There is a strong and consistent link between residential caregiving and child psychopathology (Bos et al., 2011; Wade et al., 2018). Young children with a history of residential care show poor attention, hyperactivity, emotional dysregulation, elevated levels of anxiety, and higher rates of conduct disorders (Bick et al., 2018; Mackes et al., 2020; Stamoulis et al., 2015; Teyhan et al., 2018; Wade et al., 2019). The largest prospective epidemiological study of adults who were looked after or adopted as children demonstrated far worse psychosocial outcomes for those who experienced out-of-home social care as children compared to those who did not (Teyhan et al., 2018). Some of the mechanisms linking residential care to poorer psychosocial outcomes among children in residential care facilities (RCFs) include early social deprivation often because of high ratios of children to caregivers, high caregiver turnover, frequent isolation, disruptions in continuity of care, regimentation, and inadequate social and cognitive stimulation (van Ijzendoorn et al., 2020; Wade et al., 2018, 2019).

According to the 1989 UN Convention on the Rights of the Child, the family remains the best place for raising children. The 2007 United Nations Guidelines on Alternative Care emphasise and prioritise the family-based alternative care model for reunification. The global drive to deinstitutionalise children from RCFs to alternative care arrangements is shaped by a large body of research on the harmful effects of residential care, in terms of development, health, socialisation, mental health and violence (Sherr et al., 2017) almost exclusively from Europe and the US (Akister et al., 2010; Bos et al., 2011; Huynh et al., 2019; Sonuga-Barke et al., 2017; Wade et al., 2018). Randomised controlled trials (RCTs) of foster versus institutional care have shown that early foster care placement may avert the detrimental effects of institutionalisation on the psychological wellbeing of children in residential care (Bos et al., 2011; Wade et al., 2018; Zeanah et al., 2009). This strengthens the premise that quality of care may be more important than care setting in shaping the psychological outcomes of young people (Huynh et al., 2019).

Little is known about the effect of residential care on child psychopathology specifically in Uganda where approximately 500 RCFs are in operation, looking after an estimated 50,000 children. In the absence of effective government regulation in Uganda, less than 10% of these RCFs are approved to operate in compliance with government regulations (UNICEF Uganda, 2015). It is feared that unregulated RCFs in Uganda, particularly those that are not formally registered, may expose children to risks of extreme neglect that could potentially impact their physical, cognitive, and psychosocial health. The structural risks commonly identified by the Uganda Child Protection Sector include lack of qualified social workers, the poor processes involved in admitting children to care institutions, children overstaying in residential care, lack of guidelines and standard operating procedures for reunification and reintegration of children, potential for abuse and exploitation, and limited resources that may potentially undermine the quality of care (Ddumba-Nyanzi & Li, 2018; Walakira et al., 2015).

Substantial knowledge gaps remain unaddressed by the extant literature. First, the effects of residential care on child psychosocial functioning remain under-researched in low-resource settings. Second, the longitudinal trajectories of child psychopathology following reunification with family members in Sub-Saharan Africa are undocumented, limiting our understanding of whether reunification of children in this context is in their best interest. Do reunification programmes ensure that children are not being sent back to the very contexts that led to their placement in RCFs in the first place? Can a programme of support ameliorate the original drivers of such placement and enable reunification? The current evidence on the impact of deinstitutionalisation on children’s psychosocial health comes mainly from high-income countries and is conflicted with some studies reporting benefits (James et al., 2017; Wade et al., 2018) whereas others report a potential risk of harm (Bellamy, 2008; Pitula et al., 2019; Taussig et al., 2001).

Third, whether these trajectories vary with child age, sex and type of reintegration package received is not clear. Child age and sex have been shown to moderate global and domain-specific psychopathology among institutionalised children and their recovery patterns following foster care placement (Bos et al., 2011; Fox et al., 2013). Both RCTs and longitudinal studies report that boys manifest more psychiatric symptoms than girls regardless of caregiving environment and are less likely to recover following foster placement (Bellamy, 2008; Fox et al., 2013), yet cross-sectional studies report more internalising problems among institutionalised girls and older children than boys and younger children respectively (Nsabimana et al., 2019).

Fourth, there is conflicting evidence on which broadband syndromes are impacted by deinstitutionalisation. Some studies report worse internalising and total behaviour problems among reunified than non-reunified children and no group differences in externalising problems (Bellamy, 2008; Taussig et al., 2001), whereas others found significantly more externalising problems among institutionalised than community-raised children but no group differences in internalising problems (Nsabimana et al., 2019). Lastly, we do not know all the individual and contextual factors associated with children’s psychosocial health trajectories post-reunification and data are almost non-existent for Uganda. Individual-level factors associated with persistent psychopathology among reunified children may include child sex, age at reunification, orphanhood status and self-esteem (Nsabimana et al., 2019). Child resilience-promoting factors such as having adults to talk to may be important to consider.

Salient contextual factors identified from past studies include caregiving quality (Carter, 2005; Huynh et al., 2019) with high quality responsive caregiving shown to be protective (Pitula et al., 2019; Zeanah et al., 2009), caregiver mental health (Bellamy, 2008; Panter‐Brick et al., 2014), intimate partner violence (Bellamy, 2008; Leloux-Opmeer et al., 2016) and household poverty/socioeconomic status. In this study, household social economic status was dropped from multivariate models due to strong multicollinearity, but household food security and social security access were analysed as proxy indicators for household vulnerability. Additionally, we assessed the role of child and caregiver physical health both known to be associated with child psychosocial functioning (Butler et al., 2018; Cluver et al., 2013a, 2013b; Feldman et al., 2010).

The Uganda Government through projects led by the Ministry of Gender, Labour and Social Development (MGLSD) and its partners have committed to reducing the number of children in RCFs by implementing the UN alternative care framework to reunify and reintegrate children from residential to home-based care. Longitudinal studies examining the psychosocial trajectories of younger children reunified from residential care in Uganda have not been conducted yet findings from other countries may not directly relate to the local situation. Studies from Uganda show that many children are raised under stressful familial conditions with violence against children ranging from 44% in stable settings to 72% in post-conflict contexts (Ministry of Labour; Gender and Social Development, 2018; Natukunda et al., 2019). Poverty and violence against children, the two most documented drivers of poorer childhood psychosocial wellbeing with longstanding effects into adulthood are widespread in Uganda, suggesting that families of origin may not always be as safe for the psychosocial wellbeing of children in this context as often portrayed elsewhere.

In Kenya, two-thirds (66%) of children in RCFs had experienced at least one form of abuse prior to admission (Morantz et al., 2013), suggesting that the psychopathology experienced by children in RCFs may partly be attributed to adverse experiences of children before entering care. Additionally, a large multicentre study of institutionalised children in 5 low-resource settings (Cambodia, India, Kenya, Tanzania, and Ethiopia) found no significant differences in physical, mental, cognitive and behavioural health outcomes of children in residential versus community-based care (Whetten et al., 2009). To the best of our knowledge, this is the first study in Sub-Saharan Africa to investigate the psychosocial health trajectories of young children following reunification and reintegration from RCFs. Using data from a randomised controlled trial (RCT) evaluating the impact of a parenting programme on reunification success, we conducted longitudinal analyses to assess children’s psychosocial wellbeing while still at the RCF (baseline) and at 6- and 12-months post-reunification. The study aimed to (1) determine the psychosocial health trajectories of young children who are reunified from RCFs to home-based care in Uganda, (2) assess whether these trajectories vary with child age (younger vs older), sex and reunification package, and (3) test which factors predict persistent global and domain-specific psychosocial problems at 12 months following reunification.

Method

We conducted secondary analysis of data from the main RCT. Although these analyses were not prospectively planned with the main RCT, secondary analysis of RCT data is a widely used approach to leverage these rich resources of research data (Furberg & Friedman, 2012; Thylstrup et al., 2017), with findings ranging from conclusive to hypothesis-generating.

Participants

The main study was a randomised controlled trial (ClinicalTrials.gov Identifier: NCT03498469) evaluating the impact of a parenting program on successful reintegration of 1–13-year-old children from residential care back into family-based care in 9 districts in Uganda. Despite the initial planning for a larger sample (n = 640), recruitment generated only n = 77 child-caregiver pairs. The challenges of conducting such a study have been documented for the UK (Mezey et al., 2015). We have reported the challenges encountered in Uganda during the trial (S1 Text). Additional participant information is provided in the results section under baseline characteristics of the study sample.

Procedure

All cooperating RCFs (n = 40) in the districts of Mpigi, Mukono, Masaka, and the Greater Masaka districts of Rakai, Kyotera, Ssembabule, Kalungu, Lwengo, and Bukomansimbi were evaluated for inclusion in the study (Figure S1). Children aged 1–13 years living in participating RCFs and who could be reunified with family/kin in the study area were enrolled into the study. At each participating RCF, child records were reviewed by Case Managers to assess age eligibility (first stage). Those found to be age eligible (n = 897) entered the second stage requiring completion of a 2-page biodata form to capture children’s personal information (e.g. where and with whom the child lived prior to RCF placement, location of known relatives, presence of siblings in the RCF, etc.), and the details of admission into the RCF (e.g. who referred the child, date of admission, reason for admission, contact with family etc.). The second stage eligibility assessment required the child to have a known relative or caregiver with whom they could be reunified, caregiver willingness to receive the child, and residency within the study districts (See Figure S1 for further details).

CaseManagers traced the families of potentially eligible children to assess the family’s capacity and willingness for reunification. Children who could be reunified with the available support services and without any concerns regarding their wellbeing were reunified by the project following government approval. However, following assessment, several RCFs blocked the reunification of children and withdrew from participation (S1 Text). The major reason was potential loss of donor funding if children were reunified. This left only 98 children eligible for reunification. Of these, children that required alternative care arrangements including foster care, adoption, or independent/group home living were excluded along with children with severe developmental disabilities. Additionally, given the home-based nature of the study intervention, children whose caregivers planned to send them directly to boarding school upon returning from the RCF were excluded. A total of 21 participants were excluded leaving 77 participants.

Study participants were randomized to one of two arms: the comparison arm (n = 34) and the intervention arm (n = 43). Full details on the randomisation approach and tools used are presented in the supplementary material section (S2 Text). Participants assigned to the comparison arm received a standard reintegration package that included case management support, reunification preparation and a reunification cash grant of United States Dollars (USD)125 equivalent to Uganda Shillings (UGX) 469,925 at the time of the study, administered in two disbursements. The first disbursement was provided one month before reunification for families to complete the necessary purchases in advance of children’s arrival, and the second was provided one month after reunification. In the case of sibling reunification, additional funds were provided based on the number of siblings being reunified. The intention of the grant was to offset the costs a family was likely to incur because of the child returning home (e.g. bedding, clothing, school fees etc.). Those in the intervention arm received an enhanced reintegration package of services i.e. the standard package plus participation in a bi-weekly 13-session home-based parenting programme called ‘Esanyu Mu Maka’, loosely interpreted as ‘Happiness in the Home’. The parenting programme was delivered by the Case Managers who also linked families with critical services.

Data collection started on 7th December 2017 and ended on 1st October 2019. Interviews were conducted by trained interviewers using research material that had been translated and back-translated from English to Luganda (local language) and therefore available to participants in their preferred language. Data were collected at baseline (while the child was still living in the RCF), 6 months post-reunification (at the end of the parenting intervention), and 12 months post-reunification. Information about the child was collected from the child’s main caregiver at the RCF at baseline and from the caregiver with whom the child was reunified at 6 and 12 months. Children aged 8–13 years were interviewed at baseline and follow up time points. The primary endpoint of the trial was successful reintegration at 12 months post-reunification within family-based care. Overall, the main trial comprised of 77 child-caregiver dyadic pairs. This longitudinal study analysed caregiver-reported measures on children aged 6–13 years and includes child self-reported measures from children aged 8 years and over. The caregiver-reported Child Behaviour Checklist (CBCL) for 1–5-year-old children assessed different constructs from the version for older children therefore these were excluded, leaving 55 dyadic pairs relating to children aged 6–13 years at baseline. The study was reported in accordance with the Consolidated Standards of Reporting Trials (CONSORT) 2010 Statement for Reporting Parallel Group Randomised Controlled Trials (S1 Checklist).

Informed Consent and Ethical Clearance

The RCF Directors/Wardens provided permission for the eligible children and staff to participate in the study. Eligible child-caregiver pairs identified during the assessment phase were invited to participate in the study. In the case of multiple sibling reunification, only one was randomly selected to participate in the study and the reunification grant was increased per unified child per family. The RCF caregiver who had the most contact with the child and who would be providing baseline data on the child, and the primary home-based caregivers with whom the child was being reunified provided written informed consent on behalf of themselves and the reunified child. In addition, 8–13-year-old children assented to participate in the study. The RCF and home-based caregivers were compensated with 5000 Uganda shillings (approximately USD1.33) for their time to complete the interview whereas children were given a snack and a drink. Ethical clearance was provided by Mildmay Uganda (# REC REF 0503-2017) and the Uganda National Council for Science and Technology (SS 4351).

Materials

Where possible, child, caregiver and household measures were obtained using standardised scales that have been validated for use in low-resource settings. Where no validated tools were available for the study setting, we have provided references that allude to studies that have used the tool in Uganda or elsewhere in Sub-Saharan Africa. Several measures e.g. child abuse, parenting, perceived social support were based on the Sinovuyo Parenting Program developed by one of the co-authors (Cluver et al., 2016) and implemented by UNICEF in the region (Loening-Voysey et al., 2018). Internal validity (α) values from the baseline sample of children aged 6–13 years are presented throughout.

Measures

Child psychosocial problems were caregiver-reported using a condensed age-appropriate version of the Child Behaviour Checklist (CBCL–short form), asking about past-month problem behaviours (58 items, α = 0.82). Items were scored on a 3-point Likert-type scale as not true (0), somewhat true (1), or very true or often true (2) (Achenbach, 2000) with higher scores reflecting more problems. The scale assessed 4 of the 8 small-band syndromes intended by the original developers and was collapsed into 2 broadband behaviour domains as follows: internalising problems were a summed score of the anxious and withdrawn/depressed syndromes (α = 0.77), whereas externalising problems were a summed score of the aggressive and rule-breaking syndromes (α = 0.68). The two broadband scores were summed to obtain the total CBCL score, higher scores reflected more psychosocial problems.

The CBCL has been translated to 85 languages, cross-culturally adopted, and extensively used to assess problem behaviour among children and adolescents in many settings (Achenbach, 2010; Rescorla, 2007) including in Uganda (Bangirana et al., 2009) and similar contexts in Sub-Saharan Africa (Cluver et al., 2016; Lansford et al., 2018; Weisz et al., 1993). In any context particularly low-resource settings, it is impractical to intervene on the entire population of interest. Therefore, it is imperative to distinguish between individuals who may require immediate intervention from those who may not. We used the baseline mean scores as cut-off values to create and analyse each psychosocial outcome as a binary variable (having scores higher than the baseline average vs. not).

Child-related contextual factors included child physical health, resilience, depressive symptoms, loneliness and self-esteem measures. Child physical health was a composite of 2 subjective health items. The first item asked children to rate their overall health on a 5-point Likert scale from excellent (1) to poor (5), higher scores reflected poorer perceived health. The second item asked if children had felt too ill to participate in daily activities such as going to school or playing with friends. The 2 items were summed to obtain child physical health score with higher values reflecting poorer child physical health.

Child resilience used 4 child-reported items from the California Kids Health Survey (Hanson & Kim, 2007). Items were scored on a 4-point Likert-type scale from not true at all (0) to very much true (3) (α = 0.78), higher scores reflected more resilience-promoting factors. Child depressive symptoms assessment used a condensed version of the Children’s Depression Inventory (CDI2-Short, 12 items) (Kim et al., 2014; Politi, 2012), previously validated as a screening tool for child and adolescent depression and found to be equivalent to the original 27-item scale (Binagwaho et al., 2016; Kim et al., 2014; Kovacs, 1992). Items assessed past fortnight symptoms scored on a 3-point Likert-type scale from I am sad once in a while (0) to I am sad all the time (2). Items were summed into a child depressive symptoms scale with higher scores reflecting more depressive symptoms. The CDI was prorated for the number of items used in this study relative to the original scale (α = 0.62).

Child loneliness used 3 items from the shortened UCLA scale (Hughes et al., 2004), asking about of feelings of loneliness and isolation from others, scored on a 3-point Likert-type scale from rarely (0) to often (2) and summed to obtain a loneliness score (α = 0.94). Higher scores reflected more loneliness. Self-esteem assessment used 10 items from the Rosenberg Child Self-esteem tool (Rosenberg, 1965) e.g. “I am able to do things as well as most other children of my age,” scored on a 5-point Likert-type scale ranging from strongly disagree (0) to strongly agree (3) and I don’t know (8). For analysis, the ‘I don’t know’ responses (baseline n = 9) were set to strongly disagree (0), and all items were reverse-coded such that higher scores reflected lower self-esteem (α = 0.68). The depression, loneliness, and self-esteem scales were summed to obtain a score of total emotional problems with higher values reflecting more child emotional problems.

Caregiver-related contextual factors included caregiver employment, general health, stress, mental health, caregiving quality, intimate partner violence (IPV), and access to social and community support. Caregiver employment assessment used 2 questions. The first question asked about usual work rated on a 4-point Likert-type scale as full time (1), part time (2), occasional/seasonal (3), and none (4). The second question asked about past-month work for cash or in kind (yes/no). Caregiver employment was operationalised as any usual work (options 1–3) or past-month work for cash or in kind.

Caregiver general health assessment used 3 subjective items. One item asked caregivers to rate their current health on a 5-point Likert-type scale ranging from excellent (1) to poor (5) whereas 2 items asked if the caregiver’s health status had made it difficult for them to carry out activities of daily living. The latter items were scored on a 3-point Likert-type scale ranging from very difficult (1) to not difficult at all (3) and reverse-coded such that higher scores were worse. The 3 items were summed to obtain total caregiver general health scores (α = 0.71), higher scores reflected poorer perceived caregiver physical health. Caregiver stress used 18 items from the Parenting Stress Scale scored on a 5-point Likert-type scale from strongly disagree (1) to strongly agree (5) (Berry & Jones, 1995). Positively worded items assessing parental role satisfaction e.g. “I enjoy spending time with my child(ren)” were reverse-coded and all items were summed into a caregiver stress score (α = 0.65) with higher scores reflecting more caregiver stress. Caregiver mental health assessment used 14 binary items from the Shona Symptom Questionnaire (Patel et al., 1997) asking about past-week experience of symptoms e.g. “Did you at times feel like committing suicide?” Items were summed to obtain past-week caregiver mental health scores (α = 0.89), higher scores reflected more mental health symptoms.

Low-quality caregiving was operationalised as harsh caregiving practices and unresponsive caregiving. Harsh caregiving practices were assessed as past-month caregiver perpetration of physical and emotional violence against children, using an adapted caregiver version of the IPSCAN Child Abuse Screening Tool for use in trials (ICAST-Trial) (Meinck et al., 2018; Zolotor et al., 2009). Past-month physical violence used 9 caregiver-reported ICAST items (α = 0.65) whereas past-month emotional violence used 3 caregiver-reported ICAST items (α = 0.77). Unresponsive caregiving was assessed using 2 caregiver-reported subscales of the Alabama Parenting Questionnaire (APQ) (Elgar et al., 2007). APQ monitoring and supervision subscale used 10 items e.g. “How often does your child fail to let you know where he/she is going?” Items were scored on a 5-point Likert-type scale from never (1) to always (5) and summed to obtain the APQ monitoring and supervision score (α = 0.79), higher scores reflected poorer monitoring and supervision of the child. The APQ caregiver involvement subscale used 10 items e.g. “how often do you play games or do other fun things with your child?”, scored on a 5-point Likert-type scale from never (1) to always (5). Items were reverse-scored and summed to obtain the APQ involvement score (α = 0.86), higher scores reflected poorer caregiver involvement. Low-quality caregiving was a summed composite score of the ICAST and APQ subscales, higher scores were worse.

IPV exposure was assessed using 10 items from the Revised Conflict Tactics Scale Short Form, asking about past-month frequency of partner violence perpetration e.g. “My partner pushed, shoved or slapped me,” scored on a 4-point Likert-type scale from never happened (0) to more than 3 times in the past month (4) (Straus et al., 1996). Items were summed to obtain an IPV score (α = 0.84), higher scores reflected more frequent IPV exposure. Caregiver access to social and community support was assessed using 19 items from the MOS-SSS scale (α = 0.91) (Sherbourne & Stewart, 1991) asking about how often support was available to the caregiver and scored on a 5-point Likert-type scale from none of the time (1) to all of the time (5), and 3 additional items intended to assess community belonging (α = 0.74). The 22 items were summed to obtain total social and community support scores (α = 0.93), higher scores reflected more social and community support access.

Household-related contextual factors were food security and access to any social support provisions. Food security was assessed using 3 caregiver- and child-reported binary items from the Household Hunger Scale (HHS), asking about past-month access to enough food in the home e.g. “In the past 4 weeks, did you or any household member go to sleep at night hungry because there was not enough food?”, followed by 3 additional items asking participants to rate how often the event had occurred on a 3-point Likert-type scale from rarely (1) to often (3). The items were summed to create a household food security scale (α = 0.78) with higher scores reflecting greater food security. Household social support access was assessed using 9 items asking about household receipt of any of the following: food support, clothing, savings and lending groups, education support, healthcare support, agricultural inputs e.g. seeds, income generating activities, parenting support and mosquito nets. Items were summed to obtain a social support access score with higher scores reflecting more access. Scores were collapsed into a binary measure (any social support access).

Control variables were child age, child sex, type of reunification package, and study time point/data collection wave.

Data analysis

Analyses were conducted in stages in Stata 14/IC. Firstly, the baseline characteristics of participants were analysed by study arm. Positively skewed continuous variables were log-transformed and described using means and standard deviations. Secondly, to assess how caregiver-reported child psychosocial outcomes and other key child-related measures changed over time, we fitted separate linear mixed effects (multilevel) models using maximum likelihood estimation and reported robust standard errors for broadband (internalising, externalising and total problem scores) syndromes, and key child-related measures (physical health, resilience, depressive symptoms, loneliness, self-esteem, and total emotional problems) as dependent variables. Time (study timepoint)—a categorical variable with 3 levels (0 = baseline, 1 = 6 months, 2 = 12 months) was the explanatory variable. To test whether the longitudinal changes in child psychosocial outcomes were moderated by type of reintegration package, child sex, and child age, we fitted a 4-way interaction term to allow the effect of time (categorical) on child outcomes to vary by study arm (standard vs. enhanced intervention), child sex (coded 1 for girls and 0 for boys), and age group (6–9 years & 10–13 years). The grouping variable was the child identifier, and we did not specify the covariance structure. Following each fitted model, we run the ‘margins’ post-estimation command to obtain the predictive margins for the fixed portion of the model (main effects of study arm, child sex, and child age group), contrasted over the study time-point. This aimed to predict change in caregiver- and child-reported outcomes across the 3 waves of data collection for each of the analysis subgroups.

Thirdly, to analyse and compare the average marginal effects between time-points for each of the analysis subgroups (type of reintegration package, child sex, and child age), we used the ‘margins’ command and the dy/dx option attached to the study time-point variable. This provided comparisons of the average marginal effects at follow-up time-points compared to the baseline i.e. effects at 6 months and 12 months post-reunification compared separately to the baseline. To obtain the comparison between 6 and 12 months, the ‘margins’ command was rerun with the ‘ar.’ operator attached to the study time-point variable followed by the ‘@’ operator to obtain reverse adjacent contrasts of marginal linear predictions (comparing each time-point to the previous) separately for each analysis subgroup. Fourthly, to assess the nature of change in child psychosocial outcomes between study time-points (trends over time) i.e. compare the rates and patterns of change for each analysis subgroup, we used the ‘contrast’ post-estimation command with the ‘p.’ operator attached to our time factor variable (study time-point) to obtain coefficients of orthogonal polynomials. Orthogonal polynomial contrasts allow us to partition the effects of a factor variable into linear, quadratic, cubic, and higher-order polynomial components. The ‘p.’ operator applies orthogonal polynomials using the values of the factor variable. Because our time factor variable (study time-point) had only 3 levels, only the linear and quadratic effects of time could be tested. The @ post-estimation operator was used to obtain contrasts of the orthogonal polynomial effect of time on each outcome of interest by study arm, child sex and age group. This analysis automatically applies a chi2 test of trend assessing whether the change followed the predefined trajectories, in our case linear or quadratic.

Fifthly, using the binary outcomes, we fitted logistic regression models and post-estimation marginal effects modelling to predict the proportions of children scoring above the baseline mean scores for each psychosocial health outcome at each time point. The predicted margins were plotted in Excel and graphically presented. Lastly, we fitted 3 separate linear mixed effects models to simultaneously test the influence of individual level and contextual factors on the changes in child psychosocial outcomes (internalising, externalising and total CBCL scores) through time (over the study period) adjusted for child age, and 3-way interaction between child sex, type of reintegration package received and study time point. Given that our analyses were based on a small sample of reunified adolescents, we did not adjust for multiple comparisons.

Results

Baseline characteristics of the study sample stratified by study arm

The baseline characteristics of the study population are summarised in Table 1. The mean age of children at baseline was 9.75 (2.16) years, 58.2% were boys, 81.8% had at least one living parent by caregiver report, child physical health score mean = 2.91 (1.39), range: 1–6) with 36 (81.8%) of those who were asked about their current health reporting being in good to excellent health. On average, children reported high levels of depressive symptoms (mean = 10.98 (5.07), range: 0–27) with 21 (47.7%) scoring above the baseline mean; modest levels of loneliness (mean = 2 (1.92), range: 0–6) with 15 (34.1%) scoring above the mean; and high levels of low self-esteem (mean = 6.50 (4.12), range: 0–15) with 23 (52.3%) children scoring above the baseline mean. Children had high total emotional problem scores at baseline (mean = 19.58 (8.20), range: 7–47) with 21 (47.7%) children scoring above baseline mean levels. Regarding the CBCL, RCF caregivers reported high levels of internalising problems (mean = 10.8 (4.93), range: 1–21), externalising problems (mean = 9.18 (7.84), range: 0–37) and total behavioural problems (mean = 19.98 (10.66), range: 4–54) among children at baseline. The number of children scoring above baseline mean levels for internalising and externalising problems was 23 (45.1%) and 17 (33.3%) respectively. However, children reported high levels of resilience-promoting factors at baseline (mean = 9.98 (2.41), range: 0–12) (Table 1).

Table 1 Baseline characteristics of the study sample by study arm
Full size table

Among caregivers at baseline, only 22 (40%) had worked for cash or kind in the past month. Baseline caregiver subjective health was poor (mean = 6.49 (2.07), range: 3–11); caregiver-reported stress (mean = 38.08 (7.57), range: 22–55), mental health symptoms (mean = 6.36 (3.91), range: 0–14), low quality caregiving (mean = 58.96 (12.49), range: 30–86), and IPV scores (mean = 11.76 (11.49), range: 1–48) were all high at baseline. Caregivers reported high levels of social and community support access (mean = 83.85 (14.49), range: 45–110) at baseline. Of note, baseline household food security was almost universal (94.6%), and 24 (43.6%) of the households had accessed at least one government/NGO social security provision at baseline. Children, caregivers, and households receiving the enhanced reintegration package did not significantly differ from those receiving the standard package at baseline (Table 1).

Trajectories of child psychosocial outcomes by child age, sex, and study arm

CBCL internalising problems

Compared with their baseline, children had lower internalising symptom scores at 6- and 12-months post-reunification. The longitudinal change in internalising symptoms did not differ by child age or child sex or study arm (Table 2). However, the child sex by study arm interaction term was significant in the main model (β = 5.70, CI: 0.96–10.43, p = 0.018) (S1 Table). Findings from post-estimation marginal effects modelling showed that girls in the enhanced intervention arm had significantly higher internalising symptom scores through time than boys in the enhanced intervention arm: 8.89 (0.74) vs. 6.09 (0.63), p = 0.004 respectively whereas internalising symptoms did not significantly vary between girls and boys in the standard package group: 7.70 (0.75) vs. 7.85 (0.71), p =0.885 respectively (Fig. 1). Findings from reverse adjacent comparisons of average marginal effects between timepoints (t1 to t0; t2 to t1) show significant within-group reductions in caregiver-reported internalising symptoms for all the analysis subgroups except between 6- and 12-months post reunification where further reduction in internalising problem scores was not significant for children in the standard intervention and for younger children (Table 3). Comparisons between baseline and 12 months post-reunification show significant reductions in caregiver-reported children’s internalising symptoms for all analysis subgroups (Table 3). The Chi-square tests of trend contrasted by child age, sex, and study arm show that the longitudinal change in children’s internalising symptoms between baseline and 12 months followed a significantly linear pattern, equally for younger vs. older, boys vs. girls, and the enhanced vs. standard intervention groups. The quadratic trends in caregiver-reported internalising symptoms were not significant for all analysis groups (Table 4).

Table 2 Adjusted predictive margins (standard errors) for child psychosocial outcomes overtime, disaggregated by child age, sex and study arm
Full size table
Fig. 1
figure 1

The effect of type reintegration package on child psychosocial problems at 12 months was moderated by child sex

Full size image
Table 3 Comparing the rates of change in child psychosocial outcomes between study timepoints by study arm, child sex and child age
Full size table
Table 4 Post-estimation analysis of the nature of change in child psychosocial outcomes from baseline to 12 months post-reunification
Full size table

CBCL externalising problems

Compared with their baseline, children had lower externalising symptom scores at 6- and 12-months post-reunification. The longitudinal change in externalising symptoms did not differ by child sex or study arm at any of the timepoints. However, younger children had fewer externalising problems than older children at 12 months, 2.91 (0.49) vs. 5.74 (0.81) (Table 2). Post-estimation contrasts of predictive margins by children’s age group at each time point show that the marginal difference in externalising symptoms between older and younger children at 12 months was statistically significant, diff (SE) = 2.83 (1.01), p = 0.005 (S2 Table).

Findings from reverse adjacent comparisons of average marginal effects between timepoints (t1 to t0; t2 to t1) show within-group reductions in caregiver-reported externalising symptoms for all the analysis subgroups at 6 months compared to baseline but these reductions were not statistically significant (Table 3). Comparisons of average marginal effects between 6- and 12-months post reunification show further within-group reductions in caregiver-reported externalising problems that were significant for all analysis groups except the standard intervention (Table 3). Comparisons from baseline to 12 months post-reunification show significant reductions in caregiver-reported externalising symptoms for all analysis subgroups (Table 3). The Chi-square tests of trend contrasted by child age, sex, and study arm show that the longitudinal change in children’s externalising symptoms between baseline and 12 months followed a significantly linear pattern equally for younger vs. older, boys vs. girls, and the enhanced vs. standard intervention groups (Table 4).

CBCL total psychosocial problems

Compared with their baseline, children had lower global CBCL scores at 6- and 12-months post-reunification. The longitudinal change in total behaviour problem scores did not differ by child sex or study arm at all time-points. However, younger children had fewer total behaviour problem scores than older children at baseline, 17.77 (1.32) vs. 21.39 (2.00) and at 12 months, 6.84 (1.19) vs. 10.65 (1.23) (Table 2). Post-estimation contrasts of marginal effects by children’s age groups at each time-point show that these differences were not statistically significant (p = 0.1) and (p = 0.058) at baseline and 12 months post-reunification, respectively. Post-estimation tests for interaction between child sex and study arm indicate that girls receiving the enhanced intervention package had significantly higher global problem scores than boys receiving the standard package, adjusted difference = 4.72 (1.65), p = 0.004 whereas there was no difference between girls and boys receiving the standard package (p = 0.443) (Fig. 1).

Findings from reverse adjacent comparisons of average marginal effects between timepoints (t1 to t0; t2 to t1) show significant within-group reductions in caregiver-reported global problem scores for all the analysis subgroups except between baseline and 6 months where the reduction in scores was not statistically significant for boys and children in the enhanced intervention group (Table 3). Comparisons between 6 and 12 months show further reductions in scores that were statistically significant for all subgroups except the standard intervention. Comparisons between baseline and 12 months post-reunification show significant reductions in caregiver-reported global problem scores for all analysis subgroups (Table 3). The Chi-square tests of trend contrasted by child age, sex, and study arm show that the longitudinal change in children’s global problem scores on the CBCL between baseline and 12 months followed a significantly linear pattern, equally for younger vs. older, boys vs. girls, and the enhanced intervention vs. standard package (Table 4).

Trajectories of self-reported emotional problems by age, sex and study arm

Compared with their baseline, children had significantly lower depressive symptoms at 6 months and 12 months post-reunification in all groups except for those receiving the enhanced intervention among whom scores remained relatively unchanged i.e. mean (SE) = 9.91 (0.94), 9.34 (0.95) and 9.90 (0.95) at baseline, 6 and 12 months, respectively. Child loneliness and low self-esteem scores were lower at 6 months and at 12 months compared with baseline scores in all groups. Total emotional problem scores were significantly lower at 6 and 12 months from baseline in all groups except for children receiving the enhanced intervention for whom scores declined between baseline and 6 months from 17.87 (1.47) to 15.71 (1.48), with no further improvement at 12 months, 15.76 (1.48). There were no significant between group differences in total child emotional problems at all timepoints (Table 2).

Findings from reverse adjacent comparisons of average marginal effects between timepoints (t1 to t0; t2 to t1) show significant within-group reductions in child-reported total emotional problems for all the analysis subgroups except between baseline and 6 months where the reduction in scores was not statistically significant for younger children, girls, and children in the enhanced intervention group (Table 3). Further reductions in total emotional problem scores between 6 and 12 months were not statistically significant for all subgroups whereas comparisons between baseline and 12-months post-reunification show significant reductions in child-reported total emotional problems for all analysis subgroups except the enhanced intervention (Table 3). The Chi-square tests of trend contrasted by child age, sex, and study arm show that the longitudinal change in children’s total emotional problem scores between baseline and 12 months followed a significantly linear pattern, for children in the standard but not enhanced intervention; for boys and girls but with a more significant linear trend for boys than for girls; and for both younger and older children but less significant for the younger children.

Analysing child psychosocial outcomes as binary variables

We fitted logistic regression models for each binary psychosocial outcome including a 4-way interaction term for study timepoint, age, sex and intervention package, and used post-estimation marginal effects modelling to predict the proportion of children scoring above baseline mean levels at each timepoint. Plotting the resultant predictive margins showed that the proportion of children scoring above the baseline (pre-reunification) averages significantly declined at each follow-up (Fig. 2). These findings confirmed the linear trajectories of change in child psychosocial outcomes presented in Table 4.

Fig. 2
figure 2

Trajectories of children’s psychosocial health outcomes within 12 months post-reunification

Full size image

Factors associated with persistent psychosocial problems among reunified children

Internalising problems

Independent of covariates, findings from linear mixed effects modelling show that past-month caregiver employment was associated with fewer internalising problems (β = −11.41, 95% CI: −18.04 to −4.79, p < 0.001), whereas poorer caregiver mental health (β per unit increase = 0.38, 95% CI: 0.19 to 0.58, p < 0.001) and low-quality caregiving (β per unit increase = 0.08, 95% CI: 0.01 to 0.14, p = 0.018) were associated with more internalising problems among reunified children through time (Table 5). The 3-way interaction effects on persistent internalising problems were significant but we did not attempt to interpret these findings owing to sample size limitations (S3 Table).

Table 5 Factors associated with persistent child psychosocial problems among reunified children
Full size table

Externalising problems

Independent of covariates, findings from linear mixed effects modelling show that higher child depressive symptoms (β per unit increase = 0.41, 95% CI: 0.12 to 0.70, p = 0.006), poorer caregiver general health (β per unit increase = 1.65, 95% CI: 0.89 to 2.41, p < 0.001), and low-quality caregiving (β per unit increase = 0.20, 95% CI: 0.09 to 0.31, p < 0.001) were associated with more externalising problems among reunified children through time (Table 5). The 2-way interaction term between study time-point and study arm was significant but we did not attempt to interpret these due to sample size limitations (S3 Table).

Global psychosocial problems

Independent of covariates, findings from linear mixed effects modelling show that higher child depressive symptoms (β = 0.46, 95% CI: 0.07 to 0.84, p = 0.020), poorer caregiver general health (β per unit increase = 1.98, 95% CI: 0.97 to 2.99, p < 0.001) and low-quality caregiving (β per unit increase = 0.25, 95% CI: 0.10 to 0.39, p < 0.001) were associated with more global psychosocial problems among reunified children through time (Table 4). We did not find protective effects of household food security, access to any cash or in-kind provisions, and caregiver access to social and community support on child psychosocial functioning post reunification, after controlling for other factors (Table 4). The 2-way (except study timepoint by child sex) and 3-way interaction terms were significant, but we did not attempt interpretation due to sample size limitations (S3 Table).

Discussion

Documenting the trajectories of psychosocial problems and predictors of such trajectories among children who are reunified with their families from residential care is important for our understanding of the effect of out-of-home care on the psychosocial wellbeing of children and whether this is amenable to intervention. Using a longitudinal design, this study aimed to determine the psychosocial health trajectories of reunified children, assess whether these trajectories varied with child age, sex and type of reintegration package received, and describe the factors associated with persistent psychosocial problems post-reunification. We found persistently declining and significantly linear trajectories in all the small band syndromes (rule-breaking, aggressive, anxious, and withdrawn/depressed symptoms) and the broad band syndromes (internalising, externalising and global child psychosocial problems) at 6- and 12-months post-reunification. Our findings are concordant with those from past RCTs in other countries testing the impact of family-based care placement on the psychosocial recovery of young children compared with those remaining in institutional care (Carter, 2005; Johnson et al., 2006; van Ijzendoorn et al., 2011).

We observed intervention group-by-sex interaction effects on children’s internalising problems through time i.e. significantly higher internalising problems among girls than boys receiving the enhanced reintegration package but no sex differences in internalising problems among children receiving the standard package. Though interpretation is limited by our small sample, these findings indicate that the add-on parenting programme effectively lowered internalising problems for boys but not for girls. There are several potential explanations for this finding: 1) the phenomenon that biological sex and its social ramifications may modify intervention effects i.e. interventions may not achieve the same effect for young boys and girls is well-established (Cluver et al., Cluver et al., 2013b; Cluver et al., 2016c). 2) The impact of residential care on girl’s internalising symptoms may be deeply rooted and less amenable to the effects of a parenting programme. 3) The 12-month follow-up period may have been too short to observe significant change in internalising symptoms among girls receiving the enhanced intervention. However, our findings are concordant with those from cross-sectional studies in the African setting (Nsabimana et al., 2019) but discordant with those from RCTs and longitudinal studies in high-income settings in which male gender was associated with more internalising problems among reunified children (Bellamy, 2008; Zeanah et al., 2009). Differences in study designs and settings hinder meaningful comparisons. The mechanisms underlying these gender differences in internalising symptoms that seem less amenable to parenting interventions remain unclear, but researchers suggest that girls may exhibit more risk factors for internalising symptoms including genetic, personal and environmental factors (e.g. sexual trauma and other forms of violence), which may exacerbate risk in the face of adversity (Mutumba et al., 2016). Therefore, reunified young girls may require additional support and longer recovery periods to readjust. Persistent internalising symptoms are a known precursor for future psychiatric disorders (Durbeej et al., 2019).

Trajectories of externalising problems did not differ by gender, but older adolescents (10–13 years) exhibited more externalising problems than younger adolescents (6–9 years). This finding was in line with theory considering that older adolescents will have started to form their own opinions and therefore more likely to resist conformation to expectations. This resistance to conformation may manifest as externalising behaviour. Total problems were more among girls than boys in the enhanced intervention group with no observed gender differences among those in the standard intervention arm. This finding was expected from the contribution of more internalising problems seen among girls than boys in this group but contradicted the findings of Wade and colleagues who reported that the male gender was associated with more externalising and total problems (Wade et al., 2018). Notably, the two studies were different in terms of the alternative care arrangements i.e. foster care versus family of origin/kinship care, and this difference may explain the disparate findings. In theory, it was expected that children whose parents undertook the parenting programme would manifest fewer problem behaviours, but we observed higher internalising and total problems among girls than boys receiving the enhanced package. Why the parenting programme worked for boys and not girls in our sample warrants further investigation.

Of the contextual factors tested in this study, past-month caregiver unemployment, poorer caregiver mental health and low-quality caregiving were associated with internalising symptoms among reunified children through time, indicating that household vulnerability factors are important in the emergence and persistence of youth psychopathology. Surprisingly, self-reported depressive symptoms were not associated with internalising problems in our study even though the construct included withdrawn/depressed traits. This can be attributed to differences in respondents given that the CBCL psychosocial outcomes were caregiver-reported whereas the CDI was child-reported. Child depressive symptoms, caregiver general health and low-quality caregiving were associated with externalising symptoms through time. The relationship between depression and adolescent externalising problems has been previously reported in the general population and among those with a history of residential care (Brensilver et al., 2017; Loth et al., 2014; Vinnakota & Kaur, 2018). Factors associated with persistent psychosocial problems among reunified children were caregiver general health and low-quality caregiving, both consistently known to shape children’s psychosocial functioning across geographical space and time (Cluver et al., 2009; Cluver et al., 2013a, 2013b; Nyamukapa et al., 2010). Low-quality caregiving remained consistently associated with internalising, externalising and total behaviour problems among reunified children through time, implying that returning children to families of origin where indicators of household vulnerability such as poorer caregiver physical and mental health, unemployment and low-quality caregiving are prevalent may not be in the children’s best interest. The latter findings were consistent with those from a large rigorous well-controlled study conducted across five low-resource settings, in which the authors reported null (0%) variance in youth psychosocial functioning explained by care setting (residential vs. community) whereas 8.9% variance was explained by caregiving quality (Huynh et al., 2019). Taken together, these findings underscore the importance of caregiving quality rather than care setting in shaping children’s psychosocial health and functioning.

Our analyses have limitations with which the findings should be interpreted. First, the ethics of conducting such a study required that all reunified children received a fairly comprehensive package. This study can only explore the benefits of the add-on parenting component beyond those of a standard package, which seemed to make little difference in improving children’s psychosocial health post-reunification. Second, the sample size was small and perhaps underpowered to detect some effects. Moreover, our findings may have been skewed by those that rejected reunification. Despite this there was a clear pattern of improvement for all children post reunification which endured and continued to improve from 6–12 months. In our multivariate analyses, we tested 4-way interaction effects which require a larger sample size, but we used linear mixed effects modelling which tends to be robust to the statistical effects of small samples. Nonetheless, the factors associated with persistent psychosocial problems identified in our study are in line with those from larger and similarly powered past studies and theoretical expectations. Our small sample is not representative of the entire population of institutionalised children in Uganda therefore our findings can only be extrapolated to reunified children in participating districts. Third, the short follow-up period limits our understanding of whether the observed recovery persists beyond the study period. However, using three waves of data collection, we showed persistently declining trajectories in child psychosocial problem scores at each timepoint hence eliminating the play of chance. Fourth, we used self-reported measures which are prone to recall and social desirability bias. Fifth, we did not assess biological/genetic factors which may influence risk or resilience to psychosocial problems among institutionally raised children. Lastly, our study did not use control groups (e.g. non-reunified or never-institutionalised community children) to delineate the effects of reunification or residential care on children’s psychosocial functioning, therefore, our findings may bear the effects of selection bias. Reunified children may be systematically different from those who remain in care. However, the longitudinal sample recruited from a RCT and pre-post design in which reunified children acted as their own controls i.e. assessed while still at the RCF and at 6- and 12-months post-reunification provided powerful tools in assessing the effect of deinstitutionalisation on children’s psychosocial health.

Despite these limitations, our findings have several important clinical and policy implications as they highlight the protective role of deinstitutionalisation on youths’ mental health and identify sources of persistent vulnerability. We tested a comprehensive set of contextual factors, and our findings signify that reunification of children from residential care is beneficial within the confines of favourable familial circumstances. Interventions to prevent enduring psychosocial problems among reunified children are urgently needed. Such interventions may target contextual factors such as children’s concurrent depressive symptoms, caregiver physical and mental health, increasing the quality of caregiving practices such as positive parenting, and improving livelihoods through poverty alleviation programmes. Larger studies are needed to validate our findings and future longitudinal studies may benefit from following up children who remain in care as a control group.