From about a decade ago, the decline rate of US mortality began to slow down. And for some certain age groups, the death rates even increased in the past few years. Many existing mortality models mainly focus on the forecasting of past mortality trends and characterize the trends by several latent factors. However, the interpretation of mortality changes is usually not very clear. This paper extends the Lee-Carter framework by introducing covariates into the dynamic process of the latent factor. And then we perform a comprehensive empirical analysis on the relationships between the latent trend in mortality rates and other exogenous variables. Counterfactual analysis is used to evaluate the impact of each covariate on the latent mortality trend. In addition, we also find the clustering structure in the age dependency of mortality rates. Separate analyses are performed independently for several different age-sub-groups.