Are we missing opportunities to detect acute rheumatic fever and rheumatic heart disease in hospital care? A multijurisdictional cohort study

Data sources and study population

The cohort for this study was a subset of the End RHD in Australia: Study of Epidemiology (ERASE) database.¹³ In brief, ERASE includes harmonised data from five Australian jurisdictions: New South Wales (NSW), Northern Territory (NT), Queensland (Qld), South Australia (SA), and Western Australia (WA), with cross-jurisdictional linkages between SA and NT. The database comprises routinely collected and de-identified, person-linked records from ARF/RHD registers, inpatient separations, ED presentations, a surgery registry, and death registries.¹³ ARF/RHD ascertainment and first-ever diagnosis dates were established from RHD register records, surgery data, and/or diagnosis codes from hospital separations through a validated algorithm.¹³ Our cohort comprised persons whose initial diagnosis of ARF/RHD had occurred when they were aged 8–24 years, and for whom jurisdictional ED and hospital inpatient data were available for a 3-year lookback period prior to diagnosis. A high proportion (44.2%) of missing diagnosis codes in ED records throughout the study period precluded use of WA data. The usable date range of ED data from different jurisdictions varied widely, i.e. NSW (1 July 2005–30 June 2018), NT (1 January 2000–30 June 2018), Qld (1 July 2008–31 December 2017), and SA (1 July 2003–31 December 2017). We excluded 74 persons (2.1%) who, in addition to their ARF/RHD diagnosis, had any recorded diagnosis of a congenital cardiovascular disorder and/or a specified heritable disorder known to be associated with these.

Care episodes

To account for inter- and intra-hospital transfers, a hospital admission occurring on the same day or day following a discharge was considered part of the previous episode. Similarly, an ED presentation on the same day as or day prior to an inpatient admission was considered part of the admission episode, and instances of multiple same-day ED presentations were considered a single episode. If first identification of ARF/RHD was nested within an episode, the identification date was assigned to the episode commencement date. Discharges within 7 days before ARF/RHD identification were excluded from the analysis, as mimic diagnoses within this interval were considered potentially to reflect provisional rather than final diagnostic assessments.

Mimic diagnoses

Diagnosis codes were harmonised to International Classification of Disease (ICD) Version 10 Australian Modification (ICD-10-AM) codes¹⁴ by mapping where necessary from ICD-9-Clinical Modification codes (used prior to July 1999 in all included jurisdictions) and SnoMED-CT codes¹⁵ (used solely in NSW ED data). SnoMED-CT to ICD-10-AM mapping was conducted using the SnoMAP^® tool.¹⁶

An exhaustive list of all differential diagnoses (and corresponding ICD-10-AM codes) of ARF and/or (chronic) RHD was compiled manually, and complemented with a published list of differential diagnoses.¹⁷ Subsequently, authors APR (an infectious diseases physician with expertise in ARF) and BRM (a paediatrician with expertise in RHD) categorised the codes into mutually exclusive groups (Supplementary Table S2) as those representing disorders with: (1) a high likelihood of mimicking only ARF, (2) a high likelihood of mimicking either ARF or RHD (i.e. relevant cardiac disorders other than pericarditis), (3) a medium likelihood of mimicking ARF only or either condition, or (4) (excluded from further consideration) a low likelihood of mimicking ARF only or either condition. Decision-making to assign these categories was guided by consideration of the extent to which the original diagnosis could have represented true ARF or RHD in someone later diagnosed with either or both conditions. High likelihood mimics excluded codes indicating a specific alternative aetiology (e.g. gonococcal infection, drug-induced chorea). To allow for within-episode diagnostic refinement, only diagnosis codes from the final record of an episode were considered. As the study was based on de-identified administrative data, validation of mimic diagnoses through chart review of cases with relevant codes was not undertaken.

Covariates

Sociodemographic attributes including date of birth, sex, population group (categorised as Aboriginal and Torres Strait Islander peoples, immigrant from low- or middle-income country, or other Australians) and residential remoteness were derived from the broader ERASE project.¹³ Age at first ARF/RHD diagnosis was categorised as 8–12, 13–17, and 18–25 years. In order to investigate secular trends, calendar period was collapsed into three categories: 2003–2007, 2008–2012, and 2013–2018.

Statistical analysis

Patient characteristics were summarised descriptively within age strata. The proportion of persons who experienced at least one mimic diagnosis was calculated. Determinants of experiencing one or more care episodes with a mimic diagnosis were investigated, using several univariate and multivariable logistic regression models based on different sets of diagnostic codes, with associations expressed as odds ratios (ORs) with 95% confidence intervals (CIs). Firstly, odds were estimated of experiencing ≥1 combined ARF/RHD high- or medium-likelihood mimic diagnosis. Secondly, odds were estimated with restriction of outcomes to high-likelihood combined ARF/RHD diagnostic mimics only. Thirdly, the models were restricted to diagnostic mimics representing ARF only, (a) for the full lookback period (7 days–3 years), and (b) for a lookback period restricted to 90 days–3 years before first ARF/RHD identification, in order to capture only completely missed (rather than delayed) ARF diagnoses. Residential remoteness data (Accessibility/Remoteness Index of Australia Plus [ARIA+])¹⁸ were excluded from the models presented because of separation problems and constraints on interpretability due to a substantial proportion of missing values, however, in preliminary analyses this covariate did not materially alter overall results. All analyses were conducted using Stata^® version 16.1 (Stata Corporation, College Station, TX, USA).

Ethics

The ERASE project from which this paper was produced was approved by Health Department Human Research Ethics Committees in NSW, Qld, SA, and WA (and for NT: Menzies School of Health Research), and by Aboriginal Ethics Committees from NSW, SA, and WA.

Strengths and limitations

The principal strength of this study is its unique examination at scale of precursor diagnoses among hospital presentations. The use of multijurisdictional data provided the study with a representative sample, although the rarity of new ARF/RHD cases constrained numbers. Missed opportunities to detect ARF/RHD occur in contexts that are beyond the scope of our study design. A key limitation is that we lacked data from primary care, which likely represents an important setting for diagnostic delay.²⁴ Moreover, there is recent evidence based on echocardiographic screening investigation that a proportion of ARF cases characterised by subclinical carditis present with fever but are otherwise clinically silent, and that detection of ARF can be substantially augmented in high-risk settings by echocardiographic screening of febrile children.²⁵ These findings, along with the inevitable shortcomings of symptom recall, generate uncertainty in relation to the proportion of persons detected with latent RHD through echocardiographic screening programs whose antecedent ARF was genuinely asymptomatic, even when previous joint symptoms are denied. We excluded the code indicating fever per se from our analyses, based on our judgement that the positive predictive value of this symptom for ARF would be low, even in our study cohort defined by a future diagnosis of ARF/RHD. Considering the foregoing, together with the restriction of our study cohort to individuals identified with ARF/RHD (i.e. excluding those not yet or never diagnosed), our findings possibly constitute the ‘iceberg tip’ of under-diagnosis of these conditions. However, the true total extent of missed diagnoses remains speculative. Reliance on diagnosis codes for case ascertainment is a limitation of all studies based on administrative data. In this study, it was not possible to validate cases identified as mimic diagnoses by means of chart review. Noting that the potential gain from early diagnosis is greatest among persons presenting acutely with ARF, it was not possible in our study to determine the total proportion of presentations that potentially represented a missed diagnosis of ARF specifically, because most of the cardiac disorder-related codes in our analysis were compatible with either ARF or RHD. The combination of sparse outcome data and inconsistent results for subgroups across different models mandates additional caution in the interpretation of our findings. Our selection of disease codes that represent mimics of ARF/RHD was developed with the input of expert clinicians (authors APR and BRM) but has not been independently validated. Our approach aimed to be specific rather than highly sensitive, by excluding diagnoses indicating a convincing alternative cause. In so doing, we may have under-counted mimic diagnoses. Our estimates from 3-year lookback (chosen, for consistency, as the maximum possible across all included jurisdictions) inevitably underestimated lifelong pre-diagnosis missed detection, which was not feasible to investigate with the available data. We could not determine with certainty whether an individual mimic diagnosis indicated a genuine missed ARF/RHD case or an unrecognised alternative condition (e.g. infective arthritis).²⁶