pharmacoepidemiology and drug safety 2015; 24: 885–892 Published online 22 April 2015 in Wiley Online Library (wileyonlinelibrary.com) DOI: 10.1002/pds.3776
Trends in opioid prescribing and co-prescribing of sedative hypnotics for acute and chronic musculoskeletal pain: 2001–2010† Marc R. Larochelle1,2*, Fang Zhang1, Dennis Ross-Degnan1 and J. Frank Wharam1 1
Department of Population Medicine, Harvard Medical School and Harvard Pilgrim Health Care Institute, Boston, MA, USA Section of General Internal Medicine, Department of Medicine, Boston University School of Medicine and Boston Medical Center, Boston, MA, USA
ABSTRACT Purpose Characterize trends in opioid prescribing and co-prescribing of sedative hypnotics at acute and chronic musculoskeletal pain visits from 2001 to 2010. Methods We conducted a repeated cross-sectional analysis of 15 344 visits for acute pain and 19 958 visits for chronic pain in the National Ambulatory Medical Care Survey/National Hospital Ambulatory Medical Care Survey from 2001 to 2010. The primary outcome was receipt of an opioid, and secondary outcomes were co-prescribing of a benzodiazepine or sedative hypnotic (benzodiazepine, muscle relaxant, or insomnia medications). We used multivariable logistic regression to assess temporal trends. Results Between 2001 and 2010, opioid prescribing at acute and chronic musculoskeletal pain visits increased by 50% (10.4% [95%CI 7.9–12.9%] to 15.6% [95%CI 12.5–18.6%]) and 79% (12.9% [95%CI 9.7–16.0%] to 23.1% [95%CI 18.3–27.9%]), respectively. For chronic pain visits, opioid prescribing plateaued between 2006 and 2010, and spline analysis detected a possible 2007 peak at 28.2% (95%CI 21.4–34.9%) of visits. Benzodiazepines were co-prescribed with opioids at 8.1% (95%CI 6.0–10.1%) of acute pain visits and 15.5% (95%CI 12.8–18.2%) of chronic pain visits. Sedative hypnotics were co-prescribed at 32.7% (95%CI 28.9–36.5%) of acute pain visits and 36.1% (95%CI 32.5–39.8%) of chronic pain visits. We found no evidence for decreased co-prescribing of opioids and sedative hypnotics by any of our measures. Conclusions Opioid prescribing for acute and chronic musculoskeletal pain increased from 2001 to 2010, plateauing from 2006 to 2010 for chronic pain visits. Co-prescribing of opioids and sedative hypnotics is common and may represent a target for interventions to improve the safety of opioid prescribing. Copyright © 2015 John Wiley & Sons, Ltd. key words—opioid; sedative hypnotic; benzodiazepine; musculoskeletal pain; pharmacoepidemiology Received 14 July 2014; Revised 11 February 2015; Accepted 27 February 2015
INTRODUCTION Prescription opioid-related overdose mortality has more than quadrupled between 1999 and 2010 in the USA.1,2 This trend was mirrored by a dramatic rise in prescribing of opioids for the treatment of pain. Opioid prescribing doubled from 8% of visits for chronic musculoskeletal pain in 1980 to 16% of visits in 2000.3 In a large California integrated delivery system, prevalent long-term opioid therapy
* Correspondence to: M. Larochelle, Boston Medical Center; 801 Massachusetts Avenue, Second Floor, Boston, MA 02118, USA. Email: marc. [email protected]
† Preliminary results were presented at The Society of General Internal Medicine Annual Meeting in Denver, CO, on 26 April, 2013.
Copyright © 2015 John Wiley & Sons, Ltd.
increased from 2.2% of adult members in 1997 to 3.9% in 2005.4 The epidemic of opioid-related mortality has led some to question whether opioids may now be overprescribed in contrast to their previous underuse.5–7 Opioid prescribing guidelines for chronic noncancer pain warn against co-prescribing with other sedative hypnotics.8–11 Despite this concern, cross-sectional estimates in varied populations indicate that between 17% and 42% of patients prescribed opioids are also prescribed sedative hypnotics.4,12,13 We are not aware of studies examining trends in co-prescribing of opioids and sedative hypnotics over time. Furthermore, data on opioid prescribing trends since 2005 have not differentiated between prescribing for acute and chronic pain; the
m. r. larochelle et al.
latter has received the majority of attention from guidelines and educational efforts. We sought to analyze trends in prescribing of opioids and co-prescribing of sedative hypnotics—benzodiazepines, muscle relaxants, and z-drugs (non-benzodiazepine insomnia medications)—at musculoskeletal pain visits in two nationally representative databases from 2001 to 2010. We hypothesized that trends among patients with chronic musculoskeletal pain would decline because of the focus of prescribing guidelines on such conditions, and we stratiﬁed all analyses by acute versus chronic musculoskeletal pain. We had three aims (i) determine if trends in opioid prescribing changed between 2001 and 2010; (ii) identify trends in co-prescribing of opioids and sedative hypnotics; and (iii) identify patient and visit characteristics associated with prescribing of opioids and co-prescribing of sedative hypnotics. METHODS Data sources and sample We used the National Ambulatory Medical Care Survey (NAMCS) and National Hospital Ambulatory Medical Care Survey (NHAMCS) from 2001 to 2010.14 NAMCS and NHAMCS are annual, national multistage probability samples of ambulatory visits made to physician ofﬁces and hospitals, respectively. We combined NAMCS with NHAMCS visits to hospital outpatient departments to represent outpatient clinic visits. National Ambulatory Medical Care Survey and NHAMCS were designed by the National Center for Health Statistics (NCHS) and are administered by the US Census Bureau. The sampling frame for NAMCS comprises master ﬁles of physicians maintained by the American Medical Association and American Osteopathic Association. For NHAMCS, the sampling frame is all US non-institutional general and short-stay hospitals, excluding federal, military, and veterans administration hospitals. Data collection is carried out by physicians, hospital staff, or Census ﬁeld representatives. Visit-level data include patient demographics, visit characteristics, drugs prescribed, and diagnoses. A detailed description of NAMCS and NHAMCS methodologies is available from the NCHS.14 We examined the principal or most important patientspeciﬁed reason for each visit using the NCHS reason for visit classiﬁcation, which has commonly been used to deﬁne study samples in NAMCS/NHAMCS analyses of pain treatment.3,15 Among patients 18 years or older, we included musculoskeletal pain visits, deﬁned as an NCHS reason for visit code of 1900.X through 1970.X Copyright © 2015 John Wiley & Sons, Ltd.
(with a ﬁfth digit of 0–3), or code 1980.0. We excluded patients with concomitant cancer, fracture, or nephrolithiasis diagnoses because opioid treatment is generally regarded as appropriate for these conditions. We classiﬁed visits as acute if they were categorized as a “new problem” with onset within the last 3 months. We combined visits categorized as “chronic problem, routine” or “chronic problem, ﬂare-up” as chronic. A sensitivity analysis did not detect differences in opioid prescribing for chronic visits categorized as “routine” versus “chronic.” We excluded visits classiﬁed as “pre/post surgery” or for “preventive care” because chronicity could not be determined. We excluded visits with missing data for any covariates. Variables of interest The primary outcome was prescription or continuation of an opioid medication during the visit. We identiﬁed opioid medications using Multum drug classiﬁcation category 060 (narcotic analgesics) or 191 (narcotic analgesic combinations). NAMCS/NHAMCS have listed up to eight medications per visit since 2003; however, only six were included in 2001 and 2002. For consistency, we included the ﬁrst six medications listed in each year. Between 2003 and 2010, undercounting of opioids by excluding the ﬁnal two medications ranged from 0.0% to 0.3% of visits annually. Our secondary outcome was prescribing of sedative hypnotics. There is no consensus deﬁnition of sedative hypnotics, and we applied two deﬁnitions. The ﬁrst included prescriptions for benzodiazepines (Multum category 069). Opioid prescribing guidelines frequently specify benzodiazepines when urging caution in co-prescribing of sedative hypnotics.8–11 The second deﬁnition was prescription for a benzodiazepine, skeletal muscle relaxant (Multum category 178 or 179), or z-drug (zolpidem, eszopiclone, or zaleplon). These drug categories have been included in past analyses of co-prescribing of opioids and sedative hypnotics.4,12,16 We included age, gender, race/ethnicity, payer, region, location of pain, year of visit, whether the patient had visited the clinic previously, and whether the visit was with the patient’s assigned primary care provider (PCP) as variables in descriptive and adjusted analyses. We categorized age into 18–34, 35–49, 50–64, and ≥ 65 years, and selected 50–64 years old as the referent category. In NAMCS/NHAMCS, race and ethnicity are recorded based on the data abstractor’s knowledge of the patient or information recorded in the chart. NAMCS/NHAMCS imputes race using zip code-level census data and visit characteristics for Pharmacoepidemiology and Drug Safety, 2015; 24: 885–892 DOI: 10.1002/pds
trends in opioid and sedative prescribing
visits with missing data. We combined race and ethnicity variables to classify patients as non-Hispanic white, non-Hispanic black, Hispanic, or other races. Data analysis All analyses were stratiﬁed by acute versus chronic musculoskeletal pain. To determine if the distribution of visits between acute and chronic pain changed over time, we used unadjusted logistic regression with year included as a linear term. We calculated annual, unadjusted opioid prescribing rates for musculoskeletal pain visits with Wald 95%CIs. We analyzed trends in co-prescribing of benzodiazepines and sedative hypnotics for visits with opioid prescriptions at 2-year interval (e.g., 2001–2002) to meet NCHS recommended reliability criteria of relative standard errors less than 30%. We developed multivariable logistic regression models with prescription of an opioid as the outcome, and all variables identiﬁed earlier as covariates. To assess trends in opioid prescribing, we ﬁrst included year as a linear term. To determine if there was a change in trend over the 10-year period, we developed spline functions allowing for a single knot for change in trend between 2003 and 2008, requiring at least 2 years before and after a knot.17 We identiﬁed models with a signiﬁcant (p < 0.05) change in trend at the knot. If a change in trend was identiﬁed at more than one knot, we selected as ﬁnal the model with the best ﬁt as determined by Akaike information criterion.18,19 To assess factors associated with co-prescribing of opioids and sedative hypnotics, we developed separate multivariable logistic regression models for benzodiazepine and sedative hypnotic prescribing among patients receiving a prescription for opioids. For trends in co-prescribing of opioids and sedative hypnotics, we assessed linear trends in multivariable models; however, we did not explore spline functions as the analysis was limited to ﬁve 2-year intervals. To explore if missing data may have biased our results, we compared primary and secondary outcomes for patients in the ﬁnal analysis with those excluded because of missing covariates and found no signiﬁcant differences by chi-squared tests. We also reran ﬁnal multivariable regression models including patients with missing covariate data by excluding those covariates (payer, visit with PCP, and patient established at ofﬁce) from the model. Compared with the ﬁnal analysis, effect estimates for remaining predictors changed by less than 15% with no changes in which predictors reached signiﬁcance at α = 0.05. Copyright © 2015 John Wiley & Sons, Ltd.
We used survey analysis procedures in SAS version 9.3 (SAS Institute Inc., Cary, NC, USA) that take into account the complex survey design of NAMCS/NHAMCS. RESULTS We identiﬁed 44 263 outpatient visits for adults with musculoskeletal pain from 2001 to 2010. We excluded 2136 visits with diagnosis of cancer, fracture, or nephrolithiasis, 3491 not categorized as acute or chronic, and 3334 visits due to missing covariate data. Remaining in the ﬁnal analysis were 35 302 visits: 15 344 for acute pain and 19 958 for chronic pain. The estimated total number of visits nationally increased from 68 million in 2001 to 75 million in 2010. Of visits, 47% were for acute pain and 53% for chronic pain, with no change in this distribution over time (p = 0.13). There were several differences in demographic and visit characteristics between patients seeking care for acute versus chronic pain (Table 1). Visits for chronic pain involved patients who were older and more likely to have Medicare or Medicaid versus private insurance. Combining all years, opioids were prescribed at 14.3% (95%CI 13.1–15.5%) of visits for acute pain and 20.8% (95%CI 18.9–22.6%) of chronic pain visits. Between 2001 and 2010, opioid prescribing at acute musculoskeletal pain visits increased from 10.4% (95%CI 7.9–12.9%) to 15.6% (95%CI 12.5– 18.6%; Figure 1). Adjusting for patient demographics and visit characteristics, the increase in opioid prescribing over the decade was statistically signiﬁcant (p < 0.0001), and spline functions did not detect a change in trend during 2003–2008 (p > 0.05 for knot at each year). For chronic musculoskeletal pain visits, opioids were prescribed at 12.9% (95%CI 9.7–16.0%) of visits in 2001, increasing to a possible peak of 28.2% (95%CI 21.4–34.9%) in 2007, followed by a decrease to 23.1% (95%CI 18.3–27.9%) in 2010 (Figure 1). Spline functions found a statistically signiﬁcant decrease in opioid prescribing trend following a knot at years 2006–2008; the model with a knot at 2007 had the best ﬁt and was retained in our ﬁnal model (Table 2). Benzodiazepine prescription was more likely at visits with an opioid prescribed versus those without (adjusted odds ratio [AOR] 2.8 [95%CI 2.0–3.9] for acute pain and AOR 4.3 [95%CI 3.3–5.5] for chronic pain). Similarly, sedative hypnotic prescriptions (benzodiazepine, muscle relaxant, or z-drug) were more likely for visits involving an opioid Pharmacoepidemiology and Drug Safety, 2015; 24: 885–892 DOI: 10.1002/pds
m. r. larochelle et al.
Table 1. Characteristics of musculoskeletal pain-related ambulatory visits, 2001–2010, by acute versus chronic pain
Number of visits, n Estimated national annual visits, millions (SD) Men (%) Age in years, mean (SE) Age group in years (%) 18–34 35–49 50–64 ≥65 Race/ethnicity (%) Non-Hispanic white Non-Hispanic black Hispanic Other Payer (%) Private insurance Medicare (age ≥65 years) Medicare (age