KEY POINTS
- This real-world single-centre study included 116 patients with hepatocellular carcinoma and macrovascular invasion: 50 received CyberKnife SBRT plus lenvatinib and tislelizumab, while 66 received lenvatinib and tislelizumab alone.
- Patients were selected from 215 screened cases and had preserved performance status and liver function. Stabilized inverse probability treatment weighting balanced measured baseline covariates, with additional 18-covariate and machine-learning propensity-score sensitivity analyses.
- CyberKnife SBRT was delivered 7–14 days after systemic therapy initiation, using 30–45 Gy in 3–5 fractions; the median regimen was 45 Gy in 5 fractions, corresponding to a median BED10 of 85.5 Gy.
- Confirmed objective response was 68.0% versus 37.9% before weighting and 71.1% versus 37.6% after IPTW adjustment (p<0.001) in the SBRT and systemic-only groups, respectively.
- Median local progression-free survival was 15.1 versus 7.8 months (HR 0.42, 95% CI 0.27–0.64; p<0.001), while median PFS was 11.1 versus 6.4 months (HR 0.46, 95% CI 0.30–0.71; p<0.001).
- Median overall survival was 23.0 versus 15.6 months (HR 0.57, 95% CI 0.36–0.92; p=0.022). Landmark and time-dependent analyses produced directionally consistent estimates, reducing—but not eliminating—concerns about treatment-timing bias.
- Grade ≥3 treatment-related toxicity was similar (28.0% vs 24.2%; p=0.648), with no classic radiation-induced liver disease. However, Child-Pugh class worsening was more frequent after SBRT (12.0% vs 1.5%; p=0.041), and two patients developed non-classic RILD.
CLINICAL TAKEAWAY
Adding local SBRT to targeted immunotherapy produced a striking association with deeper response, longer local control, PFS, and overall survival in HCC with macrovascular invasion. This is one of the stronger retrospective signals for integrating SBRT in this setting, but it remains non-randomized: treatment-selection bias, longer systemic therapy exposure in the SBRT group, and unmeasured confounding prevent a causal conclusion.