On the formulation of epidemic models (an appraisal of Kermack and McKendrick)