Diagnostic accuracy of surveillance tests for hepatocellular carcinoma in cirrhosis: a systematic review and network meta-analysis.

OBJECTIVE

For surveillance for hepatocellular carcinoma (HCC) to be effective, tests-including imaging, serological biomarkers (conventional and genomic) and algorithms combining multiple tests-must identify early-stage tumours. We aimed to identify, appraise and synthesise studies reporting the accuracy of all such tests in people with cirrhosis.

DESIGN

Systematic review and network meta-analysis of diagnostic test accuracy (NMA-DTA) data.

DATA SOURCES: MEDLINE and Embase (2005 to September 2025) and a published Cochrane review.

ELIGIBILITY CRITERIA: English-language, post-2005, one-gate or two-gate studies quantifying diagnostic accuracy of tests to detect HCC in populations wholly comprising people with cirrhosis, excluding those with pre-existing signs and symptoms of HCC.

DATA EXTRACTION AND SYNTHESIS: Data extracted by one reviewer, checked by a second and made available in an open-access database. We assessed risk of bias using QUADAS-2. We synthesised data using Bayesian NMA-DTA, accounting for tumour stage and incorporating continuous tests across all possible thresholds.

RESULTS

We included 170 studies (62 643 participants). Of 115 index tests, 97 were amenable to NMA-DTA. Ultrasound appears no better than alpha-fetoprotein at detecting very-early-stage HCC (sensitivity 0.34 (95% CrI 0.21 to 0.55) vs 0.39 (95% CrI 0.28 to 0.47)), only becoming superior as stage advances. Least affected by stage are contrast-enhanced MRI (sensitivity 0.70 (95% CrI 0.50 to 0.84) very-early; 0.86 (95% CrI 0.73 to 0.94) early; 0.90 (95% CrI 0.67 to 0.98) advanced) and CT (0.68 (95% CrI 0.26 to 0.93) very-early; 0.77 (95% CrI 0.29 to 0.95) early; 0.92 (95% CrI 0.49 to 0.99) advanced). No genomic biomarkers show convincing improvements over combinations of conventional blood-markers. Most studies are at high risk of bias, but conclusions are robust when restricting to studies with favourable methodological characteristics.

CONCLUSIONS

Using advanced synthesis methods, we found that tests have low sensitivity for detecting early-stage HCC. Given rising prevalence of cirrhosis and HCC, we need better tests and a stronger evidence-base to inform optimal surveillance strategies.

PROSPERO REGISTRATION NUMBER: CRD42022357163.