Neuroendocrinology and Pituitary

SUN-005 - A Machine Learning Approach to Prevent Overcorrection in Severe Hyponatremia

Sunday, June 14, 2026

9:00 AM - 4:00 PM CT

Location: ENDOExpo; Poster Floor

Abstract Poster Presenter(s)

Satoshi Naito, MD

Nagoya, JAPAN

Senior Author/Primary Investigator(s)

DH

Daisuke Hagiwara, MD, PhD

Department of Endocrinology and Diabetes, Nagoya University Graduate School of Medicine.
Nagoya, Aichi, Japan

Co-Author(s)

MK

Masaki Kitamura, MD

Department of Endocrinology and Diabetes, Nagoya University Graduate School of Medicine.
Nagoya, Aichi, Japan
RA

Ryosei Ashida, MD

Department of Endocrinology and Diabetes, Nagoya University Graduate School of Medicine.
Nagoya, Aichi, Japan
YK

Yohei Kawaguchi, MD, PhD

Department of Endocrinology and Diabetes, Nagoya University Graduate School of Medicine.
Nagoya, Aichi, Japan
TM

Takashi Miyata, MD, PhD

Department of Endocrinology and Diabetes, Nagoya University Graduate School of Medicine.
Nagoya, Aichi, Japan
YA

Yoshinori Azuma, MD, PhD

Department of Endocrinology and Diabetes, Japanese Red Cross Aichi Medical Center Nagoya Daini Hospital.
Nagoya, Aichi, Japan
TT

Taku Tsunekawa, MD, PhD

Department of Endocrinology and Diabetes, Ichinomiya Municipal Hospital.
Ichinomiya, Aichi, Japan
JK

Junki Kurimoto, MD, PhD

Department of Endocrinology and Diabetes, Konan Kosei Hospital.
Konan, Aichi, Japan
YA

Yoh Ariyoshi, MD, PhD

Department of Endocrinology and Diabetes, Konan Kosei Hospital.
Konan, Aichi, Japan
TK

Tamaki Kinoshita, MD, PhD

Department of Endocrinology and Diabetes, Nagoya University Graduate School of Medicine.
Nagoya, Aichi, Japan
SO

Shintaro Oyama, MD, PhD

Innovative Research Center for Preventive Medical Engineering, Nagoya University.
Nagoya, Aichi, Japan

Senior Author/Primary Investigator(s)

HA

Hiroshi Arima, MD, PhD

Professor
Department of Endocrinology and Diabetes, Nagoya University Graduate School of Medicine.
Nagoya, Aichi, Japan

Disclosure(s):

Satoshi Naito, MD: No financial relationships to disclose

Background
Severe hyponatremia requires urgent correction; however, overly rapid correction increases the risk of osmotic demyelination syndrome. Japanese guidelines recommend limiting serum sodium correction to ≤8-10 mEq/L within 24 hours and ≤18 mEq/L within 48 hours. Despite frequent monitoring, unanticipated rapid increases still occur. We developed a machine learning model to predict 6-hour serum sodium levels in an exploratory retrospective study and subsequently initiated a prospective study to assess its generalizability.
Methods
In the exploratory phase, we retrospectively reviewed medical records from April 2015 to March 2020. The prediction model was trained using 342 datasets from 74 patients with severe hyponatremia (male/female: 26/48) at an affiliated hospital and externally validated using 146 datasets from 15 patients (male/female: 5/10) at our hospital. Seven routinely measured clinical parameters—water intake, sodium and potassium supplementation, urine output, and serum sodium, potassium, chloride levels—were selected as predictors. The target variable was serum sodium 6 hours after the input window. Three algorithms (support vector regression, linear regression, and regression tree) were evaluated. In the confirmatory phase, a multicenter prospective interventional study is ongoing. Eligible participants are adults (≥18 years) with serum sodium < 130 mEq/L; exclusions include urinary sodium < 20 mEq/L and high-dose corticosteroid use. Clinicians choose fluid type and rate based on a matrix displaying predicted 6-hour sodium values under various treatment patterns. The primary endpoint is the proportion of participants in which the actual 6-hour serum sodium falls within ±2 mEq/L of the predicted value. The target sample size is 24 (80% power assuming 80% accuracy).
Results
In the exploratory phase, support vector regression demonstrated the best performance (internal validation: R² = 0.97, RMSE = 1.11; external validation: R² = 0.92, RMSE = 1.62). In the confirmatory phase, 19 of the planned 24 participants have been enrolled; analysis is ongoing.
Conclusions
This machine learning model using seven routinely available clinical predictors provides accurate 6-hour serum sodium predictions. The ongoing multicenter prospective study will determine its generalizability and may support safer management of hyponatremia by reducing the risk of overcorrection.

*Unless otherwise noted, all abstracts presented at ENDO must not be released to the press or the public until the date and time of presentation. For oral presentations, the abstracts are embargoed until the session begins. The Endocrine Society reserves the right to lift the embargo on specific abstracts that are selected for promotion prior to or during ENDO.*