To examine whether continuity of ambulatory asthma care can lower asthma-specific emergency department (ED) utilization by children with asthma in Taiwan.

Retrospective cohort study based on claims data. 

We used the Taiwan National Health Insurance Dataset, 2006 to 2009. The study population was new asthma patients aged 0 to 17 years in 2007, and every case was observed for 2 years. We used the Continuity of Care Index (COCI) to calculate the continuity of ambulatory asthma care in the first year, and estimated the asthma-specific ED utilization in the second year. Two-part hurdle regression was used for statistical analysis.

The 29,277 patients in our study had an average COCI of 0.68 (± 0.31), and 42.3% of patients had an index of 1. More than 1 in 20 patients—1641 (5.61%)&mdash;had at least 1 asthma ED visit, and the mean number of visits per user was 1.46 (± 0.99). After controlling for covariates, the groups with medium and low continuity of ambulatory asthma care had 21% (odds ratio [OR], 1.21; 95% CI, 1.06-1.39) and 38% (OR, 1.38; 95% CI, 1.21-1.58) higher asthma-related ED utilization, respectively, than the group with high COCI. However, among users, the number of ED visits was not statistically correlated to the continuity of ambulatory asthma care.

High continuity of ambulatory asthma care can decrease asthma-specific ED utilization risk in children with newly diagnosed asthma in Taiwan. We suggest that providers and the government reinforce the use of follow-up care and education for high-risk groups to improve the continuity of ambulatory asthma care.

	This was a retrospective cohort study using Taiwan’s National Health Insurance Dataset, aimed to examine the effects of continuity of ambulatory asthma care on asthma-related emergency department (ED) utilization for children with asthma. We used 2-part hurdle regression.

		For the first part, higher continuity of ambulatory asthma care significantly decreased the risk of having asthma-specific ED visits.

		For the second part, which examined the asthma-specific ED users only, continuity of ambulatory asthma care did not significantly correlate with the number of visits.

		Better continuity of ambulatory asthma care might improve outcomes for children with asthma.

Asthma is among the most common respiratory diseases, with 300 million affected people worldwide, according to the World Health Organization.

 About 1 in 10 American children suffered from the disease in 2009.

 Children with asthma are prone to acute, sometimes severe, attacks. Between 2005 and 2007, US children with asthma visited the emergency department (ED) 9.9 times per year on average, compared with the average 7.5 annual visits for US patients with asthma of all ages.

The US Agency for Healthcare Research and Quality has indicated that surveillance of asthma care should include daily symptom burden and acute avoidable events due to asthma, such as asthma ED visit rate and asthma admission rate. In particular, asthma ED visits reflect poor disease management on the part of providers and patients.

The US Institute of Medicine defines continuity of care (COC) as the condition of “care over time by a single individual or team of health professionals.”

 Within this long-term relationship, healthcare providers can better manage the chronic conditions of their patients because of their familiarity with their patients’ medical history and greater ability to communicate with their patients because of repeat interactions. COC leads to greater patient satisfaction; less disability and pain; and decreased ED utilization, hospitalization, and medical costs.

For patients with asthma, research has shown that higher continuity of asthma care can decrease asthma ED utilization.

 However, most of these investigations were conducted in Western countries; COC has not been well studied in Asian children with asthma. We therefore aimed to examine whether continuity of ambulatory asthma care could lower asthma ED use among children with asthma in Taiwan. We adopted 2-part hurdle regression models to analyze the data for all children with asthma, and separately for those who had ED visits.

In 1995, Taiwan implemented a single-payer, compulsory national health insurance (NHI) program that now enrolls nearly 100% of residents. Universal insurance coverage and low co-payments minimize the economic barrier to care for patients. With no gatekeeper program, patients choose the provider of their choice.

The NHI data set is a national, population-based healthcare claim database containing detailed records of outpatient visits, ED visits, and hospital admissions (including diagnosis, procedures, medication, provider information, and expense). We used the data for 2006 to 2009. By using secondary data analysis, approval from an institutional review board was not necessary for this study.

We conducted a retrospective cohort study of new patients with asthma aged under 18 years in 2007. New asthma patients were defined as those having 2 outpatient or ED visits, or 1 admission, due to asthma (The 

International Classification of Diseases, Ninth Revision, Clinical Modification 

] code for primary diagnosis of 493.XX) in 2007 but without such records in the previous year.

 The date of the patient’s first asthma visit/admission was called the index date, and all patients were observed for 2 years after the index date. Excluded subjects were those who died during the observation period, had an “unknown” study variable, or whose primary care physician or clinic/hospital could not be defined.

The first year of observation was considered the COC period. The number of asthma-related outpatient visits in this period was used to calculate the COC index, which represented the extent of care provided by the primary care physician. Previous studies have cautioned that bias may occur in calculating the COC index if the patient had too few outpatient visits; as a result, we included only patients with at least 3 asthma-related outpatient visits in the COC period.

 The second year of observation was the outcome period, used to estimate asthma ED utilization.

The 74,705 new 0- to 17-year-old patients with asthma in 2007 represented an incidence rate of 1.5% of the 5,002,123 children in Taiwan. After excluding patients who died during the observation period (n = 7), those with too few asthma outpatient visits in the COC period (n = 41,572), those with any study variable unknown (n = 834), and those whose primary care physician or clinic/hospital could not be defined (n = 3015), we studied 29,277 children with asthma.

The outcome variables included a) asthma ED visit or not, and b) the number of asthma ED visits in the outcome period. An asthma ED visit was defined as an ED visit with a primary or secondary diagnosis of asthma. Because the following were unrelated to regular health-seeking behavior, we excluded any ED visit for injuries, poisoning (

 850-995) or those with supplementary classifications (V-codes), such as chemotherapy.

The independent variable was continuity of asthma-specific ambulatory care. Jee and Cabana categorized the types of indexes for measuring COC.

 We opted to use a dispersion-type index because it is easy to calculate, is not prone to be affected by the number of visits, and considers all providers. We used the Continuity of Care Index (COCI),

 a dispersion-type index adopted by numerous studies.

The COCI for asthma-specific ambulatory care was calculated by using the number of a patient’s outpatient visits for asthma (primary or secondary diagnosis) at a clinic or hospital during the COC period. The equation is as follows:

N represents the total number of asthma-related outpatient visits, n is the number of asthma-related outpatient visits to a physician, 

The COCI value ranges from 0 to 1, with higher values representing better COC. We defined the high COC group (42.3% of subjects) as COCI = 1, indicating that all of a patient’s asthma care in the COC period was provided by the same physician. All other subjects were dichotomized by their median of COCI into the medium COC group (COCI = 0.43-0.99) and the low COC group (COCI <0.43).

Patient characteristics included sex, age (0-4, 5-9, or 10-17 years), income status, urbanization level of insurance registry district, and enrollment in the asthma pay-for-performance (P4P) program. Since health status during the COC period may impact the outcome, we used the total number of asthma outpatient visits as a proxy for health status and disease severity,

 categorizing subjects into those with low (3-4), medium (5-8), or high (≥9) visits based on the tertile. We also observed whether the patient had asthma ED visits and the total length of stay for asthma-related hospital admissions during the COC period. For providers, we included sex and mean number of asthma outpatient visits in 2007 and 2008 for the primary care physician responsible for the most asthma care for the patient. We also controlled for the place at which the patient most often received care by the primary care physician, sorting them into hospitals (with fewer than 250, 250 to 499, and 500 or more beds) and nonhospital clinics.

All analyses were conducted using SAS version

SAS Institute, Cary, North Carolina). Descriptive statistics included percentage, mean, standard deviation, and minimum and maximum of the study variables. The χ

 test and Kruskal-Wallis test were used for bivariate analyses. The significance level was set as 0.05.

In multivariate analysis, an excess of zeroes (ie, no ED visits) may bias the parameter estimation and affect the inference.

 Most previous studies analyzed all patients and did not take into account the influence of having many patients with no asthma ED visit. In this study, 94.39% of subjects had no asthma ED visit during the outcome period; we therefore used hurdle regression, as developed by Mullahy, rather than multiple regression, Poisson regression, or negative binominal regression.

 Hurdle regression uses a 2-part model: the first component was used to model whether the patient had an asthma ED visit or not by logistic regression; in the second component, which focused on the users only, the number of asthma ED visits was modeled using a left-truncated Poisson regression.

The average COCI of our study subjects was 0.68 (± 0.31) and, of all 29,277 subjects, 12,383 (42.30%) were in the high COC group (COCI = 1), 8317 (28.41%) in the medium COC group (COCI = 0.43-0.99), and 8577 (29.30%) in the low COC group (COCI <0.43). The characteristics of all study subjects and of each COC group are presented in 

. All characteristics, except sex and income status, were significantly related to continuity of ambulatory asthma care (

 shows asthma ED utilization during the outcome period. Among all subjects, 1641 (5.61%) had at least 1 asthma ED visit, and the mean number of visits for the users was 1.46 (

 0.99). In the high COC group, 3.96% of patients had at least 1 asthma ED visit; the figures were 6.60% of patients in the medium and 7.02% of patients in the low COC groups. The difference was statistically significant (

<.001). For ED users, the average number of asthma ED visits was 1.42 (± 0.89) in the high COC group, 1.44 (± 0.99) in the medium COC group, and 1.53 (± 1.05) in the low COC group; this difference was statistically significant (

The results of hurdle regression are presented in 

. The first component (binary OR) of the hurdle regression was used to analyze whether continuity of ambulatory asthma care reduced the risk of an asthma-related ED visit. As a crude result, the risks for the medium and the low COC groups were significantly higher than for the high COC group (OR, 1.72; 95% CI, 1.51-1.94; and OR, 1.83; 95% CI, 1.62-2.07, respectively). After controlling for the covariates mentioned above, the adjusted risks of an asthma-related ED visit for the medium and low COC groups were 1.21 (95% CI, 1.06-1.39) and 1.38 (95% CI, 1.21-1.58) times that of the high COC group, respectively.

The second component (count risk ratio [RR]) of hurdle regression was used to analyze whether continuity of ambulatory asthma care affected the number of asthma ED visits for the users only. As a crude result, the number of ED visits of the low COC group was 23% higher than that of the high COC group (95% CI, 1.04-1.45), but the medium and high COC groups did not differ significantly (RR, 1.03; 95% CI, 0.86-1.23). After controlling for covariates, the number of ED visits did not differ significantly between the medium and high COC groups (RR, 0.86; 95% CI, 0.72-1.04) or between the low and high COC groups (RR, 1.00; 95% CI, 0.84-1.19).

We found that children in Taiwan with newly diagnosed asthma generally had high continuity of ambulatory asthma care. The average COCI was 0.68 (± 0.31), while fully 42% of the patients had a COCI of 1, meaning they received all their asthma-related care from a single physician during the first year of observation. This high level of COC in Taiwan may be attributed to several factors. First, in addition to extensive insurance coverage and low co-payments for outpatient visits, Taiwan has no compulsory referral scheme or gatekeeper program, giving patients great freedom in choosing care providers. Second, the low birth rate and high level of education of parents allow them to focus on disease management for their children with chronic conditions like asthma. Third, the government began an asthma P4P program in 2001 based on a continuing care model, encouraging clinics and hospitals to provide patient-centered care focused on disease management, enforcement of follow-ups, and enhancement of patient self-care ability.

This study supports the results of Cree and Cyr et al.

 Higher continuity of ambulatory asthma care could lower the risk of using emergency care for children with asthma, with a trend of a dose-response effect. Hong et al studied COC in older adults in South Korea with 4 different chronic diseases, including asthma and diabetes, and found increased COC associated with a reduction in the risk of ED visits. This negative correlation was greater for asthma than for diabetes.

 In Taiwan, previous research on all patients and those with diabetes found that higher levels of COC could lower ED visits and hospital admissions.

 Thus, these findings support government programs like P4P to improve COC for patients with diabetes. Since our research showed that higher COC also had positive impacts on children with asthma, we believe this may help the government to plan programs to improve COC for patients with asthma in order to reduce the risk of ED visits.

With higher COC, patients and their physicians develop greater familiarity with each other as well as higher levels of trust. The physician can then more efficiently manage the medical conditions of returning patients, and patients are more likely to comply with instructions, subsequently leading to fewer asthma attacks. A previous study found that when providers designated someone to follow up with patients with asthma who were discharged from the ED, patients were more likely to have follow-up visits and had better quality of life, a care plan, and fewer asthma symptoms.

 Moreover, because Taiwan has very accessible healthcare, patients who trust their physician can substitute ambulatory care for ED visits. Christakis et al found no statistical correlation between COC and the risk of asthma-related ED visits in children with asthma in a health maintenance organization, but did find one for those with Medicaid.

 This finding supports the principle that universal insurance coverage and low co-payments of the NHI in Taiwan eliminate the economic barriers to high COC for children with asthma.

In terms of the number of asthma-related ED visits, we used the second component of hurdle regression, which focused on patients who had at least 1 asthma ED visit. All previous studies analyzed the number of asthma ED visits for all patients, including those with no ED visits. Cree et al found that high continuity of asthma care could decrease the number of asthma ED visits by 63%.

 In Canada, increasing the asthma COCI by 0.1 decreased the number of asthma-related ED visits by 19% for asthma patients aged 12 to 17 years.

 The 2-part hurdle regression reflects the sequential decision-making process for healthcare services. The response variable of the first part is whether to use the ED or not—a decision more likely to be decided by the patient and family, and thus, more likely correlated with personal characteristics. The response variable of the second part is the number or cost of visits, which is more likely to be associated with the characteristics of the healthcare system.

Our research showed a significant correlation in the first part—that the continuity of ambulatory asthma care influenced whether newly diagnosed children with asthma in Taiwan went to the ED&mdash;but not in the second part, the number of asthma ED visits for users only. In most cases, inexperienced parents went to the ED when the first asthma attack happened; later, they were better able to control symptoms. Most of our patients who had at least 1 asthma ED visit went to the ED only once in the outcome period, which may have caused the statistical differences between COC groups to remain undetected.

The limitation to our study is common to secondary data analysis. The NHI database contains no information on the outcomes of medical examinations, so we were unable to directly measure the severity of asthma. Therefore, we controlled for the number of asthma outpatient visits,

 whether or not the patient made an asthma-related ED visit, and the total days of asthma hospitalization during the COC period, in order to diminish the effect of disease severity.

Our study has several advantages, however. First, selecting new patients as the study subjects helped avoid the effects of past disease history and care experience on later COC and medical use. Most previous studies included all patients, whether their asthma was ongoing or newly developed, except for a Korean study of patients aged 65 to 84 years with newly diagnosed asthma or other chronic diseases, and a Taiwanese study of the effect of COC on avoidable admission for patients with diabetes, which controlled for the effect of being a new patient.

 We chose new pediatric asthma patients as our study subjects to more specifically examine the relationship between COC and ED use. Second, using a longitudinal study design helped in determining the temporality of events. Van et al believed that COC and outcome may affect each other; worse outcome or low satisfaction may trigger the patient to seek the care of other physicians and further decrease COC.

 In this study, COC was measured in the first year and the outcome was measured in the second year, to construct clear temporality.

Third, using asthma outpatient visits instead of visits for all conditions to estimate COC can more accurately reflect the actual asthma care received. For instance, Chen and Cheng chose patients with diabetes to calculate the COCI for diabetes-related physician visits.

 They mentioned that, because of the lack of a compulsory referral system and the high number of outpatient visits in Taiwan, disease-specific COCI could more sensitively examine the relationship between COC and medical use. Therefore, in our study, only asthma-related outpatient visits were counted. Fourth, and finally, Van et al found that many factors influence the correlation between COC and outcomes, such as age and severity of disease, suggesting that these covariates should be controlled for.

 In our study, we included many covariates in the models to reduce the possible effect on our results of such factors as patient characteristics, enrollment in the P4P program, severity of disease, and care provider characteristics.

Articles

WEB EXCLUSIVE

Better continuity of ambulatory asthma care can reduce the risk of asthma-related emergency department visits for children with asthma in Taiwan.

Effects of Continuity of Care on Emergency Department Utilization in Children With Asthma

The CMS core conditions—acute myocardial infarction (AMI), congestive heart failure (CHF), and pneumonia&mdash;are a focus of hospital quality reporting and its value-based purchasing program. The study's purpose was to assess national trends of in-hospital mortality and resource utilization for these core measures.

A time series study using outcomes from the 5 yearly cycles of the Nationwide Inpatient Sample (2005-2009).

 test for independence (binary or categorical parameters) or 

test for a contrasted mean (continuous parameters) were used to identify parameters that changed significantly over time (in-hospital mortality, length of stay, cost, charges, severity of illness, diagnoses per case, procedures per case). Multiple logistic and linear regression models were used to identify factors associated with in-hospital deaths, hospital charges, and length of stay (LOS).

 In-hospital mortality decreased for AMI, CHF, and pneumonia. LOS was unchanged for CHF, but decreased for AMI and pneumonia. Average inflation-adjusted charges per case increased for all 3 conditions, while the average inflation-adjusted cost per case decreased for CHF and remained stable for AMI and pneumonia. The proportion of patients with extreme disability and extreme likelihood of dying, as defined by All-Patient-Refined Diagnosis Related Group, increased for all 3 diagnoses. The number of diagnoses and procedures were independently associated with LOS, cost, and charges for all 3 conditions.

Many measures of quality of inpatient care and resource utilization for CMS core conditions improved despite increases in patient complexity and risk of mortality. Further research is necessary to determine the exact causes of these improvements.

	It will be essential to review outcomes from the implementation of national programs to control healthcare costs. Our study provides the benchmark foundation from which to evaluate 3 core measures that have increased healthcare expenditures. 

		Acute myocardial infarction, congestive heart failure, and pneumonia, the 3 core measures discussed in this study, have seen an increase in patient complexity. 

		While charges have increased in the care of these patients, costs have remained stable. 

		Inpatient mortality and hospitalizations have decreased over time, suggesting that ambulatory care settings may be absorbing more of the care of these complex patients.

Cardiovascular disease (CVD) remains the number one disease affecting Americans and consumes the most healthcare dollars.

 The American Heart Association’s statistical update on heart disease and stroke reports that CVD outpaces all cancers in healthcare resource consumption and mortality.

 CVD consumed $316.4 billion in 2010 and causes 1 in 3 deaths per year. The major contributors to these figures are acute myocardial infarction (AMI) and congestive heart failure (CHF); 1 in 9 death certificates indicates heart failure as a cause of death.

Overall, pneumonia ranks as the fifth-most frequent cause of admission, but the second-most frequent cause for those aged less than 17 years and more than 85 years.

 There were 1.1 million discharges with the primary diagnosis of pneumonia in 2010.

 Pneumonia also accounts for 16.1 deaths per 100,000 people, consumes $10.5 billion annually, and contributes 2.9% to annual costs for all hospitalizations.

Given the impact of these 3 core diseases—AMI, CHF, and pneumonia&mdash;on our public health, and the financial costs to our healthcare system, major efforts have been focused on improving their outcomes. In 2002, hospitals accredited by the Joint Commission started collecting data on standardized care process measures to improve care. In 2004, the Joint Commission and CMS synergized their efforts and developed the National Hospital Quality Measures, standard measure sets common to both organizations. Currently, there are 12 broad classifications covering major areas of potential improvement including AMI, CHF, and pneumonia.

In 2012, CMS continued efforts to improve quality care and to control costs by introducing the hospital value-based purchasing (VBP) program, funded through a 1% reduction in diagnosis-related group (DRG) payments for discharges. Of this 1% portion, 70% can be “earned back” based on adherence to the approved clinical processes of care for selected patient outcomes. The core measures of AMI, CHF, and pneumonia were selected for the fiscal year 2013 hospital VBP program.

 Hospitals responded with several initiatives (ie, broader use of new care delivery systems, such as telemedicine and hospital-based disease management programs) to reduce costs, hospital readmissions, length of stay (LOS), and inpatient mortality.

The goal of this paper, then, was to establish a base of comparison for VBP by examining—prior to its implementation&mdash;national trends for inpatient mortality, LOS, cost, and charges related to these 3 core measures in a population of all-payers and all patients 18 years or older.

For this study, we used the National Inpatient Sample (NIS) database, an all-payer database of hospital discharges maintained as part of the Healthcare Cost and Utilization Project by the Agency for Healthcare Research and Quality. The NIS approximates a 20% stratified sample of community hospitals in the United States. The sampling frame in each of the study’s surveys was a sample of hospitals that represented 90% to 95% of all hospital discharges for the particular year. Specifically, between 2005 and 2009, each NIS cycle included information on approximately 8 million discharges from approximately 1000 hospitals located in 37 (in 2005) to 44 (in 2009) states. Each NIS record represents a single hospital discharge and includes a unique identifier; basic demographics for the patient; admission and disposition type; up to 15 primary and secondary diagnoses; a list of potential comorbidities; up to 15 primary and secondary procedures; expected primary insurance payer; total hospital charges excluding physician and other professional services; LOS; type of hospital (teaching/nonteaching), bed size, and location; and mean cost-to-charge ratio.

Furthermore, the severity of illness scores defined by the All Patient Refined Diagnosis-Related Groups (APR-DRGs)

 were evaluated and reported together with the respective risk of mortality scores. The APR-DRGs expand the basic DRG structure by adding 4 severity-of-illness and 4 risks-of-mortality subclasses to each DRG: for severity of illness, minor, moderate, major, or extreme loss of function; for mortality risk, minor, moderate, major, or extreme likelihood of dying. The APR-DRG severity scores of 3 and 4 (major/extreme loss of function), likelihood-of-dying scores of 3 and 4 (major/extreme likelihood of dying), and the number of diagnoses were used as markers for the severity and complexity of cases, respectively. The outcomes included in this study were in-hospital mortality and resource utilization, which included the length of inpatient stay, as well as the total costs and charges per discharge. These outcomes were assessed in each calendar year and in patient subgroups as described below.

We included all adult (18 years or older) discharges from the NIS between 2005 and 2009 with the principal diagnoses of AMI, CHF, or pneumonia. The diagnoses and procedures were identified using the 

International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM

 codes for each principal diagnosis used for the study.

We obtained national estimates for the total number of hospitalizations for each of the 3 diagnoses and total resource utilization parameters by calendar year, by totaling individual discharge sampling weights (which are the inverse of probability selection for the sample from the entire population). Since not all participating hospitals had reported their average cost-to-charge ratios, the national estimates of total costs were reweighted to account for all discharges where cost estimates were missing.

 All charges and costs were adjusted to the year 2009 using medical consumer price indices for the study years.

All available clinical and socioeconomic parameters were compared for change between years using a stratum-specific 

 test for independence (for binary or categorical parameters) or a 

test for a contrasted mean (for continuous parameters). Additionally, resource utilization was studied separately for 2 age groups: 18 to 64 years and 65 years or older. Taylor series linearization was used to account for the stratum units used in NIS to sample hospitals based on geographic region, control, location, teaching status, and bed size.

Multiple logistic models were used to identify factors associated with in-hospital deaths while accounting for potential confounders. Similarly, factors that impacted hospital charges and LOS were assessed using multiple linear regressions after logarithmic transformation of the respective outcomes. Coefficients from these models were exponentiated to yield a percentage change in the outcomes associated with each predictor. All 

 values of .05 or less were considered to be statistically significant.

All analyses were run with SAS version 9.1 (SAS Institute, Cary, North Carolina) and SUDAAN version 10.0 (RTI International, Research Triangle Park, North Carolina). The study was reviewed and approved as meeting the exempt criteria by the Inova Health System Institutional Review Board.

Overall, 33.3 million hospitalizations were included in the 5 years of NIS, ranging from 6.54 to 6.84 million yearly, representing 164.14 million hospitalizations nationwide. Of those 33.3 million, the overall numbers of pneumonia, AMI, and CHF cases decreased. The decrease for pneumonia was from 1.17 million in 2005 (3.64% of all hospitalizations) to 1.01 million in 2009 (3.04%); for AMI, 662,000 in 2005 (2.07%) to 633,000 in 2009 (1.91%); and for CHF, 1.09 million in 2005 (3.39%) to 1.02 million (3.09%) in 2009 (

Over time, approximately 53% of patients were female; 77% were white. Reported ethnicities for the remainder, were, on average, 10.5% black, 7.5% Hispanic, 2% Asian, and less than 1% Native American. Sixty percent of the patients were 65 years or older; of these, the majority were Medicare patients, although the proportion of Medicare patients in the entire pneumonia cohort decreased from 70.22% in 2005 to 63.68% in 2009 (

The in-hospital mortality for pneumonia decreased from 4.56% in 2005 to 3.98% in 2009 (

 <.0001). The LOS decreased from 5.80 days in 2005 to 5.55 days in 2009 (

). The average charges per case increased from $25,543 in 2005 to $30,256 in 2009 (

 <.001), while the average cost per case did not change (Table 4 and 

Over the 5 years, the proportion of cases with a major loss of function increased from 36.08% to 39.74%; those with an extreme loss of function increased from 7.08% to 12.02%. Also, the proportion of patients with a major and extreme likelihood of dying increased: respectively, from 16.40% in 2005 to 24.93% in 2009, and from 4.38% in 2005 to 8.62% in 2009 (

 <.0001). The number of procedures per case increased as well, from 0.68 in 2005 to 0.75 in 2009 (

 for the individual procedures [eAppendices available at 

After multivariate analysis, major or extreme disability and the number of procedures were independent predictors of in-hospital mortality. Major or extreme disability, likelihood of death, number of diagnoses, and number of procedures were independently associated with LOS, charges, and cost (eAppendix Table 2)

The majority of patients admitted with acute myocardial infarction (AMI) were male; in 2005 and 2009, 40.72% and 39.51% were female, respectively (Table 3). The majority, 77.94%, were white. Blacks, Hispanics, Asians, and Native Americans represented an average of 8.77%, 7.16%, 2.07%, and 0.62% of patients, respectively. Fifty-eight percent of patients were aged 65 years or more, which closely paralleled the 57% of the sample who were insured by Medicare, the largest payer group. The figure, however, decreased over time, from 59.17% in 2005 to 55.94% in 2009 (

In-hospital mortality also decreased, from 6.81% in 2005 to 5.56% in 2009 (

 <.0001). Average LOS decreased, too, from 5.23 days in 2005 with to 4.95 days in 2009 (

 = .001) (Table 4). The average charges per case increased from $54,546 in 2005 to $62,917 in 2009 (

 = .0004), while the average cost per case did not change.

The proportion of patients with major or extreme loss of function increased over the 5-year period, from 24.39% in 2005 to 27.93% in 2009, and from 13.52% in 2005 to 16.99% in 2009, respectively (

 <.0001). The proportion of patients with a major or extreme likelihood of dying increased also, from 27.18% in 2005 to 27.68% in 2009, and 11.93% in 2005 to 15.95% in 2009, respectively (

 <.0001). The total number of diagnoses per case also increased from 8.19 in 2005 to 10.51 in 2009 (

 <.0001). The number of procedures per case increased from 3.93 in 2005 to 4.89 in 2009 (

 <.0001) (Table 4, Table 5 by the age groups; eAppendix Table 1 for the individual procedures).

After multivariate analysis, major or extreme disability were the only independent predictors of in-hospital mortality. The number of diagnoses per case and number of procedures per case were inversely proportional to in-hospital mortality. Major or extreme disability, risk of mortality, number of diagnoses, and number of procedures were independently associated with LOS, charges, and cost (

Female patients represented 51% of all cases of congestive heart failure (CHF) (Table 3). The majority (70.18%) of patients were white, while blacks, Hispanics, Asians, and Native Americans represented 18.71%, 8.01%, 1.68%, and 0.57% of patients, respectively. The overwhelming majority of patients (73%, a figure that remained stable throughout the 5-year period) were 65 years or older. Medicare was the largest payer group, but with a decreasing trend (dropping from 76.63% in 2005 to 74.09% in 2009; 

In-hospital mortality decreased from 3.78% in 2005 to 3.21% in 2009 (

 <.0001). The overall LOS remained stable, 5.37 days on average over the 5-year time period. Average charges increased from $31,028 in 2005 to $35,551 in 2009 (

 = .0045) in inflation-adjusted dollars. In contrast, the average cost per case decreased ($11,006 in 2005 to $10,377 in 2009; 

The proportion of patients with major or extreme loss of function increased from 34.07% in 2005 to 44.89% in 2009, and from 5.85% in 2005 to 9.68% in 2009, respectively. A similar increase was found for the proportion of admissions with major or extreme risk of dying: from 23.86% in 2005 to 33.28% in 2009, and from 6.90% in 2005 to 11.52% in 2009, respectively. The complexity of illness, reflected by the total number of diagnoses per case, increased from 8.98 diagnoses per case in 2005 to 12.22 diagnoses per case in 2009 (

 <.0001). However, the number of procedures per case did not change (Table 4, Table 5 by the age groups, eAppendix Table 1 for the individual procedures).

After adjustment for major socioeconomic confounders, the independent predictors of in-hospital mortality were found to be major or extreme disability, mortality risk, and the number of procedures per case. The number of diagnoses per case was inversely proportional to in-hospital mortality. Major or extreme disability, mortality risk, number of diagnoses, and number of procedures were independently associated with LOS, charges, and cost (eAppendix Table 2). For illustrative purposes, using eAppendix Table 2, for individuals admitted to the hospitals included in the NIS database with the primary diagnosis of CHF, each additional diagnosis on the record was associated with a 2.43% longer length of stay, 1.61% higher charges, 1.41% higher cost of treatment, and 0.975 lower odds of in-hospital death.

Our large analysis demonstrates that in the United States between 2005 and 2009, many measures of inpatient quality of care and resource utilization for pneumonia, AMI, and CHF improved—despite a universal increase in the proportion of patients with major or extreme severity of illness and likelihood of dying. In addition, charges increased but the cost of care actually decreased. These findings indicate that the push to improve quality through mandatory reporting may be contributing to improving patient outcomes and reducing cost of care.

In-hospital mortality improved for these 3 high-impact diagnoses, even after adjustment for sociodemographic changes and higher patient acuity. During the same time period, hospital LOS decreased for patients with pneumonia and AMI, and remained stable for those with CHF. The exact causes of improvement in in-hospital mortality cannot be determined from our analysis. However, LOS may have been partially impacted by the significant increase in the number of patients discharged to another hospital or care facility, or to home with healthcare assistance. Indeed, for pneumonia, discharge disposition to another hospital increased from 1.90% to 2.28% (

 = .0017); and for CHF discharge disposition to home with assistance increased from 16.31% to 18.94% (

 <.0001). These changes in discharge disposition have become common as cost-saving measures instituted under the case management model.

 Therefore, simply reporting of length of stay may be somewhat misleading and create a false sense of a decreased LOS.

Other events, occurring simultaneously with the CMS initiatives, which may have had a direct impact on the improvement in the care measures, include the increased use of outpatient services, telemedicine, and evidence-based care guidelines. The impact of diagnostic coding must also be taken into account.

 Another factor may have been patients’ increased use of ambulatory care settings to access preventive care, rather than relying on the acute care provided by hospitals’ emergency departments (EDs). Also, over the past several years, ambulatory care walk-in clinics have become more widespread, allowing patients to obtain care at lower costs. In addition, the earlier discharge patterns of hospitals have pushed ambulatory care, home health, and rehabilitation centers to absorb the care of patients at earlier stages of recovery.

Another likely reason for the reported improvements of the care measures has been the rigorous implementation of evidenced-based guidelines to deliver appropriate and timely care. The American Heart Association developed the “Get with the Guideline Campaign” to improve care for patients with heart disease, specifically CHF and AMI. Recent studies on management strategies for CHF and AMI have found that use of these guidelines resulted in lower risk-adjusted mortality.

The 3 core measures of acute myocardial infarction, congestive heart failure, and pneumonia are the leading causes of hospital admissions and expenditures. Our study sets the benchmark foundation for outcome evaluations of CMS’s value-based purchasing program and the Affordable Care Act.

Outcomes Trends for Acute Myocardial Infarction, Congestive Heart Failure, and Pneumonia, 2005-2009

To examine factors that influenced continuing care participation in patients with diabetes and factors contributing to interrupted participation for patients with diabetes enrolled in a diabetes pay-for-performance (P4P) program in Taiwan.

Data were obtained from Taiwan’s National Health Insurance Research Database on patients with a new confirmed diagnosis of type 2 diabetes during 2001 to 2008, selected as 1:1 propensity score-matched P4P program enrollees and nonenrollees (totaling 396,830). Logistic regression was performed to analyze factors associated with continuing care participation and with interrupted P4P program participation after enrollment.

Among the patients with diabetes, P4P program enrollees were 4.27 times (95% CI, 4.19-4.36) more likely to participate in continuing care than nonenrollees. Factors affecting the participation of patients with diabetes in continuing care included P4P program enrollment status, personal characteristics, health status, characteristics of the main physician, and characteristics of the main healthcare organization. Interruption of P4P program participation occurred in 78,759 (44.33%) of the enrolled patients with diabetes and was correlated with male gender, younger age (<35 years), residence in areas of highest urbanization, greater severity of diabetes complications, presence of catastrophic illness/injury, high service volume at the site of the main physician, older age (≥55 years) of the main physician, having a regional or private hospital as the main healthcare organization, and change of physician.

Taiwan’s diabetes P4P program increased continuing care participation in patients with diabetes. The rate of interruption of P4P program participation among enrolled patients with diabetes, at 44.33%, should be a focus of improvement for Taiwan’s health authorities.

	This study examined factors associated with continuing care participation in patients with diabetes and with interrupted participation for those patients enrolled in a diabetes pay-for-performance (P4P) program.

		Taiwan’s P4P program increased the rate of continuing care in patients with diabetes.

		Factors that correlated with the participation of patients with diabetes in continuing care included the patient’s P4P program enrollment status, personal characteristics, health status, characteristics of the main physician, and characteristics of the main healthcare organization.

		Factors associated with interrupted participation by patients with diabetes in the P4P program after enrollment included male gender, younger age (<35 years), residence in areas of the highest urbanization level, greater severity of diabetes complications, presence of catastrophic illness or injury, high service volume at the site of the main physician, older age (≥55 years) of the main physician, having a regional or private hospital as the main healthcare organization, and change of physician.

In 2014, an estimated 415 million individuals worldwide had diabetes, and the number of individuals with diabetes is expected to rise to 642 million by 2040.

 Previous studies have shown the challenges of diabetes prevention and management,

 but better continuity of care has been associated with improved medication compliance

 and healthcare expenses for patients with diabetes.

Pay-for-performance (P4P) is a payment scheme that rewards healthcare providers for providing high-quality continuing care services.

 In many countries—such as the United States, Australia, Germany, and the United Kingdom&mdash;P4P programs are a priority policy for promoting more efficient use of healthcare resources and enhancing the quality of care.

In Taiwan, 7% of the population (1,631,599 individuals) had diabetes as of 2012, which accounts for 3.8% of the National Health Insurance program’s total annual healthcare expenditure.

 To improve the prevention and treatment of this disease, Taiwan launched a diabetes P4P program in November 2001. As part of this program, a team of care providers, consisting of physicians, nurses, nutritionists, and other healthcare professionals work together to provide examination, testing, health education, and follow-up services in an effort to reduce the occurrence of diabetic complications and comorbidities.

 Healthcare organizations participating in the program must perform specific diagnosis and management tasks, including medical history, physical examination, laboratory evaluation, evaluation of the management plan, and diabetes self-management education. Patients who are enrolled in the diabetes P4P program must undergo a complete annual evaluation of their disease. If healthcare services have been provided as required by the program, the healthcare organization will receive a bonus payment: a value-added physician examination fee and a case management fee. For Taiwan’s diabetes P4P program, the amount of the bonus payment is calculated using a point system, and the case management fee includes: 400 points awarded for the initial physician visit (once per patient), 200 points for each follow-up visit (once every 3 months), and 800 points for the annual evaluation (once per year),

 where 1 point is worth around 1 New Taiwan Dollar (NTD) (30 NTD = US$1). In 2009, 27.56% (214,340) of Taiwan’s patients with diabetes were enrolled in the program.

Previous assessments of Taiwan’s diabetes P4P program have found it to effectively increase clinical guideline adherence

 and patient satisfaction with the quality of care,

 as well as decrease inpatient care utilization.

 In this study, we used data from the National Health Insurance database to examine factors that influence continuing care participation in patients with diabetes who either are or are not enrolled in the P4P program, and factors that affect whether interrupted participation in the P4P program occurs in patients with diabetes who are enrolled in the program. Both Taiwan and the United States provide doctors with financial incentives to enhance medical quality, and some payments are based on quality indicators. However, P4P in Taiwan is implemented by a single public insurer (National Health Insurance), whereas P4P in the United States is practiced under multiple medical insurance systems. The study results could be a policy reference for comparisons among different health insurance systems.

Previous studies have adopted variables such as gender, age, monthly salary, comorbidities,

 Diabetes Complications Severity Index (DCSI) score, and hospital accreditation level

 when evaluating diabetes P4P programs. On the basis of logical inference, we searched for possible variables that might account for patients joining or withdrawing from a diabetes P4P program; the topic has rarely been studied.

In this retrospective cohort study, analysis was performed with secondary data obtained from the National Health Insurance Research Database maintained by Taiwan’s National Health Research Institutes. The study population consisted of all patients with a confirmed diagnosis of type 2 diabetes from 2001 through 2008. Patients with diabetes were defined as individuals with 1 hospitalization, or 3 or more outpatient visits within 365 days, in which the primary or secondary diagnosis was diabetes (

 code 648.0), or diabetes that is gestational (648.8) or neonatal (775.1), or had abnormal glucose tolerance (790.2) were excluded from the study. Patients with diabetes whose outpatient records had ever been designated with the specific treatment code “E4” (for the diabetes P4P program) were defined as P4P program enrollees (N = 198,420). All study subjects were followed until the end of 2009.

Propensity score matching was performed on the basis of personal characteristics (gender, age, monthly salary, and residence urbanization level); health status (comorbidity, severity of diabetes complications, and presence or absence of catastrophic illness or injury); and characteristics of the healthcare organization (level and ownership type) in order to match P4P enrollees 1:1 with nonenrollees among patients with diabetes. This resulted in a sample of 396,830 patients for analysis. For the analysis of factors related to interrupted participation in the diabetes P4P program after enrollment, we excluded enrollees, totaling 20,744, who did not reach the time point for their annual evaluation (29 full weeks after enrollment), leaving 177,676 enrollees for the analysis. To encourage medical teams to improve their monitoring of and continual care for patients, the National Health Insurance Administration in Taiwan has established payment regulations for its diabetes P4P program. Patients with diabetes who have been in the P4P program for 29 weeks can participate in yearly evaluations, and the doctor can receive the annual case management fee for each patient.

The dependent variables examined in this study included whether participation in the diabetes P4P program was interrupted and whether continuing care participation was maintained. The independent variables analyzed included personal characteristics

 characteristics of the main physician seen (age and annual service volume), and characteristics of the main healthcare organization utilized. The National Health Insurance Administration in Taiwan has identified 30 types of severe illness or injury as catastrophic illnesses or injuries (eg, malignant neoplasm, type 1 diabetes, chronic renal failure, cerebrovascular disease, rare disease).

Further details on the variables are as follows: 1) continuing care participation was defined as having at least 1 diabetes-related physician visit every 3 months and at least 4 such visits per year following a confirmed diabetes diagnosis for patients not enrolled in the P4P program, or following P4P program enrollment for enrollees. Otherwise, patients were considered to lack continuing care participation. 2) Uninterrupted diabetes P4P program participation was defined as having regular diabetes visits during the year following P4P program enrollment, and then undergoing the annual evaluation of diabetes disease (prescription code P1409C). Enrolled patients who failed to complete the first annual diabetes evaluation were considered to have interrupted participation in the diabetes P4P program. 3) Monthly salaries were taken to be the monthly salary amounts used in the determination of National Health Insurance premiums. 4) The urbanization level of the residence area was designated as 1 of 7 levels, with level 1 corresponding to the highest level of urbanization and level 7 to the lowest. 5) Comorbidity was assessed using the Deyo modification of the Charlson comorbidity score. Each patient’s 

 primary and secondary diagnosis codes were converted to weighted numerical scores, which were then summed to give the patient’s Charlson comorbidity index (CCI).

6) The severity of diabetes complications was assessed using the categories of diabetes complications described by Young et al (retinopathy, nephropathy, neuropathy, stroke, cardiovascular disease, peripheral vascular disease, and metabolic complications). Each patient’s 

 primary and secondary diagnosis codes were converted to weighted numerical scores, which were then summed to give the patient’s DCSI.

 7) Physician service volume was computed as the annual service volume of the patient’s main physician, categorized by the quartile method as high (≥75%), medium (≥25% and <75%), or low (<25%). 8) The characteristics of the main healthcare organization utilized were classified in terms of its level and ownership type. 9) Low-income household status was defined as belonging to a household in which the average monthly income per person falls below the lowest living index, which is 60% of the living expenditure per person in the previous year in the household’s local area.

This study was approved by the Institutional Review Board of China Medical University and Hospital (IRB No. 20130326C).

First, descriptive statistics were used to examine the distribution of different variables—personal characteristics, health status, characteristics of the main physician, characteristics of the main healthcare organization, and continuing care participation&mdash;in the study participants, who were distinguished by whether they were enrolled in the P4P program and whether interruption occurred in their participation in the P4P program after enrollment. The distributions were expressed as percentages and means.

 test were used to examine each variable’s association with continuing care participation in diabetes P4P program enrollees and nonenrollees, as well as association with interrupted P4P program participation in the enrollees. Logistic regression analysis was performed to examine factors that influenced continuing care participation in patients with diabetes.

Finally, logistic regression was employed to examine factors that influenced whether interruption occurred in the participation of patients with diabetes in the P4P program after enrollment. In this study, all statistical analyses were performed using SAS 9.3 software (SAS Institute, Cary, North Carolina) for Windows. A 

value less than .05 was considered statistically significant.

Among the patients with diabetes analyzed in this study, the P4P program enrollees and the propensity score—matched nonenrollees were not significantly different in any of the matched variables, as shown in 

. The majority of the program enrollees participated in continuing care, defined as having at least 1 diabetes-related physician visit every 3 months (71.82%, vs 27.67% who did not), as shown in 

. With respect to personal characteristics, all patients with diabetes characterized by female gender (49.99%), aged 35 to 64 years (50.37%-52.35%), a monthly salary of NTD17,281-22,800 (50.60%), and residence in urbanization level 4 or 5 areas (50.77%) had higher rates of continuing care participation. As for health status, a CCI of 2 (60.19%), a DCSI of 1 (65.13%), and the absence of catastrophic illness or injury (50.88%) corresponded to higher rates of continuing care participation in the all patients with diabetes. Higher rates of continuing care were also observed in patients whose main physicians were aged 45 to 54 years (63.86%) or had a high service volume (70.67%), and in patients whose main healthcare organization was a district hospital (50.63%), a clinic (50.09%), or a public hospital (50.53%).

Factors Affecting Continuing Care Participation in Patients With Diabetes

Logistic regression analysis revealed that patients with diabetes who are enrolled in the P4P program were more likely to participate in continuing care (having at least 1 diabetes-related physician visit every 3 months) than those not enrolled in the program by a factor of 4.27 (95% CI, 4.19-4.36), as shown in 

. Increased odds of continuing care participation were associated with all patient characteristics of female gender, aged 35 to 74 years (odds ratio [OR], 1.05-1.40), and residence in urbanization level 2 or 3 areas (OR, 1.07). With respect to health status, any patients with diabetes with a CCI of 2 (OR, 1.59), with a DCSI of 1 (OR, 1.58), or without catastrophic illness or injury were more likely to participate in continuing care. As for physician and health organization characteristics, higher odds of continuing care participation were found for any patients with diabetes whose care was provided mainly by a physician aged 35 to 44 years (OR, 1.15) or with a high service volume (OR, 3.01), and mainly by a clinic (OR, 1.12) or a public hospital. In particular, physician service volume was directly correlated with the odds of continuing care participation for all patients with diabetes.

Comparison of Characteristics Between Patients With and Without Interruption of Diabetes P4P Program Participation After Enrollment

Of the patients with diabetes who were enrolled in the diabetes P4P program, 78,759 (44.33%) had interrupted participation in the program, as seen in 

. With respect to personal characteristics, higher proportions of enrollees who were male (45.61%), were aged under 35 years (52.35%), had a monthly salary of NTD22,801 to NTD28,800 (46.44%), or resided in urbanization level 1 areas (46.41%) had an interruption of program participation. Among the health status characteristics, a CCI ≥10 (47.65%), a DCSI ≥3 (50.09%), and the presence of catastrophic illness or injury (47.31%) corresponded to higher rates of interrupted program participation. Interrupted program participation also occurred at higher rates in enrollees whose main physician was aged under 35 years (48.60%) or had a high service volume (47.68%), whose main healthcare organization was a regional hospital (47.24%), and who had a change of physician (72.74%).

Factors Associated With Interrupted Diabetes P4P Program Participation in Enrolled Patients With Diabetes

Factors related to interruption of diabetes P4P program participation in enrolled patients with diabetes were analyzed by logistic regression, with the results shown in Table 4. Personal characteristics of the patients with diabetes that were associated with a greater likelihood of interrupted program participation were male (OR, 1.14), aged under 35 years, had a monthly salary of NTD ≤28,800, and residence in urbanization level 1 areas. With respect to health status, enrollees with a DCSI ≥3 (OR, 1.26) or with catastrophic illness or injury (OR, 1.24) were more likely to have an interruption of program participation. After we controlled for all relevant variables in the analysis model, the results changed so for enrollees whose main physician was 55 years or older, there were increased odds of interrupted program participation (OR, 1.19; 95% CI, 1.14-1.25); additional associated factors included having a high service volume (OR, 1.56), having a regional hospital (OR, 1.22) or a private hospital (OR, 1.02) as the main healthcare organization, and having changed physicians (OR, 7.10). In particular, there was a direct correlation between physician service volume and the odds of interrupted diabetes P4P program participation in enrolled patients with diabetes.

In this study, we showed that patients with diabetes who were enrolled in Taiwan’s diabetes P4P program were more likely to engage in continuing care (ie, have at least 1 diabetes-related physician visit every 3 months) than their nonenrolled counterparts (Table 3). This finding is in agreement with a previous study conducted in Taiwan by Lee et al (2010).

 Those authors found that patients with diabetes in the P4P program had a larger increase in the number of diabetes-specific physician visits after program enrollment (annual average increasing from 3.8 times to 6.4 times) than nonenrolled patients in the same period (from 3.5 to 3.6 times).

One of the factors that has been associated with differences in patients’ continuing care participation is gender. Chang et al (2005) showed that female patients with diabetes were more likely than their male counterparts to take their medicine regularly and practice self-care (exercise, dietary control, and weight control).

 observed a higher number of physician visits for female than male patients with diabetes. Previous research also found differing views on health-related issues between men and women and the greater social and cultural acceptability of women being weak or unhealthy.

Analyzing factors associated with continuing care participation in patients with diabetes and with interrupted participation by patients enrolled in a diabetes pay-for-performance program.

January 2016

Effects of Continuity of Care on Emergency Department Utilization in Children With Asthma

Outcomes Trends for Acute Myocardial Infarction, Congestive Heart Failure, and Pneumonia, 2005-2009

Factors Related to Continuing Care and Interruption of P4P Program Participation in Patients With Diabetes

Oral Anticoagulant Discontinuation in Patients With Nonvalvular Atrial Fibrillation

Value-Based Insurance Designs in the Treatment of Mental Health Disorders

Improving Engagement in Employer-Sponsored Weight Management Programs

Care Pathways in US Healthcare Settings: Current Successes and Limitations, and Future Challenges

Does Distance Modify the Effect of Self-Testing in Oral Anticoagulation?

Impact of a Scalable Care Transitions Program for Readmission Avoidance

The Introduction of Generic Risperidone in Medicare Part D