Cross-domain recommendation (CDR) leverages source domain information to enhance target domain recommendations. Recently, disentanglement techniques are utilized to separate domain-shared information from domain-specific information in CDR, reducing negative migration. However, previous disentanglement works only consider individual interaction preference and ignore collective preference. It fails to capture comprehensive user preferences. To address this, we propose Disentanglement-enhanced user representation via Domain-level Clusters for Cross-Domain Recommendation (DDC-CDR), a novel model integrates individual and collective preference distributions, using domain-level cluster preference distributions for collective preference distributions. To achieve our goal, we design two modules: (1) A domain classifier disentangles user representations into domain-shared and domain-specific components by information theory. (2) An enhancer promotes the adoption of domain-shared user representations across domains and utilizes hierarchical approximation to capture and align the domain-level cluster preferences of the source and target domains, ensuring consistency of domain-level cluster shared information during the cross-domain transfers. Based on them, we generate enhanced user representations containing comprehensive user preferences. Extensive experimental results on four real-world datasets demonstrate the significant superiority of DDC-CDR over state-of-the-art baselines.