Determination of a serum total calcium concentration threshold for accurate prediction of ionized hypercalcemia in dogs with and without hyperphosphatemia

Abstract Background Total serum calcium (tCa) concentrations are poorly predictive of ionized calcium (iCa) status in dogs. Hypothesis There is an optimal threshold of tCa concentration that is highly predictive of ionized hypercalcemia and this threshold is higher in hyperphosphatemic dogs as compared to nonhyperphosphatemic dogs. Animals Nonhyperphosphatemic (n = 1593) and hyperphosphatemic (n = 250) adult dogs. Methods Retrospective medical record review of paired tCa and iCa concentration measurements in dogs presented to a university teaching hospital over a 5‐year period. Positive and negative predictive values, sensitivity, and specificity were calculated for tCa concentration thresholds of 11.0‐15.0 mg/dL (upper limit of laboratory reference interval = 11.5 mg/dL) in nonhyperphosphatemic and hyperphosphatemic groups. Results In nonhyperphosphatemic dogs, an optimal tCa concentration threshold of 12.0 mg/dL resulted in a positive predictive value of 93% (95% confidence interval [CI], 84%‐98%) and sensitivity of 52% (95% CI, 43%‐61%) for ionized hypercalcemia. An optimal tCa concentration threshold was not identified for hyperphosphatemic dogs. The nonhyperphosphatemic dogs had a higher prevalence of ionized hypercalcemia than the hyperphosphatemic dogs (7 versus 3%, P = .04) and a lower prevalence of ionized hypocalcemia (23 versus 62%, respectively; P < .001). Conclusions and Clinical Importance High tCa concentrations are strongly predictive of ionized hypercalcemia in nonhyperphosphatemic adult dogs and should prompt further diagnostic testing to determine the underlying cause of hypercalcemia. In this population, dogs without increased tCa concentrations rarely had ionized hypercalcemia, but iCa concentrations still should be evaluated in patients with tCa concentrations within the reference interval if there is clinical suspicion for calcium abnormalities.


| INTRODUCTION
Hypercalcemia in dogs often is an indicator of serious disease with the most common cause being malignant neoplasia such as lymphoma or anal gland adenocarcinoma. 1 Less common causes of hypercalcemia include hypoadrenocorticism, primary hyperparathyroidism, and hypervitaminosis D. 2 Persistent hypercalcemia, regardless of underlying cause, can have serious consequences including soft-tissue mineralization and subsequent renal failure. The serious nature of such diseases and complications requires an accurate way to detect hypercalcemia so that a diagnosis can be made and appropriate intervention can be instituted as soon as possible. Ionized calcium (iCa) is the biologically active form and is the most relevant to clinical practice. 2 However, iCa concentration measurements are not readily available in all practice settings, and serum total calcium (tCa) concentration often is used in its place which is of questionable acceptability.
A large study in dogs showed significant diagnostic discordance between tCa or adjusted (corrected) calcium and iCa status. This study evaluated a single tCa concentration threshold for defining hypercalcemia (>12 mg/dL), based on the upper limit of the laboratory reference interval. 3 The positive predictive value (PPV) at this threshold was relatively low at 68%. Dogs with ionized hypercalcemia can have tCa concentrations that are normal to very high, but tCa concentration medians have been reported to be well above reference intervals. 4 This suggests that there may be an optimal cutoff for tCa concentration that can more reliably predict hypercalcemia in dogs.
When exploring the optimal diagnostic threshold for tCa concentration, its utility for predicting ionized hypercalcemia historically has been evaluated separately in subpopulations of patients with and without azotemia. Previous reports have shown that tCa concentration was less predictive in dogs with chronic kidney disease as compared to dogs that were nonazotemic. 5 In fact, in dogs with kidney disease, using the upper limit of the laboratory reference interval as the tCa concentration threshold resulted in a PPV of only 27% for predicting ionized hypercalcemia. 3 Azotemia affects the diagnostic discordance between tCa and iCa concentrations because of increased complexed calcium, with phosphorus as the most commonly measured complexed ion. 6 Citrate and lactate also are important ions that complex with calcium, but they are difficult to measure in a clinical setting.
Evaluation of the discordance of tCa concentration to predict iCa concentration has not been separately evaluated in light of the presence hyperphosphatemia, despite the high frequency of hyperphosphatemia in azotemic dog populations. 7 Therefore, it is important to reinvestigate the use of tCa concentrations in populations categorized as nonhyperphosphatemic or hyperphosphatemic.
Our objective was to evaluate the accuracy of tCa concentrations for predicting ionized hypercalcemia across multiple tCa concentrations thresholds in dogs with and without hyperphosphatemia. Our hypothesis was that there is a threshold for tCa concentration which is strongly predictive of ionized hypercalcemia. We predicted that this threshold would be higher in patients with hyperphosphatemia compared to those without.

| Statistical analysis
Thresholds of tCa concentration were selected in 0.5 mg/dL increments between 11.0 and 15.0 mg/dL. Sensitivity, specificity, PPV, and negative predictive value (NPV) were calculated at each of these thresholds for the nonhyperphosphatemic and hyperphosphatemic groups; 95% confidence intervals (CIs) were calculated using the Clopper-Pearson exact method. To determine an optimal tCa concentration threshold for predicting hypercalcemia, the PPVs were evaluated to identify the lowest threshold ( whereas ionized hypercalcemia and normocalcemia were more common in the nonhyperphosphatemic group. The final diagnoses of all dogs with ionized hypercalcemia are presented in Table 2; the most common diagnoses were hypercalcemia of malignancy (53/112, 47%) and primary hyperparathyroidism  0.67-0.72; P < .001 and r = .67; 95% CI, 0.59-0.73; P < .001, respectively).
The PPV increased rapidly for tCa concentrations thresholds >10 mg/dL in the nonhyperphosphatemic group. The PPV for the hyperphosphatemic group increased steeply at 14 mg/dL, but only a single dog fell above that threshold ( Figure 2). The sensitivity, specificity, PPV, and NPV at each of the evaluated thresholds are presented in    . For the nonhyperphosphatemic group, the optimal tCa concentration threshold for predicting ionized hypercalcemia was 12.0 mg/dL with a PPV of 93% (95% CI, 84%-98%). The NPV for this threshold was 97%, and the sensitivity and specificity were 52 and 100%, respectively.
An optimal threshold was not determined for the hyperphosphatemic group because of low PPV, sensitivity, or both at all thresholds evaluated. Figure 3 shows data from individual dogs in relation to tCa, iCa, and the optimal threshold of the nonhyperphosphatemic group.
For the nonhyperphosphatemic group, we further evaluated the performance of the tCa concentration threshold of 12 mg/dL in dogs with (n = 302) and without azotemia (n = 1291). Data are presented in Table 5. The sensitivity of tCa concentration for predicting ionized hypercalcemia was lower in the nonazotemic group (44%) than in the azotemic group (74%; P = .008). The specificity was higher in the nonazotemic group (P = .02) but was excellent for both groups (≥99%). The PPV and NPV were not significantly different between groups. We further evaluated the population of non-  The PPV was prioritized when determining the optimal tCa concentration threshold rather than sensitivity. The PPV is the probability that a patient with a positive test result has the condition in this case the probability a dog with an increased tCa concentration has ionized hypercalcemia. The PPV and NPV are the most relevant statistics for a clinical scenario when an individual patient with unknown status is being screened. 8,9 In contrast, sensitivity and specificity are useful when comparing the performance of a test to a reference standard but provide little practical application to an individual patient with unknown disease status. The disadvantage of PPV is that it is affected by disease prevalence. When a disease is common, the PPV will increase, but a low prevalence will decrease the PPV. It is also important to note that a high PPV is not always necessary. A moderate threshold can be considered when the risk, convenience, and expense of follow-up screening are low. 8,9 We determined all 4 major metrics (PPV, NPV, sensitivity, and specificity) for multiple thresholds of tCa concentration in our study so that clinicians can make their own decisions about how to apply the results in their practice. We were unable to determine an optimal tCa concentration threshold for predicting ionized hypercalcemia in hyperphosphatemic dogs. Phosphorus is 1 of the commonly measured anions that complexes with calcium. Other anions such as citrate and lactate also could contribute to complexed calcium, but these analytes were not measured. Complexes of calcium can increase the tCa concentration independent of changes in iCa concentration. In these instances, tCa concentration can be increased above the reference interval, but iCa concentration can be low, normal, or high. Hyperphosphatemia is common in certain disease processes such as advanced chronic kidney disease. Ionized hypocalcemia is more common than ionized hypercalcemia in these instances because of decreased calcitriol synthesis and the complexing effect of hyperphosphatemia on iCa. 3,10 The landmark study on predicting iCa status from tCa concentration in dogs concluded that it is not acceptable to use tCa concentration to predict iCa concentration in dogs. 3 The tCa concentration threshold used in that study was identical to the optimal threshold identified in this study, 12.0 mg/dL, and yet, we conclude that tCa concentration can be an excellent predictor of iCa status. The reason for the opposing conclusions stems from 2 considerations in the application of tCa concentration for predicting iCa status: (1) whether it is being used to predict hypercalcemia or hypocalcemia and (2) whether the dog has a condition expected to cause discordance between tCa and iCa concentrations, namely azotemia or hyperphosphatemia. In the previous study, the PPV for 12.0 mg/dL was 68% for prediction of ionized hypercalcemia and 77% for prediction of ionized hypocalcemia in the overall study population. However, when performance was tested on dogs without chronic kidney disease, PPV increased to 90% for ionized hypercalcemia; PPV for ionized hypocalcemia remained low at 74%. Our study classified dogs based on phosphorus status rather than kidney disease, and we found that azotemia does not significantly decrease the PPV of tCa concentration for prediction of ionized hypercalcemia in nonhyperphosphatemic dogs.
A recent study in cats with and without azotemic chronic kidney disease showed similar performance of tCa concentration for predicting ionized hypercalcemia as in the present study. 11 The study of cats found moderate correlation between tCa and iCa concentrations in both the nonazotemic and azotemic groups (r = .6 for both).
They reported that tCa concentration above the reference range (11.8 mg/dL) was very specific for ionized hypercalcemia in both nonazotemic and azotemic cats (100 and 99.9%, respectively), but sensitivity was low at 9% in the nonazotemic group and 28% in the azotemic group. When the tCa calcium threshold was lowered to 10.5 mg/dL, which was below the upper limit of the laboratory reference range, it remained highly specific in the nonazotemic group (90%) with much improved sensitivity (95%). A lower tCa concentration threshold of 10.3 mg/dL also improved sensitivity in the azotemic group (72%) but decreased the specificity (71%). However, although these thresholds optimize the combination of sensitivity and specificity, they do not take into account the effects on PPV. Although not directly reported, PPV can be calculated from the data. The 10.5 mg/dL threshold had a PPV of 56%, and the 10.3 mg/dL threshold in the azotemic group had a PPV of 31%. These tCa concentration thresholds therefore have low accuracy for the prediction of ionized hypercalcemia, because half or more of the tCa concentration at or above the thresholds are false positives. This study in cats did not find a significant effect of serum phosphorous concentration on the risk of hypercalcemia. However, most cats included in the study had normal serum phosphorus concentrations.
In a clinical setting, mild increases in tCa concentration may be attributed to biologic variation, especially in clinically normal dogs, owing to its high index of individuality. 12 However, our current study shows that increases in tCa concentration are strongly predictive of ionized hypercalcemia. In our study population, the PPV was 93% for a tCa concentration >12.0 mg/dL and 80% for a tCa concentrarion Ionized hypercalcemia was rare in dogs without abnormally high tCa concentrations (NPV of 97% for the threshold of 12.0 mg/dL and 98% for the threshold of 11.5 mg/dL). However, it is important to recognize the effect that disease prevalence will have on NPV. In patients with clinical suspicion for hypercalcemia, the prevalence of true hypercalcemia could be much higher than in our study population, resulting in a lower NPV and thus more false negatives if tCa concentration is used to detect ionized hypercalcemia. This is particularly true in cases with mild increases in iCa concentration. Therefore, we again emphasize the need to measure an iCa concentration in patients with clinical suspicion for hypercalcemia even if the tCa concentration is within the reference range.
Another important finding from our study is that an increase in tCa concentration is a poor predictor of ionized hypercalcemia in hyperphosphatemic dogs (PPV = 19% for tCa concentration >12 mg/dL and 15% for tCa concentration >11.5 mg/dL). Many hyperphosphatemic dogs in our study had mild increases in tCa concentration without concurrent increases in iCa concentration, and an optimal threshold for tCa concentration could not be determined in the hyperphosphatemic group. All thresholds evaluated either had low PPV, low sensitivity, or both. The NPV was ≥97% for all thresholds, indicating again that ionized hypercalcemia is rare in dogs without increases in tCa concentration, but, as discussed above, this only applies to populations expected to have a similarly low prevalence of ionized hypercalcemia. One of the explanations for the lower proportion of hyperphosphatemic dogs with ionized hypercalcemia is the effect of the most common causes of ionized hypercalcemia (malignancy and primary hyperparathyroidism) on decreasing serum phosphorus concentration. As shown in Figure 3, there was a negative correlation between tCa and serum phosphorus concentrations in dogs with ionized hypercalcemia, and hypophosphatemia was common. Additionally, the use of this model has not been validated with variables measured from multiple laboratories with different reference intervals, which would be critical before application in a clinical setting because models often perform worse on external validation. 21 In our study, the optimal tCa concentration threshold in nonhyperphosphatemic dogs had similar performance to the aforementioned multivariable model with a PPV of 93%, specificity of 100%, and sensitivity of 52%. This suggests that a simple approach using only the tCa concentration is sufficient, as long as the dog is not hyperphosphatemic.
Our study had some limitations inherent to its retrospective design. Although most tCa and iCa concentration measurements were performed in <1 hour of each other, we allowed for up to 12 hours between measurements. Treatments that patients received during this time period that may have altered either measurement were not recorded. Two other limitations pertain to the application of results in different populations of dogs. As discussed above, we selected tCa concentration thresholds based on PPV rather than sensitivity and specificity, which is disease prevalence dependent. Therefore, the PPV for the threshold designated in our study may be different in other patient populations. All of the cases included in our study either were hospitalized patients or those referred for specialty evaluation, and the prevalence of hypercalcemia could be different in patients presented for outpatient evaluation or evaluation in a general practice setting. Finally, reference ranges are calculated based on specific methodology and are laboratory specific. The optimal tCa concentration threshold for prediction of ionized hypercalcemia could differ in other laboratories using other chemistry analyzers. Of note, the previous study in dogs used a different chemistry analyzer, but also obtained a high PPV (90%) for a tCa concentration threshold of 12.0 mg/dL in dogs that did not have chronic kidney disease. 3 In conclusion, we showed that even mildly increased tCa concentration (>12.0 mg/dL) are highly predictive of ionized hypercalcemia in adult dogs without hyperphosphatemia. There is no optimal threshold for patients with hyperphosphatemia. Ionized hypercalcemia is rare in dogs without increases in tCa concentration, regardless of phosphorous status. This information has important case management and research implications. Additional studies are needed to further validate these thresholds using other reference laboratories and patient populations.

CONFLICT OF INTEREST DECLARATION
Authors declare no conflict of interest.

OFF-LABEL ANTIMICROBIAL DECLARATION
Authors declare no off-label use of antimicrobials.

INSTITUTIONAL ANIMAL CARE AND USE COMMITTEE (IACUC) OR OTHER APPROVAL DECLARATION
Authors declare no IACUC or other approval was needed.

HUMAN ETHICS APPROVAL DECLARATION
Authors declare human ethics approval was not needed for this study.