Original Research
Variations in the persistence of health expenditures and the implications for the design of capitation payments in Taiwan
Journal of Health Services Research & Policy 2015, Vol. 20(3) 146–153 ! The Author(s) 2015 Reprints and permissions: sagepub.co.uk/journalsPermissions.nav DOI: 10.1177/1355819615577711 jhsrp.rsmjournals.com
Li-Jung Elizabeth Ku1, Meng-Jiun Chiou2 and Li-Fan Liu3
Abstract Objectives: The National Health Insurance (NHI) system in Taiwan launched a trial capitation provider payment programme in 2011, with the capitation formula based on patients’ average NHI expenditure in the previous year. This study seeks to examine the concentration and persistence of health care expenditure among the elderly, and to assess the performance of the current capitation formula in predicting future high-cost users. Methods: This study analysed NHI expenditures for a nationally representative sample of people aged 65 years and over who took part in Taiwan’s National Health Interview Survey, 2005. Expenditure concentration was assessed by the proportion of NHI expenditures attributable to four groups by expenditure percentile. Four transition probability matrixes examined changes in a person’s position in the expenditure percentiles and generalized estimation equation models were estimated to identify significant predictors of a patient being in the top 10% of users. Results: Between 2005 and 2009, the top 10% of users on average accounted for 55% of total NHI expenditures. Of the top 10% in 2005, 39% retained this position in 2006. However, expenditure persistence was the highest (77%) among the bottom 50% of users. NHI expenditure percentiles in both the baseline year and the prior year, and chronic conditions all significantly predicted future high expenditures. The model including chronic conditions performed better in predicting the top 10% of users (c-statistics increased from 0.772 to 0.904) than the model without. Conclusions: Given the increase in predictive ability, adding chronic conditions and baseline health care use data to Taiwan’s capitation payment formula would correctly identify more high users.
Keywords capitation, health care expenditures, national health insurance
Introduction Research from the United States and the United Kingdom has consistently shown that a small percentage of individuals account for a large share of health care expenditures, with persistently high expenditures found from one year to the next.1–4 For example, a study showed that the top 5% of people who received any hospital care in the United Kingdom during 1991–2008 accounted for 34% of all inpatient days, while more than 90% of the sample did not have any inpatient days annually.4 A study of Medicare beneficiaries in the United States found that among the most expensive 5% of users in 2002, 23.8% still had high costs in 2003.2 Previous studies on the concentration of Taiwan’s NHI expenditures reported similar findings, for example, that the top 5% of users consumed more than 50% of total medical expenditures in 2003.5 As for cost persistence, the same study reported that 50% of high expenditure
users in 2002 remained high users in 2003. Another study found that the percentage of high users was higher among people aged 65 years and over: the percentage of high users was 56.6% in the elderly population compared to less than 26% in the younger age groups.6 As governments worldwide are faced with tighter budgets, the ability to predict expenditure persistence is increasingly important for health care policymaking. 1 Assistant Professor of Institute of Public Health, College of Medicine, National Cheng Kung University, Taiwan 2 Research Assistant in Institute of Public Health, College of Medicine, National Cheng Kung University, Taiwan 3 Professor of Institute of Gerontology, College of Medicine, National Cheng Kung University, Taiwan
Corresponding author: Li-Fan Liu, Institute of Gerontology, College of Medicine, National Cheng Kung University, No. 1, University Rd., Tainan, Taiwan. Email: [email protected]
Downloaded from hsr.sagepub.com at University of Otago Library on September 28, 2015
Ku et al.
147
National Health Insurance (NHI) was launched in Taiwan in May 1995. NHI is a compulsory, comprehensive, single payer health insurance system. Following the paradigm shift from volume-based reimbursement to value-based reimbursement,7 the NHI system which runs mainly on a fee-for-service (FFS) basis launched a trial of capitation payments in 2011 to reward health care practices for greater prevention and care coordination, rather than volume of services.8 Clinics and hospitals which joined this 3-year trial programme formed eight medical teams to care for 200,000 people, during which time they received fixed payments per person. The capitation payments are based on a formula which calculates virtual global budgets as opposed to actual spending. Medical teams receive surplus payments if patients’ actual spending turns out to be less than their virtual global budgets, but bear the related financial costs if the reverse is true. The capitation formula used to calculate the virtual budget for each practice includes three components: annual NHI expenditures per patient in the previous year; age and sex adjustments based on changes in patient distribution; and the number of patients regularly cared for by each practice. Although this study is not a formal evaluation of Taiwan’s capitation programme, as the trial is still ongoing, we seek to understand whether the payment formula based on prior health care expenditure accurately reflects how health care resources are used. If a high proportion of patients continue to be associated with high expenditures from one year to the next, the current capitation formula is a viable one. However, if individuals require high health care expenditures due to some random event in one year and return to full health in the next year, capitation payments based on prior expenditure alone cannot accurately predict health care utilization. In that case, other measures such as baseline health status and comorbidity measures may have to be used to improve the prediction of future high users. The study examines the concentration and persistence of health care expenditures using longitudinal data from a nationally representative sample of elderly people in Taiwan. We chose to focus on the elderly population, since they account for the largest share of NHI expenditures, and their numbers are expected to grow from 10.9% of the population in 2011 to 14% in 2017.8,9 This study also seeks to assess the predictive power of the current capitation formula which adjusts for age, sex and prior expenditures.
Methods Data sources This study used three datasets provided in the Collaboration Center of Health Information Application,
Ministry of Health and Welfare: the 2005 National Health Interview Survey; the National Health Insurance Research Database (NHIRD); and the National Mortality Registry. In total, 2727 elderly people aged 65 and above were interviewed in the NHIS, but only 1760 gave their consent to link their survey records to their NHIRD claims. Although our study only included those who consented, according to a previous study using the NHIS dataset, those who refused to give their consent were more likely to be male, divorced/separated/widowed, illiterate or with lower household income.10 Health care expenditures from 2005 to 2009 based on NHI claims were analysed, and the survival status for each sample person by year was verified by linking individual identification information to the National Mortality Registry. After excluding those who did not give their consent and those with missing data on the health indicators, our final sample comprised 1609 people aged 65 and older.
Total health care expenditures The health care expenditures recorded in Taiwan’s NHI system include expenses for ambulatory care, emergency room visits, hospitalization, laboratory tests, pharmaceuticals and patient co-payments. The annual total NHI expenditures were calculated by adding up the costs for all individuals in our sample. For each year, the surviving sample was ranked by annual total NHI expenditures, from the highest to the lowest percentile, and divided into four user groups: top 10%; next 11% to 25%; next 26% to 50%; and bottom 50%. We selected these percentile cut-offs based on existing literature in order to compare our results with previous findings on users in the top decile.1,3 The ranges of expenditure corresponding to each of the four percentile groups by year are shown in Table 1. The wide range in NHI expenditure shown in Table 1 supports our decision to rank expenditures instead of using the absolute numbers since the latter were easily affected by outliers.
Sample characteristics The sample was characterized in terms of: sociodemographic factors; economic factors; and health measures. The sociodemographic factors were age, sex, education level and marital status, while the economic factors were geographic region, urban or rural residence and household income per person. Since 17% of the sample had missing values on household income per person, we created a missing income category so that those individuals could be retained in the analysis. The health measures were self-rated health status, the number of limitations in basic activities of daily living (ADLs), instrumental activities of daily living
Downloaded from hsr.sagepub.com at University of Otago Library on September 28, 2015
148
Journal of Health Services Research & Policy 20(3)
Table 1. Distribution of NHI expenditure by year. Expenditure percentile
Range (NTD$) 2005
2006
2007
2008
2009
10% 11–25% 26–50% >50%
93,705–1,370,021 45,449–93,441 20,529–45,390 0–20,514
96,351–1,637,484 47,309–96,005 21,842–47,256 0–21,831
121,219–1,335,525 52,850–119,781 21,834–52,046 0–21,686
115,277–4,715,266 52,226–112,672 25,141–52,156 0–25,061
139,389-1,477,861 57,855-136,515 25,191-57,851 0–25,152
NTD: New Taiwan Dollars (the average exchange rate was 32.3 NTD ¼ 1 USD in 2005).
(IADLs) and selected chronic health conditions. Cancer, cerebrovascular disease and kidney disease were selected because they have been identified as among the most expensive conditions occurring in inpatient settings across Taiwan.9 Chronic obstructive pulmonary disease (COPD), diabetes mellitus and hypertension were also included given their high prevalence in the sample. These six conditions (yes/no) were identified by two outpatient diagnoses or one inpatient record with corresponding ICD-9-CM codes from NHI claims.
Statistical analyses The analysis began with expenditure concentration which was defined by the proportion of annual NHI expenditures attributable to each of the four aforementioned user groups. The second part of the analysis assessed expenditure persistence by comparing changes in NHI expenditure ranks from prior year t to the subsequent year t þ 1. Four transition probability matrixes examined changes in individuals’ positions over 2 years. Generalized estimation equation (GEE) models using the logit link function for binary responses were estimated to find predictors of patients being in the top 10% of users. We controlled for chronic conditions to determine whether prior health expenditure rank was itself a significant predictor of future health care spending conditional on health. The models also controlled for individual sociodemographic and economic variables likely to affect health care expenditures. Finally, we compared model performance using c-statistics – the area under the receiver operating characteristic (ROC) curve – from different sets of predictors to see if the current capitation formula could be improved by adding additional predictors.
Results Expenditure concentration The top 10% of users accounted for 54.5% to 58.3% of total NHI expenditures from 2005 to 2009 (Figure 1). During the same period, the cumulative percentage of
expenditure by adding those ranked 11–25% to the top 10% was over 75% of total NHI expenditures. Total NHI expenditures were highly concentrated among elderly beneficiaries who ranked in the top 50%, while those in the bottom 50% incurred only about 8% total spending. Although the trend in expenditure concentration increased slightly for the top 10% of users over 5 years, the trend was largely consistent for both the top 25% and top 50% of users. There were significant differences in the characteristics of the individuals in the four groups ranked by percentiles of annual NHI expenditures in 2005 (Table 2). Those in the top 10% group had higher inpatient expenditures than outpatient expenditures, while the other three groups all reported inpatient expenditures much lower than outpatient expenditures. The top 10% of users were older, more likely to be divorced/separated/widowed and living in an urban area. Compared to the individuals in the lower expenditure groups, those in the top 10% group were far more likely to be in poor health, and with a greater number of limitations in both ADLs and IADLs. Chronic conditions were also much more prevalent among this group.
Expenditure persistence Four transition probability matrixes which compare the ranks in NHI expenditure from one year to the next are shown in Table 3. In addition to the column showing the percentage of beneficiaries who died during the first year, transition probabilities on the diagonal line indicate expenditure persistence measured by the proportion of the sample who maintained their percentile ranks in both years. For example, out of the top 10% of users in 2005, only 39% retained this position in 2006, whereas 52% moved to a lower rank. Figure 2 illustrates the changes in transition probabilities on the diagonal line taken from all matrixes. Among the four user groups, the top 10% of users were found to have the greatest variation in expenditure persistence, ranging from 31% to 39%. Figure 2 also shows that NHI expenditure persistence was much higher for those in the bottom 50% than in the top 10%.
Downloaded from hsr.sagepub.com at University of Otago Library on September 28, 2015
Ku et al.
149
Figure 1. Cumulative percentages of annual National Health Insurance expenditures attributable to the top 10%, 11–25%, and 26– 50% of elderly beneficiaries, 2005–2009. Source: 2005 National Health Interview Survey and 2005–2009 National Health Insurance Research Database.
Predicting current high users from previous use Results of the GEE model show that chronic conditions and prior ranks in NHI expenditures are important predictors of being a top 10% user (Table 4). The odds of being in the top 10% for someone with a cancer diagnosis was 16.4 times greater than the odds for someone with none of the six selected chronic conditions. In fact, someone with any of the six conditions was more likely to be among the top 10% of users. In terms of previous utilization, we found expenditure ranks in both the baseline year (2005) and in the previous year to be significant predictors of current high use, although the odds ratio for the prior year were larger than the odds ratios for the baseline year. Table 4 also shows that none of the other sample characteristics such as age or income were significant predictors of being in the top 10%. Results from ROC curve analyses comparing model performance across different sets of predictors are listed in Table 5. The current capitation formula which adjusts for age, sex and previous expenditure rank (model 2) had a c-statistic of 0.754, but c-statistics were higher for more comprehensive models. Chronic conditions seemed to the best predictor of the top 10% of users as the largest increase in c-statistic was observed between model 4 and model 5 (0.132 point).
Discussion Health care expenditures on those aged 65 and older in Taiwan were concentrated in a small proportion of
patients: the top 10% of users accounted for 55% of aggregate NHI expenditures, while the top 25% accounted for 75%. These figures are similar to the findings of previous research which showed that among all NHI enrollees in Taiwan, the top 20% of users accounted for 75% of expenditures.6 Similar expenditure concentration has been found in the United States, as 23% of Medicare beneficiaries with five or more chronic conditions accounted for 68% of the programme’s spending.11 In the presence of expenditure concentration, risk-adjusted capitation payments are crucial so that providers will not suffer financially if they manage more complicated cases. Many previous studies have found that past health care use is a significant predictor of future use, after controlling for health and sociodemographic factors.1,3,4,12 Nevertheless, the pitfall of including prior expenditure as a risk adjustment factor is that the absolute number in dollars could be manipulated and promote provision of more and maybe unnecessary services.13 This study focused on the rank in health expenditure which is considered a relative measure that cannot be determined by a single health care provider. We found that less than half of the top 50% of users remained in this position for 2 years. Thus using the average annual NHI expenditures per patient as the single capitation rate for all beneficiaries is likely to be inadequate. Our finding that initial positions in the expenditure percentiles and presence of chronic health conditions were both significant predictors in addition to prior health care use deserves further discussion.
Downloaded from hsr.sagepub.com at University of Otago Library on September 28, 2015
150
Journal of Health Services Research & Policy 20(3)
Table 2. Characteristics and type of services used by percentile of NHI expenditures among elderly beneficiaries, 2005.a N ¼ 1609
Expenditure percentile
Sample characteristics (%)
10%
Total NHI expenditures, mean (SD) Outpatient Prescription drugs Inpatient Age Gender Male Female Education Illiterate Literate/Primary school Middle school and above Marital status Single/Other Married Divorced/separated/widowed Residence Rural Urban Region North Central South East
249,822 114,749 29,057 135,073 75.1
Household income per person < NT US$6000 NT US$6000–15,000 > ¼ NT US$15,000 Missing Self-rated health Excellent/ Very good Good Fair Poor No. of ADLs, mean (SD) No. of IADLs, mean (SD) Chronic conditions Cancer Cerebrovascular disease COPD Diabetes mellitus Hypertension Kidney disease
(202,295) (139,958) (26,888) (154,786) (5.8)
11–25%
26–50%
>50%
p-Valueb
62,354 51,342 24,845 11,012 74.1
30,890 28,784 13,294 2106 73.8
7807 7670 2593 136 72.4
Variations in the persistence of health expenditures and the implications for the design of capitation payments in Taiwan
Journal of Health Services Research & Policy 2015, Vol. 20(3) 146–153 ! The Author(s) 2015 Reprints and permissions: sagepub.co.uk/journalsPermissions.nav DOI: 10.1177/1355819615577711 jhsrp.rsmjournals.com
Li-Jung Elizabeth Ku1, Meng-Jiun Chiou2 and Li-Fan Liu3
Abstract Objectives: The National Health Insurance (NHI) system in Taiwan launched a trial capitation provider payment programme in 2011, with the capitation formula based on patients’ average NHI expenditure in the previous year. This study seeks to examine the concentration and persistence of health care expenditure among the elderly, and to assess the performance of the current capitation formula in predicting future high-cost users. Methods: This study analysed NHI expenditures for a nationally representative sample of people aged 65 years and over who took part in Taiwan’s National Health Interview Survey, 2005. Expenditure concentration was assessed by the proportion of NHI expenditures attributable to four groups by expenditure percentile. Four transition probability matrixes examined changes in a person’s position in the expenditure percentiles and generalized estimation equation models were estimated to identify significant predictors of a patient being in the top 10% of users. Results: Between 2005 and 2009, the top 10% of users on average accounted for 55% of total NHI expenditures. Of the top 10% in 2005, 39% retained this position in 2006. However, expenditure persistence was the highest (77%) among the bottom 50% of users. NHI expenditure percentiles in both the baseline year and the prior year, and chronic conditions all significantly predicted future high expenditures. The model including chronic conditions performed better in predicting the top 10% of users (c-statistics increased from 0.772 to 0.904) than the model without. Conclusions: Given the increase in predictive ability, adding chronic conditions and baseline health care use data to Taiwan’s capitation payment formula would correctly identify more high users.
Keywords capitation, health care expenditures, national health insurance
Introduction Research from the United States and the United Kingdom has consistently shown that a small percentage of individuals account for a large share of health care expenditures, with persistently high expenditures found from one year to the next.1–4 For example, a study showed that the top 5% of people who received any hospital care in the United Kingdom during 1991–2008 accounted for 34% of all inpatient days, while more than 90% of the sample did not have any inpatient days annually.4 A study of Medicare beneficiaries in the United States found that among the most expensive 5% of users in 2002, 23.8% still had high costs in 2003.2 Previous studies on the concentration of Taiwan’s NHI expenditures reported similar findings, for example, that the top 5% of users consumed more than 50% of total medical expenditures in 2003.5 As for cost persistence, the same study reported that 50% of high expenditure
users in 2002 remained high users in 2003. Another study found that the percentage of high users was higher among people aged 65 years and over: the percentage of high users was 56.6% in the elderly population compared to less than 26% in the younger age groups.6 As governments worldwide are faced with tighter budgets, the ability to predict expenditure persistence is increasingly important for health care policymaking. 1 Assistant Professor of Institute of Public Health, College of Medicine, National Cheng Kung University, Taiwan 2 Research Assistant in Institute of Public Health, College of Medicine, National Cheng Kung University, Taiwan 3 Professor of Institute of Gerontology, College of Medicine, National Cheng Kung University, Taiwan
Corresponding author: Li-Fan Liu, Institute of Gerontology, College of Medicine, National Cheng Kung University, No. 1, University Rd., Tainan, Taiwan. Email: [email protected]
Downloaded from hsr.sagepub.com at University of Otago Library on September 28, 2015
Ku et al.
147
National Health Insurance (NHI) was launched in Taiwan in May 1995. NHI is a compulsory, comprehensive, single payer health insurance system. Following the paradigm shift from volume-based reimbursement to value-based reimbursement,7 the NHI system which runs mainly on a fee-for-service (FFS) basis launched a trial of capitation payments in 2011 to reward health care practices for greater prevention and care coordination, rather than volume of services.8 Clinics and hospitals which joined this 3-year trial programme formed eight medical teams to care for 200,000 people, during which time they received fixed payments per person. The capitation payments are based on a formula which calculates virtual global budgets as opposed to actual spending. Medical teams receive surplus payments if patients’ actual spending turns out to be less than their virtual global budgets, but bear the related financial costs if the reverse is true. The capitation formula used to calculate the virtual budget for each practice includes three components: annual NHI expenditures per patient in the previous year; age and sex adjustments based on changes in patient distribution; and the number of patients regularly cared for by each practice. Although this study is not a formal evaluation of Taiwan’s capitation programme, as the trial is still ongoing, we seek to understand whether the payment formula based on prior health care expenditure accurately reflects how health care resources are used. If a high proportion of patients continue to be associated with high expenditures from one year to the next, the current capitation formula is a viable one. However, if individuals require high health care expenditures due to some random event in one year and return to full health in the next year, capitation payments based on prior expenditure alone cannot accurately predict health care utilization. In that case, other measures such as baseline health status and comorbidity measures may have to be used to improve the prediction of future high users. The study examines the concentration and persistence of health care expenditures using longitudinal data from a nationally representative sample of elderly people in Taiwan. We chose to focus on the elderly population, since they account for the largest share of NHI expenditures, and their numbers are expected to grow from 10.9% of the population in 2011 to 14% in 2017.8,9 This study also seeks to assess the predictive power of the current capitation formula which adjusts for age, sex and prior expenditures.
Methods Data sources This study used three datasets provided in the Collaboration Center of Health Information Application,
Ministry of Health and Welfare: the 2005 National Health Interview Survey; the National Health Insurance Research Database (NHIRD); and the National Mortality Registry. In total, 2727 elderly people aged 65 and above were interviewed in the NHIS, but only 1760 gave their consent to link their survey records to their NHIRD claims. Although our study only included those who consented, according to a previous study using the NHIS dataset, those who refused to give their consent were more likely to be male, divorced/separated/widowed, illiterate or with lower household income.10 Health care expenditures from 2005 to 2009 based on NHI claims were analysed, and the survival status for each sample person by year was verified by linking individual identification information to the National Mortality Registry. After excluding those who did not give their consent and those with missing data on the health indicators, our final sample comprised 1609 people aged 65 and older.
Total health care expenditures The health care expenditures recorded in Taiwan’s NHI system include expenses for ambulatory care, emergency room visits, hospitalization, laboratory tests, pharmaceuticals and patient co-payments. The annual total NHI expenditures were calculated by adding up the costs for all individuals in our sample. For each year, the surviving sample was ranked by annual total NHI expenditures, from the highest to the lowest percentile, and divided into four user groups: top 10%; next 11% to 25%; next 26% to 50%; and bottom 50%. We selected these percentile cut-offs based on existing literature in order to compare our results with previous findings on users in the top decile.1,3 The ranges of expenditure corresponding to each of the four percentile groups by year are shown in Table 1. The wide range in NHI expenditure shown in Table 1 supports our decision to rank expenditures instead of using the absolute numbers since the latter were easily affected by outliers.
Sample characteristics The sample was characterized in terms of: sociodemographic factors; economic factors; and health measures. The sociodemographic factors were age, sex, education level and marital status, while the economic factors were geographic region, urban or rural residence and household income per person. Since 17% of the sample had missing values on household income per person, we created a missing income category so that those individuals could be retained in the analysis. The health measures were self-rated health status, the number of limitations in basic activities of daily living (ADLs), instrumental activities of daily living
Downloaded from hsr.sagepub.com at University of Otago Library on September 28, 2015
148
Journal of Health Services Research & Policy 20(3)
Table 1. Distribution of NHI expenditure by year. Expenditure percentile
Range (NTD$) 2005
2006
2007
2008
2009
10% 11–25% 26–50% >50%
93,705–1,370,021 45,449–93,441 20,529–45,390 0–20,514
96,351–1,637,484 47,309–96,005 21,842–47,256 0–21,831
121,219–1,335,525 52,850–119,781 21,834–52,046 0–21,686
115,277–4,715,266 52,226–112,672 25,141–52,156 0–25,061
139,389-1,477,861 57,855-136,515 25,191-57,851 0–25,152
NTD: New Taiwan Dollars (the average exchange rate was 32.3 NTD ¼ 1 USD in 2005).
(IADLs) and selected chronic health conditions. Cancer, cerebrovascular disease and kidney disease were selected because they have been identified as among the most expensive conditions occurring in inpatient settings across Taiwan.9 Chronic obstructive pulmonary disease (COPD), diabetes mellitus and hypertension were also included given their high prevalence in the sample. These six conditions (yes/no) were identified by two outpatient diagnoses or one inpatient record with corresponding ICD-9-CM codes from NHI claims.
Statistical analyses The analysis began with expenditure concentration which was defined by the proportion of annual NHI expenditures attributable to each of the four aforementioned user groups. The second part of the analysis assessed expenditure persistence by comparing changes in NHI expenditure ranks from prior year t to the subsequent year t þ 1. Four transition probability matrixes examined changes in individuals’ positions over 2 years. Generalized estimation equation (GEE) models using the logit link function for binary responses were estimated to find predictors of patients being in the top 10% of users. We controlled for chronic conditions to determine whether prior health expenditure rank was itself a significant predictor of future health care spending conditional on health. The models also controlled for individual sociodemographic and economic variables likely to affect health care expenditures. Finally, we compared model performance using c-statistics – the area under the receiver operating characteristic (ROC) curve – from different sets of predictors to see if the current capitation formula could be improved by adding additional predictors.
Results Expenditure concentration The top 10% of users accounted for 54.5% to 58.3% of total NHI expenditures from 2005 to 2009 (Figure 1). During the same period, the cumulative percentage of
expenditure by adding those ranked 11–25% to the top 10% was over 75% of total NHI expenditures. Total NHI expenditures were highly concentrated among elderly beneficiaries who ranked in the top 50%, while those in the bottom 50% incurred only about 8% total spending. Although the trend in expenditure concentration increased slightly for the top 10% of users over 5 years, the trend was largely consistent for both the top 25% and top 50% of users. There were significant differences in the characteristics of the individuals in the four groups ranked by percentiles of annual NHI expenditures in 2005 (Table 2). Those in the top 10% group had higher inpatient expenditures than outpatient expenditures, while the other three groups all reported inpatient expenditures much lower than outpatient expenditures. The top 10% of users were older, more likely to be divorced/separated/widowed and living in an urban area. Compared to the individuals in the lower expenditure groups, those in the top 10% group were far more likely to be in poor health, and with a greater number of limitations in both ADLs and IADLs. Chronic conditions were also much more prevalent among this group.
Expenditure persistence Four transition probability matrixes which compare the ranks in NHI expenditure from one year to the next are shown in Table 3. In addition to the column showing the percentage of beneficiaries who died during the first year, transition probabilities on the diagonal line indicate expenditure persistence measured by the proportion of the sample who maintained their percentile ranks in both years. For example, out of the top 10% of users in 2005, only 39% retained this position in 2006, whereas 52% moved to a lower rank. Figure 2 illustrates the changes in transition probabilities on the diagonal line taken from all matrixes. Among the four user groups, the top 10% of users were found to have the greatest variation in expenditure persistence, ranging from 31% to 39%. Figure 2 also shows that NHI expenditure persistence was much higher for those in the bottom 50% than in the top 10%.
Downloaded from hsr.sagepub.com at University of Otago Library on September 28, 2015
Ku et al.
149
Figure 1. Cumulative percentages of annual National Health Insurance expenditures attributable to the top 10%, 11–25%, and 26– 50% of elderly beneficiaries, 2005–2009. Source: 2005 National Health Interview Survey and 2005–2009 National Health Insurance Research Database.
Predicting current high users from previous use Results of the GEE model show that chronic conditions and prior ranks in NHI expenditures are important predictors of being a top 10% user (Table 4). The odds of being in the top 10% for someone with a cancer diagnosis was 16.4 times greater than the odds for someone with none of the six selected chronic conditions. In fact, someone with any of the six conditions was more likely to be among the top 10% of users. In terms of previous utilization, we found expenditure ranks in both the baseline year (2005) and in the previous year to be significant predictors of current high use, although the odds ratio for the prior year were larger than the odds ratios for the baseline year. Table 4 also shows that none of the other sample characteristics such as age or income were significant predictors of being in the top 10%. Results from ROC curve analyses comparing model performance across different sets of predictors are listed in Table 5. The current capitation formula which adjusts for age, sex and previous expenditure rank (model 2) had a c-statistic of 0.754, but c-statistics were higher for more comprehensive models. Chronic conditions seemed to the best predictor of the top 10% of users as the largest increase in c-statistic was observed between model 4 and model 5 (0.132 point).
Discussion Health care expenditures on those aged 65 and older in Taiwan were concentrated in a small proportion of
patients: the top 10% of users accounted for 55% of aggregate NHI expenditures, while the top 25% accounted for 75%. These figures are similar to the findings of previous research which showed that among all NHI enrollees in Taiwan, the top 20% of users accounted for 75% of expenditures.6 Similar expenditure concentration has been found in the United States, as 23% of Medicare beneficiaries with five or more chronic conditions accounted for 68% of the programme’s spending.11 In the presence of expenditure concentration, risk-adjusted capitation payments are crucial so that providers will not suffer financially if they manage more complicated cases. Many previous studies have found that past health care use is a significant predictor of future use, after controlling for health and sociodemographic factors.1,3,4,12 Nevertheless, the pitfall of including prior expenditure as a risk adjustment factor is that the absolute number in dollars could be manipulated and promote provision of more and maybe unnecessary services.13 This study focused on the rank in health expenditure which is considered a relative measure that cannot be determined by a single health care provider. We found that less than half of the top 50% of users remained in this position for 2 years. Thus using the average annual NHI expenditures per patient as the single capitation rate for all beneficiaries is likely to be inadequate. Our finding that initial positions in the expenditure percentiles and presence of chronic health conditions were both significant predictors in addition to prior health care use deserves further discussion.
Downloaded from hsr.sagepub.com at University of Otago Library on September 28, 2015
150
Journal of Health Services Research & Policy 20(3)
Table 2. Characteristics and type of services used by percentile of NHI expenditures among elderly beneficiaries, 2005.a N ¼ 1609
Expenditure percentile
Sample characteristics (%)
10%
Total NHI expenditures, mean (SD) Outpatient Prescription drugs Inpatient Age Gender Male Female Education Illiterate Literate/Primary school Middle school and above Marital status Single/Other Married Divorced/separated/widowed Residence Rural Urban Region North Central South East
249,822 114,749 29,057 135,073 75.1
Household income per person < NT US$6000 NT US$6000–15,000 > ¼ NT US$15,000 Missing Self-rated health Excellent/ Very good Good Fair Poor No. of ADLs, mean (SD) No. of IADLs, mean (SD) Chronic conditions Cancer Cerebrovascular disease COPD Diabetes mellitus Hypertension Kidney disease
(202,295) (139,958) (26,888) (154,786) (5.8)
11–25%
26–50%
>50%
p-Valueb
62,354 51,342 24,845 11,012 74.1
30,890 28,784 13,294 2106 73.8
7807 7670 2593 136 72.4
