Infant Deaths From Medical Causes After a Maltreatment Report

Birth, CPS, and death records from California were linked to create the data set used for analysis. Vital birth and death records were obtained from the California Department of Public Health. Administrative CPS and foster care placement records were available through a data-sharing agreement with the California Department of Social Services. Birth records were used to identify the population and to control for risk factors present at birth and associated with both CPS involvement and fatality risk; CPS records were used to examine the timing and response to alleged maltreatment during infancy; and information concerning deaths classified as due to medical causes was provided in death records. Linkages and analysis were approved by both state and university committees for the protection of human subjects.

Records were linked by using open-source probabilistic linkage software (ChoiceMaker, LLC, Princeton, NJ). Match pairs with a probability score in excess of 0.80 were coded as a true match, or a pair of records containing data for the same person in different data sets. The overall record linkage from which data were drawn for this analysis involved all births occurring between 1999 and 2017. These births were linked to CPS records reflecting all children reported for maltreatment between 1999 and 2017 and death records capturing all recorded deaths between 1999 and 2017. To construct our study cohort, we identified in the birth record data all children born between 2010 and 2016 (n = 3 464 535). Using encrypted linkage keys generated during the linkage process, we then connected births in our study cohort to CPS records for the 1-year period after a child’s birth (n = 183 436). Finally, we used linkage keys to incorporate death information for our study cohort. Death records were limited to deaths in which the date of death indicated a fatality within 365 days of a child’s birth (n = 4027). We excluded neonatal deaths (n = 9601). Previous studies involving these data sources and data linkage procedures provide additional details on these data sources.^32–34 

Our dependent variable was defined as time to death from a medical cause, modeled as days from birth to death. Deaths from medical causes were identified from the death record on the basis of the medical cause-of-death codes (n = 1051) (Table 1) designated in the International Classification of Diseases, 10th Revision (ICD-10). All mechanisms of deaths from a medical cause were included; children with deaths due to nonmedical causes were censored on the date of death. Children who died during the neonatal period (≤28 days after birth) were excluded from analyses (n = 9601) given the typically congenital and perinatal nature of those deaths and the limited window for prevention (93% of deaths during that period were coded as due to congenital abnormalities).

Specific allegations of maltreatment and placement episodes in foster care were coded as time-varying covariates on the basis of event dates recorded in the administrative CPS records. This allowed us to dynamically model children’s movement between risk groups or risk sets. Children in the overall study cohort (ie, children born between 2010 and 2016) entered the reported risk set beginning on the date of a first nonfatal allegation of abuse or neglect received by CPS. If additional maltreatment reports were received during infancy, they were coded on the basis of report dates. We also modeled changes in a child’s risk of fatality during periods in which the child was placed in foster care. Specifically, on entry to foster care (ie, episode start date), the child transitioned from the reported risk set into a foster care placement risk set. If a child’s foster care placement episode ended, they returned to their last report risk profile based on the number of previous reports and their earlier categorization. The child would move back into the foster care risk set if another out-of-home placement episode began. Through this coding, we dynamically modeled medical fatality risk associated with the time infants were at home (after a report of maltreatment) versus in foster care (also after a report of maltreatment). Observations were censored if there was no death by the child’s first birthday. All reports of alleged maltreatment were included, regardless of CPS decisions to investigate or substantiate.

To control for differences in child and family risk factors associated with CPS involvement and medical deaths, we adjusted for sociodemographic fields universally measured at the time of birth: child sex (male or female), maternal race and ethnicity (white, Black, US-born Hispanic, foreign-born Hispanic, and Asian/Pacific Islander), maternal age in years (<20, 20–24, 25–29, and 30+), prenatal care (first trimester and late or no care or missing), birth payment method (public insurance and private insurance), paternity (missing or established), and a combined indicator of birth vulnerability based on a low birth weight (<2500 g) or preterm birth (<37 weeks) (yes or no). In California, mothers not covered by any health insurance at the time of birth are retroactively enrolled in the state’s Medicaid (ie, public health insurance) program, although this would not affect a mother’s access to health care prenatally.

We computed descriptive statistics for the full population of children born between 2010 and 2016 who died of a medical-cause death during the postneonatal period of infancy. For our indicators of CPS involvement (ie, reported and placed in foster care) and for each birth record covariate, we ran a χ² test to assess whether the population of children who had died differed significantly from those who did not die. We then fit multivariable competing risk models.³⁵  We started by fitting a crude model with only our measures of CPS involvement and then incorporated birth record covariates into an adjusted model to address differences in risk. Estimates are reported as hazard ratios (HRs) with corresponding 95% confidence intervals (CIs). The proportional hazard assumption was evaluated both graphically and on the basis of Schoenfeld residuals. A significant relationship between birth vulnerability and hazard of death was found (χ² = 52.7; P <.05). The adjusted model was then re-estimated with stratification by birth vulnerability status. The proportional hazard assumption held for the stratified adjusted model. All analyses were conducted by using the Stata statistical package version 16.1 (Stata Corp, College Station, TX).

As shown in Table 2, among children born in California between 2010 and 2016, the postneonatal infant death rate due to medical causes was roughly 3.0 per 10 000. Among infants with a report of maltreatment, the rate of medical death was 13.1 per 10 000, or roughly 4 times the rate in the population overall. Infants who spent any time in foster care had a cumulative infant death rate from medical causes of 3.1 per 10 000. The average age (in months) of a first report was 4.2 (SD: 3.91) overall but 3.1 (SD: 2.94) among children who died of a medical cause (t = 4.32; P < .05). Among infants with a medical-cause fatality, the average time from first report to death was 2.1 months (SD: 2.66).

Approximately 5% (0.053) of children in our study cohort were reported to CPS for maltreatment concerns during infancy (Table 2). Among those who died postneonatally from a death due to a medical cause, nearly one-quarter (0.229) had been reported to CPS for maltreatment, a notably larger proportion than children who did not die of a medical cause during infancy (P < .05). Among children who died of medical fatalities, there was a higher proportion of male patients; Black children died of medical causes at a rate more than twice their proportion of the population. Younger maternal age, late prenatal care, public insurance, presence of a birth vulnerability, and missing paternity were all associated with heightened risk of a medical-cause fatality. There were no statistically significant differences in the proportion of children placed in foster care among children who died of a medical cause versus those who did not die.

In Table 3, we model a child’s crude risk of a medical fatality on the basis of their reporting and foster care placement status, with additional estimates produced inclusive of adjustments for sociodemographic covariates observed at birth. Because minor differences in magnitude were observed between the crude and adjusted models, we focused our discussion on the final adjusted estimates. Infants with a history of reported maltreatment had a significantly heightened risk of death compared with children never reported (HR: 1.77; 95% CI: 1.36–2.30). The hazard of medical death increased among infants with ≥2 reports during the first year (HR: 3.27; 95% CI: 2.48–4.30). Among infants reported for maltreatment, periods of foster care placements emerged as protective. When in foster care, the rate of death was roughly 50% lower (HR: 0.43; 95% CI: 0.24–0.76). See Figure 1 for illustration of the cumulative hazard rates of postneonatal medical-cause deaths during infancy stratified by no maltreatment report, 1 or more maltreatment reports, and foster care placement.

In this population-based prospective study, we documented that infants with a history of reported maltreatment had significantly greater rates of death due to medical causes than those who had never been reported. Our findings also reinforce that as more reports were received, the associated risk increased. Importantly, periods during which infants were placed in foster care were associated with significantly reduced rates of deaths due to medical causes, highlighting the preventable nature of many medical-cause deaths and raising questions regarding missed opportunities to ensure health and other preventive services for infants who remain at home. We discuss these findings in greater detail in the following sections.

First, data from the current study document a heightened risk of medical-cause fatalities among infants reported to CPS. We also found that each additional report of maltreatment during infancy was associated with a heightened risk of death due to medical causes. These findings align with previous fatality studies in which researchers have concluded that maltreatment reports serve as an important signal of child risk,²⁸  including risk of death,^23,26  an established and largely objective indicator. Yet most previous studies examining CPS reports and death have focused on injury fatalities, a more direct and apparent manifestation of abuse or neglect.³⁶  No published research to date has used population-based data and a prospective analytic design to study the risk of medical fatalities after allegations of abuse or neglect. Likewise, our findings of a higher risk of death with >1 maltreatment report also align with the established relationship between multiple maltreatment reports and more health problems.^21,37 

Several family and contextual dynamics may help explain the elevated rates of medical death observed in the current study. Families who have maltreatment reports may experience acute or chronic challenges with substance abuse, mental health conditions, or extreme poverty (all of which could lead to inadequate caregiver supervision and inadequate adherence to health care recommendations).^38,39  The lack of social support in CPS-involved families can also make it more difficult for families to care for infants with chronic or acute health problems.⁴⁰  Relatedly, medical neglect may be related to the increased risk of infant deaths from medical causes.⁴¹  Pediatric providers may need to monitor adherence to home health care more closely and follow up on infants with health problems in families with a history of CPS involvement. Pediatric providers are already encouraged to provide anticipatory guidance to all families of children with medical complexities to prevent risks associated with medical fragility and unfavorable health-related outcomes.⁴²  This anticipatory guidance may be even more important for families in which challenges have already resulted in ≥1 maltreatment report.

Perhaps most notably, we found that periods of foster care placement were associated with significantly decreased rates of death from medical causes. To be clear, ∼1 in 4 infants (26.1%) reported for maltreatment spent any time in foster care during the first year of life, and only a small fraction of infants who were reported for maltreatment died of medical causes during infancy, regardless of any time in foster care. However, the statistically significant finding that foster care protected against relatively rare medical-cause fatalities during infancy reinforces that foster care as a service intervention likely protects against other newborn health and developmental adversities. Infants in foster care may find themselves with caregivers who are better equipped to care for substance-exposed or medically fragile newborns. As much as this is true, the implication is not that more infants necessarily need to be placed in foster care to obtain this level of care. It does suggest, however, that more intensive case management and health care supports may be needed generally for infants remaining at home after ≥1 CPS report.

Notwithstanding the strength of the prospective design and population-based data used for this study, the interpretation of findings and conclusions are limited in several ways. We did not have information about the frequency of health visits during infancy, including preventive care and treatment of health problems, nor did we have information about any other services that may have been offered. Likewise, we were unable to classify preexisting health risks associated with risk of medical-cause deaths (and our covariates for other social and economic risks were limited to information recorded in vital birth records). That said, previous literature suggested that infants placed in foster care would be, on average, more medically fragile, yet we found that foster care was associated with lower death rates. Sensitivity analyses indicate our findings held, regardless of whether we included or excluded deaths during the neonatal period. We also confirmed that our findings did not change if we narrowed our focus to reports of maltreatment that were screened in for investigation. Finally, although almost 1 in 8 children in the United States is born in California,⁴³  the use of any single state’s data restricts the generalizability of findings because maltreatment reporting rates vary widely across jurisdictions and reflect different definitions of abuse and neglect and varying community thresholds for reporting an infant to CPS.¹ 

The increased risk of infant death from a medical cause associated with a previous report of abuse or neglect underscores the importance of ongoing care coordination between CPS and pediatric providers. Newborns in families involved with the CPS system reflect a highly vulnerable group, so it is not surprising that their rates of death are elevated relative to infants never reported. But our findings also suggest that during periods of foster care placement, infants may receive a level of case management and care coordination that protects against death and other health and developmental adversities. Differences in death rates signal potential unmet service needs among infants who remain at home after reports.

Although the findings reported and conclusions drawn are solely those of the authors and do not reflect those of any state agency, we wish to acknowledge partners from the California Department of Social Services and the California Department of Public Health. Their ongoing commitment to evidence-driven program and policy development is truly a model for public transparency and accountability. We also wish to acknowledge collaborating colleagues at the Children’s Data Network and the California Child Welfare Indicators Project.

CI

confidence interval

CPS

child protection system

HR

hazard ratio

ICD-10

International Classification of Diseases, 10th Revision