As expected, due to their greater capacity to move between respective payment classifications, private patients were found to be more likely to switch payment classifications in their next admission than public patients, irrespective of the length of time between the two episodes. There are a number of possible explanations for this including putative structural and cognitive reasons for the observed behaviour. Structurally, the average patient who begins with a public classification is likely to be of lesser socioeconomic means than the average patient who begins with a private insured classification. The option of the former patient to use PHI at the next hospitalisation will be dependant upon them taking out, or at the very least maintaining (if they had private cover at the initial episode but did not use it) private cover in the meantime. The same pre-requisite does not exist for an initially privately insured patient accessing a public classification on the next occasion. Also patients with private insurance experiencing trauma or an acute disease event may, in some circumstances, be admitted in an emergency as a public patient, thus to some extent disallowing the patient from exerting their preference. Cognitive explanations include the possibility that PHI may not be as entrenched culturally as Medicare (the public system) in this population and, as a consequence, use of PHI may be more dependant upon marketed value propositions than use of Medicare. In other words, PHI may be perceived by the populace as a market good, whereas the public system (Medicare) may be perceived as a fundamental right.
We also found that the degree of switching from PHI towards the public system was inversly proportional to the length of time between episodes. Possibly, healthier individuals were more likely to switch to the public system than sicker individuals, assuming that a relatively short duration of intra-couplet interval can be taken as an indicator of increased morbidity. This phenomenon is consistent with reports that the decline in the proportion of the eligible population holding PHI since the introduction of Medicare in 1984 has been largely attributed to younger and healthier individuals dropping out [11
]. This finding indicated the presence of a substantial cross-subsidisation between low risk and high risk individuals further exacerbating the adverse selection price 'death spiral' of PHI [11
]. Improving the risk profile of PHI holders has been a major focus of recent federal government policy, with the introduction of a lifetime community rating in 2000 in an effort to encourage younger, healthier individuals to take out and remain in PHI funds. Our finding that in all age groups the overall rate of switching away from PHI was slightly higher in the period 1981 – 1990 compared with the rate in the period 1991 – 2001, suggests that these policies have had an effect on behaviour consistent with the government's intention.
An alternative, but less likely explanation for our findings, may be systematic differences in the cognitive decision making process between long intra-couplet intervals compared with between shorter intra-couplet intervals. For example, individual historical preferences may play a role in short interval switching but may not be important over longer intervals, where decisions may be made in isolation. While it is important to consider this alternative, we feel that if the changes observed in switches away from PHI were largely due to differences in cognitive decision making, one would expect to see a similar phenomenon in switching away from a public classification. Such a phenomena was not observed in our data.
In all age groups, our analysis indicated that the overall rate of switching from the private payment classification was slightly higher in the period 1981 – 1990 compared with the rate in the period 1991 – 2001, suggesting that the policies had an effect on behaviour consistent with the government's intention.
Assumptions and limitations of the approach
This study made use of the WA Data Linkage Project which is unique in Australia and is one of only a small number of population-based record linkage systems in the world. The use of administrative data has strengths and weaknesses. For example data can be inaccurate due to recording or coding errors [17
] or linkage errors. For this study individual patient records were linked by probabilistic matching, using an automated computer algorithm based on the probability of two records being from different people having the same identifier and two records from the same person having different identifiers. The probabilities were then aggregated into a score and checked against a threshold to determine if a match was made. This technique typically has been found to have a true positive predictive value of 95–99% and a negative predictive value of 98–99% [18
]. Extensive validation of the quality of the performance of matching has been undertaken on the WA Record Linkage Project using sampling techniques and the proportions of mismatches and missed matches found were in the order of approximately 0.11% [18
Linked data have the advantage of supporting a large and diverse research programme at relatively low cost, once the infrastructure is in place. They have the capacity to provide a population-based view of events experienced longitudinally by individuals across all institutions [17
]. Given the objectives of this study, the latter point makes the use of linked data particularly appropriate.
The approach we have taken in order to analyse the use of PHI and Medicare by the population of Western Australia is unique in at least two respects. Firstly the study was conducted at the population level, due to the use of hospital morbidity data. Therefore, the "reference population" was not merely an abstract concept as in conventional quantitative research, but an operationalised descriptor of the study sample. Secondly, our 'couplet methodology' has not been reported previously in the literature and represents a new method of analysing longitudinal data on use of health insurance. The couplet methodology has enabled patterns to be measured based the behaviour of individuals rather than average shifts in private – public mix generated from unlinked episodes of care. However, we recognise that since the analysis was based on episodes of care, patient initiated switching as part of a hospital transfer could not be analysed. This limitation may have affected our conclusions about intra-couplet intervals.
This paper is dedicated to explaining the couplet technique for the first time and applying it to address an initial set of relatively descriptive questions (ie teasing out what is happening). The next stage of investigation is a more analytic analysis which recognises that a wide range of potential explanatory variables might predict the couplet-based phenomenon of switching (the why). For example hospitalisation rates in set periods before and after a couplet, stratified by different lengths of stay and/or different DRG weights. In addition, the admission types (emergency/elective) of the members of the couplet may also be important predictors. We feel that the couplet methodology described in this paper will enable significant inroads to be made into the investigation of such explanatory variables.