Opioid Use Is More Common in Nonalcoholic Fatty Liver Disease Patients with Cirrhosis, Higher BMI, and Psychiatric Disease

More than 50 million adults in the United States had painful chronic health conditions in 2016 [1]. Prescribing rates of opioids, commonly used in the management of patients with chronic pain, have increased in the United States since 2006 with >70,000 opioid-related overdose deaths in recent years [2]. In 2016, to improve patient safety with pain management, the Centers for Disease Control and Prevention published guidelines providing recommendations for alternative treatment options [3].

Nonalcoholic fatty liver disease (NAFLD) is the most common chronic liver disease worldwide [4], yet little is known regarding the prevalence of pain and associated opioid use in this population. NAFLD is commonly associated with both obesity and metabolic syndrome due to biomechanical changes which increase the mechanical load on the joints and spine, inflammatory mediators which may alter pain modulation, and poor sleep habits [5-7]. In a single center cohort of patients with NAFLD, 30% experienced pain [8]. Furthermore, model for end-stage liver disease (MELD) and Child-Pugh score (CPS) have been positively associated with pain and pain-related disability [8-11].

Currently, there is a paucity of safety data, societal misconceptions, and fear of litigation associated with the treatment of chronic pain in patients with liver disease [12]. Many patients with severe liver disease receive opioid prescriptions for pain due to the increased risk of hepatic and renal toxicities associated with acetaminophen and nonsteroidal anti-inflammatory drug (NSAID) use [8, 9, 13]. Yet, use of opioids among patients with severe liver disease is associated with an increased risk of complications including prolonged hospitalization, hospital readmission, and decreased health-related quality of life [14-16].

Exercise is an alternative nonpharmacologic therapy for pain, which decreases hepatic steatosis among patients with NAFLD and improves mobility, potentially reducing the need for opioids [17-21]. In light of the safety guidelines for opioid use from the Centers for Disease Control and Prevention, there is promising evidence demonstrating that early physical therapy is associated with a reduction in future opioid use [22]. Therefore, it is possible that proactive identification of characteristics of patients at risk for opioid use may encourage health practitioners to consider alternative non-opioid therapies that may improve long-term health outcomes in these patients. This study estimated the real-world prevalence of prescription opioid use among patients with NAFLD enrolled in the TARGET-NASH cohort. Additionally, this study utilized traditional modeling analyses as well as machine learning techniques to develop a patient phenotype at risk for opioid use.

TARGET-NASH is an ongoing longitudinal cohort of adult and pediatric patients with NAFL, nonalcoholic steatohepatitis (NASH), or NAFLD-related cirrhosis managed according to local practice guidelines across academic and community gastroenterology, hepatology, and endocrinology clinics in the United States. A description of the study design has been previously published [23]. Briefly, enrolled patients consent to review of their medical records for 3 years prior to enrollment and are followed prospectively for 5 years. Clinical information from the electronic medical records, including patient narratives, laboratory results, pathology reports, and imaging data are extracted and uploaded into a secured database at 6 month intervals from the date of enrollment. Approvals from central and/or local institutional review boards are obtained prior to subject recruitment and at enrollment.

This analysis included all adult patients (≥18 years old) enrolled in TARGET-NASH between August 1, 2016, and March 4, 2019.

All patients were diagnosed by their treating physician with NAFLD based on liver biopsy and/or pragmatic case definitions. After enrollment, patients were classified as having NAFL, NASH, or NAFLD-related cirrhosis. NAFL was defined as the presence of hepatic steatosis not attributable to other causes (e.g., heavy alcohol consumption) and without evidence of biochemical and/or histological inflammation. NASH was defined as the presence of steatohepatitis on biopsy or both steatosis and elevated alanine aminotransferase levels in a patient with obesity, type 2 diabetes, dyslipidemia, or metabolic syndrome. Cirrhosis was based on clinical determination (i.e., clinical decompensation events, labs, imaging) or liver biopsy (see online suppl. Table 1; for all online suppl. material, see www.karger.com/doi/10.1159/000511074) [23].

Patients with documentation in the medical record of any use of the following medications codeine, buprenorphine, butorphanol, hydrocodone, dihydrocodeine, oxycodone, fentanyl, hydromorphone, methadone, morphine, oxymorphone, tramadol, pethidine, and tapentadol, within 1 year up to or at the time of enrollment were considered “opioid users.” We classified these opioids into abuse potential based on scheduling categories set by the United States Drug Enforcement Administration [24]. Information on the duration of prescription opioid use was not available.

Medication use, medical history, and procedure and laboratory values were ascertained prior to or at enrollment. Psychiatric and medical comorbidities were determined by review of medical records for keywords within clinical notes or relevant medications. For patients with cirrhosis, CPS and MELD were calculated based on labs and clinical assessment at the time of enrollment [25, 26].

The proportion of patients with documented opioid and/or non-opioid analgesic use was calculated and stratified by severity of liver disease. The distribution of opioid use by demographic characteristics, patient comorbidities, reported symptoms (abdominal pain, back pain, headache), severity of liver disease, and decompensation events was also calculated.

A χ² test was used to compare the distribution of categorical variables and an ANOVA was used to compare the distribution of continuous variables among opioid users versus nonusers.

Stepwise logistic regression was used to estimate the association between patient characteristics and the odds of receipt of opioids among all patients with NAFLD and within the strata of NAFLD cirrhosis using variables determined a priori as well as those identified via univariate analyses (significance level for entry = 0.25; significance level to remain in the model = 0.1). Collinear variables were removed, and a forward, stepwise regression model was fit. These analyses were performed using SAS version 9.4 (Cary, NC, USA).

The predicted probability of opioid use at enrollment was estimated using a classification and regression tree analysis (CART) among all patients with NAFLD. A CART analysis is a data driven, nonparametric method of estimating the association between independent variables (clinical characteristics) and the predicted probability of a given outcome (opioid use). In a CART analysis, a decision tree is created using a series of dichotomous splits of patient variables that have a similar probability of opioid use [27]. At each terminal split, the algorithm predicts the outcome by classifying the response that creates the greatest difference in the prevalence of the outcome across the splitting category [28].

CART analysis has several advantages over traditional modeling techniques. First, it allows for multiple comparisons and thus can be used to assess outcomes with many possible predictor variables. Second, it does not make assumptions about the distribution or normality of the modeled outcome and it avoids the need to inform the model with transformations of predictor variables. Third, it provides modeling of the complex interactions between predictor variables, resulting in an estimated probability of use among subgroups of patients with similar characteristics, which may be more clinically applicable.

Cross validation was performed to assess the stability of the estimate. The validation data set was created using 75% of the data as the training set and 25% as the validation set with equal balance of opioid users in each group. The misclassification rate and area under the receiver operating curve were calculated for the final model.

In addition to the CART analysis, boosted trees, which applied the classification algorithm over multiple iterations of classification trees, were used to estimate the variable importance of predictors of opioid use while decreasing the prediction error of the algorithm. The following variables were included in the tree-based analysis: age, sex, race, ethnicity, BMI, recent alcohol use, Alcohol Use Disorders Identification Test (AUDIT) score, history of diabetes mellitus, depression, anxiety, osteoarthritis, rheumatologic diagnosis, dyslipidemia, gastroesophageal reflux disease, fibromyalgia, nephrolithiasis, headache, back pain, abdominal pain, ESRD, bariatric surgery, and any use of anxiolytics, antipsychotics, and antidepressants. All tree-based analyses were performed in JMP SAS (Cary, NC, USA).

The cohort included 3,474 adult patients with NAFLD including 1,054 (30.3%) with NAFL, 1,303 (37.5%) with NASH, and 1,117 (32.2%) with NAFLD cirrhosis. At the time of enrollment, patients’ mean age (±SD) was 56.9 ± 13.0 years, 58.9% were female, 74.5% white, 5.3% black, 12.3% Hispanic, and 83.9% non-Hispanic. The mean BMI was 33.3. ± 7.6 kg/m², 50.5% of patients had diabetes, and 61.7% had dyslipidemia (Table 1).

Among all patients with NAFLD, 56.1% of patients were users of analgesics and 18.0% had documentation of opioid use within a year prior to or at enrollment. Of all opioid prescriptions, 12.2% were DEA schedule II, 1.4% were schedule III, and 7.7% were schedule IV. Opioid use was more common in patients with more advanced liver disease (12.0% NAFL, 17.7% NASH, 23.9% NAFLD cirrhosis, p < 0.001). Opioid use was more often documented among females (20.0%) compared to males (15.1%) (p < 0.001), and opioid users had higher mean BMI (35.4 vs. 33.8 kg/m², p < 0.001) compared to users of non-opioid analgesics. The prevalence of opioid use was highest among black race (23.5%) compared to white (19.9%) and other races (6.5%) (p < 0.001) (Table 1).

Non-opioid analgesic use was documented in 38.1% of patients with NAFLD. Ninety percent of non-opioid analgesic users were users of NSAIDs and 12.1% were users of acetaminophen. The proportion of non-opioid analgesic use was highest among patients with NASH (42.4%) followed by NAFLD cirrhosis (37.5%) (p < 0.001) (Table 2). Patients who were users of non-opioid analgesics were 27.5 and 47.1% less likely to have a history of a mental health diagnosis or back pain, respectively, compared to opioid users (Table 1).

In the multivariable-adjusted logistic regression model, variables associated with opioid use included NAFLD cirrhosis versus NAFL (OR 1.51, 95% CI 1.16–1.98), BMI ≥32 kg/m² (OR 1.29, 95% CI 1.05–1.59), depression (OR 1.87, 95% CI 1.50–2.33), anxiety (OR 1.59, 95% CI 1.27–1.98), osteoarthritis (OR 1.42, 95% CI 1.07–1.89), rheumatologic disease (OR 1.95, 95% CI 1.40–2.70), gastroesophageal reflux disease (OR 1.28, 95% CI 1.03–1.60), fibromyalgia (OR 2.17, 95% CI 1.55–3.04), nephrolithiasis (OR 1.69, 95% CI 1.25–2.29), back pain (OR 2.25, 95% CI 1.76–2.87), and abdominal pain (OR 1.58, 95% CI 1.28–1.95) (Table 3).

The prevalence of splenomegaly (17.2 vs. 10.1%, p = 0.003), hepatic encephalopathy (34.8 vs. 23.4%, p < 0.001), and ascites (43.1 vs. 34.4%, p = 0.010) was significantly higher among opioid users compared to patients without documentation of opioid use. The prevalence of hepatocellular carcinoma was not significantly different among patients with cirrhosis who had documentation of opioid use compared to nonusers (3.4 vs. 2.0%, p = 0.214). Among opioid users and nonusers with NAFLD cirrhosis, there was no significant difference in mean MELD score (10.3 vs. 10.1, p = 0.501) or CPS (8.0 vs. 7.7, p = 0.195) (Table 4).

In the multivariate analysis of opioid use among patients with NAFLD cirrhosis, age, sex, BMI, osteoarthritis, rheumatologic diagnosis, depression, MELD, ascites, HCC, and hepatic encephalopathy were considered in our initial model (Table 5). In our final model, age >62 years (OR 0.70, 95% CI 0.51–0.96), depression (OR 2.29, 95% CI 1.66–3.18), rheumatologic disease (OR 2.53, 95% CI 1.48–4.30), and osteoarthritis (OR 2.30, 95% CI 1.53–3.46) were associated with opioid use (Table 5).

In the boosted tree analysis, a history of fibromyalgia, depression, or back pain in the medical record had the greatest variable importance for predicting opioid use over 100 trees at a sampling rate of 0.75. The area under the receiver operating curve was 0.76 for the learning sample and 0.73 for the validation sample. The misclassification rate was 0.18 for both the training and the validation sample (Fig. 1).

Forty six percent of patients with NAFLD who had a history of depression and back pain were opioid users. The probability of opioid use among patients with a combined history of depression and fibromyalgia without back pain was 44%. Among patients without depression, the probability of opioid use was approximately 20% for patients with anxiety, abdominal pain, or back pain. The misclassification rate in the training sample and validation sample was 0.18 (Table 6).

This cross-sectional study estimates real-world opioid use among patients with NAFLD. Nearly one-fifth of those with NAFLD and one-quarter of patients with NAFLD cirrhosis reported recent or ongoing opioid use. Opioid use was twice as common among those with NAFLD cirrhosis compared with NAFL and was associated with higher BMI, painful comorbidities, and psychiatric disease.

In the CART analysis, which allows for the identification of distinct subgroups of patients with NALFD who are likely to use opioids, both depression and back pain were associated with opioid use across patient subgroups. Approximately half of patients who had both a history of depression as well as a history of fibromyalgia were opioid users. The probability of opioid use among patients with back pain and depression was twice that of patients with back pain who were not depressed.

The results of both analyses, CART and logistic regression, suggested that depression and back pain were significantly associated with opioid use in patients with NAFLD. A logistic regression analysis estimated the average effect for the association between independent variables and the odds of an outcome, while a CART analysis can define clinical subgroups of patient’s that are associated with an increased probability of opioid use. The results of this classification analysis suggest that patients with NAFLD who also suffer from back pain, fibromyalgia, and/or depression might be at particularly high risk of future opioid use and may benefit from proactive alternative therapies to address pain, such as scheduled low-dose acetaminophen or referral to physical therapy and/or a psychologist where appropriate.

Among patients with cirrhosis, the proportion of patients who were users of opioids in our cohort is slightly lower than prior studies reporting a 25–47% prevalence of opioid use among patients with cirrhosis [8, 9, 13, 29]. This may be due to differences in how opioid use was ascertained or a reflection of our “real world” population of patients with NAFLD who were receiving care from both academic and community practices. These clinical practices may have less severe liver disease and fewer comorbidities than cohorts drawn exclusively from tertiary referral centers. In addition, given that our cohort was enrolled recently, this may be a reflection of the national trend toward reduced opioid prescribing [30].

This work supports previously findings in the literature which have suggested that opioid use was more common as liver disease progressed, with the highest reported prevalence of use in patients with NAFLD cirrhosis [8]. This could reflect an appropriate avoidance of NSAIDS given their risk of precipitating acute kidney injury and gastrointestinal bleeding or misinformation about the harms of acetaminophen use in the setting of cirrhosis. However, the higher reported use of non-opioid analgesics, including acetaminophen and NSAIDs, among patients with NASH and cirrhosis compared to NAFL suggests that this observation may be due to an increased prevalence of pain in those with higher BMI or lower levels of physical activity, both of which promote liver disease progression.

Another potential explanation for opioid use among patients with NAFLD is the association between metabolic syndrome and painful complications such as osteoarthritis due to increased mechanical load or diabetic neuropathy. Even after adjusting for BMI, osteoarthritis was significantly associated with opioid use. A lower pain threshold may also be attributable to higher levels of inflammatory cytokines seen in NAFLD [9, 31]. It is possible that mood disturbances, seen more frequently in cirrhosis, and sleep disturbances, a common manifestation of hepatic encephalopathy, could contribute to opioid use even in the absence of pain [32]. In our cohort, fibromyalgia, anxiety, and depression were all associated with opioid use and hepatic encephalopathy was more commonly reported in opioid users.

Opioid use in patients with NAFLD is concerning given the potential role of opioids in sedentary behavior. Large prospective studies have demonstrated that the development of obesity is associated with a higher risk of chronic low back, neck, and shoulder pain and that this effect may be attenuated by exercise [33]. In our cohort, opioids were associated with obesity, suggesting that opioid use may predispose to the development or worsening of NAFLD.

While there are situations in which opioid prescription may be appropriate, particularly for acute pain, there are many reasons why opioids should be used cautiously in patients with NAFLD and cirrhosis [15, 34]. First, opioids have limited efficacy data for the treatment of chronic noncancer pain (>3 months) [34, 35]. Additionally, the use of long-acting opioids for noncancer pain is associated with increased risk for opioid abuse, opioid dependence, opioid overdose, and all-cause mortality compared to no opioid treatments or non-opioid analgesics [34, 36]. Last, opioids have specifically been associated with poor outcomes in patients with cirrhosis including an increase in hepatic encephalopathy, length of stay, and hospital readmission [14, 37, 38].

Given opioids’ limited long-term efficacy data and substantial risks, alternative options for patients with NAFLD and pain should be fully explored [35]. Nonpharmacologic options should be considered first-line options in patients with NAFLD [3]. Exercise, which remains a cornerstone of treatment for NAFLD, has the dual benefits of improving hepatic steatosis in patients with NAFLD and decreasing back pain, joint pain, and symptoms of fibromyalgia [17-19, 39-42]. This may be facilitated through early referral to physical therapy which is associated with reduced future opioid use [22]. Other nonpharmacologic treatment options that deserve additional study in patients with NAFLD and cirrhosis include multidisciplinary rehabilitation, mindfulness-based stress reduction, and cognitive behavioral therapy [34]. Last, treatment for coexisting depression, anxiety, substance use disorders, and sleep disturbances may improve pain and reduce inappropriate opioid use. Acetaminophen (≤2 g/day) should be considered a first-line non-opioid analgesic option in cirrhosis given its safety and efficacy data [43]. Further study is needed to determine alternative analgesics or analgesic combinations that optimize physical function and limit adverse effects in patients with NAFLD and cirrhosis.

Our study has several strengths. It includes a large, real-world cohort of patients with NAFLD and, to our knowledge, represents the largest study to date examining use of opioids in this population. Many prior observational studies of patients with NAFLD have been limited to tertiary centers or large administrative databases, which may be more prone to referral bias, misclassification bias, and limited generalizability [44, 45]. In contrast, our cohort encompasses patients with NAFLD diagnosed by histology and clinical data, rather than billing codes, and includes patients from both community and academic settings. Therefore, our estimates of opioid use are likely to more accurately reflect patients with NAFLD receiving care in usual clinical practice.

There are a few limitations of this work. The cross-sectional design of this study does not allow one to determine the temporal relationship between patient characteristics and current opioid use. Prospective studies, including the TARGET-NASH cohort, will help to shed light on the use of opioids in patients with NAFLD and the potential for associated complications. Moreover, the definition of opioid use is based on reports of a prescription for opioids at or within a year of enrollment and not on pharmacy fills. It is possible that patients who were prescribed but never took opioids were misclassified as opioid users and those who took nonmedical prescription opioids were misclassified as nonusers. Our cohort included primarily White and non-Hispanic patients who were actively receiving care in largely gastroenterology and hepatology practices across the United States. Patterns of opioid use may differ among other racial/ethnic groups or patients receiving care in other specialty or primary care clinics. Last, we were unable to differentiate between short-term and chronic opioid use.

Opioid use occurred in 1 of 5 patients with NAFLD and was more common among those who have more severe liver disease, anxiety, or depression. Given the safety concerns associated with opioid treatment, use of nonpharmacologic options and non-opioid analgesics should be considered for the treatment of pain in patients with NAFLD.

TARGET-NASH is sponsored by TARGET PharmaSolutions Inc. We would like to thank the study staff, nurses, health care providers, and participants for their contribution to this study. This is a collaboration of academic and community investigators, pharmaceutical partners, and NASH patient community advocates.

This study was approved by WCG (WIRB Copernicus Group); reference number: 20161381. Clinical information from the electronic medical records, including patient narratives, laboratory results, pathology reports, and imaging data are extracted and uploaded into a secured database at 6-month intervals from the date of enrollment. Approvals from central and/or local institutional review boards are obtained prior to subject recruitment and at enrollment.

A.M. has no conflicts related to this work. S.W. is an employee of TARGET PharmaSolutions. A.S.L. has received research grants from BMS, Gilead, and TARGET PharmaSolutions (to University of Michigan) and is an advisor for TARGET PharmaSolutions. R.F.M. has no conflicts related to this work. H.N.T. has no conflicts related to this work. J.K. has no conflicts related to this work. C.S. is an employee of TARGET PharmaSolutions. B.A.N.-T. has no conflicts related to this work. A.S.B. is a consultant for TARGET PharmaSolutions.

This research was supported in part by grants from the National Institutes of Health, T32 DK007634.

All authors approved the final version of this manuscript. A. Sidney Barritt is the guarantor of this paper. Andrew M. Moon: Study concept and design, interpretation of data, drafting and critical revision of manuscript. Stephanie E. Watkins: Study concept, analysis, interpretation of data, drafting and critical revision of manuscript. Anna S. Lok: Study concept and design, interpretation of data, and critical revision of manuscript. Roberto Firpi-Morell: Study concept and design, interpretation of data, and critical revision of manuscript. Huy Ngoc Trinh: Study concept and design, interpretation of data, and critical revision of manuscript. Justin T. Kupec: Study concept and design, interpretation of data, and critical revision of manuscript. Cheryl Schoen: Data extraction, data analysis, interpretation of the data, and critical revision of the manuscript. Brent A. Neuschwander-Tetri: Study concept and design, interpretation of data, and critical revision of manuscript. A. Sidney Barritt: Study concept and design, interpretation of data, and critical revision of manuscript.