Trajectories of opioid misuse and opioid use disorder among adults with chronic pain and HIV: an observational study

Abstract

Objectives:

In a longitudinal cohort of patients with HIV and chronic pain, we sought to (1) identify trajectories of opioid misuse and opioid use disorder (OUD) symptoms, and to (2) determine whether prescription opioid dose was associated with symptom trajectories.

Methods:

We leveraged an existing 12-month longitudinal observational study, Project PIMENTO, of persons living with HIV and chronic pain that received care at a hospital system in the Bronx, New York. A quota sampling strategy was used to ensure variability of prescribed opioid use in the recruited sample. Research interviews occurred quarterly and assessed opioid behaviors and criteria for OUD. To describe symptom trajectories, we conducted two separate longitudinal latent class analyses (LCA) to group participants into (1) opioid misuse and (2) OUD trajectories. Finally, we used multinomial logistic regression models to examine the relationship between baseline prescription opioid dose and symptom trajectories.

Results:

Out of 148 total participants, at baseline 63 (42.6%) had an active opioid prescription, 69 (46.6%) met criteria for current opioid misuse, and 44 (29.7%) met criteria for current OUD. We found three opioid misuse and three OUD symptom trajectories, none of which showed worsened symptoms over time. Additionally, we found that higher prescription opioid dose at baseline was associated with a greater OUD symptom trajectory.

Conclusions:

Opioid misuse and OUD were common but stable or decreasing over time. While these results are reassuring, our findings also support prior studies that high dose opioid therapy is associated with greater OUD symptoms.

Introduction

Chronic pain, defined as pain that persists beyond three months or beyond the normal time of healing,¹ is highly prevalent and carries significant burden in persons living with HIV (PLWH). Estimates vary widely, but approximately 25–90% of PLWH report pain lasting 3 months or more, with over 50% reporting their pain as severe.^2,3 Chronic pain can lead to burdensome consequences in PLWH, including low quality of life, functional impairment, poor antiretroviral adherence, and poor follow-up in HIV care.^4–6 Pain in PLWH is multifactorial and is thought to involve the interplay between biological factors associated with viremia, such as HIV neuropathy, augmentation of pro-inflammatory pathways, and osteonecrosis; and psychological and social factors that modify and influence pain experiences.^2,7 Additionally, as PLWH cohorts age, diseases associated with pain in aging, such as osteoarthritis and lower back pain, have become more common.⁸ Treatment of pain in PLWH has historically relied on use of opioids, with about 8–17% of PLWH prescribed long-term opioids, or opioids prescribed for longer than 90 days.^9–11 In fact, PLWH are more likely to receive opioids and more likely to receive high dose opioids, defined as opioid dose greater than 120 milligram morphine equivalents, compared to persons without HIV.^12,13

Based on risks associated with opioid prescribing, increased regulation and scrutiny have altered the paradigm of opioid prescribing.¹⁴ Opioid prescriptions peaked in 2012 and have been steadily decreasing in the ensuing decade, including among PLWH;^15,16 despite this, opioid overdoses have increased exponentially,¹⁷ often resulting from opioid misuse or opioid use disorder (OUD). PLWH are uniquely at risk for opioid overdose, as opioid misuse and OUD are highly prevalent in PLWH. Up to 20% of PLWH endorse behaviors consistent with opioid misuse, and up to 8.8% have been formally diagnosed with OUD, which may be an underestimate of the actual prevalence of OUD.^10,18–21 Nevertheless, these estimates are higher than those in patients without HIV. High prevalence of opioid misuse and OUD in PLWH may be related to high rates of chronic pain and socioeconomic factors, including homelessness and unmet medical and psychiatric need;²² additionally, injection opioid use is a common cause of HIV acquisition.

Long term opioid therapy for chronic pain, and in particular, high dose opioid therapy, are associated with incident OUD.^23–25 However, little data exist to characterize how symptoms of opioid misuse and OUD evolve over time in PLWH with chronic pain, despite its high prevalence. Leveraging an existing longitudinal cohort of PLWH with chronic pain, we examined opioid misuse and OUD over time. Specifically, we sought to (1) identify trajectories of opioid misuse and OUD symptoms over time, and to (2) determine whether prescription opioid dose was associated with trajectories of opioid misuse and OUD symptoms. Given secular trends in the general population, we hypothesized that a high proportion of participants would show increased symptoms of opioid misuse and OUD symptoms over time, and that higher prescription opioid dose would be associated with increased symptoms of opioid misuse and opioid use disorder over time.

Methods

The current study is a secondary data analysis of a 12-month prospective observational longitudinal study (Project PIMENTO) that enrolled adults living with HIV and chronic pain who were patients at Montefiore Medical Center in the Bronx, NY from May 2017 to July 2019. The overarching aim of PIMENTO was to describe patterns of prescription opioid use among PLWH and to examine how prescription opioid use patterns are associated with HIV outcomes. The current analysis extended the primary analysis by (1) identifying trajectories of opioid misuse and OUD symptoms over the course of the 12-months of follow-up and by (2) determining whether prescription opioid dose at baseline is predictive of trajectories of opioid misuse and OUD. The study was approved by the Albert Einstein College of Medicine and Montefiore Medical Center Institutional Review Board through the Office of Human Research Affairs. This study is reported in accordance with the STROBE statement for reporting of observational studies.²⁶

Setting and Participants

Participants with HIV and chronic pain were identified through the electronic medical record of Montefiore Medical Center and recruited via telephone. Inclusion criteria were: 1) HIV infection, 2) age ≥ 18 years, 3) chronic pain from low back pain, osteoarthritis, or neuropathy (ascertained through ICD-10 codes and confirmed during the screening visit), 4) in care, defined by 2 or more visits to Montefiore HIV primary care sites in the preceding 12 months, and 5) proficiency in English. A quota sampling strategy was used to ensure variability of prescribed opioid use in the recruited sample. More specifically, in April 2019 after 120 participants were enrolled, we restricted remaining enrollment to participants who were prescribed opioids in the prior 30 days based on meeting our projected quota of participants not prescribed opioids. Exclusion criteria were: 1) inability to provide informed consent, and 2) malignancy (by ICD-10 code or by self-report).

Data Collection

After written informed consent, research interviews were conducted at baseline and every 3 months for one year. Research interviews were audio computer-assisted self-interviews (ACASI) and assessed sociodemographics, pain characteristics, opioid use and misuse behaviors, and mental health symptoms. Opioid use disorder diagnoses were formally made by trained research assistants using the PRISM-OP-5, a clinical diagnostic interview procedure for diagnosing substance use disorders.²⁷ Opioid prescription data was extracted from the New York State prescription monitoring program and from the Montefiore electronic medical record.

Measures

Prescription Opioid Dose

Prescription Opioid Dose was defined as the mean daily dose prescribed over the preceding 90 days. We calculated dose at baseline using historical data from the New York State prescription monitoring program. We then used CDC conversion tables to obtain a milligram morphine equivalent (MME), including for prescribed methadone.²⁸ Buprenorphine prescriptions were not considered in MME calculations, consistent with CDC guidance.²⁸ The prescription opioid dose measure ranges from 0 with no upper bound.

Opioid Misuse

Opioid Misuse was measured using a subscale of the Current Opioid Misuse Measure (COMM) proposed by Frimerman et al.^29,30 The subscale has been used in prior research to assess the frequency of 3 opioid misuse behaviors: 1) using more opioids than prescribed, 2) using opioids for non-pain symptoms, and 3) borrowing opioids. Response options were scored on a 5-point scale (0 (never), 1 (seldom), 2 (sometimes), 3 (often), and 4 (very often)) and the subscale score is the sum of responses to the three items (from 0 to 12); we used this value for continuous measures of opioid misuse. Additionally, we used a subscale score >0 to reflect the presence of any opioid misuse. While there is no gold standard for assessing opioid misuse,³¹ we selected these three items because they are specific to self-reported opioid use behaviors, when compared to other scales that include items about dose, mood symptoms, and relationships.

Opioid Use Disorder (OUD)

Opioid Use Disorder (OUD) was measured using the PRISM-OP-5, a clinical diagnostic interview procedure for diagnosing substance use disorders.²⁷ The PRISM-OP-5 differentiates prescription OUD from all other forms of OUD by distinguishing symptoms of tolerance and dependence for those on prescription opioid therapy. For dichotomous analyses, we used a cutoff of 2 or greater to reflect any OUD, consistent with at least mild OUD per the DSM-5. In determining OUD status, tolerance and dependence criteria were not considered to be met in patients taking prescription opioids, as per DSM-5 guidelines. For continuous analyses, we used the highest number of DSM-5 criteria reported for either OUD or for heroin use disorder, from 0 to 11.³² For categorical analyses, we used standard categories: 2–3 reflected mild OUD; 4–5 reflected moderate OUD, 6 or more reflected severe OUD.

Sociodemographic variables

Sociodemographic variables measured at baseline included age, gender, race and ethnicity, employment, and disability status.

Pain variables

Pain variables included questions about duration and type of worst pain; subjective measures of pain intensity and interference (PEG score, a measure of pain intensity and interference scored out of 10 with higher scores reflecting more intensity/interference;³³ Brief Pain Inventory (BPI) intensity and interference sub-scales, separate measures of pain intensity and interference scored out of 10 with higher scores reflecting more intensity/interference³⁴), and pain catastrophizing.³⁵

Data Analysis

To describe opioid misuse and OUD symptom trajectories over time, we conducted two separate longitudinal latent class analyses (LCA) to group participants into (1) opioid misuse and (2) OUD symptom trajectories, based on their symptoms in these domains over time. Using Mplus software (Muthén & Muthén, Los Angeles, CA, USA), we developed LCA models fitted to trajectories within each of the two domains. We set each of the trajectory polynomials to be quadratic, because the preliminary analyses showed a better fit to the data as compared to a linear model. All available data were used, even if participants did not complete 12 weeks of follow-up. We applied the full information maximum likelihood approach for missing data in the analysis. To assure finding the maximum of the likelihood function, we used 200 random sets of starting values.

For each domain (opioid misuse and OUD), we started with a one-cluster latent class model that assumed all participants had the same trajectory over time. Subsequently, we fit successive models with increasing numbers of trajectories. To select the appropriate number of trajectories within each domain, we considered statistical model fit, graphical representations of the trajectories, and clinical relevance. Reviews were conducted by the investigators, 2 of whom are internal medicine physicians with long-standing experience treating patients with HIV and chronic pain (HRP, JLS). Statistical model fit was assessed using the value of the Bayesian Information Criterion (BIC), where lower values indicate better fit; the significance of the Lo-Mendell-Rubin Adjusted Likelihood Ratio Test (LMR-LRT), with significant values less than p<0.05 indicating a better fit of the X model compared to X - 1; model entropy, where values closer to 1.0 indicate a higher accuracy of classifying individuals into latent classes; and average classification probabilities for cluster membership. After determining the appropriate number of trajectories within each opioid use domain, we created an indicator variable, which had a value of 1 if the participant had the largest Bayesian posterior probability (BPP) for that cluster and 0 otherwise. The same procedure was used for each of the three opioid use domains.

For the opioid misuse domain, compared to the two-trajectory solution, the three-trajectory solution had a lower BIC (1571 vs 1583), lower entropy (0.784 vs 0.839), was not significantly different (p=0.06), and had lower though overlapping range of average probabilities for class membership (82.9%−100% vs 92.8%−97.2%). For the OUD domain, compared to the two-trajectory solution, the three-trajectory solution had a lower BIC (1574 vs 1619), higher entropy (0.905 vs 0.889), was not significantly different (p=0.065), and had overlapping average classification probabilities for class membership (95.6%−96.2% vs 96.5%−98%). In both instances, we reached consensus on the three-trajectory solutions based on statistical properties and clinical relevance.

Finally, in two separate models, we used multinomial logistic regression models to examine the relationship between baseline prescription opioid dose and opioid misuse trajectory and opioid use disorder trajectory, respectively. Logistic regression models were adjusted for age, gender, and race (black race vs non-black race). We did not adjust for ethnicity because most of the sample is Black or Hispanic and race and ethnicity were co-linear in this sample. Statistical significance for all tests was two-sided at p<0.05.

Results

A total of 148 participants were consented and completed a baseline visit. Of those, 139 (93.9%) completed the 12-month follow-up visit. The median age was 55 (range 27–76) and 84 (56.8%) were female, including transgender female (n=2). Eighty-one (54.7%) primarily identified as Black/African-American and 51 (34.5%) as Hispanic/Latino. One hundred thirty-three (89.9%) reported not working due to disability, being unemployed, retired, or not having stable employment. Ninety-eight (66.2%) were insured by Medicaid, 44 (29.7%) by Medicare, and 2 (1.4%) were uninsured.

Because we oversampled for opioid use during our study, 63 participants (42.6%) had an active opioid prescription at baseline. A small proportion (n=7, 4.8%) were prescribed a long-acting opioid at baseline. Among those prescribed opioids, the mean daily opioid dose was 91.3 morphine milligram equivalents (MME). At baseline, 5 participants were prescribed buprenorphine, 9 participants reported using methadone for OUD, and 1 participant reported being prescribed methadone for pain. Most participants described their worst pain as lower back pain (n=122, 82.4%), and the average duration of worst pain was greater than 10 years. Pain intensity and interference were severe on average; the mean PEG score was 7.4 out of 10.

Lifetime history of substance use disorder was common. Sixty-four (43.2%) met criteria for lifetime OUD, 40 (27%) for lifetime alcohol use disorder, 40 (27%) for lifetime cannabis use disorder, 67 (45.3%) for lifetime cocaine use disorder, and 83 (56.1%) met criteria for a lifetime history of nicotine use disorder. At baseline, 69 (46.6%) reported current opioid misuse and 44 (29.7%) had current OUD (see Table 1 for baseline characteristics).

Table 1.

Participant baseline characteristics (n=148)

Characteristic	n (%)
Sociodemographics
Age, mean (SD)	54.4 (8.5)
Female gender	84 (56.8)
Primary race or ethnicity
White/Caucasian	10 (6.8)
Black/African American	81 (54.7)
Hispanic/Latino	51 (34.5)
Other	6 (5.0)
Employment status
Working (full or part time)	15 (10.1)
Disabled and not working	90 (60.8)
Not working, other reason	43 (29.1)
Health insurance
Medicaid	98 (66.2)
Medicare	44 (29.7)
Other, or uninsured	5 (3.4)
Pain characteristics
Worst pain site, n (%)
Lower back	122 (82.4)
Limb pain	126 (85.1)
Upper back or neck	78 (52.7)
Duration of worst pain, mean (SD), years	10.1 (11.1)
Pain (PEG) score, mean (SD)	7.4 (1.6)
Opioid use
Current opioid prescription, n (%)	63(42.6)
Baseline opioid dose, mean MME (SD)	91.3(108.6)
Current opioid misuse	63 (46.6)
Took more opioids than prescribed	46 (31.1)
Borrowed opioids from someone else	24 (16.2)
Used opioids for non-pain symptoms	28 (18.9)
SUD diagnoses, lifetime
OUD	64 (43.2)
Prescription OUD	53 (35.8)
Heroin use disorder	29 (19.6)
Alcohol use disorder	40 (27)
Cannabis use disorder	40 (27)
Nicotine use disorder	83 (56.1)
Cocaine use disorder	67 (45.3)
Current OUD
No OUD	104 (70.3)
Mild OUD	27 (18.2)
Moderate OUD	10 (6.8)
Severe OUD	7 (4.7)

Opioid misuse symptom trajectories.

Figure 1 presents the three distinct trajectories of opioid misuse over time. These trajectories can be described as: 1) no opioid misuse (N=64), with a mean score of 0 on the COMM opioid misuse subscale over 12 months; 2) infrequent and decreasing misuse (N=48), with a mean score of 0.16–2.02 on the opioid misuse scale over 12 months; and 3) infrequent but consistent opioid misuse (N=36), with a mean score of 2.32–3.03 on the opioid misuse scale over 12 months.

Figure 1.

Trajectories of opioid misuse over the study timeline.

Opioid use disorder symptom trajectories.

Figure 2 presents the three distinct trajectories of opioid use disorder (OUD) symptoms over time: 1) no OUD (N=84), with a mean number of opioid use disorder criteria between 0.01–0.27 over 12 months; 2) sub-threshold OUD (N=52), with a mean number of OUD criteria between 1.06–2.08 over 12 months; and 3) moderate OUD (N=12), with a mean number of OUD criteria of 3.36–5.1 over 12 months.

Figure 2.

Trajectories of opioid use disorder symptoms over the study timeline.

Prescription opioid dose relationship to opioid misuse and opioid use disorder trajectories.

Figure 3 presents mean baseline opioid dose by trajectory. Increased baseline opioid dose was not associated with a linear trend in opioid misuse trajectory (F=1.28, p=0.28), while increased baseline opioid dose was associated with a linear trend in opioid use disorder trajectory (F=10.83, p<0.001). Table 2 presents results of multinomial logistic regression models adjusted for gender, age, and race (black vs non-black). Baseline opioid dose was not associated with opioid misuse group. Baseline opioid dose was associated with OUD group, with greater mean baseline opioid dose associated with trajectories with greater OUD criteria.

Figure 3.

Mean baseline opioid dose by trajectory group.

Table 2.

Logistic regression of baseline opioid dose to trajectory group.

Trajectory Group	AOR (95% CI)	p-value
Opioid Misuse
Group 1 (reference)	-	--
Group 2	1.03 (0.97 – 1.08)	0.341
Group 3	1.04 (0.99 – 1.10)	0.154
Opioid Use Disorder
Group 1 (reference)	-	--
Group 2	1.18 (1.07 – 1.29)	<0.001
Group 3	1.21 (1.09 – 1.34)	<0.001

Discussion

In this study, we sought to examine opioid misuse and OUD over time in a longitudinal cohort of PLWH and chronic pain. To do so, we identified trajectories of opioid misuse and OUD symptoms over time and examined how prescription opioid dose was associated with symptom trajectories. Most notably, we found trajectories were readily distinguishable and opioid misuse and OUD symptoms were stable or decreasing over time in this cohort of PLWH and chronic pain. Additionally, we found that prescription opioid dose at baseline was associated with OUD trajectory group in a dose-dependent fashion; greater baseline dose was associated with higher OUD symptoms during the entire follow-up.

Overall, our findings illustrated that in this cohort of patients with PLWH and chronic pain, current opioid misuse behaviors were common (46.6%) but infrequent and stable or decreasing over time. Overall, the finding that symptoms were infrequent and stable or decreasing is comforting, and supports the assertion that low levels of opioid misuse may have low chance of progressing in PLWH with chronic pain. Our data are consistent with other studies that used a similar misuse construct,³⁰ but the proportion of our sample with misuse behaviors was much higher than in other analyses of PLWH, where estimates and ascertainment methods vary widely and have ranged from 3.3% to 20.5%.^22,36,37 Whether this is due to ascertainment bias or due to secular trends in opioid misuse is unclear, but answering this question should be an important part of future research. Opioid misuse in PLWH has been associated with inadequate antiretroviral adherence and insufficient viral suppression,^22,36 but opioid misuse assessments are not universally agreed upon in research or in clinical settings and, moreover, are infrequently used in clinically practice.^38,39

We found that OUD symptoms were stable over time and a small proportion had moderate OUD symptoms. Though the severity of OUD was relatively low in our sample, the proportion of our sample with OUD (29.7%) was higher than expected. In prior research in a multistate cohort of PLWH, OUD was documented in up to 8.8% of clinical charts.²¹ The substantial difference in these estimates may reflect ascertainment bias of OUD in clinical care, which implies there may be substantial proportions of undiagnosed OUD among PLWH. A recent study supports this assertion, as a manual chart review of patients prescribed opioids showed that about 14% of low-risk patients on LTOT had moderate to severe OUD despite only 2% having been assigned a clinical diagnosis code of OUD.⁴⁰ Additionally, and more concerning, was that opioid dose was associated with higher OUD symptoms. This finding is consistent with previous literature, which has noted that opioid dose is associated greater OUD symptoms and higher risk of incident OUD.^23,25 Nonetheless, we are unable to say definitively in this study if prescribed opioid dose is predictive of OUD symptoms or prescribed opioid dose is a result of the severity of OUD symptoms. Our findings do imply that OUD may be relatively common and, moreover, is likely underdiagnosed in PLWH. This suggests that screening for OUD and treatment for OUD, if indicated, in PLWH may be an important strategy to reduce opioid related morbidity and mortality.

There are several limitations of this study worth mentioning. PLWH in this cohort had a consistent source of medical care and were living in the community as outpatients and were not incarcerated at the time of recruitment. These conditions may have resulted in a cohort with less burden of opioid misuse and OUD than other samples of PLWH and chronic pain. Nonetheless, our findings show a high prevalence of both behaviors compared to prior research, which given the limitations presented may imply greater burden of these symptoms among patients without a consistent source of medical care.

Overall, we found a mixed picture regarding opioid misuse and OUD. On one hand, patterns of opioid misuse and OUD were relatively stable or decreasing over time, providing comforting reassurance of stable risk of misuse or OUD. On the other hand, we found prevalence of misuse and OUD were relatively high and, in the case of OUD, higher than previous diagnosed estimates in similar patient populations. We believe this suggests large proportions of undiagnosed OUD in PLWH and, further, suggests that assessment of opioid misuse and OUD may be an important strategy to reduce opioid related morbidity and mortality.