Prediction of serum ionized calcium concentration by serum total calcium measurement in cats

Abstract

Feline serum samples (n = 434) were classified as hypercalcemic, normocalcemic, or hypocalcemic based on both total calcium (tCa) and ionized calcium (iCa) concentrations. Sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), positive diagnostic likelihood ratio (PDLR), and negative diagnostic likelihood ratio (NDLR) were calculated for prediction of hypercalcemia and hypocalcemia in all samples, in hypoalbuminemic cats, and in those with chronic renal failure (CRF) as compared with cats that had other conditions. Diagnostic discordance in prediction of iCa using tCa was 40%. Sensitivity of tCa in prediction of ionized hypercalcemia was low and specificity was high. The PDLR for prediction of ionized hypercalcemia or hypocalcemia was low in all cats, especially in those with CRF. Due to the high level of diagnostic discordance, tCa should not be used to predict iCa concentration. Concentration of iCa should be measured directly when accurate assessment of calcium status is needed.

Serum calcium exists in 3 fractions: ionized (iCa), complexed (cCa), and protein-bound (pCa) (1). In normal cats, pCa, cCa, and iCa account for approximately 40%, 8%, and 52% of total serum calcium (tCa) concentration, respectively (2).

Even though iCa is biologically active, clinicians rely on serum tCa measurements to predict iCa status. Variations in pCa or cCa can render the tCa concentration abnormal, even though the iCa fraction is within normal limits. Since some serum tCa is protein-bound, it had been suggested to “adjust” tCa relative to serum total protein (TP) or albumin concentration to improve diagnostic interpretation of tCa in dogs, especially when TP or albumin alterations are present (3). Adjustment formulas are not recommended for use in cats, due to a poorer correlation observed between TP or albumin and tCa (4), and adjustment formulas for dogs have been shown to offer no diagnostic advantage (5). Ion-selective electrodes are available for iCa measurement, and it is unknown if serum tCa accurately predicts iCa concentration. Thus the objective of this study was to determine the diagnostic utility of tCa measurement in predicting serum iCa status in feline serum samples.

Review of 434 consecutive feline samples presented to the clinical chemistry laboratory at The Ohio State University was undertaken for serum iCa, tCa, TP and albumin concentrations. Samples were included only if tCa, TP, albumin, and iCa measurements were performed on the same serum sample. A calcium ion-selective electrode was used to anerobically measure iCa (634 Ca⁺⁺-pH analyzer; Ciba-Corning, Medfield, Massachusetts, USA), and an automated analyzer (Hitachi 911; Roche Diagnostics Corporation, Indianapolis, Indiana, USA) was used to measure serum tCa, TP, and albumin.

Samples were subdivided by diagnosis into those cats diagnosed with CRF (n = 102), and those with conditions other than CRF (n = 332). Patients were diagnosed based on clinical signs, history, and diagnostic testing. For separate comparison, 125 samples were identified with serum albumin concentration below the reference range (OSU, 2.5 to 3.5 g/dL). Hypoalbuminemic cats were affected by a variety of conditions, including some with CRF.

Samples from cats were classified as hypercalcemic, normocalcemic, or hypocalcemic based on serum tCa, and compared to their matched serum iCa classification. Normocalcemia was defined as serum tCa within a range of 2.25 to 2.75 mmol/L (9 to 11 mg/dL), or serum iCa within a range of 1.18 to 1.38 mmol/L (4.7 to 5.5 mg/dL) (The Ohio State University Clinical Chemistry Laboratory). Hypercalcemia was defined as serum tCa concentration > 2.75 mmol/L (11 mg/dL), or iCa > 1.38 mmol/L (5.5 mg/dL). Hypocalcemia was defined as serum tCa concentration < 2.25 mmol/L (9 mg/dL), or iCa < 1.18 mmol/L (4.7 mg/dL).

Sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), positive diagnostic likelihood ratio (PDLR), and negative diagnostic likelihood ratio (NDLR) of serum tCa were calculated for hypercalcemia and hypocalcemia in all samples. Calculations were performed comparing hypercalcemic samples to non-hypercalcemic samples and hypocalcemic samples to non-hypocalcemic samples. Diagnostic discordance using tCa was also determined. Diagnostic discordance was the percentage of cat samples with incorrect calcium status identification based on tCa. In addition, all parameters were calculated in the subpopulations of hypoalbuminemic cats, those diagnosed with CRF, and in cats with conditions other than CRF.

The kappa coefficient was calculated as a measure of concordance in categorical sorting. Kappa provides a measure of the degree to which 2 tests concur when sorting based on defined criteria or categories (a reference range). The kappa coefficient relates actual agreement with chance agreement, with a maximum value of 1.00 representing perfect agreement (6). Agreement is considered poor if kappa is ≤ 0.20; fair if > 0.20 but ≤ 0.40; moderate if > 0.40 but ≤ 0.60; substantial if > 0.60 but ≤ 0.80; and good if > 0.80. Correlations of serum TP and albumin to tCa, and tCa to iCa were assessed by use of least squares regression. The correlation coefficients (r) and 95% confidence interval (CI) were calculated using statistical software (Minitab statistical software; Minitab, State College, Pennsylvania, USA).

The correlation of serum tCa to albumin or TP was poor in all cats (0.26, 0.21, respectively). Serum tCa showed a poorer correlation to serum iCa in cats with CRF (0.68) as opposed to cats without CRF (0.90).

Of the 434 feline samples, 128 were identified as normocalcemic, 26 were hypercalcemic, and 105 were hypocalcemic with both serum iCa and tCa measurement. A total of 175 samples were incorrectly identified based on serum tCa for a diagnostic discordance of 40%. In the 75 samples identified as hypercalcemic on serum iCa measurement, 48 (64%) were falsely identified as normocalcemic, and one (1%) was misdiagnosed as hypocalcemic based on serum tCa concentration. In the 119 samples identified as hypocalcemic based on serum iCa measurement, 14 (12%) were falsely identified as nor-mocalcemic based on serum tCa concentration. In the 240 samples with ionized normocalcemia, 4 (2%) were incorrectly identified as hypercalcemic, and 108 (45%) were incorrectly identified as hypocalcemic based on serum tCa concentration. Of the 434 samples, 102 were from cats with CRF. Of these, 50 were normocalcemic, 16 were hypercalcemic, and 3 were hypocalcemic based on both serum iCa and tCa concentrations. A total of 33 samples were incorrectly identified in cats with CRF based on serum tCa for a diagnostic discordance of 32%. In the 31 samples from cats with CRF identified as hypercalcemic based on iCa measurement, 15 (48%) were falsely identified as normocalcemic based on tCa measurement. In the 10 samples from cats with CRF identified as hypocalcemic based on serum iCa measurement, 7 (70%) were misdiagnosed as normocalcemic based on tCa measurement. Of 434 feline samples, 125 were associated with hypoalbuminemia. Of these, 40 were normocalcemic, 11 were hypercalcemic, and 27 were hypocalcemic based on both serum iCa and tCa concentration. A total of 47 samples were incorrectly identified in cats with hypoalbuminemia based on serum tCa for a diagnostic discordance of 38%. In the 24 samples from hypoalbuminemic cats identified as hypercalcemic based on iCa measurement, 13 (54%) were falsely identified as normocalcemic based on tCa measurement. In the 34 samples from hypoalbuminemic cats identified as hypocalcemic based on serum iCa measurement, 7 (20%) were misdiagnosed as normocalcemic based on tCa measurement. The diagnostic discordance was similar in cats with CRF, cats with conditions other than CRF, and cats with hypoalbuminemia as compared to all cats. The kappa coefficient showed fair agreement when using serum tCa to predict iCa in all cats (0.34), cats with CRF (0.36), cats with conditions other than CRF (0.29), and cats with hypoalbuminemia (0.38).

Based on serum iCa measurement in all cats, the prevalence of hypercalcemia, normocalcemia, and hypocalcemia was 17%, 55%, and 27%, respectively (Table I), whereas serum tCa predicted 7% as hypercalcemic, 44% as normocalcemic, and 49% as hypocalcemic. Hypercalcemia and normocalcemia were underestimated, and hypocalcemia was overestimated using serum tCa concentration. In the subpopulation of cats with CRF, the prevalence of hypercalcemia was higher (29%) and prevalence of hypocalcemia was lower (10%). Hypercalcemia in cats with CRF was underestimated using tCa measurement, and normocalcemia in cats with CRF was slightly overestimated by tCa measurement. In hypoalbuminemic cats, hypercalcemia was underestimated, and hypocalcemia was overestimated when using tCa.

Table I.

Percentages of cats diagnosed as hypercalcemic, normocalcemic or hypocalcemic based on serum iCa and tCa concentrations

	Hypercalcemic	Normocalcemic	Hypocalcemic
All cats (n = 434)
iCa	17.3	55.3	27.4
tCa	6.9	43.8	49.3
Cats with CRF (n = 102)
iCa	29.4	60.8	9.8
tCa	19.6	69.6	10.8
Cats with conditions other than CRF (n = 332)
iCa	13.3	53.9	32.8
tCa	3.0	35.5	61.5
Cats with hypoalbuminemia (n = 125)
iCa	19.2	53.6	27.2
tCa	10.4	48.0	41.6

Sensitivity, specificity, PPV, NPV, PDLR, and NDLR for use of serum tCa concentration to predict ionized hypercalcemia is shown in Table II. These same parameters for the prediction of ionized hypocalcemia are shown in Table III. In cats with CRF, sensitivity and positive predictive value of tCa to predict hypocalcemia was much lower, compared with cats with conditions other than CRF. The specificity of tCa was higher in the prediction of hypocalcemia in cats with CRF, compared with other cats.

Table II.

Sensitivity, specificity, positive predictive value, negative predictive value, positive diagnostic likelihood ratio, and negative diagnostic likelihood ratio of total calcium in the prediction of ionized hypercalcemia in cats

	All cats	Cats with CRF	Cats with conditions other than CRF	Cats with hypoalbuminemia
Sensitivity	35	52	22	46
Specificity	99	94	100	98
Positive predictive value	87	80	100	85
Negative predictive value	88	82	89	88
Positive diagnostic likelihood ratio	31.1	9.2	(100)a	23.1
Negative diagnostic likelihood ratio	0.66	0.51	0.77	0.55

^aThe PDLR is essentially 100 but cannot be truly calculated due to the absence of false positives.

Table III.

Sensitivity, specificity, positive predictive value, negative predictive value, positive diagnostic likelihood ratio and negative diagnostic likelihood ratio of total calcium in the prediction of ionized hypocalcemia in cats

	All cats	Cats with CRF	Cats with conditions other than CRF	Cats with hypoalbuminemia
Sensitivity	88	30	94	79
Specificity	65	92	54	72
Positive predictive value	49	30	50	52
Negative predictive value	94	92	94	90
Positive diagnostic likelihood ratio	2.6	3.9	2.0	2.9
Negative diagnostic likelihood ratio	0.18	0.76	0.12	0.28

Measurement of iCa differs from many other tests in that there are different conditions to be considered, depending on whether iCa is above or below the normal reference range. Since different diseases cause hypercalcemia or hypocalcemia, the sensitivity, specificity, PPV, NPV, PDLR, and NDLR may differ depending on iCa concentration. Thus, sensitivity, specificity, PPV, NPV, PDLR, and NDLR have been calculated for the diagnosis of both hypercalcemia and hypocalcemia.

In this study, cats were diagnosed with a number of different conditions. Serum protein abnormalities occur in many patients with CRF (7), which may have an impact on the protein-bound fraction of serum total calcium. Metabolic acidosis may also shift calcium from the protein-bound fraction to the ionized fraction. Thus, prediction of iCa by tCa may be considerably different in cats with CRF compared with those with other conditions. Considering the serum abnormalities that can potentially occur and the high percentage of cats with CRF (24%), samples from cats with CRF were analyzed separately and compared to cats that had conditions other than CRF.

The correlation of tCa to albumin or TP was poor in this study, and was poorer than in two previous experiments with cats (4,8). Cats with elevated serum urea nitrogen concentrations were excluded in 1 study (4), and the incidence of cats with CRF was higher in the present study (24%) compared with another study (20%) (8) which may account for the poorer correlation observed. The correlation of tCa to albumin presented in the current work is similar to that seen in 2 human studies and 1 dog study where the correlation of tCa to albumin was 0.32 (9), 0.25 (10), and 0.30 (5), respectively.

The correlation coefficient of tCa to iCa (r = 0.87) indicates a moderately strong linear relationship, and compares similarly with that seen in human studies (9,11,12). However, the closeness of fit is lower, and only approximately 75% of the change in tCa (r² = 0.75) can be attributed to changes in iCa concentrations. The correlation of tCa to iCa in cats with hypoalbuminemia was similar to the correlation seen in the general population. Correlation of tCa to iCa is much poorer in cats with evidence of CRF (r = 0.68), and only approximately 46% of the change in tCa (r² = 0.46) can be attributed to changes in iCa concentrations. This difference in correlation has also been observed in other species in groups of patients with different diseases (5,12).

There was a high level of diagnostic discordance (40%) when using tCa to predict iCa in all cats. In 2 human studies, the level of discordance was 31% (11) and 26% (12), which is similar to that seen in this study. The individual magnitude of discordance in classification of calcium status was not assessed. Thus only relatively small magnitudes of disagreement may lead to a difference in classification of calcium status, and may result in a higher rate of discordance. Approximately 7% of diagnostic discordance may arise from analytical variation in measurement (11,13); however, the observed diagnostic discordance of 40% clearly exceeds what might result from day-to-day variation.

Individual variation in both the pCa and cCa fractions may account for the diagnostic discordance observed when using tCa to predict iCa. In this study, there was poor correlation of tCa to TP or albumin, and it has been previously observed that cCa shows tremendous individual variation, especially in dogs with CRF (14).

Not only is there a high level of discordance when using tCa to predict iCa, but different species do not behave similarly. In this study, hypercalcemia and normocalcemia were underestimated, and hypocalcemia was overestimated when using tCa concentration. In cats with CRF, normocalcemia was slightly overestimated, and hypercalcemia was underestimated when using tCa concentration. In contrast, normocalcemia is overestimated and hypocalcemia is underestimated in dogs when tCa concentration is used (5). In dogs with CRF, hypercalcemia is greatly overestimated, and hypocalcemia is underestimated. The reasons for species differences in prediction of iCa are unclear but may relate to differences in other calcium fractions in feline diseases compared with dogs.

Sensitivity of tCa in prediction of ionized hypercalcemia was low due to the presence of a high number of truly hypercalcemic cats that were normocalcemic on tCa measurement (false negatives). Specificity was high, indicating the presence of a low number of truly normocalcemic cats that were hypercalcemic on tCa measurement (false positives). Similar values using tCa to predict iCa have been reported in humans (9) and dogs (5).

The PDLR and NDLR offer advantage over PPV and NPV as they are not affected by prevalence (15). The PDLR for prediction of ionized hypercalcemia was low in all cats, in cats with hypoalbuminemia, and especially in those with CRF. The PDLR of tCa was very high in cats with conditions other than CRF, suggesting that hypercalcemia on tCa measurement predicts ionized hypercalcemia in essentially 100% of cases, as long as there is no evidence of chronic renal disease or hypoalbuminemia.

Sensitivity of tCa in prediction of ionized hypocalcemia was fairly high in cats with conditions other than CRF. Specificity was much lower in this group indicating the presence of many truly normocalcemic cats (without CRF) that appeared hypocalcemic on tCa measurement (false positives). In those with CRF, sensitivity was much lower than in cats with other disorders, indicating that most cats with ionized hypocalcemia appeared normocalcemic on tCa measurement.

The PDLR for prediction of ionized hypocalcemia was very low in all cats. The NDLR was good for prediction of ionized hypocalcemia in cats with conditions other than CRF. The PDLR for prediction of ionized hypocalcemia was very poor in cats with CRF, suggesting that most hypocalcemic cats with CRF are misdiagnosed as normocalcemic if tCa is used.

Overall, prediction of iCa by tCa was poor, with a high level of diagnostic discordance. The only situation where tCa fairly accurately predicted iCa was in cats with hypercalcemia and no evidence of CRF or hypoalbuminemia. In general, tCa concentrations cannot be relied on to accurately assess calcium status in cats as shown by discordance with iCa measurement. This is especially important in cats with CRF where abnormalities in TP, albumin, and complexed calcium fractions may be present. Due to the high number of false negatives, many cats may appear normocalcemic when tCa is measured, but actually have a derangement in calcium homeostasis. Ionized calcium concentration must be directly measured to accurately assess calcium status.