Novel methods of predicting ionized calcium status from routine data in critical care: External validation in MIMIC-III

Background: Low ionized calcium (I-Ca) is prevalent and prognostic in critical care, but poorly detected by either total calcium (T-Ca) or albumin-corrected T-Ca (cT(Ca)). We recently derived models of I-Ca (Pred-I-Ca) and low I-Ca (Prob(HYPO)) in critical care that adjust T(Ca )for binding to albumin and small anions-represented by the anion gap's components. On internal validation, they outperformed cT(Ca) in diagnosing low I-Ca. Two other new anion gap-based models of I-Ca, derived in renal patients, have not been validated. This study tested the external validity of these 4 new models in the Medical Information Mart for Intensive Care III (MIMIC-III) database. Methods: We identified 4105 patients in MIMIC-III with I-Ca measured on an arterial blood gas panel within 20 min of chemistry panel measurements of T-Ca, albumin, sodium, chloride, and total carbon dioxide. The 4 models and cT(Ca )were assessed by their diagnostic discrimination for low I-Ca (< 1.10 mmol/l) and high I-Ca (> 1.32 mmol/ l), and by the agreement between predicted and observed values. Results: Pred-I-Ca and Prob(HYPO) had the best discrimination and agreement. Conclusions: Pred-I-Ca and Prob(HYPO) were externally validated in MIMIC-III. They can help clinicians efficiently decide when to order direct I-Ca testing in critical care.