Strategies to Identify Patient Risks of Prescription Opioid Addiction When Initiating Opioids for Pain

Data Synthesis and Analysis

The population incidence of prescription OUD after opioid prescription was estimated by collating data on opioid dependence and abuse from previous reviews on the topic (including a Cochrane Collaboration review).^7,8,16 In brief, data on the incidence of prescription OUD in opioid-naive patients being prescribed opioids for pain was extracted from the studies that met the eligibility requirement for this review (search was not intended to assess incidence of prescription OUD [eAppendix in the Supplement]). Here, summary incidence was calculated using a random-effects estimate from the included studies and performed via Comprehensive Meta-analysis software version 3 (Biostat).¹⁷

For illustrative purposes, where possible, data were extracted and used to calculate likelihood ratios (LRs), sensitivity, and specificity for each individual risk factor identified for the development of prescription OUD. Full details of the search strategy, search results, study quality review, and calculation of diagnostic accuracy estimates are available in the eAppendix in the Supplement. Contingency tables (2 × 2) were constructed to estimate the LRs, sensitivity, and specificity for each risk factor or screening tool. Data were entered into Microsoft Excel spreadsheets (Microsoft Corp) predesigned to calculate the sensitivity, specificity, LRs, and their 95% confidence intervals. When a symptom, sign, or risk factor was assessed in only 1 high-quality study, the LR and 95% confidence interval were reported. When a symptom, sign, or risk factor was assessed in 2 studies, the range of LRs was reported. If a symptom, sign, or risk factor was considered in 3 or more studies, the protocol sought to pool the LR data using separate univariate random-effects meta-analysis. Likelihood ratios greater than 2.5 or less than or equal to 0.5 were considered as potentially clinically useful.

Incidence of OUD in Pain Management

Research to date has found that the incidence of symptoms of prescription OUD varies widely in reviews of studies of patients prescribed opioids, which report that between 0.10% and 34% develop prescription OUD symptoms following an opioid prescription.^7,8,16 This review found that the incidence of prescription OUD symptoms in high-quality (level ≥3) studies varied with the patient population, care setting, and criteria for diagnosis. For instance, prescription OUD symptoms were higher in studies of chronic pain management programs, which involved direct clinical follow-up and screening of all patients, than in database studies that relied on diagnostic codes in administrative data sets (eg, criteria from the International Classification of Diseases, Ninth Revision [ICD-9], ICD-9–Clinical Modification [ICD-9-CM], or Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition, Text Revision [DSM-IV-TR]). In 2 prospective observational studies of patients in chronic pain management programs that used in-person assessments for all patients and used non-ICD or non-DSM explicit criteria (eg, aberrant behaviors, evidence of illicit substances, or abnormal findings such as presence of a nonprescribed opioid), the reported incidence of these indicators was 34% (95% CI, 26%-42%)¹⁸ and 28% (95% CI, 22%-36%),¹⁹ respectively. In contrast, the incidence of prescription OUD symptoms was lower in the 2 included large database studies that relied on physician-reported ICD-9-CM diagnoses to assess opioid abuse and dependence. In these studies, the reported incidence of these diagnoses was 0.10% (95% CI, 0.09%-0.11%)²⁰ among commercially insured persons who received an initial opioid prescription (and who received an opioid abuse or dependence diagnosis within 2 years of filling the opioid prescription [those with OUDs], vs those who did not receive such a diagnosis within 2 years [those without OUDs]), and 3.2% (95% CI, 3.0%-3.3%) among recipients of chronic opioid therapy,²¹ respectively. When subjected to a meta-analysis, across 3 studies (1 study’s data were missing) of patients prescribed opioids for pain, the incidence of prescription OUD was 2.5% (95% CI, 0.15%-30%; I² = 99%).^18,20,21 The heterogeneity was created by 2 very large analyses of insurance databases with low incidence (0.10%²⁰ and 3.2%²¹) that dominated the results of a small prospective cohort study of patients with chronic pain referred to a psychology practice (incidence, 34%).¹⁸ While further research is required, these findings suggest that incidence studies relying on conventional physician reporting that is based on overall judgment (ie, a coded diagnosis from an encounter form) and incomplete surveillance may substantially underestimate the incidence of prescription OUD.

Risk Factors for Prescription Opioid Addiction

Of 1287 identified citations that were screened, 102 were identified as being eligible for full text review, 96 of which were excluded for either not meeting the predefined study quality threshold or other reasons (eFigure in the Supplement). Overall, 6 high-quality articles were included in the qualitative (narrative) synthesis and 4 were included in the quantitative synthesis. Two studies reported on risk factors (eTable 1 in the Supplement) and 2 studies reported on risk prediction tools (eTable 2 in the Supplement). The 4 high-quality studies included in the quantitative synthesis were all retrospective studies including a total of 2 888 346 patients with 4470 cases that met the authors’ definitions of prescription opioid addiction.^18,19,20,21 Three of the 4 high-quality studies included in the quantitative synthesis examined chronic pain,^18,19,21 while 1 study²⁰ included patients “who had at least one opioid prescription claim,” of whom 58.83% also had a diagnosis of chronic pain disorder per ICD-9 criteria (Bryan Cochran, PhD, email communication, March 7, 2019).

Among included studies that met the quality threshold (see quality assessment in eTable 3 and eTable 4 in the Supplement), a history of any pain disorder (LR, 23; 95% CI, 18-29) was associated with increased risk of prescription opioid addiction (Table 1).^20,21 Furthermore, personality disorder (LR, 27; 95% CI, 18-41),²⁰ somatoform disorder (LR, 12; 95% CI, 7.18-18),²⁰ psychotic disorder (LR, 11; 95% CI, 8.5-14),²⁰ and nonopioid substance use disorder (LR range, 4.2-17)^20,21 were associated with the greatest likelihood of subsequent symptoms of prescription opioid addiction (eTable 5 and eTable 6 in the Supplement). Using a conservative estimate of pretest probability of 2.5% prescription OUD, patients with a personality disorder, somatoform disorder, or psychotic disorder would have an elevated posttest probability of 41%, 24%, and 22%, respectively, of developing OUD. While modest LRs were found for any mood disorder (LR, 6.0; 95% CI, 5.8-6.2) and any anxiety disorder (LR, 5.3; 95% CI, 5.0-5.6), only the absence of a mood disorder (negative LR, 0.50; 95% CI, 0.45-0.52) appeared to meaningfully reduce the likelihood of prescription opioid addiction.²⁰ In 1 study, having a previous history of OUD was associated with an increased risk for the subsequent development of prescription OUD following opioid prescribing for chronic pain management (LR range, 17-22).²¹ Among 2 included studies, men (LR range, 1.1-1.4) were not obviously more likely than women to develop prescription OUD.^20,21

Certain opioid prescription characteristics may also be associated with an increased risk of developing OUD. Specifically, a new prescription for any opioid for 30 days or more appeared to place patients at greater risk for prescription OUD when compared with those patients receiving a supply of opioids for less than 30 days (LR range, 3.5-4.9).²¹ This finding was consistent whether the opioids were prescribed alone, in tandem with schedule II short-acting opioids, or in tandem with opioids of another schedule. When patients’ opioid doses increased to greater than 120 morphine milligram equivalents per day (LR range, 3.2-3.4), they had a higher risk of subsequent prescription OUD.²¹ Although the presence of any concomitant psychiatric medication, such as anxiolytics (LR, 7.3; 95% CI, 6.5-8.3), appeared to be associated with greater risk of developing prescription OUD, as shown in Table 1, atypical antipsychotics appeared to be associated with the highest risk (LR, 17; 95% CI, 15-18).

Patients who received an opioid for less than 30 days (negative LR range, 0.95-0.96) or received a dose that did not increase to more than 120 morphine milligram equivalents per day (negative LR range, 0.85-0.85; the LR range is derived from 2 separate databases described in this study) did not have lower risk of prescription OUD.²⁰ Similarly, patients receiving no concomitant psychiatric medications (negative LR ranging from 0.59 for benzodiazepines to 0.93 for anxiolytics) did not have lower risk of prescription OUD (Table 1 and eTable 7 in the Supplement).²⁰

Risk Assessment Tools

While this review identified 31 studies that have evaluated risk assessment tools for prescription OUD, only 2 studies (which examined 5 individual tools) met criteria to be defined as high quality and, thus, were included (Table 2).^18,19 Furthermore, despite risk assessment tools becoming a widespread aspect of the prescribing of opioids in some pain management settings,²² their ability to identify patients at high vs low risk was limited. Among the tools evaluated through high-quality studies, the Pain Medication Questionnaire (PMQ) (cutoff score ≥30) appeared most promising, although its ability to differentiate patients at high risk from those at low risk was poor overall (Table 2). Among the PMQ validation sample, having 30 or more positive answers modestly increased the likelihood of symptoms of prescription OUD (LR, 2.6; 95% CI, 1.4-4.8). However, having fewer than 30 positive answers had marginal utility for identifying those least likely to develop prescription OUD (negative LR, 0.75; 95% CI, 0.60-0.94). In addition to its limited ability to discern patients at high risk from those at low risk and the fact that the PMQ is only useful for patients who have had pain treated in the past, the tool may be cumbersome, requiring approximately 10 minutes to administer. Similarly, no scale proved useful for assessing patients newly presenting with chronic or acute pain. Nevertheless, the PMQ is different from other scales because it assesses patient agreement with statements about their beliefs concerning past pain relief and behaviors in addressing pain, rather than traditional risk factors such as prior substance use disorders or comorbid conditions. Examples of questions include items such as “In the past, I have had some difficulty getting the medication I need from my doctor(s)” and “At times, I take pain medication when I feel anxious and sad, or when I need help sleeping.”

Other risk assessment tools we evaluated, several of which are in common use, addressed traditional risk factors (such as personal and family history of substance use disorder or mental health disorder or diagnosis). These included the Opioid Risk Tool (LR, 1.5; 95% CI, 0.76-2.9), Brief Risk Questionnaire (LR, 1.2; 95% CI, 0.96-1.6), Brief Risk Interview (LR, 1.2; 95% CI, 0.96-1.6), and Screener and Opioid Assessment for Patients with Pain (LR, 1.2; 95% CI, 0.94-1.4) (tools’ characteristics are shown in Table 3 and eTable 2 in the Supplement).^18,19 Furthermore, when their ability to discern patients at high vs low risk was critically assessed using 95% confidence intervals of LRs, none of these risk assessment tools appeared useful (Table 2).

Limitations

This study has limitations. While prior reviews have attempted to describe risk factors or opioid risk screening tools that can be used to classify patients into high- vs low-risk categories, to our knowledge, none have conducted rigorous quality assessments or used LRs as a strategy to assess the diagnostic utility of screening for risk factors or screening tools.^{16,22,29,30,31} Recently, an earlier similar review¹⁶ on this topic has been updated,^32,33 but our review used a more systematic approach and quality assessment of reviewed studies. Specifically, our study quality assessment resulted in the exclusion of most studies included in earlier reviews on this topic.^32,33 Nevertheless, the present review’s limitations highlight the gaps that exist in the literature in this area (eg, the identified heterogeneity among the included studies due to patient population, care setting, and criteria for diagnosis). In fact, the diagnosis of prescription opioid addiction in the context of chronic pain can be challenging,³⁴ and studies included in this review used varying definitions (eg, DSM-III, DSM-IV, ICD-9) for the diagnosis of prescription opioid addiction. Because its definition has not been consistent in the pain literature, we considered a wide definition of prescription opioid addiction based on the definitions used by study authors (eTable 8 in the Supplement). Although the low incidence of OUD found in the large cohort studies^20,21 might suggest that patients with an OUD diagnosis in those studies were more likely to have severe OUD, this explanation requires further investigation.

While it would be hoped that individual patient characteristics and risk screening tools might prove useful, when subjected to critical review, only a few individual patient characteristics (often identified in only 1 unreplicated study) helped identify patients at high risk, and no individual patient characteristics, other than the absence of a mood disorder, or screening tools appeared particularly useful for safely identifying patients at lower risk. Instead, most screening tools, including the commonly used Opioid Risk Tool,^35,36 were based on lower-quality studies or, when test performance was assessed by calculating LRs, demonstrated poor diagnostic performance.