Study design and study cohort
This is a retrospective cohort study in which individual-level hearing assessment results were linked to administrative datasets from multiple government agencies. The study cohort was all Aboriginal children; with audiometric assessments (for both ears) recorded in the NT Remote Hearing Assessment dataset, who were born in the NT between 1st January 1999 and 31st December 2008, and who attended school in remote and very remote geographic areas (defined using Accessibility and Remoteness Index of Australia (ARIA+)) [34]. As the focus for this study was on OM-related HI, we excluded all children with sensorineural hearing loss (n = 35) from the study (Fig. 1). Records for children in the Remote Hearing Assessment dataset were linked to records from the NT Perinatal Data Register to identify Aboriginal children born to NT residing mothers (n = 4287), which was then linked to NT school datasets to identify NT-born Aboriginal children residing in the remote and very remote areas (n = 3932). Lastly, the hospital dataset was used to identify and exclude the children with a hospital record of OM-related surgery before the age of 4 years (n = 37), under the premise that early surgery may alter the association between HI and child maltreatment. After the cohort selection process, there were 3895 children in our study cohort (Fig. 1).
Data sources
Data for the study was obtained from the NT child and youth data repository which was established through a collaboration between Menzies School of Health Research and NT Government agencies [35, 36]. Six key administrative datasets were used in the study. The first key dataset was the NT Perinatal Data Register, which is a statutory collection of information for all births in the NT. The second key dataset was the Remote Hearing Assessment dataset, which contains clinical and audiometric records for children assessed by the NT Outreach Hearing Health Program [37]. The third key dataset was the school attendance dataset (from government schools). The fourth key dataset contained results from the annual National Assessment Program – Literacy and Numeracy (NAPLAN) which is undertaken by government and non-government students in Years 3, 5, 7 and 9 [38]. The fifth key dataset contained the statutory records of all contact by children with child protection services. A sixth key dataset, the hospital separations dataset, was used to identify children receiving surgical procedures related to OM. Other datasets in the data repository included other health datasets, mortality records and youth justice records [30, 35].
Analysis
Outcome variables
In the NT, all reports of suspected child maltreatment made to child protection services are recorded as “notifications”. The preferred method of reporting suspected child maltreatment is by telephone to a Central Intake Team which operates a child protection hotline 24 h a day, 7 days a week. Reports from other agencies or reports made to police, or emergency services are also referred to the Central Intake Team. If a notification is consistent with maltreatment, it is referred for investigation, whereby the child protection service “obtains more detailed information about a child who is subject to a notification and staff make an assessment about the harm and the child’s protective needs” [39]. One possible outcome of an investigation is “substantiation”. A “substantiation” refers to an incident reported to child protection services which has been investigated with a conclusion that there is sufficient reason to believe the child has been, is being, or is likely to be abused, neglected, or otherwise harmed. In Australia, child maltreatment is reported as one or more of four types – neglect, physical abuse, sexual abuse (including sexual exploitation) and emotional abuse [39]. Exposure to family violence is reported as emotional abuse. From 1983, all adults in the NT have been required to report any child that they believed at risk of maltreatment, this statutory requirement was reinforced by the Care and Protection of Children Act 2007 (NT) which requires all adults to report any child that they believe ‘has suffered or is likely to suffer harm or exploitation’ [40]. In addition, the Domestic and Family Violence Act 2007 (NT) introduced mandatory reporting of domestic and family violence, including the reporting by police of children exposed to family violence.
The outcome variables for the study were the first child maltreatment notification and first substantiated child maltreatment notification, for each child, recorded in the child protection dataset. In additional analyses of child maltreatment outcomes by maltreatment type, we used the “primary maltreatment type”. When more than one maltreatment type is recorded for a single event, the primary maltreatment type is the one which is the greatest immediate risk to the child.
Explanatory variables
In our study, HI was determined from the first audiometric assessment (for each child) recorded in the Remote Hearing Assessment dataset, under the assumption that the first assessment result was representative of a child’s long-term HI status. This assumption is supported by previous findings that OM in NT Aboriginal children develops early in life [20], is persistent and asymptomatic, and is not diagnosed until an older age due to easier diagnosis [17]. In the NT Outreach Hearing Health Program, hearing assessments were performed using pure tone audiometry with results reported as the average threshold of hearing for the three frequencies: 500 hertz (Hz), 1000 Hz and 2000 Hz [37]. The result for each ear was classified as either normal or one of four levels of hearing loss, namely mild (16–30 dB HL), moderate (31–60 dB HL), severe (61–90 dB HL) and profound (≥ 91 dB HL). Based on these hearing results, we derived the HI variable as a categorical variable containing four mutually exclusive categories as listed below:
Normal hearing: normal audiometry results in both ears.
Unilateral hearing loss (UHL): normal in one ear and any degree of hearing loss in the other.
Mild HI: mild hearing loss in the better hearing ear.
Moderate or worse HI: moderate or worse hearing loss in the better hearing ear.
In the multivariable analysis, we selected the following additional variables that fell under three categories:
Child characteristics: sex (being female) and birth order (born as first child)
Maternal prenatal characteristics: less than seven antenatal visits during pregnancy; a record of the mother drinking alcohol during pregnancy; a record of the mother smoking during pregnancy; a record of maternal diagnosis of a sexually transmitted infection (STI)
Community characteristics: using the statistical local area (SLA) [41] in which the child first attended school [42].
Information on child and maternal characteristics were obtained from the perinatal dataset and community location was obtained from the school dataset. These variables were selected based on their availability in the NT data repository [30, 35] and evidence, from prior studies, of an association with child maltreatment. A Western Australian study reported a higher risk of child maltreatment among females [43], while an NT study reported greater risks for Aboriginal infants born to mothers with a record of a sexually transmitted infection (STI) during pregnancy, having attended less than seven antenatal visits, having consumed alcohol or having smoked during pregnancy [30]. Birth order was also included in the analysis, based on evidence from previous research that first born children are at greater risk of maltreatment [44]. The variable relating to ‘antenatal care’ is a proxy for maternal access to health care, with the cut-off point of seven antenatal visits based on clinical guidelines for remote NT settings [42]. Other maternal antenatal variables are proxies for the health status or health behaviour of the child’s mother [30, 42]. As previous research has demonstrated geographic variation in child maltreatment rates across the NT [30], we used community location (i.e. SLA [41]) to adjust (as community fixed-effects) for observed and unobserved differences between communities.
Aboriginal status
The Aboriginal status variable was derived from a hierarchy based on demonstrated accuracy with health datasets first, followed by child protection, education and youth justice records [42].
Statistical analysis
Survival analysis was used for various events. For each analysis, the survival time was defined as the time from birth to the occurrence of each event, children who did not experience the event were censored at the earliest of the following: date of death, the last observed date in the linked data or 31 December 2017. The time-scale for the survival analysis was the age of children in years (continuous variable). The Kaplan–Meier estimator was used to calculate the cumulative probability of each outcome at each age, by the four levels of HI. A Cox proportional hazard model was used in the multivariable analysis to examine the association between HI and the first record of child maltreatment. To account for the intra-group (community) correlation standard errors were clustered at the community level. Due to the substantial increase of child protection notifications/substantiations each year, separate baseline hazard rates were estimated for annual birth cohorts in the multivariable analysis, and the cumulative probability of each outcome was reported separately for the 1999–2003 and 2004–2008 birth cohorts. In additional analysis, multivariable analyses were conducted separately for the two cohorts. All statistical analyses were conducted using Stata for Windows, Version 15 (StataCorp 2015).
Ethical approval
The study was approved by the Human Research Ethic Committee of the NT Department of Health and the Menzies School of Health Research (HREC-2016-2708).