Association of Child Maltreatment With Risk of Death During Childhood in South Australia

This case-control study assesses the association of documented child protection concerns regarding child maltreatment with risk of death from infancy to 16 years of age in South Australia.


Introduction
Child abuse and neglect are prominent worldwide public health problems. 1,2 Rates vary by region and type of abuse; physical and emotional abuse are estimated to affect more than 35% of children in Asia, more than 50% of children in Africa and South America, 30% in North America, and 12.5% of children in Europe and Australia. 2 A reported 15% to 50% of children may be exposed to neglect 2 ; thus, child maltreatment, which covers abuse and neglect, affects approximately 20% to 50% of children worldwide.
Child maltreatment can start in utero (eg, with harmful drug use by the mother) but occurs throughout childhood. 3 The consequences of child maltreatment are considerable and can be lifelong. 4,5 Associations of child maltreatment with mental and physical health, including with increased risk for attempted (and completed) suicide, 6-8 mental illness, and addiction disorders 8,9 and higher rates of hospitalization, 10 have been widely documented.
The mechanisms underpinning the health consequences of child maltreatment are well described. [11][12][13][14] In children, the association of child maltreatment with injury, illness, and potential death is most direct in cases of serious physical or sexual abuse or profound neglect (supervisory, medical, and basic care). In addition, child maltreatment is associated with changes in the developing brain, 11,12 characterized as toxic stress, 13 affecting cognition, emotions, behaviors, and relational modeling, 14 which may be associated with additional harm.
The association of death with child abuse and neglect has been studied primarily through detailed case reviews. [15][16][17][18] Statutory review committees on death during childhood have been established in many jurisdictions to examine the circumstances of deaths among children to inform strategies to reduce potentially avoidable deaths. 19-21 A recent study by such a committee revealed that deaths among children with family-level child protection system (CPS) involvement (child, sibling, or parent) were associated with a substantially higher proportion of deaths coded as "undetermined/sudden infant death" or "external cause" than deaths with no CPS connection. 20 However, these studies were not population based and thus could not establish population-level risk estimates.
A review of studies of deaths among children and child maltreatment concluded that "the true incidence of fatal child abuse and neglect is unknown." 15(p265) A literature search revealed only 1 population-based study from 1992 in the US 22 involving risk of risk of death during childhood (into adolescence) by child maltreatment exposure. The authors reported 2.9 times the risk of death and 20 times the risk of assault (to <18 years of age) among children after a CPS notification compared with a matched comparison group with no notification. 22 Study estimates were not adjusted for potential confounders (such as birth outcome or socioeconomic status), nor did the authors explore risk by level of child protection concern. A 2011 study by Putnam-Hornstein 23 that used a similar method estimated the risk of injury death before age 5 years, reporting 5.85 times the risk for intentional and 2 times the risk for deaths from unintentional injury among children after a CPS report compared with children with no report. A 2021 study 24 on the risk of death from suicide after CPS contact during adolescence that used a nested case-control design reported odds ratios of 3.07 to 5.16 depending on the CPS category.
Evidence of the association of child protection concerns with all-cause death rates throughout childhood is limited. To address this evidence gap, we designed a study to estimate the association between different levels of child maltreatment concern and risk of death among children between 1 month and 16 years of age, adjusting for confounders present at birth. The age of 16 years was the chosen cutoff point, with the observation that deaths from age 16 to 18 years are more aligned with young adulthood, as reported in a recent study. 8

Study Design
This case-control study, nested within a population cohort of 608 547 children born in the state of The nested case-control design was adopted as the most rigorous study design to address our research objective to compare childhood mortality rates across CPS contact categories with consideration of the time-varying nature of CPS contact. This design allowed us to directly compare mortality rate ratios (MRRs) across CPS categories using a single, age-matched population sample while addressing immortal time bias and allowing for the complex transition pathways between CPS categories throughout childhood, thereby making the best use of the available population cohort data. Data linkage was undertaken by SA-NT DataLink, 26 the accredited data linkage agency for South Australia, using best practice determinist and probabilistic linkage. Extensive clerical review was performed, drawing on more than 50 data sets to deliver an estimated 99.6% accuracy (0.4% false-positive rate) and a 0.8% false-negative or missed links rate. 27 The research team received the deidentified data from the data custodians, with encrypted, project-specific linkage keys to enable merging across data sets.

Outcomes
Occurrence of death and month and year of death were identified from the South Australia Death Registry and used to identify study cases and children still alive at 1 month, augmented by perinatal records. Deaths before 1 month were excluded from the study because complications of pregnancy, birth trauma, and congenital defects are primarily responsible for perinatal deaths and owing to the minimal opportunity for CPS contact.

Covariates
A set of covariates was selected for inclusion in the multivariable analysis from established risk factors for death during childhood 31 or factors that might moderate the association between child maltreatment and death and included 4 birth outcomes sourced from the perinatal data: child still in hospital at 28 days after birth (yes or no), congenital abnormality (yes or no), preterm birth before 37 weeks' gestation (yes or no), and low birth weight (<2500 g) (yes or no). Also included were maternal attributes at the time of the child's birth sourced from the birth and perinatal data: age (<21 years or Ն21 years), marital status (married or in a de facto relationship, or neither), employment status (employed or not employed), and maternal smoking during pregnancy (yes or no). Area-based socioeconomic status using suburb and postal code of the mother, sourced from birth records, was mapped against the Index of Relative Socioeconomic Disadvantage 32 and classified as quintiles using Australian cutoff points. Child sex was not included as a covariate because in a nested case control design using conditional logistic regression, match variables cannot be included in the analysis. study with incidence density sampling, the odds ratios produced by logistic regression provided consistent estimators of MRRs for each CPS group compared with no CPS group. Separate models were fit with and without adjustment for the listed covariates. Unadjusted and adjusted MRRs were reported with 95% CIs and 2-sided P values. Statistical significance was set at P < .05. Separate models were run for male and female children with a dichotomized exposure of any CPS contact vs none. All analyses were conducted using Stata, version 16.0 (StataCorp LLC).

Descriptive Analysis
Of 606

Multivariable Analysis
Among children in all CPS contact categories, MRRs for death before 16 years of age, given survival to 1 month, were significantly higher than those among children with no CPS contact in the unadjusted and adjusted analyses ( Table 4). Mortality rate ratios were generally higher for CPS contact categories, suggesting more serious child protection concerns (although with overlapping 95% CIs). Unadjusted MRRs, which represent the observed risk of child death among children with  c Chapters I to XV of ICD-10 cover infectious and parasitic diseases; neoplasms; diseases of the blood and blood-forming organs and involving the immune mechanism; endocrine, nutritional, and metabolic diseases; circulatory system; mental and behavioral disorders; nervous system; the eye; the ear; respiratory system; digestive system; the skin; musculoskeletal system; genitourinary system; and pregnancy and childbirth.

Discussion
In this population-based case-control study, which used data for all deaths from infancy to 16 years of age in a 30-year South Australian birth cohort, incorporated different CPS contact levels, and adjusted for pertinent covariates at birth, an increased risk of death was found among children with  Poor birth outcomes, which were significantly associated with early death in our study, are not necessarily confounders. They may be associated with unsafe behaviors during pregnancy, such as heavy drinking or illicit drug use (representing neglect of the unborn baby). 36 As such, the multivariable estimates may be overadjusted, underestimating the association of child protection concerns with risk of death during childhood.
Children who had 1 or more notifications, even when a notification was not escalated to an investigation, had a substantially higher adjusted MRR compared with children with no CPS contact.
A possible explanation for this finding is that the opportunity for CPS contact escalation was curtailed by the censoring of CPS contact because of death. A modest further increase in risk was observed for children who underwent investigation (substantiated or not), with further risk escalation among children who had been removed to OOHC, for whom child protection concern was highest.
The risk estimates observed were somewhat higher than estimates from a previous populationbased study by Sabotta and Davis 22 of child maltreatment and child death into adolescence. They reported a 2.9 times unadjusted risk of death for children after a first notification to the CPS compared with children with no CPS contact, and Putnam-Hornstein 23 reported a 2.59 adjusted MRR for all injury-related deaths before 5 years of age.
In the present study, external causes contributed a higher proportion of deaths among children with CPS contact than did natural causes, and the highest differential proportion was from intentional injury. The percentages of deaths by assault or suicide were low among children without CPS contact (<1%) but accounted for 11.2% of deaths among children with CPS exposure. These findings were consistent with those of Sabotta and Davis, 22 who reported 18 times the risk of death from homicide among children with prior CPS contact (compared with those who had no CPS contact), 22 and those of Putnam-Hornstein, 23 who reported approximately 6 times the risk of death from intentional injury.
As this research was being conducted, the coding of child deaths was found to be incomplete, with family circumstances rarely captured. Only 2 of 1635 deaths listed any child maltreatment as contributing causes. Findings from coroners' inquests 37 and case study reports on child deaths 20 suggest that coded cause of death underestimates child maltreatment as a contributing cause of death during childhood in South Australia. This scenario may also be an issue elsewhere. Ensuring that the family circumstances surrounding a child's death are recorded as coded cause of death is important because deidentified administrative data are increasingly used to inform policy and practice.
The present study used CPS contact as the measure of child protection concern indicating possible child maltreatment exposure. Alternative case ascertainment through surveys of parents (of their abusive or neglectful behavior toward their infants and children), of teenagers (of their child abuse or neglect history), or of human services professionals (including clinicians) drawing on their observations to ascertain child maltreatment exposure was not a feasible option. Large sample sizes would be required to deliver the power to detect differences in death rates across child maltreatment exposure categories given the low occurrence of deaths among children. In addition, survey data have a number of possible sources of bias including cultural norms, social acceptability pressures, and failure to recollect early-in-life abusive or neglectful events and circumstances. Persons experiencing the worst child maltreatment outcomes are likely underrepresented in population-based surveys.
Using CPS contact as a measure of child protection concern allowed the use of large-scale administrative data. Although some children with child maltreatment exposure may not have had contact with the CPS and some children with CPS contact may not have been exposed to child maltreatment, at the group level, the CPS appeared to have sufficient discrimination to explore the based support for parents (and parents to be) to create a more nurturing environment for their child, with some programs commencing in pregnancy and continuing through middle childhood. [38][39][40][41] However, many distressed families do not have access to the intensive interdisciplinary and intersectoral family support programs and high-quality therapeutic services that they need.
Studies 42,43 have reported inadequate capacity to deliver the necessary trauma-based therapy to distressed infants, children, and their families. Clinicians and the wider human services workforce require high-level skills, expert supervision, and the opportunity to work in an interdisciplinary setting to successfully meet the needs of infants, children, and their families experiencing extreme adversity. 44 That outcomes for male individuals were worse than for female individuals in our study warrants further exploration, with possible policy and practice implications.
Researchers have demonstrated that investing in the early childhood years, especially for families with greater disadvantages, can offer a high return on investment in terms of health, education, and productivity. 45 Especially in a climate of increased stresses on families, as during the COVID-19 global pandemic and with increasing child protection concerns, society should be alert to the situation of children at risk of maltreatment and their distressed families and provide the support needed. Substantial progress has been made over many decades in reducing deaths among children.
However, there are still children who do not grow up in a safe and nurturing household, representing a clear opportunity for preventing future deaths during childhood.

Strengths and Limitations
This study has strengths. Access to linked administrative data are a considerable strength of the study, providing a sufficiently large cohort to explore deaths during childhood and access to information on a range of attributes that were potential confounders, derived at the time of the child's birth and avoiding the use of covariates collected later in life and potentially associated with child maltreatment exposure.
This study also has limitations. Death ascertainment before 1990 was incomplete. The study used the birth cohort from 1986 to maximize the number of deaths for analysis, with the observation that there was a less than 4 in 10 000 chance that a child selected as a control may have died before 1990. As a check, we reran the analysis from 1990, and the MRR for children with any CPS contact vs no CPS contact was unchanged at 2.94 (95% CI, 2.38-3.63) vs 2.99 (95% CI, 2.45-3.64).
Deaths were limited to those recorded in the South Australia Death Registry, with CPS contact also limited to the South Australia Department of Child Protection; thus, both the exposure and outcome excluded events outside South Australia. Given the case-control study design, it is unclear whether this limitation would have introduced any bias in our estimated MRRs. We had access to partial OOHC data before January 1, 1990. Out-of-home care data included all children who entered care after January 1, 1990, or entered care after January 1, 1986, and were in care on January 1, 1990.
For children who had entered OOHC, the observed MRR reflected the combined association of the child protection concern that triggered the removal, the removal itself, and possible safety concerns related to the alternate placement. Only 15 of 55 deaths among children ever removed occurred while children were in care, but further exploration of this group is needed. Cohorts larger than that in our study are needed to examine the association of child maltreatment and OOHC with risk of death among children to explore age at first (and subsequent) entry to care, care type (foster, kinship, or residential), number of placement changes, total time in care, and child maltreatment exposure characteristics.

Conclusions
The findings of this study suggest that children with reported child protection concerns have greater risk for death during childhood compared with children with no CPS contact. The high proportion of deaths from external causes suggests an opportunity for prevention.