This study spatiotemporally examines an association between unemployment and mortality. Hypothesizing the spatiotemporally heterogeneous association, we employ the Integrated Nested Laplace Approximation (INLA)-based random slope model with the spatiotemporal interaction under the control of common covariates. The model design aims to investigate the spatiotemporally transforming relationship between unemployment rates and mortality rates. The causes for all, self-harm, and mental disorders in 3,108 coterminous counties in the United States for 2001-2014, which includes two economic recessions, are considered. The results show the sporadic spatial effect and the cause-specific change of spatiotemporal interactions during the study period. The spatiotemporal patterns did not only have the same magnitude but also show the same direction of shift for causes of death. The spatiotemporal changes of the associations of one-year lagged unemployment rates are summarized as follows: (1) Dakotas and the west Appalachian counties have highly positive association in recent years; (2) the geographical shifts in high association regions were various for each cause of death: the dividing cluster for all-cause, the southerly moving cluster for the self-harm and interpersonal violence, and intensifying clusters in central and west Appalachian for the mental and substance-use disorders mortality; (3) the associations become weaker during the Great Recession period. Those patterns may be attributed to regional contexts, such as devoid of healthcare facilities and psychological deprivation. Even though there are possible mediating factors indicated by the substantial degree of residuals in some regions, our approach illustrates that the association of unemployment and mortality is spatiotemporally different across regions. It also suggests the spatiotemporal approach is effective in investigating the relationship between unemployment and mortality.