OBJECTIVE : We seek to characterize how faster tumour shrinkage rate (k) can lead to paradoxically shorter Response Evaluation Criteria in Solid Tumors (RECIST) time to progression ('TTP20' - tumour size exceeding its minimum by 5 mm and 20%)  and, therefore, progression-free survival (PFS). Specifically, we investigate under what conditions this paradoxical behaviour occurs, what fraction of patients satisfy these conditions, whether this phenomenon can invert population-level PFS hazard ratio, and consistency of an alternative time-to-event benefit metric with k. METHODS : We use a mathematical model treating tumour burden as decreasing drug-sensitive and increasing drug-resistant cell subpopulations. We fit this model to data from several clinical trials with different indications . We simulated a more effective treatment and recorded whether patients' TTP20 increased or decreased. We performed a study-level analysis to compare the relationship of speed and depth of response with TTP20 for both the administered 'control' and simulated 'more effective' drug. We propose and test an alternative benefit metric: the model-projected time that tumour size reaches 120% of baseline (TTB120). RESULTS : Depending on indication, 3-27% of patients are estimated to have a paradoxically inverse relationship between k and TTP20. Simulated head-to-head studies show that TTP20-based PFS can favour the less effective drug. In contrast, TTB120 always favours the more effective drug. CONCLUSIONS : We demonstrate the paradoxical behaviour of RECIST TTP20 - as an exemplar of percent-change-from-nadir based cancer progression criterion - both in theory and in observed patient data at the individual and trial level. We propose an alternative tumour size-based criterion (TTB120) that is directionally consistent with tumour shrinkage rate.