Month-to-month all-cause mortality forecasting: A method to rapidly detect changes in seasonal patterns

BackgroundShort-term forecasts of all-cause mortality are used retrospectively to estimate the baseline mortality and to obtain excess death after mortality shocks, such as heatwaves and pandemics, have occurred. In this study we propose a flexible method to forecast all-cause mortality in real-time and to rapidly identify short-term changes in all-cause mortality seasonal patterns within an epidemiological year. MethodsWe use all-cause monthly death counts and ratios of death counts between adjacent months as inputs. The ratio between one month (earlier month) and the consecutive month (later month) is called later/earlier ratio. We forecast the deaths one-month-ahead based on their proportion to the previous month, defined by the average later/earlier ratio over the preceding years. We provide forecasting intervals by way of a bootstrapping procedure. ResultsThe method is applied to monthly mortality data for Denmark, France, Spain, and Sweden from 2012 through 2022. Over the epidemiological years before COVID-19, the method captures the variations in winter and summer mortality peaks. The results reflect the synchrony of COVID-19 waves and the corresponding mortality burdens in the four analyzed countries. The forecasts show a higher level of accuracy compared to traditional models for short-term forecasting, i.e., 5-year-average method and Serfling model. ConclusionThe method proposed is attractive for health researchers and governmental offices to aid public health responses, because it uses minimal input data, i.e., monthly all-cause mortality data, which are timely available and comparable across countries. KeymessagesO_LIWhat is already known on this topic: There is a lack of methods to forecast all-cause mortality in the short-term in a timely or near real-time manner. C_LIO_LIWhat this study adds: The method that we propose forecasts all-cause mortality one month ahead assuming a seasonal mortality structure and adjusting it to the level of mortality of the epidemic year. These aspects make the method suitable for forecasting in a timely manner also during mortality shocks, such as the current COVID-19 pandemic. C_LIO_LIHow this study might affect research, practice or policy: The forecasts obtained with the proposed method detects changes in all-cause mortality patterns in a timely manner and can be used to aid public health responses. C_LI