Observational vs extrapolative models in estimating mortality from passive smoking