Syndromic surveillance for influenza in two hospital emergency departments. Relationships between ICD-10 codes and notified cases, before and during a pandemic.

BACKGROUND

Interest in the use of emergency department (ED) data by syndromic surveillance systems to detect influenza outbreaks has been growing. Evaluations of these systems generally focus on events during influenza seasons. The aims of this study were to identify which emergency department disease codes best correlated with confirmed influenza cases and to determine if these same codes would be useful in the non-influenza season. The 2009 influenza pandemic in Victoria, Australia, provided further opportunity to examine the performance of the syndromic surveillance system during this event.

METHODS

We undertook a retrospective analysis of data from the Victorian Department of Health's pilot syndromic surveillance programme, 'SynSurv'. SynSurv automatically captures patient information as it is entered by ED staff. This information includes patient demographics, their presenting symptoms and a preliminary diagnosis using ICD-10 coding. To determine which codes were best correlated with influenza notifications, weekly counts for each of the ICD-10 diagnosis codes ever used in the dataset were calculated and compared with the corresponding weekly count of confirmed influenza cases. Correlations between these codes and confirmed influenza cases in the non-influenza season were then undertaken. The data covered the period from July 2001 until August 2009 and included the 2009 influenza pandemic.

RESULTS

There was a marked increase in weekly counts of both laboratory-confirmed influenza cases and relevant ICD-10 codes during the influenza pandemic period. The increase in laboratory confirmed cases was more than four times greater than the previous highest number reported, in 2007, even though the influenza-like-illness activity in the community was considered comparable to 2003 and 2007. We found five ICD-10 codes to be moderately and significantly correlated with influenza cases. None of these codes was correlated with laboratory confirmed influenza notifications outside the influenza season, at least in part because of the small number of influenza cases notified during that period.

CONCLUSIONS

This study suggests that the choice of codes made by ED staff to record a case of influenza-like illness is influenced by their perceptions of how much influenza is circulating at the time. The ability of syndromic surveillance to detect outbreaks early may be impeded because case diagnosis is influenced by what ED staff believes to be occurring in the community.