Influenza‐like illness in Australia: A comparison of general practice surveillance system with electronic medical records

Abstract Surveillance systems are fundamental to detect infectious disease outbreaks and guide public health responses. We compared influenza‐like illness (ILI) rates for 2015‐2017 using data from the Australian Sentinel Practice Research Network (ASPREN) and electronic medical records from 550 general practices across Australia (MedicineInsight). There was a high correlation between both sources (r = .84‐.95) and a consistent higher ILI rate in 2017. Both sources also showed higher ILI rates among women and patients aged 20‐49 years. The use of routinely collected electronic medical records like those in MedicineInsight could be used to complement active influenza surveillance systems in Australia.


| Data source
The study included data from ASPREN and MedicineInsight.
ASPREN collects notifications of ILI 3 from a representative sample of Australian practices (one practice per 200 000 individuals in urban and one practice per 50 000-100 000 individuals in rural areas, according to representation models for sentinel systems). 6 ASPREN currently collects de-identified data from more than 200 practices.
Most data are electronically collected weekly, via automated extraction or notifications reported in a web-based system.
MedicineInsight is a national general practice database managed by NPS MedicineWise. De-identified EMR are extracted monthly from Australian practices located in all jurisdictions, varying by size and type of services offered. Extracted EMR contain sociodemographic and clinical data, laboratory results and prescribed medications. Details of the data collection have been published elsewhere. 7 MedicineInsight has been used to investigate chronic conditions, 7 but also patterns of ILI management 4 and influenza immunization. 8 A previously developed data extraction algorithm was used to identify all patients with a diagnosis of influenza, ILI diagnosis or ILI symptoms (fever + cough+fatigue). 3,4 GPs prescriptions of anti-influenza medication (ie oseltamivir, zanamivir and peramivir) were also coded as positives for ILI even without a recorded ILI diagnosis (7% of all ILI cases), as it commonly happens within primary care data. 4 All ILI consultations by the same patient within 14 days of the first ILI diagnosis were considered as part of the same event. The diagnosis of other acute respiratory infections [ie upper respiratory tract infections (URTI), acute bronchitis and lower respiratory tract infections (LRTI)] was also extracted from MedicineInsight, as they could influence the recording of ILI because of the similarity in symptoms and known variation in labelling of respiratory illnesses by GPs. 9

| Sample selection and data extraction
Patient (sex, age, Indigenous status) and practice (rurality and state/ territory) characteristics were also extracted from the database.

| Data analysis
Weekly, ILI consultation rates (or "attack" consultation rates) were calculated using the number of ILI cases per 1000 consultations.

| RE SULTS
The peak of ILI cases in any year was observed between weeks 33 and 36 in both sources ( Figure 1

| D ISCUSS I ON
This study aimed to compare weekly ILI rates between ASPREN, a sentinel general practice surveillance system, and MedicineInsight, an extensive EMR general practice database, to identify whether the latter could be used to complement influenza surveillance in Australia. Results showed a high correlation between the two, and consistency regarding the shape of the curves and peaks. The higher rates in 2017 compared with previous years reflects a longer duration and more intense season that year, which was also identified by other surveillance systems in Australia. 10 Studies in the Unites States and Portugal also found good agreement between sentinel GP surveillance data and alternative databases using EMR, with correlations ranging between 0.78 and 0.99. 11,12 In ASPREN, a rise in ILI cases started earlier each year compared with MedicineInsight which is probably related to the increase in other acute respiratory infections, particularly URTI, as identified in MedicineInsight. Because GPs can label the same set of respiratory symptoms differently, 9 it is plausible that ASPREN GPs might code other respiratory infections as ILI because of their role in the sentinel system (ie observer bias). Future studies could address this issue, using regression modelling that takes into consideration the co-circulation of other pathogens with similar symptoms to influenza during analysis. 13 The higher incidence of ILI among women in both sources could, in part, be explained by the fact that women attend general practice in Australia more frequently than men. 4  Therefore, to provide routine reports for ILI, as a complementary surveillance system, would require additional funding and a partnership between NPS Medicine Wise, government and researchers.
Notwithstanding these barriers, a combination of MedicineInsight and ASPREN and the use of innovative methodological approaches could provide more reliable syndromic and virological information, leading to improved influenza surveillance in Australia. 13,15

ACK N OWLED G EM ENTS
The authors acknowledge NPS MedicineWise and the Department of Health for their support in the development of this research.