Impact of population ageing on the costs of hospitalisations for cardiovascular disease: a population-based data linkage study

All hospitalisations occurring in Western Australia pertaining to patients aged 18 years and older with a principal or co-diagnosis CVD code from 1 July 1993 to 30 June 2004 were extracted from the Hospital Morbidity Data System (HMDS) using the Western Australian Data Linkage System (WADLS) described elsewhere [18],[19]. The HMDC captures records of separations from all public, private acute care and psychiatric hospitals and private freestanding day hospital facilities in Western Australia [20]. The relevant CVD codes of hospitalisations were identified using the International Classification of Disease (ICD) codes version current at the time of separation from hospital and are detailed in the following sub section. Prior to data extraction, ethics approval for this study was obtained from the Human Research Ethics Committee at the University of Western Australia and Curtin University.

Common types of CVD in Australia were included in this study. Relevant ICD codes to identify hospitalisations with CVD were obtained from publications produced by the Australian Institute of Health and Welfare (AIHW [17] and L Nedkoff, TG Briffa, DB Preen, FM Sanfilippo, J Hung, SC Ridout, M Knuiman and M Hobbs [21]. Events were defined as CVD when they were recorded as either heart failure, or angina pectoris, or acute myocardial infarction, or unstable angina or stroke. All records were subsequently grouped into three subgroups, namely chronic conditions (ICD 9_CM 428, and 413, ICD 10-AM I50, and I20), acute coronary syndromes (ACS) (ICD 9_CM 410, and 411.1, ICD 10-AM I21, and I20.0) and stroke (ICD 9_CM 430-432 and 436, ICD 10-AM I60-I64).

The individual records of hospitalisation were aggregated into episodes of care, defined as a period of continuous hospitalisation. Inter-hospital transfers were periods of in-patient hospitalisation that overlapped with each other or where separation (discharge) and admission dates were on sequential days. Capture of transfers between hospitals is required in order to avoid over counting of the number of hospitalisation events. In our study periods of contiguous in-patient hospitalisation were classified as episodes of care. The episode of care was assigned to one of 17 age groups (18-19, 20-24 to 90-94 in five-year age groups and 95+ years) determined by the age in years at admission time of each individual. The first age group comprised only two years (18-19 years) since data extraction was restricted to hospitalisation records for individuals from 18 years and over.

The Australian Refined Diagnosis Related Group is an Australian admitted patient classification system, which provides a way of relating the number and type of patients treated in a hospital to the resources required by the hospital. Acute admitted patient episodes of care are categorised into groups with similar conditions and similar usage of hospital resources, using information in the hospital morbidity record such as the diagnoses, procedures and demographic characteristics of patient.

The cost of each episode of care was assigned based on AR-DRG code recorded in the HMDS record for each record of hospitalisation included within an episode of care (where there were inter-hospital transfers). The cost attributed to each AR-DRG was that reported for that AR-DRG in the National Hospital Cost Data Collection Report relevant to the year of each separation [22]. The cost was assigned inclusive for all hospitalisation events in database before adjusting for the inter-hospital transfers and therefore included multiple records in the case of inter-hospital transfer. Thus, the costs of the inter-hospital transfer episodes were a sum of any records which formed the episode of care. The hospitalisation costs included medications, investigation undertaken whole admitted, physicians, nursing staff, operating theatre, emergency department and “hotel” costs but not included rehabilitation costs. All dollar values were adjusted to 2012 price levels, using health price indices calculated from the Health and Welfare expenditure series of the AIHW [23],[24].

The total number of episodes (E) and cost per episode (CPE) were calculated in five-year age groups by sex and diagnosis subgroup at two points of time, 1993/94 and 2003/04. Number of episodes per capita (NEC) for gender x and age-specific group j was calculated separately for chronic conditions, ACS, stroke and overall CVD:

N E C x j = Ex , j P x , j, where Ex,j is the number episodes for gender x and age-specific group j; Px,j is the WA population in gender x and age group j for that year (obtained from the Australian Bureau of Statistic Time Series Workbook Table 55 (ABS 3101.0)) as the denominator [25]. The rates were calculated per 100,000 population.

The cost per episode (CPE) for gender x and age specific group j was calculated for chronic conditions, ACS, stroke and overall CVD as following: C P E = C x , j Ex , j, where Cx,j is total cost in gender x and age group j; Ex,j is the number of episodes in gender x and age group j.

In our study, the impact of population ageing on the hospitalisation costs for CVD was measured using the method that was adapted from the principles of the component decomposition method used in studies in Australia and Korea [12],[13]. The previous studies, the change in healthcare expenditure was attributable to the impact of ageing, the growth of population over time, change in proportion of people using healthcare services and change in average cost per episode [12],[13]. Similar to the studies, the change in total costs of hospitalisation for CVD over a period of time were attributable to following components: population growth, ageing of the population, the increase in total number of episodes of hospitalisations and the increase in average cost per episode. Steps to calculate proportion of contribution of each component to the change in total costs of hospitalisation for CVD between 1993/94 and 2004/05 were described in detail below.

The total costs of hospitalisations for disease d in each year was decomposed by the following equation:

From equation (1), the change in the WA population over the study period can be decomposed into a change in the total population and a change in the age distribution of population. Similarly, the change in the number of hospital episodes per capita can be decomposed into a change in the total number of episodes and a change in the age distribution of episodes. Thus, a difference in the total hospitalisation costs between 1993/94 and 2003/04 (DHC) was attributable to the change in total population (CPOP), the change in the age distribution of population (CDEM1), the change in the total number of episodes (CTNE), the change in the age distribution of episodes (CDEM2) and a change in cost per episode (CCPE).

A change in total hospitalisation costs for a disease d for the time period t (1993/93 and 2003/04) was

The total of changes attributable to the change in the age structure of the population (CDEM1 _d ) and to the distribution of hospitalisation between age groups CDEM2 _d reflect the impact of ageing on total costs, thus impact of ageing = CDEM1 _d  + CDEM2 _d . Through this analysis the impact of the ageing of the population on the cost of hospitalisation for CVD was isolated from the respective costs of the increase in the general population and the total number of hospitalisations.

In order to calculate the proportion of contribution of each component, an age-specific cost profile, represents the per capita healthcare cost of a specific age group, was applied [26]. By holding age-specific cost profiles constant, this approach has an assumption that the impact of other variables with potential influence on healthcare cost does not change [26]. Although the assumption is unrealistic, it is useful for the purpose of isolating the impact of demographic changes (ageing population). The method has been used in many studies [12],[26],[27].

In our study, to capture the proportion of each component contributing to the difference in total cost of hospitalisation in the period, each component in initial year (1993/94) was moved to actual value in 2003/04 in sequence. As each component in the assumption was changed, the difference in total cost of hospitalisation between initial year (1993/94) and the assumed final year (2003/04) was attributable to whichever component was affected by the change in assumption. Each assumption using the actual value in 2003/04, the sum of the change in individual components is equal to the change in total cost between the two points of time. Details of the calculations are presented in Additional file 1.

While it is common practice that sensitivity analysis is conducted on health economic analyses to evaluate the effect of various assumptions and the averaging of values have on the magnitude of the final outcome, in this study there were no assumptions made since the data were not a sample and the number of hospitalisations and costs were those recorded directly in the data set. Thus there were no values that could be varied for use in a sensitivity analysis. Descriptive analysis was conducted using SPSS version 18. Calculations for the decomposition analysis were done in Microsoft Excel 2010.

The study used a validated and well respected source of data and linkage mechanism which has been shown to produce accurate (99% accuracy) individual level linked data [18]. The data included all hospitalisation occurring in the state (both private public and private) providing highly generalizable population level information. Thus, bias due to selection or sampling error is not present in this study, leading to an increase in study power due to the comprehensive population coverage. For the decomposition of the increase in healthcare costs, although major components contributing to the increase in total costs of hospitalisation were taken into account, the components could have been be broken down into more detail. Average cost per hospitalisation, for example, can be separated into number and type of services used within each episode and cost per service. However, the available costing data in the Hospital Morbidity Data System were not sufficient to allow further decomposition. Since the focus of the study was to evaluate the contribution of the ageing component to CVD hospitalisation costs, this limitation did not affect our ability to fulfil this aim. Another limitation is that the time period for the analysis was 1993/94 to 2003/04. Since 2004, many changes in medical management and advances in diagnosis have taken place. Nevertheless, the findings of this study on the impact of ageing on the costs of hospitalisation for CVD and subgroups provides a useful basis for projecting corresponding costs in future time periods. Despite its limitations, the study has shown the extent to which ageing has impacted on costs of hospitalisation for CVD.