Australia has a universal health care system providing access to subsidised prescription medicines to citizens and eligible residents and clients of the DVA via the PBS and the Repatriation Pharmaceutical Benefits Scheme (RPBS), respectively. People contribute a co-payment towards the cost of their medicines, which varies depending on their entitlements. Our review focusses on studies using routinely collected data on medicines dispensed through PBS and RPBS. These dispensing claims are processed by Services Australia (previously the Department of Human Services and Medicare Australia) and are provided to the Australian Government Department of Health and the DVA for monitoring, evaluation, and health service planning. These data are available to third parties, publicly or by request, for monitoring, evaluation, and research (see Supplementary Table 1).

We searched Medline and Embase from January 2013 through December 2018 using a combination of keywords and search terms describing medicines use (e.g. prescription drugs, drug therapy, drug utilisation) with PBS dispensing data sources (see Supplementary Appendix A for search strategy). We also conducted searches on key researchers in the field of medicines policy research in Australia and screened the reference lists of all included studies (Figure 1).

We included full-text English-language studies using PBS and/or RPBS dispensing claims data to measure patterns of medicines use or using medicines as a proxy of a health condition or an outcome. We excluded studies: focussing exclusively on medicine expenditure or modelling; using dispensing data obtained directly from pharmacies; requiring individual informed consent to access dispensing data; or using data derived from state-based registries.

Two reviewers (CB, SP) screened a random 20% sample of titles and abstracts independently to identify potentially relevant studies for inclusion; one reviewer (CB) screened the remainder. Two reviewers (JOC, CB) extracted data independently from all included studies and disagreements were resolved by discussion. We extracted the following key features of each study (Box 1):

We classified the broad study focus into six themes; (1) Medicine utilisation: examined trends and patterns of dispensing overall or stratified by gender, age, and medicine or additional variables; (2) Clinician practices: used individual-level data to study prescribing patterns (e.g. concomitant or inappropriate prescribing); (3) Patient practices: used individual-level data to examine patient behaviour around medicines use, such as medicine persistence or adherence; (4) Exposure and outcomes: 4A) investigated the relationship between medicine use and at least one outcome, such as death or hospital admission (‘medicine use and outcomes’), OR 4B) investigated the relationship between other exposures (e.g. device use) and at least one outcome but used dispensing claims to define a cohort, comorbidities or an outcome (‘other exposure and outcomes’); (5) Intervention impacts: examined the effect of one or more population-level interventions on prescribing or another outcome, classified as educational (e.g. prescriber feedback and education), policy (e.g. subsidy changes and restrictions), media (e.g. advertising campaigns), or multi-faceted (combination of the above); (6) Methods: used dispensing data to develop and refine pharmacoepidemiological techniques (e.g. validation of prescribing indicators) or study protocols reporting data based on dispensing claims.

We assigned WHO Anatomical Therapeutic Chemical (ATC) classifications to the medicine focus of each study [14]. We also report the proportion of studies according to their medicine focus relative to the proportion of PBS volume and spend for these classes by ATC code.

Due to the heterogeneity of study methodology, we did not assess individual study quality. However, we extracted 23 items pertaining to the REporting of studies Conducted using Observational Routinely-collected health Data (RECORD) checklist [15] to describe areas of underreporting. Two reviewers (JOC, CB) independently reviewed all articles published in the most recent year (2018); disagreements in extraction were resolved by discussion. For each item (see Supplementary Appendix B) we allocated a score of 1 if studies reported the item. As some items were not applicable for some studies, we calculated the RECORD score as the percentage of items meeting the criteria in relation to the overall applicable items for each study.

We report the results of this review according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) [16].

We observed a steady increase in the number of studies published annually and a sharp rise in 2018; this last observation year accounting for nearly one-third of all studies published in the period. Most studies used individual-level data (133 studies, 74%). The time between the study observation end date and publication was up to 2-years for 58% of studies; 16% of studies had a publication lag of more than 5 years. The median lag time for claims-level studies was 29 months and for individual-level studies, 34 months.

Approximately half of the 180 studies did not place age restrictions on their study cohorts (91 studies, 51%). The remaining studies restricted cohorts to people aged 65 years or older (55 studies, 31%) or people aged 18 years or older (26 studies, 14%). Five studies focussed on women of childbearing age and two on children. Approximately half of the studies restricted their populations to concessional beneficiaries or DVA clients (91 studies, 51%). Approximately 90% of studies using claims-level analyses used the entire PBS-eligible population. In contrast, approximately 65% of studies using individual-level analyses restricted their cohorts to the elderly, DVA clients or concessional beneficiaries.

Approximately two-thirds of studies leveraged dispensing claims and other health data (118 studies, 66%). Approximately 20% of studies used publicly available dispensing claims and 78% used data available by request, with a marked increase in the use of the PBS 10% sample and PBS ad hoc extracts over time.

Two-thirds of claims-level studies used publicly available data; 62% of these also included other unlinked health data. Individual-level studies used RPBS data (42%), the PBS 10% sample (29%) or ad hoc data extracts (22%). These individual-level studies were Australian-wide or restricted to residents of Western Australia and/or New South Wales. Seventy of these studies linked individual-level dispensing claims to other health data such as hospitalisation data, medical services claims, residential aged care claims, emergency department data, or cancer and perinatal registries; 66 were linked across jurisdictions (data not shown in table).

Approximately one-third of all studies used individual-level data to examine clinician or patient practices (47 and 16 studies, respectively). Individual and claims-level exposure-outcomes studies accounted for 21% of all studies, 27 of the 38 studies evaluated medicine safety and 6 evaluated medicine effectiveness. One-fifth of all studies used claim-level data to investigate medicine utilisation (36 studies); methodological studies and those evaluating intervention impacts each accounted for around 24% of all studies (25 and 18 studies, respectively).

The most commonly studied medicines were those acting on the nervous system (38%) and cardiovascular system (23%), followed by those acting on the alimentary tract and metabolism (14%). In general, the most commonly studied medicines groups were also the medicines groups accounting for the greatest proportion of PBS dispensing in 2018. However, this medicine focus does not align with the proportion of PBS expenditure. For example, PBS expenditure with antineoplastic and immunomodulating agents represented 32% of the PBS spend in 2018 but less than 10% of the studies published in this review.

Of the 55 studies published in 2018, we excluded seven methodological studies for which most of the RECORD items would not be applicable. From the 48 studies evaluated, the median RECORD score was 95% (interquartile range 90–100%); 13 (27%) studies scored 100%. The most underreported items were: study design, either by not reporting this item in the abstract (14 studies, 30%) or in the methods (9 studies, 19%), followed by the type of data used (14 studies, 30%), and methods of population selection (8 studies, 17%). Moreover, two-thirds of studies using linked data did not report the use of linked data in the abstract.

First, the availability of a standardised data collection of person-level dispensing claims for a 10% sample of PBS-eligible people (“PBS 10%”) has contributed to the rapid increase in the number of studies using individual-level dispensing claims over time. The PBS 10% sample dataset, established in 2005, contains the entire PBS-claims history for a 10% random selection of PBS-eligible Australians. To minimise the risk of re-identification, the data is limited to a population sample, offset dates of dispensing by up to 14 days (but identically for each person), and it is not permitted to be linked to any other dataset. The collection is provided to approved third parties on a fee-for-service basis, has a streamlined governance process and approved organisations can hold longitudinal data that is updated at least quarterly. The earliest research studies using this collection were published between 2008–2013 [17–22]. In the period of the current review, 39 studies have been published using this collection. The governance arrangements allow relatively rapid turnaround for approval of studies using contemporary data. This is a model that should be replicated across other data collections, including those with PBS dispensing claims linked to other health datasets.

Second, individual-level studies using PBS data prior to 2012 were often restricted to people receiving government entitlements to ensure complete capture of dispensing records [23]. The 2012 reform allowing the capture of all PBS dispensings (irrespective of whether they attracted a government subsidy) led to an increase in individual-level studies conducted across the entire eligible Australian population, not just in people receiving government benefits [13]. However, the collection does not contain information on private prescriptions, has limited capture of highly specialised medicines dispensed in public hospitals prior to 2013 and no information on prescription indication, prescribed daily dosage, and treatment duration. These limitations are not uncommon in community-based dispensing claims data, but it is important to consider these in pharmacoepidemiological study designs [23, 24].

With respect to undertaking exposure-outcomes studies, the Australian Institute of Health and Welfare’s development of multi-source enduring linked data assets (MELDAs) comprising continuing cross-jurisdictional, person-level linkages of medicines exposure with hospitalisation and mortality data show strong potential to further accelerate national population-based research capacity [25, 26]. At the time of writing, there were no formal policies around third-party access (including to academic researchers) to the current suite of MELDAs; this should be considered an immediate priority to realise this significant investment in public money [27].

In July 2019, quality use of medicines and medicines safety was announced as Australia’s tenth national health priority [28, 29]. Studies catalogued in this systematic review provide contemporary evidence assessing quality use of medicines including the impact of medicines policy interventions, [30–32] medicine use in populations not always represented in clinical trials, [33, 34] and adherence with current treatment guidelines [35–37]. However, there is a need for greater focus on outcomes studies, especially pertaining to medicine safety, and with greater attention to vulnerable population sub-groups [38].

Despite advances, studies examining clinician and patient practices, as well as medicines utilisation studies, still represented a large proportion of the body of literature (Supplementary Figure 3) [13]. Further, the evidence base is still dominated by studies on cardiovascular medicines and those acting on the nervous system and in elderly Australians. Significant blind spots remain in our understanding of real-world medicine effectiveness and safety, particularly in Australians who do not receive government benefits and in populations consistently excluded from clinical trials, such as women of childbearing age and children. In this context, individual-level dispensing claims linked to health outcomes data, would provide a deeper understanding of the benefits and harms derived from medicine use, including indications for prescribing, clinical diagnoses, and other patient risk factors.

Historically, researchers have faced trade-offs between the ease of using readily available individual-level data, such as stand-alone PBS dispensing claims with limited clinical information (comprising the majority of individual-level studies in our review) or investing in the long process of gaining approvals and access to linked data [39, 40]. Encouragingly, we observed an increasing number of studies based on dispensing claims linked at the individual level to other data sources and we anticipate a further upswing in these types of studies in light of the major reforms underway in Australia. Particularly the new Data Availability and Transparency legislation, designed to maximise the value of Australian Government public sector data for service delivery and research. The legislation creates roles and responsibilities to data sharing. It adopts a guidance package to allow consistent practices across jurisdictions and safe sharing of data for public good purposes, including research and development, overriding secrecy provisions [41].

Other Commonwealth countries with similar health care systems and political structures, such as Canada and the United Kingdom, have bolstered their research capability by establishing independent centres serving the specific needs of the research community and closing the gap between linkage and analysis [42, 43]. Australia would benefit from adopting a similar model to harness data from its health care system covering over 25 million citizens and residents.

Our systematic review is not without limitations. We have focussed on studies using only routinely collected data and did not include studies using PBS data that required specific individual consent. We developed an arbitrary classification to classify studies by their main focus and given the high degree of variability both within and across studies, many could have been classified under alternative categories. Finally, we only addressed the reporting quality of studies published in 2018, identifying key elements that future studies should consider increasing their transparency and reproducibility and did not assess the methodological quality or relevance of included studies.