International comparison of spending and utilization at the end of life for hip fracture patients

Abstract Objective To identify and explore differences in spending and utilization of key health services at the end of life among hip fracture patients across seven developed countries. Data Sources Individual‐level claims data from the inpatient and outpatient health care sectors compiled by the International Collaborative on Costs, Outcomes, and Needs in Care (ICCONIC). Study Design We retrospectively analyzed utilization and spending from acute hospital care, emergency department, outpatient primary care and specialty physician visits, and outpatient drugs. Patterns of spending and utilization were compared in the last 30, 90, and 180 days across Australia, Canada, England, Germany, New Zealand, Spain, and the United States. We employed linear regression models to measure age‐ and sex‐specific effects within and across countries. In addition, we analyzed hospital‐centricity, that is, the days spent in hospital and site of death. Data Collection/Extraction Methods We identified patients who sustained a hip fracture in 2016 and died within 12 months from date of admission. Principal Findings Resource use, costs, and the proportion of deaths in hospital showed large variability being high in England and Spain, while low in New Zealand. Days in hospital significantly decreased with increasing age in Canada, Germany, Spain, and the United States. Hospital spending near date of death was significantly lower for women in Canada, Germany, and the United States. The age gradient and the sex effect were less pronounced in utilization and spending of emergency care, outpatient care, and drugs. Conclusions Across seven countries, we find important variations in end‐of‐life care for patients who sustained a hip fracture, with some differences explained by sex and age. Our work sheds important insights that may help ongoing health policy discussions on equity, efficiency, and reimbursement in health care systems.


| INTRODUCTION
Health policy makers have a strong interest in developing a better understanding of how they can improve end-of-life care. [1][2][3] Prior studies have found that in the months leading to death, the need and the costs of care increase substantially, and the bulk of these expenditures come from high-acuity, high-cost individuals, such as those with persistent chronic conditions. [4][5][6] In addition, there are serious concerns that the quality of care at the end of life often falls short of expectations. [7][8][9] As more people reach old age with chronic and disabling conditions, improving the quality and efficiency of care at the end of life will continue to grow as a priority policy issue. 10 As such, end-of-life care has become an important dimension of how we measure health system performance. 11 International comparisons on end-of-life care may yield important insights into how policy makers could improve the efficiency and the quality of care. Such research is vital in setting performance benchmarks and establishing best practice models from a system-wide and policy perspective. However, to date, comparisons at the end of life across countries are quite limited, especially when it comes to comparing robust data across more than two countries and across different health care sectors.
In recent years, there have been significant advances in data infrastructure across many countries that enable international comparative research. This development is particularly true for patientlevel data, which is necessary to examine potential differences in endof-life care. Much of the relevant data are routinely collected through administrative datasets, which are increasingly accessible for research and quality improvement purposes. 12 For example, claims data from health care payers, such as health insurers or national health services, provide a solid starting point for international comparisons as shown in several international projects such as Health Basket 13 and EuroDRG. 14 In this study, as part of the International Collaborative on Costs, Outcomes, and Needs in Care (ICCONIC) project, which is a research collaborative across a set of high-income countries, we sought to evaluate differences in treatment at the end of life among frail, older adults who sustained a hip fracture across seven countries as follows: Australia, Canada, England, Germany, New Zealand, Spain, and the United States. Specifically, we sought to examine differences in utilization and spending across key health care services, including hospital care, emergency care, primary care and outpatient specialty care, and pharmaceuticals. Using a framework analyzing hip fracture patients, our goal was to provide insight into how health systems can optimally address the needs of the frail decedents by effectively accounting for resource constraints.

| SUMMARY OF PREVIOUS INTERNATIONAL STUDIES ON END-OF-LIFE CARE
There is an extensive literature on the health care costs associated with end-of-life care. Numerous studies have shown that health care costs increase manifold in the time leading up to death. [15][16][17] Riley and Lubitz, 18 for example, find that although decedents account for only 5% of the US Medicare population in any given year, expenditure on this group explains more than 25% of total Medicare expenditure. As has been shown, the bulk of these expenditures come from high-need highcost individuals, such as those with persistent chronic conditions. 19 The high costs at the end of life have led the policy makers to question if health systems are obtaining value for money in end-oflife care-particularly when considering that quality of care remains far from optimal. 20-23 At the same time, there is growing evidence that not all patients at the end of life face the same cost trajectory. [24][25][26] These studies show that there is considerable heterogeneity in the pattern in health care use and associated costs at the of end of life among different patients. Some patients face very high and persistent costs over an extended period of time, and others face a sudden decline in health status and associated rise in health care costs over a very short period of time. 27 As a result, there are widening calls to develop a greater understanding the drivers of end-of-life costs.
Another area of importance to health policy makers is awareness of gender and age disparities, including in end-of-life care. While literature found gender disparities in end-of-life spending and utilization, the direction of the disparity is not always consistent. [28][29][30] Some studies show that women receive less aggressive treatment than men receive when it comes to cancer or care in intensive care units. One potential reason is that women are more likely to have a do-notresuscitate order than men. [31][32][33] In addition, other work has found that end-of-life spending at least in the United States declines with age, indicating declining treatment intensity. 28 Thus, while there is some evidence in the area of cancer care, the evidence for frail elders with a hip fracture, while similarly important, is less established.
Further, cultural factors can influence utilization of health services. In some countries, there is variation in the proportion of older adults who live alone versus live with other family members. In the United States, evidence suggests that older adults are more likely to live alone than other countries and, therefore, may have a limited support system to care for themselves safely at home (and thus end up in a skilled nursing facility). 34 We therefore aim to expand on this literature and, importantly, show some consistency in patterns across countries by these important demographics.
Health policy makers who shape the health system have a strong interest in understanding how different countries provide end-of-life care, including current work at the Organisation for Economic Co-operation and Development (OECD). The international evidence on end-of-life care is limited-especially when it comes to comparing robust data across more than two countries. There are two notable exceptions. The first is French et al. 5

| Patient selection
We followed a two-step approach to identify a comparable set of frail elders who received treatment for a hip fracture. Hip fracture has been commonly used as a reliable marker of frailty among older adults, 38 and it accounts for the majority of fractures related to fragility globally. 39 Hip fracture is also highly associated with physical and mental disability, high mortality, and increased costs, thus requiring considerable health care resources from different parts of the health system. [39][40][41][42] As hip fractures almost always require a hospital admission and usually require surgery, the majority will be recorded in hospital admissions data and can thus serve as a robust and reliable tracer condition to explore differences in resource use across health systems. 38 We first identified a sample of comparable patients by examining all patients who received a primary diagnosis of hip fracture (S72.0-2 according to the International Classification of Diseases version 10) in 2016 and obtained a total hip replacement, a partial hip replacement, or were treated with an osteosynthesis method such as screw, plate, pin osteosynthesis, or internal fixation (see Appendix 2). From this sample, we identified those who died within 365 days from the index hospitalization associated with the hip fracture.

| Spending and utilization measures
We followed a federated data extraction approach due to data protection reasons. Each country produced means of utilization and spending by sex and age (65-69, 70-74, 75-79, 80-84, 85-89, 90-94, and older than 95 years), from individual patient-level data.
These aggregated nonidentifiable data were then collected in a central database for the analysis. It is important to note that we used the perspective of the health care payer across all countries. In most countries, this is either directly by an insurance or sickness fund (Germany and the Netherlands) or directly from a national form of health insurance (United States with Medicare program, Canada, etc.). Therefore, our study does not capture full costs (as it does not account for the fixed costs of all structures within a health system).
It only captures were actually paid for the services, which across all countries, already included the fixed costs of the system. In order to compare spending reliably, we first applied the OECD Actual Individual Consumption Purchasing Power Parities (AIC PPPs) to the expenditure data. AIC PPPs, rather than gross domestic product-based purchasing power parities, are currently used by the OECD as the most reliable economy-wide conversion rates for health expenditure. 43 Across each country, we applied 2017 AIC PPPs to all expenditures using the following exchange rates as  Figure 1A). Utilization and spending were allocated proportionally to the observation periods in case of accruals (see Figure 1B).

| Data analysis
First, we described the number of decedents by country and sex.
From each country's total sample of elderly patients who have experienced a hip fracture, we calculate the proportion of people who have died within 365 days of the fracture. 44 We also calculated the proportion of decedents in hospital relative to those who are discharged after their hip fracture admission.
To analyze within-and between-country variation, we estimate utilization and spending y i as a function of country-fixed  effects to determine baseline utilization and baseline spending, respectively. In addition, we include the interaction of country and sex, the interaction of country and age group, which we coded ascendingly from 1 to 7 for the 65-70 years to the older than 95 group, and the interaction of country and days before death into the model to disentangle potential country-specific effects of sex, age, and time. We fitted the model with no intercept term and, therefore, the age group and country estimates refer to zero, that is, we needed no reference group. We define the following six dependent variables for utilization as follows: acute hospital admissions, days in hospital, emergency department visits, medical doctor specialist visits, primary care visits, and outpatient prescription drugs. In addition, we measure the following five dependent variables spending as follows: acute hospital stays, emergency department, specialists, and primary care visits, as well as for outpatient prescription drugs. We estimate for each dependent variable the following linear regression: where x i denotes a vector of country-fixed effects and ε is a normally distributed error term. We obtaine β indicating baseline utilization and spending, ϕ sex-specific utilization and spending, γ an age gradient, and τ time-dependent utilization and spending. We considered p < 0:05 as statistically significant throughout the whole paper. All estimations were using the statistical program R 4.0.3 and the integrated lm function for linear regression. 45 4 | RESULTS

| Descriptive statistics
A total of 16,482 decedents were observed across the seven countries. In relation to the main cohort, 37 between 23.0% and 31.6% of all frail elders died within 365 days after a hip fracture across countries. Across all countries, more women (10,588) than men (5894) died after the event in absolute terms, but the relative mortality of women was lower than that of men. Mortality ranged for women between

| Regression results
Regression results with country-specific baseline utilization and spending, effects of sex, age gradients, and time effects are presented    Table 3). an elderly care facility designed to provide specialized end-of-life support services. 52 Furthermore, it is also important to not only consider the site of death but also where the majority of end-of-life care took place prior to the death. 53 According to our data, the number of days spent in hospital within the last month is strongly correlated with the rate of decedents dying in the hospital, that is, the more days spent in hospital care, the higher the in-hospital mortality rate. Further support for specialized and community palliative care services may offer patients and their carers more choice in deciding the most appropriate site of death for them. 47 We also observed important differences in the treatment by In terms of hospital days, the situation is the opposite: in Germany, decedents spend 9.6 days, and in the United States, only 5.4 days in hospital during their last phase of life. In Australia, England, Germany, and Spain, the decedents are very much supported by palliative care teams that are based in the reference hospitals or are part of the primary care organization and therefore included in the inpatient and specialist costs. 55 Similarly, in New Zealand, end-of-life care is provided in the home, with the support of homecare, medical, and hospice services, in a hospice if there are comorbidities that require active care, or in an aged care facility, which is designed to provide specialized end-of-life support services. However, the cost of aged care accommodation is often serviced by the patient, subject to an asset assessment, with only limited costs coming from the public sector. These different organizational and reimbursement arrangements make a direct comparison difficult, and our study based on administrative data should be complemented by qualitative research to explore the variation in the organization of hospice care further.

| DISCUSSION
In most international comparisons, the United States typically has the highest health care costs. 56 In part, the variation in use and costs across countries in each health care sector (e.g., hospitals) may be driven by differences in the roles and functions of that sector. For example, rehabilitation in some countries may be a part of a hospital's routine functions-whereas in other countries, much of this care may be provided in specialized clinics or in the community. This implies that rehabilitation costs will be incurred in the hospital sector for some countries, but in other countries, it may be registered as an outpatient care. Such differences in the roles and functions may be important drivers of the variation seen between countries. The results reported in this article provide an accurate description of costs and use within each sector, but further work is needed to examine the reasons for that variation-which includes further exploration of the roles and functions of each sector as well as full data capture.
Finally, we observed that some countries spend less on treating women before death. Specifically, while men and women generally receive similar care in the outpatient setting, there appears to be some differences in the type of care provided in the acute care setting, with males receiving more expensive acute hospital end-of-life care across countries. There are some potential factors that might explain this finding. First, prior work has suggested that women may have a lower biological age than men and may therefore be less frail. 60 There is also some evidence that suggests men enter the last phase of life sicker than women. For example, in Germany, male decedents scored a higher Elixhauser mortality score 61

| Limitations
The study has limitations. First, we had to rely on a federated analysis approach due to data protection regulations. Across countries, there are differences in how data are structured and collected, which may yield differences in the variables examined in this study. However, we used a patient vignette with specific diagnostic codes that are commonly used that should limit potential biases and deviations across countries. Second, the federated approach did not allow us to perform the estimations with individual-level data, which lower the efficiency of our model; however, the coefficients should remain unbiased. The small number of data points may also lead to a technical overfitting of the regression. However, given that the data points are based on thousands of observations, it still allows for a generalizability of the conclusions. Third, the method applied ignores that spending may be frontloaded during hospital stays in case of accruals (as illustrated in Figure 1B). This leads to an overestimation of the actual hospital costs incurred, most notably in the last 30 days before death. This is a problem of almost all claims data analyses and fades the longer the obser-

DISCLAIMER
The opinions, results, and conclusions reported in this paper are those of the authors and are independent from funding sources. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.