Kidney Week

Abstract: PO1187

Prediction of Ionized Hypocalcemia and Hypercalcemia: External Validation of a Novel Model

Session Information

• Mineral Homeostasis and Acid-Base Disorders: Clinical
  November 04, 2021 | Location: On-Demand, Virtual Only
  Abstract Time: 10:00 AM - 12:00 PM

Category: Fluid, Electrolyte, and Acid-Base Disorders

• 902 Fluid, Electrolyte, and Acid-Base Disorders: Clinical

Authors

• Ouyang, Jie, SUNY Downstate Health Sciences University, New York City, New York, United States

Jie Ouyang, MD

• Melaku, Yohannes, SUNY Downstate Health Sciences University, New York City, New York, United States

Yohannes Melaku,

• Puri, Isha, SUNY Downstate Health Sciences University, New York City, New York, United States

Isha Puri,

• Yap, Ernie, SUNY Downstate Health Sciences University, New York City, New York, United States

Ernie Yap,

• Goldwasser, Philip, VA New York Harbor, Brooklyn, New York, United States

Philip Goldwasser,

Background

The popular adjustment of serum total calcium (tCa) for albumin (Alb) yields a corrected value (cCa) that doesn’t detect abnormal ionized calcium (iCa) well in critical care patients (pts), possibly because it ignores the fraction of tCa complexed by small anions. To account for such anions, we derived a model that estimates iCa (iCa_EST) by adjusting tCa for Alb and the anion gap’s 3 components, Na, Cl, and tCO₂. It was far better than cCa in detecting low iCa (↓iCa) on internal validation (Yap, JALM 2020). In this study, we externally validated iCa_EST in a large, publicly available critical care database.

Methods

From the MIMIC III v1.4 database, we paired chemistry panel tCa (mg/dL), Alb (g/dL), Na, Cl, and tCO₂ values with gas panel iCa values (ref range: 1.12-1.32 mM) measured up to 20 min. apart. Limiting each pt. to the most closely-timed pair left 4105 pairs (median:10 min apart). We calculated cCa (tCa +0.8×[4-Alb]) and iCa_EST (0.219 +0.091×tCa -0.034×Alb -0.0042×Na +0.0073×Cl +0.0047×tCO₂) and compared their ROC curves (area±SE) for detecting ↓iCa (iCa<1.10; rate=33.1%), and high iCa (↑iCa) (iCa>1.32; rate=3.8%).

Results

ICa_EST was better than cCa by ROC analysis for both ↓iCa (0.834±0.007 vs 0.752±0.008, p<10^-300) and ↑iCa (0.975±0.004 vs 0.963±0.006, p<.0006). The table compares the sensitivity and specificity (SENS/SPEC) and positive and negative predictive values (PPV/NPV) of iCa_EST and cCa at similar cutoffs. ICa_EST overestimated iCa by 0.04 mM (1.17±0.002 vs 1.13±0.002, p<10^-166), a bias that was fairly consistent across the full prediction range.

Conclusion

The iCa_EST model is superior to cCa in ranking critically ill pts for both ↓iCa and ↑iCa. It can help clinicians decide when to directly measure iCa. ICa_EST overestimated iCa but applying a local correction of -0.04 would make its absolute predictions accurate, on average, in the MIMIC setting.