Covariate adjustment

Deep learning to reduce sample size requirement for adjuvant HCC trials

Covariate adjustment
AI drug development
Histology Data
HCC
+6%
statistical power with the same number of patients, and equal statistical power with 12% fewer patients
Context

No treatment is yet approved in the hepatocellular carcinoma (HCC) adjuvant setting.

Methods

Owkin developed the HCCNet model for prognosis of resected hepatocellular carcinoma patients. Using public liver patient data, we evaluate the added value of HCCnet as an adjustment covariate.

Results

Results show the use of Owkin’s HCCNet model outputs as an additional adjustment in an adjuvant trial setting. HCCnet is added to tumor stage and ECOG to evaluate its added value.

Covariate adjustment on HCCNet achieves +6% statistical power with the same number of patients, and the same statistical power with 12% fewer patients.

Impact

De-risk phase 3 trial by maximizing the probability of achieving statistical significance.

Shorten timelines by reaching significance at an earlier interim analysis.

Reduce enrolment needs, hence trial timelines, costs and time to launch.