AHCC, like other advanced oncology indications, is often modelled using partitioned survival analysis (PartSA) based on overall and progression-free survival (OS/PFS) curves. Selective internal radiation therapy (SIRT) can result in tumour down-staging, making patients eligible for treatments with curative intent (TCI). This study aimed to assess the implications of including down-staging in a PartSA of SIRT with Y-90 resin microspheres versus sorafenib in the UK.

Targeted literature reviews were conducted to identify survival after resection and ablation (the most common TCIs after down-staging) and the proportion of patients down-staged after SIRT. A lifetime PartSA was developed based on a phase III trial (SARAH). Highly informative censoring (with only one TCI patient dying within follow-up period) resulted in the traditional PartSA approach excluding the mortality effect of TCIs, underestimating long-term efficacy. A second model structure including OS without downstaging from the trial and OS after subsequent TCIs from the literature was developed and compared with the traditional PartSA.

Depending on the selectiveness of the population 5%-29% of patients received TCIs after SIRT in aHCC, while 13.5% and 2.1% received TCIs in the trial after SIRT and sorafenib, respectively. Five-year survival after TCIs in the UK was 40%-60% (from 21% including cirrhotic patients) vs. 50% in the model including down-staging. Discounted quality-adjusted life-years (QALYs) for SIRT increased by 7% (1.98 from 1.85) with the explicit inclusion of down-staging, while total costs decreased by 6% (£31,146 to £29,143), leading to SIRT dominating sorafenib in the model explicitly including down-staging compared to an incremental cost-effectiveness ratio of £4,322/QALY in the traditional PartSA.

Inclusion of TCIs can have an important effect on survival and cost-effectiveness estimates. However, careful analysis of survival within trial is required to assess the need for explicit inclusion downstaging, and external and clinical validation of estimates is essential.