The relationship between clinical subtypes, prognosis, and treatment in ICU patients with acute cholangitis using unsupervised machine learning methods
Abstract Background Acute cholangitis (AC) presents with significant clinical heterogeneity, and existing severity classifications have limited prognostic value in critically ill patients. Subtypes of AC in critically ill patients have not been investigated. Objective The study aimed to offer a nove...
Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
BMC
2025-08-01
|
| Series: | BMC Infectious Diseases |
| Subjects: | |
| Online Access: | https://doi.org/10.1186/s12879-025-11346-y |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1849345324346245120 |
|---|---|
| author | Yuanhui Sheng Shan Xu Shu Zhang Dan Zhang |
| author_facet | Yuanhui Sheng Shan Xu Shu Zhang Dan Zhang |
| author_sort | Yuanhui Sheng |
| collection | DOAJ |
| description | Abstract Background Acute cholangitis (AC) presents with significant clinical heterogeneity, and existing severity classifications have limited prognostic value in critically ill patients. Subtypes of AC in critically ill patients have not been investigated. Objective The study aimed to offer a novel approach to identify clinical subtypes and improve individualized risk assessment and treatment strategies using an unsupervised analysis. Methods We conducted a retrospective analysis of ICU patients with AC from the Medical Information Mart for Intensive Care-IV (MIMIC-IV) database. K-means clustering was applied to 24 routinely available clinical variables from the first 24 h of ICU admission to identify clinical subtypes. The primary outcome was 28-day all-cause mortality. Multivariable Cox regression was used to assess associations between subtypes, mortality, and biliary drainage strategies. Furthermore, a simplified model using the top five SHapley Additive exPlanations (SHAP) ranked variables (hemoglobin, RDW, albumin, CCI, and sodium) was developed and subsequently validated in an external cohort from MIMIC-III. Results Two distinct subtypes were identified: a mild-inflammatory subtype (n = 451) and an inflammatory-dysfunctional subtype (n = 348). The inflammatory-dysfunctional subtype was characterized by higher levels of RDW, ALP, bilirubin, creatinine, and coagulopathy markers, and exhibited significantly higher 28-day mortality (30.17% vs. 5.32%, p < 0.001). This subtype remained an independent predictor of mortality after multivariable adjustment (HR = 2.38,95%CI:1.42–3.9, p = 0.001). Within the inflammatory-dysfunctional subtype, ERCP was associated with lower mortality (HR = 0.56, p = 0.01), whereas PTCD was associated with higher mortality (HR = 1.63, p = 0.031), potentially reflecting underlying disease severity. The simplified model retained strong discriminative performance (AUC = 0.87) and was successfully validated in the external cohort, confirming the reproducibility and prognostic relevance of the subtypes. Conclusions We identified and externally validated two clinically meaningful AC subtypes with distinct prognoses. A simplified model using five readily available variables facilitates clinical application and can support more individualized treatment approaches. These data-driven subtypes offer additional prognostic discrimination beyond the Tokyo Guidelines, serving as a valuable complement to existing severity classifications for guiding precision management of ICU patients with AC. |
| format | Article |
| id | doaj-art-6892084edf65417cbea43d28f0d6b771 |
| institution | Kabale University |
| issn | 1471-2334 |
| language | English |
| publishDate | 2025-08-01 |
| publisher | BMC |
| record_format | Article |
| series | BMC Infectious Diseases |
| spelling | doaj-art-6892084edf65417cbea43d28f0d6b7712025-08-20T03:42:30ZengBMCBMC Infectious Diseases1471-23342025-08-0125111310.1186/s12879-025-11346-yThe relationship between clinical subtypes, prognosis, and treatment in ICU patients with acute cholangitis using unsupervised machine learning methodsYuanhui Sheng0Shan Xu1Shu Zhang2Dan Zhang3Emergency & Department of Critical Care Medicine, The First Affiliated Hospital of Chongqing Medical UniversityEmergency & Department of Critical Care Medicine, The First Affiliated Hospital of Chongqing Medical UniversityNursing Department, the First Affiliated Hospital of Chongqing Medical UniversityEmergency & Department of Critical Care Medicine, The First Affiliated Hospital of Chongqing Medical UniversityAbstract Background Acute cholangitis (AC) presents with significant clinical heterogeneity, and existing severity classifications have limited prognostic value in critically ill patients. Subtypes of AC in critically ill patients have not been investigated. Objective The study aimed to offer a novel approach to identify clinical subtypes and improve individualized risk assessment and treatment strategies using an unsupervised analysis. Methods We conducted a retrospective analysis of ICU patients with AC from the Medical Information Mart for Intensive Care-IV (MIMIC-IV) database. K-means clustering was applied to 24 routinely available clinical variables from the first 24 h of ICU admission to identify clinical subtypes. The primary outcome was 28-day all-cause mortality. Multivariable Cox regression was used to assess associations between subtypes, mortality, and biliary drainage strategies. Furthermore, a simplified model using the top five SHapley Additive exPlanations (SHAP) ranked variables (hemoglobin, RDW, albumin, CCI, and sodium) was developed and subsequently validated in an external cohort from MIMIC-III. Results Two distinct subtypes were identified: a mild-inflammatory subtype (n = 451) and an inflammatory-dysfunctional subtype (n = 348). The inflammatory-dysfunctional subtype was characterized by higher levels of RDW, ALP, bilirubin, creatinine, and coagulopathy markers, and exhibited significantly higher 28-day mortality (30.17% vs. 5.32%, p < 0.001). This subtype remained an independent predictor of mortality after multivariable adjustment (HR = 2.38,95%CI:1.42–3.9, p = 0.001). Within the inflammatory-dysfunctional subtype, ERCP was associated with lower mortality (HR = 0.56, p = 0.01), whereas PTCD was associated with higher mortality (HR = 1.63, p = 0.031), potentially reflecting underlying disease severity. The simplified model retained strong discriminative performance (AUC = 0.87) and was successfully validated in the external cohort, confirming the reproducibility and prognostic relevance of the subtypes. Conclusions We identified and externally validated two clinically meaningful AC subtypes with distinct prognoses. A simplified model using five readily available variables facilitates clinical application and can support more individualized treatment approaches. These data-driven subtypes offer additional prognostic discrimination beyond the Tokyo Guidelines, serving as a valuable complement to existing severity classifications for guiding precision management of ICU patients with AC.https://doi.org/10.1186/s12879-025-11346-yAcute cholangitisK-means clusteringSubtypePrognosis |
| spellingShingle | Yuanhui Sheng Shan Xu Shu Zhang Dan Zhang The relationship between clinical subtypes, prognosis, and treatment in ICU patients with acute cholangitis using unsupervised machine learning methods BMC Infectious Diseases Acute cholangitis K-means clustering Subtype Prognosis |
| title | The relationship between clinical subtypes, prognosis, and treatment in ICU patients with acute cholangitis using unsupervised machine learning methods |
| title_full | The relationship between clinical subtypes, prognosis, and treatment in ICU patients with acute cholangitis using unsupervised machine learning methods |
| title_fullStr | The relationship between clinical subtypes, prognosis, and treatment in ICU patients with acute cholangitis using unsupervised machine learning methods |
| title_full_unstemmed | The relationship between clinical subtypes, prognosis, and treatment in ICU patients with acute cholangitis using unsupervised machine learning methods |
| title_short | The relationship between clinical subtypes, prognosis, and treatment in ICU patients with acute cholangitis using unsupervised machine learning methods |
| title_sort | relationship between clinical subtypes prognosis and treatment in icu patients with acute cholangitis using unsupervised machine learning methods |
| topic | Acute cholangitis K-means clustering Subtype Prognosis |
| url | https://doi.org/10.1186/s12879-025-11346-y |
| work_keys_str_mv | AT yuanhuisheng therelationshipbetweenclinicalsubtypesprognosisandtreatmentinicupatientswithacutecholangitisusingunsupervisedmachinelearningmethods AT shanxu therelationshipbetweenclinicalsubtypesprognosisandtreatmentinicupatientswithacutecholangitisusingunsupervisedmachinelearningmethods AT shuzhang therelationshipbetweenclinicalsubtypesprognosisandtreatmentinicupatientswithacutecholangitisusingunsupervisedmachinelearningmethods AT danzhang therelationshipbetweenclinicalsubtypesprognosisandtreatmentinicupatientswithacutecholangitisusingunsupervisedmachinelearningmethods AT yuanhuisheng relationshipbetweenclinicalsubtypesprognosisandtreatmentinicupatientswithacutecholangitisusingunsupervisedmachinelearningmethods AT shanxu relationshipbetweenclinicalsubtypesprognosisandtreatmentinicupatientswithacutecholangitisusingunsupervisedmachinelearningmethods AT shuzhang relationshipbetweenclinicalsubtypesprognosisandtreatmentinicupatientswithacutecholangitisusingunsupervisedmachinelearningmethods AT danzhang relationshipbetweenclinicalsubtypesprognosisandtreatmentinicupatientswithacutecholangitisusingunsupervisedmachinelearningmethods |