Nationwide health interventions are difficult to evaluate as contemporaneous control groups do not exist and before-after approaches are usually infeasible. We propose an alternative semi-parametric estimator that is based on the assumption that the intervention has no direct effect on the health outcome but influences the outcome only through its effect on individual behavior. We show that in this case the evaluation problem can be divided into two parts: (i) the effect of the intervention on behavior, for which a conditional before-after assumption is more plausible; and (ii) the effect of the behavior on the health outcome, where we exploit that a contemporaneous control groups exists for behavior. The proposed estimator is used to evaluate one of Malawi's main malaria prevention campaigns, a nationwide insecticide-treated-net (ITN) distribution scheme, in terms of its effect on infant mortality. We exploit that the program affects child mortality only via bed net usage. We find that Malawi's ITN distribution campaign reduced child mortality by 1 percentage point, which corresponds to about 30% of the total reduction in infant mortality over the study period.