Ethics statement

The authors assert that all procedures contributing to this work comply with the ethical standards of the relevant national and institutional committees on human experimentation and with the Helsinki Declaration of 1975, as revised in 2008. All procedures involving human subjects were approved by both The University of Queensland Human Research Ethics Committee (HREC) (2021/HE001925) and the Metro South Health HREC (HREC/2022/QMS/83690). A waiver of consent was received by both HRECs relative to the inability to obtain consent from all participants whose data were used in this study. As in other work, we prospectively registered a pre-trial protocol with the Australian and New Zealand Clinical Trials Registry (ACTRN12622000870752).^{Reference Kisely, Leske, Arnautovska, Siskind, Warren and Northwood15,Reference Kisely, Arnautovska, Siskind, Warren and Najman16} We followed the STROBE (STrengthening the Reporting of OBservational studies in Epidemiology) guideline^{Reference von Elm, Altman, Egger, Pocock, Gøtzsche and Vandenbroucke17} and the RECORD (REporting of studies Conducted using Observational Routinely-collected health Data) statement.^{Reference Benchimol, Smeeth, Guttmann, Harron, Moher and Petersen18}

S.K. affirms that the manuscript is an honest, accurate, and transparent account of the study being reported; that no important aspects of the study have been omitted; and that any discrepancies from the study as planned (and registered) have been explained.

Participants

The MUSP study recruited participants who were born at the Mater Mothers’ Hospital, Brisbane's principal obstetrics unit, between 1981 and 1983.^{Reference Kisely, Leske, Arnautovska, Siskind, Warren and Northwood15,Reference Kisely, Arnautovska, Siskind, Warren and Najman16} In September 2000, agency-reported notifications of child maltreatment made to the Department of Families, Youth and Community Care were accessed and linked to the longitudinal database (n = 7214).^{Reference Strathearn, Giannotti, Mills, Kisely, Najman and Abajobir12} Child maltreatment reports were substantiated if, after investigation, there was reasonable belief that child maltreatment had occurred. There were four subtypes of child maltreatment, comprising neglect and physical, sexual and emotional abuse.

Health data sources

As previously outlined,^{Reference Kisely, Leske, Arnautovska, Siskind, Warren and Northwood15} we used encrypted identifiers to anonymously link the MUSP cohort to the two relevant Queensland Health administrative databases. One was the Queensland Hospital Admitted Patients Data Collection (QHAPDC), comprising psychiatric admissions to both public (from 1 January 2000) and private (from 1 July 2007) facilities throughout Queensland. The other was the Emergency Data Collection (EDC), comprising all emergency department presentations state-wide (from 2008). As a result, participants were at least 16 years old on admission or presentation. We used the most recent episode to identify cases of CMDs in each database. A case of CMD was defined as any admission or emergency department presentation with a primary diagnosis of depression, anxiety, PTSD or other non-psychotic disorders (ICD-10-AM codes F32–F45.9). This is consistent with the definition of CMD used by the National Institute for Health and Care Excellence in the UK.¹⁹ Bipolar affective disorder was excluded (ICD-10-AM codes F30–F31.9), as this is generally considered to be a persistent or severe mental illness, although there is no standard definition of either term.^{Reference Gonzales, Kois, Chen, López-Aybar, McCullough and McLaughlin20}

Data linkage and quality

The published protocol provides full details of the data linkage, which was performed by the Statistical Services Branch (SSB) of Queensland Health.^{Reference Kisely, Leske, Arnautovska, Siskind, Warren and Northwood15} The MUSP data custodian provided a data-set containing only the names, birth dates and sex of the birth cohort participants to the SSB. Within the SSB, these details were matched using the Queensland Health Master Linkage File (MLF). The SSB subsequently assigned a unique key to each linked record and shared this with the three data custodians, who used it as an identifier for the de-identified records that were sent to the researchers (Fig. 1).

Fig. 1 Illustration of emergency department presentations, in-patient admissions and Mater-University of Queensland Study of Pregnancy (MUSP) birth cohort database linkages.

The accuracy of linkages was optimised through a comprehensive review of all potential matches, with each linkage process undergoing cross-verification by the Project Linkage team manager. The MLF utilised various custom ChoiceMaker ‘models’ to establish the linkage map, using weighted tests to compute a match probability for pairs of records and incorporating machine learning. Identifying information pairs were manually examined to validate their accuracy. Error rates were consistently low, with 99.5% for false positives and 98% for false negatives. Further details about data quality are available in the Queensland Data Linkage Framework.²¹

Subsequent steps included cleaning and standardisation of identifying variables, accompanied by quality assurance checks for missing or duplicate data. When cross-referenced with other sources, sociodemographic characteristics such as Indigenous status were accurately identified in 89% of cases.^{Reference Kisely and Pais22}

Statistical analysis

We measured bivariate associations between the overall category of child maltreatment (agency-reported versus substantiated) and emergency department presentations or psychiatric admissions for CMDs. Hospital admissions and emergency department presentations were strongly skewed to the right. As a result, both were dichotomised into ‘never occurred’ versus ‘ever occurred’. We initially examined the association of CMD health service use with agency-reported and substantiated child maltreatment.

Logistic regression models, adjusted for covariates that had been identified in prior research as possibly confounding the relationship between child maltreatment and adverse psychological outcomes at follow-up, were also used. Covariates included baseline participant gender, parental racial origin, parental relationship status and low family income. Adjustment for marital status at the time of the most recent record in either QHAPDC or EDC was also undertaken. This was the sole sociodemographic variable consistently documented in the administrative data, apart from age and gender. We used the most recent record regardless of diagnosis (psychiatric or non-psychiatric), giving preference to QHAPDC data as these were the most complete. Single status was defined as having never been married or in a de facto relationship.

Where there were sufficient numbers, we conducted subgroup analyses examining the associations of child maltreatment with emergency department presentations and admissions for depression and anxiety/stress related disorders separately. As in previous work, we also conducted two sensitivity analyses.^{Reference Kisely, Arnautovska, Siskind, Warren and Najman16} The first was a propensity score analysis to examine the impact of adding a variable representing potential baseline confounders across the entire MUSP cohort. This was done in case specific sociodemographic characteristics were associated with difficulties in successfully linking the MUSP participants to Queensland health data. We opted for this approach over multiple imputation because it was not possible to assume that the data were missing at random. The second was to measure the effects of including marital status following imputation in the regression models. We only did this as a sensitivity analysis given that, as before, we could not assume that the missing data followed a random pattern.