Fig. 2.

Model generation for individualized prediction of postoperative outcome

Given the continuous non-linear relationship between combined aspartate aminotransferase/platelet ratio index/albumin–bilirubin grade (APRI/ALBI) score and clinical outcomes, generalized additive models with smoothing splines were used, as illustrated for a postoperative grade C liver dysfunction (LD), b postoperative 30-day mortality, and c LD-associated 30-day mortality. Shaded areas represent 95 per cent confidence intervals. Based on the respective model, the combined APRI/ALBI score was grouped into deciles and individual risk for each outcome is shown in the lower panels. Number of patients in each decile is also shown.