Building a Multivariate model to estimate and prospectively monitor excess mort...

• Author(s): Rodrigues, Emanuel ; Nunes, Baltazar ; Marques, Jorge ; Machado, Ausenda ; Matias Dias, Carlos
• Date: 2013
• Persistent ID: http://hdl.handle.net/10400.18/1895
• Origin: Repositório Científico do Instituto Nacional de Saúde
• Subject(s): Epidemiological Monitoring; Influenza; Model; Prevention; Estado da Saude e da Doença
• Document Type: Conference Object
• Language: English

Description

Background As observed in several European countries, in the winter of 2011/12 the Portuguese mortality surveillance system detected an excess mortality in elderly population that was concomitant with an influenza epidemic and a cold spell. In order to estimate the impact of specific event contribution a multivariate model was developed. Methods We used an additive Poisson regression model, with mortality rate as the outcome and season specific ILI rate above baseline, extreme temperatures events (cold wave: less then 5ºC; heat wave: above 30ºC), trend and season components as the independent variables. All cause mortality data (week 26/2007 to week 20/2012) was extracted from the national mortality surveillance system. Excess mortality associated to influenza epidemic and cold spell was obtained by respectively summing specific events components of the model during the excess period. Results We observed a mortality excess period between weeks 2 to11/2012. Within this period the total estimated mortality excess was 3994, 97% of them due to influenza epidemics (AH3) and extreme cold event. Looking into specific event contribution, 75% (2978; CI95%: 2773-3185) was associated to influenza epidemic, 22% to extreme cold (889; CI95%: 801-978) and 3% unexplained. Results also showed that the multivariate model can be used for prospectively monitoring excess mortality, by setting the extreme temperatures and influenza epidemics covariates at zero and projecting the baseline for the future. Conclusion An excess 3994 deaths was observed during 2012 winter, 75% of these was attributable to influenza and 22% to extreme cold temperatures. The multivariate model allowed us to estimate excess mortality associated to different events but also to project a baseline for mortality monitoring. This approach may be a more suitable method to build baselines to prospectively detect excess mortality since no data is removed from the mortality time series