In this study, an adaptive and variable-rate feedback reduction technique for partitioned vector quantization (P-VQ) in MU-MISO downlink precoding is proposed. Independent feed-back reduction is performed for both channel quality indicator (CQI) and channel direction indicator (CDI) over slowly-varying Rayleigh channel. The users adaptively reduce the feedback over-head for quantized channel state information (CSI) depending on the rate and direction of change in the correlated CQIs and CDIs. Simulation results show that the proposed scheme greatly lessens the communication overhead with very small performance loss and enhances the quantization efficiency.