Trajectories and correlates of opioid prescription receipt among patients experiencing interpersonal violence

Prescription and non-prescription opioid use disorder and overdose is a national public health crisis: an estimated 115 people die every day due to opioid overdose [ 1 ]. Despite efforts to reduce opioid prescriptions, the number of individuals who receive prescriptions each year remains high [ 2 ]. Opioid prescription receipt greatly increases a person’s risk for adverse outcomes, particularly when prescribed long-term; approximately 21–29% of individuals prescribed long-term opioid therapy for chronic pain use them non-medically [ 3 ] and 8–12% develop an opioid use disorder (OUD) [ 4 , 5 ]. A better understanding of factors associated with long-term opioid prescription receipt, particularly among high-risk populations, can help facilitate efforts to prevent OUD. Consistent with other substance use disorder research, evidence shows that rather than a single linear trajectory, opioid use follows multiple trajectories characterized by significant heterogeneity in behaviors [ 6 – 9 ]. Prior research has examined opioid use trajectories among diverse groups of patients including veterans, patients receiving care for HIV, patients enrolled in a Medicaid lock-in program, and individuals receiving treatment for OUD [ 9 – 12 ]. Opioid use trajectories have been shown to vary by amount of use (e.g., low, medium, high) and use patterns over time (e.g., increasing use, decreasing use). Trajectory patterns are associated with numerous factors including demographics, comorbid conditions, substance use history, and social/family conditions. These relationships, however, vary based on the population studied. One population particularly vulnerable to deleterious effects of opioid use is individuals with a history of interpersonal violence. Experiences of interpersonal violence (e.g., intimate partner violence, sexual assault), have been linked to an increased likelihood of prescription opioid use, [ 13 – 15 ] non-medical prescription opioid use, [ 13 , 14 ] OUD, [ 16 – 18 ] and heroin use [ 19 ]. Opioid prescription receipt has also been found to increase odds of OUD in this population. A recent study found that among individuals with a history of interpersonal violence, people who had received an opioid prescription in the past year had a four times greater odds of reporting opioid misuse behaviors compared to those without an opioid prescription [ 20 ]. Evidence also suggests that interpersonal violence may influence opioid use trajectories: two studies showed that physical, sexual, and emotional abuse history influences recovery trajectories for individuals seeking treatment for substance use disorders (including OUD), with higher abuse levels predicting more severe psychiatric and social problems and slower recovery times [ 9 , 21 ]. Although providing some evidence about the role of interpersonal violence in opioid use trajectories, these studies were conducted after treatment for a diagnosed substance use disorder and did not differentiate by type of substance. Thus, our understanding of prescription opioid use patterns among individuals with a history of interpersonal violence remains elusive. Identifying distinct patterns of prescription opioid use among survivors of interpersonal violence and associated characteristics will help us understand who is at risk for escalating opioid use and OUD and provide potential targets for future interventions. The purpose of this study was to examine longitudinal trajectories of prescription opioid use among patients experiencing interpersonal violence. Specifically, we aimed to: Aim 1 . Classify 5-year longitudinal patterns of opioid prescription receipt among patients experiencing interpersonal violence. Aim 2 . Identify sociodemographic and clinical characteristics associated with opioid prescription receipt patterns among survivors of interpersonal violence.

Between December 2009 and June 2019, 4,296 individuals were referred to the Beacon Program for interpersonal violence services. We excluded 2,709 people resulting in a final sample size of 1,587. The primary reason for exclusion was < 1 year of data in the EHR. Cohort characteristics are presented in Table 1 . Our cohort was largely female (87.5%), White, non-Hispanic (55.2%) or Black, non-Hispanic (29.3%), insured (77.3%), and had a mean age of 41.4 years (SD = 16.3). Over half (58.3%) had a current smoking history and almost half (44.8%) had a lifetime history of illicit drug use. Our cohort exhibited a high prevalence of chronic pain (54.6%), substance use disorders (39.0%), and mental health diagnoses (76.9%) during the study period. The most common chronic pain diagnoses were chronic low back pain (33.5%) and chronic pain, not elsewhere classified (30.0%). Alcohol (18.0%) and cannabis (16.1%) related disorders were the most prevalent substance use diagnoses; however, rates were similar across all drug related diagnoses. Notably, 13.0% received an OUD diagnosis after baseline (participants were excluded if OUD diagnosis was present at baseline). The most common mental health related diagnoses included depression (65.5%) and anxiety related disorders (53.9%). The mean Charlson comorbidity index was 1.4 (SD = 2.1), which corresponds to an estimated 10-year survival of 94% [ 23 , 26 ]. a Counts, percentages, and means are unweighted. b Counts, percentages, and means are weighted by patient’s posterior probabilities for belonging to a given class. SD = standard deviation. Table 2 shows the model fitting results for the LGCA. Four-hundred-twenty-three patients (26.7%) did not receive an opioid prescription during the study period and, thus, were not included in the LGCA. Fit indices favored solutions with an increasing number of classes. The BLRT behaved similarly, yielding significant test results comparing a K -class model relative to a K —1-class model with an increasing number of classes before becoming unstable at a 7-class solution. We selected a 6-class model for further analyses based on fit indices, interpretability, class size, and stability. Note . AIC = Akaike information criteria; BIC = Bayesian information criteria; BLRT = Bootstrap Likelihood Ratio Test. Fig 1 shows trajectories of opioid prescription receipt based on estimated probability over time. We characterized the trajectory classes based on beginning probability, ending probability, and degree of change as follows: Never: no opioid prescription receipt (26.7%) Class 1: Start medium, end medium, moderate decrease (4.8%) Class 2: Start high, end high, slight increase (2.0%) Class 3: Start medium, end low, sharp decrease (13.0%) Class 4: Start low, end high, sharp increase (3.2%) Class 5: Start low, end medium, moderate increase (13.6%) Class 6: Start low, end low, stable (36.7%) Most participants clustered around patterns indicative of no or minimal opioid prescription probability (63.4%, Never and Class 6). The remaining classes, which characterize higher probabilities of opioid prescription receipt, demonstrated considerable variability over time. The two classes that began with a medium probability (Class 1 and Class 3), both showed decreases over time. The only class that began with a high probability (Class 2), remained high over time and slightly increased. Two classes that started low (Class 4 and 5), demonstrated increases over time. A number of sociodemographic and clinical characteristics were associated with opioid prescription receipt trajectories ( Table 1 , Fig 2 ). A summary of the key characteristics for each trajectory group is provided in S2 Table . Overall, prescription opioid classes (Classes 1–6) demonstrated higher prevalence for smoking, chronic pain, substance use disorders, mental health disorders, and Charlson comorbidity index compared with the Never class. Never class were also the youngest among all classes. Class 6 (start low, end low) was most like the Never class, demonstrating the smallest standardized mean differences. The largest proportion of White, non-Hispanic patients was seen in the classes with the highest probability of prescription opioids (Class 1, 2, and 4). Males were also more likely to be represented in these classes whereas the highest proportion of females were in Class 3 (start medium, end low, sharp decrease). Black, non-Hispanic patients were more likely to be in low prescription opioid classes (Class 3, 5, and 6) and Hispanic and Other, non-Hispanic patients in the Never class. The largest proportion of insured patients were seen in higher prescription opioid classes (Class 1 and 2). Those with varying insurance status were also more likely to be represented in high prescription opioid classes (Class 2 and 4). Classes 1, 2, and 4, which demonstrated the highest probabilities of opioid use, had the highest proportions of chronic pain diagnoses, substance use disorders, including OUD, and mental health diagnoses. Individuals who clustered around Class 4 and Class 5 both started with a low probability of opioid prescription; however, Class 4 demonstrated a sharp increase in the probability of opioid prescription over time whereas Class 5 demonstrated a more moderate increase. We compared these two classes to identify factors that may contribute to the increased opioid prescription trajectory for Class 4. Both classes had similar demographics except Class 5 had a higher proportion of Black, non-Hispanics. Class 4 had a higher prevalence of chronic pain diagnoses (except for interstitial cystitis/bladder pain syndrome and migraine) and mental health disorders. Class 4 also had a higher prevalence of substance use disorders. Class 1 and Class 3 both started with a medium probability of prescription opioid use; Class 3 demonstrated a sharp decline over time and Class 1 demonstrated a more moderate decrease. We compared these two classes to identify factors that may contribute to the sharper decline in opioid prescription trajectory for Class 3. Both classes had similar demographics except Class 3 had a higher proportion of females. Class 3 had a lower prevalence of most pain diagnoses (except endometriosis and interstitial cystitis/bladder pain syndrome), substance use disorders, and mental health diagnoses compared to Class 1.

Results from this exploratory study suggest that people with exposure to interpersonal violence are at high risk for prescription opioid use (>70%) and OUD (13%). The different trajectories may help target interventions to people with interpersonal violence exposure who are at the highest risk for OUD (trajectories 1, 2, and 4). The high rates of co-morbid conditions among trajectories with higher probability of prescription opioid receipt over time highlight the importance of monitoring for co-morbidities among patients receiving care for chronic pain and the need for improved trauma-informed pain management strategies.

We conducted a secondary analysis of electronic health record (EHR) data available from January 2004–August 2019 for a cohort of patients referred for interpersonal violence services between December 2009–June 2019 at the University of North Carolina (UNC) Health System (N = 1,587). We used latent class growth analysis (LCGA) to estimate trajectories of opioid prescription receipt over a 5-year period. We then examined associations between trajectory patterns and sociodemographic and clinical characteristics. This study was approved by the University of North Carolina Institutional Review Board (#20–0988). The study population consisted of all adult patients (≥ 18 years of age at baseline) referred to the UNC Hospitals Beacon Program for interpersonal violence services between December 2009–June 2019 with at least one year of data in the EHR during the study period. The Beacon Program provides comprehensive, coordinated care to UNC Health System patients, families, and employees experiencing a interpersonal violence, including intimate partner violence, human trafficking, sexual assault, and elder/vulnerable adult abuse. Individuals were excluded if they had evidence of an OUD at baseline (diagnosis or buprenorphine, methadone, or naltrexone prescription), had a cancer diagnosis, or were receiving end-of-life care during the study period as opioid prescribing practices differ for these conditions. All data analyzed in this study were electronically abstracted from the UNC Health System EHR databases. This included data related to opioid prescription receipt and sociodemographic and clinical characteristics. Given that experiences of interpersonal violence are often chronic rather than an isolated event, we did not limit data extraction based on date of referral to the Beacon Program, but rather extracted relevant variables across the study period (January 2004–August 2019). We examined prescription drug records for opioid medications prescribed during the study period. Then, we recorded all opioid prescriptions during the 60 months following baseline. We focused on the following opioids commonly prescribed for pain: butorphanol, codeine, dihydrocodeine, fentanyl, hydrocodone, hydromorphone, levorphanol, meperidine, morphine, opium, oxycodone, oxymorphone, pentazocine, propoxyphene, tapentadol, and tramadol. Dosage and duration information was not readily available from the EHR data, especially prior to April 2014; thus, we could not calculate morphine milligram equivalents for prescriptions. Therefore, outcome variable was defined as a binary indicator of any prescription opioid receipt for each month of the study period. Demographic characteristics were assessed at the beginning of the study period and included age, race, ethnicity, and sex. We examined insurance status across the study period, categorizing patients as insured, not insured, or varying status. We used International Classification of Diseases 9 th and 10 th edition (ICD-9 and ICD-10) codes to identify medical diagnoses during the study period. Diagnoses included chronic pain conditions, substance use disorders, and mental health conditions. We calculated the Charlson comorbidity index, a method of categorizing comorbidities based on ICD codes, for each patient [ 22 , 23 ]. Lastly, we collected information on social history, namely, smoking and illicit drug use history. Social history is routinely screened for by providers within the UNC Health System and documented within each patient’s EHR. Specific information on diagnostic codes used to define characteristics is presented in S1 Table . LCGA was used to examine opioid prescription receipt over a 5-year time period and identify groups with similar trajectories (Aim 1). Baseline was defined as date of first opioid prescription during the study period (January 2004–August 2019). Those who did not have an opioid prescription during the study period were placed in a “Never” class. We fit models with 2 to 7 groups to determine the optimal number of trajectories. Model selection was based on a combination of model fit, adequacy, and interpretability. We assessed fit using Akaike’s information criterion (AIC), [ 24 ] Bayesian information criterion (BIC), [ 25 ] bootstrap likelihood ratio test (BLRT), and entropy. The LCGA was fitted using MPlus, version 8.4. To estimate the association between sociodemographic and clinical characteristics with trajectory groups, we estimated the prevalence of patient characteristics within each latent class. The analysis was weighted using patient’s posterior probabilities for belonging to a given class. We estimated and graphed standardized differences to assess the degree of similarity and dissimilarity between classes within a given covariate (“Never” class used as reference group). The analysis was conducted in Stata SE, version 6.1.

(DOCX) Click here for additional data file. (DOCX) Click here for additional data file.