Estimating the true number of people with acute rheumatic fever and rheumatic heart disease from two data sources using capture–recapture methodology

Introduction

Acute rheumatic fever (ARF) is caused by an abnormal immune response to untreated superficial Streptococcus A (Strep A) infections of the skin or throat. Repeated or severe episodes of ARF can lead to chronic heart valve damage termed rheumatic heart disease (RHD). In turn RHD may lead to heart failure and arrhythmia, stroke, endocarditis and complications during pregnancy,¹ including in young people diagnosed with initially uncomplicated disease.²

The burden of ARF and RHD is greatest among communities with frequent superficial Strep A infections in early life. These infections are most common in settings with a high circulating burden of Strep A, including from household crowding and inadequate access to health and hygiene infrastructure. In Australia, these exposures disproportionately affect Aboriginal and Torres Strait Islander peoples resulting from the ongoing impact of colonisation, racism, poorly maintained dwellings and difficulties in accessing preventative medical interventions.³ The National Aboriginal Community Controlled Health Organisation is operationalising a program of work to reduce the burden of RHD in Australia.⁴

The notification of ARF and RHD is mandated in New South Wales (NSW), Northern Territory (NT), Queensland (Qld), South Australia (SA), Western Australia (WA) and Victoria (Vic).⁵ Notified cases requiring prophylaxis are assessed and recorded on the registers. In addition to notification requirements, most Australian jurisdictions maintain case registers that are used to facilitate best practice patient management. In 2009, the Australian Government funded the Rheumatic Fever Strategy (RFS) to control RHD and reduce morbidity and mortality related to RHD into the future.⁶ The focus of the RFS-funded control programs is to identify people with potential ARF/RHD as well as to record diagnosed cases of ARF/RHD onto the registers.

The END RHD in Australia: Study of Epidemiology (ERASE) Project was an initiative to capture the epidemiological characteristics of people living with ARF/RHD across Australia’s five jurisdictions (NSW, NT, Qld, SA and WA). They developed a linked dataset of ARF/RHD diagnosed cases that combined information from ARF/RHD registers, hospitalisations, death records and other data sources.⁷ Using the ERASE database, Agenson et al.⁸ showed that 26% of Indigenous cases of ARF/RHD and 10% of non-Indigenous cases of ARF/RHD were missing overall based on hospitalisations data from four jurisdictions (NT, SA, Qld and WA). Subsequent clinical validation studies have demonstrated that under-reporting in remote regions to RHD registers is as high as 25% for hospitalised ARF and 67% for hospitalised RHD.⁹ Despite drawing on multiple data sources, even studies such as ERASE are likely to underestimate the true burden.

Capture–recapture modelling examines the overlap in case identification across different linked data sources to estimate the number of cases not identified by any of the sources. Since incomplete surveillance systems cannot identify the true number of cases, capture–recapture methods have been widely used to enhance disease surveillance^10–14 and assess the extent of under-reporting. Oliver et al.¹² identified an estimated number of 2342 true new cases of ARF/RHD arising in New Zealand between 1997 and 2015, based on incomplete hospitalisation and case notification. This analysis indicated that females, those aged above 14 years and those who did not meet the high-risk definition (i.e. Māori/Pacific ethnicity and from most deprived areas) were less likely to be identified between datasets.¹²

The aim of this study was to estimate the true number of people living with ARF/RHD in Australia using information from jurisdictional registers and hospitalisation data. The specific aims were to:

Materials and methods

Results

Fig. 1 shows the percentage of the 2586 people with ARF/RHD identified in each data source between 2011 and 2016. Of these, 896 people (34.7%) were included only in the hospital data, 844 (32.6%) people were included only in the three jurisdictional registers and 846 (32.7%) people appeared in both the hospital and register data. There appeared to be a decrease in the percentage of people with ARF/RHD identified by both sources (i.e. overlap) over the years, with similar overlaps for years during 2014–2016. By year, the percentage of ARF/RHD people identified by the hospital data only appeared to increase, whereas the annual percentage of people identified by register data only appeared to decrease.

Fig. 1.

Distribution of ARF/RHD cases per data source, by year (2011–2016).

Table 1 shows people included in the current study stratified by selected demographic features. Of the 2586 individuals, 1600 (61.9%) were female, 1895 (73.3%) were Indigenous, 1596 (61.7%) resided in Qld and 1738 (67.2%) lived outside the metropolitan and inner regional areas. The mean and median age of individuals was 29 years (standard deviation = 14 years) and 28 years (interquartile range = 24 years) respectively. Females and males were evenly distributed within hospital and register datasets. Indigenous people with ARF/RHD were mainly identified (40.4%) in the register data source whereas non-Indigenous people were mostly recorded (70.8%) in the hospital data source. While the distribution of people with ARF/RHD across the data sources was equal in Qld and WA (around 30%), a majority of people in SA (60.3%) was recorded in the hospital data source. A majority of younger people (aged 3–34 years) was identified in the register data while a majority of older people (aged 35–54 years) was identified by hospital data. The hospital data source identified most people with ARF/RHD living in the metropolitan/inner regional areas (67.1%), whereas the largest percentage of people in the register data resided in remote/very remote areas (42.4%).

Table 2 shows the estimated number of people with ARF/RHD overall and across selected demographic factors. The capture–recapture model estimated the true number of people with ARF/RHD arising during the study period as 3480 individuals (95% CI = 3366–3600 individuals). The model thus identified an additional 894 (25.7%) individuals aged 3–54 years who were likely undetected by the hospital and register datasets. From this model, 27.3% of the total females with ARF/RHD and 23.3% of the total males identified were not detected by register or hospitalisation data. Of the 2326 total Indigenous people detected by the model, 431 (18.5%) were missed, as were 314 (31.2%) of the 1005 likely non-Indigenous people. Across age groups, the percentage of missing people with ARF/RHD was lowest in the younger age groups (9.4% in those aged 3–14 years) and highest in the older age groups (40.7% in those aged 45–54 years). Across remoteness, the percentage of missing people with ARF/RHD was lowest in remote/very remote areas (16.5%) and highest in metro/inner regional areas (25.4%).

Discussion

This capture–recapture analysis estimated that hospital and register data sources in WA, Qld and SA may not be identifying 25.7% of people with ARF/RHD aged 3–54 years. Furthermore, we demonstrated that non-Indigenous, urban and older people with ARF/RHD were least likely to be identified by either the hospital or register data sources.

This analysis suggests there may be opportunities to further support best-practice care delivery for urban living people and for non-Indigenous populations at risk of ARF/RHD in Australia. The number of Aboriginal and Torres Strait Islander peoples living in capital cities is growing rapidly and care for this population is more likely to be fragmented across different providers.²⁴ Strategies to increase awareness of ARF and RHD are likely to be needed to sensitise clinicians in urban areas to the RHD risk for Aboriginal and Torres Strait Islander peoples, while also keeping vigilant for non-Indigenous case detection.

The needs of non-Indigenous people with RHD in Australia have received relatively little attention in policy, funding, resource development or health education. Māori people, Pacific Island people and migrants from low- and middle-income countries have a higher risk of ARF/RHD²⁵ than the benchmark population. The burden of ARF/RHD for Pacific people is particularly stark in jurisdictional registers with a high urban case load: 43% of newly diagnosed people in NSW from 2015 to 2017 were Pacific. In Vic, 45% of children admitted to hospital for ARF/RHD were Pacific.²⁶ These communities have specific cultural and health engagement needs that are not well met by current RHD resources in Australia.²⁶

Understanding the likely undetected case load is important for developing a complete epidemiology picture of ARF/RHD and to inform comprehensive disease control strategies. In particular, an improved understanding of which cases are at-risk of not being hospitalised or reported to a register may provide new avenues to identify and address variation in care delivery.

The main limitation of the capture–recapture method used in this study is it was based on a multinomial logit model that assumed independence between the hospital and register data sources. Although these data sources are independent (assembled through different mechanisms and held by different organisations) there clearly are connections between the two sources. For example, a person presenting to an emergency department, diagnosed with ARF and admitted to hospital, protocols in most jurisdictions should ensure that they are recorded in both hospitalisation and register data. More data sources are required to validate the presence of dependence. Alternative methods such as the inclusion of covariates in the multinomial logit model have been proposed²⁷ but were beyond the scope of this study.

Conclusion

This study has provided an estimate of the substantial under-ascertainment of ARF and RHD cases in three Australian jurisdictions using hospital records and established but not mature ARF/RHD registers. Besides the undercount from the combination of these sources, the data also shows an unanticipated lack of overlap between the sources, varying by ethnicity and age. This indicates that there is a significant proportion of ARF/RHD cases who are not receiving hospital care or being referred to the register. Some may receive care in primary health care (PHC) services only while others might be unaware of their condition. Missed cases may present to health services with preventable complications in coming years. Harnessing data from PHC sources could further enhance ARF/RHD surveillance. The likelihood of undetected cases being urban-dwelling and non-Indigenous indicates a need to support clinicians to maintain an index of suspicion of ARF/RHD when working with other high-risk groups in metropolitan settings, ensuring they are aware of the need to refer suspected cases to hospital for specialist assessment and notify cases to public health authorities to facilitate register-based follow up where necessary.