Description of a New All Cause Mortality Surveillance System in Sweden as a Warning System Using Threshold Detection Algorithms

• Published in: European journal of epidemiology Vol. 21; no. 3; pp. 181 - 189
• Main Authors: Sartorius, B., Jacobsen, H., Törner, A., Giesecke, J.
• Format: Journal Article
• Language: English
• Published: Dordrecht Springer 01.03.2006; Springer Nature B.V
• Subjects: Adolescent; Age Distribution; Age groups; Aged; Aged, 80 and over; Algorithms; Biological and medical sciences; Cause of Death; Child; Child, Preschool; Cluster Analysis; Death; Disease outbreaks; Disease Outbreaks - statistics & numerical data; Epidemiology; Fatalities; Female; Geographic Information Systems; Geography - statistics & numerical data; Humans; Infant; Infant, Newborn; Infectious diseases; Male; Medical sciences; Medicin och hälsovetenskap; Middle Aged; Miscellaneous; Monitoring systems; Mortality; Mortality - trends; Poisson Distribution; Population Surveillance - methods; Public Health Informatics; Public health. Hygiene; Public health. Hygiene-occupational medicine; Registries; Standard deviation; Statistical median; Statistics; Surveillance; Sweden - epidemiology; Sweden; Europe; Human; Warning; Threshold detection; Mortality; Algorithm; Decision tree; Epidemiology; Warning signal; Warning threshold; All-cause mortality surveillance; Surveillance; Cause; Detection threshold; Death; Detection algorithms; Public health
• ISSN: 0393-2990; 1573-7284
• DOI: 10.1007/s10654-005-5923-6

A warning system has been set up to detect unusual weekly clusters of deaths by age group and municipality using all-cause Swedish death registry data. The technique for monitoring deaths by age group (<1, 1-24, 25-44, 45-64 and 65 plus) and week uses a compound smoothing technique, which calculates a baseline of expected events from retrospective data. Due to insufficient baseline data for the geographical component of the system a different algorithm, based on the Poisson distribution, was chosen to calculate expected weekly number of deaths per municipality, adjusting for municipalities with inherently higher mortality rates. This system was designed and tested during 2004 and implemented from the beginning of 2005. Threshold settings have been adjusted to provide a realistic number of weekly alerts. An evaluation of the system will be performed prospectively from the beginning of 2005 due to the lack of a gold standard for retrospective performance evaluation.