Effect size as a tool to identify subpopulations with improved clinical outcomes in metastatic colorectal cancer.

Abstract

252 Background: ASCO defined meaningful trial endpoints in colorectal cancer (CRC) to include OS HR ≤0.67 (Ellis, JCO 2014). This measure is limited in identifying treatment benefit for subgroups from heterogeneous populations. Effect size (Glass’s Δ) calculates the absolute difference in median clinical outcomes normalized to the control group standard deviation. We hypothesized that durable effect sizes ≥2 would be useful in predicting which trials possess subgroup populations of clinical significance despite a HR > 0.67. Methods: Prospective phase II-III trials in metastatic CRC from the ASCO Meeting Library (2016-2019) were cataloged by clinical outcomes of PFS and OS. Effect size was calculated from trials reporting confidence intervals and compared with absolute difference in clinical outcome, hazard ratio and therapeutic intervention. Trials with an indeterminant HR, yet effect size > 2 were reviewed in subgroup analyses. Results: 385 abstracts were reviewed with 99 clinical analyses available for effect size calculation. Absolute difference in PFS correlated with effect size (R = 0.64) and was inversely proportional to HR (R = -0.63). The absolute difference in OS correlated with effect size (R = 0.69) and was inversely proportional to HR (R = -0.57). When stratified by clinically significant HR (defined ≤0.67), median effect size for PFS was 13.7±13.3 (SD) which was significantly different from HR > 0.67 with median effect size 1.0±3.8 (p < 0.001). Median effect size for OS when stratified by HR ≤0.67 was 3.7±2.5 which was significantly different when compared to endpoints with HR > 0.67 with median effect size 0.9±1.4 (p < 0.003). Subgroup populations with survival benefit included combination checkpoint blockade durvalumab/tremelimumab vs supportive care with effect size 3.1 (HR 0.72; NCT02870920). First-line PFS benefit was predicted in KRAS wildtype liver-limited CRC treated with FOLFOX+cetuximab vs FOLFOX+bevacizumab by effect size of 3.2 (HR 0.80; NCT01836653). Conclusions: Effect size holds potential as a measure to delineate improved clinical outcomes from heterogeneous populations and could identify those trials for which further subgroup analysis should be explored.