This paper develops a granular regime-switching framework to model mortality deviations from seasonal baseline trends driven by temperature- and epidemic-related shocks. The framework features three states: (1) a baseline state that captures observed seasonal mortality patterns, (2) an environmental shock state for heat waves, and (3) a respiratory shock state that addresses mortality deviations caused by strong outbreaks of respiratory diseases due to influenza and COVID-19. Transition probabilities between states are modeled using covariate-dependent multinomial logit functions. These functions incorporate, among others, lagged temperature and influenza incidence rates as predictors, allowing dynamic adjustments to evolving shocks. Calibrated on weekly mortality data across 21 French regions and six age groups, the regime-switching framework accounts for spatial and demographic heterogeneity. Under various projection scenarios for temperature and influenza, we quantify uncertainty in mortality forecasts through prediction intervals constructed using an extensive bootstrap approach. These projections can guide healthcare providers and hospitals in managing risks and planning resources for potential future shocks.

翻译：本文提出了一种细粒度的机制转换框架，用于建模由温度和疫情相关冲击导致的死亡率相对于季节性基线趋势的偏离。该框架包含三种状态：(1) 捕捉观测到的季节性死亡率模式的基线状态，(2) 针对热浪的环境冲击状态，以及(3) 处理由流感和COVID-19引起的呼吸道疾病大规模爆发所导致的死亡率偏离的呼吸道冲击状态。状态间的转移概率通过协变量依赖的多项式Logit函数进行建模。这些函数纳入了滞后温度和流感发病率等作为预测因子，从而允许对不断演变的冲击进行动态调整。该机制转换框架基于法国21个地区和六个年龄组的周度死亡率数据进行校准，能够解释空间和人口统计异质性。在温度和流感的各种预测情景下，我们通过使用广泛的Bootstrap方法构建的预测区间来量化死亡率预测的不确定性。这些预测可以指导医疗保健提供者和医院管理风险，并为未来潜在的冲击规划资源。