1 INTRODUCTION

International comparisons of patient trajectories across health systems can be a useful tool to help national policymakers understand whether countries are achieving comparable outcomes at similar costs for their populations. To date, most international efforts have largely focused on understanding variation across individual conditions or episodes of care in the inpatient setting for general populations.1-6 Other work focused on evaluating end-of-life care in people with cancer and revealed considerable variation in the use of intensive and hospital-centric care across high-income countries.

However, the lack of available and similarly structured patient-level information for specific types of high-need patients across the entire care trajectory limits the potential to identify improvements to be made across the health system. For these patients in particular, it is important for policymakers to understand how care is distributed across settings, such as primary care, outpatient specialty care, and even long-term care, and how use in one setting may influence utilization in another. Understanding which health systems are more effective at managing specific types of HNHC populations could offer key insights to address rising costs, waste, and inequities in the system, as well as improve patient outcomes.

In order to address this challenge, the International Collaborative on Costs, Outcomes and Needs in Care (ICCONIC) was formed in 2018. In this article, we put forward a methodological framework to enable the cross-country comparison of resource use and outcomes for specific types of HNHC patients across the entire patient pathway. Our methodology builds upon previous international comparisons work and utilizes a clinical vignette approach,3, 4, 6, 7 which allows for the systematic collection and comparison of data across countries with different structures of patient-level datasets.

Specifically, we had three key objectives. First, we outline an approach for selecting two types of HNHC patient “personas,” drawing on a typology put forward by the National Academy of Medicine (NAM), to be used as tracers across different countries and health systems.8 Second, we propose a detailed clinical vignette to be used across countries to identify two types of HNHC personas using available and accessible patient-level datasets that allow for the comparison of utilization, spending, and patient outcomes across countries. Finally, we demonstrate the comparability of these two specific personas—(1) an older adult with frailty who sustains a hip fracture and subsequent hip replacement or osteosynthesis, and (2) an older person with complex multimorbidity, specifically a person hospitalized with heart failure and a comorbidity of diabetes—across the 11 countries in the ICCONIC collaborative. Importantly, this work provides the methodological framework used in an accompanying six original research manuscripts copublished in the Health Services Research “Special Issue on International Comparisons of High Need, High-Cost Patients.”9-14 These six original research articles examine detailed variation in spending, utilization, and patient outcomes of the two specific high-need patient cohorts across different care settings.

2 METHODOLOGY AND APPROACH 2.1 Formation of the ICCONIC collaborative

To carry out this work, we formed the ICCONIC research collaborative in 2018 where we brought together partners from each of the 11 countries, representing a wide range of institutions, including universities, healthcare providers, think tanks, research centers, and international organizations.15 The research partners included collaborators with experience using routine data to compare healthcare performance at the international level and access to the datasets of interest for the study of HNHC patients.16-20 The 11 participating countries—Australia, Canada, England, France, Germany, the Netherlands, New Zealand, Spain, Sweden, Switzerland, and the United States—all represent high-income countries with high expenditures on health care, but also healthcare systems that are funded and organized differently. For a list of important health system differences, please see Table A1.

Our methodological approach to examine variations in resource use for HNHC personas combines two existing approaches that are relatively novel for international comparison of health systems. First, we propose to use linked, patient-level data to examine the entire care pathway, rather than focusing only on care in the hospital setting allowing us to trace the resources used by patients across the system. Second, our unit of analysis for comparison is the patient, who we follow throughout the system. This approach builds on the use of clinical vignette methodologies to identify similar cohorts of patients, as have been used by other projects to examine resource use in the inpatient setting,3-5, 7 and by international organizations to examine variations in clinical practice.21, 22

To advise the conceptual and methodological approach, and comment on the results, we formed an advisory board consisting of national and international experts within each of the 11 countries. The members include health economists, health services researchers, clinicians, policymakers, and representatives from payers of healthcare services (see Table A2).

2.2 Defining the HNHC patient personas

The first step of the project was to identify a group of HNHC patient subtypes to trace through the different health systems using a predefined clinical vignette, which in this article, we refer to as “HNHC patient personas.” This step is necessary for two reasons. First, HNHC patients are not a homogenous group, and while they include patients with substantial clinical need, their care needs will differ. In order to identify more actionable insights for policymakers and practitioners, we wanted to focus on certain types of HNHC patients that were defined by the same types of need. The second reason we focus on distinct HNHC patients is to ensure comparability of the patient cohorts across countries. This is because the composition of HNHC patient types may vary across countries. Therefore, looking at care trajectories and outcomes of this broader group may produce misleading policy recommendations.

To identify these HNHC patient personas, we defined clinical vignettes that were based from the NAM typology of HNHC priority populations.8 The NAM recently identified priority groups of patients that were among the most expensive to care for have substantial healthcare needs, and are particularly vulnerable to poor-quality care.23-25 Based on the NAM framework, we selected two specific HNHC patient profiles that we believed would be most identifiable across countries, given existing data collection and coding systems, and identified specific types of patients that would belong to this category using a clinical vignette approach. The first included an older adult with frailty (defined by the following clinical vignette: person above age 65 who is hospitalized with hip fracture and received a subsequent hip replacement), and a person with major complex comorbidities (defined by the following clinical vignette: a person between the ages of 65 and 90 hospitalized with heart failure and a comorbidity of diabetes) (Table 1). Both of these clinical vignettes were identifiable through an inpatient admission, which are more consistently coded through more comparable coding systems (mostly deriving from the WHO ICD-10 code system) than those in other settings (e.g., primary care, outpatient care). These decisions were made through a consensus decision-making process by all members in the collaborative, which included physicians, policymakers, data scientists, statisticians, and health economists.

TABLE 1. Identification of high-need, high-cost patient personas for international comparison National Academy of Medicine Priority Population Identified high-need patient personas for comparison Age group Identification with diagnostic codes Frail older person Older person with hip fracture 65 years and above Primary diagnoses of hospitalization S72.0: Fracture of neck of femur S72.1: Pertrochanteric fracture S72.2: Subtrochanteric fracture Procedures (using country-specific codes) Total hip replacement Partial hip replacement Osteosynthesis/pinning Person with complex multimorbidity Older person admitted with a heart failure exacerbation with a comorbidity of diabetes 65–90 years Primary diagnosis of hospitalization Secondary diagnosis of diabetes Note: Across most countries, the diagnostic classification system of ICD-10-WHO codes was used. Spain used ICD-9 codes, whereas the Netherlands used a customized approach to identify relevant codes using input from clinical experts in private insurer data. 2.3 Identifying HNHC personas across countries' datasets

In order to identify and follow each of these personas across their pathway of care over a period of a year, we required at least 2 years of patient-level data. The first year was used as the base year to identify all index cases of relevant patients that met the specific pre-specified clinical vignette definition. We then followed patients for 12 months from the index date of hospitalization to measure the service use, costs, and outcomes of the patients. To identify the index cases, we identified all patients in the base year that were hospitalized for the acute event, using a common set of diagnostic codes and relevant procedure codes (Table 1). Across most countries, we used 2 years between 2015 and 2017, except in Australia (2012–2016) and England (2014–2017), which had smaller samples and, therefore, pooled more years of data (Table A3). International classification disease (ICD-10) codes as defined by the World Health Organization were used from inpatient data files to identify hip fracture patients across all countries. We focused on the codes S72.0, S72.1, and S72.2, which represent fractures of the hip joint. Where ICD-10 codes were unavailable, such as in Spain (ICD-9-CM Codes) and the Netherlands, comparable diagnosis codes were used. Within this group, we then focused on the patients who received one of three procedures: total hip replacement, partial hip replacement, or osteosynthesis—which includes placement of a screw, plate, pin, or internal fixation. Each country used a clinical expert to identify the relevant procedure codes.

For the heart failure persona, we identified all patients hospitalized with a primary diagnosis of congestive heart failure (ICD-10 code I50.x or relevant codes in Spain and the Netherlands). Given the lack of comprehensive longitudinal data across most countries, we were unable to know if the hospitalization was the first hospitalization related to heart failure or not. We then identified the subset of patients who at the time of the first admission also had a diagnosis of diabetes, including ICD-10 codes of E11.x, E12.x, E13.x, and E14.x. Once patients were identified, we then tracked from day one of hospitalization all spending, utilization, and relevant patient outcomes that occurred for a period of 365 days (or until date of death if patients did not survive a full year) (Figure 1).

Identification of the high-need, high-cost personas across countries. The acute event for the two specific personas included an admission for a hip fracture for the persona with frailty and an admission for a heart failure exacerbation for the persona with complex multimorbidity

2.4 Country selection and datasets

In order to validate our approach, data for the two personas—the older frail adult with hip fracture and the older adult with complex multimorbidity—were extracted from the 11 country databases and examined for comparability. We examined comparability in terms of patient characteristics, including age and sex, and also explored variations in the number of chronic conditions captured in administrative data using Elixhauser definitions.26

The participating countries use a range of datasets including administrative claims data, survey data, and registry data. The priority was to have a dataset that captured patient-level and linked information across different components of the healthcare system. For further details on the representativeness of each dataset and years of data used across countries, see Table A3. The datasets differed with regards to their representativeness of the national population as well as their ability to provide linked data across all seven care settings (Table 2). Further detailed information on the datasets used across the 11 countries is listed in Table A4.

TABLE 2. Country dataset information available for public research use in the ICCONIC project Australia Canada England France Germany Netherlands New Zealand Spain Sweden Switzerland United States Inpatient hospital care X X X X X X X X X X X Post-acute rehabilitative care X X X X X X Primary care X X X X X X X X X Outpatient specialty care X X X X X X X X X X X Home health care X X X X X X Outpatient drugs X X X X X X X X X X Long-term care X X X X X

Three of the countries identified full national datasets, including New Zealand, Sweden, and Switzerland. New Zealand utilized data from Integrated Data Infrastructure, which is a linked administrative data repository that includes the entire population. Sweden identified data from their national registry and Switzerland identified data made available from the National Health Statistics provided by the Swiss Federal Statistical Office.

Five of the countries identified datasets that captured a large, regionally diverse sample of the population. For example, in the United States, a 20% Medicare fee-for-service dataset was identified as one of the data sources. Medicare is the public insurance option available for all people over the age of 65 years and some special groups, including those with disability under the age of 65 and those with end-stage renal disease. In France, a health insurance claims dataset covering all of the population in 12 regions was used, representing 70% of the French population. Germany used data from the second largest statutory health insurer, BARMER, representing approximately 10% of the German population and active in all regions of the country. England identified the Clinical Practice Research Datalink (CPRD), which is a sample representing around 10% of the England population and links general practitioner's data with hospital records. The Netherlands obtained access to claims data of an insurance company that has a 30% market share across the country.

Three countries identified regional datasets. For example, Australia used data from the Sax Institute's 45 and Up Study, a regionally representative survey of about 10% of the New South Wales population aged over 45, which is the most populous state in the country. In Spain, data from a secure anonymized health information data-lake (SAHID) covering the entire population of Aragon was used, which represents 3% of the Spanish population. In Canada, data were identified that includes all administrative claims from Ontario, the largest province in the country.

With the identified datasets, three countries—Canada, France, and the Netherlands—had the ability to comprehensively assess care across all seven care domains (Table 2). While Sweden had detailed registry data across all care domains that involve specialized medical doctors, they noted that the primary care data are only accessible at the regional level and had only aggregated and non-linked data for home health. A further three countries (Spain, Germany, and the United States) were able to analyze the care trajectory across six of the seven domains but not able to examine some of the rehabilitative, home-health, or long-term care. Similarly, England was able to identify nationally representative data that would allow for the investigation of four domains, excluding post-acute rehabilitative care, home-health, and long-term care. Switzerland was able to assess relevant patient-level data in the inpatient setting and hospital-based outpatient specialty treatment. New Zealand only had access to inpatient, outpatient specialty care, and pharmaceutical data at this time. All 11 countries were able to capture mortality and readmission outcomes specified for the two patient profiles.

2.5 Characteristics of the hip persona across countries

Across the 11 countries, sample sizes ranged from n = 1859 in Aragon (Spain) to n = 29,134 in the United States (Table 3). The mean patient age (standard deviation) ranged from 81.2 years (SD 6.9) in Switzerland to 85.4 years (SD 7.0) in Spain. The sample was predominantly female, with the proportion of women as high as 77.1% in France and the lowest at 62.8% in Australia. Countries varied in the ability to capture secondary diagnoses in the index hospitalization, ranging from an average of 3.7 comorbidities in the United States to 1.1 in New Zealand and Canada. Of note, the Netherlands was unable to calculate the prevalence of secondary diagnoses given that comorbidities could not be captured using the Elixhauser classification in the insurer data.

TABLE 3. Sample characteristics of frail older person with a hip fracture with subsequent hip replacement across countries Australia Canada England France Germany Netherlands New Zealand Spain Sweden Switzerland United States Sample size 2511 9872 2738 42,849 13,998 4463 2940 1859 14,764 6860 29,134 Age Mean (SD) 84.7 (7.7) 83.4 (8.1) 83.5 (7.9) 84.3 (7.7) 83.5 (7.7) 82.2 (8.0) 84.0 (7.8) 85.4 (7.0) 83.2 (7.6) 81.2 (6.9) 83.2 (8.3) Median 86 85 84 85 84 83 85 84 84 82 84 Female (%) 62.8% 70.7% 71.0% 77.1% 75.9% 70.8% 70.4% 76.7% 67.6% 73.7% 71.4% Chronic conditions Mean (SD) 2.9 (1.9) 1.1 (1.2) 2.2 (1.5) 1.7 (1.5) 3.2 (2.1) n/a 1.1 (1.2) 3.1 (1.5) 2.0 (1.1) 2.1 (1.8) 3.7 (2.0) Median 2 1 2 1 3 1 3 2 2 3 Diabetes 17.6% 19.5% 15.1% 12.4% 19.8% n/a 15.0% 20.2% 14.2% 14.2% 22.5% Heart failure 12.4% 6.0% 10.9% 15.3% 21.6% n/a 4.7% 8.1% 11.0% 8.4% 17.30% Depression 3.9% 1.5% 6.9% 5.8% 11.0% n/a 0.2% 8.3% 2.0% 8.4% 15.3% Hypertension 24.9% 25.9% 55.0% 41.1% 71.3% n/a 11.6% 58.5% 41.3% 51.6% 77.1% Renal failure 12.2% 4.0% 15.2% 6.3% 26.9% n/a 8.3% 9.0% 5.0% 19.4% 19.2% Chronic obstructive pulmonary disease 7.3% 6.3% 22.0% 5.0% 10.1% n/a 4.0% 8.0% 9.5% 8.1% 22.2% S72.0 1411 4953 1865 26,657 6926 n/a 1665 789 7908 3324 14,548 Total replacement 243 (17.2%) 1138 (23.0%) 220 (11.8%) 5988 (22.5%) 1558 (22.5%) n/a 297 (17.8%) 56 (7.1%) 1414 (17.9%) 1001 (30.1%) 1492 (10.3%) Partial replacement 704 (49.9%) 2860 (57.7%) 1184 (63.5%) 12,326 (46.2%) 4424 (62.9%) n/a 879 (52.8%) 626 (79.3%) 3832 (48.5%) 1934 (58.2%) 9245 (63.6%) Pinning 464 (32.9%) 955 (19.3%) 461 (24.7%) 8343 (31.3%) 944 (13.6%) n/a 489 (29.4%) 107 (13.6%) 2662 (33.7%) 389 (11.7%) 3811 (26.2%) S72.1 973 4381 762 13,543 6045 n/a 1125 843 5710 3147 13,274 Total replacement 14 (1.4%) 95 (2.2%) Suppressed 396 (2.9%) 88 (1.5%) n/a 15 (1.3%) 0 (0.0%) 33 (0.6%) 64 (2.0%) 114 (0.9%) Partial replacement 14 (1.4%) 97 (2.2%) Suppressed 213 (1.6%) 86 (1.4%) n/a 12 (1.1%) 10 (1.2%) 31 (0.5%) 68 (2.2%) 317 (2.4%) Pinning 945 (97.1%) 4189 (95.6%) 706 (92.7%) 12,934 (95.5%) 5871 (97.1%) n/a 1098 (97.6%) 833 (98.8%) 5646 (98.9%) 3015 (95.8%) 12,843 (96.8%) S72.2 127 538 111 2649 1027 n/a 150 227 1146 389 1312 Total replacement Suppressed 13 (2.4%) Suppressed 66 (2.5%) 15 (1.5%) n/a Suppressed 0 (0.0%) 11 (1.0%) 6 (1.5%) 13 (1.0%) Partial replacement Suppressed 14 (2.6%) Suppressed 29 (1.1%) 4 (0.4%) n/a Suppressed 14 (6.2%) 7 (0.6%) 20 (5.1%) 26 (2.0%) Pinning 125 (98.4%) 511 (95.0%) 108 (97.3%) 2554 (96.4%) 1008 (98.2%) n/a 147 (98.0%) 213 (93.8%) 1128 (98.4%) 363 (93.3%) 1273 (97.0%) Note: The Netherlands was unable to identify Elixhauser conditions or the specific diagnostic codes for hip in the data provided by the insurer. Clinical experts were used to identify the relevant codes in the insurer data that matched the primary diagnoses of interest.

In all countries, except in Spain, the most common diagnostic code was S72.0: ranging from 68.1% of the sample in England to 42.4% of the sample in Switzerland. This was followed by S72.1 and then S72.2 (see Figure A1). Within the different diagnostic codes, the breakdown of procedu