The multivariate predictive model to estimate ionized calcium concentration from serum biochemical results in dogs: External validation

Abstract

Background Predicted ionized calcium (piCa) can be calculated from routine biochemistry variables using a recently developed predictive model in dogs. However, it has not been evaluated with variables measured from multiple laboratories. Objectives We aimed to (a) externally validate piCa in dogs where biochemistry results were obtained from different analyzers, and (b) compare the diagnostic performances of piCa and total calcium (tCa). Methods A cross‐sectional multicentric study on 138 dogs from three different hospitals was performed. The sensitivity (Sen), specificity (Spe), positive (PPV) and negative predictive values (NPV), and diagnostic discordance of piCa and tCa were calculated using logistic regression for ionized hypercalcemia and hypocalcemia. Diagnostic performance fluctuations across hospitals were also assessed. Results For ionized hypercalcemia, the Sen (81.8%), Spe (96.1%), PPV (69.2%), NPV (97.7%), and diagnostic discordance (5.1%) of piCa were not significantly different among hospitals or from those of tCa. For ionized hypocalcemia, the Sen (range: 9.7%‐53.8%) and Spe (range: 95.6%‐99.6%) of piCa and tCa (Sen range: 16.2%‐87.8%; Spe range: 58.3%‐98.1%) varied across hospitals, although to a lesser extent for piCa. The diagnostic discordances of piCa (20.3%) and tCa (25.4%) were close. The prediction interval (PI) of piCa demonstrated high Sen to screen for ionized hypercalcemia (100%) and hypocalcemia (range: 75%‐93.3%), and high Spe to diagnose ionized hypercalcemia and hypocalcemia (100% for both). Conclusions These results support the external validation of piCa in dogs. Its PI represents a notable advantage over tCa to help clinicians explore calcium‐related disorders when ionized calcium cannot be readily measured.

Diagnostic performance of predicted ionized calcium in dogs with total hypercalcemia and total hypocalcemia

“…10 The performance of predicted ionized calcium (piCa) has been internally and externally validated in three different veterinary hospitals, using different biochemistry analyzers. 10,11 Overall, piCa had a higher PPV and positive diagnostic likelihood ratio (PDLR) than tCa for both iHyperCa (PPV: 69%-90%; PDLR: 20-157) and iHypoCa (PPV: 71%-89%; PDLR: 14-20). 10,11 The MARS model not only provides a piCa value but also a prediction interval (PI).…”

“…10,11 Overall, piCa had a higher PPV and positive diagnostic likelihood ratio (PDLR) than tCa for both iHyperCa (PPV: 69%-90%; PDLR: 20-157) and iHypoCa (PPV: 71%-89%; PDLR: 14-20). 10,11 The MARS model not only provides a piCa value but also a prediction interval (PI). When the PI was taken into account, PPV increased to 100% when the lower limit was >1.37 mmol/L and the upper limit was <1.11 mmol/L to predict iHy-perCa and iHypoCa, respectively.…”

“…Positive predictive values (PPVs) of increased tCa to predict iHyperCa have varied between 19% and 93% in previous reports, with an average value around 67%. [9][10][11][12] Analogously, the PPV of decreased tCa to predict iHypoCa has varied between 45% and 87%, with an average value around 60%. [9][10][11] Based on these data, approximately one third of dogs with decreased or increased tCa values are improperly classified as hypocalcemic or hypercalcemic, respectively.…”

“…[9][10][11][12] Analogously, the PPV of decreased tCa to predict iHypoCa has varied between 45% and 87%, with an average value around 60%. [9][10][11] Based on these data, approximately one third of dogs with decreased or increased tCa values are improperly classified as hypocalcemic or hypercalcemic, respectively.…”

“…When the PI was taken into account, PPV increased to 100% when the lower limit was >1.37 mmol/L and the upper limit was <1.11 mmol/L to predict iHy-perCa and iHypoCa, respectively. 10,11 However, these performances were assessed in a general population of dogs with a prevalence of iHyperCa and iHypoCa around 5% (0% to 21%) and 20% (15% to 28%), respectively. 10,11 Thus, the performance of piCa and its PI in a population of dogs with abnormal tCa, which implies a higher prevalence of iHyperCa and iHypoCa, is still unknown.…”