Abstract: Objective To verify the accuracy and cost of the value of serum sodium fluctuation in predicting adverse clinical outcomes, and to propose a warning value of serum sodium fluctuations for early warning of in-hospital death and summarize the methodology for setting warning values. Methods This investigation covered all patients admitted to Peking Union Medical College Hospital from January 1, 2014 to December 31, 2014. Patients with normal serum sodium at the time of admission and older than 18 years old were included. Patients without serum sodium retested during the hospitalization were excluded. Clinical characteristics including gender, age, primary diagnosis, and causes of death were documented and laboratory examinations were collected including all serum sodium values during hospitalization. The highest and lowest values among all measurements of serum sodium during hospitalization were identified. The difference between the highest and lowest serum sodium value was defined as the fluctuation value of serum sodium level.The area under the curve (AUC) of sodium fluctuations predicting in-hospital death was calculated. The specificities and sensitivities were calculated with the cut-off points of 10, 12, 14, 16, 18, 20, 22, and 24 mmol/L. A warning related to death in the hospital was defined as a true warning, and a warning unrelated to death in the hospital was defined as a false warning. The number of false warnings was defined as the cost of warning.The ratio of the true warnings to the total warnings was defined as the efficiency of warning. An appropriate cut-off point was selected for combined accuracy, cost, and efficiency of warning. Results In total, 33 323 hospitalization records met the inclusion criteria, of which 191 patients (0.57%) died. 148 records of death were analyzed. Among these patients, 53 died from tumor (35.6%), 52 died from infection (34.9%), 46 died from respiratory failure (30.9%), and 26 died from heart failure (17.4%), 6 died from sudden death (4.0%), 5 died from cerebral hemorrhage (3.4%), 6 died from bleeding from other parts (4.0%), and 1 died from cerebral herniation (0.7%).There were 75 patients (50.7%) experienced serum sodium fluctuations within 10 days before death. The average serum sodium fluctuation was (13.36±8.62)mmol/L, and the highest was 35 mmol/L. Serum sodium fluctuation predicted in-hospital death with an AUC of 85.2% (95% CI:81.8%-88.5%). Predicting in-hospital death with a serum sodium fluctuation greater than 20 mmol/L had a specificity of 99.0% and a sensitivity of 28.8%. If serum sodium fluctuation greater than 20 mmol/L was set as a warning value, 28.8% of hospital deaths would be identified, the number of warnings would be 397, the number of true warnings would be 55, the cost of warning would be 342, and the efficiency of warning would be 13.9%. Conclusions Serum sodium fluctuation greater than 20 mmol/L could be used as a biomarker value for early warning of in-hospital death with high accuracy and low management cost. Serum sodium fluctuations greater than 20 mmol/L is a warning value point and can be potentially used as a predicting warning to suppot the identification of critically ill patients and thus to improve the management of inpatients at high risk of death.

Key words: electrolyte disorder, serum sodium fluctuation, hospital management, early warning