In many practical wireless systems, the Signal-to-Interference-and-Noise Ratio (SINR) that is applicable to a certain transmission, referred to as Channel State Information (CSI), can only be learned after the transmission has taken place and is thereby outdated (delayed). For example, this occurs under intermittent interference. We devise the backward retransmission (BRQ) scheme, which uses the delayed CSIT to send the optimal amount of incremental redundancy (IR). BRQ uses fixed-length packets, fixed-rate R transmission codebook, and operates as Markov block coding, where the correlation between the adjacent packets depends on the amount of IR parity bits. When the delayed CSIT is full and R grows asymptotically, the average throughput of BRQ becomes equal to the value achieved with prior CSIT and a fixed-power transmitter; however, at the expense of increased delay. The second contribution is a method for employing BRQ when a limited number of feedback bits is available to report the delayed CSIT. The main novelty is the idea to assemble multiple feedback opportunities and report multiple SINRs through vector quantization. This challenges the conventional wisdom in ARQ protocols where feedback bits are used to only quantize the CSIT of the immediate previous transmission.