Cumulative risk effect of household dysfunction for child maltreatment after intensive intervention of the child protection system in Japan: a longitudinal analysis

Our research questions are as follows. First, does household dysfunction predict future child maltreatment occurrence? Relatedly, does MTM influence incidence of maltreatment reports? Although some studies have indicated a relationship between MTM and maltreatment recurrence [41], this relationship remains mostly unclear. Next, does household dysfunction predict future child maltreatment occurrence after adjustment of MTM? We hypothesized that the effect of household dysfunction on maltreatment recurrence would be mediated by current maltreatment status. The results would indicate that failure of casework does not only depend on the maltreatment status but also on the caregiver’s condition. Finally, does the cumulative effect of MTM and household dysfunction contribute to future maltreatment occurrence? This would help in identifying the usefulness of the ACE score in assessing the prognosis of child maltreatment casework.

The current study utilized a longitudinal, non-experimental design. This study was conducted as a part of a nationwide surveillance of temporary custody units in CGCs in 2014, which was in turn part of a child welfare research project.

CGCs are not only specially suited to handling child maltreatment cases but also dealing with the health and behavioral problems of children and caregiver’s adversity. When there is evidence of an urgent need to provide an alternative environment for a child, CGCs will place the child in a temporary custody unit and engage in investigation and treatment of the child and family. After cancelation of custody, the child either receives in-home services after sending back to their home or receives continuous treatment in out-of-home care settings (i.e., institutional or foster parental care). The subjects of the current study were children who had had entered temporary custody between August 1 and 31, 2014. We included only those stayed for more than 3 days, because important information is likely to be missed when children released within few days.

Before starting surveillance, all 207 CGCs in Japan, which spanned 47 prefectures, were contacted to request their participation. The objectives and procedures of the study were explained to the facility heads. One hundred and eighty facilities across 43 prefectures agreed to participate by returning the reply form (participation rate, 86.9%). The caseworkers in the participating facilities completed a questionnaire about the selected children’s cases on three time points, and then, they sent to researchers. The questionnaire assessed various aspects of case records, such as information on the selected child, their parent(s), family environment, history of CGC involvement, and the details of the current casework. Production of questionnaire and data collection were provided by the Aiiku Research Institute (previously the Japan Child and Family Research Institute).

The participants were children who had had entered temporary custody between August 1 and 31, 2014. We included only those who stayed for more than 3 days, because essential information is likely to be missed when children are released within few days. We collected information at three time points: first, about the day of the placement; second, about the day of the cancelation of temporary custody. According to the Japanese Child Welfare Act, “The period for temporary custody … shall not exceed 2 months …” [42]; therefore, when the child’s custody exceeded the 2-month period, we collected data on the 60th day of their placement. This yielded a total of 1081 possible cases. For the third point, the caseworkers completed a follow-up questionnaire on each index child’s condition in January 2015 (about 15 months after the child’s placement). As the government recommends that CGCs should continue providing services for at least 6 months [43], if the casework was terminated before January 2015, the follow-up information reflected their condition at the sixth month after the placement. The information was given for a total of 759 cases (70.2% of 1081 cases).

We selected six covariates from a literature review and the results of a bivariate analysis: age, sex (reference group was male), receipt of public assistance (proxy of economic status), amount of prior CPS involvement, type of post-cancelation treatment (i.e., in-home service or out-of-home service; the former was the reference), and length of stay in temporary custody. We believed that the probability of maltreatment occurrence would be lower if the length of stay was longer, because the observation period of this study was short.

All analyses were performed with R (version 3.2.0 [2015–04–16]; the R Foundation for Statistical Computing). The Cox proportional hazard model was used to investigate the relationship between the independent variables and incidence of maltreatment report after cancelation of temporary custody while controlling for the covariates. To validate the selection of covariates, we conducted a series of partial likelihood ratio and Wald tests according to the purposeful selection method provided by Hosmer et al. [45]. The proportional hazard assumption was evaluated using graphical diagnostics. Overall goodness-of-fit of each model was evaluated by comparing observed and expected numbers of events calculated from martingale residuals in ten groups based on risk score.

Most missing data resulted from CGC workers being unable to obtain sufficient information by the time to return the questionnaire. One hundred eighteen out of the 759 records (15.5%) were incomplete to some degree. Most missing data were for economic status of family (n = 37), followed by caregiver divorce (n = 17), and family structure (n = 7). To prevent data loss from listwise deletion, we filled in missing values using the single imputation method (SIM). To fill in the parent composition item, we coded missing values as a separate category (“unknown”). However, because this method might produce biased results, we employed the multiple imputation (MI) method to further validate the findings [46]. We performed 20 iterations for 30 multiple imputed datasets each. All MI calculations were performed in R using the default settings of the mice 2.25 package [47]. Model parameters were estimated with multivariate Cox regressions applied to each imputed dataset separately. These repeated estimates and their standard errors were combined into final estimates using Rubin’s rule, which ultimately yielded an MI dataset of 759 cases.

Table 1 describes the main characteristics of the sample. The mean age of the children at the time of placement was 10.6 years. Half of the children were boys, and 7.5% were non-Japanese. Six out of every ten (58.8%) children were from single-parent families, while around one fifth (20.2%) of children lived with both biological parents. One third (30.8%) of cases were from households receiving public assistance. Regarding reasons for placement, almost half (48.4%) had been placed in temporary custody for protection from maltreatment. The other common reasons were parenting problems (28.6%) and delinquency (14.6%). In total, children with history of any maltreatment accounted for 88.3%; 57.4% of children had a history of physical abuse, 71.3% had emotional abuse, 10.1% had sexual abuse, and 63.2% had neglect. More than one third (37.2%) of children had three types of maltreatment (MTM), in contrast with one eighth (12.8%) were 0—MTM. For household dysfunction, 46.6% of cases had reports of IPV, 43.5% had mental health problems, and 22.5% had substance abuse. Two thirds had some household dysfunction. The number of children with MTM + HD scores 0, 1, 2, 3, 4, 5, and 6 were 49, 79, 134, 197, 151, 116, and 33, respectively.

Overall, 76 of the 759 cases had reports of maltreatment within 15-month follow-up period. Bivariate Cox regression models with 30 multiple imputed datasets (Table 2) revealed that the following variables had significant relationships with this outcome: age (pooled hazard ratio [HR], 0.90, 95% confidence interval [CI] [0.85, 0.95]); interparental violence, caregiver mental health problem, and caregiver substance abuse (pooled HR, 1.74, 95% CI [1.10, 2.74]; 1.67, 95% CI [1.06, 2.63]; and 2.13, 95% CI [1.34, 3.40], respectively); physical abuse, emotional abuse, and neglect (pooled HR, 2.71, 95% CI [1.58, 4.65]; 3.08, 95% CI [1.54, 6.17]; and 2.92, 95% CI [1.61, 5.31], respectively); and use of the out-of-home post-temporary custody service (pooled HR, 0.21, 95% CI [0.11, 0.40]). History of sexual abuse and the amount of prior CPS involvement was marginally significantly associated with the outcome (p < .25).

Next, we constructed seven Cox regression models, each with one household dysfunction or child maltreatment variable as the independent variable and age, amount of prior CPS involvement before custody, duration of stay in custody, and post-temporary custody service as covariates. Table 3 shows the associations of each independent variable with the outcome, after adjusting for these covariates. Sexual abuse was the only variable without significant relationship with the outcome.

Then, we entered the HD and MTM scores (except for sexual abuse, given that no relationship with the outcome; Table 4). The HD score was entered as a four-level (0, 1, 2, and 3) categorical variable with 0 as the reference. For the MTM score, we combined levels 0 and 1 because of the small size; thus, it comprised three levels of 0 to 1, 2, and 3, with 0 to 1 as the reference. For the HD score, the pooled HR had a graded tendency with increasing level (HR was 2.39, 95% CI [1.26, 4.54] for score 2 and 3.17, 95% CI [1.45, 6.92] for score 3). Regarding the MTM score, the HRs for scores 2 and 3 were 2.41, 95% CI [1.09, 5.30] and 6.01, 95% CI [2.91, 12.41], respectively.

Finally, we entered the HD and MTM scores simultaneously (Table 5). When we adjusted for MTM in this model, HD score still significantly predicted the outcome, with the pooled HRs showing a graded tendency (HR for score 2 was 1.96, 95% CI [1.03, 3.73]; HR for score 3 was 2.35, 95% CI [1.07, 5.16]). Then, we added the two scores and entered them as a categorical variable, which was grouped into four levels according to sample size (0 to 1, 2, 3 to 4, and 5 to 6; the reference was 0 to 1). The summed HD score and MTM scores had also a significant relationship with outcome, and its pooled HRs showed a graded tendency with increasing level (HRs for scores of 3 to 4 and 5 to 6 were 5.00, 95% CI [1.53, 16.37] and 11.41, 95% CI [3.47, 37.52], respectively). Among the covariates, age (HR, 0.93, 95% CI [0.88, 0.99]) and a post-temporary custody service of out-of-home service (HR, 0.24, 95% CI [0.13, 0.47]) had significant relationships with the outcome.

We then conducted the above analysis using the single imputation method dataset that was coded missing values as a separate category, “unknown”. The HRs of the different levels of the variables exhibited a graded tendency and were larger than the HRs calculated using the SIM. For example, in the household dysfunction model (Additional file 1: Table S1), the HRs for scores of 2 and 3 were 2.31, 95% CI [1.21, 4.40] and 3.25, 95% CI [1.48, 7.12], respectively (in contrast, the HRs in the MI dataset were 2.39 and 3.17; see Table 4). Similarly, for the composite model (sum of the HD score and MTM scores, Additional file 1: Table S2), the HRs for scores 3 to 4 and 5 to 6 were 3.80, 95% CI [1.35, 10.74] and 8.40, 95% CI [2.94, 24.00], respectively (in contrast, the HRs in the MI dataset were 5.00 and 11.41; see Table 5). The above findings indicated that while all of the directions of the coefficients were the same, the size of the effects were larger in the composite model with the MI dataset.

Our findings have some implications. First, social workers might use them to provide risk-stratified interventions for maltreatment occurrence or recurrence according to the number of household dysfunction in each case. Moreover, the sum of the number of household dysfunction and maltreatment type and the analogue of ACE may be useful in risk stratification. Next, the child-welfare agency might be more confident in requesting coordination with agencies that provide services for adults. Furthermore, adult-focused services might be more proactive in promoting interventions for adults who exhibit risks of child maltreatment, even if the service is never involved with the index child directly. Because the Japanese CPS is highly burdened already, collaboration with agencies outside this system will be helpful. Focusing on caregiver problems would enable to use resources beyond the framework of the child welfare system. For example, the agencies dealing with mental health problems or substance abuse normally have regular contact with clients and thus can act as gatekeepers for child maltreatment. A Japanese study of 1492 patients with mental health problems including substance abuse found that 9% of patients were living with their children and 38% of them were rearing the child(ren) on their own [52]. For such patients, regardless of the possibility of maltreatment, proactive measures that help them engage in effective parenting and deliver ample social resources are needed. Researches have demonstrated the effectiveness of risk-focused parenting programs, and that parenting characteristics reduce the detrimental effects on children’s mental or behavior outcomes [53]. For IPV exposure, past research showed that resolution of IPV after CPS involvement was significantly associated with the child’s mental and behavior outcomes [54]. Despite that, a lack of coordination between CPS and IPV response agencies is not uncommon [55, 56]. In Japan, because treatment of IPV is beyond the competence of CGC, and workers tend to consider cases of witness IPV as “low risk” or safer than other life-threatening maltreatment incidents, their cases will be closed even without providing sufficient help to children or families.

Additionally, caregivers experiencing such adversity might have other vulnerabilities. Mothers engaging in substance abuse showed that CPS-involved mothers are likely to be younger, have a lower socioeconomic status, and have fewer interpersonal resources compared to those not involved in CPS [57]. Helping IPV victims requires structural interventions, including the contribution of economic, politico-legal, and social environmental factors, and a lack of social support was found to be associated with mental health and parenting functioning [58]. When we focus on these HDs, we should expand the links with other organizations that engage in distinct types of specialist area beyond child-welfare systems and construct multidisciplinary measures aimed at uninterrupted prevention and intervention of child maltreatment in providing effective casework.

A main limitation was the substantial reduction in sample size—namely, around 30% of the original sample was excluded after merging data from the entry and follow-up period. In this study, the information was obtained from separately conducted questionnaire survey in three time points and integrated into longitudinal dataset. Some CGC disagreed to participate at the third time point and that made reduce the sample size. Furthermore, some data had missing values. The small sample size and short observation period allow for limited conclusions. Constructing a comprehensive database for CGC-involved children and their families would allow for more precise analyses in this field. Additionally, because respondents were not actually the index child or family, the variables concerning their experience are likely to be inaccurate, and the severity of victimization was unknown. By obtaining this information in future research, we could provide a more precise guide in constructing effective practice to reduce the severity and chronicity of child maltreatment. In order to implement such research, we should construct self-report system targeted at victimized children and/or their family member other than perpetrators that might be feasible when we use computer-assisted interview that ensure the privacy of participants [59]. Finally, because the study sample was among the most vulnerable population in Japan, one must be cautious in generalizing the findings. Many children in this study were from adverse environment; one third received public assistance and more than half were from single-parent family and suffered from multiple adverse experiences. If we conducted investigation in children without temporary custody placement, the items related to family condition might have strong relationship to outcome, and the group of predictors would be different from this analyses. Despite these limitations, this study is valuable in its demonstration of the importance of number of household dysfunction in predicting the risk of child maltreatment after intensive intervention.