Abstract
Background. This study aimed to investigate the demographic, clinical, and laboratory characteristics of sepsis patients who were admitted to our center during 2014–2020 and to employ cluster analysis, which is a type of machine learning, to identify distinct types of sepsis in Thai population. Methods. Demographic, clinical, laboratory, medicine, and source of infection data of patients admitted to medical wards of Siriraj Hospital (Bangkok, Thailand) during 2014–2020 were collected. Sepsis was diagnosed according to the Sepsis‐3 criteria. Nineteen demographic, clinical, and laboratory variables were analyzed using hierarchical clustering to identify sepsis subtypes. Results. Of 98,359 admissions, 18,030 (18.3%) had sepsis. Respiratory tract was the most common site of infection. The mean Sequential Organ Failure Assessment (SOFA) score was 4.21 ± 2.24, and the median serum lactate level was 2.7 mmol/L [range: 0.4–27.5]. Twenty percent of admissions required vasopressor. In‐hospital mortality was 19.6%. Ten sepsis subtypes were identified using hierarchical clustering. Three clusters (clusters L1–L3) were considered low risk, and seven clusters (clusters H1–H7) were considered high risk for in‐hospital mortality. Cluster H1 had prominent hematologic abnormalities. Clusters H3 and H5 had younger ages and significant hepatic dysfunction. Cluster H5 had multiple organ dysfunctions, and a higher proportion of cluster H5 patients required vasopressor, mechanical ventilation, and renal replacement therapy. Cluster H6 had more respiratory tract infection and acute respiratory failure and a lower SpO2/FiO2 value. Conclusions. Cluster analysis revealed 10 distinct subtypes of sepsis in Thai population. Furthermore, the study is needed to investigate the value of these sepsis subtypes in clinical practice.