Association of state‐level prescription drug monitoring program implementation with opioid prescribing transitions in primary care in Australia

This study aimed to evaluate whether voluntary and mandatory prescription drug monitoring program (PDMP) use in Victoria, Australia, had an impact on prescribing behaviour, focusing on individual patients' prescribed opioid doses and transition to prescribing of nonmonitored medications.

There has been increased opioid prescribing and related harms in recent decades, particularly in Australia, the UK, Canada and the USA.
For example, in Australia, the rate of hospitalisations involving natural and semi-synthetic opioids doubled from 1999-2000 to 2018-2019 (3.5-7.0 hospitalisations per 100 000 people, respectively). 1 Furthermore, opioids have been the main drug cited in drug-induced deaths for >2 decades, comprising 2/3 of all drug-related deaths. 2 In 2019 alone, opioids were the underlying cause of 1121 deaths in Australia, among which 873 were classified as unintentional deaths.
High-dose opioid consumption is associated with an increased risk of fatal unintentional overdoses. 2Various strategies to prevent harm from prescription opioids, with a focus on those receiving higher doses have been implemented, including prescription drug monitoring programs (PDMPs).PDMPs are electronic databases that monitor patients' prescribing and dispensing information for controlled substances, including opioids, using a predefined algorithm to trigger alerts associated with high-risk scenarios, such as high-dose or highrisk drug combinations. 3idence relating to the impact of PDMPs as an opioid risk mitigation tool remains mixed.For example, 1 recent systematic review revealed US-based PDMPs reduced opioid prescribing, opioid diversion and supply, and opioid-related morbidity and substance-use disorder outcomes. 4Contrary to this, other evidence suggests that PDMP implementation can result in serious unintended consequences, such as rapid tapering or abrupt opioid cessation, resulting in poor clinical outcomes. 5Rigorous evaluations demonstrating the effects of PDMPs on nonfatal and fatal opioid overdoses are lacking. 6Furthermore, the features of individual PDMPs, including requirements and mandates to check the PDMP, the medications they monitor and the information displayed and captured within these programs appear to be important in determining the impacts of a PDMP system. 7Inconsistent effects of PDMPs across studies may also be due to variations in evaluation approaches and, to some extent, heterogeneous policy effects across time, regions, and subpopulations.
[14][15] PDMPs have now been implemented throughout Australia, with the first iteration of a real-time PDMP implemented in Tasmania in 2009. 16Victoria, the second most populous state in Australia, comprising about 26% of the Australian population, 17 implemented its PDMP in 2019.Use was initially voluntary, before being the first state to make PDMP use mandatory in April 2020, for all community-based prescribers and pharmacists.Victoria's PDMP adopts a red, amber or green traffic light notification system to highlight patients at risk of medication-related harm.A red alert is triggered if a patient has been prescribed a high dose (i.e.exceeds 100 mg oral morphine equivalents [OMEs]), a high-risk drug combination (e.g.opioid and benzodiazepine), or has multiple prescriber episodes, with alerts being triggered based on prescribing and dispensing in the past 90 days. 18Mandatory requirements stipulate that community prescribers and pharmacists are required to examine patient data within the PDMP prior to prescribing or dispensing monitored medications, and neglecting this obligation may lead to an inquiry and potential penalties.In Australia, over 50% of opioids are initiated by general practitioners (GPs), yet we have little understanding of how their opioid prescribing is influenced by PDMP use, 19 or how PDMP use impacts individual patients' pain treatment, including the use of nonmonitored medications.
Our study addresses this gap and aims to evaluate whether PDMP implementation impacts prescribing for individuals' opioid doses and if there is evidence of an effect on prescriber behaviour, such as substitution effects, that is, whether individual patients are prescribed nonmonitored medications.

| Study design and data source
This is a retrospective study using routinely collected primary healthcare data generated from general practice clinics within 3 Primary Health Networks (PHNs) in Victoria, Australia.These PHNs were Gippsland, East Melbourne and South Eastern Melbourne and represent 48% of the Victorian population.PHNs are independent

What is already known about this subject
• Inconsistent effects of prescription drug monitoring programmes (PDMPs) on prescribing have been found across different studies.
• There is little understanding of how PDMP use impacts prescribing for individual patients, including the prescribing of nonmonitored medications.
• PDMP use may lead to a rise in prescribing of alternative medications that are not monitored, known as a substitution effect.

What this study adds
• We describe the effects of voluntary and mandatory PDMP use on prescribing of opioid and nonopioid medications.
• We did not observe an increased probability of individuals transitioning from being prescribed opioids to nonmonitored medications during the implementation of PDMP.
organizations that are funded to manage health regions, of which there are 31 PHN regions in Australia and 6 in Victoria.
Prescribing data were extracted from the POpulation Level Analysis and Reporting (POLAR) general practice analytics platform, managed by Outcome Health.The details of the POLAR data have been comprehensively documented elsewhere. 20,21Briefly, the POLAR database contains nonidentifiable routinely collected data extracted from electronic health records from participating practices.
For the current analyses, data on patient demographics, diagnoses and prescription medications were obtained from 562 GP practices.A set of 2 statistical linkage keys (SHA-256 HASH key) using personal identifying information was provided for linking the sub-datasets Patients included in this analysis were at least 14 years old at the POLAR cohort entry (1 January 2017), with at least 4 opioid prescriptions over the study period.Patients were included in the analysis if they had at least 1 GP consultation during each observation period enabling confirmation they remained active patients attending the practice during the entire study period.Patients were excluded if they had a cancer diagnosis at any point during the study period, as per SNOMED CT codes and validated by Outcome Health using a systematic approach. 20

| Defining opioid dose groups
All opioid analgesics marketed in Australia during the study period were included in the calculation of opioid doses.Opioid prescriptions that are used for opioid replacement therapy or cough (e.g.high-dose sublingual buprenorphine, methadone liquid and dihydrocodeine) were excluded from this study.The World Health Organization's Anatomical Therapeutic Chemical (ATC) code was used to identify eligible opioid prescriptions. 22For opioids that are used for both analgesia and opioid dependence treatment, generic names were checked to confirm that only opioid products used for analgesia were included (Table S1).
Opioid doses in each prescription were converted into OMEs, using published opioid conversion rates. 23In line with Spooner et al.'s methodology, 24 a daily OME dose was estimated by considering all opioid prescriptions provided in a 90-day period, using a 90-day moving average calculation to avoid nonsignificant variations caused by briefly overlapping prescriptions and small intervals between prescription fills.The total volume of opioids prescribed was determined by converting each prescription into an OME dose, which was then summed over a 90-day period.The daily OME dose was then calculated by dividing this total volume over 90 days by 90.We use a 90-day period to be consistent with the approach adopted by algorithms to calculate daily opioid dose in the PDMD. 18ree prescribed opioid dose categories were defined for the current analysis: (i) high dose group (>100 mg OME), similar to that which would trigger a red alert in the PDMP, (ii) low-medium dose group (≥20 and ≤100 mg OME); and (iii) low dose group (<20 mg OME). 25 A no-opioid group was also created if patients did not receive an opioid prescription during the observation period.Patients were classified into 1 of these 4 dose categories in each of the 4 observation times, based on the highest values of the 90-day moving average of daily OME.
To account for opioid prescriptions that enabled multiple supplies (often referred to as repeats or refills), additional records were imputed at 28-day intervals for slow-release opioids or patches, and 14-day intervals for other opioids, based on standard 1-month supplies of long-term medications in Australia, with smaller supplies of immediaterelease opioids as standard supply quantities.This approach is described elsewhere. 26hey were selected as they are commonly prescribed in the treatment of pain but were not monitored by the Victorian PDMP during the study period.We assigned patients to 4 different treatment categories during each observation period (i.e.T1 to T4).These categories included being prescribed: (i) no opioids or nonmonitored medications;

| Defining prescription groups
(ii) opioids only; (iii) opioid and nonmonitored medication (i.e.any of the 4 types of medications described above); or (iv) nonmonitored medications only.

| Analysis
We described the proportion of patients in each opioid dose group and treatment group during the control period (T1), the second control period prevoluntary PDMP (T2), voluntary PDMP (T3) and the mandatory PDMP period (T4).To investigate how patients transitioned between different prescribed opioid doses, a Markov transition matrix was created.The transition probability was estimated to represent the proportion of patients who transitioned from being prescribed 1 dose category to another at the subsequent observation time.Given that tramadol was not currently monitored via Victoria's PDMP during the study period, a sensitivity analysis was undertaken, specifically by excluding tramadol prescriptions from the dose transition analysis (Figures S2 and S3).The same transition matrixes were produced to examine how patients transitioned between treatment options.Transition pathways were visualized.To compare changes between 2017 and 2020, we calculated differences in the transition probabilities from T2 to T3 (when voluntary PDMP was introduced) compared to that from T1 to T2 (2 periods with no PDMP implementation, considered the control transition period), as well as the differences in the transition probabilities from T3 to T4 compared to that from T2 to T3.

| RESULTS
We identified 118 248 patients with at least 4 opioid prescriptions who met our eligibility criteria.The characteristics of the cohort are summarized in Table 1.There were more females than males in this cohort (62.6% vs. 37.1%).Half of the cohort (56.6%) were recorded as concession card holders (i.e. are eligible for subsidized health services and medications).Only 5% of the cohort were aged <25 years old, and about 50% were aged ≥60 years.Most patients lived in metropolitan areas (77.1%).One third (30.4%) lived in the least socioeconomically disadvantaged deciles, with 12.6% living in the most disadvantaged deciles, according to the Index of Relative Socioeconomic Disadvantage.The median number of opioid prescriptions prescribed per patient was 9 (interquartile range: 5-25), and the mean number was 25.4 (standard deviation = 44.9).
Overall, during each observation period, approximately 85% of our sample were identified as being prescribed <20 mg OME or no opioids, while about 10% were prescribed 20-100 mg OME (Table 2).
The remaining $3% were prescribed >100 mg OME.The proportion of patients who were prescribed <20 mg OME doses increased by 8% during the prevoluntary PDMP period (i.e. from 42.0% at T1 to 50.1% at T2) and remained at about the same level during PDMP implementation (T3 and T4).The proportion of patients who were prescribed >100 mg OME was relatively consistent during each observation period (3.2-3.6%).
In this cohort, the majority (42.0-50.7%)were only prescribed opioids, while smaller proportions were prescribed both opioids and nonmonitored medications (13.7-16.7%)or nonmonitored medications only (3.8-4.2%;Table 2).The proportion of patients prescribed only opioids increased from 42.0-49.6% in the prevoluntary PDMP period (T1 to T2) compared with the control period and stayed stable in the voluntary PDMP period (T3) but decreased again to 44.0% after mandatory PDMP implementation (T4).A similar pattern was also observed for those prescribed both opioids and nonmonitored medications.The proportion of patients who were prescribed  nonmonitored medications only, was slightly higher during the mandatory PDMP period (T4) compared to the other 3 observation periods.
The patient-level transition probabilities across the 4 dose groups during the study period are demonstrated in the translation matrix diagram (Figure 2).The arrows in the diagram represent the percentage of patients who stayed in the same group or transitioned to another group.Of the patients prescribed low-dose opioid prescriptions (<20 mg), about 2/3 continued to be prescribed low dose opioids, while the majority of the remainder discontinued opioids during the following observation period.Transitioning to being prescribed a high dose (>100 mg) was relatively rare for this group (0.3-0.4%).Patients in the high dose group had the most stability in group membership, with about 70% staying in this same group across all 4 periods.Among patients prescribed 20-100 mg OME, about 60% remained in the same dose group across all periods, $17% moved to the low dose group, 12.9-14.5% were not prescribed opioids in the following observation time, while a small percentage switched to the high dose group ($7%).
Figure 3 describes the differences in transition probabilities among dose groups when comparing the periods before (T2) and after (T3) voluntary implementation, as well as the differences following mandatory PDMP implementation (T3 to T4) compared to voluntary PDMP implementation (from T2 to T3).Overall, there was a change in the probability of opioid dose groups after PDMP was implemented.
For those initially in the low dose (<20 mg OME) group, the transition to higher dose groups decreased following voluntary PDMP implementation (from T2 to T3) compared with the control period (T1 to T2; 0.3 vs. 0.4%), while the transition to not being prescribed opioids increased by 1%.However, when voluntary PDMP was implemented (T2 to T3), those in the low dose group (<20 mg OME) had a decreased probability (À3.9%) of continuing to be prescribed the same dose following mandatory PDMP (from T3 to T4).In contrast, the transition to no opioids over the same period increased by 4.2%.Among those who were prescribed >100 mg OME opioid prescriptions, the probability of continuing to be prescribed a high-dose decreased by 2.2% and the probability of transitioning to being prescribed 20-100 mg OME increased by 0.5% from T3 to T4 compared to the T2 to T3.For all 3 opioid dose groups, the transition to the no-opioid group increased following mandatory PDMP implementation (T3 to T4) compared to voluntary PDMP implementation (T2 to T3).
We also identified several different treatment transition pathways (Figure 4).Among patients prescribed opioids only, the majority (59.8-64.1%)continued to be only prescribed opioids, about 30% of patients discontinued receiving opioids and a small proportion transitioned to being prescribed a combination of opioids and nonmonitored medication prescriptions (7.2-8.4%) or only prescribed nonmonitored medications (1.4%).Among those being prescribed both opioid and nonmonitored medications, nearly 2/3 continued to be prescribed these medications, and about 10% switched to being prescribed nonmonitored medications only.Overall, during a subsequent observation period, 35.0-39.0% of the patients prescribed nonmonitored medications only, transitioned to being prescribed both opioids and nonmonitored medications.
Figure 5 indicates that following voluntary PDMP (T2 to T3), compared with the control period (T1 to T2), the variation of the transition rates across all treatment groups was <2%.In all treatment groups transition to the no-treatment group increased by about 6% following mandatory PDMP (T3 to T4) compared with following voluntary PDMP (T2 to T3).Those patients who were prescribed only opioids during T3 were less likely to be prescribed opioids during T4 and were more likely to be prescribed nonmonitored medications or no treatment (difference, À4.3%).In contrast, those who were only prescribed nonmonitored medications during T3 were more likely to be prescribed both opioids and nonmonitored medications (difference, 4.0%) but were less likely to be prescribed opioid treatment as the sole medication (difference, À2.8%).

| DISCUSSION
This study describes the effects of voluntary and mandatory PDMP use on opioid and nonopioid prescribing.The probability of moving  between different opioid dose groups and to nonopioid treatment groups was larger following voluntary PDMP (T2 to T3) and mandatory PDMP (T3 to T4), compared to the transitions observed during the control period (T1 to T2).An increase in the probability of moving to the no-opioid group during the mandatory PDMP period was observed, yet most patients who were prescribed high-dose opioids remained in the same dose group, with a relatively consistent rate of change at each time point, suggesting limited effects of PDMP implementation on this group.Of interest, we noted that for those who were only prescribed nonmonitored medications, the probability of being prescribed opioids increased during the mandatory PDMP when compared to other transition periods.We are not certain of the reason for this change; however, previous studies have found that for a small proportion of patients, opioid prescribing can increase following PDMP implementation, possibly due to increased prescriber confidence in prescribing opioids with the additional information provided by PDMP. 14udies investigating the impact of PDMPs on opioid prescribing doses and treatment duration have yielded varied outcomes, 6,27,28 few of which focused on transitions of care at the individual patient level, especially for patients on high-dose opioids.The Third Australian Atlas of Healthcare Variation reported an overall 18% reduction in opioid dispensing rates nationally between 2017 and 2021. 29This could be attributed to a series of national and state initia- The orange arrow suggests a transition to a higher dose group, while the green arrow suggests a transition to a lower dose group.mandated PDMP use and this was largely due to patients switching from being prescribed a low dose (<20 mg OME) to being prescribed no opioids.The dose group transitions remained relatively stable for most patients receiving high doses (>100 mg OME), with a small decrease in the number of patients being prescribed a high opioid dose following voluntary and mandatory PDMP implementation.This suggests that PDMP implementation did not influence the prescription of high-dose opioids for most patients.There is a range of possible explanations for why the intended target of the PDMP systems (i.e.patients receiving high dose opioids) were not more affected.It may be that prescribers' decision-making is more strongly influenced by the patient-prescriber relationship, 30 and therefore the effects of PDMP use may vary depending on that relationship.In these studies, the PDMP was used less routinely by GPs for existing pain patients compared with new patients, and clinicians were less likely to prescribe opioids to new patients other than for acute and verifiable conditions.Our findings are consistent with this and suggest that PDMP use may be effective in reducing prescribing low-dose opioids for naïve patients in the longer term, in contrast to influencing long-term higher dose opioid prescribing.Studies examining longer-term outcomes are needed to establish if positive longer-term effects are seen, such as fewer patients developing persistent opioid use.
To reduce reliance on opioids, alternative medications, such as gabapentin, pregabalin and tricyclic antidepressants, have garnered attention for pain management. 12Existing research has shown that PDMP use has led to substitution effects at both the population and individual levels, where alternative medications are prescribed instead of opioids. 26,31In Australia, pregabalin is subsidized by the Pharmaceutical Benefits Scheme solely for the treatment of refractory neuropathic pain and not as a first-line therapy for neuropathic or other types of pain. 32Tricyclic antidepressant medications could offer moderate pain relief in specific neuropathic pain conditions 33 but provide limited to no benefits for other types of pain.We did not observe an increased probability of individuals transitioning from opioids to nonmonitored medications after the implementation of voluntary or mandatory PDMP.Our analysis revealed an increased likelihood of individuals on nonmonitored medications switching to opioid treatment during the mandatory PDMP period.This finding suggests a potential impact of real-time access to a patient's prescription history in increasing the prescriber's confidence in prescribing. 14,34

| STRENGTHS AND LIMITATIONS
This study presents findings from 1 of Australia's largest primary healthcare datasets, leveraging comprehensive records for medications prescribed irrespective of their subsidy status.We considered opioid and nonmonitored prescriptions prior to and during However, there are some study limitations to be noted.First, Victoria's implementation of mandatory PDMP on 1 April 2020 coincided with COVID-19 restrictions over the same period.This included reduced surgeries, reduced face-to-face consultation and reduced hospital attendance.It is possible that those restrictions may have impacted overall prescription opioid initiation and lowered acute opioid prescribing. 35Although the daily OME was estimated using a 90-day moving average approach, which removes noise due to ongoing fluctuations due to overlapping prescriptions or small intervals between prescription fills, it may be affected by variations in opioid supply due to these changes in the healthcare system due to COVID-19.Secondly, changes in an individual's dose group or treatment group within a single period were not considered in this study.Third, the POLAR database only collects prescribing data, and cannot determine whether prescribed medications were later dispensed and/or taken.As such, it is best placed to look at the impact of PDMP on prescribing practices as opposed to medications supplied to patients or medication use.Notwithstanding this, previous research suggests that there is a high correlation between prescribed and dispensed medicines. 36,37Fourth, the data only capture prescribing within participating POLAR practices.There is the possibility that patients may have visited multiple practices including those not participating in POLAR, although recent Australian data suggest the impact of this would be minimal, with >80% of Australians seeing the same GP, and a further 15% seeing a GP in the same practice. 38Due to the requirement in Victoria to have a permit for strong opioids being prescribed for >8 weeks, 39 combined with the common practice of attending the same GP practice in Australia, it appears unlikely that this would have had a significant effect on our OME estimates or overall conclusions.Finally, while we were able to examine the initial outcomes of mandatory PDMP implementation in Victoria, long-term and national evaluations are still required.

| CONCLUSION
Our study suggests that the implementation of mandatory PDMP had a significant impact on patients who were prescribed low-dose opioids, with a limited impact on those prescribed higher opioid doses or nonmonitored medicines for pain.
within the POLAR database.The keys are developed through ORCA, a specialist linkage software developed and used by Outcome Health, which enables the identification of the same patients at different practices (removing patient duplication) without the need to export personal identifying data.Primary care medication records were extracted between 1 April 2017 and 31 March 2021.To enable a comparison period of transitions in the absence of PDMP implementation, the study period included 4 observation periods (Figure 1): T1: a control period where no change in PDMP access occurred (1 April 2017 to 31 March 2018); T2: second control period, 12 months prior to voluntary PDMP implementation period (1 April 2018 to 31 March 2019); T3: voluntary PDMP implementation period (1 April 2019 to 31 March 2020); and T4: mandatory PDMP implementation period (1 April 2020 to 31 March 2021).

T A B L E 1
Sociodemographic characteristics of the cohort (N = 118 248).

T A B L E 2
Changes in the proportion of patients (N = 118 248).
tives including regulatory changes to reduce opioid pack sizes, educational programs highlighting new ways to manage pain, alongside PDMP implementation.People prescribed opioids only, had an increased probability of transitioning to the no-opioid group following F I G U R E 2 Markov transition matrix of dose groups across 4 observation periods: (A) T1 to T2 (control period to prevoluntary prescription drug monitoring program [PDMP]); (B) T2 to T3 (prevoluntary PDMP to voluntary PDMP); (C) T3 to T4 (voluntary PDMP to mandatory PDMP).

F I G U R E 3
Percentage changes in transition probabilities among opioid dose groups from 2 comparisons: voluntary prescription drug monitoring program (PDMP; T2 to T3) vs. control transition period (T1 to T2) and mandatory PDMP (T3 to T4) vs. voluntary PDMP (T2 to T3).OME, oral morphine equivalent.voluntary and mandatory PDMP implementation periods to enable an examination of patient level changes in prescribing associated with PDMP availability.Patient level data allow for a more nuanced analysis, as population rates can potentially obscure the impact of interventions or treatments by overshadowing individuallevel effects.