This study investigates the issue of optimal preventive replacement scheduling for a repairable system, considering its failure/repair history and its environment or degradation, as characterised by covariates. The proposed approach is developed within a recurrent event modelling framework, in which the failure behaviour of the system is defined by its failure intensity, and perfect preventive replacements and imperfect corrective repairs are integrated following a virtual age assumption. The observed heterogeneity between systems is integrated through covariates by following a proportional hazard assumption. These covariates are assumed to be either fixed and represent, for example, the manufacturer origin, or to be dynamic and represent the monitored degradation process. The contributions of this study are twofold. First, a dynamic condition-based preventive replacement policy is developed. The maintenance decision rule integrates the available information on both the corrective repair history and covariate behaviour to determine the optimal preventive replacement time. This policy extends the state-of-the-art model of Gilardoni et al. (2016), which considers only the corrective repair history without integrating covariates. Second, the results of this study demonstrate how different levels of knowledge and information regarding the covariates can be integrated into the maintenance decision and lead to different optimal replacement times that are associated with different cost performances of the preventive replacement policy. This performance evaluation enables the optimisation of the monitoring policy and inspection frequency of the covariates. The proposed approach, which follows steps of increasing complexity, is developed and investigated first on static and then on dynamic covariates, considering minimal and imperfect repairs.