Mortality Associated With Opioid Overdose: A Review of Clinical Characteristics and Health Services Received in the Year Prior to Death

Opioid use disorder is a global public health problem with an alarming rate of morbidity and mortality, and the disorder has claimed the lives of more than 500,000 Americans since 2000 (1, 2). During this time, the number of opioid prescriptions and subsequent deaths in the United States has quadrupled, and the overall death rate from opioid-related overdoses has tripled (2). National attention to opioid use disorders has increased, with an expansion of overdose prevention initiatives. However, opioid-related overdoses were still responsible for 91 American deaths per day and accounted for 63% of all drug overdose deaths in 2015 (3).

The contribution of health care systems to lessen the impact of this opioid overdose epidemic may be improved by increasing the effectiveness of identification of high-risk cases, providing better interventions, and obtaining feedback from outcomes. However, in medical centers the quality review processes to examine fatal outcomes such as opioid overdose are constrained to incidents in the hospital or to individuals who die in the community and of whom the provider is aware. This constraint excludes review of cases of individuals who drop out of treatment or are unknown to the medical team. Surveying these cases for clinical characteristics, provider practices, the presence of screening procedures, and service use prior to the overdose could help increase knowledge of specific characteristics during clinical management and identify missed opportunities.

Assessment of the timing and type of outpatient providers who had contact with the person prior to fatal prescription opioid overdoses may help identify services with a missed opportunity to intervene. A study with veterans (N=1,813) showed that 62% were seen within the month prior to their death, and 30% had been seen in an ambulatory mental health or substance treatment program (4). Another large study of opioid-related decedents (N=13,089) in the Medicaid program showed that those with chronic pain were more likely than a group without pain to have a prescription for opioid analgesics and benzodiazepines during the last 30 days of life (5). However, although these studies used diagnoses and procedure codes to identify patterns of service use and associations of diagnoses with fatal opioid overdoses in large samples, these types of studies do not manually abstract electronic medical record (EMR)–based charts to obtain relevant details of interactions with providers and treatment services prior to death.

Thus this retrospective study evaluated EMRs and identified all opioid-related deaths that occurred in Worcester between 2008 and 2012 and in which the decedent had any encounter with the UMass Memorial Medical Center (UMMC) system in the year prior to death. UMMC is the largest provider in the metropolitan area of Worcester and a teaching center that has two level-3 hospitals and emergency departments covering 80% of occurrences; it has a full range of medical specialties, and over 70% of persons in the population have their primary care provider in this system. It was expected that most decedents had received most of their care in this health care system.

The overall objective of the study was to identify the demographic and clinical characteristics of decedents from opioid overdose and their clinical management and to describe the health care services received at the UMMC. We were particularly interested in decedents who were identified as having opioid use disorder or who had aberrant drug-related behaviors (6) with chronic pain. Considering that these two groups are highly visible and easier to identify and considered at-risk groups by most providers, we considered them together as having problematic opioid and compared them with opioid overdose decedents who had not had problematic use. This study aimed to characterize the overall population of decedents from opioid overdose who received services at UMMC and to compare decedents with and without problematic opioid use to asses all potential missed opportunities to reduce risk factors for fatal opioid overdose.

Methods

Participants

Eligible cases were Worcester residents (7) deceased by opioid overdose between 2008 and 2012 within the Worcester metropolitan area (population 513,000 in the 2010 census) who received medical services at UMMC during the year prior to their death. All cases in which the confirmed cause of death was opioid overdose were obtained from the Registry of Vital Records and Statistics at Massachusetts Department of Public Health. The following ICD-10 external cause of injury codes were used: X40–X49, X60–X69, X85–X90, Y10–Y19, Y35.2, or *U01 (Unintentional=V01–X59, Y85–Y86) + (Suicide=X60–X84, Y87.0, *U03) + (Undetermined=Y10–Y34, Y87.2, Y89.9). In addition, the following codes were used: T40.2–T40.4 (any opioid analgesic), T40.0 (opium), T40.1 (heroin), T40.2 (other opioid), T40.3 (methadone), T40.4 (other synthetic narcotic), and T40.6 (other and unspecified narcotics). Of the total of 157 decedent cases obtained, 112 (71%) were selected on the basis of any encounter with the emergency department, inpatient admission, or outpatient clinic in the 12 months prior to the date of death as confirmed with UMMC EMR data. This study was approved by the University of Massachusetts Institutional Review Board.

Study Design

This retrospective study reviewed UMMC system EMRs for 112 decedents during the 12 months prior to death by using a data abstraction form developed by the investigators. [A copy of this form is included in an online supplement to this article.] Two chart reviewers (senior medical students [CR and AS]) trained by the study principal investigator (GG), manually abstracted EMR-based charts for all decedents. The first five charts were reviewed by the principal investigator together with the chart reviewers by using the form to ensure accurate and consistent abstraction. Discrepancies after double data entry were resolved by consensus with the PI to create the final data set for analysis.

Data Collection

Demographic characteristics and diagnoses.

Individual characteristics included gender, age, and evidence of insurance coverage. The principal diagnoses for opioid use disorder, chronic pain syndromes with and without aberrant opioid use behaviors, substance use disorders, and psychiatric disorders were determined by clinical judgment, with either evidence of active treatment or evidence of repeated treatment for the same problem or diagnosis at a recent hospital discharge as assessed from free text in medical notes. Problematic opioid use was indicated for persons with a principal diagnosis of opioid use disorder or aberrant drug-related behavior during the 12 months prior to death. Aberrant drug-related behavior included documentation of any concern related to patterns of early refills, multiple calls or visits to request more opioids, lost or stolen prescriptions, use of the emergency department to obtain prescriptions, disruptive behavior associated with opioid treatment, and misuse of opioids (6).

Overdose risk assessment.

The following factors were included in the risk assessment: documented episodes of suicidal ideation, suicide attempts, and opioid overdoses within one year prior to death; number of lifetime episodes of opioid overdose; and prescriptions of opioid analgesics (with total daily morphine equivalents) or benzodiazepines (with total daily dosage).

Universal precautions.

Documented evidence was recorded of screening for substance use disorders, including smoking status; urine toxicology performed; evidence of opioid treatment agreements; patient education regarding risks of opioid treatment; and “high-risk behavior” chart alerts (8).

Medical services received.

Primary and last treating services prior to death for outpatient, inpatient, surgical, emergency department, pain clinics, and psychiatry were recorded.

Substance use management.

Documented use was recorded of treatment resources, such as the consultation-liaison psychiatry service or overdose counseling program; referrals for substance use disorder treatment or self-help group; participation in substance use disorder or medication-assisted treatment for opioid use disorder; and section 35 (Massachusetts court-mandated drug treatment).

Statistical Analyses

Analyses were conducted with the group variable problematic opioid use versus nonproblematic opioid use. Categorical variables were evaluated by chi-square tests with adjusted residual (Z>2.0 for significance, for post hoc comparisons [9]), and continuous variables were evaluated with Student’s t tests. Between-group comparisons of principal medical diagnoses were not performed. All analyses were two-tailed, with statistical significance set at ≤.05. Analyses were performed with IBM SPSS, version 24.

Results

Demographic Characteristics

Overall, most of the 112 decedents were males (64%) and Caucasian (96%). Their mean age was 41. Four decedents had documentation of an intentional overdose (Table 1). The 112 decedents were categorized as having problematic opioid use (N=53, 47%) or nonproblematic opioid use (N=59, 53%). Deaths were not significantly more likely to occur in the community (N=61, 55%) than in the hospital setting (N=47, 42%).

Clinical Characteristics

Diagnoses.

The distribution of the principal diagnoses for the entire sample was as follows: chronic pain with noncancerous condition (41%), substance use disorder (18%), opioid use disorder (17%), mental illness (14%), cancer (2%), and other (8%) (Table 1). Problematic opioid use by definition meant users had opioid use disorder (36% as a principal diagnosis), and the second most common principal diagnosis in this group was substance use disorder (34%). Nonproblematic opioid use involved the following principal diagnoses: chronic pain (54%), mental illness (24%), and other (15%) (Figure 1). However, when any secondary diagnoses were included, 62% of the entire sample had any chronic pain syndrome, 66% had any substance use disorder, and 64% had any mental illness (Table 1). Also, when all cases with any diagnoses were considered, the proportion with any mental illness was larger among the group with versus without problematic opioid use (74% versus 56%, p<.06) (Table 1).

Overdose risk assessment.

Overall, 57% of the decedents had a prescription for an opioid analgesic, and 41% had a benzodiazepine prescription within UMMC. Decedents with problematic opioid use were more likely than those with nonproblematic use to have been prescribed opioids in the last year of life (72% versus 37%, p<.001). Also, the mean total daily morphine equivalent was threefold higher for problematic opioid use than for nonproblematic use (165.4 versus 55.6 mg per day, p=.01).

Among the 112 decedents, 22 (20%) had a documented history of opioid overdoses during their last year, 36% of those with problematic opioid use and 5% of those with nonproblematic use (p<.001). Furthermore, 53% of those with problematic opioid use had a documented lifetime history of opioid overdoses, compared with only 7% of the nonproblematic use group (p<.001).

Universal precautions performed.

Including smoking status, 71% of all decedents were screened for substance use disorders, and 40% had a urine drug screen. Decedents with problematic opioid use were significantly more likely than those with nonproblematic use to have had a substance use disorders screening (89% versus 56%, p<.001) and a urine drug screen (59% versus 24%, p<.001) (Table 1). Few decedents had opioid treatment agreements (8%), documented patient education (24%), or a chart alert for high-risk behavior (5%).

Medical services received.

Overall, the mean±SD number of days between the last medical service received and death was 94±93 for emergency department visits, 100±106 for outpatient visits, and 111±93 for inpatient admissions, with no significant difference between groups. The last treating service consulted differed significantly between the two groups (p<.05). Compared with those with problematic opioid use, those with nonproblematic use were significantly more frequently last seen in the surgical and subspecialty settings (29% versus 11%, Z=2.3) and numerically seen more in a primary care setting (17% versus 9%). Compared with the decedents with nonproblematic use, those with problematic use were seen more frequently in the emergency department (55% versus 44%), as a medical inpatient (11% versus 3%), and in psychiatry (13% versus 7%) (Figure 2).

Substance use management.

In the study sample, few decedents had documented interaction with consultation-liaison psychiatry or overdose counseling (14%), substance use disorder treatment referrals (14%), active treatment for substance use disorders (15%), medication-assisted treatment for opioid use disorder (buprenorphine, 5%; methadone, 7%), referral to self-help groups (5%), or court-mandated substance use disorder treatment via section 35 (1%). Of those who had documented interactions with consultation-liaison psychiatry services, the proportion was larger among those with versus without problematic opioid use (26% versus 3%, p<.001). Similarly, as expected by group definition, significantly greater proportions of those with problematic opioid use received substance use disorder treatment referrals (23% versus 7%, p<.04), were active in substance use disorder treatment (26% versus 5%, p<.001), and had medication-assisted treatment for opioid use disorder with buprenorphine (11% versus 0%) and methadone (13% versus 2%) (p<.001).

Discussion

This retrospective study of 112 decedents from opioid overdose from 2008 to 2012 in the Worcester metropolitan area who were seen in the year before death at the UMMC identified two distinct groups. Those with problematic opioid use, who we anticipated were a high-risk population (that is, those with a principal diagnosis of opioid use disorder or aberrant drug-related behavior), represented only 47% of the deaths. Persons with nonproblematic opioid use appeared more difficult to recognize or to conveniently categorize by their providers as being at risk of dying from opioid overdose, because they had chronic pain or a mental illness as their principal medical diagnosis, without apparent red flags, and they were less likely to have a high-average-dose opioid prescription or a concurrent benzodiazepine prescription. This potentially silent group represented 53% of the deaths.

Persons with problematic opioid use were an at-risk subgroup because they had a principal diagnosis of opioid use disorder or aberrant drug-related behavior. In addition, this group had a comorbid substance use disorder, a prescription for an opioid analgesic with a high total daily morphine equivalent, and a benzodiazepine prescription and were more likely than those with nonproblematic use to have a documented last-year history of opioid overdose. Although identification of this high-risk group was documented, expected interventions (including universal precautions, referral to substance use disorders treatment and self-help groups, and encouragement to enter medication-assisted treatment) were, unfortunately, scarcely implemented. These cases represent missed opportunities; the individuals had already been identified by their providers as high risk. It is expected that treatment measures will continue to improve. These findings are, in part, contrary to those of a recent large study (N=13,089) of opioid-related decedents in the Medicaid program, which found that only 14.8% of decedents had 12-month diagnoses of any opioid use disorder, and even though 61% had a diagnosis of chronic pain, there was no report of aberrant drug-related behaviors (5).

The occurrence of death among persons with nonproblematic opioid use, as described in this study, is much more perplexing and alarming. Although it is possible that these individuals represent a subgroup of patients who were not well identified by their provider, not well documented in the EMR, or not appropriately screened as having problematic opioid use, their overall profile is highly suggestive of a silent group that needs further consideration and study. Compared with the group with problematic opioid use, this group was more likely to have a principal diagnosis of a mental illness or chronic pain without aberrant drug-related behavior, suggesting that the comorbidity of chronic pain and mental illness may alert providers (Table 1). However, nonproblematic use was less likely than problematic use to entail an opioid prescription and total daily morphine equivalents that were not alarming, suggesting a lower risk of opioid overdose. In addition, although some in this group had a concurrent benzodiazepine prescription that might be suggestive of a potential risky interaction, they were also less likely to have a documented opioid overdose in the year before death. Those with nonproblematic opioid use were more likely than those with problematic use to have been seen last in surgical or subspecialty settings, suggesting less acute problems with opioids. It is possible that this group obtained prescription opioids from other providers or from street suppliers, unknown to their providers. Alternatively, these deaths could have been unrecognized intentional self-inflicted opioid overdoses (10). However, of the 17 individuals with documented suicidal thoughts and the seven with suicide attempts in the year before death, the proportions with such thoughts or attempts were smaller among persons with nonproblematic use compared with problematic use, suggesting that this explanation is less likely. Nonproblematic opioid use accounted for the most opioid overdose deaths in our study. We believe that although this group is most at risk of death by opioid overdose, this group is least likely to be identified. Novel guidelines for detection and intervention are needed for this group. Insisting on education and distribution of naloxone kits to all patients treated with opioids may also help this silent group.

Our study had several strengths, including a population-based design, inclusion of only validated cases of decedents who had contact with the health care system during the year prior to their death, and use of a methodology to assist a local health care system in improving identification of at-risk groups and addressing weaknesses in management. Most hospital-based quality improvement efforts are focused on outcomes and incidents that occur during the treatment phase, and public health studies often are limited in their capacity to provide specific information to improve specific institutional performance. In addition, dissemination of these data can help providers in their motivation and commitment to address public health measures.

Limitations included the retrospective nature of the study that limited the predictive information and the limited reliability of the accuracy of information in the EMRs, which we could not ascertain. The EMRs reviewed were exclusively from the UMMC system, and some patients may have had encounters with other hospitals outside Worcester. Also, the prescription data were obtained from medical records and not from pharmacy records, which would have reflected refill data. By definition the sample was selected because of receipt of services, and thus there may have been a bias toward service users, which may have overstated the results of service use. Finally, the evolving nature of the epidemic, with continuing rises in mortality from synthetic opioids such as fentanyl, will likely require more rapid and contemporary feedback on recent decedents to provide pertinent information.

Conclusions

This study found that decedents who had opioid use disorder or aberrant drug-related behavior accounted only for 47% of the deaths by opioid overdose in the metropolitan area of Worcester, Massachusetts. The study found numerous opportunities for improvement in universal precautions and substance use management of this high-risk group. In addition, a harder-to-identify group with chronic pain, mental illness, and low use of prescription opioids accounted for a similar proportion of the deaths by opioid overdose. Surveillance at the population level and further characterization of this group, with specific strategies to identify and design appropriate interventions, are warranted and must be sought.