Survival of Patients With Advanced Hepatocellular Carcinoma Treated With Sorafenib in France During 2009-2018: Potentialities and Limitations of Model Adjustment in the SNDS French Hospital and Claim Database

OBJECTIVES:

Claim databases cover large populations but have little clinical information: this limits the possibility to adjust analyses on potential confounders. We aim to explore solutions to overcome the lack of clinical information in claim databases by modelling patients' survival with advanced hepatocellular carcinoma treated with sorafenib (HCC-s) in France between 2009/2018.

METHODS:

All HCC-s patients were identified in the French claims database (SNDS) between 2009/2018. OS from sorafenib initiation was estimated per year, a cox regression model was performed to model relation between OS and initiation year of sorafenib with adjustment on cofactors. Sorafénib SNDS study Abstract EASL 2022 2A combination of information was identified with specific SNDS algorithms to serve as proxies of disease severity. Aetiologies, age/gender and number of sorafenib dispensations were also identified as model confounders.

RESULTS:

17,680 HCCs patients were identified (Mean age: 66.9 years, male: 87.6%. The median OS from sorafenib initiation was 8.4 months (95%CI: [8.2;8.7]). Median OS decreased over the years: from 10.0 [9.1;10.8] to 7.3 [7.0;8.2] months between 2009 and 2018. An increase of disease severity proxies was observed: proportion of patients having complications, treatment procedures the year before sorafenib initiation and increase of delay from diagnosis to sorafenib initiation. We hypothesised that the evolution of the profile of patients treated with sorafenib in the interval may be the cause of the observed decrease of OS. OS decrease was no more observed in the multivariate model after 2012. However, the mortality increase risk is still observed in early years (2009-2012), possibly due to other unmeasured confounding factors not identified in the SNDS.

CONCLUSIONS:

Identification of disease severity proxies in claim databases are possible by using algorithms to identify complications and treatment procedures. Such factors can help to model outcome. Absence of clinical information and many unmeasured confounding factors remain a major limit to generalise findings.