2D and 3D Volumetric CT Tissue Compositional, Demographic, and MELD Predictors of TIPS Survival

To test the predictive powers of automatically generated 2D and 3D CT tissue composition volumetric measures across the entire abdomen, patient demographic data and MELD scores for mortality post-TIPS placement.

Materials and Methods:

130 consecutive adult patients were identified who underwent TIPS between January 1, 2014, and January 1, 2024, and with a CT scan of the abdomen within 30 days prior to the procedure. Median age at time of TIPS was 56.7 (range 21.7-81.9) with 62% male and a majority being minorities (56% Hispanic, 15% white, 12% black, 17% other). Median MELD 3.0 score prior to TIPS was 14 (range 7-34). TotalSegmentator {1}, a validated artificial intelligence tool, was used to segment patient’s baseline CT scans. Body composition volumes (visceral fat, subcutaneous fat and muscle) were calculated across all available vertebral bodies at both their mid-point in 2 dimensions and their entirety in 3 dimensions. 3d body compositional volumes were also calculated across the entire abdominal volume. After descriptive exploratory data analyses, COX regressions were used to test predictors separately and combined.

Results:

Crude post-TIPS mortality was 31.5% (n=41) with mean follow-up of 2.5 years. Survival at 1y was 81% (95%CI 72-87%); 3y, 67% (57-76%); and at 5y, 57% (44-68%). Higher MELD 3.0 scores, black race and older age were significantly predictive of TIPS mortality (all p< 0.007). A similar statistical strategy of Tuifua et al. (2024) {2} did not confirm that 2D skeletal muscle at the midpoint of L3 was predictive of mortality (p=0.17). Moreover, our study of 3D compositional data, hip musculature volumes at the level of the sacrum and total 3D compositional data across the entire abdomen were also not significant predictors of TIPS mortality.

Conclusion:

MELD 3.0 score, racial group and age were predictive of TIPS mortality; differences in survival according to race and ethnicity should be investigated further. However, independently or additively, 2D and 3D body compositional measures were not predictive of survival. Discrepancies with previous literature may be secondary to chance, differences in patient population or frequency of events relative to sample size. Findings suggest exercising caution when considering using body compositional measures to exclude patients from TIPS procedures.