A Note on Large-Sample Approximations in Lognormal Linear Models