Predictors of persistent opioid use in Australian primary care: a retrospective cohort study, 2018–2022

Abstract

Objective

To examine the predictors of persistent opioid use (“persistence”) in people initiating opioids for non-cancer pain in Australian primary care.

Design

A retrospective cohort study.

Setting

Australian primary care.

Subjects

People prescribed opioid analgesics between 2018 and 2022, identified through the Population Level Analysis and Reporting (POLAR) database.

Methods

Persistence was defined as receiving opioid prescriptions for at least 90 days with a gap of less than 60 days between subsequent prescriptions. Multivariable logistic regression was used to examine the predictors of persistent opioid use.

Results

The sample consisted of 343 023 people initiating opioids for non-cancer pain; of these, 16 527 (4.8%) developed persistent opioid use. Predictors of persistence included older age (≥75 vs 15–44 years: adjusted odds ratio: 1.67, 95% CI: 1.58–1.78); concessional beneficiary status (1.78, 1.71–1.86); diagnosis of substance use disorder (1.44, 1.22–1.71) or chronic pain (2.05, 1.85–2.27); initiation of opioid therapy with buprenorphine (1.95, 1.73–2.20) or long-acting opioids (2.07, 1.90–2.25); provision of higher quantity of opioids prescribed at initiation (total oral morphine equivalents of ≥750 mg vs <100 mg: 7.75, 6.89–8.72); provision of repeat/refill opioid prescriptions at initiation (2.94, 2.77–3.12); and prescription of gabapentinoids (1.59, 1.50–1.68), benzodiazepines (1.43, 1.38–1.50), and z-drugs (eg, zopiclone, zolpidem; 1.61, 1.46–1.78).

Conclusions

These findings add to the limited evidence of individual-level factors associated with persistent opioid use. Further research is needed to understand the clinical outcomes of persistent opioid use in people with these risk factors to support the safe and effective prescribing of opioids.

Introduction

Opioids are commonly prescribed for the treatment of non-cancer pain. A systematic review of international studies conducted between 1990 and 2017 reported that approximately one quarter of people with chronic non-cancer pain take opioids.¹ Clinical benefits of opioids for non-cancer pain, however, are unclear, and there is a growing body of evidence demonstrating increased risks of adverse outcomes with long-term opioid use.² Nevertheless, the global consumption of opioid analgesics has been increasing over the past few decades, with North America, western and central Europe, and Oceania being the largest contributors.³ In the United States, the rate of opioid prescribing declined by 44% in the past decade, but harms from illicit opioid consumption, including overdose and deaths, have risen dramatically during the same period.⁴ In Australia, between 2013 and 2017, more than 1.9 million people initiated opioids every year, with most opioids used for the treatment of non-cancer pain.⁵ In parallel to the increased use of prescription opioids, a rise in opioid-related harms, including fatal and nonfatal overdose, has also been reported.⁶ In Australia, opioids accounted for more than half of all unintended deaths in 2017, 56% of which involved prescription opioids.⁷ With the escalating demand for chronic pain management in primary care and ongoing risks of prescription opioid use and harms,⁸ it is critical to understand the predictors of long-term opioid use in these settings.

In efforts to curb prescription opioid use and harms, a considerable number of national and state-level policies and interventions have been seen in countries with high opioid consumption, including the United States, Canada, and Australia.^9,10 In Australia, codeine was up-scheduled from “pharmacist-only” to “prescription-only” medicine in February 2018, mandatory use of a prescription drug monitoring program was implemented in Victoria in April 2020, and smaller opioid pack sizes were introduced in June 2020.^11,12 In the past decade, reductions in opioid use have been seen in the United States and Canada, which might be attributed to similar supply-reduction interventions. In the United States, the proportion of people undergoing opioid taper has increased between 2008 and 2017,¹³ and the prevalence of people receiving high-dose long-term opioid therapy has decreased.¹⁴ In Canada, the rate of opioid dispensing declined between 2005 and 2018.¹⁵

Many of these policies aim to reduce the number of people who transition to persistent (or long-term) opioid use; however, the rate of development of persistent opioid use, and which populations are most at risk of developing persistent opioid use in the context of these system-level interventions on long-term opioid use, remain unknown. In addition, few international studies have linked certain pain conditions (eg, musculoskeletal pain), psychiatric comorbidities (eg, substance use disorder), and concurrent use of psychotic medicines (eg, gabapentinoids) with long-term opioid use.^16,17 Further understanding of the factors that lead to long-term opioid use in Australian primary care, where a majority of opioids are prescribed, can inform preventive measures targeted to specific subpopulations at risk of opioid-related harms.⁵ Therefore, the aim of the present study was to examine the predictors of persistent opioid use (“persistence”) in people initiating opioids for non-cancer pain in Australian primary care. Consistent with commonly used pharmacoepidemiological definitions, the term “initiate” refers to commencement of opioid therapy and not the intention to use opioids in the long term, and the term “persistent use” refers to a particular pattern of long-term prescription and does not indicate opioid misuse or dependence.

Methods

Study design and setting

We undertook a retrospective cohort study of people prescribed opioid analgesics between 2018 and 2022. This study used the Population Level Analysis and Reporting (POLAR) dataset, containing electronic health data of people who consult general practitioners from 562 general practices across Eastern Victoria, representing 70% of the practices in the area, and across a broad range of urban, outer urban, and rural settings.¹⁸ The database consisted of routinely collected electronic medical records between January 2017 and July 2023, including demographic data, diagnoses, and all medication prescriptions, irrespective of subsidy. Further information on the database and the opioid cohort can be found in the Opioid Prescribing, Policy Impacts and Clinical Outcomes (OPPICO) protocol and the cohort profile.^19,20 This study was reported according to the RECORD-PE (Reporting of Studies Conducted Using Observational Routinely Collected Health Data—Pharmacoepidemiology) checklist.²¹

Study sample

This study included people ≥15 years of age at POLAR patient cohort entry in 2018 who initiated opioid analgesics between January 1, 2018, and July 31, 2022 (Figure 1). Age was provided in 5-year age brackets. Thus, to include all adults (≥18 years) in our sample, we chose to include the youngest age bracket (15–19 years at cohort entry, which comprised <3.8% of the overall sample), though this might have resulted in the inclusion of a small proportion of people who were adolescents at the time of opioid initiation in our sample. To include people initiating opioid therapy, we identified people who had no prior opioid prescription in the 12 months before the “index date”—the date of the first opioid prescription during the study period. For a small proportion of people who initiated multiple times during the study period, we assessed only their first opioid initiation, consistent with existing literature.^22,23 The exclusion criteria were (1) diagnosis of terminal illness or cancer (other than non-melanoma skin cancer) during the 12 months before the index date (inclusive) and during the exposure assessment window, if the diagnosis was recorded before the assessment into the persistent or nonpersistent cohort; (2) prescription of opioid replacement therapy (eg, buprenorphine/naloxone, buprenorphine depot injections, or methadone oral liquid) during the 12 months before the index date (inclusive); or (3) having no clinic activity record (eg, general practitioner consultation, prescription, or immunization record) in the 12 months before and 12 months after the index date, to confirm their ongoing enrollment within the included practices.

Figure 1.

(I) Visualization of cohort entry and exposure and covariate assessment windows. (II) Illustration of the study time frame. ^a Age at cohort entry in 2018. ^b Excluded people with diagnosis of cancer/terminal illness during the exposure assessment window if the diagnosis was recorded before assessment into the persistent or nonpersistent cohort. Index date refers to the date of first opioid prescription or opioid initiation. ORT= opioid replacement therapy. Figure adapted from Schneeweiss et al. (2019).⁵¹

Measure of persistence

Persistent opioid use was defined as having opioid prescriptions beyond 90 days from the index date, with a gap of fewer than 60 days between subsequent prescriptions (Figure 2).²⁴ All other people in our sample were categorized as having nonpersistent (or short-term) opioid use. The proportion of people developing persistent opioid use was determined by dividing the number of people who developed persistent opioid use by the total sample (ie, the total number of people initiating opioid therapy).

Figure 2.

Diagram illustrating the definition of opioid initiation and persistence. Index date was defined as the date of first opioid prescription (or opioid initiation). Rx denotes each opioid prescription. The squared Rx indicates the last prescription between 0–90 days, and the circled Rx indicates the first prescription beyond day 90. There may be more than 2 prescriptions between 0–90 days (not shown in the diagram). People were categorized into the persistent cohort if they had at least one prescription beyond 90 days from the index date (ie, presence of a circled Rx) with <60-day gap between subsequent prescriptions during 0–90 days and <60-day gap between the squared and circled Rx.

Opioid prescription

All opioids commonly used as analgesics in Australia were examined, including long-acting opioids such as transdermal formulations, modified-release (MR) oral formulations (eg, hydromorphone MR tablet and morphine MR tablet), and methadone (classified because of its long half-life) (Supplementary Material: Table S1). Formulations primarily used as opioid replacement therapy or as antitussive agents (eg, dihydrocodeine and cold and flu medicines containing codeine, such as paracetamol/codeine/pseudoephedrine combination tablets) were excluded. Opioid doses were converted to oral morphine equivalents (OME) via published conversion factors.²⁵ The total opioid quantity supplied on a prescription was calculated by multiplying opioid strength by quantity prescribed by conversion factor.²⁶ Where multiple prescriptions were provided on the index date, opioid quantities on each prescription were summed to determine the total opioid quantity at initiation. On the basis of the standard duration of opioid supply in Australia and the definitions used in prior studies, repeats or refills that were ordered on a prescription were considered repeat prescriptions if ordered after 28 days for transdermal and slow-release formulations and after 14 days for immediate-release formulations.²³

Predictors of interest

Demographic characteristics, comorbidities, opioid characteristics, prescription of other medicines, and year of opioid initiation were examined as predictors of persistent opioid use. Specific predictor variables have been chosen on the basis of prior literature that demonstrated an association with long-term opioid use.^17,27,28 Demographic characteristics included age, sex, and concessional beneficiary status (which enables access to government-subsidized health services and medicines). Being in receipt of concessional benefits might indicate inability to undertake full-time, part-time, or casual work, or any disability or illness, or being in receipt of an Age Pension or other social security allowance. For these reasons, people who hold a Commonwealth Seniors Health Card, a Department of Veterans’ Affairs Card, a Healthcare Card, or a Pensioner Concession Card are eligible to access government-subsidized health services and medicines.²⁹ Postcodes were used to approximate remoteness (measured by the Modified Monash Model³⁰) and geographically derived socioeconomic status (measured by Socio-Economic Indexes for Areas [SEIFA] deciles, which are ranked on a scale of 1 to 10, with 1 indicating the most disadvantaged).¹⁹

Comorbidities recorded during the 12 months before the index date (inclusive) were examined. The Modified Cambridge Multimorbidity Score, a validated measure for predicting mortality in primary care,³¹ was used with small modifications of the algorithm to reflect the standard practice in Australia (Table S2). Specific comorbidities examined included depression or anxiety, sleep disorder, substance use disorder, headache/migraine, chronic pain, and musculoskeletal conditions. Consistent with methods used in prior literature,¹⁷ prescription records were used as a proxy for diagnoses, in addition to the diagnosis records, to fully capture individuals’ comorbidities. Diagnostic codes used to identify the comorbid conditions have been derived from the POLAR study protocol and prior literature^19,32,33 (see footnotes in Table S3.1–S3.6).

We examined the following characteristics of opioids prescribed at the index date: opioid type (ie, buprenorphine, codeine), short-acting vs long-acting opioids, provision of repeat prescriptions, routes of administration, and total opioid quantity prescribed at initiation. Other medicines prescribed during the 12 months before the index date (inclusive) were examined: paracetamol, nonsteroidal anti-inflammatory drugs (NSAIDs), gabapentinoids, benzodiazepines, and z-drugs (Table S4). The year of opioid initiation was examined as a predictor to examine the temporal changes in the initiation of persistent opioid use.

Data analyses

We used descriptive statistics to determine the proportions of people initiating persistent opioid use. Multivariable logistic regression was used to estimate odds ratios (ORs) and 95% confidence intervals (CIs) to examine the association between the predictors and initiation of persistent opioid use. We tested for multicollinearity between variables in the regression model using the variance inflation factor; no significant interaction was observed. All analyses were conducted in Stata Statistical Software Release 17 (StataCorp LLC; College Station, TX, USA). Graphical illustrations were created in Microsoft PowerPoint Version 16.70.

Sensitivity analysis

Given the lack of consensus in defining persistent opioid use, we undertook a sensitivity analysis using an alternative definition of a gap of <30 days between subsequent prescriptions instead of <60 days to define continuous opioid prescriptions.³⁴

Results

Sample characteristics

The sample consisted of 343 023 people who initiated opioids for non-cancer pain between January 2018 and July 2022 (Figure 3). More than half of the sample were less than 55 years of age (58.1%) and were female (58.7%) (Table 1). Approximately a third (37.1%) of the sample were concessional beneficiaries, and more than three quarters resided in metropolitan areas (80.6%). Approximately one third lived in postcodes that were categorized as being in the top 2 “least socioeconomically disadvantaged” deciles (34.9%), with more than half (60.6%) living in postcodes that were categorized in the top 4 “least disadvantaged” deciles, measured on a scale of 1–10 SEIFA deciles. The mean Cambridge Multimorbidity Score of the overall sample was 0.46 (standard deviation: 0.61). In the 12 months before opioid initiation, almost a fifth of the sample was diagnosed with musculoskeletal conditions (16.6%), and small proportions were diagnosed with depression or anxiety (6.0%) and headache/migraine (3.4%).

Figure 3.

Derivation of study cohort.

Table 1.

Baseline characteristics of people initiated on opioid analgesics and predictors of persistence.

	People with persistent opioid use (n = 16 527)	People with nonpersistent opioid use (n = 326 496)	Total sample (n = 343 023)	Adjusted odds ratioa(95% confidence intervals)
Demographic characteristics
Age, yearsb
15–44	3159 (19.1%)	139 187 (42.6%)	142 346 (41.5%)	Reference
45–54	2269 (13.7%)	54 646 (16.7%)	56 915 (16.6%)	1.40 (1.32–1.48)
55–64	2665 (16.1%)	50 841 (15.6%)	53 506 (15.6%)	1.41 (1.33–1.49)
65–74	3112 (18.8%)	43 580 (13.4%)	46 692 (13.6%)	1.24 (1.17–1.31)
≥75	5321 (32.2%)	38 236 (11.7%)	43 557 (12.7%)	1.67 (1.58–1.78)
Not specifiedl	<10 (<0.01%)	<10 (<0.01%)	<10 (<0.01%)	–
Sex
Male	6291 (38.1%)	134 070 (41.1%)	140 361 (40.9%)	Reference
Female	10 069 (60.9%)	191 194 (58.5%)	201 263 (58.7%)	1.08 (1.04–1.12)
Other/not specifiedc,l	167 (1.0%)	1232 (0.4%)	1399 (0.4%)	–
Concessional beneficiary statusd
Beneficiarym	10 719 (64.9%)	116 456 (35.7%)	127 175 (37.1%)	1.78 (1.71–1.86)
Remotenesse
Metropolitan areas	12 161 (73.6%)	264 343 (81.0%)	276 504 (80.6%)	Reference
Regional centers	1065 (6.4%)	20 699 (6.3%)	21 764 (6.3%)	1.09 (1.01–1.16)
Rural and remote	3279 (19.8%)	40 994 (12.6%)	44 273 (12.9%)	1.18 (1.12–1.24)
Not specifiedl	22 (0.1%)	460 (0.1%)	482 (0.1%)	–
Geographically derived socioeconomic statusf
1–2; most disadvantaged	2088 (12.6%)	32 337 (9.9%)	34 425 (10.0%)	Reference
3–4	1951 (11.8%)	25 575 (7.8%)	27 526 (8.0%)	1.13 (1.05–1.21)
5–6	3823 (23.1%)	68 683 (21.0%)	72 506 (21.1%)	0.99 (0.93–1.05)
7–8	3905 (23.6%)	84 334 (25.8%)	88 239 (25.7%)	0.89 (0.84–0.95)
9–10; least disadvantaged	4738 (28.7%)	115 107 (35.3%)	119 845 (34.9%)	0.79 (0.74–0.84)
Not specifiedl	22 (0.1%)	460 (0.1%)	482 (0.1%)	–
Comorbidities
Cambridge Multimorbidity Score, mean (SD) [range]	0.84 (0.78) [−0.20–5.53]	0.44 (0.59) [−0.39–5.05]	0.46 (0.61) [−0.39–5.53]	1.29 (1.26–1.33)
Depression and/or anxietym	1648 (10.0%)	19 033 (5.8%)	20 681 (6.0%)	1.39 (1.31–1.48)
Sleep disorderm	458 (2.8%)	4542 (1.4%)	5000 (1.5%)	1.14 (1.02–1.28)
Substance use disorderm	193 (1.2%)	1561 (0.5%)	1754 (0.5%)	1.44 (1.22–1.71)
Headache/migrainem	406 (2.5%)	11 273 (3.5%)	11 679 (3.4%)	1.06 (0.95–1.18)
Chronic painm	637 (3.9%)	2873 (0.9%)	3510 (1.0%)	2.05 (1.85–2.27)
Musculoskeletal conditionm	4191 (25.4%)	52 815 (16.2%)	57 006 (16.6%)	1.14 (1.10–1.19)
Opioid characteristics
Type
Buprenorphinem	2152 (13.0%)	4424 (1.4%)	6576 (1.9%)	1.95 (1.73–2.20)
Codeinem	161 (1.0%)	29 778 (9.1%)	29 939 (8.7%)	0.26 (0.22–0.31)
Codeine combinationsg,m	5676 (34.3%)	203 253 (62.3%)	208 929 (60.9%)	0.99 (0.90–1.09)
Fentanylm	205 (1.2%)	564 (0.2%)	769 (0.2%)	0.85 (0.69–1.04)
Hydromorphonem	51 (0.3%)	501 (0.2%)	552 (0.2%)	0.39 (0.28–0.54)
Methadonem	44 (0.3%)	53 (<0.1%)	97 (<0.1%)	1.43 (0.91–2.23)
Morphinem	354 (2.1%)	3782 (1.2%)	4136 (1.2%)	0.50 (0.43–0.59)
Oxycodonem	2891 (17.5%)	39 088 (12.0%)	41 979 (12.2%)	0.89 (0.82–0.96)
Oxycodone–naloxonem	2734 (16.5%)	14 931 (4.6%)	17 665 (5.2%)	0.86 (0.78–0.96)
Pethidinen	NR	NR	11 (<0.1%)	–
Tapentadolm	1310 (7.9%)	11 596 (3.6%)	12 906 (3.8%)	0.44 (0.39–0.50)
Tramadolh,m	2393 (14.5%)	27 010 (8.3%)	29 403 (8.6%)	0.82 (0.74–0.90)
Dextropropoxyphenen	NR	NR	57 (<0.1%)	–
Short-acting only	9375 (56.7%)	295 270 (90.4%)	304 645 (88.8%)	Reference
Long-actingi	7152 (43.3%)	31 226 (9.6%)	38 378 (11.2%)	2.07 (1.90–2.25)
Provision of repeat prescriptionsm	1815 (11.0%)	13 619 (4.2%)	15 434 (4.5%)	2.94 (2.77–3.12)
≥2 routes of administrationj,m	333 (2.0%)	1043 (0.3%)	1376 (0.4%)	0.89 (0.76–1.05)
Total OME at initiation, mg, mean (SD) [range]	342.34 (646.92) [6–22 400]	122.30 (202.55) [1.95–19 035]	132.90 (247.85) [1.95–22 400]	–
<100 mg	5075 (30.7%)	228 709 (70.1%)	233 784 (68.2%)	Reference
100–249 mg	6253 (37.8%)	70 336 (21.5%)	76 589 (22.3%)	2.19 (2.03–2.36)
250–499 mg	2242 (13.6%)	15 166 (4.7%)	17 408 (5.1%)	3.10 (2.83–3.40)
500–749 mg	1326 (8.0%)	7473 (2.3%)	8799 (2.6%)	4.18 (3.72–4.70)
≥750 mg	1631 (9.9%)	4812 (1.5%)	6443 (1.9%)	7.75 (6.89–8.72)
Prescription of other medicines m
Non-opioid analgesics	6328 (38.3%)	93 755 (28.7%)	100 083 (29.2%)
Paracetamolm	1437 (8.7%)	11 146 (3.4%)	12 583 (3.7%)	1.11 (1.04–1.18)
NSAIDsm	4177 (25.3%)	78 952 (24.2%)	83 129 (24.2%)	1.02 (0.98–1.06)
Gabapentinoidsm	1959 (11.9%)	13 218 (4.1%)	15 177 (4.4%)	1.59 (1.50–1.68)
Anxiolytics, sedatives, hypnotics	4333 (26.2%)	48 096 (14.7%)	52 429 (15.3%)
Benzodiazepinesm	4051 (24.5%)	44 292 (13.6%)	48 343 (14.1%)	1.43 (1.38–1.50)
Z-drugsm	516 (3.1%)	5710 (1.8%)	6226 (1.8%)	1.61 (1.46–1.78)
Year of opioid initiation
2018m	5484 (33.2%)	95 200 (29.2%)	100 684 (29.4%)	Reference
2019m	3449 (20.9%)	77 187 (23.6%)	80 636 (23.5%)	0.81 (0.78–0.85)
2020m	3564 (21.6%)	58 114 (17.8%)	61 678 (18.0%)	1.06 (1.01–1.11)
2021m	2697 (16.3%)	59 840 (18.3%)	62 537 (18.2%)	0.90 (0.86–0.95)
2022k,m	1333 (8.1%)	36 155 (11.1%)	37 488 (10.9%)	0.85 (0.80–0.91)

Abbreviations: NR= not reported because of small sample sizes; NSAIDs= nonsteroidal anti-inflammatory drugs; OME = oral morphine equivalents.

Bold values indicate statistical significance at P < .05.

^aAdjusted for all covariates listed in this table (except pethidine and dextropropoxyphene, as not included in the logistic regression analysis because of small sample size in these groups).

^bAge at cohort entry in 2018.

^cOther/not specified includes intersex.

^dBeneficiary includes those who holds one of the following: (1) Commonwealth Seniors Health Card, (2) Department of Veterans’ Affairs Card, (3) Healthcare Card, or (4) Pensioner Concession Card. Where beneficiary status was not specified, people were assumed to be nonbeneficiary.

^eRemoteness was measured on a scale of Modified Monash (MM) categories 1 to 7. The categories were classified into metropolitan areas (MM1), regional centers (MM2), and rural and remote (MM3–7).

^fGeographically derived socioeconomic status was measured by Socio-Economic Indexes for Areas (SEIFA) deciles. SEIFA is ranked on a scale of 1 to 10, with 1 indicating the most disadvantaged.

^gCodeine combinations include aspirin/codeine, ibuprofen/codeine, paracetamol/codeine, and paracetamol/codeine/doxylamine.

^hTramadol includes tramadol combinations with paracetamol (reported as a combined variable because of small proportions of paracetamol/tramadol observed in the data).

ⁱIncludes prescription of long-acting opioids with or without concurrent prescription of short-acting opioids.

^jRoutes: “Other” includes sublingual, parenteral, and rectal routes.

^kIncludes 7 months of data only (from January to July 2022).

^lThose in the “not specific” categories were coded as missing and excluded from the regression model.

^mSingle level was reported for binary variables.

ⁿVariables were excluded from the regression model because of their small sample sizes.

Codeine combinations (eg, ibuprofen/codeine, paracetamol/codeine) were the most frequently prescribed opioids at initiation (60.9%), followed by oxycodone (12.2%), codeine (single ingredient; 8.7%), and tramadol (8.6%). Most of the sample initiated on short-acting opioids only (88.8%). At opioid initiation, small proportions of the sample were provided repeat opioid prescriptions (4.5%) and opioid prescriptions with ≥2 different routes of administration (0.4%). Most buprenorphine (97.8%) and fentanyl (92.5%) prescriptions were transdermal, whereas most methadone (95.9%), oxycodone (99.99%), and tramadol (99.9%) prescriptions were oral formulations. Most of the sample were prescribed a total OME of <100 mg at initiation (68.2%), with few prescribed 500–749 mg (2.6%) and ≥750 mg (1.9%). Almost a third (29.2%) of the sample were prescribed non-opioid analgesics, with approximately a quarter (24.2%) prescribed NSAIDs in the 12 months before opioid initiation.

There were 100 684 (29.4%), 80 636 (23.5%), 61 678 (18.0%), and 62 537 (18.2%) people initiating opioids in the years 2018, 2019, 2020, and 2021, respectively. A total of 37 488 people (10.9%) initiated opioids between January and July 2022 (inclusive).

Persistent opioid use

Of the people who initiated opioids, 16 527 (4.8%) developed persistent opioid use (Figure 3). The proportions decreased from 5.4% in 2018 to 4.3% in 2021 (Table 1).

Predictors of persistence

Demographic characteristics and comorbidities

People 55–64, 65–74, and ≥75 years of age, compared with those 15–44 years of age, had 41%, 24%, and 67% higher odds of persistence, respectively (55–64 vs 15–44 years: adjusted OR: 1.41, 95% CI: 1.33–1.49; 65–74 vs 15–44 years: 1.24, 1.17–1.31; ≥75 vs 15–44 years: 1.67, 1.58–1.78) (Table 1). Those receiving concessional benefits had 78% higher odds of persistence than did those not receiving benefits (1.78, 1.71–1.86), and those residing in rural and remote areas had 18% higher odds of persistence than did those in metropolitan areas (rural and remote vs metropolitan areas: 1.18, 1.12–1.24). People with a higher geographically derived socioeconomic status had 21% lower odds of persistence (SEIFA score of 9–10 vs 1–2: 0.79, 0.74–0.84). A higher Cambridge Multimorbidity Score was associated with 29% increased odds of persistence (1.29, 1.26–1.33). Those with depression or anxiety (1.39, 1.31–1.48), substance use disorder (1.44, 1.22–1.71), and chronic pain (2.05, 1.85–2.27) had 39%, 44%, and 105% higher odds of persistence, respectively, than did those without those comorbidities.

Prescription of opioids and other medicines

Those who were initiated on buprenorphine (1.95, 1.73–2.20) had almost double the odds of persistence compared with those who were not initiated on buprenorphine, whereas those who were initiated on codeine had 74% lower odds of persistence (0.26, 0.22–0.31) compared with those who were not initiated on codeine. Those who were initiated on long-acting opioids had more than double the odds of persistence (2.07, 1.90–2.25) compared with those who were not initiated on long-acting opioids, and those who were provided repeat opioid prescriptions at initiation had almost triple the odds of persistence (2.94, 2.77–3.12) compared with those who were not provided repeat prescriptions. Those prescribed a total OME of 100–249 mg (2.19, 2.03–2.36), 250–499 mg (3.10, 2.83–3.40), 500–749 mg (4.18, 3.72–4.70), and ≥750 mg (7.57, 6.89–8.72) at opioid initiation had more than 2, 3, 4, and 7 times higher odds of persistence, respectively, compared with those prescribed a total OME <100 mg at initiation. Those who were prescribed gabapentinoids (1.59, 1.50–1.68), benzodiazepines (1.43, 1.38–1.50), and z-drugs (1.61, 1.46–1.78) in the 12 months before opioid initiation were more likely to initiate persistent use.

Year of opioid initiation

People initiating opioids in 2019 (0.81, 0.78–0.85), 2021 (0.90, 0.86–0.95), and 2022 (0.85, 0.80–0.91) had 19%, 10%, and 15% lower odds of persistence, respectively, compared with those who initiated opioids in 2018. In contrast, those initiating opioids in 2020 (1.06, 1.01–1.11) had 6% higher odds of persistence compared with those initiating opioids in 2018.

Sensitivity analysis

In a sensitivity analysis that allowed a maximum of 30-day gap between prescriptions to define persistence (instead of a 60-day gap), 6511 (1.9%) of 343 091 people initiating opioids developed persistent opioid use. Most predictors remained similar to the primary analysis (Table S5). However, some predictors (initiation with hydromorphone, oxycodone, and oxycodone–naloxone; diagnosis of sleep disorders; prescription of paracetamol; and opioid initiation in 2021) were no longer significantly associated with persistence, likely because of reduced power with the smaller sample size.

Discussion

We found that 4.8% of people initiating opioids for non-cancer pain between 2018 and 2022 developed persistent opioid use, with an overall decline in the proportions from 5.4% in 2018 to 3.6% in 2022. Some of the strongest predictors of persistent opioid use were opioid prescribing practices, such as prescribing high opioid quantities, providing repeat prescriptions, and prescribing long-acting opioids at commencement of opioid therapy. Other predictors included chronic pain, concessional beneficiary status, older age, and substance use disorders. These factors might help to identify people at risk of persistent opioid use and enable opportunities to implement early intervention and additional pain management strategies for these people.

Prior studies using the Australian national dispensing data have reported 2.6% of people initiating persistent opioid use between 2013 and 2015.²⁷ Additionally, there was a 3.0% cumulative incidence of long-term opioid prescriptions among Australians with musculoskeletal conditions between 2012 and 2018.³⁵ It is important to note that a direct comparison of these figures with our finding of 4.8% of people developing persistent opioid use is challenging because of differences in the sample population, differences in the data source (dispensing vs prescribing data), and variations in the definition of persistence.²⁴ However, our study provides more recent data around the proportion of people developing persistent opioid use, and the proportions reported in each year enables examination of the changes in the proportions between 2018 and 2022.

Our findings confirm that commencement of opioid therapy with long-acting opioids and provision of high opioid doses at commencement of therapy remain two of the strongest predictors of persistence.^17,23,27 Although our findings largely support the previous findings that prescription of strong opioids is associated with persistence,^17,23,27 our study identified a few key exceptions to this prior finding and demonstrated that the odds of persistence differ among opioid types. First, initiation with buprenorphine (excluding formulations used for opioid replacement therapy) was the strongest predictor of persistence compared with other opioid types, with people who were initiated on buprenorphine being twice as likely to develop persistent opioid use as those who did not initiate on buprenorphine. It is possible that people who were initiated on buprenorphine might have had a greater analgesic requirement for their pain, as buprenorphine is indicated for “severe and persistent pain”³⁶ and is intended for long-term use, given that most buprenorphine prescriptions were prescribed as transdermal formulations. It is also possible to speculate whether the longer duration of buprenorphine might be associated with greater persistence. Second, initiation with codeine (single ingredient) had 75% lower odds of persistence, which might be explained by the weaker potency of codeine being preferentially prescribed for people with mild pain, who are less likely to require opioids in the long term, or the common use of codeine to treat cough. Of note, this finding was not consistent with the findings for codeine combinations. Further research directly comparing the use and clinical outcomes of codeine and codeine combinations might be warranted. Third, the odds of persistence when patients were initiated with tramadol were higher than the odds when patients were initiated with other weak opioids, such as tapentadol and codeine. Interestingly, initiation with tramadol, despite it being a weak opioid, had higher odds of persistence than did initiation with morphine, a strong opioid. These findings indicate the need for a more nuanced understanding of the use of and outcomes related to individual opioid types. In addition, clinicians should take extra caution when providing repeat prescriptions at initiation, as our findings suggest that this is a strong predictor of persistence. Providing repeat prescriptions might result in a missed opportunity for monitoring and reassessment of a patient’s pain and the need for ongoing opioid therapy, although where repeat prescriptions were provided, they might indicate that there was a likely need for opioids in the longer term.

Particular conditions and prescriptions also increased the likelihood of developing persistent opioid use. People with higher multimorbidity scores, depression or anxiety, and chronic pain diagnoses could expect 29%, 39%, and 105% higher likelihoods of developing persistence, respectively. Prescription of gabapentinoids and benzodiazepines also increased these odds, by 59% and 43%, respectively. Our findings in the context of prior studies confirm that people with these diagnoses and prescriptions are consistently identified as a higher-risk group for developing persistent opioid use.^{16,17,23,27,28} Given that many of these are shared risk factors associated with nonmedical use of opioids and overdose,²⁸ comprehensive assessment of pain and other comorbidities, regular medication reviews, and provision of non-opioid treatment as additional support might be needed for these at-risk populations.

We also found that where people live and their socioeconomic status impacted the odds of developing persistent opioid use. The odds were 9% higher for people residing in regional centers and 18% higher for people in rural and remote areas than for their metropolitan counterparts. People with a concession card had 78% higher odds of developing persistent opioid use compared with people without. This includes people on low incomes, pensioners, and people who might be unable to work because of conditions such as chronic pain. Conversely, people residing in the least disadvantaged areas had 21% lower odds of developing persistent opioid use compared with those in the most disadvantaged areas. Socioeconomic disadvantage is an established determinant of health inequalities and is correlated with increased dispensing of opioids and opioid overdose.^37,38 In addition, previous research identified geographic barriers to accessing adequate pain treatment among people with chronic non-cancer pain.³⁹ It is possible that timely access to multimodal pain management strategies (eg, access to pain specialists and allied health services for non-opioid interventions) might reduce the higher reliance on opioids observed in these groups.⁴⁰

Consistent with earlier findings,^16,17,27,35 our study demonstrated that people who are older are more likely to develop persistent opioid use. The largest difference was seen between people more than 75 years of age and people 15–44 years of age, with the former having 67% higher odds of developing persistent opioid use. This might be explained by the higher prevalence of pain and multimorbidity in this population.⁴¹ However, given that older people are more vulnerable to the adverse effects of opioids, such as falls and polypharmacy,^42,43 more caution is needed when opioids are prescribed to older people.⁴⁴

Among people initiating opioids between 2018 and 2022, those initiating opioids in the later years were less likely to develop persistent opioid use than were those initiating opioids in 2018, with the exception of those initiating in 2020. This might be explained by the cessation of elective surgeries during the COVID-19 pandemic, which might have caused prolonged pain and subsequent prolonged opioid use due to delayed surgical intervention. Reduced access to nonpharmacological pain management options due to the COVID-19 restrictions may have also contributed to increased reliance on opioid therapy for pain management.⁴⁵

In Australia, the National Drug Strategy (NDS) provides an overarching framework for tackling national priorities related to drug use.⁴⁶ Consistent with the priorities outlined in the NDS 2017–2026, a series of supply-reduction strategies have addressed the rising harms due to prescription opioid use. These strategies include codeine rescheduling, implementation of a prescription drug monitoring program, and other regulatory changes, such as introducing smaller opioid pack sizes. The present study adds to the very limited evidence known about the impact of these interventions. The overall reduction in the rate of persistent opioid use demonstrated through this study might be an indicator of positive effects of the opioid-related policy changes on clinical practice. Nevertheless, as our study was observational in nature, we cannot confirm whether there is a causal relationship between the reduction in persistence and the policy changes. Further study to understand the causality and how these policy levers impacted prescription opioid use and related harm at both the individual level and the societal level might be warranted.⁴⁷

Strengths and limitations

A key strength of our study is that we used primary care data consisting of one of the largest Australian primary care cohorts of people prescribed opioid analgesics, irrespective of their subsidy status.²⁰ The rich, longitudinal patient-level data enabled a comprehensive examination of the predictors of persistent opioid use.

Our study was subject to limitations. First, it is not possible to confirm whether prescribed opioids were dispensed or used. Although prescribing and consumption have been shown to be highly correlated,⁴⁸ the present study is best positioned to examine the patterns of opioid prescribing. Second, we did not have data on patients’ responses to pain or other factors that might affect the patterns of opioid prescribing, such as patient preferences, beliefs, and psychosocial stressors or prescriber characteristics (eg, level of training or training in pain management). Further research is needed to understand how patient and prescriber characteristics might influence persistence. Third, the indication for the initial opioid prescription was unknown. Previous studies have shown differences in the prevalence of short-term and long-term opioid use among people using opioids for different conditions (eg, chronic low back pain, musculoskeletal pain).⁴⁹ Although our analyses examined and adjusted for diagnoses recorded in the year before opioid initiation, prospective studies assessing the primary reason for initiating opioids might be important to further understand the patterns of opioid use. Fourth, diagnoses are likely to be underestimated because of incomplete recording within the electronic health record data, which is a known challenge of using electronic health record–based data.⁵⁰ Therefore, it is important to note that the actual prevalence of diagnoses and prescriptions might be higher than what was reported in the present study. In addition, over-the-counter medicines, such as paracetamol or NSAIDs, might not be fully captured in our prescribing data, which could result in underestimation of their prescription.

Conclusion

A notable proportion (4.8%) of people initiating opioids for non-cancer pain in primary care developed persistent opioid use (for at least 3 months). Persistent opioid use was associated with several factors, including characteristics of opioid prescription at treatment initiation (eg, prescription of specific opioid types such as buprenorphine and long-acting opioids, and provision of repeat prescriptions and high opioid quantity), comorbidities (eg, substance use disorder and chronic pain), demographic factors (eg, age, concession beneficiary status), and prescription of other medicines in the 12 months before opioid initiation (eg, gabapentinoids and benzodiazepines). Further research is needed to understand the clinical outcomes associated with persistence, which could be coupled with the findings of this study to guide safer and more effective prescribing of opioids. Lastly, our findings showed a reduction in the rate of persistence between 2018 and 2022, potentially resulting from a series of system-level interventions aimed at reducing long-term opioid use.

Supplementary material

Supplementary material is available at Pain Medicine online.

Funding

The project was funded by the Australian National Health and Medical Research Council (#2002193). S.N. is the recipient of a National Health and Medical Research Council (Australia) Investigator Grant Fellowship (#2025894). M.J. is a recipient of a PhD Scholarship from the Monash Addiction Research Centre.

Conflicts of interest: J.I. has grants from Amgen and AstraZeneca unrelated to this research topic. All other authors have no conflicts of interest to declare.

Supplement statement

This article appears as part of the supplement titled “Pain Management Collaboratory: Updates, Lessons Learned, and Future Directions.”

This article is a product of the Pain Management Collaboratory. For more information about the Collaboratory, visit https://painmanagementcollaboratory.org/.