To our knowledge this study is the first to investigate the application of data linkage methods for the improvement of routine notifiable disease surveillance analyses in Australia. Using data linkage, the proportion of STI and BBV notifications with missing Aboriginality data was able to be substantially reduced. Following data linkage, there were negligible proportions of gonorrhoea and syphilis notifications, and less than 10 per cent of chlamydia, hepatitis B and hepatitis C notifications with missing Aboriginality data. Data linkage is a useful tool for improving the completeness of Aboriginality data, and allows the calculation of more accurate STI and BBV disease notifications rates.
Notification rates calculated by excluding cases where Aboriginality data were missing commonly underestimated disease rates in both Aboriginal and non-Aboriginal people. However, disease rates in Aboriginal people were over-estimated when notification rates were calculated by apportioning notifications with unknown Aboriginality to the Aboriginal and non-Aboriginal categories using the same proportions as those observed among notifications where Aboriginality was known. This occurs because notifications with missing Aboriginality data are more likely to be identified as non-Aboriginal via data linkage. Apportioning notifications with unknown Aboriginality to the Aboriginal and non-Aboriginal categories based on the proportions observed among notifications where Aboriginality is known will result in biased estimates of disease rates among Aboriginal people, as case notifications with unknown Aboriginality systematically differ from case notifications with known data with respect to Aboriginality.
Following data linkage, the risk of STIs and BBVs associated with Aboriginality was found to decrease. Although there is a very high incidence of STIs and BBVs in Aboriginal people, and following data linkage the significant difference between disease notification rates according to Aboriginality remained; our findings suggest that Aboriginality is most probably not as strong a risk factor for chlamydia, syphilis and hepatitis C as is suggested by the notification data when cases with unknown Aboriginality are excluded.
Our estimated age-adjusted rate ratios for chlamydia and syphilis for Aboriginal versus non-Aboriginal persons were both significantly lower than previous estimates (rate ratios of 16 and 242 respectively) based on 2002 data for Western Australia [15
]. In contrast, our estimates of the age-adjusted rate ratios of chlamydia and gonorrhoea for Aboriginal versus non-aboriginal persons following data linkage remain at least 50 per cent higher than previous estimates for 2004 which were based on the allocation of all cases of unknown Aboriginality to the non-Aboriginal classification [4
]. Our use of 2001 census data rather than extrapolations for inter-Census periods, our exclusion of census data with unknown Aboriginality, and our use of 10-year age strata to minimise low cell counts in regression analysis also contribute to differing estimates of age-adjusted rates between this study and other published estimates.
This study only addressed the issue of completeness of identification of Aboriginality. It did not seek to address the issue of validity of identification of Aboriginality. It is possible that Aboriginality as reported in the notification data or any of the linked databases may be inaccurate, and reported Aboriginality is likely to be ascertained by various methods, including through reference to previous medical records, or as an individual judgement based on a person's appearance. The estimation of disease rates based on notification data is also limited by the effect of awareness and screening programs which are often targeted at high-risk groups, and the inclusion of cases associated with contact tracing and follow-up activities as a result of screening or case reports.
The outcomes of this analysis are also limited by the linkage of only one year of notification data with the linked database, and the absence of any additional strategies used to improve the quality of the routinely collected demographic data for notified cases prior to data linkage. The potential benefits of introducing additional strategies to improve data quality and more extensive clerical review processes for pairs of records that represented possible linkages remain to be investigated in future studies; however, the aim of this analysis was to investigate the potential for improving the quality of existing routinely collected data using minimally resource-intensive methods. Further studies linking more recent case notification data, including notifications with known Aboriginality, will offer the capacity to investigate the consistency of the identification of Aboriginality across all data sources, as well as examine trends in the identification of Aboriginality, and Aboriginal and non-Aboriginal disease notification rates over time.
Even disease-specific factors, including the severity of the disease and the need for a rapid public health response have been found to be insufficient to ensure complete reporting of diagnosed cases of notifiable diseases [1
]. This study provides evidence to support the benefits of using data linkage for notifiable diseases surveillance within health jurisdictions that have high rates of missing data on Aboriginality, as non-response introduced significant bias in the estimation of disease rates by Aboriginality. Our findings also indicate the importance of introducing strategies to minimise non-response, including item non-response, in the collection of routine disease surveillance data.
Strategies that have been found to improve the completeness of disease reporting include the use of active surveillance, laboratory-notification systems and improved links with clinicians [1
]. For routinely collected notifiable disease surveillance data in Western Australia, strategies that are currently being explored to improve the completeness of reporting include the addition of a field identifying Aboriginality on laboratory request forms to allow laboratory-only disease notifications to provide data on Aboriginality, and a message directing doctors to the Western Australian Department of Health's communicable disease notification website [16
] which details the procedure for the notification of communicable diseases. Further standardisation of data collection processes in Western Australia may assist in improving the item-level completeness of disease notification data.
For the majority of disease notifications, follow-up to obtain missing information on Aboriginality is not conducted by the Department of Health unless specific public health action is required, as this is a resource-intensive process which requires resources to be withdrawn from other public health activities. In this context, and while strategies are being developed to improve the completeness of reporting, the use of minimally resource-intensive automated data linkage techniques to improve the accuracy of disease notification rates provides an efficient and effective means to improve the quality of routinely collected data and enhance its usefulness. Until the reporting of Aboriginality in disease notification data improves, we recommend that data linkage studies are performed periodically to provide an improved foundation for public health policy.