Estimating rotavirus gastroenteritis hospitalisations by using hospital episode statistics before and after the introduction of rotavirus vaccine in Australia.

INTRODUCTION

Hospital discharge records and laboratory data have shown a substantial early impact from the rotavirus vaccination program that commenced in 2007 in Australia. However, these assessments are affected by the validity and reliability of hospital discharge coding and stool testing to measure the true incidence of hospitalised disease. The aim of this study was to assess the validity of these data sources for disease estimation, both before and after, vaccine introduction.

METHODS

All hospitalisations at a major paediatric centre in children aged <5 years from 2000 to 2009 containing acute gastroenteritis (AGE) ICD 10 AM diagnosis codes were linked to hospital laboratory stool testing data. The validity of the rotavirus-specific diagnosis code (A08.0) and the incidence of hospitalisations attributable to rotavirus by both direct estimation and with adjustments for non-testing and miscoding were calculated for pre- and post-vaccination periods.

RESULTS

A laboratory record of stool testing was available for 36% of all AGE hospitalisations (n=4948) the rotavirus code had high specificity (98.4%; 95% CI, 97.5-99.1%) and positive predictive value (96.8%; 94.8-98.3%), and modest sensitivity (61.6%; 58-65.1%). Of all rotavirus test positive hospitalisations only a third had a rotavirus code. The estimated annual average number of rotavirus hospitalisations, following adjustment for non-testing and miscoding was 5- and 6-fold higher than identified, respectively, from testing and coding alone. Direct and adjusted estimates yielded similar percentage reductions in annual average rotavirus hospitalisations of over 65%.

CONCLUSION

Due to the limited use of stool testing and poor sensitivity of the rotavirus-specific diagnosis code routine hospital discharge and laboratory data substantially underestimate the true incidence of rotavirus hospitalisations and absolute vaccine impact. However, this data can still be used to monitor vaccine impact as the effects of miscoding and under-testing appear to be comparable between pre and post vaccination periods.