Investigating Gene–Environment Interplay Between Bereavement and Polygenic Risk for Attention-Deficit/Hyperactivity Disorder on Externalizing Behaviors During Adolescence

Abstract

Objective

The death of a close friend during adolescence may have a negative impact on one’s mental health. However, existing literature has focused primarily on internalizing disorders, leaving the domain of externalizing behaviors understudied. Furthermore, the role of genetics in shaping post-bereavement psychopathology is not understood. In response, we examine potential interplay between polygenic liability for attention-deficit/hyperactivity disorder (ADHD) and bereavement of a close friend in shaping risk of post-loss externalizing symptoms among adolescents.

Method

We examined self-reported loss of a close friend between ages 12 and 16 years and a polygenic risk score (PRS) for ADHD in a sample of 3,922 adolescents from the Avon Longitudinal Study of Parents and Children. Outcomes at age 16.5 included 2 subscales of the Strengths and Difficulties Questionnaire: hyperactivity/inattention symptoms and conduct problems. Zero-inflated negative binomial (ZINB) models addressed the zero-skewed outcome distribution, and likelihood ratio tests for model comparison were used to detect gene–environment interplay.

Results

Nearly 1 in 10 adolescents reported losing a close friend. After adjusting for pre-loss psychopathology, bereavement independently predicted higher hyperactivity/inattention symptom count (bereaved vs nonbereaved: incidence rate ratio [IRR] = 1.18; 95% CI = 1.05-1.31), whereas the PRS for ADHD did not; neither were associated with the odds of zero (vs any) symptoms. Similarly, a model that included bereavement but not PRS best described the observed variation in conduct problems (bereaved vs nonbereaved: IRR = 1.40; 95% CI = 1.13-1.73).

Conclusion

Our findings reinforce the negative impact that losing a friend may have on an adolescent’s mental health, and suggest that externalizing symptoms among bereaved youth warrant clinical attention. Results from ZINB models reveal that bereavement may aggravate the severity or number of existing externalizing symptoms among those who would exhibit externalizing problems regardless. Genetic liability for ADHD may not augment the prediction of risk for psychopathology after bereavement, although better-powered samples are needed.

Plain language summary

Drawing on data from a sample of 3,922 adolescents from the Avon Longitudinal Study of Parents and Children in the UK, researchers found that nearly 1 in 10 adolescents reported experiencing the death of a close friend between ages 12 and 16. Relative to those who did not experience the loss of a friend, bereaved adolescents demonstrated higher hyperactivity/inattention and conduct problems scores at age 16.5. In this study, underlying genetic liability for ADHD did not meaningfully explain externalizing behavior scores after taking bereavement into account. This study reinforces the impacts of friend bereavement on adolescent mental health and calls for more clinical and methodological attention to this topic.

The death of a close friend represents a significant psychosocial stressor.1, 2, 3 From a developmental perspective, the loss of a close friend during adolescence coincides with a sensitive period characterized by heightened vulnerability to adverse mental health outcomes.⁴ Peer relationships emerge in adolescence as central sources of intimacy and support,⁵ and predict reduced depression and anxiety in early adulthood.⁶ In addition, the impact of genetic^7,8 and psychosocial stressors—and the ability to cope with stress—vary with age.⁹ During adolescence, stressful life events disrupt adaptive emotion processing, which can lead to adverse mental health outcomes.¹⁰ Although evidence is limited compared to investigations of parental loss, bereavement of a close friend has been shown to have an adverse impact on the mental health of adolescents and young adults through increased risk of depression, substance use, externalizing behaviors, conduct problems, and poor daily and social functioning.^1,11, 12, 13, 14, 15

Nevertheless, key questions remain regarding the effects of bereavement, especially among adolescents who lose a close friend. First, research on bereavement has focused largely on internalizing symptoms, leaving the domain of externalizing behaviors understudied.^1,16,17 Robust evidence points to the co-occurrence of adverse childhood experiences and externalizing disorders,¹⁸ and evidence is accumulating that childhood bereavement may exert a similar elevation in risk.19, 20, 21, 22, 23, 24, 25 For example, in one US sample, bereaved maltreated youth were at elevated risk for externalizing symptoms relative to nonbereaved maltreated youth.¹¹ Another study identified a longitudinal association between bereavement exposure and symptoms of both conduct disorder and substance misuse in a sample of US young adults.²² Longitudinal data from large samples are needed to better understand the relationship between bereavement of a close friend and externalizing behaviors.

Second, despite their elevated risk, most individuals who experience the death of a loved one do not exhibit lasting psychological effects.²⁶ Reasons for this heterogeneity in trauma response are not well understood, but evidence suggests that genetics may play a role.²⁷ Specifically, underlying genetic liability may increase one’s risk of psychopathology following an environmental stressor, especially for externalizing disorders such as attention-deficit/hyperactivity disorder (ADHD) and conduct problems, which are highly heritable and affect an estimated 5% to 10% of youth and adolescents worldwide.28, 29, 30, 31 Diagnoses such as ADHD capture the extreme end of a continuous distribution of externalizing symptoms in the population, as supported by genetic findings.³² Existing literature highlights a potential interplay between genetics and other childhood adversities, whereby underlying genetic liability increases the risk of externalizing behaviors following a stressful life event.33, 34, 35 However, evidence is mixed,³⁴ with small samples,³³ small interaction effects,³⁵ and a predominance of cross-sectional evidence linking childhood trauma to externalizing disorders.36, 37, 38, 39 Nevertheless, these findings warrant further exploration of genetic factors that potentially shape post-bereavement psychopathology.

In this study, we examine the potential interplay between polygenic liability for ADHD and bereavement of a close friend in shaping the risk of post-loss externalizing behaviors among adolescents. We use longitudinal data from the Avon Longitudinal Study of Parents and Children (ALSPAC) birth cohort to examine the relationship of adolescent bereavement with both hyperactivity/inattention symptoms and conduct problems. We derive a polygenic risk score (PRS) for ADHD, as this singular PRS has been shown to meaningfully predict both externalizing and other neuropsychiatric disorders, including among youth.40, 41, 42, 43, 44, 45 Our aim is to examine the extent to which polygenic risk for ADHD and experiencing the death of a friend elevate—either additively or interactively—externalizing symptoms among adolescents.

Method

Study Cohort

The sample consisted of participants enrolled in ALSPAC, a birth cohort of parents and their children. All pregnant women resident in Avon, United Kingdom, with expected delivery dates between April 1, 1991, and December 31, 1992 were invited to participate.46, 47, 48 Of the 15,447 pregnancies enrolled, there were 15,658 fetuses, 14,901 of which yielded individuals who were alive at 1 year of age. Detailed health and socio-demographic data were collected via self-, maternal-, and paternal-report questionnaires as well as in-person assessment clinics. The study website contains details of all of the data that are available through a fully searchable data dictionary and variable search tool (http://www.bristol.ac.uk/alspac/researchers/our-data/).

This analysis draws on genetic data as well as phenotypic data from pregnancy through age 16 years. Given that participants of non-European ancestry were excluded during ALSPAC’s genome-wide data generation and quality control procedures, we first filtered the full ALSPAC sample to individuals of European ancestry, resulting in 11,873 adolescents. For this initial filtering, European ancestry was defined by mother’s report of her own and her partner’s race/ethnicity as White. Among participants with genetic data available, this definition aligned almost perfectly with European ancestry principal components (99.7% concordance; Table S1, available online). Next, we further filtered to adolescents with observed genetic and outcome data, resulting in an analytic sample of 3,922 adolescents.

Measures

Bereavement

At 16 years, adolescents reported on the death of a close friend since age 12 years. Adolescents who indicated a positive response were prompted to describe the effect of this loss, with response options as a 5-point scale from “very unpleasant” to “very pleasant.” To define bereavement for this analysis, we recoded responses to this question into dichotomous categories of loss vs no loss, regardless of the reported effect of the loss. This timepoint for bereavement assessment was chosen because it was the first time that adolescents reported their own bereavement status, and that data were obtained on the adolescent’s loss of a close friend (vs the mother’s loss of a friend). Loss of a close family member (eg, parent or sibling) was not investigated as an exposure because of low prevalence and high risk of confounding from parental factors (eg, genetics, ADHD symptoms) that could not be reliably accounted for with available data.

Externalizing Behaviors

Externalizing behaviors were assessed using the Strengths and Difficulties Questionnaire (SDQ).⁴⁹ The SDQ is a 25-item behavioral screening tool composed of 5 subscales capturing hyperactivity/inattention symptoms, emotional symptoms, conduct problems, peer problems, and prosocial behavior. It was administered to the mother/caregiver at the ages of 6, 9, 11, 13, and 16.5 years. Responses were recorded on a 3-point scale ranging from “not true” to “certainly true.” For each subscale, prorated scores were used if no more than 2 items were missing; otherwise, the score was set to missing. Total SDQ scores were derived by summing item responses from the hyperactivity/inattention symptoms, emotional symptoms, conduct problems, and peer problems subscales and weighted according to the nonmissing items. Internalizing symptom subscores were similarly derived but using only the emotional symptoms and peer problems subscales; externalizing symptom subscores were based on the hyperactivity/inattention symptoms and conduct problems subscales. ALSPAC relied exclusively on mother/caregiver reports across all timepoints of SDQ administration. Thus, we were not able to compare mother-reported externalizing symptoms with those reported by adolescents themselves.

Outcomes for this analysis included 2 subscales at age 16.5 years: hyperactivity/inattention symptoms and conduct problems, each consisting of 5 items (subscale score range 0-10 for both). Reliability of the SDQ was high across administration waves at ages 6, 9, 11, and 13 years (Cronbach α by time point: 0.79, 0.80, 0.81, 0.78). Reliability was not calculated for the SDQ at age 16.5 years because item-level data were not available. Prior work has shown that these subscales demonstrate strong discrimination of a clinical diagnosis.^50,51

Covariates

Child sex (male or female) was determined from the birth notification, and was included based on both an observed difference in the prevalence of bereavement in this sample and prior evidence documenting gender differences in ADHD and other externalizing behaviors.52, 53, 54 Pre-bereavement mental health symptoms were captured using the SDQ total score at age 9 years and included so as to isolate post-loss psychiatric effects. Prior psychopathology is a well-known risk factor for post-traumatic psychiatric sequelae, and may play a role in the development of psychopathology following the loss of a friend (although we do not presume all reported losses in this sample to be considered traumatic).⁵⁵ Parent educational attainment, dichotomized as at least 1 parent holding a university degree (vs neither), was derived using mother’s reports at 32 weeks’ gestation of her and her partner’s educational attainment. Parent educational attainment was selected as a proxy for childhood socioeconomic status (SES) to account for socioeconomic disparities in both bereavement exposure and psychopathology,^2,56, 57, 58, 59 under the assumption that this proxy would be largely stable from pregnancy to adolescence. Additional proxies of SES, including a more detailed categorization of parent education, were not included because of small cell counts among male bereaved children.

Statistical Analysis and Missing Data

PRS for ADHD were calculated using PRSice-2.⁶⁰ Of the 8,094,094 single-nucleotide polymorphisms (SNPs) included in the base genome-wide association study (GWAS) by Demontis et al., 7,382,297 (91%) overlapped with our target SNP list and were included in PRS calculation.³² Ambiguous, duplicated, and mismatching variants were excluded by PRSice-2, resulting in 6,298,677 final SNPs prior to clumping. SNPs were clumped based on an R² threshold of 0.25 and a distance threshold of 500 kb. PRS were generated as the number of risk alleles weighted by effect size, using a p-value threshold of 0.1 (75,466 SNPs). This p-value threshold was chosen based on Demontis et al., who found that a PRS constructed using a threshold of 0.1 was significantly predictive of ADHD in a test cohort.³² Scores were standardized using mean and SD from the full PRS sample.

Preliminary analyses explored analytic sample characteristics and assessed the performance of the PRS. Given the relatively high prevalence of zero scores on the hyperactivity/inattention and conduct subscales of the SDQ in this nonclinical sample (18% and 47%, respectively), we implemented zero-inflated negative binomial (ZINB) models to investigate the relationship of bereavement and PRS for ADHD with hyperactivity/inattention and conduct symptoms. ZINB regression models count data with excessive zeros and over-dispersion.^61,62 Theory suggests that a greater number of zeros than expected for a count distribution (eg, Poisson, negative binomial) reflects 2 underlying zero-generating processes: one governed by a binary distribution that produces excess zeros, and one governed by a discrete distribution that generates counts (including zeroes).^61,62 Thus, a ZINB model consists of 2 components: logistic regression models the odds of a zero (vs nonzero) count, and negative binomial regression models the continuous count process. Although the proportion of zero scores for hyperactivity/inattention may not seem high, to our knowledge, there is no typical threshold for what should be considered excessive zeros, and recent studies have implemented zero-inflated models for outcomes with a proportion of zeros of around 20% or less.63, 64, 65 Furthermore, preliminary analyses suggested that a standard negative binomial distribution would in fact underpredict zero scores for hyperactivity/inattention (Figure S1, available online). We ran 4 ZINB models for each outcome, examining the following: (1) the main effect of bereavement, (2) the main effect of PRS for ADHD, (3) both main effects, and (4) both main effects plus their interaction. All primary models were adjusted for child’s sex, child’s pre-bereavement mental health symptoms, and parent education. Likelihood ratio tests (LRTs) were used to compare nested models, including the 2 degrees of freedom joint test of marginal association and gene–environment interaction, which compares model (4) to model (1), and the traditional test of interaction, which compares model (4) to model (3).⁶⁶

Three sensitivity analyses were performed for our primary models. First, we explored categorical and continuous quintiles of the raw PRS instead of the standardized continuous score. These analyses may be better powered to detect smaller genetic effects by comparing risk among individuals with the highest (quintile 5; vs the lowest) PRS. Second, instead of adjusting globally for pre-bereavement mental health symptoms via total SDQ score at age 9 years, we adjusted for both internalizing and externalizing scores to examine potential residual confounding by prior psychopathology at the factor level. We adjusted for these 2 scores instead of the 4 subscales that comprise the SDQ total score given that this is a community-based sample. Finally, we ran standard negative binomial models (ie, without zero-inflation component) for comparison of model fit, comparing the Akaike information criterion (AIC) of each standard negative binomial model to its ZINB counterpart using results from the first multiply imputed dataset.

Missing data and sample filtering were handled using inverse probability weighting (IPW) and multiple imputation (MI), where IPW was used to weight the analytic sample of n = 3,922 adolescents with observed genetic and outcome data to the full European cohort of n = 11,873 adolescents.⁶⁷ Specifically, standardized IPW weights were developed based on the following predictors, as done in a recent ALSPAC ADHD paper by Niina et al.: male sex, home ownership, maternal depression, maternal age at birth, maternal education, and parity.⁶⁸ All missing exposure (26%) and covariate data (2%-7%) were then multiply imputed using chained equations in a model that included the IPW weights (imputed datasets = 20, maximum iterations = 20).⁶⁹ All analyses, with the exception of PRS derivation, were performed in R 4.1.3.⁷⁰

Results

Among our weighted analytic sample of 3,887 adolescents of European ancestry (a small difference from the crude analytic sample size is expected when estimating sample weights⁷¹), 8.3% were bereaved of a close friend between ages 12 and 16 years. Preliminary analyses highlighted important differences between our analytic sample and the full European cohort (Table S2, available online). Individuals included in the analytic sample were more likely to be female (51.2% in analytic sample, vs 48.5% in full sample) and to have a mother with a university degree (21.3% vs 13.0%). IPW/MI improved representativeness of the analytic sample relative to the full European cohort (Table S3, available online). Among our analytic sample, mean scores for hyperactivity/inattention symptoms and conduct problems were 2.7 (SD 2.2) and 1.1 (SD 1.4), respectively. Additional preliminary analyses identified a strong association of PRS for ADHD with hyperactivity/inattention symptoms and conduct problems scores, with slight variations by age (Table S4, available online). At age 16 years, every 1-SD increase in PRS was associated with a 6% higher (incidence rate ratio [IRR]=1.06; 95% CI = 1.03-1.09) expected hyperactivity/inattention score and a 7% higher (1.07; 1.01-1.12) expected conduct problems score. Table S5 and Figure S2, available online, provide additional preliminary analyses regarding the PRS and other study variables.

In primary analyses, we compared ZINB models using LRTs to identify the role of bereavement and PRS for ADHD in shaping externalizing symptoms in mid-adolescence. We found clear evidence of a main effect of bereavement on hyperactivity/inattention symptoms score (Table 1). According to the LRT, a model with both main effects of interest yielded a better fit than a model with just the PRS main effect (F = 3.95, p = .02) but not one with just the bereavement main effect (F = 2.00. p = .14); the model that included an interaction term did not fit better than the model with only main effects (F = 0.31, p = .73). According to the model with the main effect of bereavement alone, bereavement was positively associated with hyperactivity/inattention symptom score but was not associated with the odds of zero (vs any) symptoms. Specifically, the expected hyperactivity/inattention symptom score was 18% higher (IRR = 1.18; 95% CI = 1.05-1.31) for bereaved vs nonbereaved individuals. According to the model with both main effects, the estimated effect of bereavement on hyperactivity/inattention symptoms was nearly identical to that from the latter model, and the expected hyperactivity/inattention symptom score was 2% higher (1.02; 1.00-1.05) per 1-SD increase in PRS (Figure 1).

Table 1.

Gene–Environment Interplay of Bereavement of a Close Friend and Polygenic Risk Score for Attention-Deficit/Hyperactivity Disorder (ADHD) on Hyperactivity/Inattention Symptoms After Loss

Model	ZINB model component	Coefficient	IRR (count) or OR (zero-inflation)	Lower 95% CI of IRR or OR	Upper 95% CI of IRR or OR	Crude SE	LRT, model vs BER ME		LRT, model vs PRS ME		LRT, model vs BER + PRS
							F	p	F	p	F	p
BER ME	Count	Bereavement	1.176	1.053	1.312	0.055
	Zero-inflation	Bereavement	0.925	0.446	1.919	0.371
PRS ME	Count	PRS	1.023	1.000	1.048	0.012
	Zero-inflation	PRS	0.980	0.816	1.178	0.094
BER + PRS	Count	Bereavement	1.174	1.051	1.312	0.056	1.999	.135	3.947	.024
		PRS	1.023	0.999	1.047	0.012
	Zero-inflation	Bereavement	0.923	0.445	1.917	0.372
		PRS	0.979	0.816	1.175	0.093
INTERACT	Count	Bereavement	1.179	1.056	1.315	0.055	1.174	.320	2.087	.092	0.315	.730
		PRS	1.026	1.000	1.052	0.013
		Interaction	0.967	0.877	1.067	0.050
	Zero-inflation	Bereavement	0.915	0.423	1.979	0.393
		PRS	0.983	0.813	1.190	0.097
		Interaction	0.941	0.383	2.313	0.457

Note: Zero-inflated negative binomial (ZINB) models were run with inverse probability weights on multiply imputed data. Polygenic risk score (PRS) for ADHD was operationalized as standardized continuous score. BER ME denotes model with main effect of bereavement; PRS ME denotes model with main effect of PRS for ADHD; BER + PRS denotes model with main effects of bereavement and PRS for ADHD; INTERACT denotes model including main effects of bereavement and PRS for ADHD as well as their interaction. All models were adjusted for the same list of covariates. IRR = incidence rate ratio; LRT = likelihood-ratio test; OR = odds ratio.

Figure 1.

Zero-Inflated Negative Binomial Estimates of Bereavement and Polygenic Risk Score (PRS) for Attention-Deficit/Hyperactivity Disorder (ADHD) on Externalizing Symptoms After Loss

Note:Zero-inflated negative binomial (ZINB) models were run with inverse probability weights on multiply imputed data. For each outcome, 2 different ZINB model results are presented along the y-axis: one modeling the main effect of bereavement and one modeling the main effect of PRS for ADHD (plus covariates, but without mutual adjustment). Odds ratios (ORs; for the ZINB model component modeling zero vs any symptoms) or incidence rate ratios (IRRs; for the component modeling symptom count) + 95% CIs are presented along the x-axis.

For conduct problems, results suggested that bereavement alone best described the observed variation in externalizing symptoms in this sample (Table 2). The interaction model did not fit the data better than the model with both main effects (F = .17; p = .85). Furthermore, although the model with both main effects demonstrated better fit than the PRS main effect model (F = 4.07; p = .02), it did not demonstrate superior fit to the bereavement main effect model (F statistic = 0.25; p = .78). In the bereavement main effect model, the expected conduct problems score was 40% higher (IRR = 1.40; 95% CI = 1.13-1.73) among bereaved individuals than among nonbereaved individuals. In the model with both main effects, the effect estimate for PRS for ADHD contained the null (IRR = 1.01; 95% CI = 0.97-1.06), whereas the effect estimate for bereavement was nearly identical to that from the latter model.

Table 2.

Gene–Environment Interplay of Bereavement of a Close Friend and Polygenic Risk Score for Attention-Deficit/Hyperactivity Disorder (ADHD) on Conduct Problems After Loss

Model	ZINB model component	Coefficient	IRR (count) or OR (zero-inflation)	Lower 95% CI of IRR or OR	Upper 95% CI of IRR or OR	Crude SE	LRT, model vs BER ME		LRT, model vs PRS ME		LRT, model vs BER + PRS
							F	p	F	p	F	p
BER ME	Count	Bereavement	1.397	1.127	1.730	0.107
	Zero-inflation	Bereavement	1.750	0.640	4.785	0.510
PRS ME	Count	PRS	1.016	0.973	1.062	0.022
	Zero-inflation	PRS	1.007	0.739	1.372	0.158
BER + PRS	Count	Bereavement	1.395	1.125	1.730	0.108	0.249	.780	4.075	.024
		PRS	1.013	0.971	1.058	0.022
	Zero-inflation	Bereavement	1.749	0.637	4.804	0.513
		PRS	0.996	0.732	1.355	0.157
INTERACT	Count	Bereavement	1.386	1.113	1.727	0.110	0.207	.935	1.955	.118	0.167	.846
		PRS	1.010	0.964	1.058	0.024
		Interaction	1.040	0.879	1.230	0.085
	Zero-inflation	Bereavement	1.693	0.597	4.805	0.529
		PRS	0.989	0.706	1.385	0.172
		Interaction	1.077	0.327	3.551	0.604

Note: Zero-inflated negative binomial (ZINB) models were run with inverse probability weights on multiply imputed data. Polygenic risk score (PRS) for ADHD was operationalized as standardized continuous score. BER ME denotes model with main effect of bereavement; PRS ME denotes model with main effect of PRS for ADHD; BER + PRS denotes model with main effects of bereavement and PRS for ADHD. INTERACT denotes model including main effects of bereavement and PRS for ADHD as well as their interaction. All models were adjusted for the same list of covariates. IRR = incidence rate ratio; LRT = likelihood-ratio test; OR = odds ratio.

In sensitivity analyses, results that used quintile-based operationalizations of the PRS for ADHD were largely consistent with primary results (Table S6, available online). Results from the standard negative binomial (ie, not zero-inflated) models were similar for the outcome of conduct problems, but differed for the outcome of hyperactivity/inattention symptoms (Table S7, available online). Namely, LRTs suggested that a model with both main effects fit the data best, and expected hyperactivity/inattention symptom score was 17% higher (p < .01) for bereaved vs nonbereaved individuals and 3% higher (p = .04) per SD increase in PRS. However, according to the AIC criterion, the ZINB models fit better than these standard negative binomial models. Finally, adjusting for SDQ internalizing and externalizing scores in place of the SDQ total score did not have an impact on results regarding the outcome of hyperactivity/inattention symptoms, and minimally attenuated the estimated effects of bereavement on conduct problems (Table S8, available online). In the bereavement main effect model that adjusted for child sex, parental education, and SDQ internalizing and externalizing scores, the expected conduct problems score was 36% higher (IRR = 1.36; 95% CI = 1.11-1.68) among bereaved individuals than among nonbereaved individuals.

Discussion

In this study, we used longitudinal data to examine the impact of the death of a close friend during adolescence on externalizing behaviors as captured by the SDQ, and the extent to which genetic risk for ADHD shaped these behaviors. First, we found that bereavement of a close friend between the ages of 12 and 16 years was associated with increased symptom scores of both hyperactivity/inattention and conduct problems at age 16.5. Importantly, however, this relationship was observed only among those that had a positive symptom count; we did not observe differing odds of zero (vs any) symptoms by bereavement status. Second, we found evidence that after accounting for bereavement and psychiatric symptoms at age 9 years, among other covariates, PRS for ADHD did not independently associate with elevated hyperactivity/inattention symptom scores or conduct problems among adolescents. Overall, our results highlight the impact of the death of a close friend on elevated externalizing symptoms in mid-adolescence, and the potential absence of interaction between bereavement and genetic liability in predicting elevated externalizing symptoms at this age.

Our study advances the evidence base linking bereavement to externalizing behaviors during adolescence in at least 3 key ways. First, our findings reinforce the negative impact that losing a friend may have on an adolescent’s mental health. Compared to familial loss, the loss of a close friend or peer remains severely understudied in the extant bereavement literature,¹⁴ including during a period in which we expect the role of friends to be particularly salient.⁷² Compared to their nonbereaved counterparts, bereaved adolescents in our study demonstrated approximately 20% to 40% higher externalizing symptom scores at age 16.5 years, even after adjusting for pre-loss psychopathology. Although direct comparisons to other studies are challenging, our findings generally align with existing evidence. For example, Thompson et al. reported a mean of 55.2 (SD = 10.6) on a parent-reported externalizing problems scale among a group of parentally bereaved youth, compared to 50.5 (9.7) among their nonbereaved counterparts (crude percentage difference of 9%).²⁴ The mechanisms underlying this association suggest an important area for future research. It may be that externalized behavior in the aftermath of a loss attenuates social network recovery. The magnitude of this effect may be especially deleterious, given that prior work has found that young adults aged 18 to 24 years are especially responsive to social network recovery after bereavement relative to other age groups.³

By focusing on the loss of a close friend rather than the loss of a parent or sibling, we minimized the risk of bias and confounding that would be expected in a gene–environment study focused on parental loss and offspring psychopathology. Specifically, biological (eg, parent genetic liability) and environmental (eg, parent psychopathology) influences73, 74, 75, 76 on the child’s risk of bereavement and psychopathology would likely confound our estimates; in ALSPAC, insufficient genetic and phenotypic data exist from both biological parents to robustly explore these questions. Although similar influences are not implausible in the context of friend loss, given the relationship of ADHD with mortality-related outcomes⁷⁷ and the potential relationship between externalizing behaviors and both social network patterning and exposure to loss,^12,78 we argue that they are much less likely, especially after adjusting for pre-loss internalizing and externalizing symptoms. Thus, we can be more confident in our findings that speak to the enduring influence of environment on externalizing behaviors in adolescence, regardless of childhood diagnosis status.

Second, we extend existing research that has examined the impact of bereavement on mental health by applying ZINB models to the study of mental health symptom scores. Prior studies on this topic have focused on psychiatric disorder diagnoses,^20,23 which may overlook heterogeneity within diagnosis groups (eg, severity of symptoms among those diagnosed with ADHD), or count outcomes,^1,11,19 leading to models that can underpredict the number of individuals without any mental health symptoms. We identified bereavement as a predictor of higher symptom count among those with symptoms, but not of dichotomized odds of any symptoms vs none, suggesting that bereavement may aggravate the severity or number of existing symptoms among those who would exhibit externalizing problems regardless. Failure to identify an association in the dichotomized analysis may also be explained by the loss of power induced when dichotomizing a continuous variable, by power constraints driven by a binary exposure of low prevalence, by the low prevalence of participants with no symptoms at age 9 years but positive symptoms at age 16.5 (and vice versa), or by social desirability and other reporting biases in parent reports. Our findings do not necessarily reflect the absence of an association between bereavement and diagnosed ADHD or conduct disorder, as these diagnoses dichotomize symptom counts at the upper extreme.^32,79 Nonetheless, the strong relationship that we identified between bereavement and symptom count reflects the impact of bereavement on behavior severity, including among adolescents never diagnosed with prior externalizing disorders.

Finally, we highlight the dominance of this specific environmental stressor, relative to genetic liability, in shaping externalizing symptoms in mid-adolescence. After adjusting for psychiatric symptoms at age 9 years, which marks the tail end of the expected age of onset of most ADHD cases,²⁹ we observed no independent association between PRS for ADHD and higher externalizing symptoms at age 16.5. This result may be explained by potential differences in the underlying etiology of childhood-onset, persistent, and late-onset ADHD,^8,80,81 although our crude preliminary analyses demonstrated that the PRS for ADHD based on the GWAS by Demontis et al.³² was predictive of hyperactivity/inattention symptoms and conduct problems count across time points. More likely, this finding is a product of the small expected effect size of genetic liability on externalizing symptoms in adolescence, potentially coupled with statistical power constraints in this community sample.

Alongside these conclusions, key study limitations should be considered. First, our analytic sample included only individuals of European ancestry, as ALSPAC does not provide genetic data for the rest of the cohort, and is not representative of the UK population.⁴⁸ Our sample may systematically exclude marginalized groups, including those who are more likely to experience bereavement and other childhood adversities.^82,83 In that case, our study may provide a lower bound on the true prevalence of adolescents bereaved by a friend in the general UK population. Second, further details about the loss, including the specific timing of death, were not available, and externalizing symptoms were measured soon after the bereavement age window. Given the small sample of individuals (n < 10) who reported the event as having no or a “pleasant” effect, we did not explore moderation by self-reported effect of the loss. Both timing (eg, recency of loss, age at loss)20, 21, 22 and relational factors (eg, quality of relationship, type of death)^15,25 may moderate the relationship and should be investigated. Third, we were unable to account for compounded adversities, including prior losses of a friend or close family member, which may predict worse mental health outcomes among the bereaved. Residual confounding related to prior psychopathology, socioeconomic status, and other predictors of friend loss is likely as well. Fourth, we were underpowered to explore the relationship between bereavement and a diagnosis of ADHD or conduct disorder, given low prevalence in the ALSPAC cohort, and we relied on parent reports of adolescent symptomatology. Relatedly, our narrow focus on externalizing disorders limited our ability to interpret any possible association between bereavement and other conditions that could be associated with nonspecific symptoms of inattention, including substance abuse, mood, and anxiety disorders. Fifth, we did not have access to linked education data to include school-related covariates that may confound the relationship between friend loss and mental health outcomes. Finally, gene–environment interaction studies have historically been underpowered,^30,84 requiring large samples to detect small genetic effect sizes, and this study is no exception.

Despite these limitations, 3 clinical and 2 methodological implications emerge from our findings. First, our results suggest that externalizing symptoms among bereaved youth warrant clinical attention. Clinicians have resources to treat internalizing symptoms among bereaved youth,⁸⁵ but less guidance exists for treating externalizing symptoms. Second, albeit speculative, resources to help bereaved adolescents recover social networks after a loss could help reduce the risk of psychopathology. Third, our results suggest that genetic liability for ADHD does not augment the prediction of risk for psychopathology after bereavement. However, further work is needed with adequately powered samples to interrogate gene–environment interplay in the context of bereavement and the potential role of PRS in predicting individual-level responses to trauma and loss.⁸⁶ Methodologically, our application of zero-inflated models to study psychiatric symptom counts revealed nuances in the nature of the relationship among bereavement, genetic risk, and externalizing behaviors. Studies that examine mental health symptom counts may benefit from models such as the ZINB model to better address zero-skewed data and to isolate the signal of interest. Finally, better prevalence estimates of friend bereavement in diverse samples are needed to guide both research and intervention.

CRediT authorship contribution statement

Ana Lucia Espinosa Dice: Writing – review & editing, Writing – original draft, Software, Project administration, Formal analysis, Conceptualization. Henri M. Garrison-Desany: Writing – review & editing. Andrew Ratanatharathorn: Writing – review & editing, Conceptualization. Henning Tiemeier: Writing – review & editing. George Davey Smith: Writing – review & editing. Christy A. Denckla: Writing – review & editing, Writing – original draft, Supervision, Project administration, Funding acquisition, Conceptualization.