Conditional logistic regression in cluster-specific 1: m matched treatment–control designs