Cause of death (COD) data are essential to public health monitoring and policy. This study aims to determine the proportion of CODs, at ICD-10 three-position level, for which a long-term or short-term trend can be identified, and to examine how much the likelihood of identifying trends varies with COD size. We calculated annual age-standardized counts of deaths from Statistics Netherlands for the period 1996–2015 for 625 CODs. We applied linear regression models to estimate long-term trends, and outlier analysis to detect short-term changes. The association of the likelihood of a long-term trend with COD size was analyzed with multinomial logistic regression. No long-term trend could be demonstrated for 216 CODs (34.5%). For the remaining 409 causes, a trend could be detected, following a linear (211, 33.8%), quadratic (126, 20.2%) or cubic model (72, 11.5%). The probability of detecting a long-term trend increased from about 50% at six mean annual deaths, to 65% at 22 deaths and 75% at 60 deaths. An exceptionally high or low number of deaths in a single year was found for 16 CODs. When monitoring long-term mortality trends, one could consider a much broader range of causes of death, including ones with a relatively low number of annual deaths, than commonly used in condensed lists.
Short-Term Stock Price Forecasting Based on an SVD-LSTM Model
