Understanding opioid overdose characteristics involving prescription and illicit opioids: A mixed methods analysis

Drug and Alcohol Dependence 167 (2016) 49–56

Contents lists available at ScienceDirect

Drug and Alcohol Dependence journal homepage: www.elsevier.com/locate/drugalcdep

Full length article

Understanding opioid overdose characteristics involving prescription and illicit opioids: A mixed methods analysis Bobbi Jo H. Yarborough a,∗ , Scott P. Stumbo a , Shannon L. Janoff a , Micah T. Yarborough a , Dennis McCarty b , Howard D. Chilcoat c , Paul M. Coplan c , Carla A. Green a a

Kaiser Permanente Northwest Center for Health Research, 3800 N Interstate Ave, Portland, OR 97227, USA Department of Public Health & Preventive Medicine, Oregon Health & Science University, 3181 S.W. Sam Jackson Hill Road, CB 669, Portland, OR 97239, USA c Purdue Pharma, L.P. One Stamford Forum, Stamford, CT 06901, USA b

a r t i c l e

i n f o

Article history: Received 23 November 2015 Received in revised form 12 April 2016 Accepted 21 July 2016 Available online 1 August 2016 Keywords: Opioid analgesics Overdose Risk Prevention Intervention

a b s t r a c t Background: Opioid abuse and misuse are significant public health issues. The CDC estimated 72% of pharmaceutical-related overdose deaths in the US in 2012 involved opioids. While studies of opioid overdoses have identified sociodemographic characteristics, agents used, administration routes, and medication sources associated with overdoses, we know less about the context and life circumstances of the people who experience these events. Methods: We analyzed interviews (n = 87) with survivors of opioid overdoses or family members of decedents. Individuals experiencing overdoses were members of a large integrated health system. Using ICD codes for opioid overdoses and poisonings, we identified participants from five purposefully derived pools of health-plan members who had: 1) prescriptions for OxyContin® or single-ingredient sustainedrelease oxycodone, 2) oxycodone single-ingredient immediate release, 3) other long-acting opioids, 4) other short-acting opioids, or 5) no active opioid prescriptions. Results: Individuals who experienced opioid overdoses abused and misused multiple medications/drugs; experienced dose-related miscommunications or medication-taking errors; had mental health and/or substance use conditions; reported chronic pain; or had unstable resources or family/social support. Many had combinations of these risks. Most events involved polysubstance use, often including benzodiazepines. Accidental overdoses were commonly the result of abuse or misuse, some in response to inadequately treated chronic pain or, less commonly, medication-related mistakes. Suicide attempts were frequently triggered by consecutive negative life events. Conclusions: To identify people at greater risk of opioid overdose, efforts should focus on screening for prescribed and illicit polysubstance use, impaired cognition, and changes in life circumstances, psychosocial risks/supports, and pain control. © 2016 Elsevier Ireland Ltd. All rights reserved.

1. Introduction Though use of prescription opioids may be leveling off in recent years (Dart et al., 2015; Frenk et al., 2015), heroin use (Compton et al., 2016) and opioid-related overdoses have been increasing

∗ Corresponding author at: Center for Health Research, Kaiser Permanente Northwest, 3800 N Interstate Avenue, Portland, OR 97227-1110, USA. E-mail addresses: [email protected] (B.J.H. Yarborough), [email protected] (S.P. Stumbo), [email protected] (S.L. Janoff), [email protected] (M.T. Yarborough), [email protected] (D. McCarty), [email protected] (H.D. Chilcoat), [email protected] (P.M. Coplan), [email protected] (C.A. Green). http://dx.doi.org/10.1016/j.drugalcdep.2016.07.024 0376-8716/© 2016 Elsevier Ireland Ltd. All rights reserved.

(Centers for Disease Control and Prevention (CDC), 2009, 2011; Chen et al., 2014; Johnson et al., 2013; Warner et al., 2009, 2011), as have emergency department (ED) visits (CDC, 2010b; Substance Abuse and Mental Health Services Administration (SAMHSA), 2013a), and health system costs associated with opioid overdoserelated ED and inpatient visits (Yokell et al., 2014). Moreover, demographic shifts in opioid use, overdose rates, and ED visits are also occurring, redefining at-risk populations. For example, the typical individual entering substance use treatment for heroin use is a young, white, middle-class male or female from a rural or suburban community (Cicero et al., 2014). Rates of opioid-related ED visits and deaths have increased among younger (Dasgupta et al., 2014; Hasegawa et al., 2014; Jones and McAninch, 2015),

50

B.J.H. Yarborough et al. / Drug and Alcohol Dependence 167 (2016) 49–56

non-Hispanic White, non-Hispanic Black, and Hispanic (Jones and McAninch, 2015), men and women (Hasegawa et al., 2014; Jones and McAninch, 2015). Despite the public health significance of this problem (Centers for Disease Control and Prevention, 2010a; SAMHSA, 2013b), existing epidemiologic and observational studies of opioid-related overdoses are limited in that they frequently classify drugs and events in categories that make it difficult to disentangle characteristics that have very different clinical and prevention implications. For example, individuals with prescribed and illicit opioid-related overdoses are often combined in the same category, while accidental overdoses are combined with those associated with suicidal intent (Johnson et al., 2013). Moreover, these reports typically leave unaddressed the specific circumstances surrounding overdose events themselves, or the lives of the people who experience those overdoses. Increasing the depth of our understanding about opioid overdose events may help clinicians and public health professionals to identify risk and intervene to prevent overdoses and overdose deaths. As part of a larger study of rates of opioid-related overdoses following the introduction of OxyContin® with abuse-deterrent properties, we examined opioid overdoses among members in a large integrated health system. Using electronic medical records (EMR), we identified overdoses among members with and without active opioid prescriptions, completed chart audits, and conducted in-depth interviews with individuals, or family members of individuals, who experienced overdoses. Analyses reported here are based on all three data sources. 2. Material and methods 2.1. Setting This study was conducted in Kaiser Permanente Northwest (KPNW), a private, not-for-profit group-model integrated health plan which served about 500,000 members in the Pacific Northwest at study start. KPNW maintains an EMR system that captures most aspects of members’ health care (i.e., inpatient and outpatient medical, mental health, and addiction medicine encounters, imaging, labs, prescriptions, and external insurance claims). The KPNW Institutional Review Board reviewed and approved all study procedures. All interview participants provided written informed consent prior to enrollment. 2.2. Case identification process To select interview candidates, we used diagnostic codes indicating opioid overdoses and poisonings (e.g., 965.xx, E850.xx, X42) combined with pharmacy dispense records. We identified family members of decedents, when possible, by looking for other members in the same health plan subscriber unit. We reviewed pharmacy records to determine if a member/decedent had an active opioid prescription at the time of the overdose, and then randomly sampled participants from five purposefully derived groups of individuals, stratified on gender. The primary goal of the larger study was to assess the effects of the abuse-deterrent formulation of OxyContin® on overdoses, so patients with OxyContin® prescriptions were of particular interest as a group. We grouped the remaining patients by drug class or no active prescription. When individuals had more than one active opioid we categorized opioids according to a hierarchical structure so that only one active opioid medication category was associated with each overdose event. The 5 groups (in descending hierarchical order) included individuals with: 1) prescriptions for OxyContin® or single-ingredient sustained-release oxycodone, 2) oxycodone

single-ingredient immediate release, 3) other long-acting opioids, 4) other short-acting opioids, or 5) no active opioid prescriptions. Individuals in groups 1 and 5 were oversampled to ensure adequate representation; individuals in groups 2–4 were sampled proportionally, based on the number of total overdose events identified in each category. 2.3. Chart audit process As part of the larger study, we conducted chart audits to assess the validity of using EMR diagnoses to accurately identify and categorize opioid-related overdoses. Chart components reviewed included history, clinical and telephone encounters, discharge summaries, medication activity reports, and other related documentation. When data were unavailable in the EMR we reviewed associated external billing/claims data. Using a chart audit form, we documented the causal opioid(s), concomitant medication(s), contributing alcohol or illicit drug use, prescribed dose and frequency, source and route of administration for each substance, indication of abuse or misuse for each substance and, if available, any indication of suicidal intent. The first 10% and an additional randomly selected 10% of all charts were examined for quality assurance to maintain high inter-rater reliability (>95%). All chart review forms were double data-entered and verified to ensure accurate data entry. We selected interview participants from the pool of individuals whose charts were audited. 2.4. Recruitment process We mailed recruitment letters and followed up by telephone, inviting participation in a one-time interview with a $50 gift card as compensation. We addressed letters “To the family of [member name]” for deceased health plan members who had no other members associated with their subscriber unit. These letters were mailed to the deceased member’s last address. We mailed recruitment letters to 366 patients or family members, beginning in May, 2012. We were unable to reach nearly half (47%) of those to whom we sent letters: 45 letters were sent to the last known address of the deceased with no response, 66 had no available phone number or message service, and 60 were left messages with no contact or response. Two potential participants were deemed ineligible, one was 13 years old, another was not an overdose event. When recruitment ended 32 cases had not yet been contacted. Among the 203 we attempted to recruit and were able to reach, 35% refused. The most common reasons for refusal were: no reason given (n = 29), lack of interest in study participation (n = 18), logistical/scheduling difficulties (n = 11), denial of overdose event (n = 8) or lack of ability/interest in retelling it (n = 4), criminal investigation (n = 1). We completed 90 interviews (44% participation rate among those reached), 3 (described below) were later excluded. 2.5. Interview process We interviewed participants about identified overdoses during a single one-hour session. The interview guides (see Supplemental material) included open-ended questions regarding substance use/abuse history, source of substance(s), routes of administration, medical treatment following events, whether or not any substance-abuse treatment was received, and changes made to opioid treatment plans. Interviews were audio recorded and transcribed verbatim. 2.6. Data analysis process Chart abstraction data were linked to coded interview data, adding to and filling in missing data. For example, if charts did

B.J.H. Yarborough et al. / Drug and Alcohol Dependence 167 (2016) 49–56

not indicate suicidal intention and interviewees indicated the event was a suicide attempt, that information was included in the final analysis. We computed univariate and bivariate statistics using SPSS v.22 (IBM). For non-responder analyses we compared interview participants with non-responders by t-test (means) or by chi-square tests (categorical variables) to assess differences between groups. We tested differences between unintentional and intentional overdoses on several categorical variables (e.g., gender, presence of heroin at event) using chi-square tests. For qualitative analyses, our team reviewed a sample of interview transcripts, developed and refined a coding scheme, and coded additional interviews using Atlas.ti (Friese, 2011). We used constant comparative methods (Glaser and Strauss, 1967) as interviews were completed, and explored additional areas more deeply in interviews as a result of the coding and analysis process. We performed check coding on a subset of transcripts (16%), and found an agreement rate of 90% between coders. When coding was complete, we generated queries for specific codes, and members of the analysis team reviewed each query, identifying patterns and developing and refining themes using a modified grounded theory approach ˜ 2009; Strauss and Corbin, 1998). (Glaser and Strauss, 1967; Saldana, As themes emerged, we sought confirming and disconfirming cases and ultimately selected the most illustrative quotes within each theme. Early during the coding process we recognized the need to characterize the difficult life circumstances faced by those with overdose events. To capture these difficulties systematically, and to describe commonalities, coders rated the presence and severity of difficulties associated with various life circumstances, including: ability to remember the event; insight into causes; family, social, financial, medical and health circumstances; chronic pain; substance abuse/misuse; mental health; psychiatric medications; negative life events; downward spirals prior to events; and overall order/disorder in each person’s life. We coded “negative life events” when participants reported negative experiences that immediately preceded, and were reported to contribute to, overdoses (e.g., bankruptcy, divorce). We coded “downward spiral” when individuals reported a series of cascading difficulties prior to overdoses. “Life in disorder” was applied when individuals experienced multiple problems, including unemployment, unstable housing, variable mental health, or other significant life problems at the time of the overdose. Ratings were completed on a Likert-type scale ranging from one (person severely impaired by circumstance) to five (circumstance not present/problematic). We converted ratings to a binary indicator of presence/absence of a significant problem that contributed to the overdose. When describing these life circumstances, we report direct quotes, or brief case summaries when direct quotes failed to convey the complexity of the relevant information or were not parsimonious.

3. Results 3.1. Participants We reached saturation on primary themes after conducting 90 interviews between June, 2012 and February, 2014. We eliminated 3 narratives from analyses as they did not meet criteria for overdoses. The remaining 87 participants are described in Table 1. In an analysis assessing selection bias, patients who completed interviews were more likely to be from the oxycodone immediate release group and less likely to be from the other short acting opioids group. There were no differences between the responders and non-responders in the other medication groups. Responders were more likely to have mental health diagnoses and live in rural areas and less likely to be Hispanic and live in suburban areas than

51

non-responders. There were no differences between responders and non-responders based on age, gender, race, Medicaid or Medicare status, substance abuse diagnoses, or urban and large town categories. Because this was a purposively-derived, non-random sample, we report raw numbers rather than proportions throughout the results so as not to imply that any event characteristic occurred at a certain rate within the larger event-based dataset. 3.2. Opioid overdose events On average, 1.3 prescription opioids were involved in each of the 87 events. Adding heroin raised the average number of involved opioids to 1.5. The average number of substances involved in overdoses was 3.0, with benzodiazepines the most common substance in addition to opioids (involved in 28 events). Thirty-two participants had an active prescription for at least one benzodiazepine at the time of the overdose and prescribed benzodiazepines were involved in 21 events. The remaining seven individuals did not have prescriptions. Other drugs frequently present included: mood stabilizers (n = 22), alcohol (n = 17), marijuana (n = 12), sedatives/hypnotics other than benzodiazepines (n = 11), trazodone (n = 11), and atypical antipsychotics (n = 10). Muscle relaxants (n = 6) and other medications known to interact with opioids were present; so too, though less frequently, were other illicit drugs (e.g., unspecified narcotics [n = 5], cocaine [n = 4], and methamphetamines [n = 3]). Chart abstraction data, verified and updated with interview data, revealed the most common administration route for involved opioids in the sample was oral ingestion (n = 62), followed by injection (n = 17). Nine overdoses involved using pills or fentanyl patches via a route that was not prescribed (e.g., chewed, smoked, snorted); seven participants used a combination of administration routes. The source of involved opioids was most often the member’s own active prescription (n = 47). Other sources included street purchases (n = 15), family (n = 8), friends (n = 8), or expired prescriptions (n = 7). Some events involved opioids obtained from multiple sources. We could not determine source in five cases (Table 2). 3.3. Comparison of accidental opioid overdose events and those with suicidal intent We were unable to determine whether overdoses were accidental or suicide attempts/completions for nine cases. Fifty three of the remaining 78 events were categorized as accidental overdoses; 13 resulted in death. While accidental overdoses were fairly evenly split by gender, deaths were more likely to occur among men, particularly overdoses involving heroin. Six of 13 accidental overdoses resulting in death involved heroin, often in combination with other substances. In fact, 40 of 53 accidental events included more than one substance—26 included alcohol, illegal drugs (not including heroin), or prescribed medications obtained illegally. Nine accidental events included prescribed medications known to interact with opioids. While the source of the opioids involved in accidental events varied, about half (n = 27) involved actively prescribed opioids. In addition, our sample included 25 events with clear suicidal intent; 23 individuals survived. Suicide attempts were more likely among women (76%) than men (24%, 2 (1, N = 78) = 5.74, p = 0.015). We examined the association between age and overdose type (accidental vs. suicide) and found no relationship. Suicide attempts were less likely to involve heroin than accidental events. No events coded as suicide attempts involved heroin compared with 14 accidental events, 2 = 8.05, p = 0.002. Of nine events with undetermined intent, three involved heroin. Most suicide attempts involved polysubstance use (n = 23) and the member’s own active opioid prescription (n = 17). Five of these overdoses involved the

52

B.J.H. Yarborough et al. / Drug and Alcohol Dependence 167 (2016) 49–56

Table 1 Participant/decedent demographic characteristics.a Total(n = 87)

Unintentional (n = 53)

Intentional(n = 25)

Gender Male Female

39 48

28 25

6 19

Race/Ethnicity White Non-white Unknown race/ethnicity Hispanic ethnicity

64 13 9 1

42 6 5 1

18 6 1 0

Urban/rural statusb Urban Suburban Large town Small town/rural

56 3 2 2

35 3 1 1

21 0 1 1

Insurance Dual Medicare & Medicaid Medicaid Medicare

3 6 19

3 3 12

0 2 5

Mental Health & Substance Abuse Diagnoses MH and abuse/dependence diagnoses MH Dx only Abuse/dependence Dx only No MH or substance-related Dx

29 32 9 17

17 17 8 11

6 14 0 5

Age, mean (SD) ≤24c 25–34 35–44 45–54 55–64 ≥65

42.9 (16.4) 16 18 9 20 18 6

44.1 (16.9) 11 10 3 9 16 4

4 6 3 8 2 2

a The demographics for family member interviews are for the deceased patient (i.e. spouse, child), not the interviewee. All data in the table derive from chart abstraction of the electronic medical record. The total sample therefore includes 87 cases; nine events could not be determined to be either intentional or unintentional. The remaining columns for unintentional and intentional will total to n = 78. b The Rural Urban Commuting Area (RUCA) taxonomy is derived from the relationship between cities and towns as measured by work commuting flows and are based on work by the University of Washington http://depts.washington.edu/uwruca/ruca-download.php. Zip codes are used to create RUCA codes. Only patient interviews are included in this table as family member zip codes are not necessarily indicative of the residential circumstances of the decedent. c Two individuals were under 18 on the date of their overdose; both were 18 years or older on the date of the interview.

use of members’ expired prescriptions, and four included opioids obtained from family. Ten suicide attempts involved alcohol and/or non-prescribed benzodiazepines; nine events included prescribed benzodiazepines. Twenty of the 25 who made a suicide attempt had a history of diagnosed mental health problems.

3.4. Themes associated with accidental overdoses Qualitative analysis of accidental overdoses identified three primary themes related to opioid abuse, misuse, or therapeutic errors. We define abuse as any use (therapeutic or otherwise) of a drug obtained illegally, or the intentional nontherapeutic use of a drug/substance to achieve a desirable psychological or physiological effect, for example, to “get high”. We defined misuse as the intentional therapeutic use of a drug in an inappropriate way, for example borrowing a spouse’s pain medication, and therapeutic errors as unintentional medication administration mistakes (Smith et al., 2013). Some accidental overdoses involved abuse and misuse, when prescribed or illegally acquired (or both) opioids were combined with alcohol or non-prescribed benzodiazepines with the intent to address pain or emotional distress, but not to “get high”. For example, one partner of a deceased person reported that his partner was disabled and likely depressed, and was taking several prescription pills. Oxycodone and alcohol contributed to what was determined to be an accidental death. Misuse often included self-treatment that involved using medications acquired through a prescription but held, either for later use or to increase dose to “better” address

pain. In addition, some people in the sample misused medications not prescribed to them, even though they had prescriptions. For example, one participant reported that his wife had a 35-year addiction to diazepam but did not abuse or misuse her pain pills (hydrocodone/acetaminophen). The participant had his own morphine prescription. When his wife ran out of her pain medication he reluctantly let her have access to his morphine. He did not know how many she took but the dose was fatal. Chaotic and difficult life circumstances that contributed to overdose events were common in both accidental overdoses and suicidal attempts. More than a third of individuals with accidental overdoses (n = 19) and nearly a quarter of those with suicide attempts (n = 6) were rated as having lives in disorder. 3.4.1. Theme: chronic pain, substance use, and health problems increase risk for accidental opioid overdoses. Examining patterns in the life event ratings data, individuals whose overdoses were classified as accidental were most likely to report chronic pain (n = 27), current or history of alcohol or drug abuse problems (n = 26), or health problems (n = 22) as contributing to overdoses. For example, a middle-aged man with a long history of drug problems and several chronic conditions had prescriptions for fentanyl, morphine, hydrocodone/acetaminophen, and trazodone. He routinely sought additional opioids for pain treatment and recreational use, and reported he would also “talk his girlfriend out of a couple of Xanaxs.” Describing the identified overdose event, he reported that he took three Xanax and over-the-counter-sleep aides “in excess” because he wanted to “feel really good.”

B.J.H. Yarborough et al. / Drug and Alcohol Dependence 167 (2016) 49–56

53

Table 2 Event characteristics.a Total(n = 87)

Unintentional(n = 53)

Intentional(n = 25)

Category of opioids for sample selectionb OxyContin® or oxycodone single ingredient sustained release Oxycodone single ingredient immediate release Other long-acting opioids Other short acting opioids No prescription opioid

9 16 18 8 36

3 9 15 4 22

6 6 3 4 6

Single opioid event Polysubstance event Benzodiazepines involved event Involved heroin

15 72 28 17

13 40 12 14

2 23 9 0

Non-opioid substances involved in eventc Alcohol Marijuana Benzodiazepines (not through Rx) Narcotics, unspecified Cocaine/crack (Meth)amphetamine

17 12 7 5 4 3

5 7 4 5 3 2

9 4 1 0 1 1

Other active Rx at time of eventd Benzodiazepines Sedatives/hypnotics Trazodone Muscle relaxants CYP inhibitors Phenothiazines Barbiturates Antidepressants Atypical psychotics Conventional antipsychotics Mood stabilizers

32 11 11 6 4 3 2 9 10 2 22

6 9 3 1 1 1 6 5 1 15

4 2 3 3 2 1 1 4 1 6

Source of opioids involved in evente Multiple sources Own active Rx Own expired Rx Family member Friend Street Stolen (not family or friend) Unclear

4 47 7 8 8 15 1 5

3 27 2 4 6 14 0 3

1 17 5 4 0 0 0 0

Routes of administrationf Multiple routes Oral Chewed oral tablets Crushed oral tablets to mix with liquid Injected Snorted Patch, transdermal Patch, chewed Patch, smoked Unclear

7 62 2 1 17 2 5 2 2 1

3 32 1 0 14 0 4 2 2 1

2 24 0 1 0 1 1 0 0 0

a The information in this table comes from chart reviews supplemented by self-reported interview data. Family members who were interviewed do not always know all drugs, prescribed or not prescribed, involved in events, therefore this is likely an undercount. The total sample n = 87 includes 9 cases in which we could not determine the intentionality of the event. The remaining columns for unintentional and intentional will total to n = 78. b This represents how individuals were sampled based on information in the electronic health record. It does not necessarily represent the substance involved in the overdose, though it most often does. c Our chart abstraction form also included rohypnol, GHB, inhalants, ketamine, LSD, ecstasy/MDMA, hallucinogens, PCP and opium but these were never identified as contributing factors in the chart reviews nor were they mentioned in interviews. d All prescribed medications were derived from electronic health records. Interacting medications included sedatives/hypnotics, phenothiazines, barbiturates, muscle relaxants, benzodiazepines, and CYP inhibitors. Non-interacting mental health medications included anti-depressants, atypical antipsychotics, conventional antipsychotics, mood stabilizers and trazodone. e This is for the source of only the opioids involved in the event; this does not contain information on the source of other involved medications or substances. Multiple sources are counted in each applicable row, therefore counts will not add up to the total number of cases. f All seven cases (in the total column) included one oral route and one additional route (1 snorted, 3 transdermal patch and 3 injected). Multiple routes are counted in each applicable row, therefore counts will not add up to the total number of cases.

3.4.2. Theme: poor communication of dose changes increases risk for accidental opioid overdoses. Another group of events involved clinicians and/or patients making therapeutic errors. Three cases, for example, reported that their physician told them to “double” their medication and the patients interpreted this to mean they

should take twice as much, not realizing that the doctor had already increased the dosage: “Apparently it was the morphine that did it. And I kind of poisoned myself or something like that. They may have told me, but I don’t remember. . . But we aren’t quite sure where the

54

B.J.H. Yarborough et al. / Drug and Alcohol Dependence 167 (2016) 49–56

mistake came from; whether it was me or whether it was the doctor, because my husband understood the same dosage that I understood.” 3.4.3. Theme: confusion and “feeling out of it” are indicators of higher risk for accidental opioid poisoning, and people may underappreciate the seriousness of these side effects. We classified a subgroup of therapeutic errors as resulting from patients’ impaired judgment or thinking. These events were characterized by a sense of confusion about the event or their medications. Participants could not easily recall what or how much medication they had taken, could not make sense of the event, and their families were concerned that they seemed “out of it.” Most importantly, these participants tended to deny that the event was an overdose. For example, an older woman with multiple chronic conditions and sleep apnea reported always feeling very tired. Leaving her house one day, she passed out on the threshold of the apartment building. She had been taking methadone for pain but was confused about the exact dose. She did not believe this was an overdose and reported that she was just sleeping. 3.5. Themes associated with overdoses with suicidal intent Suicide attempts were less frequent in our sample than accidental overdoses. Two distinct themes were apparent in the interviews: 1) opioids were convenient for people whose pain had become unbearable, and 2) consecutive negative life events preceded attempts. 3.5.1. Theme: when treatment for chronic pain is unsuccessful and pain becomes intolerable, opioids are a convenient and promising means to end suffering. As with those who experienced accidental events, chronic pain was a common predictor of suicide attempts: “I still remember it very vividly. I was just sitting in my chair and going, you know, I’ve got the pills up there in the cupboard. I can just end this and I don’t have to deal with it anymore. The pain will go away. Even though I’m very cognizant that it will kill you, in your mind you’re not really trying to kill yourself. You’re trying to get rid of the pain.” Several participants described long histories of struggling with pain, including changing opioids many times to achieve better pain control. Clinicians’ attempts to reduce dose also appeared to place some people at risk. For example, one woman attempted suicide because of unbearable pain after tapering her opioid dose: “They dropped me from ten eighty milligram tablets a day down to three forty milligram tablets a day, within a five week period. My pain was just unbelievable. . . I would rather have been dead. I did not want to feel that anymore. There’s no way I could do it like that for the rest of my life, no way. And, I just felt that, at the time, going to sleep and never waking up again was the best option I had. The only option I had.” 3.5.2. Theme: consecutive negative life events and loss of family or social support are triggers for suicide attempts. Those with suicide attempts, in many cases, described specific, cascading stressful events that interacted, leaving them vulnerable to suicidal impulses. This group was more likely to have experienced a downward spiral (80% versus 34% of accidental overdoses; 2 (1, N = 75) = 14.11, p < 0.001) and a negative life event preceding the overdose (68% of those experiencing a suicide attempt compared to 20% who accidentally overdosed; 2 (1, N = 74) = 16.18, p < 0.001). “And I just tailspinned. You know, after that it was like, you know, I was already in a real dark hole after the fucking death with [Boyfriend], and then the financial crap and then what this

other person put me through. I had been unpacking some stuff that day and ran across the two bottles and I decided I’d had enough. So I downed about a fifth of vodka and two full big brown bottles of this stuff [liquid morphine and lorazepam]” Reduced family or social support concurrent with the overdose, and mental health problems, were more likely to be reported as causal factors among suicidal individuals: “Well, my ex-husband—he was my husband at the time—worked construction and he got laid off. The economy was bad. He wasn’t working and we lost our house. Things were just really bad financially for us. But, he just said that he was leaving and wanted a divorce, and it just crushed me.”

4. Discussion Individuals who experienced opioid overdoses, whether accidental or intentional, had multiple risk factors across multiple dimensions. Abuse and misuse of prescribed medications was common, as was access to other prescription medications that increase risk. Most overdoses involved polysubstance use. Benzodiazepines were the most common non-opioid substances involved in overdoses, followed by alcohol. Illicit drug use was also common, as was history of substance-related problems and current or past mental health conditions. Chronic pain, unstable resources, and family and social conflicts complicated the lives of the people in our sample, and in combination, set the stage for overdoses. These findings have important implications for preventing opioid-related overdoses and overdose deaths. First, though likely insufficient as a risk reduction strategy by itself, our results suggest that standard patient education and informed consent to opioid treatment should highlight that: 1) all opioids are not the same and should never be borrowed/taken from another person or taken in any manner other than prescribed; 2) prescribed opioids that remain after treatment should be disposed of immediately; 3) patients should be made aware of, and avoid taking, substances that interact with opioids, including alcohol; 4) patients should inform clinicians about all substances being used, including illicit drugs and alcohol, to ensure safety; 5) extreme disorientation or frequent falls are indicators that medications/dosages may need adjustment; and 6) prescribers should be alerted if patients experience changes in mood (particularly depression), problems with family/social support, or poorly controlled pain. Second, our results suggest several opportunities for improved screening of risk factors among opioid users. Screening for impaired cognition or motor skills after opioid initiation or dose escalation may help identify individuals at-risk for unintentional overdose. Additionally, patients should be screened for alcohol and prescribed or illicit drug use. At the same time, screening could be ineffective if patients are reluctant to disclose illicit substance use that could result in their medications being reduced or terminated. Patients with histories of mental illness should be regularly screened for suicide risk. In this sample, suicide attempts were characterized by compounding family and social stressors that interacted with mental health problems and, frequently, pain. Given the substantial proportion of individuals who are seen in outpatient mental health or substance use treatment settings within the month prior to completing suicides (Lin et al., 2015), psychiatrists and mental health clinicians who interact with patients known to use opioids should all have roles in assessing and monitoring suicide risk. Using risk algorithms, these higher risk patients could be flagged for specialized care pathways with additional monitoring and supports. Minimally, health care systems should facilitate communication between clinicians who treat a single

B.J.H. Yarborough et al. / Drug and Alcohol Dependence 167 (2016) 49–56

patient to ensure that medications with dangerous interactions are not co-prescribed and to monitor appropriate medication adherence. Detection of risk is more complicated when patients are experiencing uncontrolled pain. Clinicians trying to adequately treat pain may be inadvertently contributing to overdose risk by increasing opioid dose or adding adjunctive medications. On the other hand, our results suggest that reducing medications may increase suicide risk among some individuals as a result of intolerable pain. Pain control, coping, and suicide risk monitoring should be elements of opioid taper plans, especially for patients with histories of depression or suicidality. Finally, in an era of changing opioid prescribing guidelines, policies, and practices, it is important to understand how opioid prescribing and de-prescribing can influence overdose risk and other unintentional outcomes, such as transitions to heroin use. For example, clinician incentives tied to patient satisfaction measures could lead to higher levels of opioid prescribing, overmedication, and overdose events. At the same time dose and risk reduction efforts may also inadvertently place patients at greater risk if pain is uncontrolled and opioids are obtained elsewhere, though we did not observe many direct transitions to heroin or illegally obtained prescription opioids as a result of prescription restrictions in our sample. This may be due, in part, to the study era—these overdose events occurred prior to 2012, before the recent policy focus on opioid prescribing. Overdose events did span the period before and after the reformulation of OxyContin® however, and our data suggest that lack of availability of OxyContin® led some individuals with prescription opioid abuse histories to initiate heroin use. Reasons for heroin initiation and continuation included achieving a better high, cheaper than prescription opioids, easily attainable, and multiple routes of administration. Of the deaths associated with heroin in our sample, at least 6 had a history that included using/misusing prescription pain medications. Because we did not collect complete drug use histories in our narrative interviews, however, we cannot draw conclusions regarding medical use of prescription opioids leading to non-medical use or heroin use. Nevertheless, transitions to heroin could result in increased overdose events given street heroin’s unknown potency and unregulated status. 4.1. Limitations Nearly half of the participants we attempted to recruit were unreachable, and more than a third of those we reached refused participation, potentially limiting generalizability of our sample. Though analyses suggest few differences between respondents/non-respondents, people with substance use problems can have less stable lives and it can be difficult to find and recruit them or their family members, especially when someone has died of a drug overdose. When family members do participate, interviews may not capture all desired information. Our findings may underestimate suicides, particularly heroinrelated overdoses with suicidal intent. A strength of our method is that we combined EMR data with data abstracted from chart audits and interview data, thereby increasing the sensitivity of our ability to determine suicidal intent beyond what might be available using diagnostic codes. 4.2. Conclusion Adding context to known epidemiologic risk factors, this study documents circumstances surrounding accidental and intentional opioid-related overdoses and identifies overdose prevention opportunities. To reduce risk of opioid overdose, efforts should focus on screening for impaired cognition and motor skills,

55

prescribed and illicit polysubstance use, and changes in life circumstances, psychosocial risks/supports, and pain control. Optimal and timely screening may require shared risk management responsibility distributed across multiple health care providers who interact independently with patients prescribed opioids.

Role of funding This project was conducted as part of the study “Overdose Rates Among Patients Prescribed OxyContin® , Comparator Opioids, and No Opioids in the Kaiser Permanente Northwest & Kaiser Permanente Northern California Health Systems” which was funded by Purdue Pharma, L.P. The study was funded as part of post-marketing requirements by the Food & Drug Administration (FDA) to examine opioid overdose risks. Under Food & Drug Administration (FDA) direction and approval, the funder was involved in the overall study design. The funder had no role in the analysis or interpretation of the results; however they did review and comment on previous drafts and the final version of the manuscript. The funder was not involved in the decision of where to submit the paper for publication.

Contributors Dr. Yarborough had full access to all data, reported and unreported, and was primarily responsible for the analyses presented in the manuscript. Mr. Stumbo contributed to the qualitative analysis and revisions of the text. Ms. Janoff, Mr. Yarborough Dr. McCarty, and Dr. Green participated in the qualitative analysis and interpretation. Drs. McCarty, Chilcoat, Coplan, and Green designed the study and contributed to critical revisions of the text. All authors reviewed and approved the final version of the manuscript.

Conflicts of interest Drs. Chilcoat and Coplan are employees of Purdue Pharma LP and, as such, are members of the Industry PMR Consortium, a consortium of 10 companies working together to conduct FDArequired post-marketing studies that assess known risks related to extended-release, long-acting opioid analgesics. The Industry PMR consortium is comprised of Pfizer, Purdue Pharma, Roxane Laboratories, Janssen Pharmaceuticals, Mallinckrodt, Actavis, Endo Pharmaceuticals, Depomed and Pernix. All of the remaining authors received grant support from Purdue Pharma LP and currently receive grant support from the Industry PMR Consortium. Dr. Green has also provided research consultation for the Industry PMR. Drs. Yarborough, McCarty, and Green, Mr. Stumbo, and Mr. Yarborough receive grant support from NIMH. Drs. Yarborough, McCarty, and Green, Ms. Janoff, Mr. Stumbo and Mr. Yarborough receive grant support from NIDA. Mr. Stumbo receives grant support from the Maternal and Child Health Bureau and Autism Speaks. Dr. McCarty was PI on Research Service Agreements from Purdue Pharma LP and Alkermes.

Acknowledgements This study was funded by Purdue Pharma, L.P. as part of postmarketing requirements by the Food & Drug Administration to examine opioid overdose risks. We would like to thank Alison Firemark for her work conducting interviews, Thomas Young for his help coding transcripts, and Kevin Lutz for his editorial assistance. Finally, we would like to thank the individuals and family members who participated in this study.

56

B.J.H. Yarborough et al. / Drug and Alcohol Dependence 167 (2016) 49–56

Appendix A. Supplementary data Supplementary data associated with this article can be found, in the online version, at http://dx.doi.org/10.1016/j.drugalcdep.2016. 07.024. References Centers for Disease Control and Prevention, 2009. Overdose deaths involving prescription opioids among Medicaid enrollees—Washington, 2004–2007. MMWR 58, 1171–1175. Centers for Disease Control and Prevention, 2010a. Adult use of prescription opioid pain medications—Utah, 2008. MMWR 59, 153–157. Centers for Disease Control and Prevention, 2010b. Emergency department visits involving nonmedical use of selected prescription drugs—United States 2004–2008. MMWR 59, 705–709. Centers for Disease Control and Prevention, 2011. Vital signs: overdoses of prescription opioid pain relievers—United States, 1999–2008. MMWR 60, 1487–1492. Chen, L.H., Hedegaard, H., Warner, M., 2014. Drug-poisoning deaths involving opioid analgesics: United States, 1999–2011. NCHS Data Brief, 1–8. Cicero, T.J., Ellis, M.S., Surratt, H.L., Kurtz, S.P., 2014. The changing face of heroin use in the United States: a retrospective analysis of the past 50 years. JAMA Psychiatry 71, 821–826. Compton, W.M., Jones, C.M., Baldwin, G.T., 2016. Relationship between nonmedical prescription-opioid use and heroin use. N. Engl. J. Med. 374, 154–163. Dart, R.C., Surratt, H.L., Cicero, T.J., Parrino, M.W., Severtson, S.G., Bucher-Bartelson, B., Green, J.L., 2015. Trends in opioid analgesic abuse and mortality in the United States. N. Engl. J. Med. 372, 241–248. Dasgupta, N., Creppage, K., Austin, A., Ringwalt, C., Sanford, C., Proescholdbell, S.K., 2014. Observed transition from opioid analgesic deaths toward heroin. Drug Alcohol Depend. 145, 238–241. Frenk, S.M., Porter, K.S., Paulozzi, L.J., 2015. Prescription opioid analgesic use among adults: United States, 1999–2012. NCHS Data Brief, 1–8. Friese, S., 2011. User’s Manual for ATLAS.ti 6.0. ATLAS.ti Scientific Software Development. GmbH, Berlin. Glaser, B.G., Strauss, A.L., 1967. The Discovery Of Grounded Theory: Strategies For Qualitative Research. Aldine Publishing Company, Chicago.

Hasegawa, K., Espinola, J.A., Brown, D.F., Camargo Jr., C.A., 2014. Trends in U.S. emergency department visits for opioid overdose, 1993–2010. Pain Med. 15, 1765–1770. Johnson, E.M., Lanier, W.A., Merrill, R.M., Crook, J., Porucznik, C.A., Rolfs, R.T., Sauer, B., 2013. Unintentional prescription opioid-related overdose deaths: description of decedents by next of kin or best contact, Utah, 2008–2009. J. Gen. Intern. Med. 28, 522–529. Jones, C.M., McAninch, J.K., 2015. Emergency department visits and overdose deaths from combined use of opioids and benzodiazepines. Am. J. Prev. Med. 49, 493–501. Lin, L.A., Bohnert, A.S., Ilgen, M.A., Pfeiffer, P.N., Ganoczy, D., Blow, F.C., 2015. Outpatient provider contact prior to unintentional opioid overdose among VHA service users. Psychiatr. Serv. 66, 1149–1154. ˜ J., 2009. The Coding Manual For Qualitative Researchers. Sage, London. Saldana, Smith, S.M., Dart, R.C., Katz, N.P., Paillard, F., Adams, E.H., Comer, S.D., Degroot, A., Edwards, R.R., Haddox, J.D., Jaffe, J.H., Jones, C.M., Kleber, H.D., Kopecky, E.A., Markman, J.D., Montoya, I.D., O’Brien, C., Roland, C.L., Stanton, M., Strain, E.C., Vorsanger, G., Wasan, A.D., Weiss, R.D., Turk, D.C., Dworkin, R.H., Analgesic Anesthetic Addiction Clinical Trials Translations Innovations Opportunities Networks Public-Private Partnership, 2013. Classification and definition of misuse, abuse, and related events in clinical trials: ACTTION systematic review and recommendations. Pain 154, 2287–2296. Strauss, A., Corbin, J., 1998. Grounded theory methodology: an overview. In: Denzin, N.K., Lincoln, Y.S. (Eds.), Strategies of Qualitative Inquiry. Sage Publications, Thousand Oaks, California, pp. 158–183. Substance Abuse and Mental Health Services Administration, 2013a. The DAWN Report: Highlights of the 2011 Drug Abuse Warning Network (DAWN) Findings on Drug-Related Emergency Department Visits, http://archive.samhsa.gov/ data/2k13/DAWN127/sr127-DAWN-highlights.pdf (accessed 20.11.15). Substance Abuse and Mental Health Services Administration, 2013b. Results from the 2012 National Survey on Drug Use and Health: Mental Health Findings. NSDUH Series H-47, HHS Publication No. (SMA) 13-4805. Office of Applied Studies, Rockville, MD. Warner, M., Chen, L.H., Makuc, D.M., 2009. Increase in fatal poisonings involving opioid analgesics in the United States, 1999–2006. NCHS Data Brief, 1–8. Warner, M., Chen, L.H., Makuc, D.M., Anderson, R.N., Minino, A.M., 2011. Drug poisoning deaths in the United States, 1980–2008. NCHS Data Brief, 1–8. Yokell, M.A., Delgado, M.K., Zaller, N.D., Wang, N.E., McGowan, S.K., Green, T.C., 2014. Presentation of prescription and nonprescription opioid overdoses to US emergency departments. JAMA Intern. Med. 174, 2034–2037.