Studying Adverse Events Related to Prescription Opioids: The Utah Experience

Abstract

Background. Epidemiologists at the Utah Department of Health (UDOH) began to study prescription drug-related harm in 2004. We have analyzed several types of data including vital statistics, medical examiner records, emergency department diagnoses, and the state prescription registry to estimate the scope and correlates of prescription drug-related harm.

Objectives. To describe data sets analyzed in Utah related to the problem of prescription drug-related harm with the goal of designing interventions to reduce the burden of adverse events and death.

Results. Prescription drug-related harm in Utah primarily involved opioids and can be examined with secondary analysis of administrative databases, although each database has limitations.

Conclusions. More analyses, likely from cohort studies, are needed to identify risky prescribing patterns and individual-level risk factors for opioid-related harm. Combining data sets via linkage procedures can generate individual-level drug exposure and outcome histories, which may be useful to simulate a prospective cohort.

The Beginning: Prescription Drug-Related Research in Utah

Epidemiologists at the Utah Department of Health (UDOH) began to study prescription drug-related harm in 2004 upon the suggestion from the chief medical examiner that he was seeing more cases related to prescription drugs than illegal drugs. The first strategy was to determine if his observation could be corroborated with analysis of existing data such as death certificates and the electronic portion of the state medical examiner database. Those analyses confirmed the existence of an increased number of deaths, and we began the process of describing the decedents, quantifying nonfatal harms such as emergency department (ED) visits and looking for potential risk factors for drug-related harm. As this work progressed in Utah, other states and agencies also began analyzing the issue in different ways leading to collaborative research and prevention efforts.

What follows is a brief description of the work done in Utah on prescription drug-related harm including information on the data sets, methodologies, and findings.

Focus on Fatality: Death Certificates

Death certificate data is a logical starting point for quantifying drug-related fatalities. Death certificate data are a component of vital statistics information maintained by the Office of Vital Records and Statistics at the UDOH and are submitted for inclusion in national vital statistics data sets through the National Vital Statistics System. Each death certificate has a primary cause-of-death field and up to nine contributing diagnosis fields encoded using the International Classification of Diseases, 10th Revision (ICD-10) [1]. In Utah, cause of death is certified by the decedent's physician, the attending physician, or the medical examiner. The cause-of-death literals are entered and then machine-coded at the National Center for Health Statistics using ICD-10 nosology [2].

The ICD-10 system includes contextual information about circumstances of death in addition to cause-of-death information. Codes selected for inclusion in case definitions used for death certificate studies have varied. For example, some studies limit inclusion to deaths classified as accidents (ICD-10 codes X42 and X44), while others combine unintentional deaths (ICD-10 codes X40–44) with deaths of undetermined intent (ICD-10 codes Y10–14) [3–6]. Some researchers have advocated for use of ICD-10 codes F10–F19 (mental and behavioral disorders due to psychoactive substance use) to help identify and classify drug-related deaths. In our experience, these codes usually appear in conjunction with other more specific codes and do not provide enough specific information to assist with evaluation of drug-related fatalities using death certificates alone. Death certificates provide limited information about the specific drug(s) involved with a given death. The ICD-10 category T40 can be used to indicate poisoning by narcotics and psychodysleptics, but not all drugs of interest have a unique code. Methadone-related deaths, for example, may include the code T40.3 as a contributing cause of death, but deaths caused by oxycodone and hydrocodone would be classified as T40.2 (other opioids), and fentanyl-related deaths would be included in T40.4 (other synthetic narcotics). With nonspecific classification, the impact of individual drugs cannot be studied using coded death certificate data.

In Utah, the number of deaths in which drugs are identified as a cause of death on the death certificate increased between 1999 and 2008, the most recent year for which data are available (Table 1) [2]. The drug-related death rate has trended upward during that same time period with varying proportional contribution among the ICD-10 primary cause-of-death codes (Figure 1).

Beyond Death Certificates: Investigating Fatalities Using Medical-Examiner Data

Because of the limitations of death-certificate data, we have focused on efforts to describe decedents and circumstances of death on our medical examiner data. Utah has a centralized, state-wide, electronic, medical examiner system with the investigative information contained within a single database. Unfortunately, the data are not coded but exist as text, complicating analysis.

All decedents for the period January 2000 through December 2009 were extracted from the database for manual review by two independent reviewers. Cases were classified as primary or secondary. In primary cases, drug overdose was the most proximate cause of death. Secondary cases involved a drug, which was not the immediate cause of death (e.g., drowning, choking, fall). Cases were identified by the presence of any drug-related keywords (e.g., poisoning, drug, intoxication, toxicity, and overdose, or specific drugs such as morphine, methadone, or oxycodone) in the cause-of-death field. Indicator variables were then created for each drug mentioned as a cause of death for each decedent. The coded data files were compared, and differences were reconciled against the electronic records. In some instances, the paper files were consulted to collect more contextual information about the death, for example, to determine whether the morphine found on a toxicology test was from heroin or an existing morphine prescription.

For the period 2000–2009, there were 3,421 drug-related deaths identified in the medical examiner database. Each death was classified as caused by illicit drugs only, non-illicit drugs only, or a combination of these types. Alcohol was not counted as a non-illicit drug; therefore, the combination of an illicit drug such as heroin and alcohol would not classify the death into the combination category.

Deaths caused only by non-illicit drugs increased markedly during the study period (Figure 2). Opioid pain medications contributed to the majority of deaths in the non-illicit (90%) and combination (82%) categories. Methadone and oxycodone contributed to the largest numbers of deaths throughout the study period; however, due to the fact that many deaths involved more than one drug, the numbers cannot be summed to the total number of deaths (Figure 3).

The medical examiner report informs the death certificate, but the medical examiner database has more contextual information available and is not constrained by ICD coding. Every death that the medical examiner classifies as drug related should also be classified as drug related using death certificate data, but specifics of the case would likely be lost in the coding process. The coded data of death certificates is easier to query, and standardized queries could be generated to compare among states, whereas medical examiner data are generally restricted to a single jurisdiction, be it a state or a county.

Having established that the number of drug-related fatalities had risen in Utah, we next began to wonder whether there was a similar trend for non-fatal overdoses. To answer that question, we began analysis of our state-wide Emergency Department Encounter Database.

Administrative Data to Investigate Adverse Events: Emergency Department Data

The Emergency Department Encounter Database is a repository of all ED patient encounters in Utah from 1999 through the present and is maintained by the Bureau of Emergency Medical Services at the UDOH [7]. The database contains complete claims data for each encounter including diagnosis codes, procedure codes, patient demographic information, services received, and charges billed for each patient ED encounter. For all years, diagnosis and procedure information is encoded with ICD-9-CM. ICD-9 codes under 965 refer to poisonings by analgesics, antipyretics, and antirheumatics. The number of drug-related ED encounters has more than quadrupled from 1999 to 2008 (Figure 4).

Having quantified fatal and nonfatal drug-related harm, the next logical questions became those of context and appropriate sphere for intervention. Fundamentally, we wondered if people were being harmed by their own prescriptions, suggesting a patient-safety issue, or if the people experiencing harm had acquired the medications by illegal means, pointing to a law enforcement issue. We also wondered what had changed, if anything, about the way physicians prescribed pain medication to patients during the time we saw such dramatic increases in prescription drug-related harm. Fortunately, we had another resource of administrative data regarding filled prescriptions.

Administrative Data to Evaluate Exposure to Medication: Controlled-Substances Database

The Utah Controlled Substances Database (CSD) contains information on all Schedules II through V prescriptions dispensed from approximately 500 community pharmacies in Utah. Managed by the Utah Department of Commerce Division of Occupational and Professional Licensing (DOPL) since 1995, the CSD was established by authority of the Utah Controlled Substance Act [8] and R156-37 Utah Controlled Substances Act Rules. Medical providers may access the CSD via a secured Internet portal by telephone or fax to find the prescription histories of their patients as a means to guide patient care. Staff at DOPL use the data to detect potential drug diversion or “doctor shopping” by patients and possible inappropriate prescribing by providers. The CSD does not use a master patient index or patient-level unique identifier to produce reports. Reports are generated by searches for patient name, address, and date of birth. Potential matches are reported to the physician with the same information, and the selected record(s) then compiled into a list of filled prescriptions which includes: date prescription was filled, the doctor name associated with the prescription, the drug name, strength and form, pharmacy name and address along with the patient identifiers. Additional data items (prescription number, new/refill code, metric quantity, days supplied, National Drug Code number, Drug Enforcement Administration number with suffix, date prescription was written, and number of refills authorized) are stored within the CSD and used for research and enforcement purposes but not available to providers.

During the years 2002–2009, pharmacies reported 327,552,189 individual records for filled prescriptions to the CSD. The number of prescriptions included in the CSD has increased every year to nearly 5 million records in 2009. Opioid medications are the largest single class of drugs included in the CSD (Figure 5). Among opioids, hydrocodone is the drug dispensed most often, comprising over half of opioid prescriptions in every year (Figure 6). Methadone, a drug associated with more fatalities than hydrocodone, is prescribed much less frequently.

The majority of individuals represented in the CSD each year filled more than one prescription. Viewing person-level data rather than prescription-level data, we separated the individuals who filled prescriptions for non-opioid controlled substances from those who filled at least one opioid prescription and found that the majority of individuals filled at least one opioid prescription (Figure 7).

Linking Data Sets to Investigate Drug-Related Harm

The next question we investigated was how the individuals who died from a prescription drug overdose had gained access to controlled substance medication. To gain answers, we linked data from the medical examiner, death certificates, and CSD. From 1999 to 2004, there were 80,227 deaths of Utah residents (Figure 8). Of those deaths, 263 were identified as accidental opioid poisonings, and 971 were identified as opioid poisoning with undetermined intent resulting in 1,234 apparently non-intentional opioid poisonings. In 483 (39%) of the accidental and unknown opioid poisoning deaths, illegal substances (e.g., cocaine, methamphetamine, marijuana) were found during toxicology examination, and in 751 (61%) of the accidental and unknown opioid poisoning deaths, no illegal substances were found. Based on the linkage of death records to the controlled substance prescription records contained in the CSD, 69 of the 483 (14%) opioid deaths (accidental and undetermined intent) with illicit drug use had at least one opioid dispensed where the supply would have ended within 30 days of death if the drug was used as prescribed. In comparison, 431 of 751 (57%) of the deaths involving only non-illicit medications based on toxicology results had at least one opioid dispensed where the supply would have ended within 30 days of death. This is a strict case definition, and expanding the time window between most recently filled prescription and death indicated that the majority (75%) of opioid-related decedents had a legitimate prescription within 1 year of death.

These results indicate that a substantial proportion of the individuals who died of prescription pain medication overdose were receiving at least one of the implicated opioids by prescription from a health care provider. Those deaths represent an opportunity for prevention by better educating both patients and health care providers about the risks from these medications. These results also suggest that a substantial proportion of decedents obtained the implicated medications by some other means. Different interventions will be needed to prevent those deaths. This analysis is ongoing, along with case-crossover analyses to investigate potential increases in risk associated with drug initiation, drug combinations, and changes in dose.

To this point, all of our investigations into prescription drug-related harm relied on secondary analyses of existing data sets. We needed to gather some information about how people and patients were using prescription drugs and how drug-use behaviors might relate to future harm.

Surveying the Population: Acquisition, Use, and Disposition of Medication

In 2008, the UDOH added 12 questions about use of prescription pain medications to the Utah Behavioral Risk Factor Surveillance System (BRFSS). The BRFSS is a state-based telephone survey on health-related behaviors among adults aged 18 years and older coordinated by the Centers for Disease Control and Prevention and conducted annually since 1984. A core set of questions are collected nationally, and states have the option of adding additional modules, as Utah did, to support state-level priorities. Results of this survey appeared in the Centers for Disease Control and Prevention's Morbidity and Mortality Weekly Report [9]. A total of 5,330 adults in Utah participated in the interview. The first question asked about “any pain medications prescribed to you by a doctor.” The next question asked respondents to provide the name(s) of the medication(s) without being prompted with specific drug names or types by the interviewer. Drugs were categorized by type (opioid vs other) for this analysis. Results in this report reflect responses with at least one opioid mentioned. In 2008, 20.8% of adults in Utah was prescribed at least one prescription opioid during the previous 12 months. The majority (72%) of adults who reported receiving a prescription opioid had leftover medication from the most recently filled prescription. Of those with prescription opioid medication remaining, 71% of people reported that they kept it, 25.2% disposed of the medications, and 2.3% gave the medication to someone else. Respondents who reported receiving an opioid prescription for short-term pain were more likely to report having leftover medications (75.9%) than respondents who reported receiving a prescription for long-term pain (60.5%, P for difference in proportions <0.001). Only 3.2% of adults who had received a prescription opioid during the past year reported using the medication more frequently or in higher doses than directed by their doctor.

Receiving prescription opioids for short-term pain was more common in younger adults, while receiving prescription opioids for chronic or long-term pain was more common in older adults with no statistically significant differences by sex (Figure 9). Hydrocodone was the medication most frequently mentioned by respondents as the opioid they were prescribed during the previous 12 months (Figure 10). As expected, a larger percentage of people were prescribed hydrocodone, a short-acting drug, for short-term pain (71.0%, 95% confidence interval [CI][66.4%, 75.6%]) than for long-term pain (60.1%, 95% CI [51.7%,68.4%]). Among those who specifically reported being prescribed OxyContin® (Purdue Pharma LP, Stamford, CT, USA), 50.8% received this sustained-release drug for short-term pain, although that is not its indication. The Utah Clinical Guidelines on Prescribing Opioids for Treatment of Pain recommended that “long duration-of-action opioids should not be used for treatment of acute pain, including post-operative pain, except in situations where monitoring and assessment for adverse effects can be conducted”[10]. As these guidelines are incorporated into provider practice, it is anticipated that the number of individuals receiving sustained-action drugs like OxyContin® for short-term pain will decrease.

In the past year, 1.8% (95% CI, 1.4%, 2.3%) of adults in Utah (2.1% among males; 1.6% among females) reported using prescription opioid medication that was not prescribed to them. The proportion of males and females reporting this behavior was approximately the same for all age groups with no statistically significant differences. For both males and females, those between the ages of 35 and 44 years were most likely to report this misuse. Respondents were asked why they used pain medication not prescribed to them. Choices were not offered, and respondents could give multiple reasons that were then categorized by the interviewer. The most common reason stated was “to relieve pain” according to 72.4% of the respondents. Other reasons were categorized as “for fun, getting high, for the feeling of it” (15.3%), “to relieve other physical symptoms” (2.2%), “to relieve anxiety or depression” (3.7%), “to prevent or relieve withdrawal” (1.3%), and “other” (10.5%). Additionally, when asked how the medication was obtained, 97% reported receiving it from a friend or relative. The majority (85.2%) of individuals who used prescription opioids not prescribed to them reported that the medication was given to them for free, while 4.1% purchased it and 9.8% took it without the owner's knowledge or permission.

Enhanced Medical-Examiner Investigation to Collect Contextual Data for Fatalities

To collect additional contextual data on decedents whose deaths were investigated by the medical examiner, we interviewed the relative or friend most knowledgeable about each decedent's life. Some of results from this investigation were reported in a Utah Health Status Update [11], and data analysis is ongoing. Inclusion criteria were: Utah residents ages 12 and older whose death was identified as drug related between October 26, 2008 and October 25, 2009.

Among the 432 drug-related poisoning deaths in the study period, interviews were completed on 385 cases, 240 of which involved only non-illicit drugs and were of accidental or undetermined intent. Three characteristics were correlated to overdose deaths: financial problems, history of substance abuse, and mental health concerns.

Highlights of next-of-kin responses for 240 decedents whose deaths involved only non-illicit drugs are as follows [11]:

Financial problems:

• 63% of decedents were unemployed during the last 2 months of life.

• 59% of respondents reported that the decedent had a financial problem during the 2 months prior to death.

• 27% of individuals were uninsured at the time of death. This is higher than the state-wide uninsured rate of 14% in 2008.

History of substance abuse:

• When asked if the decedent experienced a substance abuse problem during the 2 months prior to death, 40% responded “yes.”

• Specific drugs that the decedent had ever used during his or her lifetime included high rates of marijuana (48%), cocaine (25%), methamphetamine (23%), and heroin (17%).

• 12% of decedents were reported to have used prescription pain medication for reasons other than to treat pain in the year prior to death.

• 49% of decedents had ever received treatment for substance abuse.

Mental health:

• 49% were reported to have been diagnosed with a mental illness by a health care provider.

• 24% of decedents had been hospitalized for psychiatric reasons.

Synthesizing Findings from Multiple Data Sources

The investigative path we describe was driven both by research questions and by data availability within the context of a state health department. The ideal study of drug-related harm would be a population-based prospective cohort and include users with different degrees of exposure, behaviors, and comorbidities followed through time with continuous assessment of exposure and outcomes. We are attempting to simulate such a cohort with our linked data sets, and analyses are ongoing, but in the setting of public health concern, time and resources may be insufficient to plan and conduct the ideal study. Intervention can and should sometimes occur before all of the research questions related to a health concern are answered. Administrative data—including the vital statistics, ED and CSD data described in this report—can be a resource-efficient means to answer research questions, generate hypotheses, and initiate interventions.

Bias is of concern within administrative data sets as with all data sources. Analysis is limited to the data that are present on death certificates, for example, and if there are systematic differences in how cause of death is reported that are associated with prescription drugs, then the effect estimates will be biased. In Utah, all deaths that are thought to be drug related come under the jurisdiction of the medical examiner and undergo systematic investigation. This process should minimize variation and subsequent bias in cause-of-death reporting for drug-related deaths. Deaths that were truly drug related but not suspected to be so or deaths in which the association with drugs was deliberately concealed would be missing from death certificate data. We have no way to estimate the magnitude of such misclassification. If it is significant, particularly if differently distributed among demographic or other strata, the effect would be to underestimate the burden of drug-related fatalities as measured using death certificates.

Similarly, case ascertainment bias is of concern within the medical examiner database. When drugs are not suspected as a cause of death or a motivation exists to attribute the death to another cause, such deaths are not investigated by the medical examiner. We believe that this has led to an underestimation of the impact of prescription opioids within the elderly population for which other causes of death exist and drugs may not be suspected and possibly in the adolescent population for which there might be motivation to conceal a suspected drug-related death. Diagnosis bias, or surveillance bias, is another issue. As the awareness of prescription opioid-related harm has increased over time, there may have been changes in how potential overdoses were evaluated in the ED or investigated by the medical examiner. This awareness might bias the data toward overestimating the burden of drug-related harm in the population of medical examiner cases. No changes have been made in investigative protocol (with the exception of the limited next-of-kin study) within our state Office of the Medical Examiner during the time period upon which we report, and we believe that the increasing trends we report reflect a true change in the population rather than diagnosis bias.

We have no way to estimate the impact of diagnostic suspicion regarding prescription drug-related overdose in the ED as compared with suspicion of illicit drug use. Diagnosis coding in the ED can be informed by objective, laboratory evidence of toxicology, and toxicology testing of apparent overdose victims is a common procedure. Overall, we feel confident that diagnosis codes are not unduly influenced by diagnostic suspicion bias and may be used for these and similar analyses. However, consideration is due to the following possibilities: 1) that serum dose levels may be unrelated to toxicity, and 2) that bias may exist toward judging an opioid responsible for toxicity, although a non-opioid could be responsible.

Conclusions

Investigation into the problem of adverse drug events related to prescription opioids is ongoing. From our various investigations, we have described a population at risk, but the description is so broad that we have not yet been able to target groups at heightened risk for specific interventions. The evidence from each of our analyses trends in the same direction —the burden of adverse events caused by opioids is increasing as is the exposure to opioids in the population; therefore, we expect the trend to continue. More research is needed, particularly at the individual level and to investigate risk factors among users who have not yet experienced an adverse event. Multilevel interventions are needed to minimize the harm related to prescription opioids at a population level while maintaining adequate treatment of pain among individual patients.

Acknowledgments

The authors would like to acknowledge the contributions of Eden Anderson, MPH (c), an intern at the UDOH, to the work described in this manuscript.

Disclosures

The authors have no relevant financial relationships to disclose.

References