In principle the best way to obtain estimates of per capita health expenditure by income is to collect individual-level information on the usage and associated costs of medical services, income, Indigenous origin, age and gender. Unfortunately no such Australian data exist, and we are therefore forced to combine estimates of utilisation rates of health services from the 1995 NHS with the average costs of medical services estimated from a variety of administrative and survey data sources.

The method used involves merging onto the 1995 NHS, at the level of each individual record, estimates of the average cost of medical services. This produces, for the individuals in the NHS survey, estimates of the expenditure associated with the medical services they used in the two weeks prior to the survey. Using the estimated expenditure and information on each respondent’s equivalent income, it is straightforward to estimate average per capita health expenditure for Indigenous and non-Indigenous Australians.

There are two sources of sampling errors associated with the estimates of per capita health expenditure. First, there is sampling error associated with the estimates of utilisation rates of health services from the 1995 NHS. Second, there are errors in the estimates of the average cost per medical service. The standard errors of the estimates of the utilisation rates are calculated using the ‘jackknife’ method.^[10] Unfortunately, no information is available on the standard errors of the estimates of the average cost per medical service. Given that such costs are estimated from data which relate to a substantial proportion of the total population, it is reasonable to assume that the sampling variability from this source is extremely small. Notwithstanding, the standard errors of the estimates of per capita health expenditure presented in this paper provide a lower bound estimate.

The method of estimation of per capita health expenditure by income differs in a number of respects from that used in a previous analysis of this relationship. Deeble et al. (1998) estimated total and government health expenditure by age and gender, and then allocated this expenditure across equivalent income quintiles using differences in average rates of utilisation of health services for income groups.^[11] The analysis by income group was only done for the total Australian population. Overall Indigenous public health expenditure was then compared with the estimates for the total Australian population for the respective income groups. The argument was made that Indigenous people are predominantly in the lowest income group and that per capita health expenditure should be compared with government health expenditure on the lowest income group for the total Australian population.

As outlined above, the estimates presented in this report are constructed by combining estimates of the average costs per medical service (estimated primarily from administrative data sources) and utilisation rates estimated from the 1995 NHS. The analysis is conducted at the level of the individual, and so no allocation of total health expenditure across income groups is required. The main advantage of the approach used in this report is that it allows standard errors on the estimates of health expenditure by income. As will become apparent when the results are presented, the standard errors are very high, and conclusions drawn about differences in health expenditure which ignore the standard errors may be grossly misleading.

The 1995 NHS contains information on 53 751 Australians of all ages and is representative of those living in all areas. There is information on the rates of utilisation of a variety of health services but no information on the costs of these services. The NHS also contains information on income and a range of demographic variables. The following health services are included in the estimates of per capita health expenditure by equivalent income:

The range of medical services included in the estimates of expenditure is determined by the questions asked in the 1995 NHS. More information on the definition of each medical service is presented in Appendix A.

As indicated above, the analysis of health expenditure by income uses four equivalence scales to derive separate measures of equivalent income:

Income quintiles for these four different measures of income were estimated from the 1995 NHS separately for the Indigenous and non-Indigenous components of the population. Each family’s income is ranked using the overall distribution of equivalent income in the 1995 NHS. Given that the number of Indigenous families in some of the higher quintiles is quite small, it is necessary to aggregate the top four quintiles to enhance the reliability of the estimates.^[12] The numbers of people in the respective quintiles are provided for each measure of equivalent income in Appendix B.

In the interests of transparency (i.e. to ensure that the results are replicable), Appendix C provides detailed breakdowns of all the costs per service for medical services used in this report. As discussed, these estimates were provided to CAEPR by AIHW and are derived from a variety of administrative data sources. Where possible, the estimates of cost per medical service are estimated according to Indigenous origin, gender and age group. The level of disaggregation in the estimates of cost per service varied according to what is feasible, given the administrative data available (see Table 3.1 for details of the level of disaggregation for each category of medical service). Where no disaggregation is possible, it is necessary to assume that the cost per service is identical across age groups, gender and Indigenous origin. The medical services for which disaggregated cost data are available are admitted patients, visits to general practitioners and specialists, and prescription medications.

It is important to estimate cost per medical service by as detailed a gender, age and Indigenous origin breakdown as possible due to differences in the average cost per service by demographic characteristics. The importance of this is illustrated by the differences in the estimated costs per service for admitted patients; these vary from $478 per day in hospital for Indigenous males aged 75 plus years to $900 per day in hospital for non-Indigenous females.

Note: Unfortunately, it was not possible to further disaggregate the out-of-hospital estimates for the Indigenous population because it is based on a relatively small sample size (see endnote 10).

 The medical services for which disaggregated cost data are available cover 78.7 per cent of all health expenditures included in this analysis. The inaccuracies introduced by the aggregated nature of the estimates of over-the-counter medications, non-admitted patients and other health professionals will be relatively minor. It is worth stressing that no information was provided on the standard errors associated with the estimates of cost per medical service. The standard errors, for at least some of the service types, are likely to be quite large. For example, the cost per visit to a GP for the Indigenous population is based upon information from 2000 Indigenous patient encounters. This means that the number of patient encounters in each of the gender and age groups is small for the Indigenous population.^[13]

As already noted, unlike the analysis for the Australian population as a whole presented in Deeble et al. (1998), it is not possible to separate expenditure into the private and public components. There are several reasons for this, all associated with the quality of information available from the 1995 NHS. First, due to changes in the questions about rates of medical service utilisation between the 1989 and 1995 NHS, it is not possible to estimate government health expenditure by equivalent income quintile. The main change is that the 1989 NHS asked about hospital utilisation in the previous twelve months, whereas in 1995 the NHS asked about hospital utilisation in the previous two weeks. This means that there are not enough reported visits to private hospitals to allow utilisation rates of private and public hospitals by equivalent income to be estimated.

Second, the proportion of expenditure on prescription medications that is privately funded versus the proportion publicly funded is determined by several factors. Prescription medications listed on the PBS receive a government subsidy.^[14] Prescription medications which are not listed on the schedule of PBS-approved drugs receive no subsidy, and therefore all costs are borne privately. Clearly, in order to estimate public versus private funding on prescription medications, it is crucial to separate medications according to those listed on the PBS schedule and those not listed on the PBS schedule. This is not possible using the 1995 NHS because it classifies medications according to their Anatomical Therapeutical Category, and this cannot be mapped onto PBS and non-PBS categorisation.

There are major advantages to the method used in this report to estimate per capita health expenditure by income. First, it allows standard errors to be calculated for the estimates. This is absolutely critical when interpreting the estimates of health expenditure per capita for the Indigenous population for whom the sample sizes are small. Second, it gives an accurate reflection of per capita expenditure for the sample used rather than applying the rates to aggregate data. This of course means that the estimated per capita expenditure will differ from the estimates of aggregate expenditure.