You Can Observe A Lot Just by Watching: A Cautionary Note on Establishing Baseline Influenza Mortality
Abstract:
Objectives We compare two statistical models for establishing “baseline” mortality attributable to pneumonia and influenza in the United States during the non-epidemic seasons 2013-14 through 2018-19. We aim to provide robust estimates of the burden of mortality from pneumonia and influenza during non-epidemic years in the United States, and to detail methodology for characterization of baseline mortality.
Methods: We obtained data on United States mortality attributable to pneumonia and influenza for the epidemiological seasons 2013-14 through 2018-19 from the U.S. National Center for Health Statistics Mortality Surveillance System. The data comprise weekly national mortality totals attributable to pneumonia and influenza, separately for adults 18 to 64 years old and adults aged 65 and older. We fit both generalized linear models and generalized linear mixed models to the mortality data; in contrast to the former models, the mixed models explicitly incorporated additional random components associated with intrinsic year to year variability in the mortality patterns. We also invoked a randomization procedure to ascribe uncertainty bounds to the summary descriptions of the mortality experience.
Results: The generalized linear models are analogous to averaging the mortality patterns over the 6 seasons, but failed to provide adequate representations of the annual mortality patterns. The generalized linear mixed models provided superior fits to the observed mortality data, but with a tradeoff of rather large uncertainty bounds on the mortality experience.
Conclusions: Summary estimates of “baseline” mortality attributable to pneumonia and influenza should be accompanied with an assessment of how well these summary measures represent the observed mortality experience, and metrics reflective of the variability inherent in the estimates.
