Despite advancements in road infrastructure and ongoing efforts to improve traffic safety, how infrastructure features relate to crash frequency and crash types, and how these associations vary across regions and over time, remains insufficiently understood. In this study, Sweden's national-level road network is treated as an integrated system, and traffic, land-use, meteorological, and socio-demographic data are fused to analyze crash frequency and type distributions from 2018 to 2022. An integrated crash analysis framework is proposed, coupling a hierarchical hurdle model for crash frequency prediction (comprising a binary logit for crash occurrence and a truncated negative binomial for crash counts) with a hierarchical multinomial logit for crash-type classification. To demonstrate temporal instability of how external shocks (such as the COVID-19 pandemic) affected both crash-frequency and crash-type predictions, out-of-sample simulations are performed. Key findings include: (1) Crash frequency is strongly associated with road segment design, land use, and socio-demographic variables, while crash-type distributions are mostly influenced by roadway design and occurrence time; (2) Unobserved heterogeneities are found to significantly enhance model reliability and predictive performance; and (3) The COVID-19 pandemic is shown to notably alter crash frequencies but to have a comparatively modest effect on shifts in crash-type proportions. The results can provide a foundation for infrastructure design, risk-informed policy interventions, and dynamic maintenance strategies aimed at improving the safety of national-level road network in Sweden.