Development of a new disseminated intravascular coagulation score tailored for patients with chronic liver disease: a prospective observational study.

BACKGROUND

Diagnosing disseminated intravascular coagulation (DIC) in patients with chronic liver disease is challenging because cirrhosis-related hemostatic abnormalities mimic DIC laboratory features, leading to frequent overdiagnosis using conventional DIC scoring systems.

OBJECTIVES

This study aimed to evaluate the performance of existing DIC scores in patients with cirrhosis and develop a tailored diagnostic prediction tool.

METHODS

We prospectively enrolled 80 consecutive patients with cirrhosis admitted to a single intensive care unit. Clinical manifestations of DIC, conventional DIC scores, routine hemostatic parameters, and advanced coagulation biomarkers were assessed. Machine learning techniques combining random forest, logistic regression, and extreme gradient boosting with the Synthetic Minority Oversampling Technique were used to develop a DIC prediction model using 4 routinely available parameters: chronic liver failure-sequential organ failure assessment score, platelet count, fibrinogen, and D-dimer levels.

RESULTS

Conventional DIC scores (the International Society on Thrombosis and Haemostasis, the Japanese Association for Acute Medicine, and the Japanese Ministry of Health, Labour, and Welfare) were positive in >50% of patients at study inclusion despite only 1.3% having clinical DIC manifestations, demonstrating substantial overdiagnosis. During follow-up, 10 patients (12.5%) developed clinical DIC with 90% mortality, while positive DIC scores did not predict outcomes. The newly developed ensemble prediction model achieved 100% sensitivity and 87% specificity, accurately discriminating clinical DIC from cirrhosis-related hemostatic abnormalities. Coagulation activation marker levels were elevated universally, while progressive fibrin monomer level depletion distinguished overt DIC from compensated hemostatic imbalance.

CONCLUSION

The proposed prediction tool showed excellent internal performance in identifying DIC risk in patients with cirrhosis using readily available parameters, and its implementation in a freely accessible web-based application may facilitate future clinical use after external validation.