Purpose: To determine the association of skeletal muscle CT body composition parameters derived from a fully automated machine learning algorithm with 90-day mortality after TIPS and to assess its predictive performance as a complement to MELD score for mortality risk prediction.
Materials and Methods: In this retrospective study, a total of 122 patients met the inclusion criteria for enrollment: 1) Have undergone TIPS placement between 2005-2018; 2) have an abdominal CT with contrast in a 6 - 9 months prior to intervention 3) have a clinical follow-up available for 90 days following the procedure. Clinical information was collected from the electronic health record. A fully automated machine learning algorithm was utilized to extract skeletal muscle-based body composition metrics [skeletal muscle area (SMA), skeletal muscle index (SMI), and skeletal muscle radiodensity (SMD)] at the L3 vertebral body level. Independent t-tests, univariate and multivariate logistic regression models, and ROC curve analysis were utilized to assess the association and performance of CT body composition metrics in predicting 90-day mortality.
Results: Of the 122 patients that met the inclusion criteria, 39 (32%) were female. The mean age was 58.2. 29 (23.8%) patients died within 90 days after TIPS. There were no significant associations between 90-day mortality and age (P = 0.36) or sex (P = 0.90). Patients who died within 90 days of TIPS had a significantly higher MELD scores (mean [SD], 18.9 [9.9] vs. 11.9 [6.3]; P < 0.0001), lower SMA (mean [SD], 123.0 [37.9] vs. 144.5 [30.8] cm2; P = 0.002), and lower SMI (mean [SD], 43.8 [14.4] vs. 50.5 [10.5] cm2/m2; P = 0.03) than those who survived past 90 days. There was no significant difference in SMD between those who died vs those who survived within 90 days of TIPS (mean [SD], 30.1 [9.2] vs. 31.8 [10.2]; P = 0.43). Multivariable logistic regression models showed that both SMA (OR=0.9, P = 0.003) and SMI (OR=0.9, P = 0.03) remained significant predictors of 90-day mortality when adjusting for MELD score. ROC curve analysis demonstrated that adding SMA improves the predictive power of MELD in predicting 90-day mortality after TIPS (AUC [95% CI]: 0.82 [0.75-0.90] versus 0.76 [0.66-0.86]).
Conclusion: Skeletal muscle-based CT body composition metrics are independently predictive of 90-day mortality after TIPS placement and improves the predictive performance independent of the MELD score. Further studies are needed to assess the generalizability of our findings and the potential role of skeletal muscle-based CT body composition metrics in assessing 90-day mortality after TIPS placement.