Associations between adolescent pain and psychopathology in the Adolescent Brain Cognitive Development (ABCD) study

Abstract

Pain and psychopathology co-occur in adolescence, but the directionality and etiology of these associations are unclear. Using the pain questionnaire and the Child Behavior Checklist from the Adolescent Brain Cognitive Development study (n = 10,414 children [770 twin pairs] aged 12 to 13), we estimated longitudinal, co-twin control, and twin models to evaluate the nature of these associations. In two-wave cross-lag panel models, there were small cross-lag effects that suggested bidirectional associations. However, the co-twin control models suggested that most associations were familial. Pain at age 12 and 13 was mostly environmental (A = 0–12%, C = 15–30%, E = 70–73%) and the twin models suggested that associations with psychopathology were primarily due to shared environmental correlations. The exception was externalizing, which had a phenotypic prospective effect on pain, a significant within-family component, and a non-shared environmental correlation at age 12. Environmental risk factors may play a role in pain-psychopathology co-occurrence. Future studies can examine risk factors such as stressful life events.

Introduction

Adolescent chronic pain is debilitating and common, affecting 11 to 44% of youth (Gobina et al., 2019; King et al., 2011). Pain in adolescence increases the risk of chronic pain and other physical health issues in adulthood (Hestbaek et al., 2006; Mun et al., 2021) and is associated with poorer social, vocational, and educational outcomes (Murray et al., 2020). Psychopathology frequently co-occurs with pain in adolescence (King et al., 2011; Vinall et al., 2016; Voepel-Lewis et al., 2021; Youssef et al., 2008) and is a risk factor for pain development (Kolaitis et al., 2021; Mun et al., 2021). Some research suggests that pain in adolescence also increases risk for later psychopathology (Noel et al., 2016; Youssef et al., 2008), suggesting bidirectional effects between the phenotypes. What is often not considered is that underlying risk factors such as genetic or shared environmental effects, rather than direct causal effects, may drive these associations. Furthermore, associations may reflect different mechanisms that depend on the type of psychopathology contributing to the pain-psychopathology co-occurrences. Understanding the etiological influences on pain and psychopathology is important to elucidate what drives these phenotypic associations and to identify adolescents at risk for developing pain and psychopathology. Here, we use a longitudinal adolescent sample from the Adolescent Brain Cognitive Development (ABCD) study to examine phenotypic directional effects between pain and different psychopathologies. We also leverage co-twin control and classic twin models to ascertain whether these associations survive familial confounding and reflect genetic and/or environmental relationships.

Etiology of adolescent pain

Chronic pain in adulthood appears to be due to a mix of genetic and environmental effects (Vehof et al., 2014), but there is conflicting evidence about the etiology of pain in early adolescence. For example, 41% of the variance in low back pain in 11-year-old twins was attributable to shared environmental influences, 59% to non-shared environmental influences, and 0% to genetic influences (El-Metwally et al., 2008). Another study found neck pain in 11- to 12-year old Finnish twins was influenced only by genetic and non-shared environmental effects (Ståhl et al., 2013). These conflicting results could indicate that specific pain conditions have varying etiologies. However, a longitudinal Canadian twin sample examined the latent growth factors that contributed to a general pain construct at ages 12, 13, and 14 years and found that they were accounted for by genetic and non-shared environmental effects (Battaglia et al., 2020), but no shared environmental influences. More studies are needed to clarify the sources of influences on general pain in the developmentally sensitive time of young adolescence.

Internalizing and pain

The co-occurrence between internalizing psychopathology and adolescent pain is well-studied. Children who report high somatic complaints in middle childhood are also at greater risk of anxiety, depression, and PTSD (Voepel-Lewis et al., 2021). However, there are contradictory findings in which internalizing psychopathology appears to heighten the risk of multiple pain syndromes, as well as be the outcome of pain. Higher affective disorder profiles predicted higher incidence of back, neck, abdominal, and head pain in adolescents (Stapp et al., 2022). A similar trend was observed in 3- to 6-year-old toddlers, with internalizing symptoms predicting later development of pain (Kolaitis et al., 2021). Conversely, abdominal pain in adolescents was found to increase the risk of later development of depressive symptoms (Youssef et al., 2008); and the National Longitudinal Study of Adolescent to Adult health found that adolescents who were experiencing chronic pain had 1.33 increased odds for anxiety disorders and 1.38 for depressive disorders in adulthood in comparison to the general population (Noel et al., 2016).

These phenotypic, longitudinal correlations could suggest direct, causal effects; however, they also may reflect overlapping genetic or shared environmental influences. At age 10 in the same ABCD sample examined in this study (Freis et al., 2022), internalizing behaviors display genetic and shared environmental influences that could overlap with those for pain. A predisposition to anxiety can lead to a maladaptive appraisal of acute pain and lead to chronification in children (Asmundson et al., 2012). Furthermore, a systematic review of mostly adult twin studies found that pain conditions show genetic and environmental correlations with depression and anxiety (Khan et al., 2020). A recent study found a common factor of self-reported pain, anxiety, and depression in late adolescence that displayed significant genetic, shared environmental, and non-shared environmental influences (Scaini et al., 2022), suggesting that familial influences play a partial role in increasing both pain and internalizing risk.

While it is well-established that there are shared genetic and environmental influences on internalizing and pain, the association may not be entirely familial. A longitudinal twin study of 12–14 year old adolescents found that in addition to genetic correlations between the latent growth factors of pain and anxiety/depression, there were also non-shared environmental correlations (Battaglia et al., 2020). The non-shared environmental correlations could indicate a direct effect between internalizing and pain. Co-twin control analyses can be used to assess whether the association between internalizing and pain survives when controlling for familial confounding (Carlin et al., 2005).

Externalizing and pain

Externalizing problems also are phenotypically associated with pain. For example, children with musculoskeletal pain displayed more externalizing problems (Varni et al., 1996), and externalizing symptoms have been shown to increase the odds of later musculoskeletal pain (Andreucci et al., 2021). Externalizing behaviors could directly affect pain by increasing engagement in high risk, injury-prone behaviors such as substance use or violence (Ranney et al., 2018), which could increase pain incidence. However, pain could predate externalizing, as the experience of pain could incite more irritable or aggressive behavior. To our knowledge, this will be the first study to examine the phenotypic association between pain on later externalizing.

There is less research on the etiology of the relationship between externalizing and pain. At age 10 in this ABCD sample, externalizing variance was 76% influenced by genetic effects and 26% by non-shared environmental effects (Freis et al., 2022), but to our knowledge, no study has looked at the etiological correlates with pain. A longitudinal co-twin control study found that externalizing behaviors predicted later headaches, even when controlling for familial confounding, suggesting that externalizing had an exacerbating effect on pain development rather than predisposition driving the association (Virtanen et al., 2004). However, it is unclear if this relationship extends to general pain, whether pain predicts later externalizing, and the nature of this association.

Pain and other psychopathology

Phenotypic and genetic studies examining pain with other psychopathology, such as thought, attention, and social problems, are sparse. Thought problems is a construct that encompasses obsessive thoughts, compulsions, strange behaviors or ideas, psychosis, and sleep problems in children (Achenbach, 2000). Children with orofacial pain have higher frequencies of thought problems (Al-Khotani et al., 2016); however, to our knowledge, no longitudinal or genetically informed study has examined this thought scale with general pain.

Children experiencing pain also have difficulty with social behaviors. Children with chronic multi-site pain exhibited co-morbid social anxiety, attention problems, and internalizing and externalizing problems (Skrove et al., 2015). As pain is debilitating, children experiencing chronic pain often cannot participate in social activities (Palermo et al., 2008). During a developmentally sensitive time, deficits in social experiences can have disruptive impacts into adulthood (Murray et al., 2020), suggesting pain could have an exacerbating effect on social problems. Social problems were found to be influenced by genetic and non-shared environmental factors at age 10 in the ABCD sample (Freis et al., 2022), but it is unclear whether any of these etiological components will be shared with pain.

Attention problems also appear to be associated with pain in adolescents and this relationship could be bidirectional (Voerman et al., 2017). Being in pain is cognitively burdensome and demands attention, which could intensify attention problems (Eccleston & Crombez, 1999). Meanwhile, deficits in executive functioning are associated with attention problems, including in the current sample (Freis et al., 2022). Executive function deficits have been found to predispose pre-operative patients to post-operative pain chronification (Attal et al., 2014). Abnormal brain structure in executive functioning regions implicated in attention problems have been found to differentiate whether sub-acute pain will lead to chronification or recovery (Mansour et al., 2013), suggesting that neurobiological substrates common to attention problems and executive functions could lead to increased pain.

There may also be underlying shared risk factors that predispose an individual to both pain and attention problems. One co-twin control study found that the association between childhood attention problems and neurological conditions, including migraine headaches, was entirely due to familial effects (Pan & Bölte, 2020). At age 10 in the ABCD sample, the heritability of attention problems was 66%, with no shared environmental influence and 44% non-shared environmental influences (Freis et al., 2022). We will test whether general pain shares any genetic risk with attention problems or whether there is an exacerbating effect of one on the other.

Present study

As reviewed in the prior sections, pain and psychopathology in childhood are often phenotypically correlated. There appear to be bidirectional relationships, but these may be due to familial confounds. The ABCD study collects data annually, providing a rare opportunity to look at longitudinal associations. The ABCD also includes a sub-sample of twins allowing for genetically informed analyses. Thus, in the current study, we estimate phenotypic, co-twin control, and twin models for the relationships between pain and psychopathology in early adolescence using two waves of the ABCD study. The structure of the ABCD enables us to conduct many novel associations, such as between pain and later externalizing, and etiological associations between pain and other psychopathology (i.e., thought, social, and attention problems).

While many psychopathology dimensions have distinct presentations, there is shared variance across individual disorders and clusters of disorders (Caspi et al., 2014; Clark et al., 2021), suggesting shared risk factors. For example, there are common sources of genetic and shared environmental influences for externalizing and internalizing problems (Gjone & Stevenson, 1997). Thus, it is possible that pain is associated with multiple dimensions of psychopathology because of shared risk factors across these dimensions; however, it is also possible that associations reflect unique risk factors, or a combination of shared and unique risk factors. Given that many of these associations have not been examined before, we opted to start simple by looking at these different pain-psychopathology relationships independently.

We first phenotypically characterize the directional relationships between pain and psychopathology using cross-lag panel models (CLPM) at ages 12 and 13, which can provide information on prospective effects. Second, we use co-twin control models to examine whether associations between pain and psychopathology survive when controlling for familial confounding. The co-twin control method is a quasi-experimental approach that uses a twin’s co-twin as a matched control within a regression framework (McAdams et al., 2021). Third, we use the classical twin design to estimate genetic and environmental influences on pain incidence at ages 12 and 13. Finally, we assess whether there are genetic or environmental correlations between pain and psychopathology using bivariate twin models.

Method

Sample

The ABCD study consists of 11,846 children at baseline (age M = 9.91, SD = 0.62, 48% girls) recruited from multiple sites across 21 States in the United States of America (ABCD version 4.0 https://abcdstudy.org/). The ABCD study has included 3 yearly follow-ups since the baseline assessment. Embedded in the ABCD dataset is a twin design as well as incidental twins (n = 2,108 individual twins). The rest of the sample consists of triplets (n = 30), non-twin siblings (n = 1,810) and singletons (n = 7,898). Phenotypic analyses (n = 10,515) were conducted on the entire available sample at age 12 (follow-up 2; n = 10,414, age M = 12.00 years, SD = 0.66, 48% girls) and age 13 (follow-up 3; n = 6,251, age M = 12.90 years, SD = 0.64, 47.3% girls). We only had 3-year follow up data for 62.91% of follow up 2 participants because the data release (4.0) was incomplete.

Twin analyses only included same-sex twins that were recruited from 4 dedicated twin sites (Minnesota, Colorado, Virginia, and Missouri (Iacono et al., 2018)). Of the 1725 twins, 672 were monozygotic (MZ; 53% female) and 869 were dizygotic (DZ; 49% female), as indicated by genetic data (MZ = > 0.9; DZ = 0.4–0.8). At ages 12/13, there were 529/423 MZ twins and 690/567 DZ twins who completed the pain and psychopathology data.

Measures

Pain Questionnaire

Self-reported presence and severity of a current pain episode were assessed at follow up 2 and 3. Participants reported whether they were experiencing a pain episode in the last month (0 = no; 1 = yes) and identified the physical location(s) of pain using a body map. Participants then rated their pain severity at its worst over the last month on a scale of 0–10, with 0 being no pain and 10 being the worst pain imaginable.

We combined the binary pain episode in the past month variable and pain severity variable as follows: Participants who did not have pain in the last month had a score of 0, and participants who did report pain in the past month had a score equal to its severity (0–100). While the physical location of pain is an interesting dimension of pain, we did not include location in these analyses due to low endorsements for any one location. Self-reported pain intensity is a well-established measure for pediatric pain (Varni et al., 1996).

Child Behavior Checklist

The Child Behavior Checklist (CBCL) (Achenbach, 2000) is a parent-report measure of childhood psychopathology (Albores-Gallo et al., 2007) that is often used for measuring behavioral and emotional problems at this developmental period (Achenbach, 2000). The CBCL consists of 119 items (e.g., “Withdrawn, doesn’t get involved with others.”). Parents rated whether each item described their child over the past 6 months with a 3-point scale (0, 1, 2, which correspond to ‘not true’, ‘somewhat or sometimes true’, or ‘very or often true’). Items are combined to create eight separate composite scales: anxious/depressed, social problems, withdrawn, somatic complaints, thought problems, attention problems, delinquent behavior, and aggressive behavior. A broader internalizing composite score consists of anxious/depressed, withdrawn, and somatic complaints, and a broader externalizing score consists of delinquent and aggressive behavior. The somatic complaints scale includes items that are related to pain, such as “aches or pains” and “headaches;” however, it also includes items not directly related to pain, such as “nightmares” and “feels dizzy.”

Statistical Procedures

Data transformation

The data for the current study were obtained from the ABCD National Institute of Mental Health Data Archive (NDA). We binned the pain and psychopathology variables to avoid biased estimates in both the cross-lag panel and twin modelling, due to the non-normal, asymptotic distribution of the traits (Verhulst & Neale, 2021). Pain intensity was binned into 3 groups: 0 = no pain, 1–5 = mild/moderate pain, 6–10 = moderate/severe pain. The bins were formed to keep variability and allow for polychoric correlation matrices to be estimated without any empty cells when splitting data by zygosity and twin number. CBCL bins were based on those used in a previous study (Hatoum et al., 2018): each scale was cut at 0, 1–3, 4–10. We did not include a >10 bin as the endorsement was too low in this sample. For the co-twin control models, the outcomes were ordinal, but we used square-root transformed measures as predictors. Normal distributions for the independent variables is not an assumption of regression, and keeping the predictors as continuous enabled more variability for the between family and within family predictor means.

Model fitting

All analyses were conducted in Mplus version 8.3 (Muthén & Muthén, 2017). The cross-lag panel models (CLPMs) and twin models used the diagonally weighted least squares means and variances adjusted (WLSMV) estimator. The mixed models used the robust maximum likelihood (MLR) estimator. Phenotypic analyses were conducted using Mplus’ type=COMPLEX option to adjust chi-square tests and standard errors for non-independence due to family clustering using a sandwich estimator. Good model fit was assessed using the following criteria: root-mean-square error approximation (RMSEA) < 0.06 and Comparative Fit Index (CFI) > 0.95 (Hu & Bentler, 1998). Significance was assessed as p < 0.05 for the Wald tests of the parameters, except for significance tests of variance components in the genetic models, which were assessed with chi-square difference tests using Mplus’s difftest procedure for the WLSMV estimator. We use this method because the standard errors in the genetic models are not invariant to model parameterization, but the chi-square difference test is (Neale et al., 1989). We also provide asymmetric boot-strapped 95% confidence intervals, as variance components in genetic models are not normally distributed (Neale & Miller, 1997). Specifically, we used Mplus’s OUTPUT: CINT(BOOT) option with 1000 bootstrap draws to estimate the models with 1000 samples of the same size drawn with replacement from the data; the parameter estimates for these 1000 models are then ordered from lowest to highest and the upper and lower bounds of the confidence intervals for each parameter are the 25^th and 975^th estimates, respectively.

Cross-lag panel models.

Significant phenotypic correlations between pain and psychopathology at both time points in the full sample were candidates for the CLPMs. The traditional CLPM ascertains direct influences in longitudinal panel data by controlling for within-timepoint correlations and autoregressive effects, while estimating cross-lagged correlations (Selig & Little, 2012). CLPMs may not be well-suited for causal claims as they are unable to distinguish between within-person and between-person effects, but they are useful for elucidating longitudinal associations between variables and their potential bidirectional impacts (Selig & Little, 2012). Random-intercept CLPMs are able to distinguish between within-person and between-person effects, but they require more than two timepoints (Hamaker et al., 2015).

Co-twin control models.

Co-twin control models were used to assess direct effects of psychopathology on pain as well as for pain on psychopathology, while accounting for familial confounds. We fit multi-level models that estimated both between and within family effects. The between-family predictors are created by calculating the mean CBCL or pain severity scale for each family’s twin pair. The outcome is regressed on these family means to control for familial effects, which can include both genetic and shared environmental contributors (Carlin et al., 2005). The within-family predictors are created by calculating the deviation of each twin’s CBCL or pain severity score from their family mean. Regressing the outcome on these discordance estimates shows the effect of the predictor on pain or psychopathology, controlling for familial (between-family) confounds (Carlin et al., 2005). The within-family effects were allowed to interact with zygosity. In Mplus, this interaction was accomplished by creating a random slope for the within-family effect that was then regressed on zygosity, but which has residual variance fixed to zero. Significance of this interaction estimate elucidates whether the effect differs across zygosity groups. If the within-family effect is significant in both MZ and DZ twins, this can be evidence for a causal effect. If the within-family effect is only significant in DZ twins, this can be suggestive of partial genetic confounding. Although a larger within-family effect in DZ twins is sometimes interpreted as evidence for genetic influences, absence of this pattern is not reliable evidence against genetic effects (Carlin et al., 2005). We used Mplus’ type = TWOLEVEL RANDOM to account for the random intercepts. A logit link function was employed, as the outcome was ordinal.

Twin models.

A series of twin models were conducted to ascertain shared etiological correlations between pain and psychopathology cross-sectionally and longitudinally. The classical twin design uses assumptions of the biometric genetic theory to decompose variance and/or covariance of traits into genetic and environmental components (Rijsdijk & Sham, 2002). Since MZ twins share 100% of their segregating genes and DZ twins share on average 50%, this information can be used in structural equation modelling to decompose the variance of a phenotype into three latent sources of contribution: additive genetic effects (A), shared environmental effects (C; those that lead siblings to correlate), and non-shared environmental effects (E; those that lead siblings to not correlate). Additive genetic effects are indicated when the MZ intra-twin correlation is greater than the DZ twin correlation. Shared environmental effects are indicated when the DZ correlation is more than half the MZ correlation. C can encompass shared familial factors such as nutrition in the home, parenting styles, and socioeconomic status. Non-shared environmental effects are indicated when the MZ correlation is less than 1. E includes influences that are unique to one twin like differential parental treatment, as well as measurement error.

These models can be extended to bivariate analyses, which decompose the covariance of two traits into the 3 latent sources of contribution described above (rA, rC, rE). These etiological correlations reveal whether there are shared sources of genetic or environmental contributions between two phenotypes (Rijsdijk & Sham, 2002).

Constrained correlation models were conducted to estimate twin correlations. The within twin pair, cross-trait correlations were constrained to be equal for both twins in both zygosity groups. Cross-twin, cross-trait correlations were constrained to be equal for both twins but allowed to differ across zygosity group to ascertain differences in magnitude between MZ and DZ correlations. If the MZ/DZ twin correlations were significant they were selected as suitable to perform twin models.

For the univariate twin models, probit models with delta parameterization were used, which fixed the scaling factor to 1 and estimated all the thresholds (Prescott, 2004). The bivariate twin analyses used theta parameterization in which the thresholds were fixed to their actual value (Prescott, 2004). Correlated factors models were derived from Cholesky decompositions to assess etiological correlations between pain and psychopathology (Loehlin, 1996). No genetic correlations were estimated between age 12 pain and psychopathology because the A variance component was estimated at 0.

Results

Descriptive statistics for the pain and CBCL measures for each timepoint can be found in Table 1 (descriptive statistics on the untransformed variables can be found in Supplementary Table 1). Across the two timepoints, the prevalence was 65–67% for no pain, 18–19% for mild/moderate pain, and 15–16% for moderate/severe pain. Overall prevalence of reporting any pain was 35% at age 12 with 19.94% reporting musculoskeletal pain, 7.56% reporting abdominal pain, 7% reporting head pain, and 3.37% reporting low back pain. The sample size diminished due to an incomplete data release at age 13, but the pain prevalence remained similar at 33.38%. At this age, 19.66% reported musculoskeletal pain, 7.01% reported abdominal pain, 5.63% reported head pain, and 4.59% reported low back pain. The cross-time polychoric correlation for pain was r = 0.45 (p < 0.05). The correlations in both the full and twin sample are presented in Supplementary Table 2. At both ages 12 and 13, pain intensity was significantly associated with every CBCL scale (r = 0.08–0.16, p < 0.05) except for the withdrawn scale.

Table I.

Descriptive statistics for pain & psychopathology at ages 12 and 13

Age	Measure	n	Ma	SDa	Rangea	Skewb	Kurtosisb	Frequency (%) bin 0c	Frequency (%) bin 1c	Frequency (%) bin 2c
Age 12	Pain	10,414	2.13	3.23	10	-	-	6,804 (65%)	1,935 (19%)	1,675 (16%)
	Anxious/depressive	8,085	2.32	2.96	22	0.433	−0.557	277 (34%)	3,908 (48%)	1,407 (17%)
	Withdrawn/depressive	8,085	1.22	1.91	16	0.815	−0.349	4,246 (53%)	3,240 (40%)	599 (7%)
	Somatic	8,085	1.40	1.92	16	0.528	−0.692	3,575 (44%)	3,918 (49%)	592 (7%)
	SOC	8,085	1.31	2.07	17	0.820	−0.257	4,126 (51%)	3,320 (41%)	639 (8%)
	THO	8,085	1.44	2.07	22	0.633	−0.418	3,633 (45%)	3,816 (47%)	636 (8%)
	ATT	8,085	2.70	3.31	19	0.350	−0.965	2,881 (36%)	3,270 (40%)	1,934 (24%)
	Rule breaking	8,085	1.06	1.84	23	1.004	0.181	4,610 (57%)	3,032 (38%)	443 (6%)
	Aggressive	8,085	2.87	4.02	33	0.620	−0.263	2,972 (37%)	3,303 (41%)	1,810 (22%)
	INT	8,085	4.94	5.62	50	0.362	−0.129	1,510 (19%)	3,424 (42%)	3,151 (39%)
	EXT	8,085	3.93	5.52	50	0.672	0.015	2,512 (31%)	3,145 (39%)	2,428 (30%)
Age 13	Pain	6,251	2.03	3.15	10	-	-	4,164 (67%)	1,146 (18%)	941 (15%)
	Anxious/depressive	6,133	2.35	3.02	23	0.453	−0.445	2,035 (33%)	3,041 (50%)	1,407 (17%)
	Withdrawn/depressive	6,133	1.41	2.07	15	0.679	−0.595	2,977 (49%)	2,609 (43%)	547 (9%)
	Somatic	6,133	1.38	1.91	20	0.554	−0.678	2,782 (45%)	2,907 (47%)	444 (7%)
	SOC	6,133	1.21	1.96	18	0.880	−0.071	3,219 (53%)	2,486 (41%)	428 (7%)
	THO	6,133	1.38	2.00	20	0.646	−0.411	2,807 (46%)	2,890 (47%)	426 (7%)
	ATT	6,133	2.76	3.30	19	0.300	−1.022	2,147 (35%)	2,491 (41%)	1,495 (24%)
	Rule breaking	6,133	1.06	1.85	22	1.026	0.204	3,546 (58%)	2,244 (37%)	343 (6%)
	Aggressive	6,133	2.87	3.88	34	0.558	−0.366	2,121 (35%)	2,655 (43%)	1,357 (22%)
	INT	6,133	5.14	5.79	44	0.353	−0.181	1,097 (18%)	2,531 (41%)	2,505 (41%)
	EXT	6,133	3.93	5.38	48	0.626	−0.088	1,844 (30%)	2,488 (41%)	1,801 (29%)

Note. Descriptive statistics were computed on both the untransformed and transformed variables. SOC = social problems scale; THO = thought problems scale; ATT = attention problems scale; INT = internalizing score, which includes the anxious/depressive, withdrawn/depressive, and somatic scales; EXT = externalizing score, which includes the aggressive and rule breaking scales.

^aMean (M), standard deviation (SD), and range for the untransformed pain and CBCL measures.

^bSkew and kurtosis for the square root transformed psychopathology measures, which were used in the co-twin control models.

^cFrequencies for the binned pain and CBCL measures, which were used for the cross-lag panel and twin models.

Question 1: Is the relationship between pain and psychopathology bidirectional?

We estimated a series of cross-lag panel models for pain with each CBCL scale. The auto-regressive paths as well as the within-time, cross-variable correlations were significant in all the models (see Figure 2). In the full sample, there were significant but modest bi-directional effects of internalizing predicting later pain (b = 0.07, standard error [SE] = 0.02, p < 0.001) and pain predicting later internalizing (b = 0.04, SE = 0.01, p = 0.001). Earlier externalizing predicted later pain (b = 0.07, SE = 0.02, p < 0.001), but earlier pain did not significantly predict later externalizing (b = 0.02, SE = 0.01, p = 0.130). There were significant bidirectional associations between pain and thought problems, social problems, and attention problems. Early thought problems predicted later pain (b = 0.06, SE = 0.02, p < 0.001), and early pain predicted later thought problems (b = 0.05, SE = 0.01, p < 0.001). Social problems predicted later pain (b = 0.05, SE = 0.02, p = 0.001) and pain also predicted later social problems (b = 0.06, SE = 0.01, p < 0.001). Attention problems predicted later pain (b = 0.05, SE = 0.02, p = 0.004), and pain also predicted later attention problems (b = 0.02, SE = 0.01, p = 0.039). In the twin sub-sample, the effect sizes of the cross-lagged effects were similar; however, none reached significance.

Figure 1.

Cross-lag panel models of pain with each scale of the Child Behavior Checklist (CBCL): Internalizing (INT; panel a); Externalizing (EXT; panel b); Thought problems (THO; panel c); Social problems (SOC; panel d); and Attention problems (ATT: panel e). The standardized estimates for both the full sample and sub-sample twin models are presented (full/twin). Sex was also included as a predictor for all variables (paths not shown). Bolded font indicates p < .05.

Question 2: Controlling for familial confounding, do the pain-psychopathology associations persist?

The within-pair discordances for all variables are presented in Supplementary Table 3. The results of the phenotypic regression and co-twin control models are presented in Table 2 and Table 3. The phenotypic models regressed the predictors, sex, and their interaction on the outcome variable. The phenotypic regression coefficient represents the overall effect averaging across the within and between family estimates. We present the phenotypic estimates since the mixed models used a different estimator and link function than the correlations presented earlier. The phenotypic information is in the same metric as the co-twin models, enabling easier comparison. For the cross-sectional models, the results with the predictors and outcome variables reversed can be found in Supplementary Table 4 and 5.

Table II.

Results of the multi-level co-twin control models of age 12 psychopathology as predictors and pain as an outcome

DV = Pain	IV = Psych	INT 12	EXT 12	THO 12	SOC 12	ATT 12
Age 12 pain
Model 1 Phenotypic	Phenotypic Psych	0.19 [0.03] OR = 1.21	0.15 [0.03] OR = 1.16	0.14 [0.03] OR = 1.15	0.13 [0.04] OR = 1.14	0.14 [0.03] OR = 1.15
Model 2 Co-twin Control	Between-family Psych	0.37 [0.08] OR = 1.45	0.27 [0.07] OR = 1.31	0.47 [0.12] OR = 1.60	0.46 [0.12] OR = 1.58	0.35 [0.09] OR = 1.42
	Within-family Psych	0.14 [0.11] OR = 1.15	0.32 [0.11] OR = 1.38	0.05 [0.16] OR = 1.05	0.16 [0.15] OR = 1.17	0.14 [0.12] OR = 1.19
	Zygosity main effect	−0.17 [0.17] OR = 0.84	−0.19 [0.17] OR = 0.83	−0.19 [0.17] OR = 0.83	−0.24 [0.17] OR = 0.79	−0.18 [0.17] OR = 0.84
	Zygosity interaction	−0.01 [0.19] OR = 0.99	−0.19 [0.19] OR = 0.83	−0.39 [0.23] OR = 0.68	−0.31 [0.23] OR = 0.73	−0.31 [0.19] OR = 0.73
Age 13 pain
Model 1 Phenotypic	Phenotypic	0.18 [0.04] OR = 1.20	0.12 [0.04] OR = 1.13	0.09 [0.04] OR = 1.09	0.07 [0.04] OR = 1.07	0.08 [0.04] OR = 1.08
Model 2 Co-twin Control	Between-family Psych	0.39 [0.08] OR = 1.48	0.27 [0.07] OR = 1.31	0.34 [0.12] OR = 1.40	-	0.23 [0.09] OR = 1.26
	Within-family Psych	0.06 [0.14] OR = 1.06	−0.18 [0.14] OR = 0.84	−0.07 [0.17] OR = 0.93	-	−0.07 [0.15] OR = 0.93
	Zygosity main effect	0.12 [0.16] OR = 1.13	0.11 [0.17] OR = 1.12	0.09 [0.17] OR = 1.09	-	0.09 [0.17] OR = 1.09
	Zygosity interaction	−0.41 [0.24] OR = 0.66	0.00 [0.23] OR = 1.00	−0.40 [0.25] OR = 0.67	-	−0.17 [0.25] OR = 0.84

Note. The phenotypic models regressed sex, psychopathology, and the interaction between sex and psychopathology on pain; see Supplementary Table 6 for sex and interaction results. See Supplementary Table 3 for within-pair trait discordance. Switching the predictor and the outcome for the age 12 cross-sectional models yields similar results (see Supplementary Table 4). Significant regression coefficients were candidates for the co-twin control analyses. The co-twin control models used a logit link function; the log odds and odds ratio (OR) are presented in the table. Sex was included as a covariate (estimates not shown). Results of the between row are the differences between families and are suggestive of familial effects. Results on the within row are differences within family pairs and are suggestive of direct effects. Dashes indicate non-estimated betas. Zygosity was contrast coded: MZ = −0.05; DZ = 0.05. The zygosity interaction coefficient is the difference between MZ and DZ twins. INT = internalizing score; EXT = externalizing score; SOC = social problems scale; THO = thought problems scale; ATT = attention problems scale; 12 = age 12 years; 13= age 13 years; psych = psychopathology; DV = dependent variable; IV = independent variable. Bolded font indicates p < 0.05.

Table III.

Results of the muti-level co-twin control models of pain as a predictor and age 13 psychopathology as outcomes

IV = Pain	Age 12 Pain					Age 13 Pain
	Model 1	Model 2				Model 1	Model 2
DV = Psych	Phenotypic	Between-family Pain	Within-family Pain	Zygosity Main effect	Zygosity Interaction	Phenotypic	Between-family Pain	Within-family Pain	Zygosity Main effect	Zygosity Interaction
INT 13	0.11 [0.04] OR = 1.12	0.38 [0.13] OR = 1.46	0.15 [0.11] OR = 1.16	−0.28 [0.24] OR = 0.76	−0.20 [0.18] OR = 0.82	0.12 [0.03] OR = 1.13	0.36 [0.15] OR = 1.43	0.14 [0.10] OR = 1.15	−0.31 [0.24] OR = 0.73	−0.24 [0.17] OR = 0.79
EXT 13	0.07 [0.04] OR = 1.07	-	-	-	-	0.09 [0.03] OR = 1.09	0.48 [0.14] OR = 1.62	−0.15 [0.08] OR = 0.86	0.11 [0.24] OR = 1.12	0.20 [0.15] OR = 1.22
THO 13	0.07 [0.04] OR = 1.07	0.19 [0.12] OR = 1.21	0.09 [0.11] OR = 1.09	−0.03 [0.23] OR = 0.97	−0.21 [0.19] OR = 0.81	0.07 [0.04] OR = 1.07	-	-	-	-
SOC 13	0.09 [0.04] OR = 1.09	0.31 [0.13] OR = 1.36	0.06 [0.10] OR = 1.06	0.03 [0.24] OR = 1.03	0.28 [0.18] OR = 1.32	0.04 [0.04] OR = 1.04	-	-	-	-
ATT 13	0.06 [0.03] OR = 1.06	-	-	-	-	0.03 [0.03] OR = 1.03	-	-	-	-

Note. The phenotypic models regressed sex, psychopathology, and their interaction between sex and psychopathology on pain; see Supplementary Table 7 for sex and interaction results. See Supplementary Table 3 for within-pair trait discordance. Switching the predictor and the outcome for the age 13 cross-sectional models yields similar results, besides the significance of the thought-pain association and within-family externalizing estimate (see Supplementary Table 5). Upon investigation, the externalizing within-family estimate is only significant centered on MZ twins, suggesting the result should be approached with caution. Significant regression coefficients were candidates for the co-twin control analyses. The co-twin control models used a logit link function; the log odds and odds ratio (OR) are presented in the table. Sex was included as a covariate (estimates not shown). Results of the between column are the differences between families and are suggestive of familial effects. Results on the within column are differences within family pairs and are suggestive of direct effects. Dashes indicate non-estimated betas. Zygosity was contrast coded: MZ = −0.05; DZ = 0.05. The interaction coefficient is the difference between MZ and DZ twins. INT = internalizing score; EXT = externalizing score; SOC = social problems scale; THO = thought problems scale; ATT = attention problems scale; 12 = age 12 years; 13= age 13 years; psych = psychopathology; DV = dependent variable; IV = independent variable. Bolded font indicates p < 0.05

Table 2 shows age 12 psychopathology predicting age 12 pain cross-sectionally and age 12 psychopathology predicting age 13 pain longitudinally. Cross-sectionally at age 12, the between-family associations of pain predicted by internalizing (b = 0.37 log odds, p < 0.001), externalizing (b = 0.27 log odds, p < 0.001), thought (b = 0.47 log odds, p < 0.001), social (b = 0.46 log odds, p < 0.001), and attention problems (b = 0.35 log odds, p < 0.001) were significant (see Table 2). The only significant within-family estimate was for externalizing (b = 0.32 log odds, p = 0.010), which was slightly larger than the between-family estimate. All other non-significant within-family estimates were of lower magnitude than the between-family estimates (see Table 2).

Longitudinally, the between-family estimates were significant for age 12 internalizing (b = 0.39 log odds, p < 0.001), externalizing (b = 0.27 log odds, p < 0.001), thought (b = 0.34 log odds, p < 0.001), and attention (b = 0.23 log odds, p < 0.001) predicting later pain at age 13. The within-family estimates were non-significant for all the associations and lower than the between-family estimates. These longitudinal associations appear to be due to familial confounding. There were no significant interactions of zygosity with the within-family estimates in any of the models.

Table 3 shows age 13 pain predicting age 13 psychopathology cross-sectionally and age 12 pain predicting later psychopathology. Cross-sectionally, the between-family estimates for pain at age 13 predicting age 13 internalizing was 0.36 log odds (p = 0.013) and 0.48 log odds (p < 0.001) for age 13 externalizing. The within family estimates for both these associations were not significant and were less than the between-family estimates. However, the polarity of the externalizing within-family estimate is negative, suggesting that the significant, positive within-family estimate found at age 12 should be interpreted with caution.

Longitudinally, the associations between age 12 pain and age 13 internalizing (b = 0.38 log odds, p = 0.004) and age 13 social problems (b = 0.31 log odds, p = 0.015) had significant between-family effects. The within-family estimates were not significant and smaller in magnitude. There were no significant zygosity interactions in any of the models conducted.

Question 3: What is the etiology of pain at ages 12 and 13?

The univariate variance components for the CBCL scales and pain can be found in Supplementary Table 8. At age 12, the variance of pain risk is influenced 30% by shared environment and 70% by non-shared environment. At age 13, pain risk variance is attributable to 12% A, 15% C, and 73% E, but neither of the familial components were significant at this age. However, dropping both A and C led to a significant deterioration of fit (χ²(2) = 12.529, p = 0.002, RMSEA = 0.041, CFI = 0.531), suggesting that we are unable to discriminate the influences of A and C at age 13.

The CBCL scales all exhibited a significant A component (18–62%) except for somatic problems at age 13, which had a nonsignificant A estimate of 14%. The anxious-depressive and withdrawn-depressive scales had non-significant C components at both time points (8–13%). Social problems had a nonsignificant C component at age 12 (7%) which became significant at age 13 (15%). Attention problems had C estimated at 0% for both timepoints. The rest of the CBCL scales exhibited a significant C component at ages 12 and 13 (16–57%). The non-shared environmental contribution to the CBCL scales was moderate (22–45%).

Question 4: Does pain share genetic and environmental correlations with psychopathology?

The twin correlations can be found in Table 4 and the rA, rC, and rE correlation estimates can be found in Table 5 (the model fit statistics are in Supplementary Table 8). The model fit statistics of the bivariate models can be found in Supplementary Table 9. At age 12, the MZ cross-twin cross-trait correlations were significant for pain with internalizing, externalizing, thought, attention, and social problems. The correlations were suggestive of shared environmental effects as they were similar across zygosity. Pain shared significant C correlations with internalizing (rC = 0.61, Δχ²(1) = 20.530, p < 0.001), externalizing (rC = 0.39, Δχ²(1) = 6.006, p = 0.014), thought problems (rC = 0.54, Δχ²(1) = 16.906, p < 0.001), and social problems (rC = 0.98, Δχ²(1) = 10.833, p < 0.001). Additionally, pain shared significant E correlations with externalizing (rE = 0.20, Δχ²(1) = 5.504, p = 0.019) and attention problems (rE = 0.31, Δχ²(1) = 16.543, p < 0.001).

Table IV.

Phenotypic, within-trait cross-twin, and cross-twin cross-trait correlations at ages 12 and 13

		Phenotype 2
Phenotype 1		Pain age 12		Pain age 13
Correlation Type	Cross-twin Within-trait	Within-twin Phenotypic	Cross-twin Cross-trait	Within-twin Phenotypic	Cross-twin Cross-trait
Pain age 12	0.28/0.31	-	-	0.46	0.29/0.21
Pain age 13	0.27/0.21	0.46	0.29/0.21	-	-
INT age 12	0.70/0.49	0.21	0.16/0.19	0.17	0.17/0.24
EXT age 12	0.76/0.50	0.19	0.16/0.06	0.13	0.13/0.19
SOC age 12	0.57/0.32	0.16	0.18/0.12	0.08	0.08/0.16
THO age 12	0.55/0.46	0.14	0.13/0.22	0.07	0.07/0.21
ATT age 12	0.64/0.24	0.16	0.12/0.16	0.07	0.07/0.16
INT age 13	0.67/0.56	0.17	0.05/0.15	0.10	0.02/0.13
EXT age 13	0.70/0.50	0.11	0.07/0.09	0.11	0.20/0.19
SOC age 13	0.69/0.42	0.14	0.15/0.07	0.02	0.03/0.13
THO age 13	0.61/0.44	0.08	−0.02/0.12	0.08	0.06/0.10
ATT age 13	0.63/0.22	0.09	0.08/0.08	0.02	0.01/0.13

Note. The MZ/DZ cross-twin within-trait, within-twin phenotypic, and cross-twin cross-trait correlations are presented. These estimates came from models with several constraints: Within-twin (i.e., phenotypic) correlations were constrained to be equal for both twins and across both zygosity groups. Cross-twin within-trait correlations were allowed to vary by zygosity group, and cross-twin cross-trait correlations were constrained to be equal within each zygosity group. Sex effects were regressed out of all variables. INT = internalizing score; EXT = externalizing score; SOC = social problems scale; THO = thought problems scale; ATT = attention problems scale. Bolded font indicates p < .05.

Table V.

Etiological correlations of bivariate pain and psychopathology models

	Pain age 12 A = 0% C = 30% E = 70%			Pain age 13 A = 12% C = 15% E = 73%
	rA	rC	rE	rA	rC	rE
Pain age 12	-	-	-	-	0.78	0.29
A = 0% C = 30% E = 70%					[0.57, 0.89]	[0.17, 0.42]
INT age 12	-	0.61	0.07	−0.76	1.00	−0.02
A = 41% C = 29% E = 30%		[0.34, 1.00]	[−0.09, 0.24]	[−1.00, 1.00]	[0.42, 1.00]	[−0.31, 0.17]
EXT age 12	-	0.39	0.20	−0.73	1.00	0.01
A = 50% C = 25% E = 25%		[0.09, 1.00]	[0.04, 0.37]	[−1.00, 1.00]	[0.21, 1.00]	[−0.28, 0.20]
THO Age 12	-	0.54	−0.06	-	-	-
A = 18% C = 37% E = 45%		[0.27, 1.00]	[−0.22, 0.10]
SOC Age 12	-	0.98	0.03	-	-	-
A = 50% C = 7%, E = 43%		[0.26, 1.00]	[−0.11, 0.20]
ATT Age 12	-	-	0.31	-	-	-
A = 62% C = 0%, E = 38%			[0.15, 0.50]
EXT age 13	-	-	-	0.09	0.87	−0.21
A = 40% C = 30% E = 30%				[−1.00, 1.00]	[−0.27, 1.00]	[−0.46, 0.00]
SOC age 13	-	0.50	0.06	-	-	-
A = 53% C = 15% E = 32%		[0.03, 1.00]	[−0.12, 0.27]

Note. rA, rC, and rE estimates derived from bivariate Cholesky decompositions are presented. Bivariate models were conducted based on significant twin correlations. Dashes indicate non-estimated correlations; correlations were not estimated when one of the involved variance components was zero. Asymmetric, boot-strapped (1,000 samples) confidence intervals are displayed. INT = internalizing score; EXT = externalizing score; SOC = social problems scale; THO = thought problems scale; ATT = attention problems scale. Bolded font indicates p < 0.05, determined with chi-square difference tests.

At age 13, there was a reduction in sample size, which likely limited power, so these results should be cautiously approached. The pain-externalizing cross-twin cross-trait correlations were significant for both MZ and DZ twins. The correlations were similar in magnitude (rMZ = 0.20, rDZ = 0.19), which is suggestive of overlapping shared environmental influences. None of the etiological correlations reached significance.

Based on the significant longitudinal twin correlations, we estimated twin models for associations of age 13 pain with age 12 internalizing and externalizing. There was a significant shared environmental correlation of 1.00 with internalizing (Δχ²(1) = 5.056, p = 0.026). The only significant twin correlations between age 12 pain with age 13 psychopathology was with social problems. Age 12 pain shared a significant C correlation with social problems a year later (rC = 0.50, Δχ²(1) = 4.957, p = 0.026).

Discussion

We assessed the longitudinal directionality of associations between pain and psychopathology in adolescence and tested survival of these associations when controlling for familial confounding. Furthermore, we elucidated the etiology of pain and its etiological relationships with psychopathology at ages 12 and 13. The results suggest that pain is bidirectionally associated with most psychopathology dimensions examined, and these relationships appear to be familial in origin. Pain variance was explained primarily by environmental effects and showed shared environmental correlations with most psychopathology dimensions.

Most psychopathology dimensions share bidirectional associations with pain

The observed adolescent pain prevalence of 33–35% was as expected based on prior studies at this age range (King et al., 2011). Pain correlated with all the CBCL scales except for the withdrawn-depressive scale, which is part of the internalizing sum score. The two other internalizing scales, the somatic and anxious scales, correlated with pain, suggesting that the internalizing sum was mostly driven by these constructs. Not surprisingly considering that the somatic scale includes some items related to pain, the somatic-pain correlation was largest in magnitude; however, the significant correlations with the anxious/depressive scale suggest a more general pain-internalizing association. The pain and psychopathology correlations were generally lower than expected, but the lack of correlation between pain and withdrawn-depressive symptoms was most surprising. Most of the literature on pain-psychopathology has been conducted on the relationship between pain, anxiety, and depression, and these associations are well-replicated (Khan et al., 2020; King et al., 2011). The low or absent correlations in this study may reflect the fact that the psychopathology scales were based on parent reports, whereas the pain was self-reported. However, this also is a strength of this manuscript, as the associations we did detect were based on reports from different informants.

Pain shared significant bidirectional effects with internalizing, which corroborates past research suggesting that internalizing is both a predictor and an outcome of pain in adolescence (Kolaitis et al., 2021; Stapp et al., 2022; Youssef et al., 2008). Our results contrast with one adult model, which found that cross-lagged depression/anxiety directionally predicted higher levels of pain and pain-related disability across four time-points in adults with chronic pain (Lerman et al., 2015).

We found that externalizing had a significant directional effect on increasing pain risk one year later. This result supports the previous longitudinal twin study that found externalizing phenotypically increased headache risk (Virtanen et al., 2004) and extends the finding to general pain. The current study was the first known to look at earlier pain predicting later externalizing; however, the association was not significant. Thought, social, and attention problems had significant bidirectional associations with pain in adolescence, which are novel findings. However, it is important to note that all these effect sizes were modest and did not explain substantial portions of variance. The cross-lag panel models demonstrated that most pain and psychopathology associations consist of cross-sectional and longitudinal associations, but their nature can be further elucidated with the models that incorporated familial information.

Familial effects account for most pain-psychopathology relationships

The co-twin control results suggest that the relationships between pain and internalizing, social, and thought problems at this age may be entirely familial. When controlling for familial effects, at age 12 externalizing significantly predicted concurrent pain risk.

Since the pain-internalizing relationship is due to familial effects at ages 12, 13, and longitudinally, internalizing is likely not causal for pain in adolescence, but instead, there may be common predisposing factors. Both twin (Khan et al., 2020) and molecular genetic findings (Johnston et al., 2019) have found underlying genetic and environmental risk factors contributing to chronic pain and depression/anxiety, supporting this finding.

However, in previous co-twin control studies, the extent to which familial confounding accounts for the covariation of pain and depression/anxiety seems to vary greatly depending on the population, age, and pain type studied. A study of discordant adult twins from the Netherlands Twin registry suggest migraine and anxiety/depression have bidirectional, causal effects on each other (Ligthart et al., 2010). Furthermore, a study of male veterans with chronic back pain showed that the depression/PTSD-pain relationship was not due to familial effects (Suri et al., 2017). However, familial confounds entirely accounted for the association between chronic wide-spread pain and anxiety in women (Burri et al., 2015). In the current sample, internalizing was driven mainly by somatic distress and anxiety, making it a different construct than pure depression or anxiety, which may explain the difference in findings from previous co-twin controls.

When controlling for familial confounding, the pain-externalizing relationship had significant between- and within-family effects at age 12. The similarity of the between-family and within-family coefficients is consistent with the hypothesis that there could be a partially causal mechanism (Carlin et al., 2005). However, this pattern disappeared longitudinally, with only the between-family estimate being significant. Our externalizing co-twin control model does not corroborate a previous longitudinal co-twin control study that found that externalizing in adolescents continued to directly predict later headaches (Virtanen et al., 2004).

There is not much literature on the relationships between the thought and social problems and pain. To our knowledge, no study has looked at these two CBCL measures with pain in a co-twin design. However, one discordant MZ-twin study found that obsessive compulsive disorder was not an independent risk factor for chronic widespread pain in women (Burri et al., 2015), which is similar to our findings.

The significance of the between-family association of pain and attention problems implicates shared familial effects. This result corroborates the findings of one previous study that found the association of attention problems with neurological problems, including migraine headaches, was due to familial predisposition (Pan & Bölte, 2020).

Shared and non-shared environmental effects contribute to pain variance

Pain was entirely attributable to shared and non-shared environmental effects at age 12. At age 13, the shared environment estimate was halved, and genetic influences began to emerge; however, these estimates were not significant. Across time, pain shares a significant shared environmental and non-shared environmental correlation.

These results both support and conflict with past adolescent pain etiology research. The longitudinal Quebec Newborn Twin Study found that general pain across ages 12–14 was due to genetic (37%) and non-shared environmental (62%) effects but no significant shared environmental influences (C = 1%) (Battaglia et al., 2020). However, our results are consistent with a Finnish twin study (FinnTwin12) that found low-back pain in 11 year old twins was due to shared and non-shared environmental influences with no genetic effects (El-Metwally et al., 2008).

Pain has shared environmental correlations with most psychopathology

The co-twin control findings were consistent with the twin models, with shared environmental correlations of pain with internalizing, externalizing, social, and thought problems and non-shared environment correlations of pain with externalizing and attention problems.

The shared environmental correlation between internalizing and pain both cross-sectionally and longitudinally differs from most adult twin studies. Past research in adults has found that the covariation between anxiety and depression is attributable to a mix of mostly shared genetic and some non-shared environmental variance across multiple different pain syndromes (Khan et al., 2020). Pain may have a different etiological relationship with anxious behavior in young adolescence compared to adulthood. In 11–18 year-olds, a common factor encompassing pain and anxiety/depression had significant A (27%), C (38%), and E (35%) influences, giving strength to our finding of a C correlation (Scaini et al., 2022).

We detected significant shared environmental and non-shared environmental correlations between externalizing and pain at age 12. The co-twin control findings are consistent with the twin models, as both the between-family and within-family components were significant. Taken together, externalizing seems to have shared risk factors with pain, as well as partially increasing pain risk at age 12. Thought and social problems also displayed significant shared environmental correlations with pain, which are novel findings.

Attention problems did not show a significant shared environmental correlation with pain, even though the co-twin control between-family estimate was significant. There may be overlapping genetic and shared environmental effects that we did not discriminate with these models. Attention problems had a significant non-shared environmental correlation with pain, which could suggest that there is a unique risk factor underlying both phenotypes, there is an exacerbating relationship, or that there is shared measurement error. These findings differ from a prior attention problems-migraine co-twin control result (Pan & Bölte, 2020), likely because we included multiple types of pain.

Implications

Taken together, the findings suggest that internalizing, thought, and social problems co-occur with adolescent pain entirely due to unmeasured familial factors. Externalizing possibly has a direct influence on pain that is not accounted for by shared twin-pair effects. There are multiple environmental factors that have been implicated in the epidemiology of both psychopathology and chronic pain. Some examples of these environmental factors are stress in childhood, poor nutrition, immune dysregulation and a negative family environment (King et al., 2011).

Heightened stress in childhood due to stressful life events, school anxiety, maltreatment, or family conflict can induce multiple types of psychopathology and lead to maladaptive cognitive appraisals of pain (Asmundson et al., 2012; King et al., 2011). For example, anticipation of math for people with math anxiety has been shown to activate the pain network in the brain (Lyons & Beilock, 2012). High-stress exposure in childhood has also been shown to increase inflammation throughout the lifespan (Chiang et al., 2022), which could influence both chronic pain and psychopathology incidence (Danese & Baldwin, 2017; Marchand et al., 2005). Additionally, nutritional deficiencies during a developmentally sensitive time have been implicated in pain and psychopathology (Liu et al., 2015; Mills et al., 2019).

One possible mechanism is that these shared environmental risk factors drive similar neurobiological anomalies in both psychopathology and pain. Specifically in the ABCD sample, aberrations in brain structure are associated with family conflict and low parental monitoring, and the associations are partially mediated by the children’s CBCL total problem index (Gong et al., 2021). Stressful life events, inflammation, and nutritional deprivation in adolescence have been shown to have enduring impacts on neurobiology (Liu et al., 2015; Marchand et al., 2005; Miguel et al., 2019). Gray matter abnormalities common to psychopathology, such as reduced gray matter in the insular cortex, anterior cingulate cortex, orbitofrontal cortex, middle frontal gyrus, postcentral gyrus, and thalamus, and increased gray matter in striatal regions (Bos et al., 2018; Galambos et al., 2019; Glahn et al., 2008), are often common to chronic pain patients and pain processing (Smallwood et al., 2013). A high-risk environment in childhood could incite these pervasive changes in neurobiology during a developmentally sensitive time.

The potentially direct influence of externalizing on pain could be due to heightened engagement in risky behaviors such as violence (Ranney et al., 2018), which could lead to greater injury. However, the lack of persistence of this estimate longitudinally calls for replication before firm conclusions can be drawn. While we did not find patterns consistent with causal effects between attention problems and pain in the co-twin control model, the non-shared environmental correlation in the twin model could be indicative of exacerbating effects. Experiencing pain demands attention and could cause issues with sustained attention (Weiss et al., 2018). Conversely, deficiencies in brain networks associated with attentional problems have been found to predict the transition of a subacute injury to chronic pain (Attal et al., 2014).

Limitations & Future Directions

There were several limitations in this study. The ABCD pain questionnaire does not include a pain duration measure. While the definition of clinically chronic pain varies, generally it is defined as pain that persists for 3 months (King et al., 2011). We were not able to assess whether pain was clinically chronic. Future studies using the ABCD sample could infer chronicity by using the longitudinal structure to ascertain chronic pain which has lasted a year; however, this method would miss participants whose pain duration is more than 3 months, but less than a year. Correlations between pain and psychopathology may have been lower than expected because the CBCL was parent-reported; future studies may benefit from using multiple raters for psychopathologies or by also including a self-reported psychopathology measure. Finally, the lower available sample size at age 13 limited our power for cross-sectional and longitudinal associations with age 13 measures.

CLPMs have been criticized for their inability to distinguish between within-person and between-person effects (Hamaker et al., 2015). CLPMs which include random intercepts can overcome this problem, however they require more than two timepoints. With future ABCD data releases, it would be beneficial to conduct the random-intercepts CLPMs.

Adolescent psychopathology can be modeled within multiple factor structures, and this is an active area of research (Clark et al., 2021). Since many of the psychopathology measures exhibited similar relationships with pain, there may be shared common variance that explains the consistent relationships. A future direction could be to look at the psychopathology and pain associations within different factor structures (i.e., p factor, correlated factors) to see what best explains the pain-psychopathology relationships. Pain could be a shared feature for those at risk for general psychopathology.

There have been multiple phenotypic, genetic, and neurobiological studies that suggest pain conditions and their relationships to different psychopathology dimensions may vary by sex. One study found that externalizing and attention problems predicted pain in boys and internalizing predicted pain in girls (Egger et al., 1999). We looked at interactions between sex and psychopathology (see Supplementary Table 6 and 7), and while we did not detect significant effects there appeared to be similar trends. Exploring potential sex differences in the biology of pain may help inform personalized treatment.

We looked at general pain as the main construct of interest; however, there is an abundance of literature which has found differential mechanisms underlying differing pain conditions in adolescents (King et al., 2011). It would be interesting to assess how psychopathology associates with separate pain types.

The shared environmental correlations between pain and psychopathologies could be due to a variety of shared environmental risk factors, such as stressful life events or nutrition. Future research could explore whether these constructs mediate the association between pain and psychopathology. Furthermore, brain measures can be used as an endophenotype to mediate the pain-psychopathology association.

Conclusion

Phenotypically, most psychopathology dimensions share bidirectional associations with pain, besides externalizing. Most of the psychopathology-pain co-occurrences appear to be entirely due to familial predisposition. At age 12, pain risk appears to be due to environmental effects, but some genetic effects emerge a year later. The shared environmental correlations between pain and all the psychopathology measures, minus attention, could be due to a variety of risk factors such as stressful life events or nutritional deficiencies. Possible future research could explore whether these common risk factors and brain measures mediate the association between pain and psychopathology.

Availability of data and material

The datasets generated and analyzed were accessed from the ABCD Data repository. Researchers who have been approved by the National Institute of Mental Health Data Archive (NDA) Data Use Certification may access the ABCD study data (https://nda-nih-gov.colorado.idm.oclc.org/abcd/). The ABCD data used in this report came from the ABCD release 4.0 (Study-specific NDA: 10.15154/1528696). ABCD data is publicly available.