Cohort profile: Using primary care data to understand Opioid Prescribing, Policy Impacts and Clinical Outcomes (OPPICO) in Victoria, Australia.

Abstract

Purpose

The OPPICO cohort is a population-based cohort based on non-identifiable electronic health records routinely collected from 464 general practices in Victoria, Australia, created with the aim of understanding opioid prescribing, policy impacts and clinical outcomes. The aim of this paper is to provide a profile of the study cohort by summarising available demographic, clinical and prescribing characteristics.

Participants

The cohort described in this paper comprises people who were aged at least 14 years at cohort entry, and who were prescribed an opioid analgesic at least once at participating practices for a total of 1 137 728 person-years from 1 January 2015 to 31 December 2020. The cohort was formed using the data collected from electronic health records through the Population Level Analysis and Reporting (POLAR) system. The POLAR data primarily consist of patient demographics, clinical measurements, Australian Medicare Benefits Scheme item numbers, diagnoses, pathology testing and prescribed medications.

Finding to date

In total, the cohort consists of 676 970 participants with 4 389 185 opioid prescription records from 1 January 2015 to 31 December 2020. Approximately half (48.7%) received a single opioid prescription, and 0.9% received more than 100 opioid prescriptions. The mean number of opioid prescriptions per patient was 6.5 (SD=20.9); prescriptions for strong opioids accounted for 55.6% of all opioid prescriptions.

Future plans

The OPPICO cohort data will be used for various types of pharmacoepidemiological research, including examining the impact of policy changes on coprescription of opioids with benzodiazepines and gabapentin, and monitoring trends and patterns of other medication utilisation. Through data-linkage between our OPPICO cohort and hospital outcome data, we will examine whether policy changes for opioid prescribing lead to changes in prescription opioid-related harms, and other drug and mental health-related outcomes.

Trial registration number

EU PAS Register (EUPAS43218, prospectively registered).

Strengths and limitations of this study.

• This cohort is one of the largest primary care cohorts of opioid prescribing in Australia, representing a diverse range of subsidised and non-subsidised opioid prescribing from acute to long-term opioid treatment.

• Through the rich patient-level data contained in the primary care data set, there is a detailed range of covariates for inclusion in analyses to enable exploration of opioid prescribing trajectories during this period of substantial opioid prescribing policy change.

• Codeine supply prior to 2018 for this cohort are likely to be underestimated since it could be purchased over-the-counter.

• Patients are uniquely identified at the practice level, so those who attend multiple practices may not be able to be identified as unique patients.

Introduction

Opioids are widely used for the treatment of pain, with a well-established role in the management of acute pain and palliative care. The role of opioids in the management of chronic pain is less clear, with many experiencing minimal clinical benefits with growing evidence demonstrating potential adverse events, particularly with long-term use.^1–3 Opioid prescribing in Australia has increased steadily over the past three decades,^{4 5} with approximately half of all opioids initiated by general practitioners (GPs) in primary care settings.^{6 7}

Over the past 5 years, a range of Australian policies have sought to address rising opioid supply and related harms. This includes rescheduling of codeine to a prescription only medication (1 February 2018),⁸ the implementation of mandatory real-time prescription monitoring systems in Victoria from 1 April 2020⁹ (with mandatory use planned for other states) and reduced pack sizes for immediate-release (IR) formulations of opioids prescribed for acute severe pain (1 June 2020).¹⁰ However, there is little evidence to know if and how these policy levers work. In fact, there are clear examples where opioid policies have had unintended consequences, resulting in rapid opioid tapering or cessation, with devastating outcomes.¹¹ For example, a US study of opioid cessation in primary care, following the introduction of new guidelines and prescribing policies, showed that cessation of opioid prescribing was associated with a nearly threefold increase in overdose deaths.¹² However, few other studies have assessed the impact of these new opioid policies on prescribing in primary care, or the potential unintended consequences of these policies in this setting.

To address these gaps in the literature, the OPPICO cohort was established to explore opioid prescribing patterns and trajectories within primary care, and the impact of key opioid-policy changes described above on opioid prescribing. The aim of the current paper is to describe the profile of the study cohort by summarising available demographic, clinical and prescribing characteristics.

Cohort description

Source data

The cohort was formed using the data collected from electronic health records through the Population Level Analysis and Reporting (POLAR) system that was established on 2019. POLAR is a platform designed as an asset for the comparison and analysis of aggregated and unified data for healthcare researchers.¹³ The POLAR data primarily consist of patient demographics, clinical measurements, Australian Medicare Benefits Scheme item numbers, diagnoses, pathology testing and prescribed medications. The OPPICO cohort utilises non-identifiable electronic health records routinely collected from 464 GP practices in Victoria, the second most populous state in Australia, comprising around 6.7 million people, equating to about 26% of the Australian population.¹⁴

Participating GP practices are from three health regions, called ‘Primary Health Networks (PHNs)’ in Victoria, covering the Eastern Melbourne, South Eastern Melbourne and Gippsland PHNs (see figure 1). In total, there are six PHNs in Victoria, and 31 in Australia. These three PHNs comprise both metropolitan and rural regions.

Figure 1.

Spatial distributions of cohort participants at postcode level, 1 January 2015 to 31 December 2020: the map represents the boundary of the State of Victoria, the internal borders in light grey define the geographical unit at postcode level and the green bold line draws the boundary of three geographic health areas called ‘Primary Health Networks’ in Victoria. PHN, Primary Health Network; POLAR, Population Level Analysis and Reporting.

Participants

The OPPICO cohort uses data from the electronic health records of those practices contributing data to the POLAR system. The cohort consists of people who consult GPs in participating practices including face-to-face consultations and telehealth, with all routinely collected data available for the cohort for each year of the study. Data are collected for every patient encounter in the participating GP practices. Although the study sample is not representative of the whole Australian population, or all people who are prescribed opioids, previous studies using this data have demonstrated considerable similarity in the demographic patterns of patient populations within these practices with national GP survey data.¹⁵

While collection of data through the POLAR system is ongoing, individuals in the current OPPICO cohort represent patients aged at least 14 years at cohort entry, and who were prescribed an opioid analgesic at least once at participating GP practices from 1 January 2015 to 31 December 2020. In Australia, medications and poisons are classified into schedules according to the level of regulatory control.¹⁶ For the purposes of this cohort description, opioid analgesics were classified as: (1) weak opioids (defined as ‘schedule 4’ prescription only medications such as codeine, dextropropoxyphene and tramadol, which have fewer prescribing restrictions), (2) strong opioids comprising schedule 8 (controlled drugs, eg, buprenorphine, methadone, diamorphine, fentanyl, hydromorphone, morphine, oxycodone, tapentadol and pethidine) and (3) weak opioids in combination with simple analgesics such as paracetamol or ibuprofen, which were included in schedule 3 (pharmacist only medications) and schedule 4. Lower doses of codeine (up to 15 mg) were available as a schedule 3 opioid in combination with simple analgesics up until February 2018, when these formulations became prescription medications.

Variables

The cohort data include demographic variables (age in five-year groups, sex, pension or concession card status, socioeconomic status and geographic remoteness based on postcode), information on prescribed medications and clinical care (including detailed information on opioid prescriptions), referral to other healthcare providers and initiation of GP care plans, a government-funded mechanism to allow structured care for a chronic condition involving multidisciplinary health professionals. Other prescribed medications were also extracted for the cohort including non-opioid and adjuvant analgesics such as non-steroidal anti-inflammatory drugs (NSAIDs), benzodiazepines, ‘Z’—drugs (zaleplon, zopiclone, eszopiclone and zolpidem), cannabinoids, antidepressants and antipsychotics. Diagnoses are derived from the patients’ historical and current records and are coded using systemised nomenclature of medicine (SNOMED) codes,^{17 18} with the OPPICO study focussing on pain conditions, mental health and substance use disorders. Further details of the SNOMED codes used for diagnosis identification are comprehensively documented elsewhere.¹⁹ Unique identifiers are available for prescribers within practices, and a variable captures prescriber type, enabling analyses on prescribing patterns that consider these variables.

Patient and public involvement

Patients or the public were not involved in the design, conduct, reporting or dissemination plans of our research.

Findings to date

Demographic and clinical information for the cohort are presented in table 1. Figure 1 shows the geographic distribution of cohort participants at the postcode level. In total, the cohort consists of 676 970 participants with 4 389 185 opioid prescription records from 1 January 2015 to 31 December 2020. There are more females than males in this cohort (57.0% vs 42.7%), and a small proportion (0.3%) did not state their sex or reported a category that was not male or female. Just over one-third of the cohort (37.2%) were recorded as concession card holders (ie, are eligible for subsidised health services and medications). Participants were relatively evenly distributed across the range of age categories, although planned analysis will control for variation in age grouping over time. Most patients lived in major cities (83.7%). Over half the cohort (58.4%) lived in the top five ‘least socioeconomically disadvantaged’ deciles, with more than a third (34.5%) living in the top two ‘least disadvantaged’ deciles, according to the Index of Relative Socioeconomic Disadvantage (IRSD), which was based on the patient’s residential postcode.²⁰

Table 1.

Sociodemographic and opioid prescription characteristics of the cohort n=676 970

Demographics	Number of cohort participants (N, %)
Sex
Male	289 187 (42.7)
Female	388 181 (57.0)
Not reported/other	1602 (0.3)
Pension or concession status
Seniors health card	3380 (0.5)
Department of Veterans Affairs	3737 (0.6)
Healthcare or pension concession	252 171 (37.2)
No pension/concession status specified	417 682 (61.7)
Age distribution of the cohort at cohort entry
14–24 years	74 663 (11.0)
25–34 years	114 236 (16.9)
35–44 years	110 416 (16.3)
45–54 years	109 163 (16.1)
55–64 years	97 844 (14.5)
65–74 years	83 210 (12.3)
75+	87 195 (12.9)
Not specified	243 (0.04)
Geographic remoteness
Major cities	566 592 (83.7)
Inner regional	99 460 (14.7)
Outer regional	10 715 (1.6)
Remote	187 (0.03)
Socioeconomic disadvantage*
1, most disadvantaged	44 221 (6.5)
2	26 837 (4.0)
3	14 040 (2.1)
4	36 980 (5.5)
5	63 775 (9.4)
6	70 537 (10.4)
7	57 654 (8.5)
8	129 508 (19.1)
9	138 271 (20.4)
10, least disadvantaged	95 147 (14.1)

Types of opioids prescribed	Number of prescriptions (N, %)	Cohort participants (N, %)
Strong opioids	2 439 098 (55.6)	224 100 (26.3)
Weak opioids	620 280 (14.1)	175 273 (20.5)
Weak opioids formulated in combination with simple analgesics	1 329 807 (30.3)	453 804 (53.2)

*Socioeconomic disadvantage was measured using the Index of Relative Socioeconomic Disadvantage (IRSD) deciles of patient’s residence postcode.

Overall, there was a large variation in the number of opioid prescriptions per patient. Approximately half (48.7%) of the cohort received a single opioid prescription during the study period, with the remainder receiving multiple opioid prescriptions, and 0.9% received more than 100 opioid prescriptions. The mean number of opioid prescriptions per patient was 6.5 (SD=20.9), and the median number was 2 (IQR: 1–4). Prescriptions for strong opioids accounted for 55.6% of all opioid prescriptions, with the remainder being weak opioids either alone (14.1%), or in combination with simple analgesics (30.3%). Of the 676 970 patients, 73.7% were prescribed weak opioids at least once (either as single ingredients or in combination with simple analgesics), and 26.3% were prescribed strong opioids at least once, although strong opioids made up almost half of all prescriptions, reflecting the more common use of strong opioids for longer-term prescribing (table 1).

Table 2 shows all the relevant, current-documented diagnoses within the cohort, providing an overview of the cohort characteristics. In future analyses, dates of diagnoses relative to opioid prescribing will be considered, where relevant, for specific research questions. Almost one in 15 (6.4%) cohort participants were diagnosed with cancer, about 60% were diagnosed with a pain condition, and 5.8% had more than one diagnosed pain condition. The most common pain diagnosis was ‘Musculoskeletal pain (excluding neck and back pain)’ (24.0%), followed by arthritis/rheumatism (18.5%) and back and neck pain (18.0%). Patients were identified as having a mental health condition if they had a recorded mental health diagnosis or a mental health medication prescribed. The validated RxRisk-V tool²¹ and WHO Anatomical Therapeutic Chemical (ATC) classification system for mental health-related medications were used for identifying medications commonly used in Australian mental health reporting.²¹ Further details of the SNOMED codes used for mental health medications are comprehensively documented elsewhere.¹⁹ In the OPPICO cohort, one in two people were identified as having mental health conditions, with the most common diagnoses being depression (16.8%) and anxiety (10.0%). A further 24.9% of the cohort were prescribed a medication for a mental health condition, without a mental health diagnosis being recorded. Suicidal and distress behaviours were recorded for 0.1% of all patients in this cohort, and 2.0% of the cohort were diagnosed with a substance use disorder.

Table 2.

Clinical characteristics of the OPPICO cohort*, n=676 970

Clinical characteristics	Number of cohort participants (N, %)
Cancer diagnoses	45 039 (6.7)
Pain diagnoses	409 722 (60.5)
Musculoskeletal pain (excluding back and neck pain)	162 241 (24.0)
Arthritis/rheumatism	124 918 (18.5)
Back and neck pain	121 642 (18.0)
Pain (not otherwise specified)	87 883 (13.0)
Pain related to procedure or injury	64 590 (9.5)
Migraine and headache	56 453 (8.3)
Chest and cardiac pain	37 437 (5.5)
Visceral abdominal or gastrointestinal pain	31 179 (4.6)
Neuropathic/neurological pain	27 065 (4.0)
Complex regional pain syndrome (CRPS) and chronic pain	19 993 (3.0)
Shingles-related pain	14 772 (2.2)
Fibromyalgia	140 (0.02)
Mental health diagnoses	335 122 (49.5)
Depressive disorders	113 760 (16.8)
Anxiety symptoms/disorders	67 497 (10.0)
Psychotic disorders	5263 (0.8)
Suicidal and distress behaviours	763 (0.1)
Bipolar disorders	6641 (1.0)
Other mental health conditions	27 317 (4.0)
Prescribed medication for mental health condition without diagnosis	168 338 (24.9)
Substance use disorder diagnoses	13 618 (2.0)
Alcohol	6760 (1.0)
Opioid	2121 (0.3)
Sedative/hypnotic	879 (0.1)
Cannabis and cocaine	798 (0.1)

*Note that patient could have multiple diagnoses, therefore totals may exceed the major category total.

Figure 2 shows the opioid prescriptions received by patients in participating practices, which were measured by duration of consecutive opioid prescribing. The duration of consecutive opioid prescribing was estimated using spell analysis. A spell, or period of opioid prescribing, was defined as receipt of at least two consecutive prescriptions of any type within a 60-day period, with gaps between spells defined as 60 days in which there was no opioid prescription. Patients who only received a single prescription were not included in the calculation of opioid prescription spells. Figure 2A illustrates the distribution of the total number of opioid prescription periods for patients who had more than one opioid prescription between 1 January 2015 and 31 December 2020. The majority of the cohort had relatively few periods of opioid prescription, with 32.2% and 34.6% having only one and two prescription periods, respectively. Figure 2B shows the duration of the longest period of consecutive opioid prescribing among patients who received two or more opioid prescriptions. Of those who received opioid prescriptions over more than one prescribing period (n=349 045), most (66.8%, n=233 224) received consecutive opioid prescribing for less than 4 months, and 2.7% (n=25 547) received opioids for at least 12 months. The median duration of an opioid prescribing spell was 31 days (IQR: 21–43). The median interval between periods of opioid prescribing was 8 months (IQR: 4–14) (see online supplemental appendix figure 1).

Figure 2.

Distribution of consecutive (60-day periods) opioid prescribing (A) and the distribution of duration of longest period of consecutive opioid prescribing in months (B), 2015–2020 (n=349 045 people who received >1 opioid prescription).

Figure 3 shows the proportion of patients who received prescriptions for specific opioids and coprescribed medications over time, represented as the proportion of all participants who received any opioid prescription in that year. Codeine and oxycodone were the most frequent opioids prescribed (figure 3A), while hydromorphone and pethidine were the least frequent (figure 3B). The proportion of patients receiving codeine prescriptions fluctuated over time, decreasing from 34.0% to 30.8% between 2015 and 2017, and then increasing to 35.0% in 2018 (coinciding with the removal of non-prescription codeine sales in Australia), before decreasing to 29.6% in 2020. The proportion of patients receiving oxycodone and tramadol prescriptions gradually decreased by 3% between 2015 and 2020. Tapentadol prescriptions increased substantially from 0.7% of patients receiving an opioid prescription in 2015, to 3.1% in 2020 (an increase of more than 400%) (figure 3B). A steady decline in some coprescribed medications was seen in this cohort (figure 3C). In 2015, nearly 35.0% of patients received NSAID prescriptions, decreasing to 27.9% in 2020. The proportion of patients who were prescribed antidepressants and benzodiazepines decreased by 3% and 5%, respectively. Antipsychotics, gabapentinoids and ‘Z’-drugs (ie, zopiclone, eszopiclone, zaleplon and zolpidem) were the least common type of coprescribed medications in this cohort, and small changes in the proportion of patients receiving these prescriptions were observed.

Figure 3.

Proportion of cohort participants receiving prescriptions for specific opioids in each year (A and B) and changing trends in the proportion of patients received other coprescribed medications represent as the proportion of all participants who received opioids in that year (C).

What are the main strengths and weaknesses?

Strengths of this cohort include the use of a large primary care data set to provide new insights into opioid prescribing including the impact of policy changes and patient characteristics on opioid prescribing. This data set provides diagnostic data and other rich patient-level information, and includes all opioids prescribed, irrespective of whether their cost is subsidised. Many Australian cohorts are only able to access data on subsidised opioid medicines, which does not include opioids with rapidly increasing use such as IR tapentadol formulations.²² The inclusion of diagnoses codes within the data set provides a richer clinical picture than examining prescribing data alone, and will enable examination of opioid prescribing in patients who have received diagnoses relating to specific conditions and comorbidities.

A consideration when interpreting prescribing data is the inability to confirm if prescribed opioids were dispensed and subsequently taken, so these data are best considered a representation of prescribing as opposed to opioid consumption. As with the use of all administrative and clinical data, these data were not collected for research purposes, and are subject to limitations including the potential for incomplete information. For example, even though prescribing data are likely to be relatively complete and accurate as the electronic system is used to directly create the prescriptions, a prescriber may not document a diagnosis for each occasion where a service or prescription is provided. In addition, from 1 February 2018, codeine was rescheduled to a prescription-only medicine. Therefore, prior to 2018, codeine supply is likely to be underestimated since it could be purchased over-the-counter. Our study focusses on opioid prescribing in primary care, and therefore results may not be generalisable to prescribers in other settings. A final consideration is that patients are uniquely identified at the practice level, so where patients attend multiple primary care practices, it is not possible to link their profiles. In the state of Victoria, where the study was conducted, this concern is partially mitigated for long-term opioid use because a permit is required when controlled (Schedule 8) opioids are prescribed for at least 8 weeks, limiting the patient to a single prescriber.

To ensure the data quality, we have undertaken a number of data quality assessments including (a) checking data for completeness/missingness, (b) conducting logic checks to assess the internal consistency of the database and (c) conducting an outlier analysis to identify data points outside logical value ranges (eg, number of prescriptions or prescription quantities).

Despite these limitations, this cohort represents one of the largest primary care cohorts of opioid prescribing in Australia, representing a diverse range of opioid prescribing from acute to long-term opioid treatment. With 6 years of follow-up data already available, this enables analyses of opioid prescribing over time, including among key subpopulations and during periods of key opioid policy changes. Through the rich patient-level data contained in the primary care data set, there is a detailed range of covariates for inclusion in analyses to enable exploration of opioid prescribing trajectories during this period of substantial opioid policy change.

Future plans

The data available in OPPICO can be used for various types of pharmacoepidemiological research. For example, we plan to examine trends of benzodiazepine and opioid coprescription, and high dose opioid prescribing as part of planned analyses related to this work.¹⁹ Next, as an extension of this already funded work, we will link the person-level records with hospital data across the same three health regions to determine if there are increased or decreased harms (eg, heroin or other drug-related poisonings, or suicide-related attendances) resulting from prescribing changes as captured by emergency department and hospital admissions. There are clear examples where opioid policy has had serious unintended consequences in other locations, resulting in rapid tapering or opioid cessation, with devastating outcomes.¹¹ We will identify if similar outcomes are observed in Australia. Finally, the primary care data available in OPPICO can also be used for monitoring trends and patterns of other drug utilisation before and after the implementation of a broad range of policy interventions that could influence prescribing, or to observe changes over time, in multiple areas of health.

Collaboration

All data from OPPICO presented in this article are stored by the research group of the authors on the Monash Secure eResearch Platform (SeRP) at Monash University. Data are accessible through an application to Outcome Health https://www.outcomehealth.org.au/polar.aspx. Researchers from universities, hospitals and other organisations can apply to access subsets of deidentified general practice data to conduct research relevant to PHNs and General Practice by emailing research@outcomehealth.org.au.

Data availability statement

Data may be obtained from a third party and are not publicly available. Data may be obtained from Outcome Health.

Ethics statements

Patient consent for publication

Not required.

Ethics approval

Ethical approval was received from Monash University Human Research ethics committee (Ref. 24139).