Visible light communication (VLC) may be expected to become a new member in the small-cell family of the next-generation heterogeneous network landscape for complementing the overloaded radio frequency band. In order to further improve the performance of our user-centric (UC) cluster formation design proposed in our previous work, the quality-of-service (QOS) metric in terms of statical delay requirements is considered, when constructing the UC clusters. Inspired by the concept of effective capacity (EC), we formulate the proportional-fairness-based utility function of each optical link, while satisfying the statistical delay requirement. Upon maximizing the sum utility of the system, the joint cluster formation and user association problem is solved by our proposed greedy algorithm, which is capable of providing an average user EC of about 90% of the optimal value found via exhaustive search. Furthermore, the performance of the UC cluster formation is superior than the traditional cellular designs in all scenarios considered.