Limitations of Health Insurance Commission (HIC) data for deriving prescribing indicators

We used two expert groups which sequentially and systematically reviewed each of 22 classes of drugs available on the Pharmaceutical Benefits Scheme (PBS), identifying issues of comparative efficacy, safety and cost that might be reflected in prescribing indicators. One group, convened specifically for this project, comprised general practitioners, specialist physicians, clinical pharmacologists and pharmacists from the Newcastle area and drug utilisation experts from Canberra. The second group, the Prescriber Education Reference Group (an expert group convened by the HIC to advise on a broad range of prescribing-related activities), reviewed the work of the first group, offering further expert input and critical comment. The indicators identified for testing reflected a range of issues, including high overall levels of use of individual drugs and classes of drugs, relative use of agents within and between classes of drugs (on the basis of safety and efficacy), and the use of a limited number of agents within a drug class. Additional prescribing indicators identified through a literature search were also included in this study3,11-13 (see Box 1).

The HIC data collection has been described elsewhere.14 Records are generated when the government contributes to the cost of a medical service provided under Medicare or a pharmaceutical product dispensed under the PBS. Almost all services provided by GPs generate a record in the Medicare services file.

For pharmaceuticals, there is a series of graded copayments, with a maximum patient contribution of $20.00 (in 1997) for general beneficiaries and $3.20 for concession card holders (mostly pensioners and other social security beneficiaries). When the patient pays the entire cost of the medication, there is no HIC record of the prescription.

Many commonly prescribed drugs (eg, NSAIDs, anti-infective agents, diuretics, β-blockers, benzodiazepines and tricyclic antidepressants) cost less than $20.00, so only prescriptions of these drugs to concession card holders are recorded by the HIC. During the study period, there was complete capture for more expensive drugs (including statins, proton pump inhibitors and H2-receptor antagonists). Between 1993 and 1997, the maximum patient copayment per item rose from $15.90 to $20.00, resulting in a number of drugs changing from complete to incomplete data capture.

Prescriptions recorded by the HIC cannot be attributed to individual patients, so there is no information on the age or sex of the patient, co-prescribed medications, or the indication for treatment.

Analyses were restricted to GPs providing ≥ 1500 Medicare services per year. This workload stratification has been used in other studies,15 and excludes GPs whose prescribing profiles are likely to be unreliable because of small numbers of prescriptions. Prescribing and Medicare services data from the HIC for the years 1993 to 1997 were used. Analyses were based on original prescriptions, as these represent opportunities for prescribing decisions by GPs (ie, to start, stop or continue treatment), and because the identity of the prescriber was poorly recorded for repeat prescriptions (authorised on the prescription form at the same time as the prescribing decision) in HIC datasets in the earlier years of analysis.

Drug-prescribing records for individual GPs were extracted by the HIC and summarised. Records were identified only by a recoded provider number to prevent identification of individual GPs. Prescribing rates were defined as the numbers of original prescriptions for a drug or drug class dispensed per 100 Medicare services provided by the GP.6 Ratio and proportional indicators were used to estimate the relative use of agents within a drug class, or the relative use of different classes of drugs by the GP.

Temporal trends in the prescribing indicators were compared with national trends in estimated prescription numbers obtained from Australian Statistics on Medicines (ASM).16 ASM data comprise prescription totals derived from HIC data files and an estimate of use of "under-copayment" drugs and drugs purchased with a private prescription. These estimates are derived from a statistically representative sample of Australian pharmacies. The two datasets are not directly comparable: the prescribing indicators were calculated for individual GPs and were derived from HIC records of original prescriptions, whereas data in ASM are aggregated across all practitioners (GPs and specialists) by drug, and include original and repeat prescriptions. ASM utilisation data are population adjusted and expressed as numbers of Defined Daily Doses (DDDs)/1000 inhabitants/day.17 Despite their differences, the trends in the two datasets should be similar. PBS prescription packs generally provide one month of treatment (or a standard course of treatment), so numbers of prescriptions should relate to DDD estimates.

The individual GP was the unit of analysis. Descriptive statistics were used to summarise prescribing data. Means, standard deviations, medians and ranges (defined as the 5th to 95th percentiles) were obtained using the SAS procedure Proc Univariate.18

The expert groups identified 75 prescribing indicators worthy of further analysis (the full list is available from the authors). The statistical properties of these (and other indicators identified in the literature) were examined using 1993–1997 data for approximately 14 000 GPs providing ≥ 1500 Medicare services per year (ranging from 13 206 in 1993 to 15 425 in 1997). Four indicators reflected the use of personal formularies (based on the use of a limited number of agents within a drug class) and have been reported elsewhere.19 Of the remainder, 43 prescribing indicators were identified which were thought to reflect the quality of prescribing and to be informative for GPs and regional or national authorities (Box 2).

Thirty-one indicators were prescribing rates. Sixteen applied to drugs or groups of drugs where high prescribing rates were believed to be suboptimal. These related to adverse effects: antibiotics (resistance);20,21 flucloxacillin and amoxycillin–clavulanic acid (hepatotoxicity);22 prochlorperazine (drug-induced parkinsonism and tardive dyskinesia);23 non-steroidal anti-inflammatory drugs (NSAIDs) (gastrointestinal and cardiac effects);24-26 benzodiazepines (dependence);27 or where the benefits of therapy may be outweighed by the risks of serious adverse effects (eg, cisapride28). Two prescribing rate indicators involved drugs where a relatively high rate of prescribing could be interpreted as desirable in the absence of additional clinical information (phenoxymethylpenicillin and trimethoprim). There were various reasons for the other 13 indicators. For example, some were drugs of interest because they were comparatively new and expensive, others had been rapidly taken up into prescribing and there was evidence of more cost-effective choices, and single-drug products were generally preferred to combination drugs.

Twelve indicators were ratios, reflecting relative use within (or between) classes of drugs. Concerns related to adverse effects: long-acting sulfonylureas may be associated with hypoglycaemic episodes, particularly in the elderly,29 and NSAIDs with low to medium risk of gastrointestinal adverse effects are preferred over high-risk agents24,25 (the study was conducted before the introduction of COX-2 inhibitors). Others were based on a lack of evidence of additional benefits or clinical superiority to justify the increased cost of therapy (indapamide versus cheaper thiazide diuretics;29 nebuliser therapy over other metered-dose inhalers with spacer devices30).

There was substantial variability between prescribers. This is illustrated in Box 3 for 17 indicators, where there were fivefold to more than 300-fold differences between GPs' prescribing. Generally, the variation in ratio indicators was greater than that seen for prescribing rates.

Data capture was complete for six indicators, consistently incomplete for 20 indicators and variable for 17 indicators (Box 2).

For drugs with incomplete capture (NSAIDs, benzodiazepines), there were fivefold to sevenfold differences in median prescribing rates across the five strata defined by the proportion of concession card holders in the practice (Box 6). In comparison, differences were twofold to threefold for drugs with complete capture (statins, newer antidepressants).

Ratio indicators for drugs with incomplete capture were less affected (Box 6). Estimates for the ratios of high-risk to low- + medium-risk NSAIDs and long-acting to short-acting sulfonylureas varied only slightly across the GP strata (Box 6). This was also true for estimates of the ratio of indapamide to thiazides in 1997. However, there was more variability across the strata in estimates for 1996, when there was complete capture of indapamide prescriptions and incomplete capture for thiazide diuretics (Box 6).