Kidney Week

Abstract: TH-PO0402

Development and Validation of Deep-Learning Model for Diagnosing Systemic Acidemia from Electrocardiography

Session Information

• Fluid, Electrolyte, and Acid-Base Disorders: Clinical - 1
  November 06, 2025 | Location: Exhibit Hall, Convention Center
  Abstract Time: 10:00 AM - 12:00 PM

Category: Fluid, Electrolytes, and Acid-Base Disorders

• 1102 Fluid, Electrolyte, and Acid-Base Disorders: Clinical

Authors

• Park, Won Ho, Hallym University Dongtan Sacred Heart Hospital, Hwaseong-si, Gyeonggi-do, Korea (the Republic of)

Won Ho Park, DrMed

• Park, Minje, VUNO Inc., Seoul, Korea (the Republic of)

Minje Park

• Yun, Giae, Hallym University Dongtan Sacred Heart Hospital, Hwaseong-si, Gyeonggi-do, Korea (the Republic of)

Giae Yun, MD

• An, Jung Nam, Hallym University Sacred Heart Hospital, Anyang-si, Gyeonggi-do, Korea (the Republic of)

Jung Nam An, MD, PhD

• Kim, Do Hyoung, Hallym University Kangnam Sacred Heart Hospital, Yeongdeungpo-gu, Seoul, Korea (the Republic of)

Do Hyoung Kim, MD, DrMed, PhD

• Jeon, Hee Jung, Kangdong Sacred Heart Hospital, Gangdong-gu, Seoul, Korea (the Republic of)

Hee Jung Jeon, MD, PhD

• Kim, Sung Gyun, Hallym University Sacred Heart Hospital, Anyang-si, Gyeonggi-do, Korea (the Republic of)

Sung Gyun Kim, MD, PhD

• Baek, Seon Ha, Hallym University Dongtan Sacred Heart Hospital, Hwaseong-si, Gyeonggi-do, Korea (the Republic of)

Seon Ha Baek, MD, PhD

Background

Systemic acidemia impairs cardiovascular function and is usually diagnosed via arterial blood gas, potentially delaying intervention. We developed a deep learning model using electrocardiogram (ECG) data for rapid, non-invasive diagnosis.

Methods

We developed and validated deep learning models to detect systemic acidemia, categorized by severity (mild: pH < 7.35, moderate: pH < 7.30, severe: pH < 7.20), using 12-lead ECGs from the Medical Information Mart for Intensive Care (MIMIC)-IV database. The training, validation, and test datasets included 56,249, 1,429, and 1,446 ECG-pH samples, respectively. To investigate variations in the model's efficacy depending on the type of systemic acidemia, we carried out a subgroup analysis according to pCO₂ and HCO₃^- levels.

Results

The models achieved high performance in detecting systemic acidemia, with the area under the receiver operating characteristic curves (AUCs) of 0.70, 0.73, and 0.82 for mild, moderate, and severe acidemia, respectively, in the testing cohort (Figure 1). No statistically significant differences in AUC were observed between the group with pCO₂ ≤ 45 mmHg and HCO₃^- < 22 mEq/L and the group with pCO₂ > 45 mmHg and HCO₃^- ≥ 22 mEq/L, with p-values (DeLong’s method) of 0.07, 0.07, and 0.25 for mild, moderate, and severe acidemia, respectively. The group predicted to be positive by the severe acidemia model showed a lower 30-day survival rate compared to the negative group (p<0.05), supporting the clinical efficacy.

Conclusion

Deep learning offers a non-invasive method for diagnosing systemic acidemia in critically ill patients, enabling earlier intervention and improved patient care.

Figure 1. Comparison of receiver operating characteristic curves.

Digital Object Identifier (DOI)

• doi: 10.1681/ASN.2025xrej8nqm