Background Acute fatty liver of pregnancy (AFLP) is a rare, life-threatening obstetric emergency. Early prediction of its severe postpartum complications remains a significant clinical challenge, optimal risk stratification tools have not been established.Methods This retrospective cohort study included 106 patients with AFLP admitted to a tertiary care hospital between 2011 and 2020. Severity scores (MELD, APACHE II, SOFA) were calculated using data from 24 h before delivery or within 24 h of ICU admission. Predictive performance for six major complications--acute kidney injury (AKI), disseminated intravascular coagulation (DIC), sepsis, multiple organ dysfunction syndrome (MODS), postpartum hemorrhage (PPH), and acute liver failure (ALF) was assessed using logistic regression, with discrimination evaluated by area under the receiver operating characteristic curve (AUC-ROC) and clinical utility by decision curve analysis (DCA).Results Among 106 patients, the most common complication was AKI (71/106, 67.0%), followed by DIC (30/106, 28.3%), MODS (30/106, 28.3%), PPH (29/106, 27.4%), sepsis (28/106, 26.4%), and ALF (24/106, 22.6%). The combined MELD-SOFA score was the best predictor for DIC (AUC 0.764, 95% CI 0.662-0.856), MODS (AUC 0.757, 95% CI 0.655-0.855), and ALF (AUC 0.797, 95% CI 0.692-0.886), and also demonstrated strong predictive value for AKI (AUC 0.882,95% CI 0.803-0.943). The combined MELD-APACHE II score demonstrated excellent discrimination for AKI (AUC 0.899, 95% CI 0.815-0.967). APACHE II alone best predicted sepsis (AUC 0.748, 95% CI 0.643-0.848). All models performed poorly for PPH (AUC Conclusions A complication-specific approach to score selection optimizes risk stratification in AFLP. While the combined MELD-SOFA score demonstrates broad applicability by showing high effectiveness for predicting DIC, MODS, ALF and AKI, optimal prediction for specific complications requires tailored combinations. the MELD-APACHE II score is superior for AKI, and APACHE II alone for sepsis. These findings support the targeted use of combined scoring models to guide early intervention and improve maternal outcomes.