Sodium Augmentation Prediction Model in Neurocritically Ill Patients

Hypertonic saline (HTS) is commonly used to treat intracranial hypertension in neurocritically ill patients with cerebral edema, creating an osmotic shift to pull fluid from the intracranial space. However, excessive hypernatremia and hyperchloremia can result in serious adverse outcomes including acute kidney injury (AKI), seizures, and coma. Currently, clinicians lack tools to accurately predict the effects of HTS on serum sodium when augmenting sodium beyond eunatremia. The purpose of this study was to create a model to predict effects of HTS on serum sodium levels in neurocritically ill patients.

This was a single center retrospective study of neurocritically ill adults (≥18 years of age) who received ≥2 boluses and/or continuous infusion of HTS at a level one trauma center. Exclusion criteria included vulnerable populations, preexisting hyponatremia (baseline sodium ≤135 mEq/L), or no sodium levels obtained following HTS administration. The primary outcome was the generation of a serum sodium prediction model using a Random Forest Model that was tuned to hyperparameters via grid search and used 5‑fold group‑vfold cross‑validation by patient to prevent leakage. Performance was assessed on a held‑out test set of patients using root mean squared error (RMSE), mean absolute error (MAE), R², calibration (slope/intercept), and residual diagnostics. Secondary safety outcomes included rates of hyperchloremia (>115 mEq/L), acute kidney injury, and the need for delayed neurosurgical intervention or renal replacement therapy.

Among 416 patients identified for the study, 7 were excluded for having no serum sodium levels, and a total of 409 patients with 7620 serum sodium levels were included. The first iteration of the model achieved RMSE = 0.573 mmol/L (95% CI: 0.39-0.73), MAE = 0.105 mmol/L (95% CI: 0.08-0.13), R² = 0.994 (95% CI 0.991-9.997) on the test set. Most recent sodium value, amount of sodium (in mEqs) administered over the previous 24 hours, and time since last sodium lab were identified as the parameters with the most effect on the model's outcomes.

This study developed a sodium prediction model to predict the effects of hypertonic saline on serum sodium levels for sodium augmentation in neurocritically ill patients.