Preliminary Validation of a General Factor Model of Chronic Overlapping Pain Conditions

Abstract

Chronic overlapping pain conditions (COPCs) by definition frequently co-occur, perhaps reflecting their shared etiologies. Their overlapping nature presents a methodological challenge, possibly masking associations between COPCs and health outcomes attributable to either general or specific processes. To address this challenge, we used population-based cohort data to evaluate the predictive validity of a bifactor model of nine self-reported COPCs by assessing its association with incident pain-related clinical diagnoses; pain-relevant pharmacotherapy; and other health outcomes. We obtained data from a 2005–2006 study of Swedish adult twins linked with health data from nationwide registers through 2016 (N = 25,418). We then fit a bifactor model comprising a general COPC factor and two independent specific factors measuring pain-related somatic symptoms and neck and shoulder pain. Accounting for age, biological sex, and cancer, the general factor was associated with increased risk of all pain-related outcomes (e.g., COPC diagnosis aOR, 1.71; 95% CI [1.62, 1.81]), most mental health-related outcomes (e.g., depression aOR, 1.72 [1.60, 1.85]), and overdose and mortality (e.g., all-cause mortality aOR, 1.25 [1.09, 1.43]). The somatic symptoms specific factor was associated with pain-relevant pharmacotherapy (e.g., prescribed opioids aOR, 1.25 [1.15, 1.36]), most mental health-related outcomes (e.g., depression aOR, 1.95 [1.70, 2.23]), and overdose (e.g., nonfatal overdose aOR, 1.66 [1.31, 2.10]). The neck and shoulder pain specific factor was weakly and inconsistently associated with the outcomes. Findings provide initial support for the validity and utility of a general-factor model of COPCs as a tool to strengthen understandings of co-occurrence, etiology, and consequences of chronic pain.

Introduction

Chronic pain, defined as persistent or recurrent pain lasting longer than three months,¹ is prevalent and impairing. Approximately one in five adults experience chronic pain.² Furthermore, lower back pain alone is the leading cause of disability worldwide.³ Chronic pain is often comorbid with an array of adverse health outcomes, including depression,^4,5 anxiety,^4,5 substance use disorder (SUD),⁶ and, in some studies, increased risk of all-cause mortality.^7,8 Moreover, chronic pain conditions are often comorbid with one another. In fact, several conditions, including both chronic pain conditions (e.g., fibromyalgia) and related somatic syndromes (e.g., chronic fatigue syndrome, irritable bowel syndrome), overlap so frequently that they have been conceptualized as chronic overlapping pain conditions (COPCs).⁹

High comorbidity rates among the COPCs suggest that they might share etiologies, at least partially. Possible shared underlying processes have been examined at psychological, neurobiological, and genetic levels.^10–12 In particular, central sensitization theory proposes an amplification of pain via physiological changes to central nervous processes.^12,13 Rather than understanding pain as merely resulting from a specific event or injury, central sensitization suggests that a general increased sensitivity to a range of both painful and nonpainful stimuli underlies the development of chronic pain.¹² This conceptualization of chronic pain may help explain pain that does not resolve after recovery from injury, the high—and shared—heritability across some COPCs,^11,14,15 and the often-cited comorbidity between COPCs and mental health conditions.^4,12

The co-occurring nature of COPCs poses a methodological challenge for chronic pain research for multiple reasons. First, research that collapses chronic pain into a binary indicator (i.e., any vs. none) cannot capture the complex relationships among COPCs. Second, associations between COPCs and health outcomes may be due to both general and specific processes. For example, a susceptibility to pain generally may be a risk factor for some health outcomes, whereas specific COPCs may be risk factors for others. However, statistically controlling for other COPCs in conventional regression models to examine processes specific to certain COPCs fails to consider shared etiology, which could potentially obscure meaningful relationships. Thus, methodological strategies that better reflect the underlying theoretical mechanisms of COPC comorbidities are needed.

Bifactor measurement models provide a potential solution to assessing the complex theoretical relationships among COPCs. Bifactor models decompose construct variance by modeling both general and specific latent processes. In other words, they can help identify whether constructs are derived from heterogenic (i.e., specific) processes or comorbid (i.e., general) processes.¹⁶ This makes them a useful tool for understanding associations among common and unique construct elements and specific outcomes.¹⁷ Bifactor modeling approaches have been used in various ways in mental health^17–19 and in emerging pain research.^20–23 For instance, researchers have used bifactor models to conceptualize the latent structure of questionnaire measures of medically unexplained symptoms and related conditions (e.g., functional somatic syndromes).^21–23 Moreover, we recently employed a bifactor modeling approach to demonstrate that common factors underlying different self-reported COPCs are associated with risk of suicidal behavior because of shared familial processes.²⁴ However, there is a need for broader assessment of the convergent and criterion validity of general and specific COPC factors as they relate to other, real-world clinical outcomes.

To address the methodological challenge posed by the overlapping nature of COPCs, our previous research reported a bifactor model of general and specific variation among self-reported COPCs.²⁴ In the present study, we used population-based data from Swedish adults to assess this model’s predictive validity by examining associations with pain-related clinical diagnoses and receipt of pain-relevant pharmacotherapies. We also assessed associations with a variety of other important health outcomes, such as those related to mental health, drug overdose, and all-cause mortality.

Methods

Data Source and Sample

We analyzed data from participants in the Study of Twin Adults: Genes and Environment (STAGE), a population-based assessment of a wide range of health conditions and related exposures conducted in 2005–2006. STAGE targeted the approximately 42,000 monozygotic and dizygotic twins born in Sweden between 1959 and 1985.²⁵ A total of 25,420 individuals (approximately 60%) gave informed consent and completed either an online questionnaire or a telephone interview. Previous research shows that on average, STAGE participants had completed three years of upper secondary school.²⁶ See prior research for additional STAGE sociodemographic characteristics.^26–28 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.

All residents of Sweden have personal identity numbers, which can be used to link them to data from the nationwide Swedish registers of health and other social data. For the present cohort study, the Swedish Twin Registry (STR) linked STAGE survey data with participants’ individual records from the National Patient Register (dates and diagnoses from all inpatient hospitalizations, recorded via International Classification of Diseases (ICD-10) codes starting in 1997, and from outpatient specialist services starting in 2001²⁹), the Swedish Prescribed Drug Register (all prescription medications dispensed from pharmacies starting mid-2005³⁰), the Cause of Death Register (all causes of death in 1952³¹), and the Swedish Cancer Register (all diagnoses of tumors starting from 1958³²). We included register data through December 31^st, 2016, providing approximately ten years of register-based follow-up for the STAGE cohort. The current study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE³³) reporting guidelines, modified given the large number of exposures.

Measures

Self-Reported COPCs

We assessed self-reported symptoms of nine COPCs. Further detail regarding COPC assessment criteria can be found in our previous work.²⁴

Chronic Widespread Pain.

We created a five-level variable of symptom counts of chronic widespread pain based on the 1990 fibromyalgia criteria proposed by the American College of Rheumatology,³⁴ which has been used in other STR studies.^4,10,11,35

Chronic Fatigue Syndrome (CFS).

We created a four-level variable of symptom counts for CFS based on a previous STR measure,³⁶ which has also been used in other studies.^10,11,35

Irritable Bowel Syndrome (IBS).

We created a seven-level variable of symptom counts for IBS. This operationalization is similar to Svedberg et al.’s³⁷ definition based on the Rome criteria.

Recurrent Headache.

We created a three-level variable for headache (i.e., no headache, non-recurrent headache, and recurrent headache), which closely aligns with other STR studies of recurrent headache.^10,11,15

Bladder Pain.

Our operational definition of bladder pain was based on Altman et al.’s³⁸ STAGE bladder pain syndrome classification, which comprises a five-level symptom count.

Joint Pain.

We assessed joint pain with a dichotomous variable capturing pain lasting more than six months. This operational definition of joint pain is similar to prior STR studies¹⁰; however, we extended the time length criterion for prolonged joint pain from four weeks to six months.

Neck Pain, Shoulder Pain, and Lower Back Pain.

Each of these three pain conditions was assessed via a four-level variable capturing pain intensity and resulting disability (in everyday activities, social life, and work), which follows Nyman et al.’s^14,39 operational definition in STAGE.

Pain-Related Outcomes

We assessed pain-related diagnoses made by clinicians in inpatient hospitalizations and outpatient specialist visits and recorded via ICD codes. Diagnoses included COPCs, such as fibromyalgia (ICD-10 code: M79.7), CFS (G93.3), IBS (K58), headache (G43, G44, R51), interstitial cystitis (N30.1), and neck and back pain (M54). We included other musculoskeletal-pain-related diagnoses as well (M00-M99 other than M79.7 and M54). We also assessed healthcare utilization for pain as the count of inpatient and outpatient specialist encounters (on distinct dates) resulting in a pain-related diagnosis.

We additionally assessed receipt of pain-relevant pharmacotherapies. Prescription medications of interest included opioids (Anatomical Therapeutic Chemical codes N02Axxx, N02BE51) and benzodiazepines (N05BAxx, N05CDxx, N03AE01). We assessed long-term opioid therapy (LTOT) as filling an opioid prescription on at least two distinct days within a six-month period, consistent with previous Swedish register-based research.⁴⁰

Mental Health-Related Outcomes

We identified clinically diagnosed mental health conditions, including depressive disorders (F32, F33, F34.1), anxiety disorders (F40-F45, F48), bipolar disorder (F30-F31), schizophrenia spectrum disorders (F20-F25, F28-F29), and (non-tobacco) SUD (F10-F16, F18-F19). We also assessed healthcare utilization for mental health as the count of inpatient or outpatient specialist encounters resulting in a mental health diagnosis.

Overdose and Mortality-Related Outcomes

In addition to examining all-cause mortality, we also assessed both fatal (from the Cause of Death Register) and nonfatal (from the National Patient Register) overdoses (T36-T51.0, T96, X40–45, Y10-Y15).

Statistical Analysis

To capture common and specific variation among COPCs, we fit a bifactor measurement model to STAGE derived from our prior work.²⁴ In that previous study, we applied exploratory factor analysis (EFA) to assess the covariance among the self-reported COPCs. This resulted in the extraction of three exploratory factors, which we rotated towards a bifactor solution. Next, we re-fit the EFA using confirmatory factor analyses (CFA), wherein we forced the EFA specific-factor indicators loading less than |0.30| to load to 0.0. Because the measures for chronic widespread pain, lower back pain, and joint pain loaded less than |0.30| on all specific factors, we allowed them to only load onto the general factor. This way, the general factor can be interpreted as capturing chronic widespread pain, back pain, joint pain, and their pain comorbidities, while avoiding potentially anomalous specific factor loading patterns.²⁴

For the current study, we applied CFA to fit the same bifactor model. The model allowed for the nine self-reported COPCs to load onto one general factor and two specific factors. The general factor captured the covariation among the nine COPCs, whereas the specific factors captured the unique covariations among specific EFA-defined subsets of the conditions above and beyond the general factor. The first specific factor in our model represents the unique covariation among pain-related somatic symptoms (CFS, IBS, headache, and bladder pain), while the second specific factor represents the unique covariation among neck and shoulder pain conditions.²⁴

Our primary analytic goal was to examine the predictive validity of the COPC general factor model for subsequent health outcomes. To examine incident outcomes, each outcome analysis excluded participants who had experienced that outcome, other than healthcare utilization and mortality, before completing the baseline STAGE questionnaire. For LTOT, we excluded participants who had records of any prior opioid receipt. Given the rarity of any loss to follow-up (i.e., due to emigration or death only), we considered all outcomes as dichotomous rather than time-to-event. Previous studies of STAGE have reported comparable results with logistic and Cox regression, suggesting that this choice would produce little bias.^24,41

As a preliminary step, we assessed associations between each self-reported COPC and a selection of key outcomes: clinically diagnosed COPCs, other pain conditions, LTOT, depressive disorder, and all-cause mortality. Next, we applied a structural equation modeling framework using Mplus software version 8.3⁴² to examine associations for the latent general and specific factors. All models were adjusted for age (as a linear covariate), biological sex, and registered cancer diagnosis, retrieved from both the Swedish Cancer Register and the National Patient Register. We included only these three covariates because our goal was descriptive. We ran one regression per outcome, resulting in 15 models. Each model regressed the outcome onto covariates and the general and specific factors simultaneously, adjusting standard errors for the nonindependence of twins. Because the COPC symptom counts were ordinal variables, we fit all models using the mean and variance adjusted weighted least squares estimator (WLSMV). WLSMV handles missing data via pairwise deletion and may be robust to missingness at random.⁴³ We estimated odds ratios (ORs) rather than risk ratios (RRs) for associations between the latent factors and outcomes due to computational challenges when deriving risk ratios for associations between latent exposures and dichotomous outcomes. However, ORs will approximate relative risk when outcomes are rare.^44,45 Additionally, we estimated standardized linear regression coefficients for the healthcare utilization counts.

We conducted four sensitivity analyses to examine the robustness of the associations. First, we examined the impact of estimating relative risk with ORs, which can be biased upward for common outcomes. We used R Studio⁴⁶ (package FactoMineR⁴⁷) to compute individual scores on the first principal component (PC1) of the nine self-reported COPCs. For the principal components analysis, all COPC scales were treated as categorical variables, with missing values as a separate category. We then assessed associations between PC1 and the five selected outcomes examined in preliminary analyses. PC1 is an alternative means of assessing the general pain factor that makes no assumptions about latent distributions.⁴¹ We used extracted PC1 scores as exposures in logistic and Poisson regression to estimate ORs and RRs, respectively, for associations between the general factor and the selected incident outcomes. We also estimated RRs for univariate and multivariate associations between the self-reported COPC measures and selected outcomes to further facilitate comparisons between ORs and RRs. Second, we considered the outcomes as time-to-event data, using Cox regression to estimate hazard ratios (HRs) accounting for censoring by death. Because it is not possible to apply Cox regression using the WLSMV estimator, we estimated HRs for associations of PC1 and the self-reported COPC measures with the selected outcomes to facilitate comparisons among HRs, ORs, and RRs. Third, in our main analysis, we treated healthcare utilization for both pain and mental health as linear outcomes because Mplus cannot apply count regression with latent predictors. However, these outcomes were positively skewed. To examine whether the associations derived from our main analysis persisted when accounting for distributional assumption violations, we analyzed dichotomized indices of any healthcare utilization. Lastly, to examine generalizability to individuals with pre-existing conditions before STAGE baseline, we repeated the main analyses including participants in whom the outcomes had occurred prior to STAGE.

Results

After excluding two people with potentially invalid dates of death, an analytic sample of 25,418 individuals remained (M_age = 33.2; SD_age = 7.7; 55.7% female). Tables 1 and 2 provide more information on our sample, including self-reported COPC frequencies and the number of participants who experienced each outcome.

Table 1.

Descriptive statistics for the self-reported COPCs.

Self-reported COPC	n	% (n/25,418)
Chronic widespread pain
0 symptoms	16,981	66.8
1	1,305	5.1
2	144	0.6
3	270	1.1
4 symptoms	794	3.1
Missing	5,924	23.3
Lower back pain
0 symptoms	12,771	50.2
1	1,336	5.3
2	1,318	5.2
3 symptoms	1,798	7.1
Missing	8,195	32.2
Joint pain
No	20,558	80.9
Yes	1,453	5.7
Missing	3,407	13.4
Chronic fatigue syndrome
0 symptoms	16,376	64.4
1	1,729	6.8
2	2,281	9.0
3 symptoms	1,752	6.9
Missing	3,280	12.9
Headache
None	7,212	28.4
Non-recurrent headache	8,058	31.7
Recurrent headache	6,955	27.4
Missing	3,193	12.6
Irritable bowel syndrome
0 symptoms	20,389	80.2
1	688	2.7
2	607	2.4
3	453	1.8
4	500	2.0
5	337	1.3
6 symptoms	107	0.4
Missing	2,337	9.2
Bladder Pain
0 symptoms	16,293	64.1
1	3,462	13.6
2	736	2.9
3	497	2.0
4 symptoms	154	0.6
Missing	4,276	16.8
Neck pain
0 symptoms	13,012	51.2
1	1,356	5.3
2	1,370	5.4
3 symptoms	1,510	5.9
Missing	8,170	32.1
Shoulder pain
0 symptoms	13,717	54.0
1	1,042	4.1
2	1,145	4.5
3 symptoms	1,346	5.3
Missing	8,168	32.1

Note. COPC = Chronic overlapping pain condition.

Table 2.

Descriptive statistics for each outcome.

Outcomes	n	% (n/25,418)
Pain-related outcomes
COPC	2,711	10.7
Other painful condition	6,726	26.5
Any opioid	8,512	33.5
LTOT	2,941	11.6
Any benzodiazepine	2,476	9.7
Healthcare visits for pain
0 visits	17,390	68.4
1	3,153	12.4
2–10	4,233	16.7
11+ visits	642	2.5
Mental health-related outcomes
Depressive disorder	1,199	4.7
Anxiety disorder	1,738	6.8
SUD	598	2.4
Bipolar disorder	232	0.9
Schizophrenia spectrum disorder	165	0.7
Healthcare visits for mental health
0 visits	22,828	89.8
1	836	3.3
2–10	1,132	4.5
11+ visits	622	2.5
Overdose and mortality-related outcomes
All-cause mortality	220	0.9
Overdose mortality	25	0.1
Nonfatal overdose	251	1.0

Note. COPC = Chronic overlapping pain condition; LTOT = Long-term opioid therapy; SUD = Substance use disorder.

Figure 1 depicts the bifactor measurement model, which fit the data well (χ² (41) = 212.32, p < .001, CFI = .996, TLI = .996, RMSEA = .01). Standardized factor loadings on the general factor ranged from 0.33 to 0.98, with chronic widespread pain loading close to unity. COPC loadings on the somatic symptoms specific factor ranged from 0.33 to 0.50. Neck and shoulder pain were constrained to load equally (0.48) onto the neck and shoulder pain specific factor for model identification purposes.

Figure 1.

Bifactor measurement model of chronic overlapping pain conditions.

N = 23,929. Fit χ² (41) = 212.32, p < .001, CFI = .996, TLI = .996, RMSEA = .01. Measured variables are depicted as squares and latent factors are depicted as circles. Variance for all latent factors was fixed at 1.0. Numbers denote standardized factors loadings with standard errors in parentheses. The specific factors were permitted to correlate with each other (r = 0.18; 95% CI [0.11, 0.25]). Neck and shoulder pain specific factor loadings were constrained to be equal to ensure model identification.

Predictive Validity

Most self-reported COPCs were positively associated with clinically diagnosed COPCs, other pain conditions, LTOT, depressive disorder, and all-cause mortality in univariate analyses. Further, the associations at least partially attenuated when all COPCs were included in multivariate analyses (Table S1). These preliminary analyses supported the possibility of shared etiology among COPCs and thus the rationale for implementing a general factor approach as described below.

General Factor

Pain-Related Outcomes.

Over and above the specific factors, the general factor was positively associated with all pain-related outcomes, with demographically adjusted ORs (aORs) ranging from 1.54 (95% CI [1.47, 1.61]) for other musculoskeletal-pain-related diagnoses to 2.02 (1.91, 2.12) for LTOT receipt (Figure 2). The general factor was also associated with greater pain-related healthcare utilization (β = 0.18; [0.17, 0.18]).

Figure 2.

Adjusted odds ratios (ORs) estimating associations between the three factors and pain-related outcomes.

All associations adjusted for age, biological sex, and cancer. ORs scaled per one standard deviation of the latent factor.

Mental Health-Related Outcomes.

The general factor was positively associated with most clinically diagnosed mental health outcomes among our sample. With the exception of schizophrenia spectrum disorders (aOR = 1.17; [1.00, 1.38]), aORs ranged from 1.47 (1.28, 1.68) for bipolar disorder to 1.72 (1.60, 1.85) for depressive disorder (Figure 3). The general factor was also associated with greater mental healthcare utilization (β = 0.08; [0.07, 0.08]).

Figure 3.

Adjusted odds ratios (ORs) estimating associations between the three factors and mental health-related outcomes.

All associations adjusted for age, biological sex, and cancer. ORs scaled per one standard deviation of the latent factor.

Overdose and Mortality-Related Outcomes.

The general factor was positively associated with all-cause mortality (aOR = 1.25; [1.09, 1.43]), fatal overdose (aOR = 1.84; [1.39, 2.43]), and nonfatal overdose (aOR = 1.56; [1.37, 1.78]; Figure 4). See Table S2 in supplementary materials for all model results.

Figure 4.

Adjusted odds ratios (ORs) estimating associations between the three factors and overdose and mortality-related outcomes.

All associations adjusted for age, biological sex, and cancer. ORs scaled per one standard deviation of the latent factor.

Somatic Symptoms Specific Factor

Pain-Related Outcomes.

Independent of the general factor and the neck and shoulder pain specific factor, the somatic symptoms specific factor was positively associated with COPC diagnosis (aOR = 1.38; [1.24, 1.53]), LTOT receipt (aOR = 1.19; [1.08, 1.32]), any opioid receipt (aOR = 1.25; [1.15, 1.36]), and benzodiazepine receipt (aOR = 1.81; [1.62, 2.03]; Figure 2). With the exception of benzodiazepine receipt, these associations appeared weaker than the general factor associations. The somatic symptoms factor was weakly and not statistically significantly associated with other musculoskeletal-pain-related diagnoses (aOR = 1.01; [0.92, 1.12]) and pain-related healthcare utilization (β = 0.01; [−0.01, 0.02]).

Mental Health-Related Outcomes.

Regarding mental health, the somatic symptoms specific factor was positively associated with most outcomes, above and beyond other factors (Figure 3). With the exception of SUD (aOR = 1.15; [0.96, 1.39]), aORs ranged from 1.74 (1.54, 1.96) for anxiety disorder to 2.14 (1.68, 2.73) for bipolar disorder. Furthermore, the somatic symptoms specific factor was also associated with mental healthcare utilization (β = 0.18; [0.16, 0.19]). Mental health associations not containing the null appeared greater in magnitude relative to those for the general factor.

Overdose and Mortality-Related Outcomes.

The somatic symptoms specific factor was positively associated with fatal overdose (aOR = 1.94; [1.14, 3.31]) and nonfatal overdose (aOR = 1.66; [1.31, 2.10]; Figure 4). The somatic symptoms factor was also associated with a greater risk of all-cause mortality, although the confidence interval included the null (aOR = 1.25 [0.98, 1.59]; see Table S2).

Neck and Shoulder Pain Specific Factor

Pain-Related Outcomes.

In contrast to results for the other factors, adjusted associations with the neck and shoulder pain specific factor were all weak and negative for pain-related outcomes. aORs ranged from 0.87 (0.78, 0.97) for any benzodiazepine receipt to 0.97 (0.87, 1.07) for COPC diagnosis (Figure 2). Additionally, the neck and shoulder pain specific factor was weakly and negatively associated with healthcare utilization resulting in a pain diagnosis (β = −0.02; [−0.03, −0.01]).

Mental Health-Related Outcomes.

The neck and shoulder pain specific factor was also weakly and/or negatively associated with all mental health-related outcomes, with aORs ranging from 0.69 (0.51, 0.94) for schizophrenia spectrum disorders to 1.03 (0.91, 1.15) for anxiety disorder diagnoses (Figure 3). Furthermore, the neck and shoulder pain specific factor was associated with little difference in mental healthcare utilization (β = 0.00; [−0.03, 0.04]).

Overdose and Mortality-Related Outcomes.

The neck and shoulder pain specific factor was associated with lower risk of all-cause mortality (aOR = 0.73; [0.59, 0.90]), nonfatal overdose (aOR = 0.83; [0.66, 1.04]), and fatal overdose (aOR = 0.66; [0.40, 1.10]), although confidence intervals for fatal and nonfatal overdose were wide and included the null (Figure 4).

Sensitivity Analyses

When we estimated adjusted RRs (Table S3) and HRs (Table S4) for univariate and multivariate associations of PC1 and the self-reported COPC scales with selected outcomes, results were comparable to the ORs reported in Table S1. This similarity suggests that our results would not be substantively biased by variations in outcome incidence or by not considering time-to-event information, respectively. See Figure S1 for time-to-event descriptive data.

When we dichotomized our indices of healthcare utilization, the general factor remained positively associated with healthcare utilization for pain and mental health, and the somatic symptoms specific factor remained positively associated with mental healthcare utilization but not pain-related healthcare utilization. Lastly, the neck and shoulder pain specific factor was weakly, negatively associated with healthcare utilization for pain and mental health. aORs from this analysis are presented in Figures 2 and 3 to facilitate comparison with aORs for other outcomes in the primary analyses.

When we included individuals with pre-existing conditions before STAGE baseline, associations for the general factor followed the same patterns (Table S5). Associations for the somatic symptoms specific factor were also similar, with the exception that the factor was also statistically significantly associated with SUD diagnosis. Lastly, associations for the neck and shoulder pain specific factor generally followed the same pattern, with associations remaining weak and negative.

Discussion

Overall, we found support for the validity of a general-and-specific-factors modeling approach to COPCs. The self-reported COPC general factor was associated with increased odds of experiencing all subsequent register-based, independently measured pain-related outcomes, most mental health-related outcomes, and all mortality and overdose-related outcomes. Further, the somatic symptoms specific factor was associated with pain-related outcomes related to medication receipt, most mental health-related outcomes, and overdose over and beyond the general factor. In contrast, the neck and shoulder pain specific factor was weakly and negatively associated with most outcomes. Sensitivity analyses supported the robustness of the results.

General Factor

Our results provide evidence of the convergent validity of the self-reported general factor, in that it was associated with COPC diagnoses made by clinicians. A one-standard-deviation increase in the COPC general factor was associated with 1.71-fold greater odds of receiving a real-world clinical COPC diagnosis. Moreover, the COPC general factor’s criterion validity was supported by its association with increased risk of outcomes conceptually related to a general liability for pain, including opioid receipt, LTOT receipt,⁴⁸ pain-related healthcare visits, and outcomes related to mental health,^4,12 overdose,⁴⁹ and mortality.^7,8 The current study is also among the first to suggest that general processes related to pain are associated with increased mortality risk, contrary to other studies that have focused on any chronic pain or chronic widespread pain and report inconsistent findings.^7,8 Further, our results build upon previous cross-sectional research showing that a general factor of symptom distress is associated with health anxiety, depressive symptoms, somatosensory amplification, and any existing self-reported functional somatic syndrome.^21,22

A possible interpretation of our findings is that the general factor captures underlying etiologic processes contributing to COPCs. This interpretation would be consistent with central sensitization theory, which posits that COPCs share etiologic amplification of pain via changes to central nervous system processes.¹² Thus, the general factor may offer a means of capturing broad liability for COPCs. However, it is important to note that COPC covariation could also be a product of an underlying network of symptom co-occurrence, wherein having one COPC increases one’s risk for other COPCs due to underlying interconnected mechanisms.⁵⁰ Other interpretations include the possibilities that the general factor reflects a shared consequence of various specific types of pain or that the pain-related or other outcomes actually increase risk of self-reported COPCs. Within the context of substance use specifically, for example, the Catastrophizing, Anxiety, Negative Urgency, and Expectancy (CANUE) model proposes a reciprocal relationship between chronic pain and substance use.^51,52 Further research is needed to investigate these possibilities.

Somatic Symptoms Specific Factor

Theoretically and empirically, pain-related somatic conditions such as IBS and CFS appear to be related to mental health conditions.^4,5,10,21,53 The somatic symptoms specific factor was derived from the four included COPCs arguably most related to mental health, and it was associated with most outcomes related to mental health, usually to a seemingly greater extent than was the general factor. Importantly, a one-standard deviation increase in the somatic symptoms factor was also associated with 1.38-fold increased odds of receiving a clinical COPC diagnosis, suggesting an additional etiologic contribution to COPCs. The somatic symptoms factor was also associated with increased risk of pain-related medication receipt and overdose. Our data cannot explain why the somatic symptoms specific factor was associated with increased odds of receiving prescription opioids and LTOT, as these therapies are typically contraindicated for conditions such as IBS or headache. One explanation could be the documented trend of individuals with mental health conditions being more likely to receive prescription opioids in greater quantities relative to those without such conditions.^54–61 This line of evidence, termed ‘adverse selection,’ suggests that patients who are at greatest risk for adverse outcomes tend to be prescribed the most potentially harmful therapies.^54,59,60

Neck and Shoulder Pain Specific Factor

Independent of the general and somatic symptoms factors, the neck and shoulder pain specific factor was weakly and inconsistently associated with most outcomes. At first glance, our findings may appear to suggest that neck and shoulder pain, independent of the general and somatic symptoms factors, could be a modestly protective factor against some adverse outcomes. However, univariate analyses showed that neck and shoulder pain were individually associated with increased risk of various outcomes (Table S1). It may simply be that once we control for the general and somatic symptoms factors, there is little meaningful variance remaining in the neck and shoulder pain measures. Additionally, specific factors from bifactor models can be limited in their reliability and validity. Thus, inferences regarding this factor should be made with caution.^62–64

Implications

Our findings provide support for measuring pain across traditionally disparate conditions rather than only measuring pain specific to one location. Because using a bifactor model to conceptualize pain is a relatively novel area of research, the goal of the study was descriptive. However, we believe the findings can be used to generate further clinical and research avenues. For instance, the preliminary validation of the self-reported general factor provides further support for the development of clinical assessments that measure a wide range of COPCs concurrently, such as the Chronic Overlapping Pain Conditions Screener.⁶⁵ Moreover, there has been growing discussion in clinical contexts of the challenges of pain classifications.^66,67 For example, some argue for a new, common classification of functional somatic syndromes, in which these disorders are not split between whether they are more somatic or more mental health-related but includes elements of both domains.⁶⁶ This approach has been proposed to strengthen therapeutic partnerships between physicians and patients, develop novel services for people with COPCs, and generate intervention-relevant research. The current study bolsters the need for and contributes to this discussion, as findings may support a more overarching, less-location-specific pain etiology.⁶⁶

Additionally, because the general factor was associated with register-based COPC diagnoses and other pain-related healthcare outcomes—whereas the somatic symptoms specific factor appeared to be a stronger predictor of mental health outcomes—it is possible that our approach could be used as a method of assessing common underlying mechanisms contributing to the COPCs. For example, research has identified central sensitization as a potential general underlying neurobiological mechanism of pain.^12,68 However, it has recently been suggested that some commonly used survey measures of central sensitization may not accurately capture the central sensitization construct but instead may assess pain-related psychological constructs.⁶⁸ Thus, the COPC general factor approach may be a useful alternative measurement strategy for exploring central sensitization etiology and consequences, in addition to other psychosocial processes, such as pain catastrophizing.⁶⁸ Further investigation of the COPC general factor could potentially reveal important information regarding the etiology of chronic pain, as well as causal processes linking pain to other adverse health outcomes.

Limitations

Our findings should be considered in the context of the current study’s limitations. First, our results should not be interpreted as examining causal relations with clinical outcomes. Second, we included in-depth measures of a wide array of COPCs, but there are others that future research could examine (e.g., temporomandibular disorder, vulvodynia, endometriosis, and migraine or other headache types⁹). Third, although outcomes came from nationwide records over 10 years, the Swedish National Patient Register only provides information on inpatient and outpatient specialist encounters. Thus, individuals who did not seek medical care or received only primary care treatment were not identifiable. Assuming misclassification is nondifferential, this could lead to an attenuation of associations towards the null. Fourth, the current study applied a bifactor model because of its ability to model theoretically relevant COPC comorbidity and heterogeneity simultaneously. However, this does not guarantee that a bifactor model is the best or only way to conceptualize the mechanisms underlying COPCs. Future research is needed to compare the bifactor model with other measurement approaches. Fifth, although our sample was large and population-based, selection bias is possible, as STAGE participants likely differ from nonparticipants.²⁷ However, research shows that STAGE participants and nonparticipants are roughly comparable regarding age,²⁷ education, parental education, and parental criminal history.^26,28 Moreover, in similar research using STAGE, selection bias did not appear to impact observed associations.⁴¹ Sixth, missingness on the self-reported COPC measures, likely driven by the STAGE survey’s size and scope, may have biased observed associations.²⁵ Seventh, we did not consider the social and cultural context of the COPCs, which may contribute to their co-occurrence and associations with related outcomes. It may be beneficial to incorporate general factor models into future research on sociocultural processes in COPC etiology. Lastly, the generalizability of our findings is unknown because we relied on a sample of young-to-midlife Swedish-born twin adults, and we lacked data on race and ethnicity.

Conclusions

The current study supports the utility of a general-and-specific-factor approach to conceptualizing the complex relationships among COPCs. In addition to supporting a measure of self-reported general liability for chronic pain, we identified a distinct and potentially etiologically important additional source of variation specific to pain-related somatic symptoms that appeared more associated with mental-health related outcomes compared with pain-related outcomes. Our findings suggest that future research on the etiology, classification, treatment, and consequences of chronic pain may benefit from applying similar modeling approaches.

Perspective:

This article presents associations between a novel measurement model of chronic overlapping pain conditions (COPCs) and various health outcomes. Findings provide support for measuring pain across multiple domains rather than only measuring pain specific to one physical location in both research and clinical contexts.

Highlights:

• Bifactor models may help capture theoretical similarities and differences among pain conditions.

• General chronic pain liability predicted increased risk of most health-related outcomes.

• A somatic symptoms factor also predicted increased risk of most health-related outcomes.

• Findings preliminary support the utility of a general factor model of chronic pain conditions.