Multi-connectivity enables a 5G cellular system to meet the challenging reliability requirements of downlink ultra-reliable low-latency communication (URLLC) data traffic. In it, multiple base stations (BSs) transmit to the URLLC user by pre-empting time-frequency resources assigned to enhanced mobile broadband (eMBB) users. We derive insightful expressions for achievability, which is the probability that the URLLC user’s block error rate (BLER) requirement is met by multi-connectivity. We do so for both joint transmission (JT) and orthogonal transmission (OT) modes of URLLC for the general case in which the transmissions occur over frequency-selective channels. We then propose a low-complexity algorithm to jointly select the set of cooperating BSs and their modulation and coding schemes (MCSs) to minimize the eMBB throughput loss. For time-varying channels with feedback delays, we present an alternate stochastic reliability requirement for URLLC traffic. The MCS selected on the basis of this requirement has a markedly higher probability of meeting the BLER target over the grid of URLLC user locations. Our results highlight the different trade-offs between URLLC achievability, eMBB throughput loss, and channel state information feedback overhead of OT and JT. They bring out the significant impact of feedback delays even at moderate Doppler spreads.