Oxycodone initiation in Australia (2014–2018): Sociodemographic factors and preceding health service use

Oxycodone is the most commonly prescribed strong opioid in Australia. This study describes health service antecedents and sociodemographic factors associated with oxycodone initiation.

hazardous, longer-term use. 3 Australian dispensing data suggest that increases in opioid use between 2006 and 2015 were driven by a growing number of people treated with strong opioids at lower doses. 4The controlled release, fixed-dose combination oxycodone/ naloxone is widely used, including low strength (oxycodone ≤5 mg) formulations. 5No studies have reported on trends in the prevalence of persistent use.
Information is limited about why, and in which settings, opioids are prescribed in Australia.Australian data from 2013 to 2017 showed that cancer pain accounted for only a small proportion of opioid use in the general population. 2 Primary care survey data from 2011 to 2012 showed that almost 60% of opioid prescriptions were for musculoskeletal problems (most commonly back problems) and 44% for some type of chronic, noncancer pain. 6One study found that half of all opioids were initiated by general practitioners, 2 which implies that initiation in other settings, such as emergency departments (EDs) and for hospital inpatients (i.e. in acute care) accounted for the remainder.
With increases in harmful use of prescription opioids including deaths by oxycodone overdose, 7 research has examined associations between opioid use and sociodemographic factors that might inform interventions to improve prescribing practice and quality use of opioids.][10][11] Those who initiated oxycodone rather than another strong opioid were more likely to be younger with no previous cancer treatment history. 12is study reports on oxycodone use in the state of New South Wales (NSW), Australia, in the 2014-2018 period.We conducted a population-based, descriptive study investigating sociodemographic factors associated with oxycodone initiation and the health services immediately preceding it, particularly acute care presentations, to inform opioid prescribing practice in Australia and elsewhere.We aimed to describe the: i. proportion of new opioid users per year initiating on oxycodone relative to other opioids; ii.rate per 1000 population of people who were new or prevalent oxycodone users per year; iii. sociodemographic characteristics and preceding health service use in those initiating oxycodone; and, iv.long-term use over 1 year after initiating oxycodone.POPPY II consists of deidentified records from 10 NSW and national datasets, linked on personal identifiers using a probabilistic algorithm.The study protocol has been described elsewhere including details of the source datasets and linkage. 13The current analysis used linked records from 8 of the 10 data sources.Study data provide information on all: subsidized medicine dispensings; inpatient stays in NSW hospitals; subsidized medical and hospital services; notifications of primary malignant neoplasms; and deaths; and >90% of visits to NSW EDs.The data sources are described in detail in Table S1.

| Setting
Government funding and delivery of health care in Australia is divided between state and federal levels.Acute care is the responsibility of Australian states and territories and is available to all Australian citizens, permanent residents and eligible foreign visitors through public hospitals and EDs.NSW is the most populous state, with a population of 7.9 million in 2018. 14The federal government provides universal health insurance that includes primary care as well as other private health care services such as elective surgery and pathology services through the Medicare Benefits Scheme (MBS) and medicines through the Pharmaceutical Benefits Scheme (PBS).Only a small proportion of medical care and medicines falls outside these government schemes; recent data suggest that privately funded prescriptions and use by hospital inpatients accounted for 12% of the dispensed amount of opioids in oral morphine equivalents. 15at is already known about this subject • Oxycodone is the most commonly prescribed strong opioid in Australia.
• Oxycodone and other opioid use is higher in rural locations and areas of socioeconomic disadvantage.
• About half of all courses of opioids are initiated by general practitioners.

What this study adds
• About half of all people initiating oxycodone had an episode of acute care or a therapeutic procedure in the previous 5 days.
• 4.6% of people initiating oxycodone were on treatment 1 year later.
• Higher rates of new oxycodone use in rural and regional areas were not explained by age, sex or socioeconomic disadvantage.

| Study population
Our cohort was defined as all NSW residents aged 18 years and over dispensed a PBS prescription for any opioid in the period from 1 January 2014 to 31 December 2018 (Anatomical Therapeutic Chemical classification and PBS item codes are listed in Table S2).Opioids prescribed for treating opioid dependence (opioid agonist therapy) are not included in these PBS data.We determined NSW residence at cohort entry by patient postcode.Within the cohort, we defined a subcohort of those dispensed oxycodone, including all PBS-listed oxycodone medicines (single ingredient and combination, both immediate and controlled release) during the study period.We identified new episodes of oxycodone use as those without any opioid dispensing during a lookback period of 365 days before the initial dispensing; individuals meeting the initiation criteria more than once contributed multiple episodes.For long-term use analyses, a treatment episode ended at the earliest of death, end of the study period (31 December 2018) or 365 days after initiation.

| Study measures
We described the subcohort of new users by age group (18-24,   25-34, 35-44, 45-54, 55-64, 65-74, 75-84 and 85+ years) and sex at first oxycodone dispensing.Based on residential postcode, we described each new user's location of residence according to the Australian Statistical Geography Standard remoteness area category (Major Cities, Inner Regional, Outer Regional or Remote/Very Remote) and Index of Relative Socioeconomic Disadvantage (IRSD; as quintiles, with 1 = most disadvantaged and 5 = least disadvantaged). 16,17 calculated the proportion with recent (the same day or up to 4 days before) injury, therapeutic procedure, admitted patient hospitalization or public ED attendance, as well as a history of cancer (1 year or any) or opioid dependence (any receipt of opioid agonist therapy for dependence or opioid use disorder diagnosis in health service records); the categories are not mutually exclusive and new use could be counted under >1 of them.We chose the length of the exposure window as 5 days to be consistent with other published work; a longer window would increase the chance of coincidental associations, while a shorter window would not allow sufficient time for patients to fill prescriptions. 18We also calculated the proportions of different prescriber types responsible for prescribing the first dispensing (medical practitioner, dental practitioner, medical/dental practitioner or other) using PBS data.
We defined the date of health service use as the date of discharge from hospital, the service date for MBS reimbursement records, date of diagnosis for cancer records and date of dispensing for PBS records.Most therapeutic procedures in private hospitals were recorded in both an admitted patient record and an MBS record.We addressed this overlap when calculating the total proportion of preceding health service use.For the purposes of describing characteristics of oxycodone new users and health service use preceding first oxycodone dispensing, we reported on cohort members initiating oxycodone between 1 January 2014 and 1 July 2017 only, as ED and controlled drugs records did not extend past this date.
We detail the study measures in the Supporting Information including the definitions of diagnosis groupings, hospital procedures, ED diagnoses, therapeutic procedures, cancer history, opioid dependence and injury.

| Statistical methods
All analyses were done using SAS 9.4 (SAS Institute Inc, Cary, NC, USA).

| Descriptive analyses
For new users of any opioid, we categorized their first dispensing as oxycodone or other opioid and calculated a proportion, by calendar year.If the first dispensing included both oxycodone and other opioids, it was classified as oxycodone use.
Within each calendar year, we categorized each person dispensed a PBS-listed oxycodone item as a prevalent user.For prevalent users who had a new episode of use in that year, we also counted them as a new user.To calculate dispensing rates for new and prevalent (continuing) oxycodone users, we took the counts for each calendar year and converted them to population rates using mid-year estimated resident populations for NSW for persons aged ≥18 years. 14e study cohort was described by number (%) for categorical variables and median (interquartile range) for age.

| Long-term use
Within the subcohort of people initiating oxycodone, we estimated the proportion of people covered curve for 1 year following first oxycodone dispensing.This method measures the proportion of a population of new users who are considered exposed to a medicine on each day of follow up, accounting for stopping and restarting. 19We assigned each day of follow up as currently on or off oxycodone treatment using the individualized dispensing pattern method, 20 censoring observations at death or end of follow-up.We used exposure at day 89 and day 364 of follow-up (day of initiation defined as day 0) as a measure of long-term use. 21,22We also calculated the proportion of people who had no further oxycodone dispensings in the period from day 1 to day 364, that is, who had a single dispensing at initiation only, as an alternative, crude indicator of treatment duration (for people who initiated oxycodone 1 January 2014 to 31 December 2017 only).

| Area-level population-based multivariable model of oxycodone new use
We explored factors related to population oxycodone initiation rates in a multivariable model, accounting for unexplained spatial variation using postcode-level random intercepts and overdispersion of counts with a negative binomial model.Small-area population data by age and sex were only available for the census year of 2016, so our analysis was limited to dispensing for this year.We aggregated counts of new use during 2016 by postcode, sex and age group.These were modelled with sociodemographic factors (age group nested within sex, remoteness area, IRSD decile) as covariates using log mid-year populations for each postcode-age-sex stratum 23 as offsets, yielding multivariate (adjusted) incidence rate ratios (IRRs) for each covariate.
We excluded dispensings where the postcode was for a post office rather than a street address.

| Sensitivity analyses
We did sensitivity analyses to look at the effect of a longer 30-day window for identifying injuries, therapeutic procedures and admitted patient hospitalizations or ED attendances preceding first oxycodone dispensing.

| Characteristics of oxycodone new users
New users of oxycodone in the period from 1 January 2014 to 1 July 2017 (n = 565 357) had a median age of 54.7 years, and 52.5% were female (Table 2).In 18.9% of new users, the initial oxycodone dispensing was for fixed-dose combination oxycodone with naloxone.Starting oxycodone was slightly less common among those in the most disadvantaged quintile of the population (16.0%) and slightly more common among those who were in the median quintile (22.2%) and in the least disadvantaged quintile (22.3%).Most people lived in a major city (66.4%).
Medical practitioners were the prescribers in most cases (98.0%).

| Health service use preceding first oxycodone dispensing
In the 5 days preceding their first oxycodone dispensing, 39.3% of people initiating oxycodone had been discharged from hospital ( Only 1.2% of people prescribed oxycodone had any history of opioid dependence. The most frequent primary hospital diagnosis preceding first oxycodone dispensing was knee arthrosis, occurring in 5.3% of preceding admissions (Table S3).Other frequent diagnoses were internal derangement of the knee, delivery by caesarean section, inguinal hernia and shoulder lesions.The most frequent primary hospital procedures were generally consistent with the diagnoses, and included physiotherapy, total arthroplasty of the knee, cholecystectomy, shoulder reconstruction and caesarean section (Table S4).ED diagnoses were most commonly for pain and injury (Table S5).The most frequent MBS therapeutic procedures were sleeve gastrectomy (4.0%), knee surgery, shoulder surgery and removal of tonsils (Table S6).
The sensitivity analysis with a 30-day window preceding first oxycodone dispensing generated larger percentages than the main analysis: 47.0% had been discharged from hospital, 29.4% had visited an ED, and 30.1% had an MBS-subsidized therapeutic procedure; an injury occurred among 8.5%.

| Long-term oxycodone use
The proportion of people using oxycodone decreased sharply at 40 days' postinitiation and then slowly thereafter (Figure 2).It was 7.7% after 90 days; 4.6% of people initiating oxycodone were using it 1 year later.Overall, 65.0% of people initiating oxycodone received the initial dispensing only, and no further dispensings of oxycodone within 1 year.

| Area-level population-based multivariable model of new oxycodone use
In the population-based multivariable model of new oxycodone use in 2016, initiation rates increased with age (Table 3 and Figure 3).IRRs for women were significantly higher than for men in the age bands of 18-24, 25-34, 35-44 and 85+ years.Relative to residents of major cities, rates of new use were significantly higher for residents of inner regional (IRR 1.43, 95% confidence interval [CI] 1.36-1.51),outer regional (IRR 1.48, 95% CI 1.39-1.57)and remote (IRR 1.35, 95% CI 1.16-1.58)areas, while the IRR for residents of very remote areas could only be estimated imprecisely (0.95, 95% CI 0.69-1.32).Rates of new use fluctuated insignificantly between deciles of IRSD with no evidence of a trend.After accounting for covariates, there was appreciable variation in incidence rates between postcodes (τ = 0.047).A total of 1269 first dispensings with a non-street-address postcode (comprising <1% of observations) were excluded from the analysis.

| DISCUSSION
Oxycodone accounted for 30% of all opioid initiations between 2014 and 2018.In 2018, almost 3% of the NSW population were new users of oxycodone, and almost 6% were prevalent users, receiving 1 or more dispensings.
About half of all new use was associated with acute care and procedures, with a substantial contribution from elective surgery.
F I G U R E 2 Proportion of people exposed to oxycodone (i.e.proportion of people covered, PPC) on each day over 365 days following initiation (n = 830 941).
The most frequent primary diagnoses for these admissions were joint replacement surgery, delivery by caesarean section and other surgeries; these were reflected in the data on inpatient procedures and medical service claims.In addition, a history of recent ED attendance was common and most frequently involved an acute pain finding such as back pain or abdominal pain.New use among people who had a recent history of cancer constituted only 7% of total initiations.Prescribing by dentists was negligible at <2%, and only about 1% of those initiating oxycodone had a history of opioid dependence.
Those with musculoskeletal pain received more initial prescriptions with a duration exceeding 7 days (34.2%).While there is a stark contrast in the rates of dental use between NSW and Ontario results, the proportions of other indications are broadly similar.
Consistent with our finding that acute pain conditions frequently preceded oxycodone use, 2/3 of new users received a single dispensing only.This pattern can also be observed the proportion of people covered curve where the large drop at 40 days corresponds to the coverage days assigned to a single dispensing of the most commonly prescribed pack of oxycodone tablets.The proportion covered at 1 year (4.6%) provides an estimate of long-term or persistent use, noting that this number does not consider the use of other opioids.
The proportion of persistent use at 1 year was higher than that found in another Australian study covering 2013-2015 (2.6%), but the difference may be accounted for by varying methods to define persistence and the exclusion of people with cancer in the earlier study. 24We used a high-sensitivity, low-specificity measure of longterm use, but nonetheless, we found a proportion of people who went on to long-term use that was similar to North American studies of opioids initiated in ED for acute pain. 25,26e area-level model of oxycodone initiation during 2016 revealed a strongly increasing incidence with age.Incidence rates in women aged 18-44 years (i.e. of childbearing age) were significantly higher than for men of equal age, highlighting the contribution of opioid initiation after delivery by caesarean section.In the same model, initiation in areas outside the major cities was significantly elevated, but the area-based index of relative social disadvantage had no consistent effect.Finally, initiation rates varied substantially from postcode to postcode after accounting for age, sex, remoteness and disadvantage.
0][11] Studies from the USA also report higher rates of opioid prescribing in rural than urban areas, and mixed results for the importance of income. 28Higher rates of injury in rural areas compared with urban areas-injuries that are also more severe and lead to greater morbidity-may be an important reason for urban-rural differences. 29,30Further explanation will require data on service accessibility as well as detailed, population-level data on acute and chronic pain conditions.Unexplained small-area variation in rates may also speak to local differences in provider behaviour and nonmedical use.that large proportions of people presenting to ED with musculoskeletal pain received opioids [31][32][33] and that postsurgery prescription of opioids was common. 34Understanding opioid prescribing in these settings is important, particularly given a recent study showed oxycodone for acute low back or neck pain did not improve pain at 6 weeks. 35In US and Canadian studies, opioid treatment of low back pain in the ED increased the risk of prolonged opioid use, 25,26 motivating in-hospital interventions that might reduce the use of oxycodone postsurgery and in the ED. 36,37 found that oxycodone is frequently initiated for acute pain, particularly after ED presentation.While 2/3 of the opioid initiators ceased after a single dispensing, the proportion of people taking oxycodone 12 months later (4.6%) underlines the need for early, careful assessment of ongoing pain to maximize appropriate use of nonopioid and nonpharmacological pain control.New Australian hospital care standards require that hospital prescribers prepare an opioid weaning and cessation plan, to be started as soon as clinically appropriate, and provide it to the patient's general practitioner on discharge. 38General practice guidelines also state that opioids should be weaned and ceased as an acute injury heals and that even in complex postoperative cases, this is expected to be within 90 days.General practitioners should be competent with multimodal approaches to analgesia and should be ready to escalate care to specialist services including pain management, drug and alcohol, clinical pharmacy and allied health as required. 39However, the recommended care escalation options for general practitioners and their patients may be limited and difficult to access in many settings.
One in 5 people were treated initially with modified release fixeddose combination of oxycodone with naloxone.In part, this may represent appropriate management of chronic pain after initiation on nonoxycodone immediate release opioids.Prescribing modifiedrelease opioids for acute pain conditions such as postsurgical pain is prevalent while having few benefits in postoperative pain management and being a risk factor for persistent use. 40Potential motivations include the perception that oxycodone with naloxone causes less constipation than oxycodone alone, along with availability of a very low strength product containing oxycodone 2.5 mg/tablet.
Australian care standards for opioid stewardship in acute pain preference immediate-release opioids over modified release. 38

| Strengths and limitations
The strengths of our study are that it used a general population-based cohort of new users of oxycodone, with rich, individual patient data from multiple sources, covering cancer diagnoses, acute care and procedures.We used longitudinal data to investigate long-term use and a multivariable, small-area model to disentangle sociodemographic factors relating to initiation.We anticipate that these results will generalize to the overall Australian population, as well as to other countries with similar health systems.
Among its limitations, our study underestimates true total prescription opioid use as the data do not cover medicines privately funded or administered among public hospital inpatients, nor do they F I G U R E 3 Age group and sex effects from a postcode-level population-based random effects model of oxycodone initiation by age group and sex, age-sex interaction, index of relative social disadvantage and remoteness.
include medicines dispensed on discharge from public hospitals.
Beyond simple underestimation of oxycodone use, we would have misclassified individuals with a history of privately funded opioid use as new users, and there may have been concurrent use and/or switching between subsidized oxycodone and privately funded opioids.People who were dispensed oxycodone on discharge from a public hospital but had no further PBS dispensings were omitted from our analysis, potentially biasing our observed characteristics of new users.
Furthermore, in examining antecedents of oxycodone new use, we did not have the data to confirm that oxycodone initiations were a direct consequence of the preceding events; by contrast, the results of the sensitivity analysis suggest that the true proportions of antecedents may be higher.Ascertainment of injuries was based on admitted patient diagnoses only, and we could not identify injuries as such when only treated elsewhere such as ED or primary care.Finally, we may have missed some cases of cancer because the register data did not cover the entire follow-up period, although ascertainment by medicine use and hospital diagnosis was still possible.

| CONCLUSION
A substantial proportion of the NSW population was dispensed oxycodone in each of the years we examined, and about half of the new use was associated with a recent episode of acute care or a therapeutic procedure.There was an important role for acute pain conditions and elective surgery and the expected association with age.These findings imply that interventions in acute care settings could have a substantial impact on the proportion of the population exposed to opioids.Examining spatial variation, the initiation rate outside cities was higher than that among city dwellers even after accounting for age, sex and areabased social disadvantage, noting there was no independent effect of disadvantage.Long-needed efforts to improve the quality use of oxycodone and other opioids in rural and regional areas will require a closer examination of the reasons for this disparity.

2 | METHODS 2 . 1 |
Study design and data sources This was a cohort study of new users of oxycodone based on linked clinical and administrative data from NSW for the 2014-2018 period with a 1-year lookback.We used the POPPY II study data, a cohort based on all adult (≥18 years) NSW residents initiating a new opioid dispensing episode from 1 July 2003 to 31 December 2018.
The study was approved by the Australian Institute of Health and Welfare (AIHW) Ethics Committee (EO2016/4/314), NSW Population and Health Services Research Ethics Committee (2017/HRE0208), the Australian Capital Territory (ACT) Health Human Research Ethics Committee (ETHLR.18.094) and the ACT Calvary Public Hospital Bruce Ethics Committee (5-2019).Data from ACT were not used in this study.
Our population-based results complement previous Australian studies in specific care settings.The large proportion of acutecare-associated oxycodone initiation aligns with published findings T A B L E 3 Factors associated with oxycodone initiation rates in an area-level multivariable model.

Table 1 )
Annual proportion of new opioid users initiating oxycodone vs. other opioids. .New use, prevalent use and the proportion of oxycodone use relative to other opioids increased over the years of 2014-2017 and decreased slightly in 2018.
2.8 | Nomenclature of targets and ligands Key protein targets and ligands in this article are hyperlinked to corresponding entries in http://www.guidetopharmacology.org and are permanently archived in the Concise Guide to PHARMACOLOGY 2019/20.

Table 2
Characteristics of new opioid users initiating oxycodone, 1 January 2014 to 1 July 2017.
lence of 15.0%.After excluding those with a 12-month history of cancer, in total, 54.7% of new users had evidence of a hospital discharge, ED visit and/or therapeutic procedure in the preceding 5 days.T A B L E 2 Note: Counts of <10 are suppressed to preserve confidentiality.a Individuals were counted again if they ceased and reinitiated oxycodone use.b Within 5 days before oxycodone initiation.