We evaluated three established statistical models for automated early warnings of disease outbreaks; counted data Poisson CuSums (used in New Zealand), the England and Wales model (used in England and Wales) and SPOTv2 (used in Australia). recommend the SPOTv2 model over the England and Wales model, mainly because of a better sensitivity. However, the impact of previous outbreaks on baseline levels was less in the England and Wales model. The CuSums model did not adjust for previous outbreaks. INTRODUCTION With recent developments in world politics, monitoring infectious diseases statistically has increased in importance. Bioterrorism and biological warfare have sparked the development of computer systems for automatically detecting sudden changes in public health. Both the United States and the European Union invest large amounts of money for protection against these threats [1, 2]. This adds to more traditional reasons for surveillance of communicable disease, e.g. outbreak detection, monitoring trends of infectious diseases, and evaluating public health interventions [3]. In the detection of outbreaks of communicable diseases, it is desirable to minimize the time period between the actual start of the outbreak and the time Rabbit polyclonal to ACADL the system provides a warning. Different statistical models have been developed for this purpose, but we have been unable to find a systematic comparison between the different systems. In preparation, before the introduction of an automated system for outbreak detection of communicable diseases in Sweden, we evaluated three commonly used models designed to identify outbreaks sufficiently early to allow time for interventions. In order to evaluate the models, we used retrospective epidemiological data from the national Swedish surveillance system of communicable diseases. METHODS Data The Swedish Institute for Infectious Disease Control (SMI) is a governmental expert agency, with the task of protecting the Swedish population from communicable diseases. An important part of national communicable disease control is surveillance based on statutory notifications of 58 infectious diseases regulated by the Communicable Disease Act. A double notification system is used for each case of such disease. The two GSK2118436A reports emanate from the clinician treating the patient and from the laboratory having diagnosed the causative agent. Reports for the same patient are linked using a personal identification number issued to all Swedish residents, and used in all contacts with the GSK2118436A Swedish health care system. This double reporting system considerably increases the sensitivity of the surveillance system [4]. Whenever a laboratory performs microbiological typing, e.g. serotyping and phage typing for salmonellosis, such data are included in the laboratory report and used in the detection and investigation of outbreaks. All analyses were based on the date of registration at the national database at the SMI. The flow of information and timeliness in the surveillance system has previously been studied in detail, and the median delay between diagnosis and registration of the report was previously (1998C2002) 1C2 weeks [5]. Since 2004, a new electronic surveillance system has been in use with automatic reporting from the laboratories, allowing the detection of events in real time. For the evaluation of the three statistical models we used retrospective epidemiological data for three diagnoses with different outbreak patterns compiled by the SMI between 1992 and 2004; i.e. campylobacteriosis, hepatitis A and tularemia. Campylobacteriosis is the most commonly reported bacterial intestinal infection reported in Sweden with several previous large and small outbreaks; hepatitis A GSK2118436A has previously given rise to many small outbreaks both secondary for returning travellers and in intravenous drug users, and tularemia typically produces outbreaks when the rodent host population of the causative agent is increasing. The number of cases per week was studied. Thus, it was assumed that the population was constant during the study period. A baseline of 5 years starting with data between 1992 and 1997 was the base for estimating the expected number of cases for the.