Identifying patterns of pain, depression, anxiety, interpersonal trauma exposure, and nonmedical prescription opioid use: Latent class analysis among patients with chronic pain

Abstract

Background

Chronic pain in the context of certain factors may be associated with potential for nonmedical prescription opioid use; however, identifying this risk can be challenging and complex. Several variables alone have been associated with non-prescribed opioid use, including depression, anxiety, pain interference, and trauma exposure. Prior research has often failed to integrate these assessments together, which is important as these factors may cluster together in important and complex ways. The current study aimed to identify classes of patients with chronic pain who have differential risk for use of nonmedical prescription opioid use, depression and anxiety, and pain severity, interference, and catastrophizing, and interpersonal violence exposure.

Methods

Self-report and medical record data from patients (N = 211; M_age = 48, 69.0% women, 69.0% white) at a pain management center were collected.

Results

Latent class analysis revealed 3 classes with (1) low probability of clinically significant depression, anxiety, pain, and nonmedical prescription opioid use (44.7%), (2) high probability of clinically significant depression, anxiety, pain, pain catastrophizing, trauma, and nonmedical prescription opioid use (41.3%), and (3) high probability of severe pain and nonmedical prescription opioid use (14.0%).

Conclusions

High-risk classes had either high levels of depression and anxiety, pain catastrophizing, and interpersonal violence exposure, or pain severity and interference. Future research should continue to explore these classes in large, diverse samples, and prospective study designs. Finally, results underscore that opioid use is complex, not easily identified by a single factor, and may be motivated by complex unmet clinical needs.

Introduction

Chronic pain is a major public health issue affecting up to 1 in 3 adults in the United States.¹ Chronic pain is challenging to treat and causes immense suffering to individuals who experience it.^2,3 To cope, many individuals may seek and receive prescription opioids to manage pain symptoms.⁴ Some individuals benefit from prescription opioids, particularly to manage acute pain episodes in the context of chronic pain. However, there is potential for nonmedical prescription opioid use (NMPOU), possibly to further manage pain or related symptoms or due to the development of an underlying opioid use disorder (OUD), which can increase the risk of adverse outcomes like overdose.^5,6

Several factors are prospectively associated with NMPOU, including childhood sexual abuse,⁷ sleep disturbance,⁸ history of a substance use disorder (SUD), and certain mental health conditions.⁹ However, it remains challenging to identify those who may be at risk for NMPOU. Indeed, some measures (eg, the Opioid Risk Tool) have been found to perform no better than chance in predicting future NMPOU.¹⁰ Furthermore, a major issue with traditional statistical methods for detecting risk for NMPOU is testing one predictor at a time or using multivariate approaches that are unable to determine how risk factors cluster together. This can lead to statistically significant but misleading findings, as it is unlikely a single variable will accurately identify those at risk. Indeed, utilizing one variable alone may inappropriately categorize individuals into high-risk groups in a biased fashion, exacerbating known health disparities. For example, considering childhood trauma exposure alone as a risk factor could exacerbate existing racial inequities in opioid prescribing,¹¹ as Black and Latino populations are more likely to experience these types of trauma.¹² Instead, there are likely complex multifactorial patterns of risk that must be captured to identify at-risk groups. Indeed, utilizing person-centered analytic approaches may be one way to determine how risk factors may cluster together, which then may be used to develop more refined risk prediction tools. Considering this, latent class analysis (LCA) may be a promising method to identify chronic pain patient groups at risk for NMPOU. Unlike traditional methods, LCA can integrate multiple variables simultaneously¹³ to identify classes of individuals who have differential levels of risk for NMPOU and other important features.

Prior LCA work has identified discrete classes of pain patients. For example, one study of adult pain patients with an opioid prescription identified 3 latent classes of individuals: Poor mental health, poor physical health, and hazardous alcohol use, with the poor mental health class being at increased risk of opioid misuse.¹⁴ Another study integrated pain catastrophizing, which is defined as the tendency to think about pain as catastrophic, such as ruminating about the pain, magnifying the threat associated with pain, and feeling helpless to handle the pain.¹⁵ This study also included individuals with chronic pain currently prescribed opioids and identified four groups with low pain catastrophizing/low mental health symptoms, low pain catastrophizing/high attention deficit hyperactivity disorder (ADHD) symptoms, high pain catastrophizing/high anxiety, and high pain catastrophizing/high mental health.¹⁶ Finally, a study examining a large sample of 1514 people prescribed opioids for chronic non-cancer pain in Australia found 3 classes of child abuse exposure, with the group with the highest levels of childhood abuse having the highest levels of pain, mental health symptoms, and risk for OUD.¹⁷ Although prior studies are informative, they were all conducted among patients who were prescribed opioids, rather than individuals who may be seeking opioids or other treatments to manage pain. Furthermore, no studies included all key variables we propose may be relevant as indicators of NMPOU: Pain severity¹⁸ and pain interference (ie, functional impairment in activities of daily living caused by pain and may predict negative pain-related outcomes above and beyond pain severity alone),¹⁹ mental health symptoms,⁷ pain catastrophizing,²⁰ and interpersonal violence exposure.²¹ Each of these factors represent potentially unmet clinical needs that may motivate NMPOU.

To address this gap in research, we conducted an LCA among outpatients with chronic pain, integrating various potentially important variables associated with risk for NMPOU, including depression and anxiety, pain catastrophizing, pain severity and interference, and previous interpersonal violence exposure²¹ (which includes but is not limited to child abuse exposure). We elected to use LCA (vs latent profile analysis) to ensure results using established clinical cut-offs that are easily interpretable to clinicians. To validate classes with exogenous variables not included in the LCA analyses, we examined whether suicidal ideation and history of prior suicide attempts varied across identified classes.

Methods

Participants and procedure

Patients (N = 211) presenting for a psychological intake evaluation at the University of North Carolina (UNC) at Chapel Hill Pain Management Center from May 2016 to May 2022 were included in the current study. As part of their evaluation, participants completed self-report measures and a standardized, semi-structured clinic interview. This study was reviewed and considered to be exempt by the Institutional Review Board at the University of North Carolina at Chapel Hill as data were collected as part of routine clinical care.

Self-report measures

Brief Pain Inventory (BPI). The BPI is a 15-item self-report assessment of pain intensity and interference. Participants rate their levels of pain severity and interference within the past 24 hours, as well as on average. Pain interference total scores were averaged to maintain rating out of 10. The BPI has good psychometric properties,²² including among individuals with chronic pain.²³ The BPI can be used to discriminate patients with severe pain intensity and interference with cut-offs of 7 and 6, respectively.²⁴

Pain Catastrophizing Scale (PCS). The PCS is a 13-item self-report measure assessing the rumination, magnification, and helplessness that defines catastrophizing thought patterns in the context of pain. Participants report the degree to which they have certain thoughts and feelings related to pain (eg, I feel I can’t go on; I become afraid the pain will get worse). Total scores range from 0 to 52, and scores above 30 indicates pain catastrophizing greater than the 75th percentile in an original sample of injured workers filing for disability.¹⁵

Generalized Anxiety Disorder-7 (GAD-7). The GAD-7 is a brief, 7-item self-report measure designed to screen for generalized anxiety occurring within the past two weeks. It has demonstrated good validity and reliability in clinical settings.²⁵ Total scores range from 0 to 21, with clinically significant symptoms categorized as mild (5–9), moderate (10–14) or severe (15+).²⁵ A cut-off of 10 representing at least moderate anxiety symptoms was used in the current study.

Beck Depression Inventory—2nd edition (BDI-II). The BDI-II is a 21-item self-report measure assessing the severity of depression symptoms in the past 2 weeks.²⁶ The BDI-II has been used widely to assess and diagnose depression. It has demonstrated validity and reliability in chronic pain patients.²⁷ Total scores range from 0 to 63, with clinically significant symptoms categorized as mild (14–19), moderate (20–28), or severe (29+).²⁶ In the current study, a cut-off of 14 was used to represent those with at least mild symptoms of depression.

Current Opioid Misuse Measure (COMM) – adapted. The COMM is a 17-item measure used to assess NMPOU.²⁸ The COMM short form (COMM-SF) reduced the number of items to 5²⁹ and was tested among 134 patients receiving chronic opioid therapy. Specific items include going to someone other than one’s prescribing physician to get pain medications; taking medications differently than prescribed; spending a lot of time thinking about opioid medications; others worrying about the way one is taking one’s medications; and using opioid medications for symptoms other than pain. The COMM-SF was validated and shown to improve ease of administration.²⁹ Due to inclusion of items assessing potential for opioid misuse, and to avoid use of stigmatizing language toward chronic pain patients and recognize that there are multiple reasons why patients may use opioids in ways other than prescribed (eg, underlying OUD, poorly managed pain symptoms), we use the phrase NMPOU rather than traditionally used phrase “opioid misuse.” The total score was used in LCA analyses, with scores ≥1 indicating risk for NMPOU. All COMM-SF items were administered in the current study, as well as 2 additional items from the original COMM assessing taking more medications than prescribed and borrowing pain medication from someone else. Internal consistency was somewhat low (α = 0.51). Therefore, we examined the prevalence of each individual item within classes identified by our LCA.

Medical record coding

During the intake evaluation, patients completed a semi-structured clinical interview with a licensed clinical psychologist with specialized training in Pain Psychology. Results of the evaluation were entered into a templated note, allowing for more reliable extraction of data. Undergraduate and/or medical student research assistants (RAs) were trained to code these notes by completing a training with the first author or a trained coder, and conducting practice coding sessions which were reviewed with trained coders. All records were coded by two separate RAs and then compared for accuracy. Any discrepancies were adjudicated by consensus discussion, with input from the first author as needed. Variables extracted from templated intake note included: Demographics (age, race/ethnicity, marital status, insurance coverage), exposure to physical or sexual assault or abuse (clinicians were required to enter whether previous trauma/abuse was reported and if so, what type, which was then coded by trained RAs), substance use (clinicians were required to input endorsement of current tobacco, alcohol, cannabis, or other substance use in the templated note), history of suicide attempts, and the presence of current suicidal ideation.

Data analytic plan

LCA analyses were conducted in SAS 9.4.³⁰ In LCA, 2 parameters are estimated using maximum likelihood estimation: Latent class prevalences, indicating the size of each class, and item response probabilities, indicating the probability of a particular factor within each class.³¹ We included a set of 7 variables, which were dichotomized into either absence or presence of clinically significant symptoms. The variables were clinically significant anxiety symptoms (GAD-7 ≥ 10), clinically significant depressive symptoms (BDI-II ≥ 14), pain catastrophizing (PCS > 30), opioid misuse risk (COMM ≥1), history of physical or sexual abuse or assault as reported during clinical interview, severe pain interference (≥6 on the BPI interference scale) and severe pain (≥7 on the BPI severity scale).

To determine the optimal number of classes that best fit the data, we fit a series of latent class models specifying 1 to 7 classes. We used several fit indices to determine the best-fitting model, including the Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC), consistent Akaike Information Criterion (cAIC), and Akaike’s Bayesian Information Criterion (ABIC), with lower values indicating a more optimal balance of model fit and model parsimony.¹³ We also evaluated entropy, a measure of the degree to which the model produces classes that are well separated, as a descriptive measure of the final model. We did not use it as a model selection tool, as distinction and separation between classes may not always be reflective of reality, as naturally occurring groups may often have some degree of overlap.³¹ Additionally, we considered the absolute and relative frequencies of the smallest class and the interpretability of each identified class. To avoid locally optimal solutions, we specified that each estimated latent class model begin with 500 random start values and confirmed that the maximum log likelihood value was replicated in the best fitting model.

Once we selected the best fitting model, we assigned individuals to the latent class for which they had the highest posterior probability of membership to examine the demographic characteristics including age, gender, race, health insurance type, and marital status, of each class. As an additional validation of class structure, we also examined history of suicidal attempts and current suicidal ideation across classes. To account for the uncertainty associated with latent class membership (ie, the fact that each individual does not perfectly “belong” to one class), we weighted individuals in each class according to their posterior probability of class membership and generated weighted counts and percentages of each demographic characteristic, history of suicide attempts, and current suicidal ideation.³²

Results

Preliminary analyses

Patients (N = 211) were an average of 48 years of age, and most were women (69.0%; Table 1; Table S2). Most self-identified as White non-Hispanic (69.0%), followed by Black non-Hispanic (21.5%), and Hispanic or other races (9.5%). More than half had Medicaid/Medicare health insurance. The greatest proportion were married or cohabitating (41.3%). Regarding reasons for referral, participants were being evaluated for: Diagnostic assessment and pain coping skills (89.1%), suitability for long-term opioid prescription (28.4%), or suitability for spinal stimulator implantation (6.6%). More than one referral reason was possible, resulting in percentages >100%. Many participants reported clinically significant levels of anxiety (40.5%), depression (65.8%), pain catastrophizing (30.8%), potential for non-prescribed opioid use (59.3%), interpersonal trauma exposure (45.0%), pain interference (59.2%), and severe pain (59.1%).

Table 1.

Demographic characteristics of participants by class.

	Class 1 N = 81.4 (44.7%)a	Class 2 N = 75.4 (41.3%)a	Class 3 N = 25.5 (14.0%)a
	N (%)a	N (%)a	N (%)a
Age M (SD)	48.9 (15.1)	44.9 (11.7)	51.0 (10.4)
Gender
Female	53.3 (67.6)	51.8 (71.2)	16.9 (66.1)
Male	25.5 (32.4)	21.0 (28.8)	8.7 (33.9)
Race
White non-Hispanic	61.5 (79.6)	46.4 (63.9)	13.1 (52.6)
Black non-Hispanic	9.0 (11.7)	18.1 (24.9)	9.2 (37.1)
Hispanic or other races	6.7 (8.7)	8.1 (11.2)	2.5 (10.3)
Insurance type
Medicaid/Medicare	32.8 (42.4)	44.0 (61.9)	15.6 (60.9)
Private	32.7 (42.4)	13.5 (19.0)	5.7 (22.3)
Uninsured or other	11.7 (15.2)	13.5 (19.0)	4.3 (16.8)
Marital status
Single	25.8 (33.8)	28.3 (38.5)	12.9 (50.4)
Married or living together	34.9 (45.7)	28.7 (39.0)	9.8 (38.5)
Divorced, separated, or widowed	15.6 (20.4)	16.5 (22.5)	2.8 (11.1)

^aCounts and percentages were weighted by the posterior probability of belonging to a given class to account for the uncertainty associated with latent class membership (ie, the fact that each individual does not perfectly “belong” to one class).

Primary analyses

Model Fit Across Class Structures. Table 2 displays the fit indexes for the latent class models with 1 to 7 classes. AIC values decreased sequentially from the 1- to 4-class model; however, improvements from the 3- to 4-class structure were minimal. Specifically, the 4-class model had slightly lower LL, AIC, and ABIC levels, but the reductions were relatively small in size. The 3-class model performed better on BIC and cAIC. Given that improvements in model fit were inconsistent and minimal from Class 3 to 4, and considering the principle of parsimony, the 3-class model was preferred. BIC values decreased sequentially from the 1 to 3-class model, and increased in the 4-class model, again supporting the selection of a 3-class structure. Similar patterns were observed for cAIC and ABIC, all suggesting the 3-class model was the best fitting model for the data. Although not used for model selection, entropy for the 3-class value was 0.69, indicating a moderate separation of classes.

Table 2.

Fit statistics by number of classes in latent class analysis.

Number of Classes	LL	df	AIC	BIC	cAIC	ABIC	Entropy
1	−840.26	120	286.27	309.70	316.70	287.52	1
2	−757.11	112	135.96	186.16	201.16	138.64	0.71
3	−745.41	104	128.57	205.56	228.56	132.68	0.69
4	−736.53	96	126.80	230.56	261.56	132.33	0.70
5	−731.94	88	133.62	264.15	303.15	140.58	0.69
6	−727.03	80	139.81	297.12	344.12	148.20	0.69
7	−724.70	72	151.14	335.23	390.23	160.96	0.72

ABIC = Akaike’s Bayesian Information Criterion; AIC = Akaike Information Criterion; BIC = Bayesian Information Criterion; cAIC = Consistent Akaike Information Criterion; df = degrees of freedom; LL = log likelihood.

3-Class Model Features. Figure 1 depicts the item response probabilities for each construct across classes (see Table S1 for item response probabilities and corresponding standard errors). Class 1 (low probability of clinically significant depression and anxiety, pain, and potential for NMPOU) was the largest class and included about 81 participants (44.7%). The most commonly endorsed items from the COMM-SF was spending a lot of time thinking about opioids, including whether they had enough opioids (26.0%). It was characterized by relatively low probabilities of clinically significant anxiety and depression symptoms, pain catastrophizing, potential for NMPOU, interpersonal violence, pain interference, and severe pain. Demographically, it was majority women (67.6%) and white non-Hispanic patients (61.5%). Less than half (42.0%) reported alcohol use, 22.9% currently used tobacco, 9.6% used cannabis, and 1.3% reported other drug use. A relatively low prevalence of suicide attempt history (16.5%) and current suicidal ideation (4.9%) was observed.

Figure 1.

Item response possibilities by class in the 3-class model. BDI = Beck Depression Inventory; COMM = Current Opioid Misuse Measure; GAD = Generalized Anxiety Disorder-7; PCS = Pain Catastrophizing Scale. Clinical cut-offs were used: GAD-7 > 10,²⁵ BDI > 14,²⁶ PCS > 30,¹⁵ COMM > 1,²⁸ Brief Pain Inventory interference > 6 and severity > 7.²⁴

Class 2 (high probability of clinically significant depression and anxiety, pain, pain catastrophizing, trauma, and potential for NMPOU) was the second largest class and included about 75 individuals (41.3%). In this class, the most commonly endorsed items from the COMM-SF were having to go to someone other than their prescribing physician to get sufficient pain relief from medications (34.2%) and spending a lot of time thinking about opioids, including whether they had enough opioids (32.5%). Class 2 was characterized by the highest probabilities of anxiety, depression, pain catastrophizing, interpersonal violence exposure, and pain interference, as well as relatively high probabilities of severe pain and potential for NMPOU. Demographically, Class 2 was mostly women (71.2%), with a slightly higher composition of women compared to Classes 1 and 3, and mostly White non-Hispanic patients (63.9%). Regarding substance use, over a third (37.6%) reported using tobacco, 27.7% alcohol, 14.3% cannabis, and 1.1% other drugs. A higher prevalence of past suicide attempts (37.6%) and current suicidal ideation (20.5%) was observed as compared to Classes 1 and 3.

Class 3 (high probability of severe pain and potential for NMPOU) was the smallest class and included about 25 individuals (14.0%). The most commonly endorsed items from the COMM-SF in this class were going to someone other than their prescribing physician (eg, another physician, the emergency department, a friend) to get sufficient pain relief from medications (36.8%) and taking medications differently than prescribed (36.8%). Class 3 was characterized by moderate probabilities of depression and interpersonal violence exposure, a high probability of pain interference and severe pain, and an equally high probability of potential for NMPOU to that of Class 2. Demographically, Class 3 was also mostly women (66.1%) and a relatively lower proportion White non-Hispanic patients (52.6%), with a relatively higher proportion of Black non-Hispanic (37.1%) patients. In terms of substance use, nearly a third (27.7%) used tobacco, followed by alcohol (23.1%), cannabis (7.3%), and other drugs (2.1%). There were a relatively low prevalence of past suicide attempts (12.6%) and current suicidal ideation (7.2%).

Discussion

The 3 classes displayed quantitatively and qualitatively distinct patterns of symptoms characterized by: (1) low probability of clinically significant depression and anxiety, pain, and potential for NMPOU (44.7%), (2) high probability of clinically significant depression and anxiety, pain, pain catastrophizing, trauma, and potential for NMPOU (41.3%), and (3) high probability of severe pain and potential for NMPOU (14.0%).

The current findings are consistent with and extend on prior work in a variety of ways. Specifically, we found one high-risk class (Class 2) for NMPOU was characterized by high probability of clinically significant depression and anxiety, severe pain, pain catastrophizing, and interpersonal violence exposure. The racial composition of this class was majority White/non-Hispanic. The most common items from the COMM-SF endorsed were having to go to someone other than their prescribing physician to get sufficient pain relief from medications and spending a lot of time thinking about opioids, including whether they had enough opioids. Furthermore, this class had relatively higher rates of tobacco and cannabis use. The composition of this class is consistent with research using LCA and other methods demonstrating high levels of depression and anxiety can be associated with high risk of NMPOU among those with chronic pain.^7,14,16 Indeed, it is likely that these mental health symptoms are associated with motives to use opioids to cope with both physical and emotional pain.³³

Additionally, Class 2 was the only one with elevated pain catastrophizing. Prior research indicates pain catastrophizing tends to co-occur with high levels of mental health symptoms,¹⁶ and may be a pain-related factor associated with risk for NMPOU.^20,34 The high levels of interpersonal violence exposure in this class is also consistent with research suggesting interpersonal violence history may be associated with NMPOU.⁷ However, it is important to note that 30% of Class 1, which was low-risk for NMPOU, also reported interpersonal violence exposure. Therefore, violence exposure may be common among individuals with chronic pain, and not all individuals with this experience exhibit risk for NMPOU. Finally, as further external validation of these classes, we found that members of this high depression and anxiety symptom class had the highest probability of reporting previous suicide attempts and current suicidal ideation. This again underscores the high levels of distress and risk experienced by individuals in this class, and is consistent with prior work linking suicidal ideation and attempts with use of NMPOU.³⁵

The second high-risk class was characterized by high probability of severe levels of pain intensity and interference, with relatively lower (but still moderate) probabilities of clinically significant depression and anxiety, and interpersonal violence exposure. This group had the highest probability of severe pain, and the most commonly endorsed items from the COMM-SF were going to someone other than their prescribing physician to get sufficient pain relief from medications, like Class 2, and taking medications differently than prescribed. There was relatively less substance use compared to Class 2 but not the low risk class (Class 1). Class 3 was composed of relatively more Black participants than other classes. Overall, severe pain and pain interference may indicate risk for, and possibly motivate NMPOU when in the context of moderate depression, anxiety, and interpersonal violence exposure. However, in the absence of these co-occurring symptoms, prior work has found that pain severity alone may not be associated with risk for NMPOU.³⁶ Again, this underscores the need to consider multiple factors together, which may also represent unmet clinical needs, leading patients to use opioids to attempt to treat unmanaged pain.

It is important to note that both high risk classes endorsed spending a lot of time thinking or worrying about opioid medications. While this can be a symptom of SUD (American Psychiatric Association, 2013), it could also be related to barriers to receiving prescription opioids due to stigma or logistics, particularly after the release of the 2016 Centers for Disease Control and Prevention guideline for prescribing opioids for chronic pain.^37,38 To understand the nature of this item, and whether it is related to an underlying SUD, a clinical interview is likely required. However, other items did assess behaviors that can increase risk for adverse outcomes like overdose, such as NMPOU.

The low risk class was the largest class and characterized by low probability of clinically significant depression and anxiety, pain, and potential for NMPOU. Interestingly, this class reported the highest rates of alcohol use compared to the other 2 classes (but still lower than the general US population)³⁹ but slightly lower rates of tobacco use. Of note, rates of depression and NMPOU were still relatively high, which is consistent with prior research in general pain outpatient samples.^40,41 Finally, this class had the highest proportion of individuals with private insurance, potentially indicating economic privilege. This may allow for more opportunities to address clinical needs, and therefore lower risk for NMPOU.

The current findings, particularly in the context of prior research, have clinical implications. First, our findings coalesce with prior research indicating that there is no one factor in isolation, or even a simple algorithm of factors, that accurately portrays potential for NMPOU.¹⁰ Therefore, research should continue to examine whether patterns of symptoms that cluster together may be useful indicators for clinicians to attend to when monitoring patients for behaviors that may increase risk for adverse outcomes (eg, overdose). Additionally, these patterns of symptoms may inform unmet treatment needs. For example, patients showing patterns such as Class 2 may be in need of mental health treatment, as their opioid use may be in part motivated by attempting to cope with distress.³³ This group may particularly benefit from treatment targeting pain catastrophizing, which research has also found may be associated with increased opioid cravings.^20,34 On the other hand, those in Class 3 may need assistance with physical or occupational therapy, as pain severity and interference may motivate NMPOU behaviors to perform activities of daily living. Finally, there may be different markers of risk for use of NMPOU among different racial groups. For example, Class 2 was primarily composed of white/non-Hispanic individuals, while Class 3 included a relatively higher proportion of Black participants. Though more research is needed, this highlights that focusing on some risk factors over others could lead to biased perceptions of risk.

Results from the current study should be considered in the context of its limitations and directions for future research. First, the current study is cross-sectional; therefore, we cannot speak to the ability to predict later opioid use behaviors. Future research should determine whether clusters derived through LCA can prospectively predict engagement in NMPOU. Second, results are based on patients from a single clinic utilizing a relatively small sample. Future work should be conducted in larger, more diverse samples to ensure generalizability. Third, we did not have the ability to objectively assess opioid use behaviors, such as via urinalysis.⁴² Therefore, some patients may have not reported specific behaviors due to the desire to present in a socially desirable manner. Although most estimates suggest that patient self-report is fairly accurate compared to urinalysis, ⁴² future research should integrate objective assessments to avoid this issue. Fourth, the sample comprises individuals who were referred to and attended a psychological intake session. Participants were not necessarily referred to address mental health problems, as many were referred for an evaluation of suitability for opioid prescribing; however, results from this sample may not generalize to all pain clinic outpatients. Fifth, although we measured substance use, we were unable to assess whether participants met criteria for a SUD. Future research should include measures of problematic substance use outside of opioids. Similarly, some variables, including substance use, were abstracted from templated medical records, and clinicians may have varied in how they assessed these items. Utilizing standardized measures in future research would improve future replicability. Sixth, although the COMM is a validated measure,²⁸ the COMM-SF²⁹ has not been discussed in a peer-reviewed publication, and evidenced low internal consistency in the current study. However, the full COMM²⁸ includes many items that do not directly assess NMPOU (eg, anger, memory problems, suicidal ideation), which may be closely interrelated, but do not directly assess NMPOU. Therefore, results must be considered in the context that more research is needed to validate the psychometric properties of the COMM-SF. Seventh, the research took place over a period of time that included significant societal and environmental changes, including a change in opioid prescribing guidelines in 2016⁴³ and the COVID-19 pandemic. Our sample size does not allow us to analyze whether this pattern of result has changed due to such events, but this would be a valuable direction for future research.

Overall, results from the current study add to the literature suggesting complex patterns of symptoms that contribute to NMPOU among patients with chronic pain. In particular, 2 high-risk groups were identified, one with high probability of clinically significant mental health symptoms, pain catastrophizing, and trauma exposure, and one with high probability of severe pain intensity and interference. Pending replication, future research may use these results to inform the development of future assessments to support effective pain management and mitigate risks associated with opioid use. Additionally, future work should continue to evaluate how to best serve these patients’ diverse and often unmet clinical needs.

Supplementary material

Supplementary material is available at Pain Medicine online.

Funding

The study was supported through the University of North Carolina (UNC) Injury Prevention Research Center (IPRC) under exploration grant (R49CE003092-04). Additionally, work on this study was supported by the National Institute on Drug Abuse (NIDA; K23DA054299-01, PI: N.A. Short). Neither funding source was involved in the study design, data collection, analysis and interpretation of data, writing of the report, or the decision to submit the article for publication.

Conflicts of interest: None declared.

Data availability

Data are not publicly available.