Table 3 Characteristics of ML-based predictive model of HCC patients after transplantation
Author | Study type | No. of patients | Model | Outcomes | AUC/C-statistics | Conclusion |
|---|---|---|---|---|---|---|
Marsh [58], 1997 | Retrospective Single center | 214 | ANN | 1,2,3-year RR | D: 0.962 ± 0.01 (1-year) 0.944 ± 0.05 (2-year) 0.952 ± 0.04 (3-year) V: 0.962 ± 0.043 (1-year) 0.966 ± 0.025 (2-year) 0.971 ± 0.034 (3-year) | The ANN model can identify post-LT HCC patients with or without recurrence |
Marsh [59], 2003 | Retrospective Single center | 214 | ANN | 1,2,3-year RR | 0.98 (1-year) 0.95 (2-year) 0.96 (3-year) | The ANN has genotyping as input parameter, which is possible to predict recurrence risk of post-LT HCC |
Rodriguez-Luna [60], 2005 | Retrospective Single center | 19 | ANN | Recurrence | – | This study validates the result conducted by Marsh et al., which the model had the discrimination power of 89.5% |
Zhang [61], 2012 | Retrospective Single center | 290 | MLP | 1,2,5-year survival | 0.909 (1-year) 0.888 (2-year) 0.845 (5-year) | The MLP model had high accuracy to predict post-transplant mortality risk for HCC recipients |
Nam [24], 2020 | Retrospective Multicenter | 563 | DNN | Recurrence | 0.75 | The DNN model showed promising predictive performance and outperformed other traditional predictive model to predict HCC recurrence after LT |