Customize your JAMA Network experience by selecting one or more topics from the list below.
Bohnert ASB, Valenstein M, Bair MJ, et al. Association Between Opioid Prescribing Patterns and Opioid Overdose-Related Deaths. JAMA. 2011;305(13):1315–1321. doi:10.1001/jama.2011.370
Author Affiliations: Department of Veterans Affairs, Health Services Research & Development (HSR&D) Center of Excellence, and Serious Mental Illness Treatment Resource and Evaluation Center, Ann Arbor, Michigan (Drs Bohnert, Valenstein, McCarthy, Ilgen, and Blow and Ms Ganoczy); Department of Psychiatry, University of Michigan Medical School, Ann Arbor (Drs Bohnert, Valenstein, McCarthy, Ilgen, and Blow and Ms Ganoczy); Department of Veterans Affairs, HSR&D Center of Excellence, Indianapolis, Indiana (Dr Bair), and Department of Medicine, Indiana University School of Medicine, Indianapolis (Dr Bair).
Context The rate of prescription opioid–related overdose death increased substantially in the United States over the past decade. Patterns of opioid prescribing may be related to risk of overdose mortality.
Objective To examine the association of maximum prescribed daily opioid dose and dosing schedule (“as needed,” regularly scheduled, or both) with risk of opioid overdose death among patients with cancer, chronic pain, acute pain, and substance use disorders.
Design Case-cohort study.
Setting Veterans Health Administration (VHA), 2004 through 2008.
Participants All unintentional prescription opioid overdose decedents (n = 750) and a random sample of patients (n = 154 684) among those individuals who used medical services in 2004 or 2005 and received opioid therapy for pain.
Main Outcome Measure Associations of opioid regimens (dose and schedule) with death by unintentional prescription opioid overdose in subgroups defined by clinical diagnoses, adjusting for age group, sex, race, ethnicity, and comorbid conditions.
Results The frequency of fatal overdose over the study period among individuals treated with opioids was estimated to be 0.04%.The risk of overdose death was directly related to the maximum prescribed daily dose of opioid medication. The adjusted hazard ratios (HRs) associated with a maximum prescribed dose of 100 mg/d or more, compared with the dose category 1 mg/d to less than 20 mg/d, were as follows: among those with substance use disorders, adjusted HR = 4.54 (95% confidence interval [CI], 2.46-8.37; absolute risk difference approximation [ARDA] = 0.14%); among those with chronic pain, adjusted HR = 7.18 (95% CI, 4.85-10.65; ARDA = 0.25%); among those with acute pain, adjusted HR = 6.64 (95% CI, 3.31-13.31; ARDA = 0.23%); and among those with cancer, adjusted HR = 11.99 (95% CI, 4.42-32.56; ARDA = 0.45%). Receiving both as-needed and regularly scheduled doses was not associated with overdose risk after adjustment.
Conclusion Among patients receiving opioid prescriptions for pain, higher opioid doses were associated with increased risk of opioid overdose death.
The rate of overdose mortality increased sharply in the United States in the past decade and overdose mortality is a pressing public health problem.1 Between 1999 and 2007, the rate of unintentional overdose death in the United States increased by 124%,2 largely because of increases in prescription opioid overdoses.1,3 Achieving a better understanding of the factors contributing to prescription opioid overdose death is an essential step toward addressing this troubling and dramatic increase in overdose mortality.
There is some evidence that higher prescribed doses increase the risk of drug overdose among individuals treated with opioids for chronic noncancer pain.4 Specifically, the risk of drug-related adverse events is higher among individuals prescribed opioids at doses equal to 50 mg/d or more of morphine. The association of opioid prescribing patterns with risk of overdose may vary across groups of patients; opioid treatment recommendations for pain are typically specific to particular subgroups such as those with chronic noncancer pain,5 cancer-related pain, and substance use disorders.6 However, potential subgroup differences in overdose risk related to opioid prescribing have not been examined.
Additionally, the use of supplemental “as-needed” opioid doses to augment regularly scheduled opioid regimens has yet to be examined as a risk factor for overdose. Short-term studies (10-16 weeks) suggest that prescribing an opioid to be taken as needed to supplement a regularly scheduled opioid is an effective strategy for addressing pain exacerbations,7,8 but recent treatment guidelines5,9 concluded that there is no evidence of the long-term safety of this practice. To address these gaps in the evidence base, we examined the relationship between opioid prescribing patterns and risk of opioid-related deaths over 5 years among diagnostic subgroups of patients (chronic pain, cancer, acute pain, and substance use disorders) in a national sample of Veterans Health Administration (VHA) patients.
This study used a case-cohort design.10 In this design, all incident cases in a defined cohort are sampled in addition to a random sample from this same cohort. We included two 5% random samples of VHA patients (one each for fiscal year [FY] 2004 and FY2005), as well as all FY2004 and FY2005 VHA patients who died of an opioid overdose before the end of FY2008 (“cases”). The study sample was further restricted to individuals treated with opioids. Individuals with indicators of palliative care consultations or hospice care in their VHA medical records were excluded. The sample size was 155 434.
For cases, observation time began on the date of the first opioid fill that occurred after the first medical visit within the study period to avoid potential immortal person-time.11,12 Similarly, for all noncases, observation time began on the date of the first opioid fill that occurred after the first medical visit in the year for which the individuals were in the random sample. Observation time ended on the day of death or the end of FY2008, whichever came first.
We used data from the VHA's National Patient Care Database to identify all VHA encounters. Outpatient prescription medication data came from the VHA's Pharmacy Benefits Management Services. We obtained cause of death indicators from the National Death Index,13 which compiles death certificates from all state vital statistics offices and has the greatest sensitivity among population-level sources of mortality data.14 The National Death Index searches for the VHA population have been described elsewhere.15 We used established methods16 to ascertain “true” matches with these 2 definitions of a match: (1) full match on Social Security number (SSN) and sex and match on at least 2 of the 3 parts (day, month, year) of date of birth; or (2) match on at least 7 digits of the SSN plus full match on date of birth, sex, first name, and last name and middle initial when present. More than 99% of deaths among VHA patients had a full match on SSN. Study procedures received approval from the Ann Arbor VA human studies committee, which waived the requirement for informed consent.
Opioid overdose death was determined using National Death Index files and defined as death with an underlying cause-of-death code from the International Statistical Classification of Diseases, Tenth Revision (ICD-10) of X42, X44, Y12, or Y14, in combination with a T-code of 40.2.17 These underlying cause-of-death codes include all deaths due to medications and narcotics that were ruled unintentional (X42 and X44) or indeterminate in intent (Y12 and Y14). Intentional overdoses, which are represented with underlying cause-of-death codes X60 through X69, were not included. The T-code restricts the cases to overdoses due in whole or in part to prescription opioids.
The opioid analgesic medications dispensed by VHA pharmacies during FY2004 through FY2008 were codeine, morphine, oxycodone, hydrocodone, oxymorphone, and hydromorphone. Morphine-equivalent doses were calculated using established methods.18,19 Semisynthetic opioids were not included because they correspond with ICD-10 T-code 40.4, which includes opioids not currently used for pain in VHA (ie, buprenorphine). Methadone (ICD-10 T-code 40.3) was not included in this analysis because of its uniqueness in indications, duration of action, and inconsistent morphine equivalency.20
Maximum daily opioid dose was measured as a time-varying covariate. We used an “as-prescribed” approach,21 which assumes that patients take all prescribed opioids at the maximum dose and on the schedule recommended by their clinicians. Patients who received refills or new prescriptions for the same drug at the same dose and schedule while they still had opioid medication on hand from a prior fill were assumed to have taken the medication from the prior fill, as prescribed, before taking medication from the second fill. However, patients taking opioid medications who received prescriptions for a different dose or schedule or a different opioid medication were assumed to have begun taking the new prescription on the date that it was filled.
Next, each patient's total maximum daily dose for each day of the study observation period was calculated by adding the daily doses of all fills that covered that particular day. The specific daily dose contributed by each fill was determined by dividing the total morphine-equivalent milligrams dispensed in that fill by the number of days supplied. This measurement of dose reflects the maximum daily dose prescribed and not necessarily the actual amount consumed. Morphine-equivalent maximum daily dose was converted into a categorical variable with the values of 0 mg, 1 mg to less than 20 mg, 20 mg to less than 50 mg, 50 mg to less than 100 mg, and 100 mg or more.4 In addition, a time-varying indicator of whether patients were prescribed a regularly scheduled opioid plus a simultaneous as-needed opioid was coded for each day of the study observation period that a patient had at least 1 opioid prescription using the following 3 mutually exclusive categories: 0, only regularly scheduled opioids; 1, only as-needed opioids; or 2, both a regularly scheduled opioid and as-needed opioid prescriptions.
We obtained demographic variables and diagnosis codes from the International Classification of Diseases, Ninth Revision, Clinical Modification, from patient records. Race (white, black, other) and Hispanic ethnicity were determined by VA clinicians in the course of medical care and recorded in the medical records for purposes related to evaluation. All diagnoses reflected whether the patient had been diagnosed with each specific condition in the year prior to the start of observation time. We combined all non–substance use psychiatric disorders into 1 category and combined a number of chronic bodily pains into 1 category. Acute pains were defined using codes for injuries, acute pain, and postsurgery pain. Cardiovascular disease (CVD), chronic obstructive pulmonary disease (COPD), and sleep apnea were combined to capture diseases affecting the cardiovascular and pulmonary systems. For the purposes of subgroup analyses, the “chronic pain” category included all individuals who had a diagnosis for at least 1 of the following: headache, neuropathy, or chronic bodily pains.
We examined bivariate associations of patient characteristics with case status using χ2 tests. We developed Cox proportional hazards models to examine the relationship of opioid prescribing and risk of opioid overdose death adjusting for demographic and clinical characteristics. Multivariable modeling was restricted to periods when individuals were prescribed at least 1 opioid. We used a risk-sets approach and a robust variance estimator for the multivariable modeling.22 We calculated approximated absolute risk differences for significant associations by first estimating the overall rate of opioid overdose among individuals receiving opioid treatment, next multiplying this rate by the hazard ratio (HR), and then taking the difference between these 2 estimated rates.
All individuals were included in modeling when possible. Ten individuals were missing data on age. Analyses were conducted using SAS version 9.2 (SAS Institute, Cary, North Carolina) and Stata version 11.1 (StataCorp, College Station, Texas). The threshold for statistical significance was P < .05. Power calculations for Cox proportional hazards models in Stata indicated that 66 overdose deaths would need to be observed to detect an HR of 2.0 and 191 deaths to detect an HR of 1.5 at 80% power and α = .05.
Of the 1136 individuals who died of a prescription opioid overdose during FY2004 to FY2008 in the source population, 752 were treated with prescription opioids for pain during FY2004 to FY2008. Two of the opioid overdose deaths among individuals who received opioid therapy for pain occurred among the 2112 individuals who had received palliative care or hospice (who were not included in analyses). The source population included 5 719 542 individuals, and 31.92% of individuals in the random sample were treated with opioids for pain but did not receive palliative or hospice care. Consequently, we estimated that the 750 deaths occurred among approximately 1 834 250 individuals in the base population who would have met study criteria if they had been selected for the random sample. We therefore approximated the rate of overdose among individuals treated with opioids to be 0.04%. Opioid overdose decedents were statistically significantly more likely to be middle-aged and white; more likely to have chronic or acute pain, substance use disorders, and other psychiatric diagnoses; and less likely to have cancer (Table 1).
Table 2 reports the unadjusted rate of opioid overdose within the study sample for each clinical subgroup. For 326 of 750 cases (43.5%), opioid overdose death occurred when the maximum prescribed dose was equal to 0. The overdose rate was higher at higher maximum daily doses compared with lower maximum daily doses across all subgroups examined. Of the different dosing schedules (as-needed, regularly scheduled, and both), the periods with the highest unadjusted overdose rate were those when patients had concurrent fills for both as-needed and regularly scheduled opioids. Across all opioid regimens, the unadjusted rate of overdose in the sample was highest in the group with substance use disorders.
The models reported in Table 3 included maximum prescribed dose, dosing schedule, and demographic and clinical characteristics entered simultaneously into Cox proportional hazards models of risk of opioid overdose death. The association of simultaneous as-needed and regularly scheduled opioids was not significant after adjustment. Having as-needed opioids only compared with having regularly scheduled opioids was associated with an increase in risk of opioid overdose among patients with cancer (HR = 2.75; 95% confidence interval [CI], 1.31-5.78; absolute risk difference approximation [ARDA] = 0.072%). For those in the cancer subgroup, maximum prescribed doses of 100 mg/d or more were associated with an HR of 11.99 (95% CI, 4.42-32.56; ARDA = 0.45%). The HRs associated with the same comparison were as follows: among those with substance use disorders, HR = 4.54 (95% CI, 2.46-8.37; ARDA = 0.14%), among those with chronic pain, HR = 7.18 (95% CI, 4.85-10.65; ARDA = 0.25%), and among those with acute pain, HR = 6.64 (95% CI, 3.31-13.31; ARDA = 0.23%).
We repeated the models substituting a continuous variable (where 0 = dose of 1 to <20 mg/d, 1 = dose of 20 to <50 mg/d, etc) in place of the categorical variable (eTable 1). The estimates for this variable for each subgroup were as follows: chronic pain, HR = 1.96 (95% CI, 1.74-2.21); cancer, HR = 2.42 (95% CI, 1.76-3.34); acute pain, HR = 1.97 (95% CI, 1.58-2.46); and substance use disorders, HR = 1.69 (95% CI, 1.38-2.07). These findings indicate a dose-response relationship between maximum daily prescribed dose of opioid and risk of opioid overdose death. We further modeled risk of death by opioid overdose within the subgroups of chronic pain, cancer, and acute pain stratified by substance use disorder status. The HRs of the dose categories 50 mg/d to less than 100 mg/d and 100 mg/d or more compared with 1 mg/d to less than 20 mg/d were elevated for all subgroups except in the subgroup of individuals with acute pain and a substance use disorder, for which the HR estimates were not statistically significant (eTable 2).
Dunn et al4 found that risk of drug-related adverse events among individuals treated for chronic noncancer pain with opioids was increased at opioid doses equivalent to 50 mg/d or more of morphine. Our analyses similarly found that the risk of opioid overdose increased when opioid dose was equivalent to 50 mg/d or more of morphine. The present study also extended prior research in several important ways. We used a large, national sample of individuals and focused exclusively on opioid overdose deaths. Because the circumstances that lead to overdose death may vary by the condition for which the opioid is prescribed and substance use disorder status, we conducted analyses for subgroups of patients, including those with cancer, acute pain, and substance use disorders.
The present study also extended the prior research by exploring a novel risk factor, ie, concurrent prescriptions of regularly scheduled and as-needed opioids. We found that those patients who were simultaneously treated with as-needed and regularly scheduled opioids, a strategy for treating pain exacerbations,7,8 did not have a statistically significant increased risk of opioid overdose in adjusted models. Recent treatment guidelines have indicated that the long-term safety of this strategy for pain exacerbation has not been established,5,9 and in the present study we did not find evidence of greater overdose risk associated with this treatment practice after accounting for maximum daily dose and patient characteristics.
The analyses of opioid risk among subgroups of patients resulted in several novel findings. Prior work on overdose risk has excluded patients with cancer.4 The present findings indicate that although the overall rate of overdose is lower among patients with cancer compared with other patients, there was a statistically significant association of prescribing patterns with overdose risk among patients with cancer receiving opioid therapy. Renal and liver impairment can interfere with the metabolism of opioids, and high doses of opioids may place patients with cancer at risk for overdose. Patients with cancer were also at an increased risk of overdose if they were prescribed as-needed opioids alone compared with being prescribed regularly scheduled opioids alone, adjusting for total dose. Pain for patients with cancer can be particularly rapid in onset, unpredictable, and severe,23 and taking opioid doses as needed may result in high doses being taken without the benefit of tolerance developed through a regularly scheduled opioid.
Despite “double effect,” which allows use of high doses of opioids for patients in palliative care settings regardless of potential risk (without intention to hasten death),24 only 2 deaths were observed among individuals who had palliative or hospice care. This may be the result of misclassification of cause of death, but this finding provides some preliminary evidence that overdose death is not common among patients in palliative care. Patients in these settings may be better monitored in a way that mitigates the risk of overdose.
There were several aspects of opioid-taking behavior not examined in the present study but that may further explain risk for opioid overdose. Patients may obtain opioids from non-VHA medical settings (eg, through doctor shopping25) or from individuals who are not clinicians. Patients may save medications and take them later in large doses for pain relief or recreationally, particularly when a dose change occurs before the end of a fill or with as-needed doses. These factors may play a particularly important role in opioid overdose among patients with substance use disorders and in explaining those overdoses that occurred when maximum prescribed dose was equal to 0 in the present study.
The present findings highlight the importance of implementing strategies for reducing opioid overdose among patients being treated for pain. McLellan and Turner6 recommended the following: ascertaining patient history of substance use and dependence, using treatment contracts, and scheduling frequent follow-up visits that include a urine toxicological screen for patients at special risk. Increasing use of specialty pain care may reduce the burden on primary care clinicians for these frequent follow-up visits.
The present study found large HRs in the association of maximum daily dose with risk of death by opioid overdose. However, the estimated overall risk of opioid overdose among individuals treated with opioids (0.04%) and the approximated absolute risk increases for significant associations, which ranged between 0.072% and 0.45%, were small. Opioid overdose death represents a particularly important outcome, but a rare one, and the findings should be interpreted accordingly.
We were not able to measure actual dose consumed by patients on any given day. Semisynthetic opioids, which may have particularly high risk for overdose,26 were excluded from analysis, and use of nonopioid medications was beyond the scope of this analysis. Some of the deaths may have been suicides misclassified as unintentional overdoses. However, misclassification of suicides as unintentional overdose is uncommon.27 Additionally, the use of medical records likely results in underdetection of medical conditions. Nonetheless, this study used a large national sample linked to mortality data. Although we were not able to measure opioid dose consumed, the method we used to measure opioid dose reflects the clinician's intention for the maximum frequency for taking the medication.
This study documents a relationship between opioid prescribing and opioid overdose in a large, national, prospective cohort of individuals receiving opioid therapy for a variety of medical conditions. The risk of opioid overdose should continue to be evaluated relative to the need to reduce pain and suffering and be considered along with other risk factors.
Corresponding Author: Amy S. B. Bohnert, PhD, University of Michigan Medical School, 4250 Plymouth Rd, Box 5765, Ann Arbor, MI 48109 (email@example.com).
Author Contributions: Dr Bohnert had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Study concept and design: Bohnert, Valenstein, Bair, McCarthy, Ilgen, Blow.
Acquisition of data: McCarthy, Blow.
Analysis and interpretation of data: Bohnert, Valenstein, Bair, Ganoczy, McCarthy, Blow.
Drafting of the manuscript: Bohnert, Bair.
Critical revision of the manuscript for important intellectual content: Valenstein, Bair, Ganoczy, McCarthy, Ilgen, Blow.
Statistical analysis: Bohnert, Ganoczy.
Obtained funding: Bohnert, Blow.
Administrative, technical, or material support: McCarthy.
Study supervision: Valenstein, Ilgen, Blow.
Conflict of Interest Disclosures: All authors have completed and submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest. Dr Bair reported having a 1-time advisory board involvement with Purdue Pharma. No other disclosures were reported.
Funding/Support: This work was financially supported by the Department of Veterans Affairs Office of Mental Health Services, Patient Care Services, and a Health Services Research and Development Service (HSR&D) Career Development Award (CDA-09-204; A.S.B.B.).
Role of the Sponsor: The sponsor had no role in the design and conduct of the study; in the collection, analysis, and interpretation of the data; or in the preparation, review, or approval of the manuscript or the decision to submit.
Additional Contributions: Anna Eisenberg, MPH, Department of Psychiatry, University of Michigan Medical School, provided editorial assistance, for which she was compensated as part of regular employment.
Create a personal account or sign in to: