Background Locoregional therapy (LRT) is frequently used as bridging therapy to transplantation/resection or palliative treatment for hepatocellular carcinoma (HCC). Multiple prediction models have been developed for prognosis and treatment response among patients undergoing LRTs. Objective We aimed to systematically review the methodological quality and performance of clinical risk scores predicting outcomes in patients with HCC treated with LRT. Design EMBASE and PubMed were searched from inception to 18 March 2026. Our main outcome was the concordance statistic, or area under the receiver operating characteristic curve (AUROC), to predict survival and other tumour-related and liver-related outcomes. Results 130 studies met the inclusion criteria, resulting in 179 individual scoring systems. The total population was 70 061 patients, 21% female (4–51%), most with Barcelona Clinic Liver Cancer Stage B (39%). Risk scores commonly incorporated tumour parameters, liver function tests, cirrhosis staging and comorbidities. AUROC values ranged from 0.56 to 0.94. 16 studies (12.3%) had low risk of bias, while most had high risk of bias. Of these, the Y-scoring system, Cheng et al nomogram and Li et al nomogram showed the highest discrimination for overall survival in palliative LRT (AUROC >0.87). Separate pooled analyses showed that the Hepatoma Arterial-embolisation Prognostic (HAP) and modified Hepatoma Arterial-embolisation Prognostic II (mHAP-II) scores had the highest performance (pooled AUROC >0.72), while the Six-and-Twelve (0.68) and Albumin-Bilirubin (ALBI) scores (0.60) demonstrated lower performance. Conclusion Several outcome scoring systems are promising for specific LRTs and populations. However, most models have high risk of bias, underscoring the need for further development and validation.
Daruwalla et al. (Thu,) studied this question.