This section describes some of the common challenges encountered when using the PBS/RPBS database and its various extracts to obtain utilisation estimates.
Seasonality
PBS data are subject to seasonality due to the effect of the Safety Net (Fig. 1). As previously mentioned, when a family spends over a specified amount on PBS medicines in one calendar year (i.e. exceeds the Safety Net threshold), the cost of all subsequent PBS medicines are reduced to the concessional rate for general beneficiaries and are free for concessional beneficiaries. This reduced medicine price on reaching the Safety Net can lead to a phenomenon known as stockpiling: Safety Net entitlements result in some patients obtaining extra quantities of their medicines toward the end of the year, stockpiling for the new year when they revert back to paying standard prices. This results in increased rates of dispensing of the medicine toward the end of the year followed by a trough at the start of the next year. Despite attempts to reduce this phenomenon through the introduction of the Safety Net 20 day rule on 1 January 2006, stockpiling continues to result in pronounced seasonality in utilisation data based on date of supply.
Date of supply vs. date of processing
Each dispensing claim records the date a medicine is supplied to a patient by the dispensing pharmacy or the date the dispensing pharmacy’s claim for reimbursement is processed by DHS. Some of the data extracts include both of these variables while others include only one. For example, ASM reports are based around date of supply, PBS Statistics online use date of processing, and Section 85 online extracts are available by either date. There is often a discrepancy in utilisation measures based on date of supply versus those based on date of processing as the processing of the claim occurs some time after the prescription is dispensed and the interval of time between dispensing and processing is variable [25]. As such, caution must be employed when using data based around date of processing, particularly when examining medicine use at particular time points; these figures are primarily useful for obtaining rough approximations of utilisation. Date of supply should be used preferentially for examining medicine use.
Figure 2 depicts the discrepancy between utilisation estimates based on date of supply compared to date of processing for all s85 PBS-subsidised medicines. Troughs in utilisation according to date of processing can be seen around late 2011 and late 2013, indicating delays in the processing of claims by DHS. The 2011 trough is followed by compensatory peaks, indicating a period of increased processing. Note that the data by date of supply show seasonal fluctuations as mentioned above. Seasonal fluctuations are less apparent in date of processing graphs due to the variable delay between dispensing and processing.
The risks of using date of processing data are well demonstrated by a recent example. In October 2013, the Australian Broadcasting Corporation’s Catalyst program aired a two part series questioning the link between high cholesterol levels and heart disease and suggested that the benefit of statins for preventing cardiovascular disease had been exaggerated. Much public debate followed the program. In May 2014, Australian Doctor published an article based on PBS statistics online data, reporting that there had been no change in statin dispensing for up to 3 months after the program aired [31]. However, our interrupted time series analysis using date of supply showed that there was an immediate and sustained 2.6 % reduction (equating to 500,000 fewer prescriptions) in statin dispensing persisting up to 8 months after the program aired [32].
One caveat concerning use of date of supply data is that dispensing records are not available in the dataset until the claim has been processed by DHS. Due to the variable delay in processing, this can result in incomplete ascertainment of claims dispensed on a given date for a number of months. It is therefore advisable to truncate the observation end date by at least 3 months, and preferably 6 months, to avoid under-reporting of utilisation. Indeed, the Section 85 Date of Supply report does not contain data for the most recent 6 months to ensure that the data are of a satisfactory level of completion before publication.
Ascertainment and under co-payment medicines capture
As previously mentioned, the PBS database did not capture data on the dispensing of under co-payment medicines until at least April 2012 (July 2012 for Section 85 Date of Supply), thereby under-ascertaining the utilisation of certain medicines prior to this time. As under co-payment prescriptions comprised approximately 18 % of medicine use in 2011 [7], this issue significantly impacts utilisation estimates for certain drugs.
To determine if a particular PBS item was under-ascertained it is necessary to track the time points at which the cost of the medicine fell below the general co-payment threshold in the study period (prior to April 2012). As the co-payment threshold increases yearly and medicine prices change over time, the medicine’s price must be compared to the yearly co-payment threshold throughout the period of interest. The inclusion of under co-payment data in 2012 also must be considered when examining utilisation of under co-payment medicines over this period. It should also be noted that while DHS now records all under co-payment dispensing claims, not all of the collections detailed in this document incorporate under co-payment data as part of the extract (see Table 2).
An example of how these issues affect the data is provided in Fig. 3. Oxycodone suppositories (30 mg; PBS item code 2481N) were under co-payment between January 1998 and December 2004. Oxycodone utilisation was under-ascertained in the PBS dataset for the duration of this under co-payment period; only use by concessional beneficiaries or general beneficiaries qualifying for the PBS Safety Net were captured. In December 2004, the price of item 2481N increased from AUD$18.42 to AUD$29.84, exceeding the co-payment threshold of AUD$23.70. This transition to over co-payment resulted in more complete capture of medicine use; PBS-subsidised utilisation increased as a result, and under co-payment utilisation dropped off. As this medicine was above co-payment in 2012, it was not affected by the change in data collection by DHS.
However, the impact of this change on medicine ascertainment can be demonstrated by examining trends in antidepressant utilisation over this period. As many antidepressants are off-patent, there is a high rate of under co-payment utilisation of this class. Accordingly, Fig. 4 demonstrates a sharp increase in antidepressant prescriptions coinciding with the uptake of under co-payment medicines into the dataset.
The capture of under co-payment medicines has implications for analyses using unit-level data. If the medicine of interest is under co-payment for all or part of the study period, restriction of the study population to concessional beneficiaries or DVA clients can ensure more complete ascertainment of medicine use. This is because the concessional co-payment threshold is lower than the cost of any medicine on the PBS.
This method has been widely used in Australian studies using unit-level PBS data: of 113 such studies published between 1987 and 2013, 75 % employed a study population comprised of concessional beneficiaries or veterans [1]. As the concessional status of a patient can change over time, inclusion should ideally be further restricted to individuals for whom all dispensed medicines are provided at the concessional rate during the study period. Alternatively, under co-payment use of medicines requiring written or telephone authority approval can be tracked using the Authority Approvals database (such as use of dexamphetamine and methylphenidate in attention deficit hyperactivity disorder).
Changes to medicine coding: ATC and PBS item code changes
Both PBS item and ATC codes are subject to change and this requires consideration when examining utilisation trends. For example, the antidepressant venlafaxine was listed under the ATC code N06AE06 until 1995, when its code changed to N06AA22. The code was further changed in 1999 to N06AX16 [33]. Similarly, a particular formulation of tobramycin, a systemic antibiotic, had the PBS item code 1356J until December 2005, when the code changed to 8872Y [34]. While the original item code still exists, it no longer refers to this formulation.
It is therefore important to ensure that all relevant historical and current ATC and/or PBS item codes are included in the analysis to avoid errors in utilisation estimates. Defining medicines of interest by ATC codes rather than item codes can help to overcome this problem, as ATC codes capture all current and historical PBS item codes, are less prone to change, and any historical changes in ATC code can be easily determined from the World Health Organisation Collaborating Centre for Drug Statistics Methodology (WHOCC) website [33]. One caveat is that there are occasional differences between the WHO-defined ATC codes and the ATC codes present in the PBS dataset. For example, lithium carbonate is classified as an antipsychotic by WHO (code N05AN01) but as an antidepressant by the PBS (code N06AX) [35]. Changes in PBS item codes are more difficult to track, but are recorded from 2003 in PBS monthly reports [36].
It is also worth noting that a medicine may concurrently have more than one ATC code (when the medicine has multiple indications affecting different body systems) or item code [according to the indication, strength, or prescriber (i.e. medical practitioner, nurse practitioner, or dentist) of a particular formulation]. Therefore, in some cases, the item code may provide a proxy of the indication or reason for prescribing. For example, the antineoplastic bevacizumab is available under item code 10114H for epithelial ovarian, fallopian tube or primary peritoneal cancer, but under 4400N for colorectal cancer. However, other drugs have multiple indications for prescribing combined under a single code (e.g. one item code for the antidepressant paroxetine is used for major depressive disorder, obsessive compulsive disorder, and panic disorder). Additionally, the validity of item codes for inferring patient diagnosis is uncertain. Depending on the research question, the researcher may choose to include all codes or only those referring to a certain indication or prescriber type.
Policy changes
Changes in the medicine reimbursement process can impact data capture and estimates of utilisation. For example, the introduction of the Public Hospital Pharmaceutical Reforms from 2001 increased access to PBS-subsidised medicines by allowing participating public hospitals to provide PBS medicines to patients at discharge and outpatients. These Reforms are governed by individual agreements between each state and territory and the Australian Government. Agreements were initially established in Victoria (September 2001), followed by Queensland (August 2002), Western Australia (2002), Northern Territory (January 2007), South Australia (August 2008) and Tasmania (December 2010), with reforms implemented gradually across each state [37]. New South Wales and the Australian Capital Territory do not participate in the Reforms. Researchers conducting state-by-state comparisons should consider whether the introduction of the Reforms may influence utilisation estimates for the medicines of interest. A variable indicating the type of dispensing pharmacy (hospital, community) can be provided with PBS data by request.
In late 2011 the Pharmaceutical Reforms were varied to enable the introduction of a new scheme governing the subsidy of chemotherapeutic agents, the Revised Arrangements for the Efficient Funding of Chemotherapy measure [37]. These arrangements came into effect in December 2011 for private hospitals and community pharmacies, and April 2012 for public hospitals [38]. S100 medicines dispensed in public hospitals have traditionally been processed in bulk by DHS at the end of each month, and therefore were not recorded as individual-level dispensing claims or included in the dataset. However, the Efficient Funding of Chemotherapy resulted in a shift from bulk to unit-level processing of s100 chemotherapeutic items and increased capture of these medicines in the dataset. As such, an increase in utilisation of s100 medicines dispensed through public hospitals can be observed following the introduction of the scheme. These examples highlight the need to question significant and unexpected changes in the data to determine whether they represent a true change in utilisation or an artefact of the way the data are ascertained.
Measures of utilisation
Medicine use can be quantified in a variety of ways in the PBS dataset, including by number of dispensings or costs. The strength of the medicine and quantity supplied can also be used to calculate DDD/1000 pop/day, a widely used measure of utilisation allowing for standardisation of drug use across countries and different forms of the drug. The DDD metric, established by the WHOCC, is based on the estimated mean daily dose of the drug when used for its main indication in adults [6]. DDD/1000 pop/day can be calculated for both plain products (which contain only one active ingredient) and combination products (with more than one active ingredient). The DUSC calculates DDD/1000 pop/day for combination products by counting the DDD for each constituent separately [7]; this method contrasts with that used by the WHOCC, who assign DDDs by counting the entire combination as one daily dose [39]. This methodological difference should be considered when making international comparisons of utilisation using DDDs.
Different measures of utilisation may yield differing results, and researchers must determine which measure(s) is most appropriate for their research question and dataset, considering the strengths and limitations of the chosen measure. Prescription-based measures such as ‘number of dispensings’ do not standardise utilisation across populations, or across different medicine strengths and pack sizes. While DDD/1000 pop/day is useful for standardising population-based measurements, the DDD on which it is based does not necessarily accurately reflect the dose recommended or prescribed. DDD is also limited for quantifying medicines use in children and the elderly, for whom different doses may be used. As with ATC codes, DDDs can change over time; a list of changes can be accessed from WHOCC [40].
Figure 5 demonstrates the differing results obtained when measuring utilisation by number of prescriptions dispensed, DDD/1000 pop/day, or medicine cost to government for the antipsychotic quetiapine; the antidepressant desvenlafaxine; the benzodiazepine diazepam; and the stimulant methylphenidate. Each of these measures have certain strengths and weaknesses. For example, the DDD of quetiapine is 400 mg, which is the average dose for an adult patient diagnosed with psychosis. However, a recent analysis by DUSC revealed that 23 % of patients taking quetiapine are using only the 25 mg strength of the drug, likely for the treatment of non-psychotic disorders such as anxiety and insomnia [41]; DDD/1000 pop/day would therefore likely under-estimate true use. In addition, quetiapine is one of the most costly medicines to the Australian Government [42], and analyses relying solely on cost would over-estimate its utilisation. Assessment of utilisation by number of prescriptions dispensed can also be problematic when comparing between different drugs, strengths and pack sizes. Quetiapine, for example, is usually dispensed in packs of 60 tablets, while desvenlafaxine has a pack size of 28. Use of different measures of utilisation may also impact trends in medicine use. For example, the increase in quetiapine use between 2006 and 2011 is more pronounced when measured by number of prescriptions (232 % increase) than by cost (154 % increase) or DDD/1000 pop/day (123 % increase).