In this study we have examined the independent effects of socioeconomic, locational accessibility to services and private patient status on hospital utilisation rates in the last years of life in patients who died of the same underlying cause after adjustment for potential confounders.
Although the literature is replete with studies recognising that socioeconomic status has a marked effect on utilisation of health services, the majority of studies have been subject to criticism due to potential confounding by severity of illness or comorbidity. For example, in a study examining trends in patterns of hospitalisation of the New South Wales population, Walker et al [20] found that the poorest socioeconomic group had a 21% higher hospitalisation rate in 1996/1997 compared with that observed in the richest sub group, an association which was reversed for private patients. However, in that study no adjustment was made for severity of illness or comorbidity, nor was stratification by disease attempted. Thus it is difficult to determine if the differential utilisation was explained by structural differences such as variations in social support or by real differences in need, or both.
Other studies have looked at access to specialised services as a function of socioeconomic status. Coory et al [21] investigated the effects of socioeconomic status on utilisation of invasive coronary procedures in Queensland. The authors found, as has been demonstrated in numerous other studies, wide disparities in access. They concluded that free access to health care did not necessarily ensure equitable access. However, in that study there was no adjustment for severity of illness, although adjustment was made for comorbidity, for this reason the results reported by Coory et al [21] were unable to assess equity issues definitively, given that equity requires 'equal access for equal need'.
Studies that have assessed health outcome as a function of socioeconomic status have found an inverse relationship between socioeconomic status and mortality. Hall et al [22] examined the influence of social, economic and locational disadvantage on lung and breast cancer case fatality in Western Australia, finding that survival was poorer in patients treated in public hospitals, rural hospitals and relatively disadvantaged socioeconomic groups. The authors acknowledged that since no staging information was available, adjustment for severity of disease was not possible.
In this study, in addition for adjusting for demographic characteristics and comorbidity, we have at least partially adjusted for severity of illness by restricting the analyses by both the time to death and underlying cause so as to generate, as far as possible, a relatively homogenous group of patients with respect to need. Further, we have accounted for the possibility that patients from disadvantaged groups may have a shorter average duration of life-threatening illness due to more advanced disease at initial diagnosis, by undertaking separate analyses for the last, second and third year prior to death. Given that in the last year of life needs will most likely be the most homogenous, any differences between the utilisation rate in that period for a particular cause of death is more likely to be attributable to causes other than illness severity rather than differences observed in the second or third years prior to death. Therefore, comparison of utilisation trends across socio-demographic variables between look-back periods has enabled us to separate severity of illness effects from true socio-demographic effects.
Our results suggest that the independent effects of socioeconomic status differ markedly across diseases and that the consistency of effects between three and one-year observation periods before death also differs across diseases. For colorectal cancer we observed no significant socioeconomic effect in the last year of life; however, in the third year prior to death socioeconomic status had a statistically significant effect suggesting that disease severity may have confounded the effect of socioeconomic status observed earlier in the course of the disease. This would be consistent, for example, with a screening effect of higher rates of colonoscopy leading to higher and earlier hospital utilisation in more advantaged groups. For lung, breast cancer and ischaemic heart disease the reverse association was observed with respect to highly disadvantaged individuals. In the third year prior to death no statistically significant difference was observed for between highly disadvantaged and highly advantaged individuals, but in the last year of life a marked difference was observed, suggesting that socioeconomic status is a major driver of differential utilisation. Such differential hospital utilisation in terminal breast cancer patients might, for example, be due to a lower level of accessibility to hospice or home-based palliative care services.
With respect to disadvantaged, average and advantaged individuals who died of ischaemic heart disease we found that utilisation was associated with socioeconomic advantage regardless of time to death indicating that socioeconomic status was the major driver of differential utilisation. Severity of illness did not appear to confound this association because there was little difference in results across look-back periods. The effect of socioeconomic status in cerebrovascular disease was difficult to interpret since the relationship was inconsistent across disadvantage categories.
Distance to health services has long been recognised as an important factor in access to care with numerous studies describing an inverse relationship between distance and utilisation across many diseases. Jong et al [8] showed that people living in areas of Australia with limited access to services had poorer health than people living in metropolitan areas. Jones et al [5] examined the relationship between asthma mortality and access to health services, finding that asthma mortality increased with travel time to hospital. An inverse relationship between geographic proximity to services and mortality has also been demonstrated for ischaemic heart disease in Australia. Sexton et al [6] found that populations living outside capital cities had higher death rates than those living in capital cities in Australia. In addition, Hall et al [12] found that locational accessibility had a significant effect on patterns of surgical care in people with colorectal cancer and the mortality of patients with breast and lung cancer [22]. As has been the case with socioeconomic status, a major limitation of all of these studies was the lack of adjustment for severity of illness and often comorbidity as well.
Our results suggest that locational accessibility to services is generally not an independent predictor of utilisation during the years leading up to death, with the exception of in those dying from ischaemic heart disease, where we found that the incidence rate ratios were significantly different in the last year of life. These results are somewhat concordant with a study of colorectal and lung cancer which suggested that disease stage accounted for differential utilisation effects rather than locational accessibility per se [23].
It has been suggested that the existence of optional private health insurance in addition to the universal publicly funded Medicare system in Australia has led to a two tier system, for the rich and poor, with implications for treatment patterns and survival in economically disadvantaged groups [24]. However, proponents of the dual system argue that private health insurance is able to fund extra demands in a regulated way by providing a mechanism to ensure choice in service provision and reducing pressures on the public system [25]. Our results indicate that admission as a private patient (adjusted for hospital type, locational accessibility and socioeconomic status) did not consistently effect utilisation across all causes of death. However, we found that payment classification had a statistically significant independent effect when colorectal cancer or cerebrovascular disease was the underlying cause of death and that in both cases the effect was most marked in the last year of life when we found that private patients were more likely to be hospitalised.
Strengths and limitations of this study
The use of probabilistically matched administrative health data made available a comprehensive patient-based data set, as opposed to a separations-based data set, enabling access to historical information on health service utilisation for the total population of WA. The study was therefore population-based rather than sample-based; allowing the health experience of all individuals who died between 01/01/1997 and 31/12/2000 to be incorporated into the modelling process. We feel this is likely to lead to greater accuracy and improved external validity compared with studies using institutional or response-based sampling from a base population [26, 27].
Significant information error in our results is highly unlikely as the data used were population-based and classification regimes were applied consistently throughout the data set. Exhaustive validation research on the WA Data Linkage System [15] has shown that missing demographic data items are very uncommon (<1%). One limitation of our study is that we used time to death as a proxy for severity of illness, which was in itself a proxy for level of need. The measurement of severity of illness is a major problem when undertaking studies of this type as it cannot be readily determined from administrative data sets. Our use of time to death defined a relatively homogenous group of patients and this, combined with stratification by underlying cause of death, may have reduced the extent of confounding by stage of disease. However, the level of control of confounding was likely to have been incomplete.