Format

Send to

Choose Destination
Epidemiol Infect. 2015 Aug;143(11):2399-407. doi: 10.1017/S0950268814003276. Epub 2014 Dec 12.

Using winter 2009-2010 to assess the accuracy of methods which estimate influenza-related morbidity and mortality.

Author information

1
Group Health Research Institute,Group Health Cooperative,Seattle,WA,USA.
2
Department of Population Medicine,Harvard Medical School and Harvard Pilgrim Health Care Institute,Boston,MA,USA.
3
Kaiser Permanente of Southern California,Los Angeles,CA,USA.
4
Epidemiology Research Center, Marshfield Clinic Research Foundation,Marshfield,WI,USA.
5
Vaccine Study Center, Kaiser Permanente of Northern California,Oakland,CA,USA.

Abstract

We used the winter of 2009-2010, which had minimal influenza circulation due to the earlier 2009 influenza A(H1N1) pandemic, to test the accuracy of ecological trend methods used to estimate influenza-related deaths and hospitalizations. We aggregated weekly counts of person-time, all-cause deaths, and hospitalizations for pneumonia/influenza and respiratory/circulatory conditions from seven healthcare systems. We predicted the incidence of the outcomes during the winter of 2009-2010 using three different methods: a cyclic (Serfling) regression model, a cyclic regression model with viral circulation data (virological regression), and an autoregressive, integrated moving average model with viral circulation data (ARIMAX). We compared predicted non-influenza incidence with actual winter incidence. All three models generally displayed high accuracy, with prediction errors for death ranging from -5% to -2%. For hospitalizations, errors ranged from -10% to -2% for pneumonia/influenza and from -3% to 0% for respiratory/circulatory. The Serfling and virological models consistently outperformed the ARIMAX model. The three methods tested could predict incidence of non-influenza deaths and hospitalizations during a winter with negligible influenza circulation. However, meaningful mis-estimation of the burden of influenza can still result with outcomes for which the contribution of influenza is low, such as all-cause mortality.

KEYWORDS:

Influenza; modelling; statistics

PMID:
25496703
DOI:
10.1017/S0950268814003276
[Indexed for MEDLINE]

Supplemental Content

Full text links

Icon for Cambridge University Press
Loading ...
Support Center