Assessing the association between child maltreatment and chronic pain: a cross-sectional co-twin control study

Abstract

Although associations exist between child maltreatment (CM) and multiple chronic painful conditions, it remains unclear to what extent associations might be attributable to unmeasured confounding. We leveraged the co-twin control approach, which rules out genetic and shared environmental confounding by design, with cross-sectional, national Swedish data from the Study of Twin Adults: Genes and Environment (N = 25,418; M_age = 33.2; SD_age = 7.7; 55.7% female) to assess associations between the number of self-reported experiences of CM (ie, experiencing emotional or physical abuse/neglect, sexual abuse, or witnessing family violence before age 18 years) and self-reported endorsement of criteria for chronic widespread pain (CWP), lower back pain (LBP), and irritable bowel syndrome (IBS). In negative binomial generalized estimating equation models adjusting for age and sex at birth, a one-unit increase in CM counts was associated with 37%, 18%, and 34% more endorsement of CWP, LBP, and IBS criteria, respectively (CWP-adjusted incidence rate ratio [aIRR] = 1.37 [95% confidence interval: 1.32–1.42]; LBP aIRR = 1.18 [1.16–1.21]; IBS aIRR = 1.34 [1.29–1.39]). In comparisons of differentially exposed co-twins, associations attenuated only slightly for CWP (aIRR, 1.21 [1.12–1.31]), LBP (aIRR, 1.15 [1.09–1.21]), and IBS (aIRR, 1.24 [1.14–1.35]). Analyses restricted to monozygotic twins to rule out virtually all genetic confounding produced similar results (CWP aIRR, 1.20 [1.05–1.38]; LBP aIRR, 1.10 [1.01–1.21]; IBS aIRR, 1.14 [1.00–1.30]). Altogether, the results suggest that associations between CM and CWP, LBP, and IBS are not entirely attributable to genetic or shared environmental confounding.

1. Introduction

Research demonstrates that individuals experiencing chronic pain often report histories of maltreatment during childhood.^{7,9,12,17,19,30,54,59} Child maltreatment (CM) comprises experiencing physical abuse/neglect, emotional abuse/neglect, sexual abuse, or witnessing family violence before age 18 years.^6,7 Child maltreatment consistently demonstrates robust associations with chronic pain. For instance, CM is associated with roughly a 2-fold greater prevalence of lower back pain (LBP) and chronic widespread pain (CWP)-related conditions (eg, fibromyalgia), in addition to 15% greater prevalence of irritable bowel syndrome (IBS).⁷ Relatedly, experiencing a greater number of adverse childhood events, a construct overlapping with CM, is associated with 1.31-fold greater prevalence and 2.05-fold greater odds of LBP⁵⁹ and IBS,³⁸ respectively.

Still, the extent to which the association between CM and chronic pain is confounded by unmeasured genetic and early, shared environmental factors (ie, familial confounding) is unclear. For instance, familial-level environmental risk factors (eg, lower socioeconomic status) shared between victimization during childhood^40,45 and chronic pain⁴⁹ may confound associations. Furthermore, chronic pain is partially influenced by genetics,^23,36,52,56 and research has demonstrated the importance of accounting for genetic confounding when assessing associations involving self-reported trauma.^37,39,40 However, it is difficult to wholly account for all possible genetic and early environmental confounders through standard methods (eg, statistical covariates) due to various measurement considerations (eg, measurement availability, measurement error).⁵⁵ Hence, it is critical to apply designs that help account for unmeasured confounding when assessing the association between CM and chronic pain. Such knowledge would provide further evidence of the extent to which addressing CM may be beneficial for alleviating chronic pain.²⁸

The co-twin control design is a causal inference method that accounts for unmeasured familial confounding by design. In the context of the present study, the co-twin control design compares chronic pain among twins in a pair who differ from each other (ie, co-twins) in CM.^32,50 Because twins share 50% or 100% of their segregating genes and are typically reared in similar environments, the associations estimated from the co-twin control method rule out everything that makes twins similar, inherently controlling for genetic and shared environmental confounding.^32,50 To the best of our knowledge, no study has applied the co-twin control approach to estimate the association between CM and chronic pain.

This study assessed the extent to which self-reported CM is associated with endorsement of criteria for pain and related somatic conditions using a population-based, cross-sectional sample of Swedish twins. Given that pain conditions are clinically heterogeneous but also frequently comorbid with one another, we examined 3 conditions associated with CM in previous research.^7,38,59 Specifically, we assessed relationships between CM and CWP, LBP, and IBS while implicitly accounting for unmeasured familial confounding by design using the co-twin control method.

2. Methods

2.1. Data source and sample

We obtained data from the 2005 to 2006 Study of Twin Adults: Genes and Environment (STAGE), a population-based, cross-sectional survey of health conditions and related exposures conducted by the Swedish Twin Registry (STR). STAGE targeted virtually all twins born in Sweden between 1959 and 1985, with approximately 60% (N = 25,420) providing informed consent and responding to the survey either online or through telephone interviews.²⁷ STAGE was approved by the Stockholm Regional Ethics Committee (Reference Number 2010-322-31/1), and this secondary analysis was determined to be exempt human subjects research by the Indiana University Institutional Review Board. This study follows the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guidelines.⁵³

2.2. Measures

2.2.1. Exposure

2.2.1.1. Child maltreatment

Guided by previous STAGE studies,^5,11,15 we used questions from the Life Stressor Checklist-Revised (LSC-R)^33,57 included in the STAGE survey to operationalize CM. The LSC-R assesses the occurrence of a variety of stressful events within an individual’s lifetime. A strength of the LSC-R is its detailed measurement of the core CM subtypes that define CM, including questions assessing whether individuals have ever experienced emotional abuse/neglect, physical neglect, physical abuse, sexual abuse, or witnessed family violence before 18 years of age. The LSC-R measures each CM subtype with one question (yes/no), with the exception of sexual abuse being measured by 2 separate (yes/no) questions (ie, “Were you ever touched or made to touch someone else in a sexual way because you felt forced in some way or threatened by harm to yourself or someone else?” and “Did you ever have sex because you felt forced in some way or threatened by harm to yourself or someone else?”). We combined these 2 sexual abuse questions, such that if participants responded yes to either question, they were considered to have experienced sexual abuse. Additional follow-up questions probed the age at which any endorsed experiences occurred. In this study, we only counted CM as endorsing having experienced any of the 5 CM subtypes before 18 years of age. We created a 0 to 5 count variable representing the number of reported CM subtypes occurring before age 18 years, as research suggests that the accumulation of CM experiences may be of particular importance to consider when assessing the association between CM and pain-related outcomes.^9,19,59 However, it is worth noting that this definition of CM does not necessarily represent the total accumulation of CM experiences, as it only assesses the number of CM subtypes rather than the total number of times each CM subtype occurred.

2.2.2. Outcomes

The STAGE survey included an array of questionnaires assessing multiple pain-related syndromes.²⁷ We chose to include CWP, LBP, and IBS as outcomes in this study because our prior research suggests that they capture the comorbidity among and unique heterogeneity between the pain conditions assessed in STAGE.^8,46

In brief, our previous research analyzed self-reported criteria for 9 pain-related syndromes assessed in STAGE.⁸ We found that co-occurrence among the pain conditions could be explained by one general pain liability factor, plus 2 specific factors capturing unique covariance among pain-related somatic syndromes (IBS, headache, bladder pain, and chronic fatigue) and neck and shoulder pain, respectively.^8,46 Both the general and somatic syndromes factors predicted subsequent pain- and mental health–relevant clinical outcomes, supporting their validity.⁴⁶ In this study, to represent the breadth of conditions in the chronic overlapping pain condition framework, we included CWP, LBP, and IBS, as they were the conditions most strongly correlated with the general factor and the somatic syndrome-specific factor, respectively. Further detail regarding our bifactor model of chronic overlapping pain conditions is provided in our previous work.^8,46

2.2.2.1. Chronic widespread pain

We used 4 questions to create a 0 to 4 count variable representing the number of CWP criteria participants self-reported. Notably, given its length, the STAGE survey used a gating strategy, such that not all participants were asked to answer each survey item. To be administered the CWP items, participants had to endorse pain in at least 2 parts of the body within the 6 months before the STAGE survey using a body map. On endorsing this gate question, participants answered whether they suffered from general pain during the past 3 months (yes/no). This question gated the second question, which asked whether participants had continuous pain during all 3 months (yes/no). Those who endorsed continuous pain during all 3 months were then eligible to answer whether they suffered from pain in both the upper and the lower body (yes/no) and whether they suffered from pain in both the right and left sides of the body (yes/no). As in our prior work,^8,46 the 4 questions regarding general pain, continuous pain, pain in the left or right side of the body, and pain in the upper or lower region of the body defined the 0 to 4 count variable used to operationalize CWP. If participants reported pain in less than 2 parts of the body within the past 6 months, they did not receive any CWP questions and were assumed to meet 0 CWP criteria. This operationalization of CWP is based on the definition presented in the 1990 fibromyalgia criteria proposed by the American College of Rheumatology⁵⁸ and has been used in previous STR studies.^22–25

2.2.2.2. Lower back pain

Following previous STAGE research,^35,36 including our prior work,^8,46 we created a 0 to 3 count variable representing endorsement of LBP criteria, including pain presence (1), intensity (2), and debilitation (3). First, using the body map described above, participants self-reported whether they had experienced pain or discomfort in their lower back during the past 6 months. Second, those who endorsed pain or discomfort in their lower back during the past 6 months responded to 3 questions on 10-point Likert scales assessing their pain’s level of intensity at the moment, at its worst during the past 6 months, and on average during the past 6 months (1 = no pain; 10 = worst conceivable pain). Participants were considered to have endorsed the LBP intensity criterion if the average of the 3 pain intensity questions was greater than or equal to 3. Third, these participants also responded to 3 questions using 10-point Likert scales to assess how much their LBP affected their everyday activities (1 = not disturbed at all; 10 = impossible to continue with these activities) and social life and possibility to work (1 = no change; 10 = extremely changed). We determined that participants endorsed the LBP debilitation criterion if their average total among the 3 debilitation questions was greater than or equal to one. Participants who did not endorse lower back pain or discomfort using the body map were assumed to have met 0 LBP criteria.

2.2.2.3. Irritable bowel syndrome

We used questions assessing IBS criteria in STAGE to create a 0 to 6 IBS count variable based on the Rome criteria and previous STR research,^23,25,47 including our own.^8,46 This variable represents the number of IBS criteria self-reported by participants with abdominal discomfort lasting at least 7 days per month. Specifically, we included endorsement of lower abdominal pain (1), bloating (2), constipation (3), diarrhea (4), abdominal pain when feces become looser and defecation more frequent (5), and alleviation of abdominal pain after defecation (6) in the definition of IBS. It is important to note that all these items were gated by questions assessing whether participants had recurrent abdominal discomfort and whether the abdominal discomfort lasted at least 7 days per month. Moreover, participants only received questions regarding whether they experienced abdominal pain when feces become looser and defecation more frequent and alleviation of abdominal pain after defecation if they endorsed any abdominal pain in their upper, lower, or another part of the abdomen. Those who did not endorse recurrent abdominal discomfort that also lasted at least 7 days per month were assumed to meet 0 IBS criteria.

2.2.3. Covariates

All models controlled for age (linearly) and sex at birth given known differences in CM^18,44 and chronic pain³⁴ by age and sex. In addition, these are the only demographic factors available within STAGE that are not potentially influenced by CM and thus could not be possible mediators within the relationship between CM and chronic pain. We did not adjust for other factors assessed in STAGE (eg, psychopathology) beyond those implicitly controlled by the co-twin design, as we did not want to introduce potential bias due to unknowingly adjusting for potential mediators or colliders rather than confounders.⁴²

2.2.4. Statistical analysis

We used SAS software version 9.4⁴³ for all data management and analysis. We started by applying the multiple imputation by chained equations (MICE) approach^26,51 to address missingness (≤34.6%; Table 1). The MICE model included all variables either used in the primary analysis (ie, 3 pain-related outcomes, 2 covariates) or used to create variables needed for the primary analysis (ie, 5 CM indicators). Given the extent of missingness on exposure and outcome variables, we added 8 auxiliary variables to the MICE model in an effort to increase the precision of the values generated through multiple imputation.^13,20 We chose auxiliary variables from those available in the STAGE survey based on their correlations with the primary analysis variables (r ≥ 0.35),¹³ leading to the inclusion of 5 additional measures of other pain-related syndromes (recurrent headache, chronic fatigue syndrome, joint pain, neck pain, shoulder pain) and 3 measures of mental health (depression with and without the loss of a loved one, anxiety). We specified the finalized MICE model to perform 50 imputations, which resulted in a data set with no missing values. All analysis variables were subsequently created from the MICE-derived data set.

Table 1.

Overall descriptive statistics by exposure status.

	Overall sample (N = 25,418)	No CM (N = 10,430)	Any CM (N = 6194)
	n (%)	n (%)	n (%)
Age (mean (SD))	33.2 (7.7)	33.5 (7.6)	33.2 (7.7)
Sex at birth
Female	14,153 (55.7)	6032 (57.8)	3863 (62.4)
Chronic widespread pain
0 criteria	16,981 (66.8)	8011 (76.8)	4250 (68.6)
≥1 criterion	2513 (9.9)	820 (7.9)	945 (15.3)
1 criterion	1305 (5.1)	467 (4.5)	448 (7.2)
2 criteria	144 (0.6)	60 (0.6)	54 (0.9)
3 criteria	270 (1.1)	70 (0.7)	119 (1.9)
4 criteria	794 (3.1)	223 (2.1)	324 (5.2)
Missing	5924 (23.3)	1599 (15.3)	999 (16.1)
Lower back pain
0 criteria	12,771 (50.2)	6133 (58.8)	2926 (47.2)
≥1 criterion	4452 (17.5)	1577 (15.1)	1352 (21.8)
1 criterion	1336 (5.3)	529 (5.1)	385 (6.2)
2 criteria	1318 (5.2)	476 (4.6)	381 (6.2)
3 criteria	1798 (7.1)	572 (5.5)	586 (9.5)
Missing	8195 (32.2)	2720 (26.1)	1916 (30.9)
Irritable bowel syndrome
0 criteria	20,389 (80.2)	9177 (88.0)	4985 (80.5)
≥1 criterion	2543 (10.0)	945 (9.1)	934 (15.1)
1 criterion	637 (2.5)	289 (2.8)	192 (3.1)
2 criteria	558 (2.2)	220 (2.1)	190 (3.1)
3 criteria	428 (1.7)	145 (1.4)	168 (2.7)
4 criteria	488 (1.9)	164 (1.6)	191 (3.1)
5 criteria	328 (1.3)	99 (1.0)	142 (2.3)
6 criteria	104 (0.4)	28 (0.3)	51 (0.8)
Missing	2486 (9.8)	308 (3.0)	275 (4.4)

Overall sample column includes individuals missing CM reports; No CM and Any CM columns only include individuals who provided full CM data. Only participants who endorsed pain or discomfort in at least 2 parts of the body within the 6 months before the STAGE survey, endorsed pain or discomfort in their lower back within the 6 months before the STAGE survey, and endorsed recurrent abdominal discomfort that lasted at least 7 days per month received questions assessing chronic widespread pain criteria, lower back pain criteria, and irritable bowel syndrome criteria, respectively. Otherwise, participants were assumed to not meet criteria for the corresponding pain-related syndromes.

CM, child maltreatment; STAGE, Study of Twin Adults: Genes and Environment.

Our analysis assessed the association between CM counts and CWP, LBP, and IBS criteria at the phenotypic level (ie, among all individuals regardless of relatedness) and the within-twin–pair level (ie, among differentially exposed twin pairs). To account for violating the overdispersion assumption of the Poisson distribution and the clustering of individuals within twin pairs, we estimated adjusted incidence rate ratios (aIRRs) using negative binomial generalized estimating equation (GEE) models with robust standard errors. These models specified an independence working correlation structure, which was selected over an exchangeable structure using the quasi-likelihood information criterion.

We first assessed phenotypic associations by regressing CWP, LBP, and IBS criteria counts separately onto CM counts and demographic covariates. Next, to derive within-twin–pair estimates accounting for familial confounding, we used what is known as the hybrid approach.^1,50 In brief, the hybrid approach involves calculating the average value for each twin pair on CM, in addition to each individual twin’s deviation from their twin-pair average. Each twin’s deviation score from their pair-mean CM value becomes the exposure of interest in co-twin control GEE models (ie, those testing within-twin–pair associations), whereas a twin’s pair-mean CM value is used as a covariate. Critically, the co-twin control approach capitalizes on within-family variability. Thus, twin pairs that are differentially exposed to CM (represented as twin deviation scores from their pair-mean CM value) and vary in their endorsement of pain-related syndrome criteria contribute critical information to association estimates. Using this deviation approach is conceptually akin to controlling for being from a family that is either more or less likely to experience CM due to unmeasured between-family genetic and environmental differences,⁵⁰ thereby ruling out confounding from all sources of twin-pair similarity. Within-twin–pair associations can thus be interpreted as quantifying the relationship between CM counts and counts of pain-related syndrome criteria endorsement controlling for the average level of CM within each twin pair. That is, the within-twin–pair association asks whether a twin who experienced more CM relative to their co-twin (inherently matched on genetic and familial environmental factors) would also experience more CWP, LBP, or IBS. We also created mean and deviation variables for age at survey participation and sex at birth to account for within-twin–pair differences on model covariates. For primary analyses, we estimated within-twin–pair associations among all twin pairs, regardless of zygosity, in addition to within-twin–pair associations among monozygotic (MZ) twin pairs only. Restricting the sample to MZ twins provides us with the strictest genetic control, as it rules out all genetic confounding under the assumption that MZ twin pairs share 100% of their segregating genes.³²

We conducted a series of sensitivity analyses to probe findings from the primary analyses and assess their robustness to different measurement considerations. First, research yields mixed findings as to whether specific CM subtypes play a greater role than others in the relationship between CM and chronic pain.^{14,19,29,30,41,54,59} Thus, we repeated phenotypic and within-twin–pair analyses assessing associations for each of the 5 CM subtypes separately. We included all twins, regardless of zygosity, in these and subsequent sensitivity analyses to maximize statistical precision. Second, there are multiple ways of indexing CM, with some studies measuring CM as a count variable^14,19,59 and others measuring CM dichotomously (any vs none).^4,7,29 To assess the robustness of findings to these different measurement considerations, we repeated phenotypic and within-twin–pair analyses using an any vs none exposure created from total CM counts. Third, our primary analysis assessed associations between CM counts and counts of endorsed pain-related syndrome criteria. However, other STR studies have used dichotomized definitions for these conditions. Acknowledging that the association between CM and criteria counts may differ from the association between CM and meeting the threshold for a pain-related syndrome, we conducted a sensitivity analysis examining dichotomous CWP, LBP, and IBS syndrome definitions from previous STR research. Specifically, participants were considered to meet the threshold for CWP if they endorsed all 4 CWP criteria.^22–25 Similarly, participants were considered to meet the threshold for LBP if they endorsed all 3 LBP criteria.^35,36 Finally, participants were considered to meet the threshold for IBS if they endorsed at least one IBS criterion, consistent with operationalizations of IBS in previous STR studies.^22,24,47 This analysis used GEE models with logit links and binomial distributions to estimate adjusted odds ratios for the dichotomized pain-related outcomes. Fourth, to assess for the presence of sex differences, we repeated phenotypic analyses stratified by sex at birth. Finally, we repeated primary analyses on the original, nonimputed data, excluding twin pairs with missing data to provide a comparison between estimates that accounted for missingness through MICE and those that did not.

3. Results

After excluding 2 individuals with potentially invalid dates of death, our sample included 25,418 individuals within 8937 twin pairs (M_age = 33.2; SD_age = 7.7; 55.7% female). Among those who provided data on pain outcomes, 12.9% (n = 2513) endorsed at least one CWP criterion, 25.8% (n = 4452) endorsed at least one LBP criterion, and 11.1% (n = 2543) endorsed at least one IBS criterion (Table 1). Furthermore, of the 16,624 participants who provided CM data, 37.3% (n = 6194) reported experiencing at least one CM subtype, with emotional abuse/neglect being endorsed the most (29.2%) and physical neglect being endorsed the least (3.9%; Table 2). Notably, 2003 twin pairs were differentially exposed to CM. Among these pairs, 321 twin pairs, 409 twin pairs, and 438 twin pairs differed in their endorsement of CWP, LBP, and IBS criteria, respectively.

Table 2.

Exposure descriptives among overall sample.

	n (%)
CM count
0 subtypes	10,430 (41.0)
≥1 subtype	6194 (24.4)
1 subtype	3558 (14.0)
2 subtypes	1372 (5.4)
3 subtypes	820 (3.2)
4 subtypes	358 (1.4)
5 subtypes	86 (0.3)
Missing	8794 (34.6)
Emotional abuse/neglect
No	12,219 (48.1)
Yes	4853 (19.1)
Missing	8346 (32.8)
Physical neglect
No	17,502 (68.9)
Yes	647 (2.6)
Missing	7269 (28.6)
Physical abuse
No	16,246 (63.9)
Yes	1783 (7.0)
Missing	7389 (29.1)
Sexual abuse
No	17,075 (67.2)
Yes	1019 (4.0)
Missing	7324 (28.8)
Witnessing family violence
No	14,769 (58.1)
Yes	3332 (13.1)
Missing	7317 (28.8)

N = 25,418.

CM, child maltreatment.

3.1. Primary analysis

At the phenotypic level, a one-unit increase in CM was associated with 37% more endorsement of CWP criteria accounting for age and sex at birth (aIRR, 1.37 [95% confidence interval: 1.32–1.42]). On adjusting for unmeasured familial confounding through the co-twin control approach, the association between CM and CWP attenuated slightly yet persisted, with a one-unit increase in CM being associated with 21% more endorsement of CWP criteria in the within-twin–pair comparison (aIRR, 1.21 [1.12–1.31]). Finally, on restricting the sample to MZ twins only for 100% genetic control, the results were similar in that a one-unit increase in CM was associated with 20% more endorsement of CWP criteria (aIRR, 1.20 [1.05–1.38]; Fig. 1).

Figure 1.

Associations between CM and pain-related syndromes. Phenotypic and within-twin–pair estimates of associations between CM and pain-related syndrome criteria derived from separate negative binomial generalized estimating equation models with robust standard errors. All models adjusted for age and sex at birth. Differentially exposed twin pairs 5 321 (chronic widespread pain), 409 (lower back pain), and 438 (irritable bowel syndrome). CM, child maltreatment; MZ, monozygotic.

Regarding the association between CM and LBP, a one-unit increase in CM was phenotypically associated with 18% more endorsement of LBP criteria (aIRR, 1.18 [1.16–1.21]) adjusting for age and sex at birth. Within-twin–pair analyses adjusting for familial confounding showed that a one-unit increase in CM was associated with 15% more endorsement of LBP criteria (aIRR, 1.15 [1.09–1.21]). The results were similar on restricting the sample to MZ twins, in that a one-unit increase in CM was associated with 10% more endorsement of LBP criteria (aIRR, 1.10 [1.01–1.21]; Fig. 1).

Finally, a one-unit increase in CM was associated with 34% more endorsement of IBS criteria in phenotypic models adjusting for age and sex at birth (aIRR, 1.34 [1.29–1.39]). In the co-twin control model, associations attenuated slightly yet persisted, demonstrating that a one-unit increase in CM was associated with 24% more endorsement of criteria for IBS (aIRR, 1.24 [1.14–1.35]). Associations also persisted when we restricted the sample to MZ twins, with a one-unit increase in CM being associated with 14% more endorsement of IBS criteria (aIRR, 1.14 [1.00–1.30]; Fig. 1).

3.2. Sensitivity analyses

Sensitivity analyses demonstrated that results from the primary analyses were robust to various exposure, outcome, and analysis considerations. First, all CM subtypes were associated with each pain-related syndrome at the phenotypic level. Moreover, associations for CM subtypes persisted after adjusting for familial confounding, although confidence intervals for the associations between sexual abuse and CWP and those between physical neglect and LBP and IBS contained the null (Fig. 2; see Table S1, http://links.lww.com/PAIN/C386 in the supplementary materials section for model results). Second and third, the results were similar to the primary analysis when we dichotomized CM exposures and pain-related outcomes, respectively (Table S2, http://links.lww.com/PAIN/C386). Fourth, phenotypic associations were similar between men and women (Table S3, http://links.lww.com/PAIN/C386). Finally, the complete-case estimates derived from nonimputed data followed the same pattern as results from the primary analyses, although the complete-case estimates were greater in magnitude (Table S4, http://links.lww.com/PAIN/C386).

Figure 2.

Associations between CM subtypes and pain-related syndromes. Phenotypic and within-twin–pair estimates of associations between CM and pain-related syndrome criteria derived from separate negative binomial generalized estimating equation models with robust standard errors. All models adjusted for age and sex at birth. Analyses to derive within-twin-pair estimates included all twin pairs, regardless of zygosity. CM, child maltreatment.

4. Discussion

The association between CM and chronic pain is multifaceted and complex.²¹ Research demonstrates that both genetic and familial-level factors are associated with CM^40,45 and chronic pain.^{23,36,49,52,56} These factors may also be potential sources of confounding that could inflate the observed association between CM and chronic pain. Thus, it is imperative to apply designs that account for this potential confounding when assessing the association between CM and chronic pain.

This study applied the co-twin control approach to a large, population-based twin study to assess the association between CM and pain-related syndromes (ie, CWP, LBP, IBS) accounting for sources of familial confounding. First, we assessed the association between CM and pain-related syndromes among unrelated individuals to obtain phenotypic estimates that do not account for unmeasured familial confounding. The results through phenotypic models controlling for age and sex at birth demonstrated that a one-unit increase in CM was associated with 18%, 34%, and 37% more endorsement of LBP, IBS, and CWP, respectively. In other words, for each increase in CM subtype endorsed, individuals reported between 18% and 37% more pain-related syndrome criteria compared with individuals who did not report a history of CM. These findings align with previous research assessing associations between CM and CWP,^7,60 LBP,^7,59,60 and IBS.^7,38,60

Second, we assessed associations among twins to adjust estimates for sources of unmeasured familial confounding. In models including both MZ and DZ twins, a one-unit increase in CM was associated with 15%, 21%, and 24% more endorsement of LBP, CWP, and IBS, respectively. That is, twins who differed by one CM unit had between 15% and 24% more endorsement of pain-related syndrome criteria. The results were similar on limiting the sample to MZ twins who share virtually all their genes and thus provide even stronger control of genetic confounding. Specifically, MZ twins who differed from their co-twin by a one-unit increase in CM had 10%, 14%, and 20% more endorsement of LBP, IBS and CWP, respectively, although they shared genetics and many early environmental experiences.

All associations attenuated in within-twin–pair analyses, suggesting that familial confounding may inflate associations of CM with CWP, LBP, and IBS if not controlled for. Notably, associations for CWP and IBS attenuated the most when implementing the co-twin control approach, suggesting that their phenotypic associations may be more biased due to familial confounding compared with the association with LBP. Furthermore, analyses restricted to MZ twins demonstrated the greatest aIRR reduction for IBS, suggesting that the association between CM and IBS may be more confounded by genetic factors, in particular. It is important to note that this study cannot speak to the specific sources of unmeasured familial confounding that contributed to the attenuation of estimates. Hence, future research is needed to disentangle the specific contributions of genetic, shared environmental, and nonshared environmental factors, in addition to further research on identifying potential factors that may confound the association between CM and chronic pain.

Although all associations in within-twin–pair analyses attenuated, they persisted, such that associations between CM and CWP, LBP, IBS were still statistically significant. The persistence of associations in within-twin–pair analyses suggests that familial confounding does not fully explain the relationships between CM and CWP, LBP, and IBS. Theoretically, associations that persist even after applying the co-twin control approach can be attributed to either nonshared unmeasured confounding that the design cannot account for (eg, childhood mental health conditions preceding CM) or to CM being an independent risk factor for CWP, LBP, and IBS. However, this study is not able to distinguish between which is more likely. Thus, additional research is needed to assess the extent that CM is associated with chronic pain. For instance, triangulating family-based research designs (eg, the co-twin control approach, sibling comparison designs) with prospective data on CM (eg, child protective service records),⁴ clinical diagnoses of pain-related syndrome diagnoses, and potential sources of nonshared unmeasured confounding not accounted for in this study would provide a rigorous assessment of the extent to which intervening on CM would be expected to decrease subsequent chronic pain.

Still, whether the findings of this study are due to nonshared unmeasured confounding or to CM being an independent risk factor for CWP, LBP, and IBS, the general patterns of attenuation in associations demonstrate the importance of adjusting for unmeasured familial factors when assessing associations between CM and CWP, LBP, and IBS. Thus, future research on the association between CM and pain-related syndromes should consider the impact of sources of unmeasured familial confounding on associations. Based on the findings of this study, it may be best to use family-based research designs (eg, twin comparisons, full sibling comparisons,¹⁰ children-of-twins comparisons)^10,31 when studying associations between CM and CWP, LBP, and IBS, as all these designs inherently control for sources of unmeasured genetic and shared environmental confounding.

Of note, we found that associations with pain-related syndrome criteria were consistently positive for all specific CM subtypes in both phenotypic and within-twin-pair models. However, it is important to recognize that decreased statistical power may have contributed to some confidence intervals containing the null in within-twin–pair analyses of physical neglect and sexual abuse. Although there is some evidence that physical neglect may not be as consistently associated with chronic pain compared with other CM subtypes,³⁰ our findings largely align with previous research suggesting that the accumulation of CM, regardless of type, may be a risk factor for chronic pain.^9,19,59

This study has several strengths, including the application of the co-twin control design to rule out genetic and shared environmental confounding among a population-based sample. To the best of our knowledge, this is the first study to use this design-based adjustment to account for familial confounding of the relationship between CM and pain-related syndromes. However, the findings from this study should be interpreted in the context of its limitations. First, the co-twin control design has limitations that may affect the results of this study. For instance, comparing twins with their co-twins leads to compounded exposure measurement error, which can attenuate within-twin–pair estimates and lead to type II error.^16,32 Furthermore, as noted above, co-twin comparisons can only account for unmeasured factors shared between twins, leaving nonshared, unmeasured factors that may affect twin-pair discordance unaccounted for. In fact, bias due to unmeasured nonshared confounding can be exacerbated by the design.^16,32 Thus, causal interpretations are limited. Second, due to the cross-sectional nature of this study, we relied on retrospective self-reports to capture CM, which may be subject to recall bias.^2,3 Third, our measures of CM did not capture some aspects that have been shown to be important for chronic pain and other outcomes, such as when in childhood CM occurred and the frequency of repeated CM events.⁴⁸ Fourth, our pain-related outcomes were not actual diagnoses, as participants self-reported CWP, LBP, and IBS criteria. Moreover, our measures of pain-related syndromes varied in assessment timeframes across conditions. However, our recent research has provided initial validation of these measures.⁴⁶ Furthermore, sensitivity analyses using suggested diagnostic-like criterion cut-offs produced similar results. Finally, this study used a population-based data set composed of Swedish individuals. Thus, the generalizability of our findings outside of Sweden is not known.

5. Conclusion

The findings from this study demonstrate that individuals who report CM endorse more CWP, LBP, and IBS criteria, even after accounting for genetic and early, shared environmental confounding. The persistence of associations may either be attributable to nonshared sources of unmeasured confounding or to an independent association specific to CM, although this study cannot distinguish between these explanations. Still, the general patterns of attenuation among all within-twin-pair associations underscores the value of the co-twin control in strengthening estimates of the putative effect of CM on CWP, LBP, and IBS. The results highlight the importance of further research to understand the extent to which intervening on CM would be expected to decrease subsequent chronic pain.

Supplementary Material

Supplemental digital content

Supplemental digital content associated with this article can be found online at http://links.lww.com/PAIN/C386.

Supplemental digital content is available for this article. Direct URL citations appear in the printed text and are provided in the HTML and PDF versions of this article on the journal’s Web site (www.painjournalonline.com).