Using Machine Learning to Explore the Impact of Borderline Hyponatremia on Adverse Outcomes in Heart Failure Patients: A Data Mining Approach

An association between slightly low serum sodium levels within the normal range and increased mortality risk in heart failure patients has been observed. However, the precise borderline level of serum sodium in heart failure patients remains undetermined, and the impact of borderline hyponatremia on heart failure outcomes remains unclear. This study aims to identify the threshold for borderline hyponatremia and assess the relationship between borderline hyponatremia and adverse outcomes in hospitalized heart failure patients. The adverse events, encompassing all-cause deaths and cardiac transplantation. This retrospective cohort study included 1,249 heart failure patients treated at the Heart Failure Center of Fuwai Hospital, Beijing, China, between 2009 and 2013, of which the incidence of adverse events was 23.1%. The optimal threshold for borderline hyponatremia were identified using Youden index and machine learning. Furthermore, we used Cox proportional risk analysis, smooth curve fitting, and Kaplan-Meier survival curve to assess the relationship between sodium levels and the incidence of adverse outcomes in patients with heart failure while considering potential confounding factors such as age and gender. We created three different models using univariate and multivariate Cox proportional-hazard regression models, including an unadjusted model, a minimally adjusted model, and a fully adjusted model. For patients with borderline hyponatremia, the HRs and 95%CIs for their unadjusted, minimally adjusted, and fully adjusted models were 1.90 (1.46, 2.48), P=<0.0001, 1.86 (1.43, 2.44), P<0.0001, and 1.37 (1.03, 1.81), P=0.0280. There was a significant association between borderline hyponatremia and adverse outcome.