Serum Calcium and Serum Albumin Are Biomarkers That Can Discriminate Malignant from Benign Pelvic Masses Free

Annually, more than 239,000 women worldwide are diagnosed with ovarian cancer and the majority will die from their disease (1). The high case-fatality rate reflects the lack of an effective early detection test for ovarian cancer and the lack of specific symptoms in early stage disease (2). Approximately 60% of ovarian cancers are detected at advanced stages when the cancer has spread throughout the peritoneal cavity (3).

In addition to efforts to develop more effective therapies, efforts to improve ovarian cancer survival focus on two areas: early detection and the identification of malignant masses in women with a pelvic mass (4). Improving the identification of malignant masses is critical because survival rates are significantly better for women treated by a gynecologic oncologist (5). However, only one third to one half of U.S. women with ovarian cancer are referred to a gynecologic oncologist for primary surgery (6). In part, the low referral rate reflects the difficulty of accurately predicting, prior to surgery, which ovarian masses are malignant.

Methods to improve the differential diagnosis of ovarian cancer include ultrasonography, serum cancer biomarkers, e.g., CA-125, HE-4, multivariate tests such as the OVA1 test, and algorithms that use multiple biomarkers plus imaging data (7, 8). In theory, a complementary approach would use biomarkers that predict ovarian cancer in asymptomatic women. Other than mutations in the BRCA1/2 genes, few such biomarkers have been identified. However, in population-based studies, we found that women without apparent cancer whose serum calcium levels were within the normal range but were high (“high normocalcemia”) were significantly more likely to be diagnosed with and die of ovarian cancer (9). We hypothesized that the association between high normocalcemia and ovarian cancer reflects a paraneoplastic effect of occult ovarian tumors. Many cancer cells, including ovarian cancer cells, upregulate the production of parathyroid-hormone related protein (PTHrP). PTHrP is an oncofetal protein that acts like parathyroid hormone (PTH) and causes serum calcium levels to rise (10).

Because high normocalcemia predicted ovarian cancer among women without evidence of cancer, we reasoned that it might predict ovarian cancer among women with pelvic masses. Specifically, we hypothesized that high normocalcemia would discriminate women with malignant pelvic masses from women with nonmalignant pelvic masses. We tested this hypothesis in an unselected series of women with pelvic masses.

Data were abstracted from the medical records of a consecutive series of 514 women with pelvic masses who underwent surgical resection at Wake Forest University Baptist Medical Center from July 2009 through June 2013. The independent variables included total serum calcium, serum albumin, albumin-adjusted serum calcium, age, body mass index, menopausal status, and parity. The primary outcome was tumor pathology. “Cases” were any malignant histopathology and “controls” were benign and low malignant potential (LMP; “borderline”) tumors. Data on clinical and histopathologic characteristics were obtained from pathology reports. The study was approved by the Wake Forest University Institutional Review Board (#IRB00022763).

Serum calcium and serum albumin were measured in the university hospital laboratory, which undergoes routine quality control. Calcium was analyzed via o-cresolphthalein complexone and the calcium ion complex. Albumin was measured using bromocresol green on a Beckman Coulter Clinical Chemistry analyzer (Beckman Coulter, Inc.). We used data on serum calcium and albumin measured prior to surgery. The biologically active fraction of serum calcium was estimated using the formula: serum albumin–corrected calcium (mg/dL) = total calcium (mg/dL) + 0.8 × (4 – albumin (g/dL); ref. 11). We defined “hypercalcemia” as an albumin-adjusted serum calcium beyond the upper limit of the normal range in our laboratory, 10.5 mg/dL, and “high normocalcemia” as an albumin-adjusted serum calcium ≥ 10 mg/dL.

We divided the patient data into a “training” set, to identify associations and build predictive models, and a “testing” set, to confirm them and to test model performance (Fig. 1). We used a stratified random sampling approach that preserved the proportion of malignancy in both sets. Using the training set, we built multivariable logistic generalized additive models with variables selected based on our previous work (9). Age was centered about the mean. Serum calcium, serum albumin, and body mass index were centered and scaled. Continuous values were modeled alternatively as linear and non-linear terms, thin-plate regression splines with the degree of smoothness selected by generalized cross validation. Model fits were compared using the Akaike Information Criterion (AIC) and explained deviance. The AIC can be used to select among statistical models to identify those that provide an adequate fit to the data using the fewest parameters (12). The final Overa predictive model was selected as the model with the best AIC score in the training data.

Using the testing data set, we computed a risk prediction score based on each of our selected logistic models. For each model, we computed sensitivity, specificity, and positive and negative predictive values (PPV and NPV) at an optimum threshold. We constructed receiver operating characteristic (ROC) curves and computed the area under the curve (AUC) to evaluate the models' predictive power. All analyses employed R using the “caret” package for sampling, “pROC” package for AUC analyses, and “mgcv” for generalized additive models.

There were 170 malignant tumors and 344 benign/borderline tumors. Cases were more likely to be older (mean age = 62 vs. 52), menopausal (80% vs. 57%), and to have a lower body mass index (28 vs. 31 kg/m²). Among cases, 55% were stage IIIC and 74% were high grade. Characteristics of the tumors are shown in Table 1. Eighty-one percent (81%) of the malignant cases were epithelial neoplasms and the most common was a serous cystadenocarcinoma. Epithelial neoplasm was the most common (58%) benign lesion.

Values for serum albumin, calcium, and albumin-corrected calcium by histology and tumor stage are shown in Table 2. The mean total serum calcium was similar in cases and controls (9.34 and 9.31 mg/dL, respectively). Serum albumin was significantly lower in cases (3.23 g/dL vs. 3.85 g/dL, P <0.01). Cases had a higher albumin-adjusted calcium than controls (9.95 mg/dL vs. 9.53 mg/dL, P < 0.01). High normocalcemia occurred in 53% of malignant tumors versus 12% of benign and LMP tumors. Total serum calcium levels did not vary by tumor stage (Table 3). Conversely, serum albumin levels were significantly lower with higher stage (P < 0.0001), and albumin-adjusted serum calcium levels were significantly higher with higher stage (P < 0.0001). Serum albumin levels were particularly low among the 14 malignant cases with mucinous histology (mean = 2.78 g/dL; range, 1.3–3.7; Table 2).

Characteristics of the patients in the training dataset are shown in Table 4. Total serum calcium, unadjusted for serum albumin, was not significantly different in malignant (9.40 mg/dL), LMP tumors (9.39 mg/dL), and benign masses (9.32 mg/dL). Conversely, serum albumin levels were significantly lower among women with malignant tumors (3.26 g/dL) versus women with LMP tumors (3.83 g/dL) and benign masses (3.90 g/dL). The mean albumin-adjusted serum calcium in the training set was significantly higher among cases than controls, 10.0 mg/dL versus 9.5 mg/dL (P < 0.01). In the training dataset, the odds of malignancy for a woman with high normocalcemia, adjusting for age and body mass index, was 15.2 [95% confidence interval (CI), 6.3–36.7].

The distribution of variables in the testing dataset was very similar to that of the training set. As in the training set, total serum calcium levels were not different in cases and controls (9.3 vs. 9.3 mg/dL); albumin levels were significantly lower in cases (3.2 vs. 3.8 g/dL), and albumin-adjusted serum calcium was significantly higher (9.9 vs. 9.5 mg/dL, P < 0.01). In the testing dataset, high normocalcemia was associated with a risk of malignancy that was increased approximately 14-fold (OR, 13.57; 95% CI, 5.7–32.1).

Modeling the variable, “total serum calcium,” in isolation and uncorrected for serum albumin, provided little predictive information. However, total serum calcium corrected for serum albumin explained nearly 20% of the deviance in the outcome. Modeling calcium and albumin as separate variables, rather than adjusting total serum calcium for albumin, provided significant additional information. The best fit, most parsimonious model (Overa) incorporated serum calcium, serum albumin, age, and body mass index as nonlinear terms and also included menopausal status and parity.

At an optimum threshold, albumin alone had a sensitivity of 58% and a specificity of 90%, yielding a PPV of 76% and an NPV of 79% with an AUC of 0.77. Overa achieved an AUC of 0.83 with a sensitivity of 72% and specificity of 83%, a PPV of 71% and an NPV of 85%. The P value comparing the performance of the Overa model to albumin-corrected calcium alone was highly significant (P < 0.001; Table 5). We conducted analyses separately for pre- and postmenopausal women using the Overa model. For postmenopausal women, the sensitivity is 71.4% and specificity is 84.9%. For premenopausal women, the sensitivity is 91.7% and specificity is 71.7%.

An important goal for biomarkers in this setting is the detection of potentially curable ovarian cancers, i.e., stage I disease. The mean albumin-adjusted serum calcium of women with stage I disease in the testing dataset, (9.82), was significantly higher than for control women, 9.52 mg/dL (P = 0.0004). The AUC of the Overa model for discriminating stage I versus benign/borderline tumors was 0.70, with sensitivity of 0.94 and a specificity of 0.44.

Between 5% and 10% of U.S. women undergo surgery for suspected ovarian cancer at some point in their lives and approximately 13% to 21% of these will have cancer (4, 13). Accurately predicting which tumors are malignant prior to surgery could improve survival by triaging women with ovarian cancer to specialist care. In our series of 514 women with pelvic masses who underwent surgical resection, the majority of women with cancer, 53%, had high normocalcemia versus 12% of women with benign/LMP tumors. We confirmed that women with high normocalcemia were approximately 14 times more likely (95% CI, 5.7–32.1) to have a tumor that was malignant. These data extend our previous findings that high normocalcemia predicts ovarian cancer in asymptomatic women (9).

Women with ovarian cancer had significantly lower serum albumin than women with benign and LMP tumors. Low serum albumin is well known in association with ovarian cancer and is an adverse prognostic index for survival (14, 15). The causes of hypoalbuminemia in ovarian cancer include poor nutrition, small bowel obstruction, ascites, and the metabolic effects of the tumor mass (16). In normal women, serum albumin levels decline slightly with age. Among non-Hispanic whites in the National Health and Nutrition Examination Survey (NHANES) III, the mean serum albumin among women 25 to 59 years (N = 1,427) was 41.6 g/L (SE = 0.25) and among women 60 to 89 years (N = 1,195) was 40.1 g/L (SE = 0.23; 17, 18). Although low serum albumin is common in ovarian cancer, to our knowledge, it has not previously been considered as an aid in its differential diagnosis.

Approximately 50% of serum calcium is bound to serum proteins, principally albumin. Because low serum albumin may mask high serum calcium, serum calcium levels commonly are adjusted for serum albumin < 4.0 g/dL (19). Many apparently normocalcemic women with ovarian cancer would be classified as hypercalcemic if their calcium levels were albumin-adjusted; e.g., a woman with a total serum calcium of 9.9 mg/dL and serum albumin of 3.2 g/dL (the mean albumin of cases in our training set) has an adjusted serum calcium of 10.54 mg/dL (normal: 8.5–10.5 mg/dL).

Different cutpoints have been used to define hypercalcemia, though a value > 10.5 mg/dL (2.6 mmol/L) is typical (20). Hypercalcemia occurs in approximately 10% to 30% of patients with cancer and is most often observed in multiple myeloma and cancers of the lung and breast. Hypercalcemia has been considered uncommon in most ovarian cancers. In a large series of gynecologic cancer patients (N = 5,260), using a sensitive definition of hypercalcemia (> 10.2 mg/dL), Jaishuan and colleagues reported a prevalence of 5%. That value underestimates the true prevalence as the samples were not albumin-adjusted (21). Using albumin-adjusted serum calcium and a more conservative definition of hypercalcemia (> 10.5 mg/dL), 8.4% of our ovarian cancer patients were hypercalcemic versus 1.2% of women with benign masses. Although hypercalcemia was present in a minority of cases, high normocalcemia was present in the majority (53%). Our data support the insights of Allan and colleagues, who, based on 5 cases of ovarian cancer (2 clear cell carcinomas and 3 cystadenocarcinomas, none of whom had bony metastases), concluded in 1984 that “paraneoplastic hypercalcemia due to ovarian carcinoma may be more common than generally recognized” (22).

Hypercalcemia has been described in several rare forms of ovarian cancer, especially small cell carcinoma, which accounts for approximately 1% of ovarian cancers. Young and colleagues reported that 62% of a case series of small cell carcinoma were hypercalcemic (23). The high prevalence was based on 49 of 79 patients (from a series of 150) in whom preoperative total serum calcium was measured. Because serum calcium likely was measured for cases with suspected hypercalcemia, the true prevalence of hypercalcemia in small cell carcinoma may be lower. Our study had no cases of this rare histology.

Hypercalcemia also has been reported in 5% to 10% of clear cell ovarian cancers (24). Our findings are consistent with this, as 1 of 11 (9%) of our clear cell patients was hypercalcemic. It is noteworthy that these two ovarian cancer histologies, small cell and clear cell, have poor prognoses, supporting the view that high serum calcium is a marker of aggressive ovarian cancer (25, 26). We have proposed that high serum calcium represents a spectrum in ovarian cancer, with cancers like small cell and clear cell cancer inhabiting the far end of the spectrum (9).

We hypothesize that a plausible mechanism for the association between high normocalcemia and malignancy in pelvic masses involves two elements: First, malignant ovarian tumors depress serum albumin more than benign tumors do. This likely includes a metabolic effect of the malignant mass. This effect is seen in Table 3, where the mean serum albumin in benign/LMP tumors is 3.85 g/dL and declines progressively in malignant tumors from 3.68 in stage I, to 3.23, 3.17, and 2.82 g/dL in stages II, III, and IV (P <0.0001). Second, malignant tumors increase total serum calcium levels more than benign tumors do. A likely mechanism is overexpression by the tumors of PTHrP. PTHrP is expressed in the normal ovary and increased expression has been shown in many types of ovarian cancer (10). PTHrP promotes the release of calcium from bone and inhibits its excretion in the kidney, causing serum calcium levels to rise. This is the fundamental mechanism of humoral hypercalcemia of malignancy (27). We propose that ovarian cancers exert an analogous but subtler effect, i.e., a “humoral high normocalcemia of malignancy.”

The use of biomarkers for the diagnosis of malignancy in women with pelvic masses is an area of large current interest (4, 8). Currently, there are two FDA-approved biomarker panels to help discriminate malignant from benign ovarian masses in women scheduled for surgery: the OVA1 test, and the combination of CA-125 and HE-4. The performance of these tests depends greatly upon factors such as the prevalence of cancer in the study population and the proportion of cases that are postmenopausal. The sensitivity of OVA1 (which includes CA-125) for all stages combined was reported to be 96% in postmenopausal women and 84% in premenopausal women with specificities of 28% and 40%, respectively (28). Although the Overa model cannot be compared fairly with these tests except via a “head to head” comparison, its high sensitivity for premenopausal women, 91.7%, accompanied by a high specificity, 71.7%, suggests that the Overa model might be especially useful for premenopausal women. The high sensitivity of Overa for stage I disease, 0.94, is particularly noteworthy. This contrasts with the sensitivity of CA-125 alone, which is approximately 0.50.

The use of serum calcium and albumin to predict malignancy in this setting has several advantages over present technologies. Unlike ultrasonography of adnexal masses, whose interpretation is highly subjective, serum calcium and albumin are well-characterized, objective parameters (29). In addition, the cost of the OVA1 test is approximately $650.00 (in 2013 prices; ref. 30). Conversely, total serum calcium and albumin are measured very inexpensively. Values for these analytes likely are already in the patient's medical record and thus their incremental cost is zero. A test that uses serum calcium and albumin therefore may be particularly useful in a low resource environment. In addition to its use as a triage test, we speculate that increasing values of serum calcium and decreasing values of serum albumin over time may be useful as a screening tool for the early detection of ovarian cancer.

Our study has several limitations. First, due to its cross-sectional design, a temporal relationship between serum albumin and serum calcium and ovarian cancer cannot be conclusively established. The known association between hypoalbuminemia and ovarian cancer argues for a causal role of cancer as the cause of the low albumin. It is possible that high normocalcemia could predispose to, rather than be caused by, ovarian cancer. However, the known relationship between several types of ovarian cancer and hypercalcemia, as well as an established mechanism by which ovarian cancers raise serum calcium, supports the latter interpretation. Secondly, our study was retrospective and was performed at a single institution. It is conceivable that unknown biases present at a single institution could have contributed to the associations observed. Conversely, this study has several strengths, including a clear a priori hypothesis, large sample size, and the use of a discovery and verification approach.

Neither high normocalcemia nor hypercalcemia is specific for cancer. In the community setting, most hypercalcemia is due to primary hyperparathyroidism, whereas in hospitalized patients, most reflect malignancy. Undiagnosed primary hyperparathyroidism should be rare in this population. For example, Press and colleagues analyzed the medical records of 2.7 million patients at a tertiary referral center and found a prevalence of hypercalcemia (unadjusted total calcium > 10.5 mg/dL) among women of < 2% (31). In practice, primary hyperparathyroidism can be ruled out by a serum PTH that is not elevated (32).

Although total serum calcium adjusted for serum albumin gives an approximation of biologically active calcium, it is relatively insensitive to true calcemia, which is measured more accurately using ionized calcium (33, 34). In our population-based studies, ionized calcium gave a higher relative hazard for ovarian cancer (RH, 2.4; 95% CI, 1.45–4.09) than albumin-adjusted serum calcium (RH, 1.47; 95% CI, 1.02–2.13; ref. 9). This suggests that measurements of serum albumin and ionized calcium may yield a better discrimination of malignant masses than albumin and total serum calcium.

In conclusion, in an unselected series of 514 women who underwent surgical resection for a pelvic mass, hypercalcemia was found in 8.4% of cases with malignancy and in 1.2% of women with benign and LMP masses. High normocalcemia occurred in 53% of women with malignant tumors versus 12% of benign/LMP tumors. An albumin-adjusted serum calcium ≥ 10 mg/dL conveyed a risk of malignancy that is increased approximately 14-fold. The use of serum calcium and serum albumin to predict malignancy in this setting has a high sensitivity for early stage disease and has several advantages over present, more expensive technologies. Our model may help triage women with ovarian cancer to appropriate surgical care.

H.G. Skinner and G.G. Schwartz are co-inventors on a patent application using serum calcium to detect ovarian cancer that is assigned to the Wisconsin Alumni Research Fund and Wake Forest University. No potential conflicts of interest were disclosed by the other authors.

Conception and design: M.G. Kelly, G.G. Schwartz

Development of methodology: M.G. Kelly, S.S. Winkler, H.G. Skinner, G.G. Schwartz

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): M.G. Kelly, S.S. Winkler, S.S. Lentz, S.H. Berliner, M.F. Swain

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): M.G. Kelly, S.S. Winkler, H.G. Skinner, G.G. Schwartz

Writing, review, and/or revision of the manuscript: M.G. Kelly, S.S. Winkler, S.S. Lentz, M.F. Swain, H.G. Skinner, G.G. Schwartz

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): H.G. Skinner

Study supervision: G.G. Schwartz

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.