition Examination Survey (NHANES) and the NHANES Epidemiologic Follow-up Study (6-9). We included males ages 24 to 77 y at baseline examination in 1971 to 1975. Participants who reported lung, colon, or prostate cancers before the baseline exam (n = 102) and those who reported prostate cancer or died within 1 y of baseline (n = 90) were excluded. The final analytical cohort included 2,814 men for whom baseline serum calcium measurements were available.

Person-time at risk was calculated as the interval between the dates of the initial examination and prostate cancer diagnosis or death (for cases) or date of last contact (for noncases). Prostate cancer was ascertained by self-report (with or without a recorded hospital stay) and by death certificate. Cases with a death record were considered fatal cases, and all others were considered nonfatal cases. For cases with a hospital record, the date of diagnosis was the date of first admission for prostate cancer. For cases without a hospital record, the diagnosis date was defined as June 30 of the year of self-report. For cases confirmed by death certificate only, the date of diagnosis was the date of death. The earliest diagnosis date was used for cases with multiple dates. Fatal cases were identified through a death record linkage with prostate cancer mentioned as a cause of death.

We estimated the relative risks using Cox proportional hazards models relating tertiles of serum calcium to prostate cancer while controlling for potential confounders. Potential confounding was evaluated by age (years), body mass index (BMI; as both a continuous and a categorical variable, cut point at 25 kg/m²), race (black or white), and self-reported family history of prostate cancer. The linear trend was modeled as the median serum calcium value for each tertile as a continuous value. The statistical interaction was tested by a likelihood ratio test of saturated models and Wald's test of cross-product terms. Analyses were done with the use of SAS 9.1 (SAS, Inc.).

Eighty-five incident cases of prostate cancer, including 25 fatal cases, were observed among 2,814 men over 46,188 person-years of follow-up (1971-1993). The 25 fatal cases included 10 cases for which the death certificate was the only record of prostate cancer (Table 1). Men with higher serum calcium tended to be younger (46 versus 52 years in the 3rd and 1st tertiles, respectively) and were more likely to be black (15.4% versus 11.6% black in the 3rd and 1st tertiles).

There was no evidence of increased risk of incident disease with increasing serum calcium levels. Conversely, comparing men in the top with men in the bottom tertile of serum calcium, we observed a significantly increased hazard for fatal prostate cancer that persisted after adjustment for age (age-adjusted relative hazard = 2.59; 95% confidence interval, 1.00-6.72; P_trend = 0.05), BMI (age-adjusted and BMI-adjusted relative hazard = 2.62; 95% confidence interval, 1.01-6.80; P_trend = 0.04), and race (age-adjusted, BMI-adjusted, and race-adjusted relative hazard = 2.68; 95% confidence interval, 1.02-6.99; P_trend = 0.04). The tests for linear trend were statistically significant. However, most of the excess risk was concentrated in the top tertile of serum calcium. There was no substantive difference between adjusting for BMI as a continuous or categorical variable. The adjustment for family history of prostate cancer did not alter the point estimate but widened the confidence interval (P_trend = 0.06). We did not detect statistically significant interactions between serum calcium and other major risk factors for prostate cancer, but power was limited (Table 2).

In this prospective cohort, we observed an ∼3-fold increased risk for fatal prostate cancer among men in the upper tertile of the distribution of serum calcium. We observed a significant dose response. To our knowledge, this is the first study to examine prostate cancer risk in relation to serum calcium. The measurement of serum calcium preceded the diagnosis of prostate cancer by an average of 9.9 years (SD = 4.5 years).

Numerous studies have investigated the role of dietary calcium in prostate cancer, with mixed results (e.g., refs. 10-12). Calcium levels in serum are tightly controlled over a wide range of dietary calcium and generally are not correlated with dietary calcium levels (13). For example, in a random subsample of 354 subjects drawn from a cross-sectional study of >3,400 participants, Mataix et al. found no relationship between the measurements of calcium in diet and in serum (14). The concentration of serum ionized calcium, the biologically active fraction of calcium, is tightly regulated by parathyroid hormone and normally does not deviate by >2% from its set point. This contributes to the stability of total serum calcium levels, of which ∼50% is ionized (15).

High serum calcium significantly predicted fatal but not incident prostate cancer. This may reflect the fact that, in the “post-PSA era,” incident prostate cancers are predominantly screen-detected cancers that are not life threatening (i.e., are not true cases). Additionally, some self-reported cases of prostate cancer may have been missed (i.e., underreporting), and some may have been misreported (i.e., overreporting). These forms of misclassification are likely to be nondifferential and, therefore, bias results toward the null. Our finding is consistent with the results of a large population-based study of serum calcium and survival. In that study of 33,346 Swedes, Liefsson and Ahrén found an increased risk for fatal (cause not further specified) but not for incident malignancy among men <50 years of age with serum calcium >2.5 mmol/L (10.0 mg/dL; ref. 16). It is noteworthy that the increased cancer mortality observed in this Swedish study was observed for men with serum calcium in the upper normal range.

The normal reference range for serum calcium varies by laboratory and assay methods. A commonly used normal reference range is 9 to 10.5 mg/dL (2.2-2.6 mmol/L; ref. 17). Few cases in our study were hypercalcemic (2 of 25 or 8% had serum calcium >10.5 mg/dL). Rather, the increased risk we observed was due largely to men with serum calcium in the upper portion of the reference range (cf. Liefsson and Ahrén). We hypothesize that this increased risk may be attributable to factors that control serum calcium levels, such as genetic variation in the genes for the calcium-sensing receptor (18).

This study benefits from its prospective design and a priori hypothesis. The principal threat to the validity of prospective cohort studies, loss to follow-up for mortality, was low (3.7%). Our findings could be influenced by bias and by confounding. For example, it is possible that some advanced prostate cancers were undetected at the time of calcium measurement. This would underestimate the positive association between serum calcium and prostate cancer because serum levels of calcium in men with advanced prostate cancer typically are normal or low due to the transfer of calcium from the serum into bony lesions (2, 19). Second, it is conceivable that some men failed to report advanced prostate cancer that had been treated. The serum calcium levels for such men could therefore reflect the effects of treatment. This possibility is unlikely to have influenced our results because the standard treatment for prostate cancer, androgen deprivation, does not significantly alter the serum levels of calcium or parathyroid hormone (20). Lastly, it is possible that serum calcium levels are confounded by vitamin D deficiency, which is known to elevate the serum levels of parathyroid hormone and may be associated with the risk of prostate cancer (21). Vitamin D deficiency is unlikely to be an important confounder in this study because vitamin D deficiency is associated with normal or with low serum calcium (i.e., the opposite of the result observed).

Our results are based on a small number of fatal cases (n = 25) and require confirmation by other prospective studies. If confirmed, these findings have important implications for risk stratification and for prostate cancer prevention. Presently, the only established risk factors for prostate cancer are race and family history. Our finding of a >2.5-fold increased risk for men in the highest tertile of serum calcium is comparable in magnitude with the increased risk associated with family history (∼2.5) and could add significantly to our ability to identify men at increased risk for fatal prostate cancer (22). Most importantly, unlike family history, serum calcium and parathyroid hormone are factors that can be modified by lifestyle and by pharmacologic means (23).

The complex sampling design of the NHANES surveys involves the clustering and weighting of observations. Whether statistical analyses of longitudinal data derived from these surveys need account for the sample design is debated. We examined the positive association between serum calcium and prostate cancer mortality in NHANES III using SUDAAN, which adjusts for the sample design. We confirmed relative risks similar to those reported here for NHANES I using SAS, which does not adjust for the sample design (Skinner and Schwartz, in preparation).

No potential conflicts of interest were disclosed.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.