I think that algorithms/ML people will have something to offer here. This essentially changes triage into a hard optimization problem. IANAD, but it's not just

I think that algorithms/ML people will have something to offer here. This essentially changes triage into a hard optimization problem. IANAD, but it's not just a problem of matching people in the current pool. You also have to plan for incoming patients. Having the parameters of the ventilator not perfectly match the patient likely affects the probability of survival in a smooth way (with in some bounds). So now the problem becomes minimizing the total death by maximizing the average likelihood of survival. You have to take the patient ventilator parameters into account (tidal volume, lung-compliance, weaning, etc) , as well as information about the distribution of those likely to be sick at the same time (which will change over time based on behavior).