Non-Suicidal Self-Injury in Pediatric Bipolar Disorder: Clinical Correlates and Impact on Psychosocial Treatment Outcomes

Non-suicidal self-injury (NSSI) in childhood is not well documented, especially among youth with pediatric bipolar disorder (PBD). The current study evaluated prevalence and correlates of NSSI, and its impact on intervention response, in a randomized trial of Child- and Family-Focused Cognitive-Behavioral Therapy (CFF-CBT) versus Treatment As Usual (TAU), adjunctive to pharmacotherapy. This study included 72 children ages 7–13 (58% male) with PBD. NSSI and correlates were assessed at baseline; mood and psychiatric severity were measured longitudinally. NSSI was common: 31% endorsed NSSI behaviors; 10% reported thoughts of NSSI, in the absence of behaviors. Children engaging in NSSI reported higher depression, psychosis, suicidality, and hopelessness; lower self-esteem; and reduced family help-seeking in univariate analyses. In a multivariate logistic regression, high child depression and psychosis, and low family help-seeking, remained significantly associated with baseline NSSI. In mixed-effects regression models, presence of NSSI at baseline did not influence the response of depressive symptoms to treatment. Children who endorsed NSSI experienced steeper response trajectories for psychiatric severity, regardless of treatment group. Youth who denied NSSI showed poorer response to TAU for manic symptoms; mania trajectories in CFF-CBT were similar across youth. Thus, NSSI in PBD is common and associated with impairment. As children might engage in NSSI for different reasons, the function of NSSI should be considered in treatment. Since children without NSSI fared worse in TAU, it may be important to ensure that youth with PBD receive structured, intensive interventions. CFF-CBT was efficacious regardless of NSSI, and thus shows promise for high-risk children with PBD.


Pediatric Bipolar Disorder
Pediatric bipolar disorder (PBD) is a serious condition associated with significant morbidity and mortality. PBD affects approximately 1-2% of children and adolescents (Van Meter et al. 2011). Compared to adults with bipolar disorder, youth with PBD spend more time symptomatic with mixed depressive and manic presentations, rapid mood fluctuations, and subthreshold symptoms Geller et al. 2008). Early-onset of bipolar disorder is also associated with greater functional impairment and poorer quality of life (Perlis et al. 2009). In addition, youth with PBD often present with psychotic features and psychiatric comorbidity Geller et al. 2008;Van Meter et al. 2016), psychosocial impairment ), and familial dysfunction (Keenan-Miller et al. 2012).
Importantly, these children and adolescents are at increased risk for suicide. According to a recent systematic review calculating weighted mean prevalence rates across 13 studies (N = 1508), 50-60% of youth with PBD report suicidal ideation (SI), and 20-25% attempt suicide (Hauser et al. 2013). Findings from three studies indicate that SI in PBD is associated with Caucasian race, younger age, narrow PBD phenotype (youth with hypomanic or manic episodes, and hallmark symptoms of elevated mood or grandiosity), more severe depressive symptoms, hopelessness, low self-esteem, family rigidity, and poor quality of life (for review see Hauser et al. 2013;Weinstein et al. 2015a). In contrast, significant correlates of suicide attempts in PBD reported in 10 studies include female gender, older age, earlier onset of illness, more severe/episodic PBD, mixed episodes, comorbid disorders, prior non-suicidal self-injury (NSSI) and suicidality (SI and attempts), physical/sexual abuse, parental depression, family history of suicidality, and familial dysfunction (for review see Hauser et al. 2013). Thus, youth with PBD represent a highrisk group for which additional research on suicidality and associated predictors is needed.
Non-Suicidal Self-Injury NSSI, or the deliberate destruction of one's body without suicidal intent (for review see Nock 2010), is one such correlate that is associated with suicidality in community and clinical samples broadly (for review see Jacobson and Gould 2007;Stewart et al. 2017), and in PBD specifically. NSSI is more prevalent in youth with PBD who have a history of suicide attempts compared to those who do not (Algorta et al. 2011;Goldstein et al. 2005). NSSI was also more common among youth with PBD who attempted suicide prospectively over 5 years (Goldstein et al. 2012). Interestingly, emerging evidence suggests that NSSI may be a stronger predictor of suicidal behavior than prior attempts in depressed adolescent samples Wilkinson et al. 2011). Thus, the link between NSSI and suicidality is strong and well documented, especially among youth with mood dysregulation. While the exact mechanisms linking NSSI to suicidal behaviors remain unclear, according to the interpersonal theory of suicide, NSSI may facilitate an acquired capability for suicide via habituation to the pain and fear involved in suicide attempts (Joiner 2005). This theory has recently received empirical support in clinical samples of adolescents (Horton et al. 2016;Stewart et al. 2017).
In addition to being a risk factor for future suicidal behaviors, research suggests that NSSI is a dangerous and distinct phenomenon in its own right (for review see Muehlenkamp 2005). Prevalence of NSSI in heterogeneous youth samples ranges from 13% to 23% in the community (for review see Jacobson and Gould 2007) and 40% to 60% in clinical settings (for review see Barrocas et al. 2011;DiClemente et al. 1991). In addition to suicidality, NSSI is associated with female sex, peer NSSI, hopelessness, emotion dysregulation, alexithymia, dissociation, internalizing and externalizing psychopathology, borderline personality disorder symptoms, eating disorder symptoms, substance use, and physical/sexual abuse (Fox et al. 2015; for review see Jacobson and Gould 2007). Regarding functions, youth appear to engage in NSSI for intrapersonal negative reinforcement (to reduce tension or negative emotions), intrapersonal positive reinforcement (to create a desirable physiological state or punish oneself), interpersonal negative reinforcement (to escape interpersonal tasks/demands or stop bullying/fighting), and interpersonal positive reinforcement (to gain attention, support, or access to desired materials; for review see Nock 2010;Nock and Prinstein 2004). Intrapersonal negative reinforcement is the most commonly endorsed function (Nock et al. 2009).
Although NSSI has received less empirical attention than suicidality, recent research offers improved conceptualization of this behavior in youth. Nevertheless, further investigation is needed. Research with younger, high-risk samples is crucial, as limited work has evaluated NSSI in prepubertal children (for review see Barrocas et al. 2011). Also, early identification and intervention among those predisposed to NSSI may prevent its persistence and the development of suicidality in adolescence and adulthood.

Non-Suicidal Self-Injury in Pediatric Bipolar Disorder
As aforementioned, youth with PBD are at increased risk for both NSSI and suicidality (Algorta et al. 2011;Goldstein et al. 2005;Goldstein et al. 2012). While at least 15 studies have examined either SI or suicide attempts in PBD (for review see Hauser et al. 2013;Weinstein et al. 2015a), to our knowledge, only one study has examined correlates and potential functions of NSSI in PBD. In a sample of 432 youth with PBD, Esposito-Smythers et al. (2010) reported NSSI prevalence of 22% during the most recent mood episode, and a lifetime prevalence of 34-37%. Bipolar I and II disorders, a mixed episode, depression, separation anxiety, psychosis, suicide attempt, and psychosocial dysfunction were associated with NSSI. The authors noted that these correlates were consistent with the functional model of NSSI (for review see Nock 2010;Nock and Prinstein 2004). Specifically, they concluded that youth with illness features associated with high or low emotional reactivity (bipolar I or II disorders, mixed episode, depression, psychosis, separation anxiety, suicide attempt) might have engaged in NSSI for intrapersonal negative or positive reinforcement. In contrast, they suggested that youth with psychosocial dysfunction potentially used NSSI for interpersonal negative or positive reinforcement.
Thus, prevalence of NSSI was high in this sample of youth with PBD. In addition, NSSI was associated with unique predictors not commonly documented in heterogeneous community or clinical samples. Therefore, additional research on NSSI in PBD is essential, as it is relatively common, dangerous, and related to suicidality. Better understanding of the subset of youth who engage in these high-risk behaviors also has important clinical implications. Specifically, improved conceptualization of NSSI can indicate treatment targets, and importantly, signify youth who may be susceptible to NSSI, suicidality, or both.
Impact of Non-Suicidal Self-Injury on Psychosocial Treatment Outcomes Given increased rates of suicidality and NSSI in youth with mood disorders, and PBD specifically, it is also important to examine whether these deleterious behaviors influence psychosocial intervention response, as such information can inform clinical decisions and treatment planning. However, few studies have examined the impact of these behaviors on treatment outcomes. Limited research suggests that both SI (Asarnow et al. 2009;Curry et al. 2006;Rohde et al. 2006) and NSSI (Asarnow et al. 2009;Vitiello et al. 2011) are associated with worse outcomes across groups in randomized controlled trials (RCTs) of Cognitive-Behavioral Therapy (CBT) for adolescents with depression. However, one study found evidence of moderation via enhanced response of suicidal, depressed adolescents to CBT versus non-directivesupportive therapy (Barbe et al. 2004). These results suggest that depressed youth with SI and NSSI generally fare worse in treatment, and thus may require more supports and interventions that target these behaviors directly.
Even less research has examined the predictive and moderating effects of suicidality and NSSI in psychosocial treatments for PBD. One study found that SI did not impact intervention response for PBD (Weinstein et al. 2015b); however, no studies have specifically examined NSSI. As such, more research is needed to better understand the role NSSI plays in intervention response for children with PBD. These insights can signify those who may require therapeutic provisions and specialized treatments for suicidality and NSSI, such as Dialectical Behavior Therapy (Goldstein et al. 2015).

Current Study
Given limited research on NSSI in prepubertal children in general and PBD in particular, and lack of knowledge regarding the impact of NSSI on treatment trajectories, this study aimed to identify correlates of NSSI among children with PBD, and examine the influence of NSSI on psychosocial intervention response. We conducted secondary analyses in an RCT of Child-and Family-Focused CBT (CFF-CBT) versus Treatment as Usual (TAU) for PBD (West et al. 2014). Results from this RCT demonstrated superiority of CFF-CBT versus TAU for treatment attendance and satisfaction, children's manic and depressive symptoms, and global functioning (West et al. 2014).
Hypotheses were based on findings from aforementioned studies of NSSI in PBD (Esposito-Smythers et al. 2010) and heterogeneous youth samples (Fox et al. 2015; for review see Jacobson and Gould 2007), and the functional model of NSSI (for review see Nock 2010;Nock and Prinstein 2004). Consistent with prior research and the intrapersonal negative and positive reinforcement functions of NSSI, we expected that illness features exacerbating emotion dysregulation (via low or high reactivity) would be correlated with NSSI. Specifically, we hypothesized that children who endorsed NSSI would have higher rates of bipolar I and II disorders and mixed episodes; more severe depression, mania, anxiety, psychosis, and suicidality; higher hopelessness; and lower self-esteem. Also in line with past research and the interpersonal negative and positive reinforcement functions of NSSI, we anticipated that characteristics facilitating social dysfunction would be associated with NSSI. Specifically, we hypothesized that children with NSSI would have higher rates of disruptive behavior disorders, and greater impairment in global functioning, quality of life, and coping skills. Similarly, we hypothesized that parents of children who reported NSSI would have more mood symptoms and deficits in family functioning, coping, and help-seeking. Finally, based on findings from RCTs of CBT for depressed adolescents (Asarnow et al. 2009;Vitiello et al. 2011), and considerable impairment associated with NSSI, we expected that presence of NSSI at baseline would be predictive of worse outcome trajectories across treatment conditions.

Method Participants
Seventy-two children with PBD and their parents were recruited from a specialty pediatric mood disorders clinic in an urban, Midwestern Academic Medical Center from 2010 to 2014. Inclusion criteria required children to be ages 7-13 and meet DSM-IV-TR criteria for a bipolar spectrum disorder, including bipolar I, II, or not otherwise specified (NOS). Bipolar disorder NOS was defined using DSM-IV-TR criteria as the presence of depressive and manic symptoms that met symptom severity threshold but not minimal duration criteria, or the presence of recurrent hypomanic episodes without intercurrent depressive symptoms. Children were also required to be stabilized on medication, defined as scores of ≤20 on the Young Mania Rating Scale (YMRS; Young et al. 1978) and <80 on the Children's Depression Rating Scale-Revised (CDRS-R; Poznanski et al. 1984). Children were excluded if they had an IQ < 70; a primary psychotic disorder, substance abuse/dependence, neurological conditions, or other medical problems that complicated psychiatric symptoms, per the Washington University Kiddie Schedule for Affective Disorders and Schizophrenia (WASH-U-KSADS; Geller et al. 2001); active suicidality requiring hospitalization; and primary caregiver with severe mood symptoms.

Procedures
Baseline Diagnosis and Randomization Procedures were overseen by a University Institutional Review Board. Study personnel completed informed consent/assent, screening, the WASH-U-KSADS (Geller et al. 2001), and the Kiddie SADS-Present and Lifetime Version (K-SADS-PL; Kaufman et al. 1997) with parents and children. Interviews were reviewed during study meetings to ascertain consensus diagnoses. Children with PBD subsequently completed the baseline assessment and were randomized to treatment condition. Additional evaluations were conducted at 4-, 8-, 12-(posttreatment), and 39-weeks (6-month follow-up) by blind raters.
Treatment Conditions Families were randomly assigned to CFF-CBT (n = 35) via a specialty mood disorders clinic or TAU (n = 36) via a general psychiatry clinic, both within the same outpatient psychiatry program (n = 1 excluded with baseline data only). See West et al. (2014) for the Consolidated Standards of Reporting Trials (CONSORT) diagram, additional study details, and primary outcomes. Treatment in both conditions consisted of 12 weekly 60-90 min sessions and up to six monthly booster sessions.
CFF-CBT is a manualized, family-based intervention that integrates CBT with psychoeducation and complementary mindfulness and interpersonal/family therapy techniques. Family, parent, and child session content is structured around the treatment acronym BRAINBOW,^including: Routine (development of consistent routines); Affect Regulation (psychoeducation about emotions; mood monitoring; coping strategies); I Can Do It! (improvement in child self-esteem and parent self-efficacy); No Negative Thoughts/Live in the Now (cognitive restructuring and mindfulness to reduce negative thoughts); Be a Good Friend/Balanced Lifestyle (social skill-building and parent self-care); Oh How Do We Solve this Problem? (family problem-solving and communication training); and Ways to Find Support (enhanced support networks and parental advocacy; Pavuluri et al. 2004a;West et al. 2014).
Though CFF-CBT and TAU sessions were intended to be matched on treatment dosage, more CFF-CBT participants (n = 30) completed the core treatment (12 weekly sessions) than TAU (n = 17). However, attrition by the 6-month assessment did not differ by condition. Other than efforts to match number of sessions across conditions, TAU sessions were not manipulated in terms of content or structure. In addition to psychotherapy, all participants received medication management according to an evidence-based algorithm (Pavuluri et al. 2004b).
Twenty-three therapists, consisting of clinical psychology pre-and post-doctoral trainees novice to treatment of PBD, facilitated CFF-CBT following a 3-h training on the manual. They also received weekly supervision from the treatment developers. Control therapists (n = 23) consisted of clinical psychology doctoral trainees, post-doctoral fellows, psychiatry fellows, and social work interns. They received a 1-h training on PBD symptoms, course, and associated impairments, making this condition Benhanced^compared to typical TAU. Per session audio-recordings coded by trained raters, therapists demonstrated strong fidelity to CFF-CBT (93%), and there was minimal presence of CFF-CBT concepts in TAU (4%; West et al. 2014).

Measures
Diagnoses The WASH-U-KSADS (Geller et al. 2001) is as semi-structured interview used to assess DSM-IV diagnoses, including bipolar spectrum disorders and most recent/ current episode. This assessment has well-established reliability and validity. Trained study personnel administered the interview at baseline and demonstrated adequate reliability, k > .74.

Demographics The Conners-March Developmental
Questionnaire (Conners and March 1996) assessed children's age, sex, and race at baseline.

Clinical Characteristics
The clinician-rated Clinical Global Impressions Scale for Bipolar Disorder-Severity Scales (CGI-BP; Spearing et al. 1997) assessed overall severity of children's psychiatric illness longitudinally, and baseline levels of mania, depression, and psychosis. Scores range from 1 (not at all ill) to 7 (extremely ill). The CGI-BP-S has strong inter-rater reliability, and evidenced good reliability in the current study, ICC = .81.
The Child Mania Rating Scale (CMRS; Pavuluri et al. 2006) and companion Child Bipolar Depression Rating Scale (CBDRS; Unpublished Data, August, 2003) assessed children's mood longitudinally. The CMRS contains 21 items and the CBDRS contains 22 items; both are completed by parents. Responses range from 0 (never/rarely) to 3 (very often); total scores range from 0 to 63 (CMRS) and 0 to 66 (CBDRS). The CMRS has good internal consistency, testretest reliability, and validity (content, construct, criterion). The CBDRS is face valid (consistent with DSM-IV-TR criteria for a major depressive episode) and had modest construct validity, as it was significantly correlated with the clinicianrated CDRS-R (Poznanski et al. 1984), r = .32, p < .01, and the CGI-BP-Severity of Depression (Spearing et al. 1997), r = .44, p < .01. In this study, reliability was strong for both the CMRS, α = .90, and CBDRS, α = .88.
Finally, the clinician-rated Children's Global Assessment Scale (CGAS; Shaffer et al. 1983) measured children's baseline psychosocial functioning. Scores range from 1 to 100. The CGAS has strong inter-rater reliability, test-retest reliability, and validity (discriminant, concurrent). In the current study, the CGAS had good inter-rater reliability, ICC = .76.
Suicidality The Columbia Suicide Severity Rating Scale (C-SSRS; Posner et al. 2011) is a semi-structured interview that evaluated current and lifetime suicidality at baseline. Suicidal behaviors assessed include preparatory acts (acts or preparation toward making a suicide attempt, such as assembling a method, like buying pills), aborted attempts (individual begins to take steps toward making a suicide attempt, but stops him/ herself), interrupted attempts (individual is interrupted by someone/something from starting a suicide attempt), and actual suicide attempts (individual commits a potentially lifethreatening act, with at least some wish to die). In current analyses, binary variables were created to assess passive/ nonspecific ideation versus severe ideation (with a method, intent, and/or plan), and absence versus presence of suicidal behavior (preparatory acts; aborted, interrupted, and actual attempts). The Suicidal Ideation Intensity scale measured intensity of ideation; scores range from 0 to 25. The C-SSRS has shown good sensitivity and specificity for suicidal ideation and behavior. Reliability was strong in this sample, α = .88.

Child Psychosocial Factors
All child measures were selfreports completed at baseline.
The Piers-Harris Self-Concept Scale-2 (PHSCS-2; Piers and Herzberg 2002) measured children's self-esteem. The PHSCS-2 contains 60 yes/no items that assess attitudes about physical appearance, anxiety, intellectual/school status, behavior, satisfaction, and popularity. Reliability and validity are well established; reliability was excellent in this sample, α = .90.
The Hopelessness Scale for Children (HSC; Kazdin et al. 1986) assessed children's feelings of hopelessness via 17 yes/ no items. The HSC has good reliability and validity. Internal consistency in this sample was adequate, α = .63.
The Youth Coping Index (YCI; McCubbin et al. 1996) measured children's coping. The YCI contains 31 5-point Likert-scale items that evaluate personal development, problem solving, and stress management. The YCI has strong internal consistency, test-retest reliability, and predictive validity; current sample α = 82.
Lastly, the Questionnaire for Measuring Health-Related Quality of Life in Children (KINDL; Ravens-Sieberer and Bullinger 1998) assessed children's quality of life. The KINDL contains 24 5-point Likert-scale items that measure physical, emotional, self-esteem, family, friend, and school domains. The KINDL has good reliability and excellent validity; current sample α = .75. The FCOPES has adequate internal consistency and testretest reliability for the overall scale and subscales, and good factorial and concurrent validity. The current study used the total scale and acquiring/accepting help subscale; both demonstrated adequate reliability, current sample α = .84 and .77, respectively.

Data Analysis
Frequency counts were conducted to determine prevalence of NSSI. Participants were grouped based on presence/absence of NSSI behaviors, consistent with prior research (Esposito-Smythers et al. 2010). Specifically, youth who endorsed any NSSI behaviors, including mild (infrequent, 1-3 times/year), moderate (frequent, 4-11 times/year), and severe (very frequent, ≥ 12 times/year), were compared against those without NSSI behaviors, including both youth who denied NSSI and those who reported thoughts of NSSI only, in the absence of behaviors. Because we were interested in examining the most severe form of this phenomenon in PBD, we did not group those who reported only thoughts of NSSI with those who endorsed NSSI behaviors.
Preliminary Pearson correlations indicated that both child factors and parent/family variables were moderately correlated with one another, suggesting that these were related but distinct constructs. Differences in child and parent/family demographic, clinical, and psychosocial factors were compared using independent-samples t-tests and Chi-square analyses. Given the exploratory nature of this study, the significance level was set at p < .05. Significant correlates from univariate findings were subsequently entered into a single multivariate logistic regression model simultaneously, examining associations with presence/ absence of NSSI. Only demographic, clinical, and psychosocial variables found to significantly differ across NSSI groups at the univariate level were controlled for in logistic regression analyses.
Exploratory two-level, mixed-effects regression models (MRMs) were used with the intent-to-treat sample to examine the impact of NSSI on PBD dependent variable symptom outcomes (mania, depression, and overall psychiatric severity). As primary outcome analyses indicated a significant difference in mania across groups initially (West et al. 2014), models examining mania trajectories controlled for baseline manic symptom severity. MRMs included effects for Time (centered at 12 weeks), Treatment (coded 1, reflecting post-treatment differences in intercept), NSSI (coded 1), Treatment x Time (group differences in slope through follow-up), Treatment x NSSI (post-treatment differences in intercept by NSSI group), Time x NSSI (NSSI as a predictor of overall slope over time), and Treatment x Time x NSSI (NSSI as a moderator of treatment slope over time). Models also included random effects for intercept and slope, and linear and quadratic time effects, which were retained when significant.

Multivariate Logistic Regression Model of NSSI
Significant correlates identified in aforementioned univariate analyses were entered into a single multivariate logistic regression, including child depression, psychosis, lifetime and current severe SI, SI intensity, suicidal behaviors, hopelessness, self-esteem, and family help-seeking behaviors. According to the multivariate logistic regression, high levels of child depression, OR = 2.84, and psychosis, OR = 5.15, and low levels of family help-seeking behaviors, OR = 0.67, remained significantly associated with NSSI, after controlling for other variables found to be significant at the univariate level (see Table 2).

Impact of NSSI on Psychosocial Treatment Outcomes
In MRMs, NSSI did not influence the treatment effect on depressive symptoms (see Table 3). However, NSSI was a predictor of overall psychiatric severity outcomes, with a medium effect size, d = 0.47 (see Table 3, Fig. 1). Children who endorsed NSSI experienced steeper response trajectories for overall psychiatric severity over time, regardless of treatment group. In addition, NSSI moderated the response of manic symptoms to intervention, with a medium effect size, d = 0.45 (see Table 3, Fig. 2). Youth who denied NSSI showed a poorer response to TAU for manic symptoms than those who endorsed NSSI, whereas mania trajectories in CFF-CBT were similar across youth.

Discussion
To our knowledge, this is the second study to examine prevalence and correlates of NSSI in PBD, and the first to evaluate the impact of NSSI on psychosocial treatment outcomes for this population. NSSI was common. The strongest unique correlates of NSSI in multivariate analyses were depression, psychosis, and impaired family helpseeking. Suicidality, hopelessness, and low self-esteem were also associated with NSSI at the univariate level. Children who endorsed NSSI experienced steeper treatment response trajectories for overall psychiatric severity, regardless of treatment group. Youth who denied NSSI showed a poorer response to TAU for manic symptoms versus those who endorsed NSSI; mania trajectories in CFF-CBT were similar across youth, regardless of NSSI. These findings offer a broader conceptualization of NSSI in PBD, and suggest how these behaviors might influence response to intervention.

Impact of NSSI on Psychosocial Treatment Outcomes
The impact of NSSI on psychotherapy outcomes has been rarely examined in RCTs of CBT for depressed adolescents (Asarnow et al. 2009;Vitiello et al. 2011), and unstudied in intervention trials for PBD, until now. Children who denied NSSI showed a poorer response to TAU for manic symptoms compared to those who endorsed NSSI. In contrast, mania trajectories in CFF-CBT were similar across youth, and superior to TAU. These results are similar to prior moderation analyses in the same RCT, which found that children with milder depression and greater self-confidence fared worse in TAU (Weinstein et al. 2015b). Thus, it is possible that TAU providers feel more concern and pressure to provide effective therapy to children presenting with more severe symptoms and NSSI, resulting in greater time and effort spent preparing for sessions and researching therapeutic strategies. Alternatively, it is possible that parents of children with more severe symptoms and NSSI are more motivated to engage in the therapy process. Thus, youth with NSSI may receive more intensive treatment in TAU, leading to enhanced outcomes. In contrast, mildly ill youth may inadvertently receive less powerful interventions and be less motivated to engage in therapy, resulting in attenuated treatment effects. These findings suggest that it may be important for children with PBD to receive intensive interventions, and amelioration of mania in particular may require more specialized strategies. Since CFF-CBT is manualized and targeted for PBD, all children are exposed to core treatment elements, which are based on dysfunctions particular to this population (Pavuluri et al. 2004a;West et al. 2014). This structure and specificity appears to facilitate improvement in symptoms and functioning across severity levels of PBD. Presence of NSSI at baseline also influenced the treatment effect on children's overall psychiatric severity, such that children who endorsed NSSI experienced steeper response trajectories. This is in contrast to prior research, which found NSSI to be a predictor of poorer psychotherapy outcomes among depressed adolescents (Asarnow et al. 2009;Vitiello et al. 2011). One potential explanation for this is that in younger children with PBD, these symptoms and behaviors may not be as ingrained. Another reason this might occur is because children with more severe impairments initially may have more room for growth and improvement. Additional research is needed on this topic. Finally, NSSI did not impede the response of depressive symptoms to psychosocial treatment in PBD. This is promising, given that NSSI and depression commonly co-occur (Esposito-Smythers et al. 2010;Fox et al. 2015;Hankin and Abela 2011; for review see Jacobson and Gould 2007), and NSSI is predictive of worse outcomes for depressed adolescents (Asarnow et al. 2009;Vitiello et al. 2011). One possible explanation for this lies in the heavy family involvement common in the treatment of PBD versus adolescent depression. For instance, parents may be able to easily monitor and intervene on maladaptive or unsafe behaviors. In addition, involvement in therapy might have provided basic strategies for ameliorating depressed mood, which appeared to be effective regardless of NSSI. Nevertheless, presence of NSSI seems to influence the response of children's manic symptoms and overall psychiatric severity to psychotherapy, and thus should be assessed, monitored, and considered in treatment.

Prevalence and Correlates of NSSI
Prevalence of NSSI (31%) was similar to studies reporting rates of NSSI in PBD (Esposito-Smythers et al. 2010) and other heterogeneous clinical samples (for review see Barrocas et al. 2011;DiClemente et al. 1991). This high prevalence, coupled with data indicating that an additional 10% of children endorsed thoughts of NSSI, highlights the severity of the sample. These results also suggest that NSSI in PBD is unfortunately common, even among school-aged children, and thus should be carefully evaluated and tracked. Depression, psychosis, and impaired family help-seeking were correlates of NSSI in both univariate and multivariate analyses, suggesting that these constructs are strongly and uniquely associated with NSSI in PBD. However, NSSI was not associated with demographics, comorbidity, mania, coping skills, functioning, quality of life, or parental mood symptoms. Both depression and psychosis were predictive of NSSI in the one other study that examined this phenomenon in youth with PBD (Esposito-Smythers et al. 2010). Research with heterogeneous youth samples also implicated depression (Fox et al. 2015;Hankin and Abela 2011; for review see Jacobson and Gould 2007) and psychosis (Martin et al. 2015) as predictors of NSSI. Thus, the relationship between these constructs is robust, and may indicate a subset of youth with more severe symptomatology.
There are several other potential explanations for the strong association between NSSI, depression, and psychosis in PBD. For instance, NSSI may be closely linked to both depression and psychosis due to a common underlying deficit in emotion regulation (for review see Dickstein et al. 2009;Fox et al. 2015;Fusar-Poli et al. 2014). The functional model of NSSI also offers insight into possible connections between these constructs (for review see Nock 2010;Nock and Prinstein 2004). For example, children who experience intense sadness or irritability might engage in NSSI to regulate overwhelming feelings, consistent with the intrapersonal negative reinforcement function of NSSI. This rationale is feasible, given that PBD is characterized by intense emotional reactivity Geller et al. 2008), and intrapersonal negative reinforcement is the most commonly endorsed function of NSSI (Nock et al. 2009). However, during anhedonic depressive episodes, children might use NSSI to induce positive feelings or a desirable physiological response. In the context of psychosis, it is possible that youth engage in NSSI to escape intense and negative affective states associated with hallucinations and delusions, as psychosis may stem from a core dysfunction in emotion regulation (Fusar-Poli et al. 2014). These conclusions are purely speculative, given the general lack of literature on NSSI in PBD, and because functions of NSSI were not measured in this study. Nevertheless, NSSI in PBD should be assessed and considered in treatment planning, with a focus on understanding children's motivations, as alternate functions may indicate different strategies. If children endorse NSSI as a means of stabilizing intense emotions, affect regultion strategies might be helpful. Alternatively, if children use NSSI to create positive emotions or punish oneself, behavioral activation or cognitive restructuring could be salient. Impaired family help-seeking was also a strong and unique correlate of NSSI, which is consistent with prior research (for review see Arbuthnott and Lewis 2015). Thus, in addition to Fig. 2 Impact of non-suicidal self-injury on intervention response for children's manic symptoms. CMRS Children's mania rating scale; CFF-CBT Child-and family-focused cognitive-behavioral therapy; TAU Treatment as usual. Data collection wave (Time): 0 baseline; 1 4 weeks; 2 8 weeks; 3 12 weeks/post-treatment; 4 39 weeks/6-month follow-up deficits in emotion regulation, children with PBD might also engage in NSSI for interpersonal reasons (for review see Nock 2010;Nock and Prinstein 2004). For example, it is possible that children use NSSI for interpersonal negative reinforcement, or to disrupt fighting or other familial processes commonly impaired in PBD (Keenan-Miller et al. 2012). Alternatively, if families are unable to ascertain assistance and appropriate therapeutic services, youth may resort to maladaptive coping strategies (like NSSI) as a cry for help. Similarly, families' inability to acquire resources and solve problems could model and foster helplessness in children. In addition, NSSI may be a response to distress for children who find it hard to seek assistance from family members. Indeed, as youth often do not disclose NSSI, parents may not be aware of their children's difficulties (for review see Arbuthnott and Lewis 2015). Thus, fostering open communication and targeting dysfunctional processes in families affected by PBD may be crucial. If children use NSSI to gain attention or support, teaching children and families alternative strategies via relationship-building activities and effective communication skills might be relevant. In addition, interventions that help parents mobilize resources and model help-seeking could improve NSSI in youth.
Finally, suicidality, hopelessness, and low self-esteem were correlated with NSSI at the univariate level only, suggesting that these constructs are related to NSSI, though not as strongly as depression, psychosis, and impaired family help-seeking. NSSI is more prevalent in youth with PBD who have a history of suicide attempts (Algorta et al. 2011;Goldstein et al. 2005;Goldstein et al. 2012), which may be due to an acquired capability for suicide according to the interpersonal theory of suicide (Joiner 2005). However, further empirical research is needed to confirm the relevance of this theory in youth with PBD. Regarding the relationship between NSSI, hopelessness, and low self-esteem, children may engage in NSSI to ameliorate negative feelings, induce positive feelings or physiological states, or punish oneself, consistent with intrapersonal reinforcement functions of NSSI (for review see Nock 2010;Nock and Prinstein 2004). If future research confirms these associations, then emphasis on affect regulation and other coping skills for stabilizing mood might be important for youth with PBD who self-injure.
While NSSI is a risk factor for suicidality in and of itself, there are also some unique and overlapping correlates for these dangerous behaviors, which warrant discussion. Depression, hopelessness, and low self-esteem are correlates of both NSSI and suicidality in PBD (Esposito-Smythers et al. 2010; for review see Hauser et al. 2013;Weinstein et al. 2015a). Thus, it might be important to assess for and address these impairments in treatment, and when they are present, closely monitor for both NSSI and suicidality. However, the strongest unique correlates of SI and NSSI differed in this sample. Quality of life and family rigidity were specific to SI (Weinstein et al. 2015a), while psychosis and impaired family help-seeking were exclusively associated with NSSI. Thus, families of children with NSSI might require additional training in advocacy, problem-solving, and effective communication. Importantly, children with PBD and NSSI should be closely monitored for psychosis, given that these youth seem to be at increased risk for thought problems. In sum, NSSI appears to be a common and complex phenomenon in PBD, which requires further investigation.

Limitations and Future Directions
Several limitations should be noted. First, these analyses were secondary and not adequately powered to examine NSSI in PBD. In addition, as NSSI was assessed over the past year, the more proximal relationship between NSSI and baseline constructs is unknown. The modest sample size also precluded the examination of NSSI across different severity levels, and may have resulted in reduced power to detect small effects. Similarly, grouping all youth together who endorsed NSSI may have obscured differences between children who engaged in repeated versus less frequent NSSI. Also, the outpatient clinical sample likely restricts the generalizability of results, as findings may not apply to adolescents with PBD or youth in intensive treatment. Further, equal distribution of bipolar diagnoses was not observed. A majority of children had bipolar disorder NOS, which is expected given the age of the sample, but this limited power to compare NSSI across bipolar subtypes. Also, correlates of NSSI were evaluated via cross-sectional data, and NSSI was only assessed at baseline in this study. Thus, the temporal relationship between NSSI and associated constructs cannot be certain. Relatedly, analyses focused on the association between lifetime suicidal behaviors and NSSI in the past year; thus, findings cannot speak to the proximal relationship between these behaviors. Finally, as functions of NSSI were not specifically evaluated, the reasons youth engaged in these behaviors remain speculative at this point. While prior work offers a theoretical explanation for why youth might engage in NSSI, these functions are not necessarily mutually exclusive (for review see Nock 2010;Nock and Prinstein 2004). Therefore, such functions may lack the ability to fully discriminate and explain why individuals engage in NSSI.
Despite randomization, baseline differences in children's manic symptoms emerged. Controlling for baseline mania reduced power in these models. Similarly, differential treatment completion in CFF-CBT versus TAU and attrition over follow-up may have limited power to detect longitudinal group differences. In addition, the medium-sized sample precluded examination of differing severity levels of NSSI on treatment outcomes. Also, although providers in both CFF-CBT and TAU were trainees at comparable levels and all were novice to the treatment of PBD, it is possible that differences in characteristics and training backgrounds of therapists across groups impacted findings. Thus, additional studies should evaluate the effectiveness and generalizability of CFF-CBT across settings and clinicians.
Future research should also replicate findings in larger samples. We were unable to examine cyclothymia in the current sample, but future work should evaluate NSSI across the full spectrum of bipolar disorders. Differences across severity levels of NSSI should also be examined, especially among those who engage in NSSI versus those who report thoughts of NSSI only, to determine predictors of progression from thoughts to behaviors. In addition, studies should longitudinally measure NSSI trajectories, functions, and the influence of potential risk factors, such as the frequency and severity of mood symptoms. Also, more nuanced measurement of NSSI behaviors may discern intricacies in the relationship between NSSI and intervention response in PBD. Further research should also examine associations among NSSI and suicidal behaviors in PBD, and determine the relevance of the interpersonal theory of suicide (Joiner 2005) for this population. Finally, additional work is needed to discern whether youth with PBD and NSSI represent a distinct subgroup that may be at increased risk for more severe emotion dysregulation and other disorders characterized by NSSI later in life, such as borderline personality disorder.

Clinical Implications and Conclusions
Findings highlight the importance of evaluating NSSI in youth with PBD. Children with more severe depression and psychosis, and low-resource families, might be at the highest risk and benefit from extra supports. Interventions aimed at understanding why children engage in NSSI and offering more effective means of fulfilling these needs may be beneficial. Identified correlates suggest that youth with PBD may engage in NSSI due emotion dysregulation, lack of family coping skills, and for intrapersonal and interpersonal negative and positive reinforcement. If future research confirms the relevance of these factors in understanding NSSI in PBD, then therapeutic strategies linked to these deficits might be important, such as affect regulation, behavioral activation, positive family relationship-building activities, effective communication, and parental advocacy. Of note, these elements are incorporated in CFF-CBT, which may be one of the reasons why this psychosocial treatment was efficacious in treating PBD, regardless of NSSI.
Results also suggest that TAU may look different for children with varying presentations. Mildly ill youth with fewer severe presenting issues (e.g., absence of NSSI) may inadvertently receive less potent interventions in TAU and thus fare worse than children with greater dysfunction. Therefore, structured interventions that target dysfunctions specific to PBD, such as CFF-CBT, may be most effective for ameliorating symptoms and optimizing outcomes across severity levels of PBD.