“…Considering a large number of potential covariates with complete information, most of which have been considered by previous studies (e.g., age [years], sex, height-SDS, weight-SDS, BMI-SDS, abdominal wall thickness [cm], fasting duration [hours], cooperation [calm vs. restless; dummy-coded], liver echogenicity, liver lower edge [dichotomized and dummy-coded: concave shape; pointed shape, rounded shape], liver size standardized to height-related mean value in %, spleen size standardized to height-related mean value in %), a two-step procedure as implemented in the FWRselect R-package was used to identify the most appropriate subset of covariates (independent variables) concerning the analysis of ATI, SWE, and SWD levels (dependent variables) in BBS patients by multiple linear regression (31). As previously described (23), a greedy forward selection algorithm was employed in a rst step, changing one variable at a time until no further improvement in model t assessed by the Akaike information criterion (AIC) was attained. Second, a bootstrap-based procedure evaluating the number of signi cant covariates as a tradeoff between model size and model t was performed at a signi cance level of p < .05.…”