 To examine the costs of hospitalizations related to heart failure (HF) among patients aged 18 to 64 years by diagnosis status.

There are few reports on HF-related hospitalization costs in patients aged 18 to 64 years, although this group has had increased HF hospitalization rates in recent decades.

Using the 2005 MarketScan Commercial Claims and Encounters inpatient data set, we identified 23,216 hospitalizations for which HF was the primary or secondary diagnosis among patients aged 18 to 64 years who had a noncapitated health insurance plan. We used multivariate regression to analyze the association between patient characteristics and both hospitalization costs and length of stay (LOS).

For the entire population, the mean cost of hospitalization was $23,077; the cost was higher when HF was a secondary rather than the primary diagnosis ($25,325 vs $17,654; 

 <.001). After controlling for covariates, hospitalizations with HF as a secondary diagnosis resulted in $3944 higher costs than those with HF as the primary diagnosis (

 <.001). Among those with HF as a secondary diagnosis, the cost of hospitalization with the primary diagnosis of ischemic heart disease was $14,989 higher than it was when the primary diagnosis was noncardiovascular disease and nonrespiratory conditions (

 <.001). However, patients in the latter group had a longer LOS.

Information on the costs of HF-related hospitalizations can be used as inputs in economic evaluations such as cost-effectiveness analyses and as references for policy makers in making resource allocation decisions.

The costs of 23,216 hospitalizations related to heart failure (HF) among patients aged 18 to 64 years were analyzed.

The mean cost of HF-related hospitalizations was $23,077 and was higher when HF was a secondary rather than the primary diagnosis. 

For those with a secondary diagnosis of HF and the primary diagnosis of ischemic heart disease, the costs of hospitalizations were $14,989 higher than they were when the primary diagnosis was noncardiovascular disease and nonrespiratory conditions. 

Among all the categories of primary diagnoses, the primary diagnosis of noncardiovascular disease and nonrespiratory conditions resulted in the longest length of stay.

	In the United States, rates of hospitalization for heart failure (HF) increased substantially in the past 3 decades, and HF has become an epidemic.

 Although HF affects adults 65 years or older more than those below age 65 years, the younger group had a higher relative increase in hospitalization rates for HF, regardless of sex, than did the older group.

 In addition, the percent increase was greater for hospitalizations with HF as a secondary diagnosis than it was for hospitalizations with HF as the primary diagnosis.

 Nationally, approximately 70% of HF-related hospitalizations listed HF as a secondary diagnosis.

 From a public health perspective, promoting prevention and improving medical management based on an understanding of the costs for HF-related hospitalizations in people aged 18 to 64 years could have a large payoff in terms of containing the ever-increasing healthcare costs.

	For 2007, the medical cost of HF in the United States for those aged 65 years or older was estimated at $30.2 billion.

 The hospitalization cost for this age group was estimated at $20.1 billion in 2009.

 The average cost for an HF hospitalization more than doubled from about $7000 in the 1990s to $18,086 in recent years.

 On the national level, HF has an extraordinary impact, consuming 1% to 2% of the total healthcare resources in industrialized countries, including the United States, with this percentage likely to increase in the future.

 Moreover, HF is the fastest-growing cardiac disease entity in the United States, affecting 2% of the population, with nearly 1 million hospitalizations for acute decompensated congestive HF occurring annually.

	Although many researchers have investigated the medical costs related to HF, nearly all the cost studies have been for the population 65 years or older because of the high hospitalization rates for HF in this age group.

 Correspondingly, information is lacking on the costs of HF in adults under 65 years. Furthermore, although many previous studies had focused on hospitalization costs,

 the costs by detailed diagnosis codes and diagnosis status have not been well examined. Because the types of HF vary considerably and at least 70% of HF diagnoses are listed as a secondary diagnosis, it would be helpful to have cost analyses for this disease that consider these characteristics, especially as the information obtained could influence the cost-effectiveness of HF management programs. Thus, we investigated HF-related hospitalization costs among patients aged 18 to 64 years by diagnosis status.

	We identified HF-related hospitalizations from the 2005 MarketScan Commercial Claims and Encounters inpatient data set, which in that year had data on more than 1 million inpatient hospitalizations for privately insured patients 64 years or younger. The data contain comprehensive, highquality coding of medical services.

 Researchers have previously used this database for medical cost analyses of cardiovascular diseases (CVDs).

	We used the primary diagnosis and 14 secondary diagnoses to identify hospitalizations for patients aged 18 to 64 years with a diagnosis of HF; our analysis was restricted to the 

International Classification of Diseases, Ninth Revision 

) codes listed in the American College of Cardiology/American Heart Association HF performance measures (

). We identified 23,216 HF-related hospitalizations for patients with a noncapitated health insurance plan and with hospitalization costs between the 1st and 99th percentiles. We excluded the hospitalizations of patients with a capitated health insurance plan because their costs did not reflect the medical services they received; excluding hospitalizations with a cost below the 1st or above the 99th percentile reduced the influences of extreme values on the cost estimates. We further created 2 subsamples for the cost analyses: (1) HF as the primary diagnosis and (2) HF as a secondary diagnosis. Hospitalizations with both the primary diagnosis and a secondary diagnosis of HF were included in the first subsample.

	For hospitalizations with a secondary diagnosis of HF, we identified the primary diagnoses of (1) ischemic heart disease (IHD) (

 codes 410-414); (2) other diseases of the heart (

 codes 392-398, 402, 404, 415, 416, 420-427, and 429); (3) other CVDs (

 codes 390-459, excluding the codes for IHD and other diseases of the heart); (4) respiratory disease (

 codes 460-519), including pneumonia and chronic obstructive pulmonary disease; and (5) non-CVD and nonrespiratory conditions.

Outcome Measures and Other Control Variables

	The main outcome for this analysis, hospitalization costs (total payments to providers, not charges), represented the total costs per hospitalization, which included the costs for physician services, all diagnostic tests, therapeutics, supplies, and room fees during the stay. We excluded the hospitalizations with a cost below $1060 (cutoff point of the 1st percentile) and above $208,507 (cutoff point of the 99th percentile).

	Length of stay (LOS) was a second outcome measure. Because LOS is a major factor influencing the cost, reducing it might have important implications for managing HF and for cost containment.

 The LOS was categorized as 1 for hospital stays of fewer than 7 days and as 2 for hospital stays of 7 or more days. We used 7 days as the cutoff point because the average number of days for an HF-related hospitalization in the United States is about 6 to 8 days.

	We included age, sex, urbanization, and the Charlson Comorbidity Index as control variables. We used the Charlson Comorbidity Index, which measures the likelihood of death or serious disability in the subsequent year by diagnosis codes for up to 18 different diseases,

 to control for the effects of disease severity and comorbidities on the costs.

	Statistical analyses were performed using SAS version 9.2 (SAS Institute Inc, Cary, NC). Significance was assumed when the 2-tailed probability value was less than .001. For cost comparisons between hospitalizations with the primary diagnosis of HF and those with a secondary diagnosis of HF, the Wilcoxon 2-sample test was used. We used regression models to conduct multivariate analysis of the costs and LOS. Medical cost data were usually skewed and not normally distributed; thus, many researchers have used log-transformation in cost analyses. Many other studies, however, did not use the transformation because of the difficulties in interpreting the results.

 We excluded hospitalizations with total costs below the 1st or above the 99th percentile to avoid the effects of outliers, and our large sample sizes also made the analysis relatively robust to the violations of normality. Therefore, we did not transform the costs in our analysis. Because the study samples consisted of hospitalizations rather than patients, some patients might have had multiple hospitalizations during the year. To account for the possibility of the repeated measures in the analysis, we used the PROC MIXED procedure to estimate the impact of patient characteristics and other covariates on hospitalization costs, and we used the PROC GENMOD procedure to estimate the impact of these variables on LOS.

	For all 23,216 HF-related hospitalizations, the mean cost was $23,077; for the 6806 hospitalizations (29%) with HF listed as the primary diagnosis, the mean cost was $17,654, and for the 16,410 hospitalizations (71%) with HF listed as a secondary diagnosis, the mean cost was $25,325 (

). Hospitalized men incurred higher costs than hospitalized women, and among the 6 groups formed by combining sex and diagnosis status, men with HF as a secondary diagnosis had the highest cost ($27,277), whereas women with HF as a primary diagnosis had the lowest cost ($16,058). The costs across age groups varied slightly regardless of diagnosis status.

	The regression results indicated that sex significantly influenced costs, but age did not (

). In the entire sample, hospitalized men incurred costs that were $4317 (P <.001) higher than those of hospitalized women. Regional variations in the hospitalization costs were significant, with the highest cost in the West. Hospitalized patients with a LOS of 7 days or more incurred significantly higher costs than those with a LOS of less than 7 days. Hospitalized patients with a secondary diagnosis of HF incurred costs that were $3944 (P <.001) higher than hospitalized patients with the primary diagnosis of HF.

	Among patients with a secondary diagnosis of HF, 17% had a primary diagnosis of IHD; 14%, other diseases of the heart; 7%, other CVDs; 18%, respiratory diseases; and 44%, non-CVD and nonrespiratory conditions. Among these groups, hospitalized patients with IHD had the highest costs per hospitalization ($36,442). After controlling for other covariates and using hospitalizations with the primary diagnosis of non-CVD and nonrespiratory conditions as the reference, hospitalized patients with IHD incurred costs that were $14,989 higher (P <.001); those with other diseases of the heart had costs that were $9480 higher (P <.001) (Table 3).

	For both the entire sample and hospitalizations with a secondary diagnosis of HF, age 40 to 54 years, urbanization of hospitals, and the Charlson Comorbidity Index were significant predictors of LOS status (>7 days vs <7 days) (

). For example, patients aged 40 to 54 years were less likely to have a stay of 7 days or more than the reference population (those aged 55-64 years). Hospitalized male patients had a lower probability of having a LOS of 7 days or more than hospitalized female patients. For the entire sample only, having a secondary diagnosis of HF increased the probability of a longer stay. For the group of hospitalizations with a secondary diagnosis of HF, the primary diagnosis of other CVDs decreased the probability of having a longer LOS when compared with the primary diagnosis of non-CVD and nonrespiratory conditions. Hospitalized patients with the primary diagnosis of IHD, other diseases of the heart, or respiratory diseases might have a shorter LOS than the reference group.

	Our cost analysis provided new information in that it was based on all possible ICD-9 diagnoses of HF as well as diagnosis status. The detailed information we obtained could be useful in developing or evaluating the  cost-effectiveness of HF management programs as well as related policies, such as evaluating the impact on hospitalization costs of the classification of HF by the Agency for Healthcare Research and Quality as an ambulatory care—sensitive condition.

 This classification indicates that HF is a condition for which good outpatient care might prevent the need for hospitalization or for which early intervention might prevent the need for hospitalization or prevent complications or more severe diseases. Effective ambulatory care for HF will reduce hospitalization of patients with HF as the primary diagnosis, but will increase hospitalization of patients with HF as a secondary diagnosis. Moreover, improving survival of patients living with acute myocardial infarction (AMI) and the aging of the US population will increase the number of patients living with HF.

 Therefore, despite an overall decline in age-adjusted mortality from IHD in the United States, HF incidence and hospitalization rates have increased.

	According to Dr J. Fang, MD, an epidemiologist at the Centers for Disease Control and Prevention, as mortality from AMI and other IHDs continues to decline and the population continues to age, it is expected that the incidence and prevalence of HF will continue to increase. Management of HF in order to reduce hospitalizations with HF as the primary diagnosis was one of the Healthy People 2010 objectives. It also will be an objective in Healthy People 2020 (E-mail communication, July 2010).

	As we expected, we found that HF was more likely to be listed as a secondary diagnosis than as the primary diagnosis. We also found that the costs of hospitalizations were much higher with HF as a secondary diagnosis than as the primary diagnosis (the difference was $7671). After controlling for covariates, the difference still was substantial ($3944). Further study is needed to investigate how much of the hospitalization costs were for the medical management of HF, so the economic burden of HF can be quantified correctly.

	Earlier research found that hospitalization costs for HF management averaged $7000 in the early 1990s in patients 65 years or older and increased to an average of $18,086 in that group recently.

 However, it is unclear whether the HF in those studies was based on a primary or secondary diagnosis. If the most recent average was based on HF as the primary diagnosis, our results of $17,654 in 2005 for patients aged 18 to 64 years with HF as the primary diagnosis are quite comparable with the costs for patients 65 years or older.

	Among hospitalized adults with a secondary diagnosis of HF, we found that patients with IHD or with other diseases of the heart as the primary diagnosis incurred higher costs than those who had non-CVD and nonrespiratory conditions as the primary diagnosis. However, the hospitalized adults with IHD and other diseases of the heart were not more likely to have a LOS of 7 days or more than were those with non-CVD and nonrespiratory conditions. That may be because patients with IHD have a greater chance of getting percutaneous transluminal coronary angioplasty and coronary artery bypass grafting, which are very expensive, whereas treatment for acute myocardial infarction with revascularization reduced the total hospital LOS.

 The literature has shown that among HF-related hospitalizations in the United States, the percentage for patients with IHD, other diseases of the heart, and other CVDs as the primary diagnosis declined, while the proportion with a primary diagnosis of non-CVD increased.

 Because IHD causes reduced pump efficiency that may lead to HF, IHD is a major risk factor for HF. Primary prevention and appropriate management of IHD are critical steps in the reduction of HF admissions.

Articles

FOCUS ON HOSPITAL MEDICINE

Hospitalization costs associated with heart failure averaged $23,077 and were higher when heart failure was a secondary rather than the primary diagnosis.

Costs of Heart Failure-Related Hospitalizations in Patients Aged 18 to 64 Years

To assess whether health plan members who used retail clinics chose that setting for minor conditions and continued to see other providers for more complex conditions.

Retrospective analysis of claims data in a commercially insured population.

Health plan enrollment data were used to identify and describe the analysis population. Episode Treatment Groups were used to identify members with chronic conditions and to analyze reasons for retail clinic use, complexity of retail clinic visits, and care for chronic conditions in non—retail clinic settings. Logistic regression was used to study predictors of retail clinic use.

Retail clinic users differed significantly from nonusers. The most significant predictors of retail clinic use were age, sex, and proximity to a retail clinic. Episodes of care treated in the retail clinic appeared to be less complex than similar episodes treated in other settings. Chronically ill members who used the retail clinic saw another provider for their chronic condition at rates similar to or higher than those of members who did not use the retail clinic.

Individuals may be able to identify when conditions are minor enough to be treated in a retail clinic and serious enough to be treated by a traditional provider.

Appropriate use of retail clinics depends on an individual’s ability to identify when a condition is minor enough to be treated in that setting. We used claims data from a health plan to study the patterns of care for members who use retail clinics. We found that:

Episodes seen in the retail clinic setting appeared to be less complex than similar episodes seen in an office setting. 

 Members with chronic conditions who used retail clinics sought care for their chronic conditions from non—retail clinic providers at rates similar to those of members who did not use retail clinics.

	Retail clinics, also called convenience clinics, provide preventive care and treatment for minor, acute illnesses. They are located in retail stores, supermarkets, and pharmacies, and typically are staffed by certified physician assistants or nurse practitioners. For the consumer, retail clinics have advantages compared with traditional primary care, including lower cost and the availability of services without an appointment. The first retail clinics emerged in 2000 in Minnesota; the number has continued to grow nationally from 62 in 2006 to nearly 1000 in late 2008.

 Despite this growth, a 2007 survey found that only 1.2% of households had used a retail clinic in the previous year,

 even though a recent study estimated that more than one-fourth of the US population resides within a 10-minute drive of one,

 and a 2008 survey revealed that almost half of consumers would consider using a retail clinic.

	What does increasing use of retail clinics mean for consumers and for the healthcare system? These clinics offer convenience to the consumer and lower costs to the payer.

 Their appropriate use depends on the consumer’s ability to distinguish minor acute conditions that can be treated at a retail clinic from more serious conditions that would be better addressed by traditional providers. Given research suggesting that one-third of American adults have a health literacy measured at a “basic” or “below basic” level, this ability is not self-evident.

 Retail clinics have been criticized by the American Medical Association, the American Academy of Pediatrics, and the American Academy of Family Physicians, who claim that the clinics may create a conflict of interest (because of their relationships with retail pharmacies) and promote fragmented or substandard care.

 Providers have expressed concerns about use of retail clinics and the practice of waiving the copay for retail clinic visits.

 Critics also have cited concerns about the qualifications of clinic staff and their ability to diagnose serious medical problems,

 and about the disruption of existing physician—patient relationships, especially for those with chronic conditions.

	Research to date has focused on the purpose and cost of the visit, the quality of care provided,

 and the characteristics of retail clinic users.

 This research has not substantiated the concerns voiced about retail clinics, but rather suggests that minor conditions can be treated appropriately in that setting. A question that remains unanswered is whether individuals are adept at assessing the seriousness of their own conditions and are seeking care consistent with that assessment.

	The goal of this study was to extend the research to whether members choose retail clinics for minor concerns and whether they still choose non—retail clinic providers for less minor and chronic conditions. In particular, we estimated predictors of retail clinic use, compared complexity of visits for similar episodes in retail clinic and non–retail clinic settings, and assessed the degree to which members with chronic conditions who used a retail clinic also saw other providers for their chronic conditions.

	This study was set at Blue Cross and Blue Shield of Minnesota, a national health plan headquartered in Minnesota. Health plan claims and enrollment data were analyzed for a population of commercially insured members residing in Minnesota and continuously enrolled in 2007. The analysis included only services rendered by providers contracting with the health plan and located in Minnesota. This set included 72 separate retail clinic locations. A retail clinic is a unique category for contracting purposes and is defined as a clinic located in a retail or nonmedical setting with a limited scope of services. All retail clinics are covered by the same reimbursement arrangement. The services provided at retail clinics are limited by design to address noncomplex ailments that fall within the scope of practice of the provider staff. Some health systems in Minnesota have added clinics that function as retail clinics in terms of scope of practice and reimbursement, but are staffed by physicians in addition to nurse practitioners and physician assistants. These physician-staffed clinics were classified as retail clinics for the purposes of this analysis.

	To study whether members chose retail clinics for less complex concerns, we identified members with at least 1 visit to a retail clinic in 2007. To draw a comparison with members who did not use a retail clinic, we identified members with at least 1 office visit associated with diagnoses thought to be treatable in a retail clinic (see 

). We excluded members younger than 18 months (as of January 1, 2007) because the state of Minnesota requires that retail clinic patients be at least that old. We used geocoding to identify whether each member lived within 10 miles of a retail clinic.

 We then assessed whether members with at least 1 chronic condition who used a retail clinic continued to use traditional providers for chronic conditions. Episode Treatment Group (ETG) data from the prior years 2005 and 2006 were used to identify chronic conditions in the study population. Episode Treatment Groups use proprietary algorithms to group medical and pharmacy claims into episodes of care based on clinically similar conditions.

 Individuals with a chronic condition were identified using the following ETGs: diabetes (27-30), asthma (386-389), depression (95, 96), coronary artery disease (251-254, 256-265), congestive heart failure (267, 268), chronic obstructive pulmonary disorder (390-395), hyperlipidemia (47), and hypertension (278-281). We used episodes from the prior years to exclude new diagnoses of chronic conditions in the analysis year. For the comparison group of retail clinic nonusers with a chronic condition, we did not require an office visit as an inclusion criterion.

	We also used ETGs to describe the reason for the retail clinic visit and to describe the overall health of the members. Episode Risk Groups (ERGs) aggregate ETGs of similar clinical and risk characteristics.

 A person with multiple ETGs will have 1 or more ERGs. Each member was assigned a prospective ERG risk score, a weighted combination of those ERGs that predicts utilization. We defined high health risk as those members with a score of 7 or higher. This cutoff point captured those members incurring the top 0.5% of total claims spending.

	Demographic characteristics were compared between retail clinic users and nonusers using 

2 tests to assess whether clinic use was strongly dependent on sex, age, or type of health plan product. All available data describing an individual member were included in the analysis, including sex, age, health plan product, presence of a chronic condition, residence near a retail clinic, and ERG risk score. Health plan products were categorized as preferred provider organization (PPO), health maintenance organization (HMO), or consumer-directed health plan (CDHP). The PPO plans involve a network of preferred doctors and hospitals. Generally, services incurred with providers not in the network result in higher out-of-pocket expenses (ie, deductibles, copayments, and/or coinsurance). The HMO products require members to choose a primary care provider and require a copayment for most services. The CDHP products include major medical plans with high deductibles and out-of-pocket maximums, and some form of savings vehicle (eg, Health Reimbursement Arrangement, Health Savings Account) to fund expenses that are subject to the deductible. In the time period covered by the analysis, none of these products included an incentive for members to use retail clinics.

	We assessed whether members chose the retail clinic for minor conditions using the percentage of retail clinic episodes resolved in 1 visit. For people with chronic conditions, we analyzed whether they also saw a non—retail clinic provider for that condition. This study received no external funding.

	In 2007, among commercially insured members who were continuously enrolled, 40,686 members (2.8%) had at least 1 visit to a retail clinic. This constituted an increase of 71% over the previous year (results not shown). Among members who did not have a visit to a retail clinic, 651,514 members had at least 1 office visit for an immunization or for diagnosis of a condition that was potentially treatable in a retail clinic. 

compares the demographic characteristics of retail clinic users and nonusers. They differed significantly on nearly every measure analyzed. A higher percentage of retail clinic users were female, healthier, and younger compared with retail clinic nonusers. In both populations, a majority of members were enrolled in a PPO plan. A lower percentage of retail clinic users had chronic conditions and an ERG prospective risk score of 7 or higher. More than 90% of retail clinic users lived within 10 miles of a retail clinic, but fewer than 50% of nonusers did.

	We performed a logistic regression to identify the predictors of retail clinic use (

). All of the characteristics listed in Table 2 were significant predictors. This table indicates that compared with the reference group (members aged 25-44 years), all other age groups were less likely to use the retail clinic, after controlling for other characteristics. Members enrolled in CDHP plans, members enrolled in HMOs, and women were more likely to use retail clinics, whereas sicker members and members with chronic conditions were less likely. The most significant predictor in the model was whether the member lived within 10 miles of a retail clinic.

	We next sought to understand whether members appropriately chose the retail clinic for their care by analyzing the percentage of episodes of care resolved in 1 visit. We reasoned that if the episode that prompted the retail clinic visit resolved completely in that visit, the ailment was minor and using the retail clinic for its treatment was fitting. Overall, 94.8% of episodes with a retail clinic visit began with that visit. Moreover, 89.9% of episodes involved no other providers, and 88.3% were resolved in 1 visit, suggesting that the majority of visits were for minor conditions. 

 shows the percentage of episodes resolved in 1 visit for the top 10 episode types (as described by ETG and sub-ETG) seen in the retail clinic, and compares this percentage with that for episodes seen in a non—retail clinic setting. We excluded 2 of the most frequent episode types, routine inoculation (791, 0) and screenings and immunizations incidental to other services—immunizations (901, 7). As one would expect, for both of these episode types, nearly every episode was resolved in 1 visit, in both the retail clinic and the other settings. For all but 2 episode types, a greater percentage of episodes were resolved in 1 retail clinic visit than in 1 visit to other providers. This conclusion held when we applied the Bonferroni correction. For chronic sinusitis, there was no difference between the 2 settings. For routine exams, fewer retail clinic episodes were resolved in 1 visit. This may reflect the more open-ended nature of a routine exam, which could result in referrals for follow-up. We also performed this analysis separately for those members with and without chronic conditions, and separately for children aged 2 through 11 years, and the conclusions did not change (results not shown).

	Finally, we explored how often in the study year people with chronic conditions who were seen in a retail clinic also were seen by another provider for care of their chronic condition, which reflected a member’s ability to choose a care setting appropriate for more complex conditions. We compared that with the rate at which retail clinic nonusers with chronic conditions had provider visits for their chronic condition. For this comparison, we included all commercially insured members residing in Minnesota who had at least 1 chronic condition. When comparing the retail clinic users to the nonusers, we found that they differed in the same way as in the larger population (

). We limited this comparison to members 18 years and older, yielding 8859 retail clinic users and 386,600 nonusers with 1 or more chronic conditions. Retail clinic users with diabetes were less likely to see non—retail clinic providers for the care of their chronic conditions than were nonusers, and retail clinic users with hyperlipidemia or depression were more likely (

). There were no significant differences in these measures for the other conditions studied. The number of retail clinic users with chronic obstructive pulmonary disorder was too small to make a meaningful comparison. When we applied the Bonferroni correction, the difference for members with diabetes was no longer significant, although the differences for hyperlipidemia and depression remained significant.

	The number of retail clinics has increased significantly in the last decade. With growing acknowledgment of their ability to provide quality care at lower cost, insurers have developed strategies to encourage the use of such clinics. These policies have been criticized by those who worry that retail clinics will adversely affect the care received. In this study, we compared the characteristics of people who used retail clinics with the characteristics of those who did not, and estimated the predictors of retail clinic use. We explored whether members appeared to be making appropriate choices by examining the complexity of retail clinic visits and visits to non—retail clinic providers by members with chronic conditions.

	Overall, our findings of age and sex being significant predictors reaffirm the findings of previous research.

 We also found that proximity to a retail clinic was a significant predictor of retail clinic use. The percentage of care episodes resolved in 1 visit was higher for retail clinic users than for nonusers. This finding is consistent with an earlier study that found that retail clinic users were no more likely to have an “early return” visit.

 One interpretation of this finding is that the episodes seen in the retail clinic setting were less complex and were suitably treated there. Another interpretation is that non—retail clinic providers may have been more likely to schedule follow-up visits or to refer to other providers, and that retail clinics may discourage repeat visits. Claims data did not provide adequate information to judge if those episodes (whether occurring in a retail clinic or not) that included more than 1 visit were due to the proper referral of a patient with a more serious condition to a primary care provider or specialist, unnecessary referrals, or poor quality of care provided by the originating provider. For the retail clinic users, another possible explanation for the episodes requiring additional visits may be that the consumer underestimated the complexity of the condition and chose an inappropriate setting for treatment.

	Concerns about retail clinics disrupting the patient—provider relationship have not been fully explored. When we focused on members with chronic conditions, we found that retail clinic users were at least as likely as nonusers to have had a visit with another provider for their condition. No studies to date have addressed the impact of retail clinics on coordination of care. When a patient visits a retail clinic located in a clinic system with electronic health records, that record makes it possible for the patient’s primary care provider to know about the visit. Some retail clinics provide a record of the patient’s visit to the primary care provider at the patient’s request. In the absence of either of these options, a physician may not know about a patient’s retail clinic visit, which may affect coordination of care.

	This study has several limitations. Some chronic conditions (eg, depression) may have been overidentified because of limitations of the episode grouper logic. Missing episodes and orphan records resulting from the logic may have slightly affected the results by under- or overidentifying retail clinic visits. We could not ascertain how many members attempted to use the retail clinic but were turned away because their presenting condition was not appropriate for treatment there. One study suggested 2% of retail clinic patients were triaged to an emergency department or a primary care provider.

 In our analysis of why people with chronic conditions went to the retail clinic, we did not control for clinically significant combinations of acute and chronic conditions such as asthma and bronchitis. Finally, claims data presented an incomplete picture of the decision-making process that led the member to the retail clinic. The member’s stated symptoms may not match the diagnosis eventually coded on the claim by the retail clinic provider. Furthermore, we cannot know from the claims whether additional diagnoses (beyond the presenting complaint) were uncovered during the course of the retail clinic visit and resulted in follow-up visits to other providers. These new diagnoses may have resulted in a subsequent episode of care that was not captured by this analysis.

	Previous research on retail clinics focused on the purpose of the visits, the characteristics of the patients, cost, and quality of care. In this study, we explored whether members could ascertain when the severity of their condition indicated that a retail clinic would be an appropriate treatment setting. For the most common acute and preventive conditions, the choice of a retail clinic does not appear to have been associated with problematic self-triage, with more than 88% of episodes resolved in 1 visit. Sicker members appeared to be less likely to use retail clinics. Retail clinic use was not associated with a lesser frequency of ongoing visits to other providers for the treatment of chronic conditions. The degree to which patients with chronic conditions who use the retail clinic receive high-quality care for those conditions from their regular provider should be further studied. Another area for future study is whether healthier patients using retail clinics for routine and minor acute care affect the patient mix in primary care settings.

 From Blue Cross and Blue Shield of Minnesota (ARW, XTZ, WS, HR, EPB, NAG, TJS), Eagan, MN. Dr Wilson is now with the Medical University of South Carolina, Charleston, SC.

This study was funded by Blue Cross and Blue Shield of Minnesota.

MANAGERIAL

Care episodes treated in retail clinics appeared to be less complex than those treated in office settings.

Retail Clinic Versus Office Setting: Do Patients Choose Appropriate Providers?

To examine the relationship between participation in a large wellness and care management company's transitional case management (TCM) program and hospital readmission.

A total of 10,258 members were identified as either participants or nonparticipants in TCM from data archives of a large healthcare company. Engagement and claims data were analyzed using multivariable logistic regression. Readmission predictors that were studied included TCM engagement, the major diagnostic categories of "musculoskeletal" and "digestive," length of stay for the initial hospitalization, cost of initial inpatient stay, risk score, age, and sex.

Readmission rates were lower among individuals who were engaged in TCM compared with those who were not engaged. Within 30 days, 12.66% of individuals participating in TCM were readmitted to the hospital compared with 35.85% of those not participating (

 <.0001). In the first 30 days, individuals who did not participate in TCM were almost 4 times more likely to have a hospital readmission than those who did participate. The most important predictor of hospital readmission was engagement in TCM. Individuals who were engaged in the program were less likely to be readmitted than those not engaged in the program (

Implementation of a telephonic TCM program was associated with lower rates of readmission within 30 days. Timely engagement in TCM was associated with a lower likelihood of readmission.

A telephonic transitional case management (TCM) program for patients discharged from an acute care facility was effective in reducing readmissions within 30 days.

The readmission rate for participants in TCM was 12.66% compared with 35.85% for nonparticipants. 

Patients who did not participate in TCM were almost 4 times more likely to be readmitted compared with those who did participate in the program. 

The most important predictor of hospital readmission was participation in TCM.

Interest in reducing healthcare costs continues to grow in the national debate on healthcare reform. An important element in this debate is the impact of hospital readmissions on overall healthcare costs. Appropriate discharge planning and transitional case management (TCM) programs are frequently touted in health service literature as potentially successful interventions to improve care and curb cost by reducing or preventing unnecessary readmissions. These programs facilitate care transition from acute care to the home setting, and strive to reduce or prevent unplanned readmissions. Additionally, it has been suggested that program interventions are associated with increased patient satisfaction and overall well-being.

 Despite tremendous efforts by healthcare providers and hospitals, readmissions are a common occurrence, as many patients experience difficulties in the first few days and weeks after discharge from an acute care facility. Communications among healthcare providers and between providers and patients/caregivers remain inadequate or lacking.

 These difficulties commonly result in readmissions to acute care facilities that could have been avoided. Depending on diagnosis, between 5% and 30% of adult medical—surgical patients are readmitted to the hospital within a month.

 Transitional case management programs are expected to reduce preventable hospitalizations by providing a framework for assessing patients’ and caregivers’ understanding of their condition(s) and increasing adherence to their management plan after discharge.

 Patients may have problems understanding discharge instructions or adherence to medications. They also may ignore their follow-up appointments with doctors and other healthcare providers. Almost all of these care gaps and barriers are frequently addressed by most TCM programs. It is imperative that healthcare programs engage individuals in their care during this critical time of their illness. Engaging patients or caregivers can occur in many ways, including face-to-face discussions and via telephone or mail.

 This study illustrates the impact of a TCM program offered and operated by a national healthcare company. The program proactively monitored and managed hospitalized patients immediately after their discharge to ensure a safe transition to home. Patients identified for this activity were considered high risk for readmission due to their underlying illness, length of hospitalization, or other complex discharge plans.

 The TCM program being studied in this analysis was a postdischarge, telephonic, patient-centric program that aimed to close gaps in care for patients after leaving the hospital and returning home. Many patients leave acute care facilities without adequate discharge planning, resulting in unnecessary readmissions. Premature and unnecessary rehospitalizations may result from patients’ failure to schedule outpatient follow-up visits; lack of  follow-up plans with other providers; lack of scheduled necessary laboratory, radiologic, and specialized testing; and absent or inappropriate therapeutic pharmaceutical interventions (such as missing evidence-based therapy, drug duplications, or severe adverse drug reactions).

 Case managers used specific tools and assessments to find and then close gaps in care. They discussed with patients their discharge/follow-up plans, including their medications, and assessed their level of functioning (including activities of daily living and cognitive status). Moreover, they conducted a depression screen, safety risk assessment, and caregiver status assessment (if applicable), and identified economic barriers to care such as lack of transportation, caregivers, or personal funds. Case managers educated patients and/or caregivers to increase their understanding of the disease process and facilitated adherence to treatment plans including self-care and other home-based care such as dressing changes and infusions. Lastly, case managers conducted a condition-specific assessment relative to the primary discharge diagnosis (eg, coronary artery disease, cancer, stroke, asthma).

 Because it was critical to address potential care gaps in a timely manner to prevent readmission, this TCM process started with 2 attempts by the case manager to contact the patient by telephone within 3 business days of  discharge. If the first call failed to contact the patient, the case manager left a toll-free number and requested a call back. If the case manager did not receive a call back within 2 business days of the first attempt, a second call was attempted. If that failed, the case manager sent an “unable to reach” letter to those who could not be reached. In this letter, the patient was urged to contact the case manager at the toll-free number provided in the letter. The case was left open on the case manager’s list for 15 business days. If the patient was readmitted to the hospital (resulting in an inability to reach the patient), the process was started again after discharge. If the patient did not respond to the letter and did not initiate a contact with the case manager, a third and final call attempt was made. If there was no answer, the case manager left a final message to let the patient know that calls would no longer be forthcoming. The case manager provided a contact number again for future use and then closed the case.

 At all times, TCM staff encouraged patients to participate in the program. Case managers used their critical thinking and clinical judgment to identify gaps in care relevant to the disease, current patient condition, discharge instructions, and adherence to care plans. If the patient was not well versed in his or her discharge instructions, the case manager contacted the physician to better understand the discharge instructions and then educated and supported the patient to boost compliance with discharge instructions. For example, after assessing barriers to follow-up care (as mentioned above), the case manager might educate the patient about available options or solutions such as community resources (for financial barriers), the need for making or keeping follow-up appointments with providers, medications gaps, or referrals to other internal or external specialized programs (eg, behavioral health referrals for depression management, cancer or transplant management programs, mail order programs).

 Case managers had full access to a physician medical director whom they could ask for guidance on all clinical and nonclinical gaps in care. This communication was encouraged and could occur via instant messaging, e-mail, telephone, or in person. The medical director contacted the physician primarily responsible for out-of-hospital care, as necessary, to produce an appropriate readmission prevention plan. In addition to their clinical judgment, case managers could use specific job aids that guided them in determining which cases needed to be referred to the medical director. Afterward, those activities were recorded in real time and stored on a proprietary computer platform. This computer system enabled tracking of all activities and documentations, and linked them to other desired data on process or outcome of care. At their discretion, TCM staff could use 1 or more calls to close outstanding care gaps and prevent unnecessary and costly readmissions.

 In patients under the age of 18 years or those unable or unwilling to talk, case managers conducted the above assessments by talking to the legal guardian or a known caregiver. Legal permission may have been required and was obtained prior to discussing patients’ care issues, including gaps and interventions.

 The purpose of this study was to examine the relationship between participation in a TCM program and hospital readmission.

 This was a retrospective cohort study. Patients were identified as eligible for this study if they had an initial hospitalization between April 1, 2007, and June 30, 2008. (Additional inclusion/exclusion criteria are detailed in 

.) Individuals with an initial hospitalization were then followed for 30 days to determine whether a readmission occurred.

 To accurately reflect the impact of a TCM program, readmission rates were calculated for those participating and not participating during the same time periods: within 7 days, within 15 days, and within 30 days. These time periods were chosen to clearly establish the temporal relationship between the case manager intervention and readmissions. The intervention (case manager call and patient participation in the program) had to precede the readmission. For example, if program participation occurred 2 to 6 days after discharge and a readmission was averted within 7 days after discharge, that was considered an intervention success and counted in the study. On the contrary, if the intervention occurred 10 to 14 days after discharge and the averted readmission occurred only 7 days after discharge, that case was not counted (ie, the intervention was not within the time frame for the outcome of interest; therefore, there was no association between outcome and intervention).

 The participant group was identified as patients who had an initial hospitalization during the selected time period (April 2007-June 2008), who were identified for the TCM program within a certain time period of initial discharge date (within 7 days, within 15 days, and within 30 days), and who participated in TCM during the same category time period. Patients were considered participants in the program if they met at least 1 of the following criteria: they had an assessment done, an intervention such as a scheduled follow-up visit to their doctor was completed by their TCM nurse, or an indicated therapeutic intervention was arranged.

 The nonparticipant group was identified as patients who had an initial hospitalization during the selected time period (April 2007-June 2008), who were identified for the TCM program within a certain time period of initial discharge date (within 7 days, within 15 days, and within 30 days), and who did not participate in TCM during the same category time period.

 Data for this study were extracted from 2 data sources that were merged to form a single data set. The first step was identifying and then extracting admission data from a large healthcare company’s data files. Information on contacts, enrollment, and participation in TCM was obtained from a data set specially designed to house intervention and engagement efforts in our study population. After linking these secondary data to the original admission cohort (see Study Design), the following information was collected: identification for TCM participation, actual participation (or nonparticipation) in TCM, date of participation, and time (in days) from discharge to participation.

 Univariate analyses were conducted using c2 tests or t tests as appropriate. Odds ratios (ORs) were calculated when applicable.  All analyses were performed with SAS version 9.1.3, executed on an AIX 5.3 platform (SAS Institute Inc, Cary, NC).

 A backward elimination multivariate logistic regression analysis was conducted to further study the relative impact of various independent variables on readmission (dependent variable) and to produce a predictive model for evaluating the program intervention. Independent variables included TCM participation, major diagnostic categories, length of stay for the initial hospitalization, age of patient, sex of patient, and cost of initial hospitalization.

 Several criteria were applied to create the initial data pull and to produce a final study data set of 10,258 individuals, as noted in the Data Sources section. Figure 1 shows the number of initial and final TCM program cohorts (or patient cohorts) after application of appropriate program inclusion/exclusion criteria.

 Among the study population, 49% were female and 51% were male. The average age of participants was 50 years, and the average length of the initial hospitalization was 5.2 days. The most common major diagnostic categories were musculoskeletal, circulatory, digestive, and respiratory. The average reimbursed charges for each admission were approximately $20,000. The average severity score, as measured by Symmetry Episode Risk Groups (ERGs), was 16.7 Episode Risk Groups predict current and future healthcare use for individuals by creating risk measures that incorporate episodes-of-care methodology, medical and pharmacy claims information, and demographic variables.

 In the 30-day time period, 88% of individuals participated in the TCM program and 12% did not participate. Among the nonparticipants, 60% (771) could not be reached by the case manager using the previously described outreach protocol. Possible reasons for inability to reach those identified for the TCM program include incorrect telephone numbers, not being at home at the time of contact, and inaccurate mailing addresses. Another 30% (388) of the nonparticipant group was composed of individuals who were contacted by a case manager but after assessment were determined to need no further intervention. This may have been because the patient was adherent to, or adequately progressing through, the posthospitalization treatment plan. The remaining 10% (127) refused to participate in the TCM program or had disenrolled from the insurance plan.

 Readmission rates were statistically different between the participant group (13%) and nonparticipant group (36%) (P < .0001). Age, sex, and cost of initial hospitalization were comparable between the 2 groups (

). Risk scores were not substantially different between the 2 groups, with an average ERG score of 16 for TCM participants (meaning they were 16 times more likely than a person with a risk score of 1, or the risk of an average person in the study population, to use healthcare resources) and 17 for nonparticipants (meaning they were 17 times more likely than a person with a risk score of 1 to use healthcare services). The average length of hospital stay among participants within 30 days was 5.1 days, and the average length of stay among nonparticipants within 30 days was 5.9 days (P <.0001). The major diagnostic categories of circulatory, digestive, and musculoskeletal were different between the groups.

 In summary, the 2 groups were generally similar to each other with regard to age, sex, and cost of initial admission. The differences in length of stay, which could have been because of illness severity and/or diagnosis, were not disconcerting given the above ERG risk scores. In any event, adjustment for these confounding factors was carried out in multiple regression modeling.

Costly new breast cancer therapies augment the significant burden this disease places on healthcare resources, but in context they may still provide value to society.

Hospital Readmission Among Participants in a Transitional Case Management Program

	Accountable care organizations (ACOs) have emerged over the past 2 years as the newest big thing in healthcare. Hospitals, physician organizations, community health centers, and lots of other organizations are holding strategic planning meetings to figure out how to get some of the ACO action that the Patient Protection and Affordable Care Act promises.

	The specific provisions of the Affordable Care Act that people are excited about are Sections 3022 and 10307, which delineate a Medicare Shared Savings Program and include the broad requirements for organizations that as ACOs would be eligible for participation under the new payment approach. As a program, rather than a pilot or demonstration, any entity that meets the eligibility criteria gets to participate. And because it is a program, the Centers for Medicare & Medicaid Services (CMS) will issue program regulations rather than a request for proposals, with proposed regulations expected to be issued by year-end.

	The impetus for establishment of ACOs, which are charged with holding down spending while meeting quality criteria, comes from the growing pressure of escalating healthcare costs on public and private insurers combined with evidence that at least some of this spending appears to not improve either quality of care or patients’ experience with care. Although experts dispute the exact amount of wasted spending, it is generally thought to be substantial.

	Even a small reduction in the excessive spending—more precisely, a small reduction in the rate of increase in healthcare spending&mdash;on the order of about 2 percentage points in real, per capita growth per year would essentially forestall the need for possibly more arbitrary approaches to cost containment and the specter of actual rationing of care. Accountable care  organizations are thought by many to hold great promise for cost containment because, unlike the “regulatory” approaches used by many managed care plans, the ACO approach would give responsibility for containing costs to clinicians, who can better target waste and inefficiency if motivated to do so, rather than distant third-party public and private payers.

	Further, ACOs offer the promise of decreasing the fragmentation of healthcare delivery by bringing under one virtual roof the various medical specialists and other health professionals and institutions that need to coordinate care for the growing number  of patients with multiple chronic conditions. In one recent accounting, half of Medicare beneficiarieshad 5 or more chronic conditions and were responsible for 76% of Medicare spending in 2006.

 A Medicare patient with 5 or more chronic conditions averages 37 physician visits, sees 14 different physicians, and fills 50 prescriptions a year.

Patients with multiple chronic conditions often receive suboptimal care despite the fact that individual clinicians may be practicing according to their own professional standards.

	In theory, the ACO would address the fragmented care that too often results from competent clinicians practicing in silos, producing different diagnoses and treatment plans, prescribing incompatible medications, and delivering redundant, costly care. Although a lot of current activity is focused on the addition of promising chronic care coordination and management programs to current healthcare delivery, the ACO—again, in theory&mdash;would take practice redesign to a broader level, producing a different business model for success and an altered culture to achieve that success with active involvement of all clinicians and other professionals across the spectrum of care that patients receive.

	In practice, numerous provider organizations appear to be well positioned to meet the expectations one would have for ACOs. These include very successful physician-led, multispecialty group practices, such as Scott and White Healthcare in central Texas and Group Health of Puget Sound in Seattle; physician—hospital organizations such as Advocate Health in Chicago; hospitals with employed medical staff such as Virginia Mason in Seattle and Carilion Health System in Virginia; and independent practice associations (IPAs) and management services organizations such as Monarch Health Care in southern California and Physician Health Partners in Denver. Some already own their own health plans and can participate in Medicare through the Medicare Advantage (MA) program.

	A real-world model for developing ACOs is “delegated-capitation,” which is being used by health maintenance organizations (HMOs) in much of California and some other markets where medical groups and IPAs (generally without the organizational  inclusion of hospitals) are functioning and, subsequent to a market shakeout in the 1990s, are now doing pretty well.

 In fact, the ACO idea is not really new; in the 1990s, these organizations were labeled provider-sponsored organizations (PSOs), and there were great hopes for them to enter traditional Medicare. Indeed, the Balanced Budget Act of 1997 established the opportunity for direct contracting between Medicare and PSOs, without having a health plan intermediary.

	The fact that only a couple of PSOs have participated in what is now called the MA program should offer caution about the prospects for ACOs.

 It appears that organizations capable of meeting the program criteria for risk-taking under MA rules chose either to obtain insurance licenses themselves to become full-fledged MA plans or to contract with insurance companies to serve Medicare patients. The latter activity has been spurred on in recent years by the substantial extra payments MA plans have received since 2004.

 Indeed, many would-be ACOs, such as Geisinger Health System and Kaiser Permanente, already participate in Medicare with their own MA plans. Given this experience, 2 questions need to be asked. First, why would the opportunity to become an ACO produce a different provider response than PSOs did? Second, which kinds of organizations would choose to participate in Medicare under the Shared Savings Program as ACOs?

PROBLEMS WITH THE SHARED SAVINGS PROGRAM MODEL

	Proponents of the shared savings approach for ACOs believe that an incremental, nonthreatening program design is the best way to gain initial participation by diverse provider systems across the country and to nudge them in the desired direction of change, with the possibility of strengthening the program design over time to more fundamentally change the business model and culture of the ACOs. Consistent with this notion of a glide path to transformation of healthcare delivery systems, some have suggested a tiered approach that over time would implement different design features, including the use of financial risk, in ACOs based on the extent of their organizational integration and capabilities.

	Unlike the PSO program, the shared savings design does offer the possibility of broad participation. However, if CMS is not careful, broad participation will likely happen because the members that make up the ACOs will not actually have to change the way they deliver care. Unless CMS chooses to put some teeth into the program through reasonably demanding eligibility criteria, the Shared Savings Program expects relatively little of ACOs.

	The basic statutory requirements of the program are that ACOs need to have the capacity to deliver or arrange for the continuum of care for those patients assigned to it, to have a sufficient number of primary care professionals to provide services to at least 5000 beneficiaries (achievable by as few as about 10 primary care physicians with typical practices), and to report data on cost, quality, and overall patient experience for beneficiaries in traditional Medicare.

 Although Sections 3022 and 10307 give the Secretary discretion in using additional payment approaches, they specify in detail a shared savings payment approach whereby groups would be paid their usual Medicare fee-for-service reimbursements, with no penalties at all for higher spending, and could share in savings if the group provides care to assigned beneficiaries for less than a Medicare benchmark spending target, while passing readily achievable thresholds for patient service and quality of care.

	Under the legislated approach, there can be no limitation on patient choice of provider at the point of service, in contrast to the lock-in approach that many MA plans adopt. It is even possible that the Secretary could assign beneficiaries “invisibly” (without their knowledge) to an ACO on the basis of concurrent fee-for-service claims that indicate where they receive the preponderance of their primary care services, as was done in the Medicare Physician Group Practice (PGP) demonstration (discussed below). Under such an approach, it is possible that the ACO wouldn’t know which of its patients qualify it for shared savings payments.

	To set the spending target, the ACO’s 3-year spending history for the beneficiaries assigned to it would be compared with a spending target that trends the historical spending forward by the projected increase in spending in Medicare Parts A and B nationally. If actual spending ends up being less than the projected spending by a certain percentage (eg, 2% in the PGP demonstration), the ACO would split the savings with Medicare (eg, 80%-20% in the PGP demonstration). The threshold and the percentage shares for the parties will be determined by CMS in forthcoming regulations.

	The problem with this approach is there are no downside penalties for missing the spending targets. Given the reliance on fee-for-service, it is hard to see how a speculative bonus attained in the future would counter the direct fee-for-service rewards for providing more reimbursable services, especially for hospital-sponsored ACOs. The ACO might engage in some targeted programs to address a particular care area, but not engage in a broader approach to change the “more is better” culture that is pervasive under fee-for-service.

	A further problem is that the ACO’s historic spending pattern for its assigned patients, “risk-adjusted” for health status, is used  as the base spending amount the ACO is workingwith. Therefore, previous practice inefficiency is baked into the calculations. Indeed, prospective ACOs even have an incentive to spend more before onset of the program to build up their 3-year historical  spending base. It is no wonder that some have described the anticipated ACO program as the “lottery.” Why not take a chance on sharing in savings that might occur for reasons having little to do with new organizational efforts to alter care delivery (eg, experiencing a season without influenza)?

	As referred to above, the model for the ACO program, including invisible, claims-based assignment and shared savings with no risk payment approach, has been tested in the PGP demonstration project that began in 2005. In the demonstration, 10 group practices (with 1 practice actually functioning like an IPA, and most with strong hospital affiliations or ownership) were permitted to receive bonus payments if they passed quality-of-care thresholds and achieved savings on Part A and B Medicare spending compared with spending for a matched control group cared for by others in the same geographic area.

	In the first 3 years of this demonstration, 5 of the 10 practices bettered the 2% savings threshold compared with the control group and so received shared savings; 3 practices were within 1% of the control target; and 2 practices had spending that exceeded the target by more than 2%, with no financial penalties.

 However, the year 2 evaluation report documented that the essential reason for the overall savings across the 10 sites of about 1% compared with the control group was from diagnosis coding changes the PGP sites initiated that produced different risk-adjustment scores for their patients.

 In effect, the coding pattern changes produced apparent savings that resulted in shared savings payments to some of the demonstration sites, but not actually fewer dollars spent. In addition, the evaluators were not able to link savings to specific interventions, and found that the formal programs the PGP sites adopted were targeted to particular small patient subpopulations, with little attempt to reorient the basic care provided to all patients.

 Another barrier to successful organizational change, according to the PGP program managers, was that individual PGP physicians were still paid based on fee-for-service—oriented productivity.

	Given that the initial CMS commitment for the PGP demonstration was only 3 years, it is not surprising that the participant organizations adopted targeted programs that did not attempt to change the fee-for-service—oriented business model they have been long accustomed to. To do so based on a mere 3-year commitment from one of its payers— even one as significant as Medicare&mdash;would have been foolhardy.

	Theoretically, a program that offers shared savings without any financial downside provides weak incentives for providers to change behavior. Empirically, the approach did not succeed in the PGP demonstration, assertions to the contrary notwithstanding.

	History tells us that meeting the requirements for being a full-fledged MA plan is not something that a diverse set of provider organizations have been willing to undertake, presumably because it represents too much change and uncertainty for most providers. A refreshed approach, whereby Medicare directly contracts with organized provider groups, reasonably avoids beneficiary lock-ins, does not require providers to assume full insurance risk, and bypasses many regulatory requirements designed to protect beneficiaries and the Medicare program. But in the process of taking this refreshed approach, the Shared Savings Program may have swung too far in the opposite direction. The question at this point is whether the ACO design, based on the not particularly successful PGP demonstration, represents 

	Fortuitously, although the PGP demonstration design was essentially adopted as the Shared Savings Program, there appears to be flexibility for CMS to alter the design somewhat to provide an opportunity for a better test of something between unfettered fee-for-service (with all of its perverse incentives) and full-risk contracting (with its inherent limitations on patient prerogatives and incentives for stinting on care). The Centers for Medicare & Medicaid Services has been holding “listening sessions” in which many prospective ACOs, including many experienced group practices and IPAs, are suggesting the need for making changes to the PGP model. Further, the Affordable Care Act specifically allows the Secretary to create a “partial capitation” ACO payment model or “any payment model that the Secretary determines will improve the quality and efficiency of items and services furnished under this Title,” instead of the PGP-style payment approach outlined in the legislation. With this flexibility in mind, we recommend making the following program design changes:

 Recommendation #1: Avoid Invisible Assignment

	Many want to avoid the PGP demonstration’s passive, invisible assignment, but not require a formal, locked-in enrollment by  beneficiaries as under MA. There are 2 basic reasons why individuals should know up front that their care is being analyzed as part of an ACO’s cost calculations.

	First, although the shared savings model built on unchanged fee-for-service may provide weak incentives for change, the approach in fact does have incentives that could well affect how care is provided—both for better and for worse. Physician group ACOs without a hospital component would benefit financially under a shared savings approach by keeping patients out of the hospital. In contrast, hospitalbased ACOs might not have an incentive to keep patients out of the hospital because the actual fee-for-service profits would likely be greater than the potential share of savings coming many days later. And all ACOs would have an incentive not to refer patients to other providers even when another institution might be more capable of addressing a patient’s clinical situation; in this way they would maximize fee-for-service income while reducing “out of network” spending attributed to their patients.

	We now have good measures of primary and secondary preventive care that ACOs could be required to meet to be eligible for bonuses; but we lack measures for “rescue” care withheld, such as for a patient who might be referred to a specialized cancer center, often at much greater cost than if the ACO provided the services itself. Patients need to know about these incentives and perhaps have an opportunity to opt out of such arrangements, while being able to continue seeing the same providers. Some process protections are likely needed as well, although probably fewer than what are required under the appeals procedures applicable to MA plans. One idea to consider is an ombudsman program to help patients navigating healthcare and making choices, perhaps by facilitating access to a non-ACO second opinion about a recommended course of treatment.

	The other reason for rejecting invisible assignment is to promote a kind of social contract between beneficiaries and the ACO, outlining the parties’ responsibilities to each other but not otherwise restricting patient freedom of choice— call it a “soft lock-in.”

 An ACO should be expected to (1) make a positive case to would-be ACO participants (ie, patients who have been seeing its clinicians) about the potential benefits that their involvement in enhanced patient care activities would provide and (2) attempt to achieve their affirmative participation.

	Under an approach that uses prior-year claims to attribute patients to an ACO ahead of time and offers them the opportunity to opt out, provision also should be made for other beneficiaries to seek out care from an ACO electively without first having to spend a year or more seeing that ACO’s physicians. For these individuals, the approach is like an enrollment, but again, their freedom of choice of provider at the point of service would not be taken away. Finally, consideration should be given to permitting the ACO to provide some additional services that are currently prohibited as illegal inducements to care (eg, reducing cost-sharing for those without supplemental insurance, covering travel expenses) to support development of a social contract emphasizing ACOs’ and patients’ responsibilities to each other.

 Recommendation #2: Introduce Shared Risk

	Many organizations that want to be rewarded for prudent husbanding of healthcare resources while providing high-quality care are not in a position to take on full insurance risk for care provided by all providers that see their patients, as they would as a PSO under the MA program, or even to take full risk for the services they provide themselves. Because of the skewed nature of healthcare spending, in which a small number of individuals are responsible for a highly disproportionate share of spending (many as a result of random, unpredictable events that often are outside of providers’ control), ACOs— especially the smaller ones without deep pockets&mdash;correctly are leery of entering into any risk arrangements. Many providers assumed risk in the 1990s and did not have the tools to manage cost control; most subsequently gladly gave up their risk arrangements and reverted back to fee-for-service.

	A middle ground between full risk and fee-for-service would be to adopt risk corridors on top of risk payment so that the downside financial exposure is limited and manageable. Risk corridors would limit to some extent the profits or losses the ACOs would incur if their spending deviated beyond preset limits from the spending target. In essence, Medicare would share both savings and losses with the ACOs. These risk corridors should probably apply both to spending on individual patients and to spending in the aggregate, much as stop-loss reinsurance works. Large ACOs based in hospitals might want to purchase reinsurance themselves and assume full risk for their Medicare patients, although organizations taking full risk raise solvency and beneficiary protection issues that might require an MA-like regulatory regime. For most ACOs, Medicare would essentially become the reinsurer by having the risk corridors, in an approach similar to what was set up for Part D prescription drug plans.The payments to physicians might continue to be fee-for-service based on existing reimbursements for services rendered with reconciliation later (as under the PGP demonstration) or actual per capita payments, with the risk corridors. Such a shared-savings and shared-risk approach would simulate capitation (now often called “global” or “comprehensive” payment to avoid the stigma often associated with the term “capitation”), with the marginal incentive to avoid provision of unneeded services clear. However, because of the risk corridors, and because the financial stakes are not as great as they are under the MA program, many organizations would likely be willing to participate. This approach would meet what should be a key objective of ACOs and the Medicare Shared Savings Program: to fundamentally change the business case for providers such that an unnecessary hospital admission, test, or invasive procedure is considered an avoidable expense rather than profitable revenue.

 Recommendation #3: Decrease Variation in Spending Targets

	The PGP-style payment approach contemplated for the Medicare Shared Savings Program in the legislation would accept ACOs’ historical spending levels for assigned patients as the base for determining the ACO’s applicable spending target, presumably to stimulate provider interest in trying the ACO approach. But this approach ignores the substantial variation in wasteful, baseline spending and also the reality that there is a somewhat negative correlation between baseline spending levels and rates of spending growth

The Shared Savings Program design is at risk of not providing a sufficiently strong business case to convince provider organizations to change their practice cultures.

October 2010

Costs of Heart Failure-Related Hospitalizations in Patients Aged 18 to 64 Years

Retail Clinic Versus Office Setting: Do Patients Choose Appropriate Providers?

Hospital Readmission Among Participants in a Transitional Case Management Program

Shared Savings Program for Accountable Care Organizations: A Bridge to Nowhere?

Scheduled and Unscheduled Hospital Readmissions Among Patients With Diabetes

Variation in Prescription Use and Spending for Lipid-Lowering and Diabetes Medications in the Veterans Affairs Healthcare System

Mental Health Services Utilization Among Women Experiencing Intimate Partner Violence

Effect of a Patient Panel-Support Tool on Care Delivery

Initial Experience With a First-to-Market Member Accountability-Based Insurance Product

Assessing the Accuracy of Drug Profiles in an Electronic Medical Record System of a Washington State Hospital