Can linked emergency department data help assess the out-of-hospital burden of acute lower respiratory infections? A population-based cohort study

Bronchiolitis, pneumonia and other acute lower respiratory infections (ALRI) cause substantial morbidity in children. Indigenous populations such as American Indian and Alaskan Native, New Zealand Maoris, Canadian and Australian Aboriginal populations suffer a higher burden than their non-Indigenous counterparts [1–4]. In these studies and many others from developed countries, the estimates of burden are based on hospitalisation rates. There are limitations to using these data to assess the burden of ALRI. Hospitalisations represent the severe end of the ALRI spectrum and therefore underestimate the true burden of ALRI. To prevent transmission and population spread of ALRI, we need to investigate the burden of ALRI at the community level, through general practitioner and emergency services. To our knowledge, there is only one general practitioner dataset, the Bettering the Evaluation and Care of Health, or BEACH, program in Australia [5]. However, BEACH is based on a population sample of general practices in Australia, and not total population-based. While not community-level data, emergency department (ED) data which can be captured through administrative datasets may be able to provide an insight into the out-of-hospital burden of ALRI.

In addition to being a vital component of assessing the burden of ALRI [6], ED data can be used for syndromic surveillance as an early warning of epidemics and to forecast future epidemics [7–9]. For estimates to be meaningful, population-based data (through administrative health datasets) are needed. Unlike some parts of the United States where outpatient data are recorded as part of hospital morbidity datasets [10], ED data in Australia are collected through separate data systems. A single Western Australian (WA) study documented the epidemiological characteristics of ED presentations to four major teaching hospitals in metropolitan WA, but data were not stratified by Aboriginality and focused on upper respiratory infection such as tonsillitis [11]. There are no published data documenting the out-of-hospital burden of ALRI in WA or in Indigenous populations in the developed world.

WA has a strong history of data linkage infrastructure through the Western Australian Data Linkage System (WADLS), one of few such systems worldwide [12]. The WADLS encompasses systematic record linkage and uses probabilistic matching to link records from numerous datasets together for the same individual. The linkage of identifying data is conducted by the data linkage branch at the Western Australian Department of Health and subsequent de-identified data are given to the researcher. ED data are now available as a core dataset within WADLS. Previously we have used linked data from WADLS to investigate the burden, causal pathways and aetiology of ALRI hospitalisations in a WA birth cohort [3, 13–15]. Here, our primary aim is to investigate the feasibility of using linked ED data to describe the epidemiology of ALRI presentations to gain insight into the out-of-hospital burden of ALRI. Secondly, we aim to document the factors influencing presentation to ED with ALRI in metropolitan WA Aboriginal and non-Aboriginal children.

We have attempted to provide an insight into the out-of-hospital burden of ALRI in a cohort of Aboriginal and non-Aboriginal children using linked ED data. Bronchiolitis and croup were the most common diagnoses given to children, there was a clear seasonal peak of presentations in winter, and those aged <12 months had significantly higher presentation rates than older children.

We noted more ED presentations for croup than for bronchiolitis in non-Aboriginal children, despite bronchiolitis being the main reason for hospitalisation with ALRI in this population [13]. This is most likely due to the use of steroids over the last two decades in children presenting with croup in WA, who are then likely to be treated as outpatients with very few being admitted [21]. The presentation rates for croup were similar in Aboriginal and non-Aboriginal children. Overall, metropolitan-born Aboriginal children presented to ED with ALRI more often and more frequently than non-Aboriginal children, highlighting the continuing disproportionate burden of ALRI that Aboriginal children in WA suffer: 1 in 6 metropolitan-born Aboriginal children attended an ED with ALRI at least once before their ninth birthday as opposed to 1 in 14 metropolitan-born non-Aboriginal children.

We concentrated on infant and maternal predictors of ED presentation that were identified as risk factors for ALRI hospitalisation from our previous analyses [13]. We restricted this analysis to metropolitan-born children as those children being transferred from rural or remote areas to a metropolitan ED may have a more severe ALRI or have different family circumstances than children attending ED who live in Perth. Consistent with our previous analysis, we also reported predictors of ED presentation separately in Aboriginal and non-Aboriginal children due to the differences in disease burden and risk factors to hospitalisation with ALRI [13]. For all children, male sex and maternal age <30 years were the strongest predictors of ED presentation for ALRI. For non-Aboriginal children, there were many other important factors influencing presentation, such as previous pregnancies, autumn-births and elective caesarean delivery which we identified and offered explanations for in our previous work using hospitalisation as the outcome [13, 22]. For Aboriginal children, there were no significant predictors to ED presentation aside from male sex and being born to a teenage mother. This is in contrast to additional risk factors for hospitalisation for ALRI which included low optimal birthweight and maternal smoking during pregnancy [13]. These results suggest that the factors influencing admission to hospital with ALRI are different to the factors influencing presentation to ED in Aboriginal children. Also, importantly, in contrast to our previous analysis of hospitalisation with ALRI, socio-economic status was not a significant predictor of ED presentation in either Aboriginal or non-Aboriginal children, which may be due to the analysis being restricted to metropolitan births. A data linkage study of ED visits in infants from one jurisdiction in the United States found insurance status at birth to be the biggest predictor of ED visits for any diagnosis [10].

The ED presentation rates we have reported here provide us with an estimation of the burden to EDs with ALRI and an insight into the out-of-hospital burden of ALRI. However, our rates presented here are still likely to underestimate the true burden of ALRI to EDs, and are therefore minimum estimates, due to several important limitations of the available ED datasets. Indeed our ED presentation rates for ALRI were lower in all age groups than those reported in a similar study in Boston [6]. First, the EDDC from the nine metropolitan EDs contains only one ICD diagnosis code. Second, data being collected in rural and remote departments cannot be used to identify or differentiate between specific ALRI diagnoses because of the very broad and limited diagnostic categories available. Furthermore, of those metropolitan records with the capacity to record an ICD diagnosis code, this was often missing or too broad to be clinically meaningful for analysis. For example, we identified only 131 ED presentations for influenza from 2001 to 2005 which is an underestimate according to our previously analysed virology data from Princess Margaret Hospital for Children that identified 1802 specimens tested and 199 positive for influenza virus from 2001 to 2005 in children presenting to ED (Moore HC, unpublished data). A previous study in WA using the EDDC has suggested that there is no systematic bias in the failure to record a discharge diagnosis [23]. However, here we have shown differences between those presentations with symptom-coded ALRI (that often had missing ICD codes) and ICD-coded ALRI with respect to Aboriginality, sex and age, suggesting some level of systematic bias of missing ICD diagnosis codes.

Third, based only on one diagnosis code, or the primary discharge diagnosis code, there is a greater chance of inconsistent recording of diagnoses between various EDs. This was the experience in a data linkage study in the United States where the classification of the ALRI diagnosis was partly dependent on which ED children attended in the same geographical area [24]. Other ED registers worldwide have the capacity to record 3 to 10 ICD diagnosis codes [25, 26]. The allowance for multiple diagnosis codes increases the amount of diagnostic information in order to identify specific causes of presentation. While we acknowledge that we restricted our dataset by excluding broader or non-specific codes (e.g. viral infection of an unspecified site, which is most likely to be a respiratory infection) we opted to maintain a high specificity in identifying ALRI presentations.

The fourth limitation of these data relates to the representativeness on a population level. Our dataset included nine EDs from metropolitan Perth. While our results presented here accurately reflect the burden of ALRI ED presentations in metropolitan-born children in WA, we cannot extrapolate these findings to rural and remote WA. Additionally, due to the staggered entry of the nine metropolitan EDs to the EDDC system, we have been unable to investigate presentation rates over time, and ED presentations to the hospitals that did not commence data collection until 2004/2005 will have been missed. However, as the data in more recent years are complete, temporal trends will be possible for future data extractions and analyses.

Although, to our knowledge, data from the EDDC system in WA have not been validated against medical records, there are examples of administrative data being used to gain a picture of out-of-hospital burden of ALRI with few concerns over data quality. In one study, there was an overall lack of agreement between discharge diagnoses from administrative ED data and medical chart review but agreement was high for croup (90.4%), pneumonia (86.5%) and bronchiolitis (84.9%) [25].

We acknowledge that data reporting to EDDC needs to be improved as in its current form it has restricted ability to provide accurate estimates of ED attendance and an insight into the out-of-hospital burden at a population level. We offer the following recommendations: first, similar to coding of hospital admissions throughout the state [27], trained clinical coders should enter ICD diagnosis codes in an effort to reduce the amount of missing data. Second, the capacity to record multiple ICD diagnosis codes should be introduced. Third, rural and remote ED datasets that currently record data on broad level diagnostic categories should migrate to an ICD diagnosis system, as occurs with the hospital morbidity data from rural and remote areas. This would standardise diagnostic coding throughout the state and result in ED data coding being in-line with hospital morbidity coding. These recommendations for improvements to the EDDC have also been made by researchers investigating child maltreatment [28]. Other jurisdictions wanting to develop or improve existing ED data sources should ensure that these recommendations are met. Due to the population-based data linkage infrastructure in WA and the increasing ability to link other datasets such as microbiology records [15], these improvements are justifiable. If achieved, ED could be a useful data source in WA to provide measures of the out-of-hospital burden, in light of the lack of population-based general practitioner data, and improve surveillance systems to detect outbreaks of future infections.