With the increasing digitalization of the power system, cyber attacks that threaten physical disruption, such as power outages, are increasing. Hence, understanding system resilience and the operator responses is crucial for anticipating and mitigating threats that may cause outages through power line disconnections. A common approach to assessing grid resilience is to consider the worst-case attack, in which the attack is assumed to maximize the potential damage while the operators react to minimize such loss. This assessment, however, can have a vast number of possible actions by the attacker as the size of the power network and the severity of the attack increase, making it computationally expensive. We propose a topology reduction technique on radially operated distribution networks, which reduces the set of lines to be considered in the worst-case attack. The reduced network can determine such an attack more efficiently compared to the original network. Case studies on 33- and 119-bus systems showed that the introduced method reduced the network sizes by 25% and 38%, respectively, and its effectiveness on the worst-case attack computation increased as larger attacks with more line disconnections were considered.