Mathematical descriptions of prostate cancer progression and treatment response often rely on deterministic dynamics, yet \emph{in vivo} behaviour is shaped by stochastic variation across scales. Randomness can be linked to clonal heterogeneity, phenotype switching, metabolic plasticity, spatial structure in the tumour micro-environment and fluctuating host–microbiome interactions, among others. These sources of variability influence observed treatment response and resistance, and motivate modelling approaches in which noise is treated as a structural component rather than a perturbation. This talk considers how stochasticity can be incorporated into dynamical systems for prostate cancer under phytocannabinoid-based therapy, with the aim of identifying which forms of variability are most relevant for describing experimental observations. A set of stochastic and hybrid models is introduced to represent tumour–microbiome–therapy interactions. Continuous-time dynamics describe tumour burden, metabolic state and microbial populations, while discrete stochastic events account for phenotype switching and treatment administration. Random effects are introduced through stochastic differential terms and probabilistic transitions to capture variability in drug response, metabolic coupling and micro-environmental conditions. The models are informed by experimental observations from phytocannabinoid treatments in prostate cancer, including differential responses under distinct dietary and metabolic contexts. Numerical simulations of these models consider both spatially structured and well-mixed settings, allowing comparison between intrinsic cellular variability, environmental fluctuations and therapy-related effects. The aim is to assess how different sources of stochasticity shape simulated treatment response and resistance patterns under mono- and combination phytocannabinoid therapies. By isolating and quantifying distinct noise mechanisms, the framework provides a basis for interpreting heterogeneous experimental outcomes and for developing data-informed stochastic models of refractory prostate cancer.