The Course of Grief in Children Bereaved by Sudden Parental Death

Abstract

Context

There have been major advances in our understanding of the phenomenology and course of grief in adults. However, little is known about the course of grief in children.

Objective

We report on course and impact of children’s grief reactions following parental sudden death on subsequent psychiatric and functional status.

Design

Longitudinal study of bereaved children and families with yearly comprehensive assessments up to three years after parental death.

Setting

Bereaved children and their surviving parents recruited through the coroners’ records and advertisement.

Participants

182 parentally bereaved children between 7 and 18 years of age whose parent died from suicide, accident, or sudden natural deaths.

Main Exposure

Sudden parental death.

Main Outcome Measures

Grief, functional impairment, incident depression.

Results

There were 3 distinct trajectories of grief reactions with a group, consisting of 10.4% of the sample, with increased grief reactions that showed no change 33 months post-death. Youths with prolonged grief reactions had higher rates of previous personal history of depression. Prolonged grief made unique contribution to increased levels of functional impairment even after controlling for clinical characteristics antecedent and following the death. Conversely, prolonged grief, both in youths and the surviving caregiver, predisposed to an increased hazard of incident depression. Another group (30%) showed increased grief reactions 9 months following the death, which gradually decreased over time. Despite this, grief reactions in this group were also associated with functional impairment and increased risk of incident depression.

Conclusions

Grief reactions abate over time for most children bereaved by sudden parental death; however, a subset show increased or prolonged grief reactions, which in turn increases the risk for functional impairment and depression. Research on interventions designed to relieve the burden of grief in bereaved children are needed. Such efforts should also assess and address grief reactions in the surviving parent.

INTRODUCTION

The death of a parent is consistently rated as one of the most stressful life events that a child can experience^1,2. In Western countries, 4% of children experience the death of a parent and approximately 1 in 20 children in the United States experience the loss of a parent before they reach 18 years of age³. There have been major advances in our understanding of the phenomenology and course of grief in adults^4-29 . We define “grief” as the subjective experience of loss, whereas bereavement refers to status with respect to loss, regardless of subjective experience. While less is known about grief reactions in children, preliminary evidence from cross-sectional studies suggests that, similar to adults, some bereaved children experience a complicated or prolonged grief reaction^30-37. Reflecting the current terminology in the literature, we have used the terms “complicated” and “prolonged” interchangeably.

Relatively little is known about the course of grief in children. In our ongoing longitudinal study of the impact of parental sudden death on children, we have previously reported that exposure to parental death results in an increased risk for depression and post-traumatic stress disorder (PTSD) during the first year after the death that persist into the second year^38,39. Herein, we report on course and impact of parentally-bereaved children’s grief reactions on subsequent psychiatric and functional status. We hypothesize that, over the first three years after parental death, (1) a subset of parentally-bereaved children will have prolonged grief reaction (PGR) consistent with adult complicated grief, which will be predicted by a prior personal or family history of psychiatric disorders; (2) conversely, those bereaved children with PGR will have more functional impairment above and beyond other psychopathology; and (3) Youths with PGR will have increased incidence and earlier onset of depression.

METHODS

Sample

The sample is from the impact of parental death on children and families study on which we have reported previously. Bereaved children were interviewed at three points in time, at baseline which occurred 8.5 [mean=8.5; Standard Deviation (SD) =3.7] months after the death^36,38; then around 1 year later³⁹, 21.4 (SD=4.2) months after the death; and around two years later, 33.2 (SD=5.7) months after the death. The sample consisted of 182 children less than 18 years of age at baseline. The sample was 54.4% male and with a mean age of 12.4 years (SD=2.8). The majority of children (91.8%) were biological to the proband. Of these, 165 (90.7%) were followed up a year later, and 141 (77.5%) two years later. We also interviewed the surviving parent, mostly female (89%) and a biological parent of the child (88.4%).

We compared subjects lost to follow-up to those retained in the study on demographic and clinical characteristics at baseline (number of variables tested=24, corrected α=.002). There were no differences in grief reactions, our main outcome, as measured by the Inventory of Complicated Grief-Revised for Children (ICG-RC)^36,40 between subjects retained and those lost to follow-up [55.5 (SD=19.7) vs. 53.7 (SD=18.6), t=0.33, df=152, p=.74] . Subjects lost to follow-up had higher scores on the negative coping frequency scale [5.2 (SD=.6) vs. 4.1 (SD=1.6), t=−4.49, df=17.65, p<.001] and were more likely to blame others for the death (66.7% vs. 24.4%, Fisher’s exact test, p=.001) as compared to those retained.

Recruitment

Deceased probands were between the ages of 30-60, had children 7-18 years old, and died within 24 hours from suicide (n=42), accident (n=31), or sudden natural death (n=51). Bereaved families were recruited via coroner’s reports (49.7%) and newspaper advertisement (50.3%). Details on the recruitment and representativeness of the sample were published previously^36,38,39. In brief, monthly lists of deaths by suicide, accident, or sudden natural causes, occurring between July 2002 and December 2006, were obtained from the coroner offices of Allegheny County and neighboring counties after obtaining approval from the Institutional Review Board at the University of Pittsburgh. Suicides are those determined as definite suicides by the coroner whereas all undetermined or ambiguous deaths were excluded (e.g., death by falling, death by firearms thought to be accidental). Only accidental and natural deaths that resulted in sudden death are included. Accidents in which multiple family members died or were seriously injured were also excluded. The causes of sudden natural death included in this sample were: myocardial infarction (n=21), other heart conditions (n=18), infections (n=1), and others (n=11) (e.g., cancer, diabetes, stroke, aneurysm, gastric bypass surgery, etc.). The accidental deaths consisted of 11 drug overdoses, 8 motor vehicle accidents, and 12 from other causes (e.g., drowning, exposure to cold, etc.). Drug overdoses were carefully reviewed to rule out those with possible suicidal intent. Probands who died from drug overdoses had no previous history of suicide attempts as compared to suicide victims (0% vs. 35.7%, Fisher’s exact test, df=1, p= .02).

A letter was sent from the coroner’s office to the next of kin of 1773 deceased probands, of which we were able to reach 1638 (92.4%) families. The letter included a description of the study and a stamped refusal card. If no refusal card was received within two weeks, the interviewers called the next of kin to check their eligibility and enlist their participation. Of those contacted, 16.2% sent back the refusal card, 52.5% were not eligible because the deceased was either single and had no children or had no children in the eligible age range, and 23.7% were found eligible. The eligibility rate was similar across types of death and 71% of eligible subjects participated. We also expanded our efforts to include other sources such as radio and paper advertisement and distributing posters and brochures to physicians, hospitals, funeral directors, community mental health facilities, and other service providers. Probands recruited through the coroner’s office and those recruited by advertisement were compared on demographic and clinical characteristics (number of variables tested=12, corrected α=.004) and were found similar except that those recruited through the coroner had higher rates of alcohol/substance abuse disorders (73.1% vs. 50.0%, χ²₁=8.09, p=.004). A higher proportion of sudden natural death probands (70%) were recruited by advertisement, because people who die from sudden natural death frequently do not come to the attention of the coroner. Finally, the demographic characteristics of the suicides and accidents were similar to suicide and accident victims in Allegheny County (metropolitan Pittsburgh)⁴¹.

Assessment

Our modified version of the adult Inventory of Complicated Grief (ICG), the ICG-RC¹⁶, was used to assess grief phenomenology in children under the age of 18. We have previously examined its psychometric properties³⁶. The ICG-RC has high internal consistency as measured by Cronbach’s α of 0.95 and shows evidence of convergent and discriminant validity in relation to self-reported measures of depression, anxiety, PTSD, and functional impairment. For the surviving parent, the 19-item ICG⁴⁰ was used. We use a previously established cut-off of 25 or greater, which identifies adult individuals with complicated grief with significantly worse general, mental and physical health, and social functioning⁴⁰. The Circumstances of Exposure to Death (CED) was used to assess the children’s experience surrounding and following the death of their parent⁴². Past (prior and at time of death) and current psychiatric disorders in children younger than 18, were assessed using the School Age Schedule for Schizophrenia and Affective Disorders, Present and Lifetime Version (K-SADS-PL)⁴³. For children who turned 18 during the course of the study and for surviving parents, we administered the Structured Interview for DSMIV Axis I and II Disorders (SCID)^44,45. At follow-up, the K-SADS-PL and SCID were used in conjunction with the Longitudinal Interval Follow-up Evaluation (LIFE)⁴⁶ to assess the longitudinal course of psychiatric disorders. Psychiatric assessment of the proband was conducted using a psychological autopsy procedure⁴². Functional status was determined using the Children’s Global Assessment Scale (CGAS)⁴⁷ or the Global Assessment Scale (GAS)⁴⁸ for adult children and surviving parents. High inter-rater reliability was maintained on psychiatric diagnoses and global impairment (Kappa ranged between .74 and .85; ICC= .88). A history of physical or sexual abuse was assessed using screens from the PTSD section of the psychiatric interview and the Abuse Dimensional Inventory⁴⁹.

A battery of self-reported instruments was also administered to assess the severity of symptomatology in offspring. Self-reported symptoms of depression, anxiety, post-traumatic stress disorder (PTSD), and suicidal ideation were assessed using the Mood and Feelings Questionnaire (MFQ), the Screen for Child Anxiety Related Emotional Disorders (SCARED), the Child PTSD Symptom Scale Interview (C-PSSI) and the Suicide Ideation Questionnaire-Jr., respectively^50-53. Parallel measures in offspring who turn adults were obtained using the Beck Depression Inventory (BDI), the Beck Anxiety Inventory (BAI), the PTSD Symptom Scale-Interview (PSSI-I), and the Suicide Ideation Questionnaire^54-57.

Factors that might buffer the impact of bereavement were also assessed in offspring. Socioeconomic status and household income were rated using Hollingshead’s scale⁵⁸. Intercurrent life events were assessed using the Life Events Checklist (LEC) and the shortened Social Readjustment Rating Scale of Holmes & Rahe, in offspring under and over age 18, respectively^59-61. Family cohesion was assessed using the Family Adaptability and Cohesion Evaluation Scales-II (FACES-II)⁶². Social support was assessed using the Survey of Children’s Social Support (SOCSS)⁶³ and the Multidimensional Scale of Perceived Social Support (MSPSS)⁶⁴. Aggression and self-esteem were assessed using the Aggression questionnaire⁶⁵ and the Weinberger Adjustment Inventory self-esteem sub-scale⁶⁶. Finally, coping style was assessed using the Kidcope⁶⁷ in offspring under 18 years of age and the Ways of Coping questionnaire⁶⁸ in older offspring. When different measures were used for offspring who were under or over age 18, scores from these measures were standardized.

Statistical Analysis

We conducted Latent Class Growth Analysis (LCGA), using Mplus⁶⁹ (version 5.21), to identify grief trajectories using ICG-RC. We started with a 1-class model then increased the number of classes until we reached the best-fitting model using goodness of fit statistics, Bayesian information Criteria (BIC) and Lo-Mendell-Rubin Adjusted Likelihood Ratio Test. We included a cluster effect to account for sibling pairs in the dataset.

The rest of the analyses were conducted in STATA⁷⁰. We analyze 3 outcomes (ICG-RC class, functional impairment, and incident depression). We used a Bonferroni correction to correct for multiple comparisons for univariate analyses with alpha of 0.05/3=0.017. Post-hoc pair-wise comparisons within ICG-RC class were conducted only following a statistically significant test. There were 3 classes resulting from LCGA and hence 3 possible pair-wise comparisons, for which α was set at 0.017/3=0.006. For all multivariate models, we use p<0.05 because variables included in these models were already significant at the corrected level. Since multiple children per family were recruited, we included a cluster effect in all regression models. Our power analyses for regression models, using Cohen⁷¹ tables of the Effect Size Index f², indicate that we have power to detect effect sizes in the order of f²=0.23 with 20 variables (main effects and interaction terms) included in the model with α = 0.01 and power of 0.80. This corresponds to medium to large effect sizes for which this study was designed to detect. We define the most parsimonious model as the simplest model with the highest predictive value as defined by pseudo-R². We removed variables from the final model if they were not significant at p<0.05. Finally, we tested for two-way interactions between variables in the final model.

Paired t-tests were used to examine changes in ICG-RC scores between 9 and 21 months and between 21 and 33 months within each class or trajectory identified and analysis of variance (ANOVA) were used to compare scores between class at 9, 21, and 33 months. We compared children in the three classes identified on the following domains assessed at baseline: demographics; clinical characteristics antecedent to the death, such as personal and family history of psychiatric disorders; circumstances of exposure to death; and clinical characteristics at 9 months following the death. We usedunivariate statistics for these comparisons (chi-square test, Fisher’s exact test, and ANOVA) with post-hoc pair-wise comparisons as described above. Since we identified three classes, multinomial logistic regression was used to examine the most parsimonious set of predictors. Regression, using the hierarchical method, was applied including the variables (n=17) that were significantly different between the ICG-RC classes (Table 1). We control for the effect of time of the baseline interview in relation to time of death. We included 4 blocks in the hierarchical model in the order in which they are listed: 1) characteristics antecedent to the death; 2) circumstances of exposure to death (e.g., type of death), 3) clinical diagnoses following the death in offspring and surviving parent; and 4) finally, self-reported symptoms and psychosocial characteristics. We examined collinearity between variables within and across blocks. The order of blocks 1 to 3 follows the temporal order of variables. Block 4 was separated from 3 because self-reported symptoms and psychosocial characteristics assessed are correlated with diagnostic variables in block 3 and their effects are of interest only if they were statistically significant in the presence of variables from preceding blocks (previous psychiatric history and clinical diagnosis following the death).

Table 1.

ICG-RC Scores at Baseline and Follow-up by ICG-RC Class

	Class 1	Class 2	Class 3	Between class test statistics
Sample size	107	56	19	F	df	P
9 months	43.4 (9.7)a	69.2 (16.5)b	86.8 (10.9)c	116.0	2, 151	<.001
21 months	38.4 (6.7)a	57.2 (8.1)b	79.9 (12.2)c	228.3	2, 153	<.001
33 months	37.8 (8.4)a	53.9 (10.1)b	79.0 (12.3)c	117.9	2 ,125	<.001

Within class test statistics
9 vs. 21 months	t=4.4, df=73, p<.001	t=4.1, df=44, p<.001	t=0.8, df=9, p=.47
21 vs. 33 months	t=0.1, df=70, p=.88	t=2, df=38, p=.05	t=0.4, df=10, p=.70

Letter superscripts represent post-hoc differences. Similar letters represent no significant post-hoc difference at p<0.006

We examined next whether the course of functional impairment differed by ICG-RC class above and beyond other clinical characteristics. We used mixed effects regression models with GAS scores as the outcome and with main effects of class (or correlate), time, and class (or correlate) by time interaction. Of these analyses, 6 variables including ICG-RC class were significantly associated with GAS scores, either as main effect or in interaction with time. Next, we examined a multivariate mixed model, which included ICG-RC class, time, and the other correlates that were significantly associated with GAS scores, and testing for two-way interactions with time. A total of 13 variables (main effects and interactions) were included in the multivariate model. We compared models with time as random vs. fixed using a Likelihood Ratio Test and used time as random or fixed accordingly. ICG-RC class was represented with a 2-degree of freedom parameter. The time of baseline and follow-up assessments in relation to the time of death varied by subjects and as such a logarithmic transformation was used.

We previously reported depression to be the most commonly incident disorder following bereavement in this sample compared to non-bereaved controls³⁸. We report the incidence rate ratio (IRR) of depression by ICG-RC class. We examine whether ICG-RC scores at 9 months predicts incident depression during the course of follow-up above and beyond other correlates significantly associated with incident depression (n=4 variables). Logistic regression is used to identify the most parsimonious predictors testing for main effects and two-way interactions. We also examine the time to onset of depression by ICG-RC class using Kaplan Meier survival analysis. We use Wilcoxon and log-rank tests to test the equality of survival functions. There were 15 variables, including ICG-RC class, significantly associated with time to onset of depression. These were included in a multivariate Cox regression to examine whether ICG-RC class predicts time to onset above and beyond other characteristics.

We conducted all analyses on the original and an imputed^72,73 dataset. There were missing data in our study: 21.4% of the sample had no missing data for any variable, 53.8% had missing data for 1 or 2 variables, 18.7% had missing data for 3 or 4 variables, and 6% had missing data for more than 4 variables. We applied Multiple Imputation by Chained Equations (MICE) technique in STATA (ice) with an inclusive strategy as recommended by Collins and Colleagues⁷³ (2001).

RESULTS

1. a) Grief trajectories

The best fitting model of ICG-RC trajectories using LCGA consisted of three classes (Figure 1), which correspond to: 1) Class 1 consisting of 107 (58.8%) subjects who scored in the lower 50^th percentile on ICG-RC at 9 months after the death (Inter-Quartile Range=68-40), who experienced a significant decrease in grief scores between 9 and 21 months, and whose scores remained low afterwards (Table 1); 2) Class 2 consisting of 56 subjects (30.8%) with grief scores in the 75^th percentile at 9 months and who showed a steady decline of their manifestations of grief from 9 through 33 months; and 3) Class 3, consisting of 19 subjects (10.4%) with grief scores in the 75^th percentile at 9 months, with virtually no change at 21 months and 33 months later. There were significant differences in ICG-RC scores among the three trajectories (p<0.001) at each of the 9, 21, and 33 months assessments since the death with pairwise comparisons resulting in significantly higher scores for Class 3, followed by Class 2 and Class 1. Of the 19 subjects in Class 3, 6 (31.5%) met criteria for depression, anxiety, or PTSD throughout the period of follow-up; 7 (36.8%) had one of these diagnoses at the time of death that was either prolonged or recurred; and 6 (31.5%) did not meet criteria for any of these disorders.

Figure 1.

Latent Class Growth Curve Modeling on the Inventory of Complicated Grief-Revised Child version (ICG-RC). Red, green, and blue lines represent classes 1, 2, and 3, respectively. Circles indicate the sample mean; triangles indicate the estimated mean. To avoid non-convergence and local solutions, we used 100 random sets of starting values with 10 final optimizations, and we verified that the final highest likelihood value was replicated at least twice.

b) Baseline predictors of grief trajectories

We compared ICG-RC classes on demographics, clinical characteristics antecedent to the death, circumstances of exposure to death, and clinical characteristics 9 months after the death (Table 2). Using multinomial logistic regression (Table 3), the most parsimonious model (χ²₁₆=61.9, p<0.001, Pseudo R²=0.29) that differentiated Class 2 and 3 from Class 1 included parent loss due to accidental death and higher self-reported depression in the child at 9 months. In contrast, child functional impairment at 9 months, previous history of depression, and incident PTSD significantly differentiated Class 3 from each of Class 1 and Class 2. A previous history of bipolar disorder in the proband significantly differentiated Class 2 from Class 1 only.

Table 2.

Baseline Demographic, Clinical, and Psychosocial Characteristics Significantly Associated with ICG-RC Class

	Class 1	Class 2	Class 3
				Test	df	P*
Sample size	107	56	19
Clinical Characteristics Antecedent to the Death
Offspring Previous History of Depression, N (%)	7 (6.5)a	6 (10.9)ab	7 (36.8)b	FET**	--	.002
Proband Lifetime Bipolar, N (%)	6 (5.6)a	13 (23.2)b	4 (21.1)ab	FET	--	.002
Proband Lifetime Alcohol/Substance, N (%)	53 (51.0)a	38 (67.9)ab	16 (84.2)b	χ2=9.6	2	.008
Surviving Parent Previous History of Anxiety, N (%)	31 (29.0)a	28 (50.0)b	7 (36.8)ab	χ2=7.0	2	.03
Circumstances of Exposure to Death
Months since death, M±SD	9.0 ± 3.8a	7.3 ± 3.1b	8.0 ± 4.9ab	F=4.2	2, 179	.02
Type of death, N (%)
Natural	57 (53.3)	18 (32.1)	3 (15.8)
Accident	21 (19.6)	17 (30.4)	9 (47.4)	FET	--	.005
Suicide	29 (27.1)a	21 (37.5)ab	7 (36.8)b
Clinical Characteristics Following the Death
Offspring PTSD, N (%)	4 (3.8)a	9 (16.1)b	5 (27.8)b	FET	--	.001
Offspring Functional Impairment, Mean±SD	78.9 ± 11.6a	74.7 ± 10.9ac	65.8 ± 11.7b	F=11.3	2, 178	<.001
Surviving Parent PTSD, N (%)	17 (15.9)a	19 (35.2)b	11 (57.9)b	FET	--	<.001
Surviving Parent Functional Impairment, Mean±SD	76.6 ± 11.1a	70.6 ± 13.2b	68.3 ± 13.6b	F=6.7	2, 177	.002
Surviving Parent Complicated Grief (ICG ≥25) , N (%)	18 (19.4)a	23 (46.9)b	7 (46.7)b	FET	--	.001
Self-Reported Symptoms and Psychosocial Characteristics
Depression, M±SD	4.0 ± 4.1a	7.4 ± 4.5b	10.3 ± 5.2b	F=21.9	2, 167	<.001
Anxiety, M±SD	14.4 ± 12.3a	22.0 ± 10.9b	31.2 ± 13.0c	F=18.4	2, 171	<.001
PTSD, M±SD	3.4 ± 4.6a	7.8 ± 7.7b	11.4 ± 10.2b	F=13.7	2, 137	<.001
Suicidal Ideation, M±SD	20.6 ± 9.2a	25.2 ± 14.3ab	31.2 ± 20.8b	F= 6.5	2, 167	.002
Negative Coping, frequency, M±SD	3.9 ± 1.6a	4.6 ± 1.3b	5.1± 1.5b	F= 6.0	2, 156	.003
Hostility, M±SD	14.9 (5.9)a	17.5 (6.1)b	21.0 (6.6)b	F= 9.1	2,169	<.001
Indirect Aggression, M±SD	11.5 (4.3)a	13.5 (4.3)b	15.6 (5.2)b	F= 8.1	2,169	<.001

^*α=0.017;

^**FET: Fisher’s exact test;

Letter superscripts represent post-hoc differences, similar letters represent no significant post-hoc difference at p<0.006.

Table 3.

Results from Multinomial Logistic Regression with the Final Parsimonious Model Predicting ICG-RC Class

Outcome	Predictor	β	95% CI	Z	P£
	Functional impairment**	−0.01	−0.05, 0.03	−0.7	.51
	Type of death££, Accident	1.39	0.28, 2.51	2.5	.01
	Type of death, Suicide	0.71	−0.42, 1.85	1.2	.22
Class 2 *	Months since the death	−0.16	−0.27, −0.05	−2.8	.006
	Proband lifetime bipolar disorder	1.18	0.14, 2.22	2.2	.03
	Self-Reported symptoms of depression	0.18	0.09, 0.28	3.9	<.001
	Previous history of depression	0.43	−0.74, 1.61	0.7	.47
	Incident PTSD	1.94	−0.06, 3.94	1.9	.06
	Functional impairment	−0.08	−0.14, −0.02	−2.8	.005
	Type of deathc, Accident	4.01	1.47, 6.55	3.1	.002
	Type of death, Suicide	1.53	−0.60, 3.67	1.4	.16
Class 3 *	Months since the death	−0.23	−0.45, −0.02	−2.1	.04
	Proband lifetime bipolar disorder	1.61	−0.49, 3.71	1.5	.13
	Self-Reported symptoms of depression	0.30	0.15, 0.45	4.0	<.001
	Previous history of depression	1.93	0.50, 3.36	2.6	.008
	Incident PTSD	2.44	0.05, 4.82	2.0	.045

^*Class 1 is the reference class;

^**Lower GAS scores correspond to more functional impairment;

^£Significant at α<0.05.

^££Type of death is introduced as a dummy variable (or 2-df parameter) with natural death group as the reference group

2. Course of functional impairment by grief trajectories

There were no significant differences in the course of functional impairment, as measured by the CGAS, by ICG-RC class as evident by the non-significant class by time interaction (Table 4). However, a main effect for Class 3 showed significantly lower scores on the GAS, which reflects worse functioning (β=−14.1, z=−2.7, p=0.006). Using multivariate mixed effects regression (χ²₅=87.6, p<0.001, Pseudo R²=0.26), there was an overall worsening in functional impairment of bereaved children over time (β=−1.6, z= −2.7, p=0.006). This decline in functional impairment was more prominent in Class 2 (β=−3.7, z=−2.1, p=0.03) relative to Class 1 and most marked in Class 3 relative to Class 1 (β=−9.3, z=−3.3, p=0.001). These findings persisted even after controlling for other significant correlates namely, current anxiety symptoms (β=−0.7, z= −2.9, p=0.003), and PTSD symptoms (β=−0.52, z=−5.5, p<0.001).

Table 4.

Mixed Effects Regression Models for Functional Impairment

	Variable		Time*		Variable by Time
	β	P	β	P	β	P
Characteristics Antecedent to the Death
Offspring Relationship to the deceased	18.2	.002	6.5	.001	−7.3	<.001
Clinical Characteristics Following the Death
ICG-RC class**
Class 2	−2.6	.45	−0.05	.94	−1.1	.31
Class 3	−14.1	.006	−0.05	.94	0.4	.81
Offspring Sexual Abuse	−38	.001	−0.7	.17	13.1	.004
Surviving Parent Complicated Grief (ICG ≥25)	−9.5	.007	−1.5	.02	2.9	.02
Self-Reported Symptoms
Anxiety	−3.4	.01	−0.6	.24	0.09	.87
PTSD	−0.6	.005	−1.3	.08	−0.05	.46

^*Time defined as ln(months since the death);

^**ICG-RC Class is introduced as a dummy variable (or 2-df parameter) with Class 1 as the reference group

3. a) Grief trajectories and incidence of depression

There were a total of 59 (32%) incident cases of depression that occurred within three years following the death. Class 2 (IRR=2.9; 95% confidence interval [CI], 1.6-5.3; p=0.0001) and Class 3 (IRR=2.5; 95% CI, 0.98-5.8, p=0.04) had almost three-fold increased incidence of depression as compared to Class 1, but there was no difference between Classes 2 and 3 (IRR=0.87; 95% CI, 0.34-1.97, p=0.38).

In order to assess whether ICG-RC scores at 9 months predict incident depression by 21 or 33 months, we excluded children who met criteria for depression at time of death (n=8) and those who already had incident depression by the 9-months assessment (n=41). We compared subjects with incident depression by 21 or 33 months (n=18) and to those who did not experience depression on baseline characteristics, including ICG-RC scores. ICG-RC scores at 9 months, child rating of aggression, feeling other people were accountable for the death, life events since the death, and complicated grief in the surviving parent were significantly associated with incident depression. Using logistic regression, the most parsimonious model (χ²₄=19.6, p=0.003, Pseudo R²=0.33) included a significant interaction between ICG-RC scores in children and surviving parent’s complicated grief (OR=1.2, z=2.0, p<.05), meaning that children with higher ICG-RC scores and whose surviving parent had complicated grief were at increased risk of incident depression. In addition, feeling other people were accountable for the death (OR=7.4, z=3.0, p==0.003) and life events since the death (OR=1.2, z=2.2, p==0.03) were associated with increased risk for incident depression (Hosmer-Lemeshow Goodness of Fit: χ²₈=7.4, p=0.49).

b) Grief trajectories and time to onset of depression

Parallel analyses conducted to examine differences in time to onset of incident depression also found significant difference among the three classes (Wilcoxon: χ²₂= 13.5, p=0.001; Log-rank test: χ²₂=15.2, p<0.001) (Figure 2). Classes 2 and 3 had significantly earlier onset of depression compared to Class 1, but no differences were found between Classes 2 and 3. Cox regression (χ²=63.7, df=11, p<0.001) showed that, Class 2 (HR=3.3, z=3.1, P=0.002) and Class 3 (HR=2.5, z=2.1, P=0.04) showed an increased hazard ratio, compared to Class 1, even after controlling for other characteristics associated with earlier onset of depression, namely, indirect aggression (HR=1.3, z=4.5, P<0.001), attention deficit hyperactivity disorder (ADHD) in the offspring (HR=2.6, z=2.1, P=0.04), and having had a last confiding conversation with the deceased parent (HR=2.8, z=2.3, P=0.02).

Figure 2.

Kaplan Meier Curve of Time to Onset of Depression by ICG-RC Class. Red, green, and blue lines represent classes 1, 2, and 3, respectively.

DISCUSSION

More than half of children bereaved by parental sudden death experienced relatively rapid resolution of their manifestations of grief within 1 year of the loss of their parent. However, sometimes the course of grief was more problematic, as 30% show a more gradual diminution in grief symptoms, and around 10% show high and sustained prolonged grief nearly 3 years after parental death. The experience of PGR was predicted by a prior personal history of depression and was associated with greater functional impairment above and beyond that accounted for by other psychopathology. Youths with PGR also show an increased incidence and earlier onset of depression predicted by both their grief reactions and those of their surviving parent, 9 months after the death, even after controlling for characteristics antecedent and following the death.

Before we discuss the implications of these findings, we put them in the context of the strengths and limitations of this study. This is the first population-based longitudinal study of the impact of sudden parental death on children, has the largest sample of suicide-bereaved offspring, and includes a comprehensive assessment of offspring and surviving parent, in addition to a psychological autopsy assessment of the deceased proband. However, it is very difficult to determine whether this sample is representative or not because coroners’ records do not routinely list surviving offspring, and because natural deaths are not routinely the province of the coroner. However, our sample of suicides and accidents was demographically similar to those in Allegheny County overall, and our response rate was 71%, which is relatively high for these types of studies. While it is possible that there were referral biases, insofar as more disturbed families who were more worried about their children might have been more likely to participate, it is of interest that those bereaved families recruited by advertisement showed lower rates of proband alcohol/substance abuse disorders, which would argue against such a referral bias. Our profile of the deceased parents is similar to those obtained in previous psychological autopsy studies, which makes it more likely that our results are not simply the result of sampling bias^74,75. A second limitation is the use of an informant rather than self-report in the case of the deceased probands. However, the psychological autopsy procedure is a specific and fairly sensitive method for determining psychiatric disorder⁷⁶, and if anything, is likely to underestimate the rate of disorder, thus biasing the results more towards the null hypothesis⁷⁷. Although there were differences between subjects retained and those lost to follow-up on negative coping and blaming others for the death; nevertheless, subjects retained were not different than those lost to follow-up on grief scores. Finally, the study sample is mostly Caucasian, and does not include youth bereaved by homicide or anticipated parental death, which limits generalizability of these findings to those bereaved by these other types of death.

We find the profile of children with a problematic course of grief, i.e. those with gradual diminution of grief symptoms and those with sustained prolonged grief, to include a previous history of depression and a family history of bipolar disorder. Higher rates of psychiatric disorders are expected in suicide victims and hence in their families; however, we previously reported that early parental death from a wide range of causes, namely suicide, accidents, and sudden natural death was associated with a history of bipolar disorder, alcohol and substance abuse, and personality disorders in the deceased parent³⁸. Increased rates of psychiatric disorders in the deceased parents suggest a pre-existing vulnerability in their offspring that antedated parental death and thus puts them at increased risk of adverse outcomes following bereavement.

Children with PGR will show greater functional impairment within the first year after the death, and over the three-year course of follow-up. Moreover, PGR makes a unique contribution to the functional impairment of bereaved children, since the association between prolonged grief and impairment persists, even after controlling for personal and family history of psychiatric disorders antecedent and following the death. We previously reported, in a cross-sectional analysis of this sample³⁶, that grief scores are associated with functional impairment, above and beyond other psychopathology. Herein, we show that the association between prolonged grief and impairment persist over time. These results are convergent with findings in bereaved adults^14-66,29.

We found a bi-directional relationship between prolonged grief and psychiatric vulnerability. Prior psychiatric disorder in the child and parent increased the likelihood of the occurrence of prolonged grief in the child. On the other hand, even controlling for these factors, prolonged grief was associated with an increased hazard of incident depression. Thus, prolonged grief makes a unique contribution to the functional impairment and psychiatric morbidity of bereaved children, which justifies a clinical focus not only on psychiatric sequelae, but also on prolonged grief. In adults, interventions that target complicated grief have been shown to be effective in restoration of function and the relief of grief, compared to interpersonal therapy²⁷. While preventive interventions for community samples of bereaved youth and their families have been shown to reduce grief reactions⁷⁸, the development of clinical interventions to target prolonged grief is likely to be as important for helping bereaved youth as has been demonstrated in the treatment of bereaved adults. In addition, preventive interventions should not only target bereaved children with prolonged grief but also those with increased grief reactions 9 months following the death. Grief reactions in children in Class 2 gradually diminished over 33 months but they still showed functional impairment and were at increased risk of incident depression.

One of most consistently reported findings is that caregiver’s well-being following parental bereavement is a significant predictor of children’s well-being^38,39,78-81. We find that the combination of complicated grief in the surviving parent and in the child were particularly potent in predicting incident depression in children up to three years after the death. These findings have important clinical implications on intervention and prevention efforts. It is imperative to assess the surviving parent and intervene when appropriate to improve the outcome of parentally bereaved children. Adaptations of successful adult treatment approaches^27,82-84 to the treatment of prolonged grief in youth may require interventions that are family, rather than individually focused.

Future studies are needed to examine the long-term mental health and developmental outcomes in bereaved children, to examine the etiological and biological pathways by which prolonged grief exerts its effect, and to develop interventions to promote the relief of grief in parentally-bereaved youths.