Confirmatory randomized clinical trials are designed to provide definitive information on the efficacy and safety of a new drug. While the outcome based on the overall population is the basis for the final conclusion, sound subgroup outcomes may provide additional insights of the drug effects and information for possible development of personalized treatment strategies. This presentation is to revisit issues related to subgroup analysis, to highlight pitfalls of potential biased interpretations and to explore ways for improvements. For the latter, the use of extensive stratifications to reduce the chance of biased subgroup findings is discussed with simulation results to illustrate the idea.