Introduction
Older adults in the US have experienced increases in opioid use disorder (OUD) and opioid overdose (OD) in the last 2 decades. Between 2000 and 2020, the age-adjusted rate of deaths involving synthetic opioids that are chemically made in laboratories (excluding methadone) increased by 14-fold (from 0.2 to 3 deaths per 100,000 standard population), and the rate of deaths involving natural opioids that come from poppy plant and semisynthetic opioids that are chemically modified from natural opioids increased by 5.7-fold (from 0.3 to 2 deaths per 100,000 standard population) among older adults [17]. The deaths include those from prescription and illicit opioids. The finding parallels prior studies that showed an upward trend in the diagnosis of OUD between 2013 and 2018 among older adults [27] and observed that older adults had the largest increase of all age groups in the incident diagnosis of OUD or nonfatal OD between 2006 and 2016 [34]. The increased trends in OUD and OD have prompted studies into understanding use of prescription opioids and other pain medications before the disease onset in older adults [33; 34] and their unsafe use of prescription opioids (e.g., high dose)[13; 14; 21; 22; 33]. However, the decline in opioid prescribing and unsafe prescription opioid use[3; 6]contrasts with the increases in OUD and OD [25], signaling a need to identify factors beyond the volume of opioid prescribing that may be associated with OUD and OD to inform interventions in older adults.
Unrelieved physical pain is the most commonly reported motivation for opioid misuse in older adults [24]. The 2015 US National Survey on Drug Use and Health showed that 4 of 5 older adults (85%) engaged in prescription opioid misuse to relieve pain, compared with 66.4% of their younger counterparts [24]. Prior studies also revealed that inadequately controlled pain is one of the major pathways to OUD [30] and is a potential risk factor for drug overdose death [32]. These findings, taken together, indicate that unrelieved pain may be a key element in the pathway from prescription opioid use to opioid misuse, leading to a high risk of OUD or OD among older individuals.
Today, limited data exist to quantify the associations of unrelieved pain with risk of OUD and OD among older adults with prescribed opioids. The lack of such data has hampered our understanding of the potential harms of continued opioid therapy when patients’ pain remains uncontrolled. Today, only 1 study explored the correlation between pain and prescription OUD using 2 waves (2001–2002 and 2004–2005) of the National Epidemiologic Survey on Alcohol and Related Conditions [5]. While demonstrating a positive correlation, the prior study is limited by outdated data, no consideration of confounders (eg, health-related behaviors), and the study population, over two-thirds of which were younger than 65 years old.
To address this research gap, we conducted a population-based cohort study to examine the associations of uncontrolled pain and high-impact pain with risk of OUD and OD among older adults with prescribed opioids.
Methods
Study Design and Source
This retrospective cohort study analyzed data from a national sample of older Americans who participated in the Health and Retirement Study (HRS) and consented to the linkage of their Medicare data from January 1, 2006, to December 31, 2021. HRS is a national survey that has been conducted biennially since 1992 among a nationally representative sample of community-dwelling Americans aged 50 years or older [28]. We used biennial HRS surveys to measure pain intensity and high-impact pain (key exposures) and important confounders, such as health-related behaviors (eg, smoking and drinking). Medicare claims data contain beneficiaries’ medical billing records for Parts A (inpatient), B (office-based visits), and D (prescription drugs, implemented on January 1, 2006). We used Medicare Parts A and B data to capture OUD and OD (primary outcomes) and Part D data to measure prescription opioids and important drug-related covariates (eg, polypharmacy). The Ohio State University’s Institutional Review Board approved this study and waived the requirement for obtaining patient informed consent because of the use of deidentified data. This study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline.
Study Sample
The study sample included HRS-Medicare participants aged 65 years or older who entered a cohort on the date of their first observed biennial HRS non-missing pain assessment (ie, cohort entry), had a diagnosis of chronic pain, and received ≥1 opioid prescription within 1 year before or on that date between January 1, 2008, and December 31, 2020. Within that sample, we further implemented the following exclusion criteria: (1) no continuous enrollment in Medicare Parts A, B, and D; (2) presence of enrolment in Medicare Advantage plan; (3) presence of OUD or OD (outcomes measurement); (4) use of medications for treatment of OUD; and (5) diagnosis of substance use disorder in 1 year before each cohort entry (baseline). The first two exclusion criteria were used to ensure the complete capture of baseline medical and prescription claims for measuring key variables of interest.
We created 2 independent cohorts to detect the risk of OUD and OD. For each cohort, participants were followed up from cohort entry until the incident outcome, death, disenrollment in Medicare Parts A and B, or study end (December 31, 2021), whichever came first. To ensure that the study population remained at risk for OUD or OD, we further required individuals in each cohort to have ≥1 prescription opioid fill during the follow-up. Figure 1 shows the sample selection details. The medications of interest and diagnostic and procedure codes for conditions and services considered in the sample selection are given in eTables 1 and 2.
Figure 1. Cohort Inclusion Flowchart for the Study Sample.
HRS represents Health and Retirement Study; OUD, opioid use disorder.
Key Exposures
The primary exposure was uncontrolled pain, as measured by HRS participants’ responses to 2 pain-related questions in each biennial wave: “Are you often troubled with pain?” and “How bad is the pain most of the time: mild, moderate, or severe?”[35] Patients who reported having moderate or severe pain were classified as having uncontrolled pain; otherwise, controlled pain [15]. The secondary exposure was high-impact pain, as measured by HRS-reported uncontrolled (vs controlled) pain, and whether pain made it difficult for participants to do their daily activities, such as household chores. Individuals who reported moderate to severe pain and had pain affecting daily activities were categorized as having high-impact pain; otherwise, having no high-impact pain [29]. Both pain measures of interest were treated as time-varying exposures in analyses because older adults’ pain experience changes over time [18].
Outcomes
The primary outcomes were incident encounters of OD and OUD (eTable 3). OD that included both fatal and nonfatal encounters was defined as an emergency department, inpatient, or outpatient encounter claim with International Classification of Diseases, Ninth or Tenth Revision, Clinical Modification (ICD-9 or ICD-10 CM) diagnosis codes of opioid poisoning [16; 33]. OUD was defined based on a conservative algorithm used by previous studies [4; 31] to increase true OUD cases, including receipt of (1) an inpatient encounter with an opioid dependence diagnosis; (2) 2 or more outpatient visits with an opioid dependence diagnosis; (3) an encounter with an opioid misuse or dependence diagnosis plus an emergency department or hospitalization with an injection drug use–related infection; or (4) an encounter with an opioid misuse or dependence diagnosis plus an inpatient or residential rehabilitation or detoxification care. The algorithm of all but the first criteria required 2 or more encounter claims that occurred within 3 months of each other during the follow-up, with the first observed encounter indicating the presence of an OUD event.
Covariates
A broad list of potential confounders was identified from the HRS survey and Medicare claims data. These variables included demographic characteristics, socioeconomic variables, type of pain conditions, health indicators, comorbidities, health care use, pain management, and medication utilization. Self-reported race and ethnicity were examined given racial differences in OUD and OD occurrence [19]. Race and ethnicity were grouped into 3 categories, Black, White, and other, which included Hispanic individuals or Asian, Pacific Islander, and Native American individuals, based on the Medicare–Research Triangle Institute race code available from Medicare claims data. Health indicators included overall self-rated health status (classified into poor or fair; good; and very good or excellent), ever smoked (yes vs no), ever consumed alcohol (yes vs no), body mass index (calculated as weight in kilograms divided by height in meters squared and classified into underweight [<18.5], normal weight [18.5–24.9], overweight [25.0–29.9] or obese [≥30]), physical impairment (difficulty in performing ≥3 of 7 activities of daily living) [11], and physical dependence (difficulty in performing >1 of 5 instrumental activities of daily living), all of which were derived from HRS survey data. Pain management included the receipt of medical procedures (eg, nerve blocks) and therapies (eg, physical and occupational therapy) for managing chronic pain, use of prescription nonopioids, use of adjuvant analgesics, opioid dosage and duration (the total sum of days), and use of long-acting opioids during the 1-year baseline. Medication utilization included polypharmacy and use of central nervous system medications. Finally, to account for increases in the diagnosis of OUD or OD, we adjusted for the year of the index wave.
We treated the following characteristics as time-varying covariates: marital status, income status, health indicators, health care use, use of procedure or therapy for chronic pain management, types of pain conditions, presence of substance use disorder (excluding OUD or OD), presence of cancer, mental health disorders, fall, urinary incontinence, and seizure, and the total number of comorbidities. The full list of covariates, their data source, and assessment periods are given in eTable 4.
Missing Data
Small percentages of participants had missing data on the HRS-assessed covariates (<1.3% at baseline and ≤1.0% during follow-up) (eTable 5). These missing data were replaced with the last observed value carried forward. Because all eligible participants continued to be followed for detecting OUD or OD after HRS discontinuation, there were 911 participants who dropped out from HRS before the end of their follow-up period. To handle the missing data on HRS-assessed pain measures and time-varying covariates between dropout from HRS and the end of follow-up, we carried over the values from the last HRS survey during the follow-up.
Statistical Analysis
We assessed baseline covariates of participants according to the status of pain control and high-impact pain, with a standardized mean difference (SMD) higher than 0.100 indicating covariate imbalance [2]. Differences in baseline variables between groups were accounted for via inverse probability of treatment weighting (IPTW), calculated as the inverse of the propensity score for the exposed group (eg, uncontrolled pain) and the inverse of 1 minus the propensity score for the nonexposed group (eg, controlled pain).
To account for time-varying confounders that simultaneously acted as confounders and intermediate variables, we used the marginal structural model (MSM) approach [23]. In this approach, we estimated the treatment weights by fitting pooled multivariable logistic regression models with the HRS-assessed pain exposure of interest as the dependent variable and baseline and time-varying variables as independent variables. Weights were truncated at the first and 99th percentiles to reduce the influence of outliers on estimates.
To examine the associations of pain exposures with risk of OUD or OD, we fitted a separate Cox regression model with MSM weights for each outcome to estimate the adjusted hazard ratios (AHRs) and their 95% CIs for the study vs comparison group. In all models, we used a robust sandwich-type estimator to account for within-individual correlations and applied the baseline HRS survey weights to extrapolate to the US older population.
We performed 5 sensitivity analyses to assess the robustness of the estimates by (1) additionally accounting for censoring due to death via inverse probability of censoring weighting (IPCW) in all models [7]; (2) restricting participants to those with prescription opioids for at least 30 days during the 1-year baseline period; (3) excluding OUD cases that occurred within the first year of follow-up to address possible reverse causality due to under- or delay-diagnosis of OUD (for the OUD cohort only); (4) examining the associations of pain exposures with a composite outcome consisting of OUD and OD; and (5) stratifying analyses for the composite outcome according to demographic characteristics and socioeconomic variables at baseline. For the last analysis, we analyzed the composite outcome to ensure a sufficient number of outcome cases within each subgroup to yield reliable results. All analyses were performed using SAS, version 9.4 (SAS Institute Inc). Statistical significance was set as P < .05, and all tests were 2-sided.
Results
We identified 3,104 older HRS-Medicare participants who had chronic pain and received at least 1 opioid prescription (mean [SD] age, 76.2 [7.9] years; 2062 [66.4%] female; 1042 [33.6%] male), 1 year before or at cohort entry between the 2008 and 2020 HRS wave (Table 1). The mean (SD) length of follow-up was 6.4 (4.1) years (median [IQR], 5.6 [2.9–9.5] years) for both the OUD and OD cohorts (eTable 6). During the follow-up period, 51.6% of participants died and were censored at the time of death. Proportions of participants lost to follow-up due to death differed between those with uncontrolled vs controlled pain and between those with vs without high-impact pain, which were adjusted via IPCW in a sensitivity analysis.
Table 1.
Clinical and Demographic Characteristics of the HRS-Medicare Study Participants, Overall and Stratified by Uncontrolled vs Controlled Pain
| Baseline Characteristic a | Participants, No. (%) | SMD b | |||
|---|---|---|---|---|---|
| Overall sample (n=3104) | With uncontrolled pain (n=1359) | With controlled pain (n=1745) | Before IPTW | After IPTW | |
| Age, y | |||||
| Mean (SD) | 76.2 (7.9) | 75.4 (7.9) | 76.9 (7.9) | 0.189 | 0.010 |
| 65–74 | 1484 (47.8) | 719 (52.9) | 765 (43.8) | 0.182 | 0.022 |
| 75–84 | 1087 (35.0) | 438 (32.2) | 649 (37.2) | 0.104 | 0.001 |
| ≥85 | 533(17.2) | 202 (14.9) | 331 (19.0) | 0.110 | 0.030 |
| Sex | |||||
| Male | 1042 (33.6) | 381 (28.0) | 661 (37.9) | 0.211 | 0.004 |
| Female | 2062 (66.4) | 978 (72.0) | 1084 (76.3) | ||
| Race and ethnicity | |||||
| White | 2351 (75.7) | 1019 (75.0) | 1332 (76.3) | 0.032 | 0.019 |
| Black | 421 (13.6) | 175 (12.9) | 246 (14.1) | 0.036 | 0.005 |
| Otherc | 332 (10.7) | 165 (12.1) | 167 (9.6) | 0.083 | 0.031 |
| Dual eligibility (yes) | 857 (27.6) | 439 (32.3) | 418 (24.0) | 0.187 | 0.005 |
| US region | |||||
| Northeast | 367 (11.8) | 179 (13.2) | 188 (10.8) | 0.074 | 0.003 |
| Northcentral | 798 (25.7) | 324 (23.8) | 474 (27.2) | 0.076 | 0.028 |
| West | 453 (14.6) | 212 (15.6) | 241 (13.8) | 0.051 | 0.047 |
| South | 1486 (47.9) | 644 (47.4) | 842 (48.3) | 0.017 | 0.007 |
| Educational level | |||||
| <High school | 845 (27.2) | 404 (29.7) | 441 (25.3) | 0.100 | 0.011 |
| High school or equivalent | 1611 (51.9) | 726 (53.4) | 885 (50.7) | 0.054 | 0.004 |
| ≥College | 648 (20.9) | 229 (16.9) | 419 (24.0) | 0.178 | 0.017 |
| Marital Status | |||||
| Married or partnered | 1576 (50.8) | 644 (47.4) | 932 (53.4) | 0.121 | 0.012 |
| Separated or divorced | 387 (12.5) | 200 (14.7) | 187 (10.7) | 0.120 | 0.019 |
| Widowed or never married | 1141(36.8) | 515 (37.9) | 626 (35.9) | 0.042 | 0.026 |
| Household income | |||||
| ≤$14,000 | 771(24.8) | 380 (28.0) | 391 (22.4) | 0.128 | 0.021 |
| $14,001-$26,000 | 769 (24.8) | 374 (27.5) | 395 (22.6) | 0.113 | 0.017 |
| $26,001-$50,000 | 754 (24.3) | 326 (24.0) | 428 (24.5) | 0.013 | 0.035 |
| ≥50,001 | 810 (26.1) | 279 (20.5) | 531 (30.4) | 0.229 | 0.003 |
| Self-reported health status | |||||
| Poor or fair | 1422 (45.8) | 863 (63.5) | 559 (32.0) | 0.664 | 0.024 |
| Good | 980 (31.6) | 361 (26.6) | 619 (35.5) | 0.194 | 0.007 |
| Very good or excellent | 702 (22.6) | 135 (9.9) | 567 (32.5) | 0.574 | 0.036 |
| Body mass index | |||||
| Underweight | 74 (2.4) | 32 (2.4) | 42 (2.4) | 0.003 | 0.006 |
| Normal weight | 888 (28.6) | 357 (26.3) | 531 (30.4) | 0.092 | 0.013 |
| Overweight | 1104 (35.6) | 462 (34.0) | 642 (36.8) | 0.059 | 0.026 |
| Obese | 1038 (33.4) | 508 (37.4) | 530 (30.4) | 0.143 | 0.016 |
| Ever smoker | 1741 (56.1) | 808 (59.5) | 933 (53.5) | 0.121 | 0.003 |
| Ever consumed alcohol | 1193 (38.4) | 452 (33.3) | 741 (42.5) | 0.191 | 0.009 |
| Any physical impairment | 518 (16.7) | 324 (23.8) | 194 (11.1) | 0.339 | 0.028 |
| Any physical dependence | 944 (30.4) | 536 (39.4) | 408 (23.4) | 0.351 | 0.022 |
| Chronic pain diagnosis | |||||
| Musculoskeletal | 3035 (97.8) | 1342 (98.7) | 1693 (97.0) | 0.120 | 0.031 |
| Neuropathic | 1217 (39.2) | 623 (45.8) | 594 (34.0) | 0.243 | 0.007 |
| Idiopathic | 416 (13.4) | 241 (17.7) | 175 (10.0) | 0.224 | 0.028 |
| Comorbidity | |||||
| Mental health | 809 (26.1) | 422 (31.1) | 387 (22.2) | 0.202 | 0.016 |
| Cancer | 842 (27.1) | 331(24.4) | 511 (29.3) | 0.111 | 0.002 |
| Diabetes | 1319 (42.5) | 610 (44.9) | 709 (40.6) | 0.086 | 0.000 |
| Cardiovascular disease | 1873 (60.3) | 829 (61.0) | 1044 (59.8) | 0.024 | 0.031 |
| Stroke | 358 (11.5) | 164 (12.1) | 194 (11.1) | 0.030 | 0.002 |
| Hypertension | 2465 (79.4) | 1088 (80.1) | 1377 (78.9) | 0.028 | 0.015 |
| Pulmonary condition | 1973 (63.6) | 882 (64.9) | 1091 (62.5) | 0.050 | 0.003 |
| Kidney disease | 675 (21.7) | 324 (23.8) | 351 (20.1) | 0.090 | 0.039 |
| Liver disease | 253 (8.2) | 131 (9.6) | 122 (7.0) | 0.096 | 0.011 |
| Gastrointestinal tract disorder | 992 (32.0) | 471 (34.7) | 521 (29.9) | 0.103 | 0.002 |
| Fall injury | 538 (17.3) | 252 (18.5) | 286 (16.4) | 0.057 | 0.028 |
| Urinary incontinence | 758 (24.4) | 378 (27.8) | 380 (21.8) | 0.140 | 0.022 |
| Seizure | 78 (2.5) | 41 (3.0) | 37 (2.1) | 0.057 | 0.000 |
| Neurodegenerative disorder | 294 (9.5) | 131 (9.6) | 163 (9.3) | 0.010 | 0.012 |
| Total number of comorbidities, Mean (SD) | 20.4 (12.1) | 21.7 (13.2) | 19.4 (11.2) | 0.188 | 0.012 |
| Health care utilization | |||||
| Any hospitalization | 857 (27.6) | 414 (30.5) | 443 (25.4) | 0.113 | 0.024 |
| Any ED visit | 1252 (40.3) | 580 (42.7) | 672 (38.5) | 0.085 | 0.001 |
| Pain management | |||||
| Use of nonpharmacological therapy for pain management | 1099 (35.4) | 520 (38.3) | 579 (33.2) | 0.106 | 0.004 |
| Use of adjuvant analgesic | 1055 (34.0) | 598 (44.0) | 457 (26.2) | 0.380 | 0.006 |
| Use of nonopioid drug | 2049 (66.0) | 975 (71.7) | 1074 (61.5) | 0.218 | 0.006 |
| Opioid dosage, MME/d | |||||
| <20 | 1582 (51.0) | 640 (47.1) | 942 (54.0) | 0.138 | 0.013 |
| 20–50 | 1095 (35.3) | 509 (37.5) | 586 (33.6) | 0.081 | 0.017 |
| >50 | 427 (13.8) | 210 (15.5) | 217 (12.4) | 0.087 | 0.005 |
| Use of long-acting opioid | 140 (4.5) | 109 (8.0) | 31 (1.8) | 0.292 | 0.066 |
| Use of opioids for ≥90 d | 774 (24.9) | 519 (38.2) | 255 (14.6) | 0.555 | 0.017 |
| Duration of opioid use, Mean (SD), d | 75.2 (111.1) | 111.5 (129.0) | 46.9 (84.8) | 0.592 | 0.017 |
| Medication use | |||||
| Polypharmacy | 2676 (86.2) | 1194 (87.9) | 1482 (84.9) | 0.086 | 0.017 |
| Use of other CNS drugs | 1176 (37.9) | 607 (44.7) | 569 (32.6) | 0.250 | 0.004 |
| Year of index HRS (cohort entry) | |||||
| 2008 | 1044 (33.6) | 481 (35.4) | 653 (37.4) | 0.066 | 0.019 |
| 2010 | 448 (14.4) | 202 (14.9) | 246 (14.1) | 0.022 | 0.020 |
| 2012 | 481 (15.5) | 202 (14.9) | 279 (16.0) | 0.031 | 0.063 |
| 2014 | 477 (15.4) | 217 (16.0) | 260 (14.9) | 0.030 | 0.024 |
| 2016 | 273 (8.8) | 113 (8.3) | 160 (9.2) | 0.030 | 0.026 |
| 2018 | 207 (6.7) | 88 (6.5) | 119 (6.8) | 0.014 | 0.040 |
| 2020 | 174 (5.6) | 56 (4.1) | 118 (6.8) | 0.117 | 0.022 |
Abbreviations: CES-D, Center for Epidemiological Studies-Depression; CNS, central nervous system; HRS, Health and Retirement Study, ED, emergency department; IPTW, inverse probability of treatment weighting; MME, morphine milligram equivalent; SMD, standardized mean difference
HRS-assessed characteristics and cognitive function were derived from the index HRS wave; Medicare-assessed utilization and medication use were measured during 1 year before the index HRS wave.
SMD greater than 0.100 indicates an imbalance between the groups for the measured characteristic.
Included Asian, Pacific Islander, and Native American individuals and Hispanic individuals.
Of 3104 participants with prescribed opioids, 1359 (43.8%) experienced uncontrolled pain and 1044 (33.6%) had high-impact pain at cohort entry (Table 1 and eTable 7). After IPTW, the distributions of all measured baseline characteristics were well balanced between groups, with SMD for characteristics less than 0.100 (Table 1 and eTable 7). The proportions of participants with uncontrolled pain and high-impact pain increased from 43.8% to 47.3% and from 33.6% to 36.3%, respectively, over time during the follow-up period (eTable 8).
Risk for OUD and OD
Among older adults with chronic pain who received prescription opioids, the crude incidence rate of OUD was 4.27 per 1000 person-years and of OD was 3.28 per 1000 person-years (Table 2). Patients with uncontrolled (vs controlled) pain had a higher crude rate of OUD (8.44 vs 1.23 per 1000 person-years) and of OD (3.90 vs 2.82 per 1000 person-years) (Table 2). As indicated in Figures 2A and 2B, the unadjusted cumulative probability of patients who remained OUD or OD event-free during the follow-up period was lower in patients with uncontrolled (vs controlled) pain. Similarly, patients with vs without high-impact pain had a higher crude rate of OUD (9.20 vs 1.95 per 1000 person-years) and of OD (4.19 vs 2.84 per 1000 person-years) (Table 2). The unadjusted cumulative probability of patients who remained OUD or OD event-free during the follow-up period was lower in patients with vs without high-impact pain (Figure 2C and 2D). In the MSM-adjusted Cox regression model, patients with uncontrolled (vs controlled) pain had a higher risk of OUD (AHR, 9.70 [95% CI, 4.56–20.63]; P < .001) and OD (AHR, 2.46 [95% CI, 1.30–4.66]; P < .001) (Table 2) and had lower cumulative probability of remaining OUD or OD event-free over time (Figure 3A and 3B). We also observed higher risk of OUD (AHR, 6.74 [95% CI, 3.76–12.08]; P < .006) and OD (AHR, 1.96 [95% CI, 1.07–3.60]; P = .029) (Table 2) and lower cumulative probability of remaining OUD or OD event-free for patients with vs without high-impact pain after accounting for confounders (Figure 3C and 3D). Similar results for associations of OUD and OD were obtained when additionally adjusting for censoring due to death via IPCW (Table 2).
Table 2.
Associations With Risk of Opioid Use Disorder and Opioid Overdose Among Study Participants by Pain Exposure Status
| Overall sample (n=3104) | Uncontrolled pain (n=1359) | Controlled pain (n=1745) | HR (95% CI) for patients with uncontrolled vs controlled pain | ||||||
|---|---|---|---|---|---|---|---|---|---|
| Outcome | Crude incident rate per 1000 PY (No. of events/py) | Crude | P-value | MSM-adjusteda | P-value | MSM-adjusted with IPCWb | P-value | ||
| Opioid use disorder | 4.27 (84/19673) | 8.44 (70/8295) | 1.23 (14/11378) | 7.86 (3.98–15.51) | <.001 | 9.70 (4.56–20.63) | <.001 | 9.75 (4.48–21.20) | <.001 |
| Opioid overdose | 3.28 (65/19805) | 3.90 (33/8465) | 2.82 (32/11340) | 1.37 (0.77–2.43) | .28 | 2.46 (1.30–4.66) | .006 | 2.68 (1.40–5.13) | .003 |
| Overall Sample (n=3104) | With high-impact pain (n=1044) | Without high-impact pain (n=2060) | HR (95% CI) for patients with vs without high-impact pain | ||||||
| Outcome | Crude incident rate per 1000 PY (No. of events/py) | Crude | P-value | MSM-adjusteda | P-value | MSM-adjusted with IPCWb | P-value | ||
| Opioid use disorder | 4.27 (84/19673) | 9.20 (58/6307) | 1.95 (26/13366) | 4.78 (2.77–8.24) | <.001 | 6.74 (3.76–12.08) | <.001 | 5.63 (2.94–10.77) | <.001 |
| Opioid overdose | 3.28 (65/19805) | 4.19 (27/6444) | 2.84 (38/13361) | 1.49 (0.83–2.69) | .18 | 1.96 (1.07 −3.60) | .029 | 2.12 (1.12–4.03) | .021 |
Abbreviations: HR, hazard ratio; IPCW, inverse probability of censoring weighting; MSM, marginal structural modeling; PY, person-years
Cox regression model with MSM regression that adjusted for baseline covariates and time-varying covariates.
Sensitivity analysis that additionally adjusted via IPCW for censoring due to death.
Figure 2. Unadjusted Risk of Opioid Use Disorder and Opioid Overdose During Follow-up Periods By Participants According to Pain Control and High-Impact Pain.
Unadjusted estimates of the survival curves by pain measures of interest are shown for opioid use disorder and opioid overdose outcomes. Participants were grouped into uncontrolled vs controlled pain and those with vs without high-impact pain according to pain measures at the index HRS wave.
Figure 3. Adjusted Risk of Opioid Use Disorder and Opioid Overdose During Follow-up Periods By Participants According to Pain Control and High-Impact Pain.
Marginal Structural Model-Adjusted estimates of the survival curves by pain measures of interest are shown for opioid use disorder and opioid overdose outcomes. Participants were grouped into uncontrolled vs controlled pain and those with vs without high-impact pain according to pain measures at each biennial HRS wave.
Sensitivity and Subgroup Analyses
Results consistent with those for the main analyses were obtained for the composite outcome that considered both OUD and OD (eTable 9). The sensitivity analysis excluding OUD cases (n=11) within the first year of follow-up generated similar results to those of main analyses for OUD (eTable 10). As well, the sensitivity analysis restricting the study sample to individuals with opioid refills for at least 30 days during the 1-year baseline yielded results similar to the main analysis for associations of OUD and OD (eTable 11). Similar results were also obtained in subgroups stratified according to baseline demographic and socioeconomic variables (eTable 12).
Discussion
This cohort study using a national sample of HRS participants with linked Medicare claims data is among the first to provide population-based data on pain control and risk of OUD and OD among older adults with chronic pain and prescribed opioids. Using an MSM approach to account for time-varying pain exposure and time-varying confounders, we observed that the risk of OUD and OD was substantially higher in patients with uncontrolled (vs controlled) pain and in patients with (vs without) high-impact pain. Findings in both sensitivity and subgroup analyses were similar to those of the main analyses. Our results demonstrate that uncontrolled pain and high-impact pain are associated with an increased risk of OUD and OD among older adults with prescribed opioids.
Unrelieved pain is the most frequently noted motive for prescription opioid misuse among older adults [12; 24] and has been considered a key risk factor for OUD or OD [32]. However, few data exist on the magnitude of pain intensity associated with OUD or OD in the older population. The only notable prior study, nearly 2 decades ago, indicated that moderate to severe pain is a predictor of prescription OUD among the noninstitutionalized US population 18 years of age or older [5], a result consistent with our observation. The present study is also among the first to estimate the prevalence of unrelieved pain after prescription opioid therapy among older adults, with nearly half having uncontrolled pain and one-third having uncontrolled pain that affected daily activities.
Our observed high risk of OUD and OD associated with uncontrolled or high-impact pain confirmed our hypothesis that inadequately controlled pain is a key component in the pathway from prescription opioid use to OUD or OD in older individuals. Persistent pain after opioid treatment may lead to an inability to reduce opioid use among older adults or may lead older adults to increase their use of opioids or to use prescribed opioids for reasons or in ways other than as prescribed, symptoms considered for the diagnosis of OUD.[8] These symptoms put older adults at risk of OD, especially when their tolerance to prescription opioids grows[9] or they have difficulty accessing opioid prescriptions and turn to nonprescribed opioids (ie, diverted or illicit sources)[1; 26].
Our study sample of HRS-Medicare older participants has comparable demographics to the overall older Medicare patients with chronic pain and receipt of prescription opioids in terms of mean age (76 vs 76 years old), female gender (66.4% vs 65.4%), white race (75.7% vs 81.4%), and residing in US South (47.9% vs 40.4%)[33]. Similarly, the proportion of participants reporting ever alcohol consumption in our HRS-Medicare sample (38.4%) is comparable to that (31.7%) of HRS participants with self-reported pain [35]. Regarding physical impairment, 16.7% of our HRS-Medicare older participants with chronic pain and prescribed opioids reported having difficulty in performing ≥3 of 7 activities of daily living (including picking up dimes, dressing, walking, bathing, eating, getting in/out of bed, and using the toilet), which is higher than that (6.0%) of HRS-Medicare participants who experienced hospitalizations [11].
The study findings underscore the importance of continuously monitoring pain intensity for older individuals with prescribed opioids for managing chronic pain [10]. Unrelieved pain after prescription opioids may be considered as a key component along with other established risk factors of OUD or OD [20; 32] to assess the need for continuing prescription opioids. For older individuals whose pain remains unrelieved after receiving opioid therapy, clinicians should consider alternative therapies to mitigate the risk of OUD or OD [10].
Strengths and Limitations
A notable strength of our study is the use of both Medicare claims and HRS assessment data to optimize adjustments of potential confounders (eg, health indicators and comorbidities) associated with pain intensity and OUD and OD outcomes. This study also has limitations. First, while we used a conservative definition of OUD developed by prior studies [4; 31] to increase the likelihood of including patients with true OUD, misclassification and especially underdiagnosis of OUD was possible. However, underdiagnosis is likely nondifferential between comparison groups (defined by pain status). Second, while great effort was made to establish the temporal relationships between uncontrolled pain and the onset of OUD, there exist concerns about reverse causality, where underdiagnosed OUD might lead to uncontrolled pain. We addressed such concerns in a sensitivity analysis that excluded OUD cases detected within the first year of follow-up; the analysis yielded similar results to the main analyses for OUD, suggesting that reverse causality is not a major problem. Third, because we relied on patient-reported pain exposures, self-report bias is possible. Fourth, Medicare prescription data do not capture self-paid prescriptions or opioids covered by non-Medicare programs. Fifth, our results remained subject to residual confounding from unknown and unmeasured confounders (eg, reasons and duration of chronic pain before cohort entry). The adjustment of the duration of chronic pain, for example, could attenuate our estimates. Sixth, our data did not provide information on individuals’ illicit opioid use and thus are limited in discerning whether opioid overdose was caused by prescribed opioids or illegally acquired opioids. Finally, our findings are generalizable only to HRS participants with linkage of Medicare fee-for-service claims data. Further studies should be conducted among Medicare Advantage enrollees, which has increased from 25% in 2010 to 54% in 2024 of all eligible Medicare beneficiaries [36].
Conclusions
The findings of this cohort study indicated that among older adults prescribed opioids, uncontrolled (vs controlled) pain and high-impact (vs without high-impact) pain) were associated with increased risk of OUD and OD. To reduce the risk of OUD and OD, clinicians should closely monitor pain control after prescribing opioids to older adults and consider alternative pain management, especially for older adults whose pain remains unrelieved.
Supplementary Material
Acknowledgments
Funding/Support:
This research was supported by grant R01HRS029001 to Dr Wei from the Agency for Healthcare Research and Quality
Role of the Funder/Sponsor:
The funder had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.
Footnotes
Declaration of interest: None
Data Access, Responsibility and Analysis:
Dr. Wei had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Data Sharing Statement:
Data for these analyses were licensed to the authors through a data user agreement with the Health and Retirement Study (HRS) and NIH/NIA. Individual researchers can access the data via a license through the HRS at https://hrs.isr.umich.edu/ and the NIA Data LINKAGE Program (LINKAGE) at https://www.nia.nih.gov/research/dbsr/nia-data-linkage-program-linkage.
References
- [1].Allen LD, Pollini RA, Vaglienti R, Powell D. Opioid Prescribing Patterns After Imposition of Setting-Specific Limits on Prescription Duration. JAMA Health Forum 2024;5(1):e234731. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [2].Austin PC. Balance diagnostics for comparing the distribution of baseline covariates between treatment groups in propensity-score matched samples. Stat Med 2009;28(25):3083–3107. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [3].Ball SJ, Simpson K, Zhang J, Marsden J, Heidari K, Moran WP, Mauldin PD, McCauley JL. High-Risk Opioid Prescribing Trends: Prescription Drug Monitoring Program Data From 2010 to 2018. J Public Health Manag Pract 2021;27(4):379–384. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [4].Barnett ML, Meara E, Lewinson T, Hardy B, Chyn D, Onsando M, Huskamp HA, Mehrotra A, Morden NE. Racial Inequality in Receipt of Medications for Opioid Use Disorder. N Engl J Med 2023;388(19):1779–1789. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [5].Blanco C, Wall MM, Okuda M, Wang S, Iza M, Olfson M. Pain as a Predictor of Opioid Use Disorder in a Nationally Representative Sample. Am J Psychiatry 2016;173(12):1189–1195. [DOI] [PubMed] [Google Scholar]
- [6].Bohnert ASB, Guy GP Jr., Losby JL. Opioid Prescribing in the United States Before and After the Centers for Disease Control and Prevention’s 2016 Opioid Guideline. Ann Intern Med 2018;169(6):367–375. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [7].Bolch CA, Chu H, Jarosek S, Cole SR, Elliott S, Virnig B. Inverse probability of treatment-weighted competing risks analysis: an application on long-term risk of urinary adverse events after prostate cancer treatments. BMC Med Res Methodol 2017;17(1):93. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [8].Boscarino JA, Hoffman SN, Han JJ. Opioid-use disorder among patients on long-term opioid therapy: impact of final DSM-5 diagnostic criteria on prevalence and correlates. Subst Abuse Rehabil 2015;6:83–91. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [9].Boyer EW. Management of opioid analgesic overdose. N Engl J Med 2012;367(2):146–155. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [10].Dowell D, Ragan KR, Jones CM, Baldwin GT, Chou R. CDC Clinical Practice Guideline for Prescribing Opioids for Pain - United States, 2022. MMWR Recomm Rep 2022;71(3):1–95. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [11].Greysen SR, Stijacic Cenzer I, Auerbach AD, Covinsky KE. Functional impairment and hospital readmission in Medicare seniors. JAMA Intern Med 2015;175(4):559–565. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [12].Han B, Compton WM, Blanco C, Crane E, Lee J, Jones CM. Prescription Opioid Use, Misuse, and Use Disorders in U.S. Adults: 2015 National Survey on Drug Use and Health. Ann Intern Med 2017;167(5):293–301. [DOI] [PubMed] [Google Scholar]
- [13].Hernandez I, He M, Brooks MM, Zhang Y. Exposure-Response Association Between Concurrent Opioid and Benzodiazepine Use and Risk of Opioid-Related Overdose in Medicare Part D Beneficiaries. JAMA Netw Open 2018;1(2):e180919. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [14].Jena AB, Goldman D, Weaver L, Karaca-Mandic P. Opioid prescribing by multiple providers in Medicare: retrospective observational study of insurance claims. BMJ 2014;348:g1393. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [15].Jenny Wei YJ, Chen C, Fillingim RB, DeKosky ST, Schmidt S, Pahor M, Solberg L, Winterstein AG. Uncontrolled Pain and Risk for Depression and Behavioral Symptoms in Residents With Dementia. J Am Med Dir Assoc 2021;22(10):2079–2086 e2075. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [16].Jones CM, Shoff C, Blanco C, Losby JL, Ling SM, Compton WM. Overdose, Behavioral Health Services, and Medications for Opioid Use Disorder After a Nonfatal Overdose. JAMA Intern Med 2024. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [17].Kramarow EA, Tejada-Vera B. Drug Overdose Deaths in Adults Aged 65 and Over: United States, 2000–2020. NCHS Data Brief 2022(455):1–8. [PubMed] [Google Scholar]
- [18].O’Neill A, O’Sullivan K, O’Keeffe M, Walsh C, Purtill H. The change of pain classes over time: a latent transition analysis. Eur J Pain 2020;24(2):457–469. [DOI] [PubMed] [Google Scholar]
- [19].Office of the Assistant Secretary for Planning and Evaluation USDoHaHS. Substance Use and Substance Use Disorders by Race and Ethnicity, 2015–2019, Vol. 2024, 2021. [PubMed] [Google Scholar]
- [20].Quality. AFHRa. Opioids in Older Adults: Adverse Event Red Flags, Vol. 2025. AHRQ Publication No. 24–0014-4-EF, 2023. [Google Scholar]
- [21].Ramachandran S, Salkar M, Bentley JP, Eriator I, Yang Y. Patterns of Long-Term Prescription Opioid Use Among Older Adults in the United States: A Study of Medicare Administrative Claims Data. Pain Physician 2021;24(1):31–40. [PMC free article] [PubMed] [Google Scholar]
- [22].Raman SR, Bush C, Karmali RN, Greenblatt LH, Roberts AW, Skinner AC. Characteristics of New Opioid Use Among Medicare Beneficiaries: Identifying High-Risk Patterns. J Manag Care Spec Pharm 2019;25(9):966–972. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [23].Robins JM, Hernan MA, Brumback B. Marginal structural models and causal inference in epidemiology. Epidemiology 2000;11(5):550–560. [DOI] [PubMed] [Google Scholar]
- [24].Schepis TS, Wastila L, Ammerman B, McCabe VV, McCabe SE. Prescription Opioid Misuse Motives in US Older Adults. Pain Med 2020;21(10):2237–2243. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [25].Schieber LZ, Guy GP Jr., Seth P, Losby JL. Variation in Adult Outpatient Opioid Prescription Dispensing by Age and Sex - United States, 2008–2018. MMWR Morb Mortal Wkly Rep 2020;69(11):298–302. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [26].Sedney CL, Haggerty T, Dekeseredy P, Nwafor D, Caretta MA, Brownstein HH, Pollini RA. “The DEA would come in and destroy you”: a qualitative study of fear and unintended consequences among opioid prescribers in WV. Subst Abuse Treat Prev Policy 2022;17(1):19. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [27].Shoff C, Yang TC, Shaw BA. Trends in Opioid Use Disorder Among Older Adults: Analyzing Medicare Data, 2013–2018. Am J Prev Med 2021;60(6):850–855. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [28].Sonnega A, Faul JD, Ofstedal MB, Langa KM, Phillips JW, Weir DR. Cohort Profile: the Health and Retirement Study (HRS). Int J Epidemiol 2014;43(2):576–585. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [29].Spector AL, Quinn KG, Wang I, Gliedt JA, Fillingim RB, Cruz-Almeida Y. More Problems, More Pain: The Role of Chronic Life Stressors and Racial/Ethnic Identity on Chronic Pain Among Middle-Aged and Older Adults in the United States. Chronic Stress (Thousand Oaks) 2023;7:24705470231208281. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [30].Stumbo SP, Yarborough BJ, McCarty D, Weisner C, Green CA. Patient-reported pathways to opioid use disorders and pain-related barriers to treatment engagement. J Subst Abuse Treat 2017;73:47–54. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [31].Wakeman SE, Larochelle MR, Ameli O, Chaisson CE, McPheeters JT, Crown WH, Azocar F, Sanghavi DM. Comparative Effectiveness of Different Treatment Pathways for Opioid Use Disorder. JAMA Netw Open 2020;3(2):e1920622. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [32].Webster LR. Risk Factors for Opioid-Use Disorder and Overdose. Anesth Analg 2017;125(5):1741–1748. [DOI] [PubMed] [Google Scholar]
- [33].Wei YJ, Chen C, Lewis MO, Schmidt SO, Winterstein AG. Trajectories of prescription opioid dose and risk of opioid-related adverse events among older Medicare beneficiaries in the United States: A nested case-control study. PLoS Med 2022;19(3):e1003947. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [34].Wei YJ, Chen C, Schmidt SO, LoCiganic WH, Winterstein AG. Trends in prior receipt of prescription opioid or adjuvant analgesics among patients with incident opioid use disorder or opioid-related overdose from 2006 to 2016. Drug Alcohol Depend 2019;204:107600. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [35].Whitlock EL, Diaz-Ramirez LG, Glymour MM, Boscardin WJ, Covinsky KE, Smith AK. Association Between Persistent Pain and Memory Decline and Dementia in a Longitudinal Cohort of Elders. JAMA Intern Med 2017;177(8):1146–1153. [DOI] [PMC free article] [PubMed] [Google Scholar]
- [36].Xu L, Welch WP, Ruhter J, Nguyen NX, Sheingold S, De Lew N, Sommers BD. Medicare Advantage Overview: A Primer on Enrollment and Spending: Issue Brief. Washington (DC), 2023. [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Availability Statement
Dr. Wei had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Data for these analyses were licensed to the authors through a data user agreement with the Health and Retirement Study (HRS) and NIH/NIA. Individual researchers can access the data via a license through the HRS at https://hrs.isr.umich.edu/ and the NIA Data LINKAGE Program (LINKAGE) at https://www.nia.nih.gov/research/dbsr/nia-data-linkage-program-linkage.