 To illustrate the use of cluster analysis for identifying sub-populations of complex patients who may benefit from targeted care management strategies.

 We identified a cohort of adult members of an integrated health maintenance organization who had 2 or more of 17 common chronic medical conditions and were categorized in the top 20% of total cost of care for 2 consecutive years (n = 15,480). We used agglomerative hierarchical clustering methods to identify clinically relevant subgroups based on groupings of coexisting conditions. Ward’s minimum variance algorithm provided the most parsimonious solution.

 Ward’s algorithm identified 10 clinically relevant clusters grouped around single or multiple “anchoring conditions.” The clusters revealed distinct groups of patients including: coexisting chronic pain and mental illness, obesity and mental illness, frail elderly, cancer, specific surgical procedures, cardiac disease, chronic lung disease, gastrointestinal bleeding, diabetes, and renal disease. These conditions co-occurred with multiple other chronic conditions. Mental health diagnoses were prevalent (range 28% to 100%) in all clusters.

 Data mining procedures such as cluster analysis can be used to identify discrete groups of patients with specific combinations of comorbid conditions. These clusters suggest the need for a range of care management strategies. Although several of our clusters lend themselves to existing care and disease management protocols, care management for other subgroups is less well-defined. Cluster analysis methods can be leveraged to develop targeted care management interventions designed to improve health outcomes.

	This study illustrates the use of cluster analysis to identify sub-populations of complex patients for potential targeted care management within an integrated health maintenance organization.

		Among a cohort of adults with multimorbidity and high healthcare utilization, we identified 10 clinically relevant clusters of complex patients.

		While care management protocols may already exist in many healthcare settings for some common clusters, other clusters identified present opportunities for new or enhanced care management.

		Data mining methods such as cluster analysis can be applied in other settings where electronic diagnosis data are readily available.

	By 2020, over 81 million persons in the United States will have 2 or more chronic conditions.

 Multimorbidity results in adverse health outcomes and higher healthcare costs, and challenges current models of care delivery.

 Care management has the potential to improve health outcomes for persons with multimorbidities. However, most disease and care management strategies have been developed to improve specific health outcomes for populations defined by single diseases or specific circumstances (such as hospital discharge).

 There is a need for strategies that can identify sub-populations with multiple, interacting diseases, in order to provide them with appropriate and relevant care management support.

	Investigations to identify these populations of complex patients have traditionally relied upon multivariable regression analyses to identify patient-level characteristics (such as demographics and diseases) that predict the outcome of interest (such as hospitalization).

 As compared with investigations that use multivariable regression analyses to identify individual disease predictors of specific outcomes, data mining techniques provide an opportunity to empirically identify groups of patients with similar patterns of multimorbidities. One such technique, cluster analysis, refers to classification methods that are used for discovering groups or “clusters” of “highly similar entities” within data sets.

 Cluster analyses are common in psychology, sociology, and marketing research, and the methods have been used to a limited extent in health services research.

 While cluster analyses previously have been used to discover patterns of multimorbidities,

 in this study we demonstrate the application of such methods for identifying clusters of patients with high utilization that may suggest opportunities for enhanced care management in a managed care setting.

	We used cluster analysis to explore a large, 2-year cohort of health maintenance organization members with 2 or more chronic conditions. We hypothesized that within a large, complex patient population, cluster analysis would reveal groups of patients with distinct patterns of comorbid conditions. Although some of these subgroups would be characterized by well-known patterns of co-occurring medical conditions with established care management strategies, other subgroups would reveal combi-nations of comorbidities that might benefit from new, proactive, and targeted care management.

	Kaiser Permanente Colorado (KPCO) is an integrated, not-for-profit health maintenance organization. During the years studied (2006 and 2007), KPCO had approximately 430,000 members. This study was approved by KPCO’s Institutional Review Board.

	The study population consisted of KPCO members 21 years or older on January 1, 2006, categorized in the top 20% of total cost of care in both 2006 and 2007, each with 2 or more of 17 common chronic medical conditions. Annual cost estimates combined general ledger costs with direct and indirect utilization-related costs to provide cost-of-care estimates for KPCO members.

 We excluded members with a long-term care facility stay, chronic kidney dialysis, or an inpatient visit of greater than 30 days during the 2 years based on the premise that their unique and significant care management needs are likely to already be well defined. Six extremely high cost outliers were also removed from the cohort.

	We compiled a list of 17 chronic medical conditions based on prevalence in the general population, prevalence in our specific cohort, and a literature search of conditions likely to predict hospitalization or adverse health outcomes in complex patients.

 The selected conditions were diabetes, chronic obstructive pulmonary disease (COPD), chronic kidney disease, stroke, obesity, dementia, fall, hip fracture, chronic pain, skin ulcer, orthopedic surgery, back surgery, abdominal surgery, gastrointestinal bleeding, cancer (excluding non-melanoma skin cancer), cardiac disease (which included coronary artery disease and congestive heart failure), and mental health conditions—primarily depression, but also including generalized anxiety and bipolar disorders. Determinations of whether cohort members had a given condition were based on inpatient and outpatient 

International Classification of Diseases, Ninth Revision (ICD-9)

 diagnosis and procedure codes in 2006 and 2007. In addition, we used KPCO’s cancer registry to determine cancer diagnoses in 2005, 2006, and 2007. We considered a cohort member to have obesity if they had an 

 diagnosis code for obesity or a median body mass index (BMI) greater than or equal to 30 in 2006 and 2007. BMI data were available for 98.6% of cohort members; if a cohort member did not have a BMI value or an obesity diagnosis in 2006 or 2007 then we did not consider them to be obese.

	SAS version 9.2 (SAS Institute, Cary, North Carolina) was used for all analyses. We described demographic attributes, healthcare utilization, comorbidity score (using the Quan adaptation of the Elixhauser comorbidity index),

 and prevalence of clinical conditions within the cohort using frequencies and medians with 25th and 75th percentiles.

 We used agglomerative hierarchical clustering to identify clinically relevant groups of cohort members with similar multimorbid conditions. With this method of cluster analysis, each cohort member starts as its own cluster. The 2 most similar clusters are merged and this new cluster replaces the 2 former clusters. The process continues until there is only 1 cluster containing all observations.

 After the clustering algorithm is run, the user must select the appropriate cutoff point for the number of clusters desired based on clinical importance or other pre-specified criteria.

 Various algorithms are available for cluster analysis. For this study, we used Ward’s minimum variance method as the primary algorithm. With this algorithm, every possible cluster combination is considered at each step of agglomerative hierarchical clustering, and the combination thatresults in the smallest addition to the error sum of squares is selected.

 Ward’s method is a widely used algorithm which minimizes the variance within clusters, and is also known to produce clusters of similar sizes.

 We compared results from Ward’s method to results using the flexible beta algorithm, where the user sets different levels of beta, and beta values less than zero optimize the dissimilarity between clusters.

 In the analytic data set, the presence or absence of each of the 17 conditions was represented with a 1 or 0 for each cohort member. We first randomly split the full analytic data set into 2 equally sized data sets. We then converted each split data set into a dissimilarity matrix using Jaccard’s coefficient. This is an appropriate distance measure for clinical conditions, as it considers the number of conditions that 2 people have in common and ignores conditions that neither person has.

	We ran our primary algorithm, Ward’s minimum variance method, on both split data sets. The pseudo F, pseudo T, and 

 statistics were examined for different numbers of clusters to identify possible clustering solutions.

 These statistics pointed to several desirable numbers of clusters, and membership in these clusters was described by examining the prevalence of each condition in the cluster. Cluster membership was compared between the 2 split data sets to assess the consistency of the clustering process. Since cluster membership was similar between the 2 split data sets, thus reinforcing the stability of the algorithm in this population, Ward’s algorithm was run on the entire data set. We subjectively determined that a 10-cluster solution produced the most clinically relevant clusters. For comparison, we then produced 10 cluster solutions using the flexible beta method, with beta set at -0.25 and -0.5, and compared these results with Ward’s method. This confirmed that Ward’s algorithm resulting in 10 clusters appeared to be the most parsimonious solution and provided the most clinically relevant groups. We then described the 10 clusters by the number of cohort members in the cluster, median age of cluster members, and percentage of cluster members with the most prevalent conditions in that cluster. We also described relative cost of care ratios for each cluster.

 provides descriptive demographic and disease characteristics of the study cohort (n = 15,480). The median age of cohort members was 65 years, and 59.1% of cohort members were women. Cohort members had a median of 5 chronic medical conditions (including, but not limited to, the 17 conditions included in the cluster analysis).

	In this study we tried 3 clustering algorithms—Ward’s, flexible beta with beta set to -0.25, and flexible beta with beta set to -0.5. Full results from the 3 methods are in the 

. All 3 methods identified some of the same distinct clusters, including clusters of patients with chronic pain and mental health conditions and gastrointestinal bleeding and mental health conditions. However, we preferred Ward’s method because the flexible beta methods tended to produce many small homogeneous clusters and several large heterogeneous clusters, while Ward’s method produced more similarly sized and more homogeneous clusters. Eight of the 10 clusters produced by Ward’s method had what we referred to as “anchoring” conditions, or chronic conditions shared by almost all (>98%) of cluster members. We considered such anchoring conditions potentially useful given the ultimate goal of proactively identifying characteristics of persons who may benefit from targeted care management. With the flexible beta results, only 5 and 7 of the clusters from the method with beta set at —0.25 and –0.5, respectively, had anchoring conditions.

 provides a description of cluster membership using Ward’s algorithm. Of the 10 clusters, several described groups of patients with well-known and highly prevalent comorbidities such as diabetes and obesity, cardiac disease and obesity, renal disease and diabetes, and the multiple diseases and conditions common in the frail elderly. Other clusters were defined by combinations of comorbidities that are less frequently described including abdominal and orthopedic surgeries with obesity. Two clusters included individuals who were substantially younger than the others: mental health conditions and chronic pain, and mental health conditions and obesity. Two single conditions were highly prevalent across all groups: mental health conditions (primarily depression) and obesity. These conditions were present in all clusters with prevalence rates that ranged from 28% to 100% (mental health conditions) and from 38% to 100% (obesity). Prevalence of all 17 conditions in each cluster is available in the Appendix, as are relative cost ratios.

	In this investigation, we demonstrated the use of cluster analysis to identify distinct subgroups of patients with specific combinations of co-occurring conditions in a managed care population. Exploratory by nature, cluster analysis provided a unique method for investigating the co-occurrence of multiple conditions. When working with large data sets, simple tabulations of the number of chronic conditions can be difficult to interpret. In our population in which cohort members had at least 2 of 17 conditions of interest, there are theoretically over 100,000 different possible combinations of coexisting conditions. In actuality, our cohort only had 1507 different combinations of the 17 conditions of interest. Even this number of naturally occurring clusters is much too large to allow identification of targeted care management opportunities.

	Compared with people with only 1 chronic condition, persons with multiple chronic conditions experience worse health outcomes, including, but not limited to, lower quality of life, poorer functional status, and excess morbidity and mortality.

 Chronic conditions do not occur in isolation, yet many existing care management strategies are directed toward single conditions. Cluster analysis provided a data-driven approach to identifying 10 clinically relevant groups of patients with patterns of comorbidities that could be targeted with new, enhanced care management strategies.

	In our analysis, the largest cluster was characterized by older age coupled with conditions associated with frailty, such as stroke and dementia, along with chronic conditions including diabetes, obesity, and cardiac disease. Along with this group, our analysis produced several other clusters of wellknown, high-cost co-occurring conditions for which chronic disease management programs are widespread, such as cancer with obesity and/or mental health conditions (primarily depression) and cardiac disease co-occurring with obesity and diabetes. Identification of these expected groups in our analysis provided additional assurance of the validity of our data mining method.

	Our analysis produced several clusters that suggest potential care management opportunities in our managed care setting, especially in relatively younger adults. Two clusters from this cohort of individuals had median ages of 50 years (cluster 4) and 55 years (cluster 1). All 1676 members of cluster 4 had both mental health conditions and obesity, but none of the other 15 conditions of interest. This highly homogeneous cluster represented 10.8% of the entire study cohort. Individuals in cluster 1 were characterized by chronic pain co-occurring with mental health conditions and/or obesity. Chronic pain is associated with depression and anxiety, and has been linked to increased utilization of health resources.

 Coordinated care strategies to relieve depression and improve chronic pain outcomes have been evaluated to a limited extent.52,53 Cohort members in cluster 1 also had the third highest median total cost of care of all clusters, after the clusters of cancer and surgical patients. Identification of these 2 clusters supports the notion that multimorbidity and patient complexity are not limited to the elderly.

 It is likely that the relatively younger adults in these 2 groups would benefit from care management that emphasizes integrated depression care.

 Finally, we noted a high prevalence of mental health conditions and obesity across all clusters, suggesting that these 2 conditions should be addressed in most, if not all, comprehensive care management programs.

	Our study has several limitations. Cluster analysis is an exploratory classification method that is supported by a relatively small body of statistical evidence, and different clustering algorithms produce different results.

 Knowing this inherent problem, we tried several algorithms. We chose the ultimate clustering solution based on subjective review and clinical relevance of cluster membership. In addition, agglomerative hierarchical clustering forces everyone into a cluster. There are likely small groups of patients within some of our clusters who have unique combinations of comorbid conditions that differ from the majority of people in that cluster. To assess the consistency of the clustering process within our patient population, we compared results from a randomly split data set of cohort members in 2006 and 2007. To further assess the stability of these clusters over time, analyses should be conducted on cohorts from different years. As with any investigation, the characteristics of our clusters are limited to our data and setting. Replicating these analyses in other settings and other patient populations may potentially reveal different clusters. However, these differences would and should inform management strategies specific to populations in those settings.

	An additional limitation is that we identified the 17 conditions of interest based on electronic diagnosis and procedure data in 2006 and 2007, with the exception of cancer, for which we also included 2005 data and used a cancer registry, and obesity, for which we used BMI data. We anticipate that for most chronic conditions, such as COPD or diabetes, the member would have had at least 1 diagnosis in these years; however, the sensitivity and specificity of diagnostic data are imperfect, leading to potential misclassification of conditions. Lastly, it is likely that individuals in our cohort carried multiple other diagnoses not specifically assessed. Based on the Quan adaptation of the Elixhauser comorbidity index, cohort members had a median of 5 chronic conditions.

 We selected the 17 conditions examined in this study based on prevalence in the general population and in our specific cohort, and on prior knowledge of or association with increased healthcare utilization; however, it is likely that including other conditions in the analysis would result in different clusters or subgroups.

	In this investigation, we demonstrated how cluster analysis can be used to identify homogeneous groups of complex patients from a large heterogeneous population. Such data mining methods can be applied in other settings where electronic diagnosis data are readily available. Alternatively, it is possible to use the conceptual results from this investigation to increase awareness of the need for a diverse array of care management services for individuals with high levels of healthcare utilization. However, further understanding of the care management needs of clusters of patients with similar comorbidities is warranted before designing specific tailored interventions. For example, the specific needs of persons with mental health conditions and obesity should be explored in detail in order to develop relevant care management strategies for that group.

	Cluster analysis methods can be leveraged for targeted care management interventions designed to improve health outcomes and potentially lower healthcare costs. This cluster analysis of a large cohort of individuals with multiple morbidities suggests that complex patients with high healthcare utilization represent a highly diverse group of individuals. While some subgroups may respond well to existing approaches to care management (such as those designed for populations with diabetes, cardiac disease, and frailty), others will likely require new and/or individualized care management strategies to achieve favorable health outcomes.

 From Institute for Health Research (SRN, JFS EAB), Kaiser Permanente Colorado, Denver; Department of Family Medicine (EAB), University of Colorado, Aurora.

Articles

WEB EXCLUSIVE

Cluster analysis can aid in identifying subgroups of patients with similar patterns of comorbid conditions for targeted care management.

Identifying Subgroups of Complex Patients With Cluster Analysis

To compare healthcare costs and resource utilization among patients with post-traumatic stress disorder (PTSD) vs control subjects with major depressive disorder (MDD) in populations covered by Medicaid or private insurance.

 Retrospective analysis of Medicaid and private insurance administrative claims data.

Patients with at least 2 PTSD diagnoses during or after 1999, and at least 1 PTSD diagnosis during or after 2003, were identified from deidentified Medicaid claims from Florida, Missouri, and New Jersey (1999-2007) and from a privately insured claims database (1999-2008). Patients had continuous eligibility 6 months before (baseline) and 12 months after (study period) the index date and were aged 18 to 64 years. Potential control subjects having MDD without PTSD diagnosis were identified using similar selection criteria. Control subjects with MDD were matched to patients with PTSD on age, sex, state or region, employment status (private insurance only), index year, and race/ethnicity (Medicaid only). Study period per-patient utilization and costs, calculated as reimbursements to providers for medical services and prescription drugs, were compared using univariate and multivariate analyses.

 Patients with PTSD had higher rates of other mental health disorders (eg, anxiety and bipolar disorder) and higher mental health—related resource use and costs than control subjects with MDD in both Medicaid and privately insured populations. The mean study period total direct healthcare costs were higher for patients with PTSD than for control subjects with MDD ($18,753 vs $17,990 for Medicaid and $10,960 vs $10,024 for private insurance, 

 <.05 for both). The difference in total direct costs was driven by higher mental health—related resource use for patients with PTSD.

 Patients having PTSD had 4.2% to 9.3% higher mean annual per-patient healthcare costs compared with matched control subjects having MDD among patients covered by Medicaid or private insurance.

	While post-traumatic stress disorder (PTSD) is common among Veterans Affairs (VA) patients, it is costly even to payers outside the VA system, including those covered by Medicaid or private insurance.

		 A comparison of patients diagnosed as having PTSD vs demographically matched patients diagnosed as having major depressive disorder (MDD) suggests that patients with PTSD had 4.2% to 9.3% higher annual per-patient healthcare costs among patients covered by Medicaid or private insurance.

		 Patients diagnosed as having PTSD had higher mental health—related resource utilization and costs compared with control subjects having MDD; the higher mental health–related costs were the primary driver of the cost difference.

	Post-traumatic stress disorder (PTSD) is an anxiety disorder that may develop following a traumatic event or series of events (eg, car accident, military combat, natural disaster, or assault). Patients with PTSD usually are seen with the following 3 categories of symptoms: reexperiencing (eg, flashbacks and nightmares), avoidance (eg, staying away from activities that remind them of the trauma and developing emotional numbness), and hyperarousal (eg, difficulty sleeping and constantly feeling on edge). Persistent presence of these symptoms lasting more than 1 month causing distress or impairment of functioning may indicate that a patient has PTSD.

	The National Comorbidity Survey Replication estimated that the lifetime prevalence of PTSD is 6.8%, and the 12-month prevalence is 3.5% among general adults in the United States.

 Even after controlling for the type and degree of trauma, studies

report that PTSD occurs approximately twice as frequently in women compared with men. Studies estimated high PTSD prevalence rates among military veterans (16.6%)

 and among urban primary care patients (23%).6 In contrast, analyses of mid to late 1990s Medicaid claims data estimated the annual prevalence of PTSD among the general adult US population to be much lower, at 0.4% to 0.5%,

 possibly because PTSD cannot be captured among individuals not seeking medical care.

	Post-traumatic stress disorder is associated with other psychiatric comorbidities, including substance abuse or dependence, major depressive disorder, panic disorder, agoraphobia, generalized anxiety disorder, obsessive-compulsive disorder, eating disorder, social phobia, and bipolar disorder.

An analysis of the National Comorbidity Survey by Kessler

suggested that PTSD may cause secondary mental health disorders: respondents with PTSD were more likely to develop anxiety, mood, and substance disorder than other respondents without PSTD, and the elevated risk of secondary mental health disorders disappeared with remission of PTSD. Post-traumatic stress disorder is associated with increased risk of congestive heart disease and with higher rates of hypertension, arthritis, asthma or bronchitis, kidney or liver disease, thyroid or autoimmune disease, stomach or gallbladder problems, epilepsy or neurologic disorders, and ulcers.

describing healthcare utilization by patients with PTSD suggest that PTSD is associated with higher resource use relative to that by patients without any mental health disorders or by patients with non-PTSD mental health disorders. Compared with veterans without mental health diagnoses, veterans diagnosed as having PTSD had higher adjusted relative risk (aRR) of using primary care services (aRR, 1.3), medical or surgical subspecialty services (aRR, 1.6), ancillary services (aRR, 1.4), emergency services (aRR, 1.6), and inpatient services (aRR, 3.2).

 report healthcare costs of patients with PTSD in the United States. A study

 of female health maintenance organization members found that costs increased with increasing severity of PTSD symptoms; women having more severe PTSD symptoms had about twice the annual costs compared with women having less severe symptoms after adjusting for comorbidities and demographics ($3060 in 1997 US dollars among women with high scores, $1779 among women with median scores, and $1646 among women with low scores). Another study

 found that 12-month mental healthcare costs of patients having major depressive disorder (MDD) with PTSD were higher than those of patients having MDD without PTSD ($1196 vs $332).

	Given the limited research available in the literature, the objective of the present study was to characterize the healthcare resource utilization and direct (medical and pharmaceutical) costs among patients with PTSD in Medicaid and privately insured populations. Based on the comorbidity burden and higher resource use reported in previous studies,

it was expected that patients with PTSD would have higher costs compared with average patients without PTSD and that a more relevant comparison might be another well-studied mental health condition. Because depression often occurs after PTSD and the symptoms can overlap,

 the study sought to compare patients with PTSD vs matched control subjects with MDD, a well-characterized and costly population.

 The estimated lifetime prevalence of MDD among the adult US population is 16.2%, and the 12-month prevalence is 6.6%.

 The disorder is often comorbid with other psychiatric comorbidities. Among patients with lifetime prevalence of MDD, 72.1% had another mental health comorbidity, 59.2% had anxiety disorder, 24.0% had substance use disorder, and 30.0% had impulse control disorder.

 Annual healthcare costs per patient were estimated to be $8368 (in 1998 US dollars) among patients with treatment-resistant depression, $3571 among patients with depression that is not treatment resistant, and $2359 among a random sample of patients from a private insurance database.

	Two patient populations were analyzed in this study, one covered by Medicaid and the other by private insurance. The Medicaid study sample was selected from Medicaid claims databases in 3 states (Florida, Missouri, and New Jersey), covering approximately 12 million lives (1999-2007). The privately insured study sample was selected from a private insurance administrative database (Ingenix Employer Solutions, Rocky Hill, Connecticut) that included approximately 12 million beneficiaries (including employees, spouses, and dependents) from 55 large self-insured companies in the United States with claims for services provided from 1999 to 2008. Collectively, the companies have operations nationwide in a broad array of industries and job classifications.

	The Medicaid and privately insured data contain deidentified information on patient demographics (eg, age and sex), monthly enrollment history, and medical and pharmacy claims. Medical services use was recorded, with dates of service, billed charges, actual amounts paid to providers, procedures performed (

 codes for privately insured data using the 

International Classification of Diseases, Ninth Revision, Clinical Modification [ICD-9-CM])

. The databases also include pharmacy claims, with prescribed medications identified by National Drug Code, date of prescription fill, days of supply, quantity, and actual payment amounts.

	Patients were identified as having PTSD if they had at least 2 claims for PTSD (

 code 309.81) occurring on 2 different dates during or after 1999, with at least 1 claim for PTSD during or after 2003. Two PTSD diagnoses on 2 different dates were required to identify patients with PTSD. Because the duration of symptoms in patients who are receiving treatment is approximately 36 months and rises to 5 years or longer in patients not receiving treatment,

 the criterion of being positive for PTSD at 2 time points is an approximate validation of the diagnosis. The index date was defined as the first PTSD diagnosis during or after 2003 that was not the first overall PTSD diagnosis. To focus on prevalent rather than newly diagnosed patients with PTSD, the index date was selected as a PTSD diagnosis that was not the first PTSD diagnosis. Only patients with index dates starting in 2003 were selected so that healthcare resource use and costs reflected more recent treatment patterns. To ensure that complete claims data were available, patients with PTSD were required to have continuous healthcare coverage eligibility during the 6 months before the index date (baseline period) and during the 12 months following the index date (study period). Patients were required to be aged 18 to 64 years throughout the study period. Patients with PTSD having health maintenance organization, capitated, or Medicare coverage were excluded from the analysis because payment information may be incomplete for those patients.

	Control subjects with MDD were selected using similar criteria. Controls included beneficiaries with no diagnosis of PTSD in the available data and with at least 2 claims for MDD (

 code 296.2 [MDD, single episode] or 296.3 [MDD, recurrent episode]) occurring on 2 different dates during or after 1999, with at least 1 claim for MDD during or after 2003. The index date for controls with MDD was defined as the first MDD diagnosis during or after 2003 that was not the first overall MDD diagnosis. Controls with MDD were also required to have continuous healthcare coverage eligibility during the 6 months before the index date (baseline period) and during the 12 months following the index date (study period), be aged 18 to 64 years throughout the study period, and have no health maintenance organization, capitated, or Medicare coverage. Note that this does not require the controls with MDD to have an MDD diagnosis during the 6-month baseline period.

	Patients having PTSD were randomly matched one-toone to controls having MDD within each of the databases on age, sex, state or region, employment status (privately insured only), index year, and race/ethnicity (Medicaid only). Almost all (99.9%) identified patients with PTSD were matched to controls with MDD. Patients who were not matched were generally younger, in the range of 18 to 19 years.

	Patients with PTSD and controls with MDD were compared on the following characteristics: demographics (age, sex, state or region, and race/ethnicity [the latter for Medicaid only]), employment status (privately insured only), and baseline rates of selected comorbidities over the 6-month baseline period. Baseline comorbidities included the following: mental health disorders, such as anxiety (excluding PTSD), bipolar disorder, depression (MDD or other depression), substance abuse or dependence, insomnia, and schizophrenia; chronic pain conditions, such as arthritis, back or neck pain, fibromyalgia, and migraine; and the Charlson Comorbidity Index (CCI), which includes 17 physical conditions predictive of 1-year mortality (including common conditions like cancer, diabetes mellitus, and cardiovascular disease).

	Healthcare costs, including medical service and prescription drug costs, were calculated for the 12-month study period. Cost analyses were conducted from the payer’s perspective (ie, costs were defined as Medicaid or private insurer payments to providers). Medical service costs were calculated for inpatient, emergency department (ED), and outpatient or other medical services. Prescription drug costs were calculated as mental health—related prescriptions (ie, antidepressants, antipsychotics, anxiolytics, anticonvulsants, and hypnotics or sedatives), cardiovascular drugs, antidiabetic drugs, gastrointestinal drugs, and other prescription drugs. Direct healthcare costs included both mental health–related and all-cause costs. Mental health–related costs included costs for mental health–related drugs, medical claims with a primary or secondary mental health disorder diagnosis (

 codes 290-319), and outpatient psychiatric treatment.

	All costs were inflated to 2008 US dollars. These were obtained using the Consumer Price Index for medical care.

	Medical services were classified into the following mutually exclusive categories: inpatient care, ED visits (identified using 

 procedure codes associated with ED care and place and type of service codes), and outpatient or other services. Prescription drugs were identified and classified using National Drug Codes from a pharmacy database (Master Drug Database; MediSpan, Indianapolis, Indiana). Resource utilization also included mental health— related and all-cause resource use.

	Patient baseline characteristics, study period resource use, and costs were compared between beneficiaries treated for PTSD and matched controls with MDD. Categorical variables were compared using 

 test was used for univariate comparisons of resource use counts and costs.

	Multivariate models were used to compare risk-adjusted mental health—related costs, total direct costs, and mental health–related and all-cause hospitalization rates between patients with PTSD and controls with MDD. The models adjusted for differences in patient characteristics, such as age, sex, state or region, CCI, and baseline comorbidities of back or neck pain, arthritis, schizophrenia, substance abuse or dependence  fibromyalgia, and migraine among the Medicaid population. Schizophrenia, substance abuse or dependence, arthritis, fibromyalgia, and migraine were not included as covariates in the model for the privately insured sample because the prevalences of these baseline diagnoses were low among the privately insured participants with PTSD or MDD. Logistic regression analysis was used to compare hospitalization rates. A generalized linear model with log link and gamma distribution for the error term was used to compare costs between patients with PTSD and controls with MDD.

	All analyses were conducted using commercially available statistical software (SAS version 9.2; SAS Institute Inc, Cary, North Carolina). 

 <.05 was considered statistically significant.

Baseline Characteristics and Comorbidities

	The prevalence of PTSD in this study was 0.5% among the Medicaid sample (1999-2007) and 0.2% among the privately insured sample (1999-2008; 

). The study samples included 9114 Medicaid patients diagnosed as having PTSD and 9720 privately insured patients diagnosed as having PTSD. Medicaid patients, compared with privately insured patients, were younger (37.5 vs 43.3 years) and had a higher proportion of women (78.9% vs 70.0%; 

). Medicaid patients having PTSD had a significantly lower mean CCI compared with matched controls having MDD (0.55 vs 0.69;

 <.001), while there was no significant difference in CCI between the privately insured patients with PTSD and the controls with MDD (0.31 vs 0.29, 

 = .20). Medicaid patients having PTSD had higher rates of mental health—related and chronic pain comorbidities compared with privately insured patients having PTSD, notably schizophrenia (11.1% vs 1.0%), bipolar disorder (25.4% vs 8.1%), and drug abuse or dependence (18.3% vs 3.6%). Both Medicaid and privately insured patients having PTSD had significantly higher baseline rates of anxiety (excluding PTSD), bipolar disorder, substance abuse or dependence, and chronic pain conditions compared with controls having MDD.

All-Cause and Mental Health—Related Resource Use

, during the 12-month study period, patients with PTSD had significantly higher rates and mean numbers of days or visits for almost all measures of mental health—related resource use. The measures for all-cause healthcare resource use showed more modest differences between patients with PTSD and matched controls with MDD.

	For all-cause healthcare resource use, Medicaid patients having PTSD had a significantly higher proportion of patients with at least 1 ED visit (63.7% vs 60.0%), mean number of total ED days per patient (3.4 vs 2.9 days), and total outpatient or other days per patient (166.5 vs 157.8) compared with controls having MDD (Table 3). Privately insured patients having PTSD had a significantly higher proportion of patients with at least 1 inpatient stay (19.9% vs 18.7%) and 1 ED visit (32.2% vs 28.4%), as well as a significantly higher mean number of total ED days per patient (0.9 vs 0.8) and total outpatient or other days per patient (35.0 vs 28.6) compared with controls having MDD. Privately insured patients with PTSD had a signnificantly lower proportion of mental health—related prescriptions, which was driven by a lower rate of antidepressant use.

	For mental health—related resource use, patients having PTSD had a significantly higher rate and mean annual number of days with ED, outpatient or other, and psychiatric or psychological treatment visits, as well as a higher mean number of hospitalization days compared with controls having MDD. Patients diagnosed as having PTSD had a significantly higher proportion of prescriptions for antipsychotics (58.5% vs 48.9% for Medicaid and 18.6% vs 14.8% for privately insured) and anticonvulsants (57.1% vs 49.2% for Medicaid and 29.9% vs 26.2% for privately insured) compared with controls having MDD (

). Medicaid patients with PTSD also had a significantly higher proportion of patients with at least 1 prescription for anxiolytics and hypnotics or sedatives. Privately insured patients with PTSD had a significantly lower proportion of patients with at least 1 prescription for antidepressants (62.5% vs 78.9%).

Unadjusted All-Cause and Mental Health—Related Costs 

	Annual total direct healthcare costs were $18,753 per Medicaid patient with PTSD, 4.2% higher than for controls with MDD ($17,990), while privately insured patients with PTSD had total healthcare costs that were 9.3% higher than those of controls with MDD ($10,960 vs $10,024; 

). The 3 largest cost components of all-cause total healthcare costs for patients with PTSD were outpatient or other services (45.8% for Medicaid and 49.8% for privately insured), inpatient services (22.9% for Medicaid and 24.3% for privately insured), and mental health—related prescription drugs (17.3% for Medicaid and 10.0% for privately insured).

Post-traumatic stress disorder was associated with 4.2% to 9.3% higher annual per-patient healthcare costs compared with MDD among patients covered by Medicaid or private insurance.

Cost of Post-Traumatic Stress Disorder vs Major Depressive Disorder Among Patients Covered by Medicaid or Private Insurance

 This report presents evidence about “what works” in value-based purchasing strategies emphasized in national healthcare reform legislation, with a focus on service delivery integration, payment, and value-based insurance design. We review key findings to address the question: What is the evidence that a value-based strategy will improve health outcomes or reduce healthcare costs?

 This report is a review of peer-reviewed and field-based research associated with several value-based strategies purchasers can use to increase value in their healthcare market decision making: Accountable Care Organizations, Patient Centered Medical Homes, bundled payment methods and incentives, and value-based insurance designs.

 Environmental scans and evidence reviews of value-based purchasing were conducted to inform this report.

 Many value-based approaches are new to the healthcare arena, and evidence of performance in improving health outcomes or reducing healthcare costs is limited

but increasing as experience with these approaches grows.

 Purchasers should be aware that evidence is contingent on market environment and study assumptions in considering specific strategies, and be cautious implementing them. The need for a new measurement paradigm for cost and quality that reaches across the care continuum is evident.

	Value-based purchasing approaches including service integration, payment methods, and value-based insurance design are emphasized in national healthcare reform legislation, but evidence of “what works” varies by strategy and is often inconclusive.

		Purchasers should be aware of evidence about performance, improving health outcomes, or reducing healthcare costs in considering specific strategies, and be cautious implementing them.

		Many value-based approaches are new to the healthcare arena, and evidence of performance is limited but increasing as experience with these approaches grows.

A criticism of the 2010 national healthcare reform, enacted in the Patient Protection and Affordable Care Act (PPACA) and amended by the Health Care and Education Reconciliation Act, is that it does not address cost control. The national healthcare reform, however, did make several forays into encouraging value in healthcare purchasing. The message from President Barack Obama’s administration is that healthcare reform efforts should pursue value, not volume purchasing, and quality, not quantity, in healthcare services.

 A value-based strategy strives for improved healthcare quality and health outcomes while achieving efficiency. So, what is the evidence that a value-based strategy will improve health outcomes or reduce healthcare costs?

The PPACA legislation calls for implementation of demonstrations, pilots, and programs involving accountable care organizations (ACOs), patient-centered medical homes (PCMHs), payment bundling, and pay-for-performance (P4P) initiatives; extends a gain-sharing demonstration; and, in section 2713(a), calls for the creation of policy to “permit a group health plan and a health insurance issuer offering group or individual health insurance coverage to utilize value-based insurance designs.” We address 3 areas of value-based purchasing developed in the legislation of interest to purchasers: service delivery integration, payment, and value-based insurance design.

Service delivery integration is not a new concept in healthcare. Past examples of similar efforts range from the comprehensive planning legislation of the 1970s to Medicaid’s primary care case management model and from managed care’s integration and behavioral health carve-outs to physician hospital organizations. The results of these early experiences were inconsistent, due to challenges with implementation and measurement of cost savings and outcomes. As of late, there are high hopes for 2 models of service delivery integration: ACOs and PCMHs, which may employ payment bundling methodologies or financial incentives, such as P4P or gain sharing, to encourage providers to reduce cost and improve quality. In theory, these models generate cost savings from reduced hospitalizations and readmissions.

Accountable care organizations are networks or sets of providers and institutions, such as primary care physicians, specialists, and hospitals, which have joint responsibility for the quality and cost of care for a population. This care delivery model combines integration of care across providers to improve coordination and reduce duplication of services, and provides incentives for shared savings and insurance design focused toward population health improvement as strategies to enhance healthcare value. Congress mandated in 2000 that the Centers for Medicare and Medicaid Services (CMS) conduct the Physician Group Practice (PGP) Demonstration to test an ACO-like hybrid payment methodology that combines Medicare feefor- service payments with new incentive payments. The PGP encouraged the coordination of Medicare Part A and Part B services, promoted efficiency through investment in administrative processes, and rewarded physicians for improving health outcomes.

 A US Government Accountability Office evaluation concluded that the care coordination programs initiated by the participants showed promise, but that wider applicability of the payment methodology is questionable. In particular, the large size of the 10 participating physician groups, all with more than 200 physicians, gave the participants “certain size-related advantages that may make broadening the payment approach used in the demonstration to other physician groups and non-group practices challenging.” 

 Beyond the CMS PGP demonstration, few studies have rigorously examined whether ACO-type models do, in fact, improve quality and reduce costs. Given the recent introduction of ACOs, research in this area is limited and additional analysis is needed.

Delivery models where each patient has an ongoing relationship with a primary care provider who leads a team that takes collective responsibility for the patient’s care, PCMHs infuse traditional primary care approaches with service delivery integration through interdisciplinary team-based care coordination to reduce duplicative care, and related payment innovations to control costs and provide quality care. Because PCMH models coordinate the spectrum of patient care needs, they are well suited to leverage innovative communication tools and information technology and new reimbursement structures to realign incentives in favor of providing highquality care efficiently. Like other innovative models of delivery, the PCMH is most advantageous for systems or purchasers that can realize cost reduction benefits from decreased utilization and that are already well connected and technologically advanced. However, the additional administrative costs of integration at start-up may hinder the model’s cost control benefits in the short term.

An evidence base is building for improved quality, but evidence regarding patient and provider satisfaction and decreased costs under the PCMH model is inconclusive. Current research shows the PCMH has been adapted to a wide variety of populations and markets. PCMH examples for Medicaid populations are seen in Community Care of North Carolina and within the large integrated health plans and delivery systems of UnitedHealthcare, Group Health Cooperative of Puget Sound, Intermountain Healthcare, and the Geisinger Health System. An evaluation of Community Care of North Carolina found significant net savings for fiscal years 2004 through 2006. Fiscal year 2006 savings, for example, were estimated at close to $300 million for the state. When compared with fee for service, savings were especially high for inpatient care and for care to patients less than 1 year old.

Payment methodologies such as bundled payment, financial incentives such as P4P, or gain-sharing approaches may be combined with integrated delivery systems or employed alone to achieve value in healthcare purchasing. Even though bundled payment approaches have been attempted for at least 2 decades, evidence—whether bundling is by episode, condition, treatment, or admission&mdash;is sparse beyond studies of hospital care bundling by diagnosis-related groupings. However, select notable studies have been conducted that show promising results. For example, Geisinger’s coronary bypass program, called ProvenCare, is designed as a flat fee payment for surgery and all related care for 90 days after discharge, which places accountability for outcomes on the care team.

 Results showed a 10% reduction in readmissions, shorter average length of stay, and reduced hospital charges. The program achieved a 44% drop in readmissions over a course of 18 months.

A financing incentive, P4P, rewards providers for achieving the purchaser’s established objectives. Evidence of P4P, especially regarding cost reduction, is mixed.

 The evidence of the effects of P4P on quality improvement is stronger. One systematic review of the literature found partial or positive effects of financial incentives on measures of quality in 5 of 6 studies of physician-level financial incentives and 7 of 9 studies of the provider group-level incentives.

 P4P effectiveness depends on who receives the incentive (individual providers responsible for treatment decision making or health plans), on knowledge of the incentive, and on the level of incentive.

Gain sharing, once controversial but now in popular use, is another approach to reward incentivized behavior. Gainsharing arrangements, particularly those used in hospital and integrated delivery systems, provide bonus payments to physicians and other providers, to reward cost savings resulting from their efforts to reorganize delivery of clinically appropriate care at lower cost, eg, by being prudent in their clinical choices, including selection of procedures, supplies, and devices. The existing studies of individual programs show cost savings, but few studies have rigorously examined this issue. CMS implemented a 3-year demonstration project in 2007 that was extended by the 2010 national health reform legislation. CMS launched another 3-year demonstration with a consortium of 12 New Jersey hospitals and their participating physicians in 2009 that is intended to reduce healthcare costs and improve quality of care.

Value-based insurance design (VBID) seeks to increase value in healthcare through insurance design and incentives, and includes various strategies and approaches that can be used together or separately in different VBID models. One model of VBID reduces or eliminates cost sharing for services that have strong evidence of clinical benefit in order to improve healthcare quality and efficiency.

 Another model increases cost sharing for services that do not show evidence of clinical benefit. While VBID has broader healthcare system applications, many attempts—and available evidence&mdash;involve pharmaceutical benefits. In general, coverage and cost-sharing provisions address the problem of moral hazard, which is the overconsumption of health services in the absence of a consumer having a financial stake due to insurance coverage. By reducing or removing cost-sharing for benefits where utilization is highly desired, as in the case of clinical benefits, VBID leverages the cost-sharing incentives on consumer behavior.

The empirical evidence of VBID is mostly related to select attempts to implement the approach by employers and other purchasers, including Pitney Bowes and Oregon’s Public Employees’ and Educators’ Benefit Boards. For example, Pitney Bowes shifted all diabetes drugs and devices from tier 2 or 3 formulary status to tier 1. This single change in pharmaceutical benefit design immediately made critical brand-name drugs available to most Pitney Bowes employees and their covered dependents for 10% co-insurance, the same co-insurance level as for generic drugs compared with the previous cost share of 25% to 50%.

 A study conducted 2 to 3 years after the policy change found that for plan participants with diabetes total pharmacy costs decreased by 7%, emergency department visits decreased by 26%, and hospital admission rates, although slightly increased, remained below the demographically adjusted Medstat benchmark. Overall direct healthcare costs per plan participant with diabetes decreased by 6%.

 Another Pitney Bowes effort eliminated copayments for cholesterol-lowering statins and reduced them for a blood clot inhibitor called clopidogrel. The change resulted in an immediate 2.8% increase in adherence to statins relative to controls and, for clopidogrel, a 4-percentage-point difference in adherence rate between intervention and control patients a year later.

 Forthcoming evaluations are pending on the University of Michigan’s MHealthy: Focus on Diabetes project and Aetna’s Post-MI Free Rx Event and Economic Evaluation trial.

The end game in healthcare purchasing is not strictly one of cost control nor one of quality improvement in isolation; it is a search for value. The evidence about value-based purchasing is mixed but promising for some care delivery models and markets. Integrated healthcare systems, Intermountain and Geisinger, and public payers with dominant market share have experienced the greatest success to date, as have select large employers that are dedicated to VBID (Pitney Bowes in particular). It is not known whether these same results are transferable to other contexts. In an era when individuals increasingly change employers many times throughout their working lives, it is likely that many employers and purchasers may never see the economic benefit of their investments in value-based care. A danger is that this dynamic may push selective applications of VBID with potential to produce gains in the short term, with lesser emphasis on applications generating longer-term results.

Ultimately, value-based purchasing is part of a much broader policy “experiment” to advance value as a remedy for spiraling health benefit costs and quality concerns in US healthcare. To date, experience with value-based purchasin and evidence about what works for which populations and contexts has been limited, drawn largely from Medicare demonstrations and a few pioneering employers and private health systems. However, small pilots and programs have shown sufficient promise to maintain public interest in these approaches. What is needed now are rigorous studies that examine the performance of value-based strategies and compare these approaches to one another and to competing approaches to cost control and quality improvement. Studies are needed to determine how to effectively optimize the impact of value-based approaches in different contexts and in combination with direct patient care management strategies.

Conducting these studies has proved difficult and sometimes controversial. Coming to consensus about common frameworks and methodologies for measuring impacts and outcomes of value-based interventions and system design, conducting analyses with less than ideal administrative data, and completing a systemic view of patient utilization and cost across and within fragmented care settings are key challenges in attempting to show impacts of value-based approaches on healthcare utilization, quality, and cost. Practical issues such as how episodes and duration of care are determined; how quality of patient care and outcomes are to be assessed; how costs, cost savings, and cost offsets are determined; and to whom costs and benefits accrue remain fundamental concerns to purchasers, payers, providers, patients, and other stakeholders. Ultimately, addressing these issues will lead to the development of a performance measurement system that incorporates care coordination—a new paradigm that is meant to incorporate data from across the care continuum.

For example, in the case of VBID, better understanding of costs and benefits—how and to whom to target incentives, at which levels of risk&mdash;is essential if these strategies are to be economically viable and sustainable. If prescription drug costs are calculated in exclusion of metrics such as hospital admissions, readmissions, emergency department use, and costs of adverse events, then the “value” of VBID may be missed altogether. When cost components are considered independently, using an uncoordinated measurement approach, the cost of increasing adherence in a prescription drug regimen for depression where the average person in the VBID program increases use by a dose per year at an additional cost to the insurance plan of $30 per member per year will show on the balance sheet as an additional cost. However, from a more comprehensive, integrated systems perspective, the cost of enhanced prescription drug adherence would be measured against the offset, or prevented, costs of averted adverse events. Are the savings from these prevented adverse events greater than $30 per member per year? Until the measurement paradigm connects these types of data in a coordinated effort, it will be difficult to measure “value.” Value is not free. However, if value-based purchasing techniques are applied with the benefit of coordinated measurement and from a systems perspective, cost control can be demonstrated&mdash;and for specific populations.

While some short-term benefits may be demonstrated in value-based purchasing, it typically takes years before improved patient outcomes and related cost savings can be realized. At this time, select value-based purchasing mechanisms have been proved effective in select circumstances under particular assumptions. Now, the challenge is to replicate these results in varied environments, as the American health system is made up of diverse markets operating under different conditions. Purchasers will need to assess if the environments and assumptions of the existing evidence are close enough to their own to merit a value-based purchasing experiment. Further, the PPACA has established a new Center for Medicare and Medicaid Innovation in CMS to test and study the most promising models of payment and service delivery. After reviewing the value-based purchasing evidence, the need for a new measurement paradigm for cost and quality that reaches across the care continuum is evident. Given the length of time from concept to evaluation results, the need for development of such a framework is pressing. The existing evidence in combination with the new revelations of pioneering purchasers and the recommended measurement framework will push forward our understanding of how cost and quality can be interrelated in the value proposition.

 From Altarum Institute (GNE), Washington, DC; Altarum Institute (HK), Portland, ME.

 The authors (GNE, HK) report no relationship or financial interest with any entity that would pose a conflict of interest with the subject matter of this article.

 Concept and design (GNE, HK); acquisition of data (GNE); analysis and interpretation of data (GNE, HK); drafting of the manuscript (GNE, HK); critical revision of the manuscript for important intellectual content (GNE, HK); administrative, technical, or logistic support (GNE, HK); and supervision (GNE, HK).

 Holly Korda, PhD, MA, Altarum Institute, Systems Research and Initiatives, 4 Milk St, 3rd Fl, Portland, ME 04102. Email: holly.korda@altarum.org.

	1. Sebelius K. Speech at the National Business Coalition on Health Annual Meeting; November 15, 2010; Washington, DC.

	2. US Government Accountability Office. Medicare Physician Payment: Care Coordination Programs Used in Demonstration Show Promise, but Wider Use of Payment Approach May Be Limited. February 2008. GAO-08-65.

	3. Lurito K. Mercer Consulting Letter to Jeffrey Simms, State of North Carolina, Office of Managed Care, Division of Medical Assistance. Community Care of North Carolina Web site. http://www.communitycarenc.com/PDFDocs/Mercer%20SFY05_06.pdf. Published September 19, 2007. Accessed December 7, 2010.

	4. Mechanic RE, Altman SH. Payment reform options? episode payment is a good place to start. Health Aff (Millwood). 2009;28(2):w262-w271. http://content.healthaffairs.org/cgi/content/abstract/28/2/w262. Accessed December 11, 2010.

	5. Peterson LA, Woodard LD, Urech T, Daw C, Sookanan S. Does pay-for-performance improve the quality of health care? Ann Intern Med. 2006;145(4):265-272.

	6. Pizzi R. CMS approves New Jersey gain-sharing demonstration project. Healthcare Finance News. http://www.healthcarefinancenews.com/news/cms-approves-new-jersey-gainsharing-demonstration-project. Published August 19, 2009. Accessed December 7, 2010.

A review of evidence about what works in 3 areas of value-based purchasing emphasized under healthcare reform: service delivery integration, payment, and value-based insurance design.

Value-Based Purchasing: The Evidence

 To examine variation among commercial health plans in resource use and quality of care for patients with diabetes mellitus or cardiovascular disease.

 Cohort study using Healthcare Effectiveness Data and Information Set data submitted to the National Committee for Quality Assurance in 2008.

 Composite measures were estimated for diabetes and cardiovascular disease resource use and quality of care. A “value” classification approach was defined. Obtained were descriptive statistics, Pearson product moment correlations between resource use and quality of care, and 90% confidence intervals around each health plan’s composite measures of resource use and quality of care. Health plans were classified based on their results.

 For patients with diabetes, the correlation between combined medical care services resource use and composite quality of care is negative (−0.201, 

 = .008); the correlation between ambulatory pharmacy services resource use and composite quality of care is positive (0.162, 

 = .03). For patients with cardiovascular disease, no significant correlation was found between combined medical care services resource use and composite quality of care (—0.007, 

 = .94) or ambulatory pharmacy services resource use (0.170, 

Measures of resource use and quality of care provide important information about the value of a health plan. Although our analysis did not determine causality, the statistically weak or absent correlations between resource use and quality of care suggest that health plans and practices can create higher value by improving quality of care without large increases in resource use or by maintaining the same quality of care with decreased resource use.

	An examination of variation among health plans in resource use and quality of care for patients with diabetes mellitus or cardiovascular disease suggests that plans can demonstrate higher value under either of the following conditions:

		By improving quality of care without large increases in resource use

		By maintaining the same quality of care with decreased resource use.

Reducing the rate of increase in healthcare costs, while addressing gaps in quality of care, is the most challenging and pressing healthcare issue facing the United States and other developed countries.

documenting problems with quality and cost of healthcare in the United States have given rise to a growing interest in understanding the relationship between cost and quality. The science of measuring quality has advanced considerably over the past 2 decades; given the widespread use of Healthcare Effectiveness Data and Information Set (HEDIS), ORYX (an initiative by the Joint Commission on Accreditation of Healthcare Organizations), and related measures,

 there is substantial consistency around the content and scope of quality measurement. While methodological approaches to measuring differences in cost are emerging, whether through a focus on resource use or through various formulations of expenditures, there is no widely used way to directly compare costs among entities.

APPROACHES TO MEASURING RESOURCE USE AND QUALITY OF CARE

Premiums represent the cost of care to the employer and employee but have an imperfect relationship with the actual costs of medical care and reflect relative market power, benefit design, and other factors. By contrast, actual prices paid reflect resource use and local market forces but often vary among providers, making it difficult or impossible to use in cost comparative analysis among entities. Using a standardized set of prices applied to actual utilization is a common method for analysis that provides more transparency than premiums alone and avoids the challenges associated with using actual prices. Finally, resource use largely reflects the framework of interactions among patients, providers, and others within the healthcare delivery system.

The rise in chronic diseases has been directly linked to the increase in healthcare spending to treat them.

 estimates that in 2007 the total estimated cost of diabetes was $174 billion, including $116 billion in excess medical expenditures and $58 billion in reduced national productivity. Furthermore, the American Heart Association and the National Heart, Lung, and Blood Institute estimated that in 2009 the cost of cardiovascular diseases and stroke in the United States was $475.3 billion.

While HEDIS has for years included measures reporting health plan utilization of services, these measures do not directly relate to any HEDIS quality measure, are not case mix adjusted, and cannot be aggregated across categories, thus providing a limited amount of useful comparative information. In 2007, the National Committee for Quality Assurance (NCQA) introduced HEDIS measures of relative resource use (RRU) for 6 prevalent high-cost areas of care, including diabetes mellitus, cardiovascular disease, uncomplicated hypertension, chronic obstructive pulmonary disease, asthma, and low back pain, that use standardized prices and case mix adjustment to better define the cost portion of efficiency. Roski et al

 provide an overview of the NCQA HEDIS resource use measurement approach and results from a small pilot test project (<25 health plans) related to diabetes care. Turbyville et al

 examine the implications of wide national variation in RRU and quality results among commercial health maintenance organizations (HMOs) for their members with diabetes. The present study focuses on results for RRU and quality of care among commercial HMOs but goes further by examining the ability to construct a health plan classification approach based on resource use and quality-of-care performance in patients with diabetes or cardiovascular disease.

The data used in this study are part of the HEDIS 2008 health plan submission to the NCQA based on services rendered in 2007. Before submitting performance data to the NCQA, health plans must undergo an independent audit review that assesses the validity of data collection and measure calculations. Analyses reported herein are based on commercial HMO plans that submitted summarized member month, claim, and medical record data for diabetes and cardiovascular disease resource use and quality performance. Individual variables in the RRU measures are based on a patient population perspective and include all services rendered (not just those that are attributed to the disease) in the specified service categories. Health plans submit their summarized data within specified patient cohorts, which allow for risk adjustment once the data are submitted (

). The denominator inclusion criteria for HEDIS RRU measures are designed to be representative of patients included in the related quality measure denominator population, allowing for a more direct examination of the relationship between resource use and quality of care.

To estimate the quality of care, we calculated 2 combined composite quality measures based on calendar year 2007 results, 1 for diabetes and 1 for cardiovascular disease. The diabetes composite quality measure components are as follows: glycated hemoglobin testing, glycated hemoglobin poor control (>9.0% [inverted]), retinal eye examination, low-density lipoprotein cholesterol screening, low-density lipoprotein cholesterol control (<100 mg/dL [to convert cholesterol level to millimoles per liter, multiply by 0.0259]), medical attention for nephropathy, and blood pressure control (<130/80 mm Hg). The cardiovascular disease composite quality measure components are as follows: cholesterol screening, cholesterol control (<100 mg/dL), blood pressure control (<140/90 mm Hg), and persistence of B-blocker treatment after a heart attack. We further indexed each health plan’s results for diabetes or cardiovascular disease against the mean HMO commercial plan performance for each condition, allowing us to demonstrate its relative performance (eg, an index result of 1.12 implies 12% above average, whereas a result of 0.90 implies 10% below average).

To examine RRU among health plan members with diabetes or cardiovascular disease, we used 2 resource use variables for each condition, the combined medical care services variable and the ambulatory pharmacy services variable. The discrete service categories of the RRU measure for total medical care services include inpatient facility services, procedure and surgery services (inpatient and outpatient), and evaluation and management services (inpatient and outpatient). Within these service categories, the measures examine all resources expended during the 2007 calendar year. Detailed characteristics of these measures are included in the NCQA’s HEDIS technical specifications.

The category of inpatient facility services accounts for a patient’s stay and is adjusted for severity of diagnosis, occurrence of major surgery (defined in the specifications), and length of stay. The procedure and surgery services and the evaluation and management services include inpatient-based or outpatient-based care, such as that provided by an internist or a surgeon, and are adjusted for intensity. Ambulatory pharmacy services include all prescriptions rendered in ambulatory settings and are based on National Drug Code identifiers. Ambulatory pharmacy services were analyzed separately because of variation in benefit status or eligibility in pharmacy services and owing to past research on their relationship to quality. Benefit status variation results in the pharmacy component having a different per-member divisor than the medical service categories, rendering a total resource use composite computationally difficult when using summarized data. Furthermore, evidence shows that resource uses for medical and pharmacy services have different, sometimes inverse, correlations with quality.

 The RRU index is the ratio of the health plan’s observed (submitted) per-member-per-month resource use to the expected per-member-per-month resource use across the service categories. Health plans estimate their observed per-member-per-month resource use by following the RRU HEDIS specification that includes application of the NCQA standardized prices to the specified health service units. Pricing algorithms are designed to reflect service pricing levels for HMOs for the most recent period. The NCQA estimates the expected amounts based on all submitted data using indirect standardization to adjust for each health plan’s patient population distribution by age, sex, the presence of comorbid conditions, and severity of condition (eg, congestive heart failure vs angina). The ratio of each health plan’s observed results to expected results is indexed to the mean HMO commercial plan performance.

Statistical software (SAS version 9.1; SAS Institute, Cary, North Carolina) was used to calculate health plan descriptive statistics and Pearson product moment correlations for the diabetes and cardiovascular disease resource use and quality-of-care measures (combined medical care services and pharmacy services) for the ratio of observed results to expected result. We excluded health plans that did not submit complete RRU and quality-of-care data, submitted RRU data for fewer than 400 members with the given disease, or had a ratio of observed results to expected results of less than 0.33 or greater than 3.00. The NCQA based these exclusion decisions on previous research, consideration of standard errors and sample size needs for reliability purposes, and investigation of larger-than-expected observed variation among health plans, which revealed that a ratio of observed results to expected results of less than 0.33 or greater than 3.00 was caused by plan data submission errors in almost all cases. We did not report regression estimates because the direction of causality is unclear: do changes in resource use lead to changes in quality, or do changes in quality lead to changes in resource use? Furthermore, the study design did not permit us to test the direction of causality.

This study examines variation among health plans in resource use and quality of care for patients with diabetes mellitus or cardiovascular disease.

August 2011

Identifying Subgroups of Complex Patients With Cluster Analysis

Cost of Post-Traumatic Stress Disorder vs Major Depressive Disorder Among Patients Covered by Medicaid or Private Insurance

Value-Based Purchasing: The Evidence

Classification of Health Plans Based on Relative Resource Use and Quality of Care

Lynch Syndrome Screening Implementation: Business Analysis by a Healthcare System

The FLU-FIT Program: An Effective Colorectal Cancer Screening Program for High Volume Flu Shot Clinics

Diabetes and Employment Productivity: Does Diabetes Management Matter?

Accountable Care Organizations: The Proposed Regulations and the Prospects for Success

Comparative Efficacy of Ezetimibe/Simvastatin, Rosuvastatin, and Atorvastatin in Uncontrolled Hyperlipidemia Patients

HCV Screening Practices and Prevalence in an MCO, 2000-2007