Can synthetic controls improve causal inference in interrupted time series evaluations of public health interventions?