Economic Burden and Mortality Associated With Prescription Opioid Use

Prescription opioid use and misuse poses a significant public health challenge to the United States. In this dissertation we use the three-manuscript format to address some the areas of unmet research. Each manuscript has an abstract, introduction, background, methods, results, discussion, and conclusions sections. Manuscript 1: We used a retrospective cohort design to examine the association between the patterns of initial prescription opioid use for non-cancer pain and risk of all-cause mortality among an insured opioid-naïve patient population in the U.S. Multivariable Cox regression model was used to estimate the association of initial pattern of opioid use with all-cause mortality, adjusting for baseline covariates to control for confounding. We found that incident chronic opioid use was associated with an increased risk of all-cause mortality that persisted for up to 5 years after the initiation of opioid therapy. Manuscript 2: We used a cross-sectional study of the Rhode Island Prescription Drug Monitoring Program data to estimate the annual statewide spending for prescription opioids in Rhode Island. A generalized linear model with gamma distribution with a log link function was used to estimate the relative differences in per-patient annual adjusted average opioid prescription cost. We found that in 2015 the annual expenditure for opioid prescriptions in Rhode Island was $44,271,827. Commercial insurance bore the majority of the cost of prescription opioid use, but cost per patient was highest among Medicare beneficiaries. Manuscript 3: Using the 2015 Prescription Drug Monitoring Programs data for Rhode Island we examined the association between potential prescription opioid misuse and method of prescription opioid payment used. A multivariable log-binomial regression model was used to examine the risk of potential opioid misuse, controlling for sex, age group, type of opioid used, and concurrent benzodiazepine use. We found that patients on chronic opioid therapy who pay for some, but not all, opioid prescriptions in cash could be associated with potential opioid misuse only when the patient has other health insurance coverage.


INTRODUCTION
Pain is one of the most common reasons for an emergency department or outpatient office visit. [1][2][3][4] Opioid analgesics are often prescribed for acute pain in these settings because there is an established role of prescription opioid therapy in the management of moderate to severe acute pain and cancer pain. 5 However, the use of prescription opioids for chronic non-cancer remains controversial. [6][7][8][9][10][11] That notwithstanding, each year about 20% of adults in the United States are prescribed opioid analgesics for a variety of painful conditions ranging from acute injury to chronic non-cancer pain and terminal comfort measures, in part, because adequate and rapid pain relief has been used as a metric for patient satisfaction ratings. 12,13 Initial opioid therapy often intended by prescribers for short-term use often leads to unintended long-term use with adverse health outcomes including opioid misuse, opioid use disorder (OUD), overdose, and opioid-related deaths. [14][15][16][17][18] In recent years, deaths from drug overdose have increased dramatically, exceeding the number of deaths from motor vehicle accidents, to become the leading cause of accidental death in the United States. [19][20][21][22] The role of the initial pattern of opioid use in premature death is largely unknown.

BACKGROUND
Since 2013, mortality rates from drug overdose in the United States have frequently exceeded those from motor vehicle accidents and homicide. In 2015, drug overdose was responsible for over 50,000 deaths in the United States, with almost 63% related to opioid use -representing an age-adjusted opioid-involved death rate of 10.4 per 100,000 people. 23 More than 60% of all opioids involved in drug overdose deaths are prescribed by healthcare providers. 20, 24,25 The number of deaths attributed to prescription opioid has now exceeded the total number of deaths from suicide and motor vehicle accidents combined. 26 Over the past decade, the rate of opioid-related overdose death has increased by about 200%. 23 Certain groups of opioid users are particularly vulnerable to opioid overdose, including those who seek care from multiple providers to obtain the same prescription opioid analgesic 27,28 , those on high daily dosages of opioids [28][29][30][31][32] , those with a history of mental illness 33 , and Medicaid beneficiaries. 33,34 Prior studies have also established the use of long-acting opioids, especially at higher doses ( ≥ 50 morphine milligram equivalent [MME] per day) for chronic pain, use of ≥4 prescribers or pharmacies per year, and concurrent use of psychoactive medications including benzodiazepines and gabapentin as significant risk factors for opioid-related overdose death. 21,32, [35][36][37][38][39][40] Several studies have found concurrent benzodiazepine use in 30% to 60% of opioid-related deaths. [41][42][43] In accordance with changes in recent opioid prescribing guidelines, the Food and Drug Administration has issued a boxed warning limiting the concurrent use of benzodiazepines with opioids. 11,44,45 Another study of opioid-naïve patients who had low-risk surgery found that the use of prescription opioids within 7 days of surgery was associated with an increased risk of long term opioid use. 46 Other populations at greater risk of harm include patients with sleep disorders, renal and liver failure, older adults, patients with mental health conditions, and patients with alcohol or other substance use disorders. [47][48][49][50] The pattern of prescription opioid use varies from one individual patient to another depending on the indication, comorbidity, and local opioid prescribing norms.
Some patients require opioid medications on a regular basis; some require it asneeded, and others take it both regularly and as needed. A recent case-control study used the Veteran Health Administration data to examine the relationship between the patterns of opioid use and health outcomes found that receiving as-needed and regularly scheduled opioid therapy was not associated with increased risk of overdose. 32 Little attention has been given to the effects of the initial pattern of incident opioid use on the risk of death from all causes among patients receiving opioid therapy for non-cancer, non-terminal care. Current guidelines for opioid use address risks associated with chronic opioid therapy, which represents only about 5% of episodes of opioid use each year in the United States and the remaining patients take prescription opioids somewhat irregularly and the risk associated with this pattern of use warrants further investigation. 51 Understanding the risk of mortality associated with the incident chronic opioid use is an important step towards decreasing opioidrelated mortality. To address this gap in the evidence base, we examined the relationship between initial patterns of incident opioid use on the five-year mortality rate among opioid users without a cancer diagnosis.

Study design and data source:
This was a retrospective cohort study using de- Study population: All patients in the database who filled at least one opioid prescription were eligible for the study. Therapeutic class codes were used to identify all opioid prescriptions filled between January 1, 2010 and December 31, 2015 ( Figure   1). Prescriptions with improbable or missing quantity or days' supply, and bulk containing agents were excluded. We focused on opioid prescription fills among opioid-naïve patients, defined as having no opioid prescription or OUD claim in the first observed 6 months of enrolment in the database. 52 The index date was defined as the date of the first opioid prescription among opioid-naïve patients ( Figure 2). To allow for adequate ascertainment of baseline characteristics and initial pattern of prescription opioid use, we required patients to be continuously enrolled for at least 6 months before (i.e., pre-index baseline period) and 6 months after the index date (i.e., index period). Patients were excluded if they were younger than 18 years old as of the index date, or if their first opioid prescription was for a medication used to treat OUD, implying prior exposure to opioids was likely. 53 The study sample was further restricted to patients with no claims for any cancer diagnosis, or indication of palliative or hospice care. All patients were required to have a confirmed cancer diagnosis was defined as having at least two separate medical claims with a cancer diagnosis, with service dates at least 30 days apart. The cancer diagnosis codes could be in either a primary or secondary position in an inpatient or outpatient claim. 54 Hospice and palliative care were identified with ICD-9 and CPT codes. The final study sample consisted of 4,005,001 patients.

Exposure definition:
The index period (6 months following the first opioid prescription claim), was used to ascertain the initial pattern of prescription opioid use.
Based on the Consortium to Study Opioid use and Trends criteria, 38,55 patients were categorized either as daily (chronic) prescription opioid users if they had more than 90 non-overlapping days' supply of prescription opioids during the index period or as non-daily users otherwise. We computed the average daily morphine milligram equivalents (MME), and the number of providers and pharmacies used during the index period to fulfil opioid prescription requirements. Opioid prescription dosing information was converted to daily MME by multiplying the quantity of each prescription by the strength of the prescription, and multiplying this total by conversion factors published by the Center for Disease Control and Prevention (CDC). 56 For patients who received more than one opioid prescription on any given day, the MME of all prescriptions were added together. Based on recent CDC treatment guidelines, the average daily MME was categorized as < 50, 50-90, or >90 MME daily. 11 Covariate information: The dataset contains information on gender (male or female), type of insurance (commercial or Medicare), type of health plan, and US states that were grouped into five census regions. Age in years was calculated as of the index opioid prescription date. Clinical characteristics that may influence the initial pattern of prescription opioid use and/or time to mortality, such as Charlson comorbidity index (a common measure of overall medical comorbidity), common pain conditions, mental health disorders, surgical procedures, pregnancy, occurrence of sprains and fractures, and substance use disorders such as alcohol, smoking, and OUD were recorded during the baseline/index period using ICD-9 and ICD-10 codes. 57 Tobacco and alcohol were the most commonly used substances and were observed in 2.2% and 0.35% of the study cohort, respectively. For every 10,000 patients in the cohort, 4 had at least one medical claim for an overdose during the index period; 2 out of 10,000 had at least one claim for an OUD; and 1 out of 10,000 had at least one claim for an opioid overdose. Join and arthritis pain (15.2%) were the most common pain syndromes with a medical claim during the index period followed by back pain (7.5%). Almost 20% of patients had a pregnancy-related claim during the baseline period, 10% had a surgical claim, and 6% had a claim for a sprains and fractures. The proportion of patients who filled at least one antidepressant (13.4%) or a benzodiazepine (7.3%) was much higher than the proportion of claims for a mental health disorder (6%). Overall, almost 90% of patients had no comorbid physical conditions. However, 20% of patients on incident daily prescription opioid had one or more comorbid physical conditions compared to 10% among non-daily incident prescription opioid users. Similarly, on average, the number of opioid prescriptions, number of different types used, daily MME, number of providers and pharmacies involved were higher among incident daily opioid users than non-daily incident opioid users.
Patients were followed for up to 6 years (median of 2.5 years) with a total of 11,294,819 person-years during which 39,417 (1%) died (Table 2). Overall crude death rate was 349 deaths per 100,000 person-years. Among incident daily prescription opioid users, the crude incidence rate of all-cause mortality was 1,476 per 100,000 person-years compared to 317 per 100,000 person-years among non-daily users. The incidence rate difference (IRD) of 1,159 per 100,000 person-years represents the mortality rate among prescription opioid users that is attributable to a pattern of incident daily use, assuming daily users (i.e., exposed) would have had a mortality rate equal to non-daily users (i.e., unexposed) had they not been daily users.

DISCUSSION
We found that an initial pattern of incident chronic opioid use increased risk the risk of all-cause mortality, even after adjusting for potential confounding factors measured at baseline. Almost 3% of our study population was exposed to incident daily prescription opioid use which was comparable to that found in previous studies. 62 About 27.8% of patients were exposed to at least one prescription opioid medication during a six-year period for which the data was available in the database of a and non-daily prescription users in our study population represent the same study base.
We applied the same inclusion and exclusion criteria to all patients, and the exposure of interest and the outcome were measured in a similar manner. Because both groups had comparable observation times, it is unlikely that censoring was related to the initial pattern of prescription opioid use. We adjusted for measured potential confounding variables, including the use of restrictive quadratic splines for continuous variables to allow for a non-linear relationship with time-to-all-cause mortality in our model. Patients with a known history of OUD were also excluded and we adjusted for OUD and opioid overdose diagnosed during the index period because continued opioid use after an overdose is associated with a higher risk of repeated overdose and death. 67 Drug abuse, overdose, and OUD tend to occur at higher rates among long term opioid daily users. 38 After adjusting for demographic characteristics, substance use disorders, psychiatric medication use, and other medical and surgical conditions at baseline, we found that incident prescription opioid use was associated with a higher rate of mortality that persists long after treatment was initiated. Despite these challenges, to the best of our knowledge, this is one of the largest observational studies of the association between patterns of opioid use and mortality carefully designed to emulate the essential tenets of a randomized trial. 68 Pain is a common symptom among patients seen in office and emergency room settings and prescription opioid analgesics can be beneficial in when used appropriately. There is very little evidence of long term benefits for chronic noncancer pain or palliative care and several studies have reported significant risks associated with chronic opioid therapy. 6

LIMITATIONS
The results of this study should be considered with certain study limitations in mind.
First, we lacked data on socioeconomic status, and this likely limited our ability to include a sufficient set of covariates to adjust for confounding. For example, for the first time in nearly a quarter century, life expectancy decreased in the US in 2015, disproportionately affecting black men. 74,75 Racial differences in rates of opioidrelated deaths have been well-documented. 76 Second, we assumed that patients were opioid-naive if they did not fill any opioid prescription during the baseline period although this may not imply that they were never exposed to opioids previously.
Third, some patients may not have used their opioid prescription as recommended by the prescriber, and that any excess opioid medications from a previous prescription filled were completed before a subsequent prescription was taken. 38 Fourth, although outpatient pharmacy data reflect a complete and accurate representation of prescription opioid utilization, we did not account for prescriptions paid for in cash that may not be captured in the database. Lack of information of prescription opioid use at other health plans or prior to start of the database may have led to misclassification of some prevalent prescription users as incident user. Such misclassification was minimized by at least 6 months of enrollment with medical and pharmacy benefits prior to the index date. Fifth, ICD-9 and ICD-10 codes in administrative data may lack sensitivity and specificity in identifying certain lifestyle medical conditions such as tobacco and alcohol use.

CONCLUSIONS
A small but significant proportion (3%) of all opioid-naïve patients became incident daily prescription opioid users within the first 6 months of opioid therapy. Incident chronic opioid use was associated with a 2.6-fold increased risk of all-cause mortality that persisted for up to 5 years after the prescription opioid regimen was first started.
Our findings are consistent with current CDC guidelines to use the lowest effective prescription opioid dose for the shortest duration of treatment possible, especially among opioid-naïve patients.

19.
Centers for Disease C, Prevention.      an identity link function, and separately with a log link function, was used to estimate the absolute and relative differences in per-patient annual adjusted average opioid prescription cost, respectively, by potential predictors.

CONCLUSIONS:
This study provides the first estimate of the statewide direct cost burden of prescription opioid use using PDMP data and standardized pricing benchmarks. Total annual cost increased with age up to 65 years, mean daily dose, and concurrent use of benzodiazepines or stimulants. Commercial insurance bore the majority of the cost of prescription opioid use, but cost per patient was highest among Medicare beneficiaries. In addition to reducing harms associated with opioid overuse and misuse, substantial cost savings could be realized by reducing unnecessary opioid use, especially among middle-aged adults.

What is already known about this subject
What is Already Known About This Subject: • The overuse and misuse of prescription opioids is associated with increased morbidity and mortality, and places a significant cost burden on health • In 2016, the total annual expense for almost 18 million prescription drugs were filled at retail pharmacies in Rhode Island was about $1.2 billion.

What This Study Adds:
• Total annual retail expenditure for opioid analgesic prescriptions in Rhode Island for 2015 is estimate at $44,271,827.
• Total annual per-patient cost for opioid prescriptions is $17.65 higher among adults age 45-54 years, and $28.47 higher among patients of age 55-64 years as compared with those ages 65 years or older.
• Commercial insurance bears the majority of the cost of prescription opioid use but cost per patient is highest among Medicare beneficiaries.
• More than 10% of patients paid for at least one opioid prescription with cash but cash payment accounts for only 5.1% of the total annual prescription opioid expenditure.
• Among patients who also received prescriptions for benzodiazepines, annual cost for opioid dispensings was approximately twice as such as patients who did not receive benzodiazepines; while among patients who also received prescriptions for stimulants, annual cost for opioid dispensings was approximately 50% higher as compared with patients who did not receive stimulants.

INTRODUCTION AND BACKGROUND
Pain is one of the most common reasons for an outpatient office visit. [1][2][3] According to a 2011 Institute of Medicine report, approximately 100 million adults in the United States suffer from chronic pain each year. 4 Prescription opioid analgesics are commonly used to treat a variety of painful conditions that range from acute injury to palliative care for terminal illness. In 2012, U.S. health care providers issued over 259 million opioid prescriptions. 5 Higher rates of opioid prescribing over the past 2 to 3 decades may be attributed in part to historical changes in pain management guidelines, the Joint Commission requirements for routine pain assessment, promotion by the pharmaceutical industry, and changes in reimbursement and out-of-pocket payment for prescription opioids. [6][7][8][9][10] Although use of prescription opioid analgesics is often clinically necessary, their addictive and euphoric properties easily result in misuse and often lead to opioid use disorders. In recent years, deaths from drug overdose have increased dramatically, exceeding the number of deaths from motor vehicle accidents, and have become the leading cause of accidental death in the United States. 5,[11][12][13] In 2015 alone, drug overdose was responsible for 52,404 deaths, with 33,091 (63.2%) associated with opioids. More than 60% of opioids involved in drug overdose deaths are prescription opioids. 14,15 Benzodiazepines and opioids are often prescribed together, which has the potential risk of respiratory depression and overdose death. 16 The data provided by the Rhode Island Department of Health for our study included only de-identified patient, prescriber, and dispensing pharmacy information.
Available information included patient age (in years), gender, a unique prescriber and dispensing pharmacy identifier, the National Drug Code (NDC) number, product name, strength, formulation, and therapeutic class code of each prescription, as well as the number of days supplied, metric quantity dispensed, method of payment, and date the prescription was filled. The pharmacist estimated days supply based on the quantity prescribed and the daily dosage. The payment type field was populated by the dispensing pharmacist, and a payment type of "commercial" was presumably incorrectly assigned for many older adults who were more likely enrolled in a Medicare plan. Therefore, for patients who were aged 65 years or older with a payment type of commercial, their payment type was reclassified as Medicare. We hypothesized that the total annual expenditure for opioid prescriptions was associated with the use of benzodiazepines or stimulants, use of higher daily doses of opioids, and use of multiple providers and pharmacies. This study determined if opioid prescription expenditure was associated with patient age groups, gender, or payment type. Use of benzodiazepines or stimulants was defined as the use of 1 or more prescriptions of either a benzodiazepine or a stimulant within the study year. The dosage of each prescription opioid dispensed was converted to its morphine milligram equivalent (MME) per day using NDC-based conversion factors published by the Centers for Disease Control and Prevention's National Center for Injury Prevention and Control in June 2015. [35][36][37][38] Less than 1% of opioid prescriptions did not have enough information to calculate the daily MME. Finally, a measure of potential health system abuse was defined as patients who received prescriptions for opioids from 5 or more prescribers and filled by 5 or more pharmacies within the 12-month period With patients as the unit of analysis, a generalized linear model with gamma distribution with an identity link function, and separately with a log link function, was used to estimate the absolute and relative differences in per-patient annual adjusted average prescription opioid cost, respectively, by potential predictors. 40,41 Independent variables included in the model were age group, gender, payment type, mean daily MME per patient, use of benzodiazepines or stimulants, and conditional on the link function. The gamma family was selected, which has a constant coefficient of variation and assumes that the variance is proportional to the square of the mean. 41,42 All analyses were performed with SAS version 9.4 (SAS Institute, Cary, NC), and all statistical tests were 2-sided and performed at the 0.05 significance level. This study was approved by the institutional review board at the University of Rhode Island.

RESULTS
The      Coefficients of gamma regression and standard errors estimated using the log link function; c If age ≥ 65 and Commercial Ins. then payment method was reclassified as Medicare; d Includes worker's compensation, Indian Nation, the Veterans Health Administration, Tricare, other federal sources; e,f Use of other benzodiazepines and stimulants defined as one of more prescriptions in a 12 month period. f 5/5/12/ Criteria: Opioid prescriptions from at least 5 different pharmacies and 5 different prescribers during the 12 month period.  Excess relative risk due to interaction was 0.32 on the risk ratio scale which means there is a positive modification the association between cash payment and the risk of potential prescription opioid misuse on the additive scale. In both age groups, exclusive cash payment was associated with a much lower risk of potential prescription opioid misuse.

INTRODUCTION
The use of chronic opioid therapy (COT) for chronic non-cancer pain has increased tremendously over the last three decades despite lack of evidence for long-term effectiveness. There is evidence that excessive opioid prescribing has contributed to the current opioid epidemic since misuse of prescribed opioids has been associated with accidental opioid overdose-related deaths. 1,2 The risk of prescription opioid misuse is the primary reason why almost all states now monitor the use of controlled substances using their prescription drug monitoring programs (PDMP). Validated criteria have been developed to help identify patterns of prescription opioid consistent with prescription opioid misuse among patients on chronic opioid therapy using administrative claims data. 3 Limited data suggest that information on the method of payment may help detect potential prescription opioid misuse and contribute to improved opioid prescribing among physicians by identifying patients at high-risk for overdose. Almost all state PDMPs collect information on method payment, and a recent study showed that the proportion of controlled substances paid for in cash varies by about 3-fold across states. 4 High percentages of opioid prescriptions paid for in cash are generally regarded with suspicion because patients who seek prescriptions of the same controlled substances from multiple prescribers are more likely to pay for their prescriptions in cash. 5 Furthermore, the National Association of Boards of Pharmacy, PDMP administrators and law enforcement investigators believe that cash payment indicates potential prescription opioid misuse, especially when the patient has other health insurance coverage. 6 "Pill mills" also accept cash payment as a way to avoid detection. 7 The practice of concealing potential opioid misuse could have significant implications for drug utilization reviews since these evaluations are often restricted to specific health systems and could miss opioid prescriptions paid in cash.
Understanding how cash payment relates to potential prescription opioid misuse may shed light on the public health magnitude of the problem and provide new ways to identify potential opioid misuse from a payer perspective.

BACKGROUND
Each year about twenty percent of adults in the United States are prescribed opioid analgesics for a variety of painful conditions ranging from acute pain due to injury to chronic cancer pain and pain occurring during terminal illness. Initial opioid therapy often intended by prescribers for short-term use sometimes leads to unintended longterm use, with adverse health outcomes including opioid misuse, overdose, and opioid-related deaths. [8][9][10][11][12] Prescription opioid misuse (POM) often precedes heroin use. 13 The rise in POM has been associated with a corresponding increase in opioid prescribing for therapeutic uses especially among patients with chronic pain. A recent study of controlled substance prescribing patterns in eight US states found that opioid analgesics were prescribed twice as often as stimulants or benzodiazepines. 4 A relatively small proportion of prescribers were responsible for a large proportion of prescriptions. For example, overall 10% of prescribers accounted for more than 50% of opioid prescribing and in one state 25% of opioid prescription were prescribed by 1 percent of prescribers. Prescription opioids for misuse are most commonly obtained from a family member or friend, but patients at highest risk of overdose are as likely to get them legally from a licensed clinician. 13 There is no universally accepted definition of prescription opioid misuse with adequate support for its use. 3 According to a recent survey, the most common reason for POM among patients on opioids is to relieve physical pain suggesting that misuse most commonly occurs among patients on COT. 15 Compared with opioid-naïve patients, the use of cash payment and POM behavior were more common among patients on COT, especially those taking schedule II opioids. 5 Several risk factors have been associated with POM among patients on COT; these include being young or middle aged adults, white race, and a history of mental health disorders (including depression or posttraumatic stress disorder), and a family or personal history substance use disorder, including tobacco use. [16][17][18][19][20][21][22][23][24][25][26] The risk of POM was the primary reason why states monitor controlled substances using PDMP and in recent years, there has been a growing interest in using PDMP data for epidemiologic and health services research despite absence important clinical and demographic variables. 27 One of the unique features of most PDMP is their ability to capture almost all outpatient opioid prescriptions filled at a retail pharmacy with the method of payment used by the patient irrespective of their insurance coverage.
The decision to pay for an opioid prescription with cash depends on many factors including promotional cheaper generic formulations, lack of health insurance or prescription drug coverage, the existence of a gap in prescription drug coverage especially among the elderly, or simply to minimize information made accessible to drug utilization reviewers. 28     There is limited evidence that patients who seek prescriptions of the same controlled substance from multiple prescribers are more likely to pay for their prescriptions in cash. 5 We found that 17% of patients on long-term opioid therapy who had insurance coverage during the study year paid for at least one opioid prescription with cash. 28  These studies and ours suggest that POM is a serious public health problem in the United States. Our study suggests that use of cash payment when a patient has other health insurance coverage may be a surrogate marker of potential POM.

LIMITATIONS
Our study has some limitations. First, being a cross sectional study we could only demonstrate associations between method of payment and potential POM, not causality. We assumed that our modified validated measure of opioid misuse correctly identified patients who actually misused opioids. However, we did not have any clinical documentation or patient reported opioid use to ascertain the way prescribed opioids were used. Multiple prescribers may be part of a coordinated care within a large group practice and thus not necessarily represent risk of opioid misuse. Second, we could not account for opioid prescriptions filled in neighboring states since the data source does not provide that information. In a recent study looking at potential misuse of medications used to treat attention deficit hyperactivity disorder, patients with potential misuse behavior traveled longer distances and visited more states to fill their medications than patients without similar behaviors. 40 It is estimated that about 20% of patients who misuse opioid prescriptions get their prescriptions filled in more than one state compared to only 4% of patients with no evidence of opioid misuse. 41 Third, the method of payment may not have captured Medicare or Medicaid payment accurately.
Reclassification of patients aged 65 years and older with commercial insurance as Medicare may have led to some misclassification of some methods of payment.
However, such misclassification did not involve cash payment and therefore would not be expected to affect our results. Fourth, we lacked information on socioeconomic status, indications for opioid prescription, and use of illicit opioids.

CONCLUSIONS
Our study provides one of the first estimates of the relationship between cash payment and potential POM among patients on COT using a data source that directly captures the method of payment irrespective the patient's health insurance. Seventy percent of patients with insurance having drug coverage paid for at least one opioid prescription with cash. While there may be legitimate reasons to pay for opioid prescriptions with cash, this study suggests that patients on COT who pay for some, but not all, opioid prescriptions in cash should raise special concerns for potential opioid misuse only when the patient has other health insurance with drug coverage. Among patients aged 65 years and older, the strength of the association between cash payment and potential POM is stronger among women than men. Further research is needed to determine the public health relevance of these findings and whether cash payment is associated with adverse health outcomes.

29.
CDC. Pharmacists: on the front lines addressing prescription opioid abuse and overdose. Atlanta, GA: Center for Disease Control and Prevention; 2016.

FUTURE RESEARCH WORK
In the first manuscript, we used a baseline model to evaluate the association of the initial pattern of prescription opioid use on the risk of all-cause mortality. We plan to perform additional analysis with time-varying exposure and time-varying confounding using marginal structural models. Prescription opioids use tends to vary over time and standard approaches for adjustment of confounding discussed here can be biased in the presence of time-dependent confounders that are also affected by previous prescription opioid exposure. In a situation like the one just described, marginal structural models may allow for improved adjustment of confounding.