Hidden baseline heterogeneity can mask treatment effects in randomized xenograft studies

Abstract Randomization and standardized analytical workflows are central to preclinical oncology, yet they do not necessarily ensure biological comparability at baseline. Using a real-world xenograft efficacy dataset generated under standard CRO conditions, we show that biologically meaningful treatment effects can remain statistically undetected when baseline heterogeneity is not explicitly considered. Although treatment groups appeared balanced at study entry by conventional summaries, baseline tumour volume exhibited latent structure within the cohort. Pooled group-level analyses reported no significant treatment effect, whereas baseline-aware evaluation of tumour growth trajectories revealed a responder subpopulation whose signal was masked by aggregation. This discrepancy did not arise from deficiencies in experimental conduct or data quality, but from limitations of pooled inference in the presence of structured baseline variability. Extending this framework to publicly available xenograft datasets, we observed that baseline tumour volume heterogeneity is common under routine preclinical workflows and can similarly alter efficacy interpretation. Together, these findings indicate that baseline tumour burden represents a biologically consequential source of hidden heterogeneity and that baseline-aware analytical strategies can improve the sensitivity and interpretability of preclinical efficacy assessment without deviation from standard experimental practice.