	Between 2007 and 2009, Intermountain Healthcare, an integrated delivery system in Salt Lake City, Utah, used delivery optimization techniques to standardize lung care for premature infants born in its hospitals. These efforts yielded a 75% reduction in the use of postnatal mechanical ventilation, an improvement in quality that generated hundreds of thousands of dollars per year in savings.

 Since 1995, Intermountain has successfully undertaken more than 100 similar performance improvement initiatives, resulting in improved clinical outcomes and lower care delivery costs.

 Yet because payers reimburse healthcare delivery organizations based on the episodes of care that these quality improvement initiatives seek to reduce, investments in quality and efficiency improvement have often resulted in revenue losses for Intermountain and a large share of cost savings accruing to commercial health plans and government payers.

	As the Intermountain case illustrates, healthcare delivery organizations have historically faced high net costs resulting from improvement, due to a combination of largely unreimbursed input costs (eg, hiring care coordinators), realization of revenue reductions when improvements are successful (eg, reductions in avoidable hospital admissions), and the accrual of most of the cost savings to insurers or government programs. As a result, critical tools that are known to improve care and produce greater efficiency—electronic health records and information sharing between providers (health information technology [HIT]), care process redesign, and business process management&mdash;are underused because of the longstanding misalignment of incentive between the historically dominant payment policy and the means to improve efficiency and quality.

	A new legislative and market context, driven by the perceived unsustainability of an industry whose spending growth rates have exceeded those of the rest of the economy for several generations without proportionate gains in quality and health, has the potential to gradually establish a stronger business case for performance improvement. The current strategies for testing and bringing to scale innovative HIT and provider payment models in the public and private sector are intended to gradually lower the net costs that healthcare delivery organizations like Intermountain face when their efforts to redesign care processes and improve outcomes are successful.

	First, the Health Information Technology for Economic and Clinical Health Act provisions of the American Reinvestment and Recovery Act subsidize the adoption of HIT in the delivery of care, thereby temporarily offsetting some of the unreimbursed input costs that healthcare delivery organizations face when modernizing their information systems. This law provides up to $63,750 in temporary payment incentives for individual physicians and more than $2 million for hospitals to demonstrate the meaningful use of electronic health records.

 “Meaningful use” creates a minimum floor for improvement by incentivizing physicians and hospitals to apply HIT to record health information and improve processes that can lead to better outcomes.

	Second, several provisions in the Affordable Care Act and new payment innovations among private payers and delivery organizations are together introducing new reforms in the market, including accountable care organizations (ACOs), a range of bundled or “episode-based” payment models, and expanded implementation of advanced patient-centered medical homes. The Affordable Care Act also chartered a new Center for Medicare and Medicaid Innovation in the Centers for Medicare and Medicaid Services that is testing and will scale new models of care and payment that prove effective. Similarly, many state Medicaid programs (eg, those in Utah, Colorado, and Oregon) are experimenting with new forms of “accountable care” payment arrangements as alternatives to “pure” fee-for-service systems.

	Together, temporary federal subsidies for HIT and new provider payment innovations in the public and private sectors are designed to gradually alter the economics of performance improvement, making healthcare delivery organizations’ pursuit of excellence more technically and financially viable than was possible in a health system characterized by paper records and fee-for-service payments.

	To be clear, as capital investments, HIT implementations, and new care delivery strategies are implemented in the short run, most healthcare delivery organizations participating in new payment models are unlikely to see their earnings grow as rapidly as they might in an environment of unfettered fee-for-service payments. For example, consider an ACO model offering healthcare delivery organizations 50% of the savings it generates from improvements once the first 2% in savings is achieved. (These are the basic parameters of the Medicare Physician Group Practice Demonstration, an early ACO-like model launched in 2005.) If the ACO earns 5% in gross cost reductions through, for example, implementation of new care coordination programs that achieve reductions in hospital admissions,

 it will receive a 1.5% return in sharedsavings bonuses—not accounting for the infrastructure and other human investments that will be required to achieve these savings.

	Accordingly, in the short run, reforms that support HIT adoption and the proliferation of new payment models should not be viewed as mechanisms for physicians and healthcare delivery organizations to achieve dramatically higher earnings. 

 These models would be better viewed as mechanisms to support delivery organizations’ transition to care models delivering better-coordinated, safer, and more efficient care by lowering the net costs that providers face when investing in successful clinical improvements. Hence, even modest shared-savings payments from new payment models can help temporarily offset some of the input costs and revenue losses

	for delivery organizations that undertake efforts to improve quality and lower per capita healthcare costs.

	This notion is consistent with the experience of the 10 demonstration sites that participated in Medicare’s Physician Group Practice demonstration,

 implemented in 2005, and the early experience from the Office of the National Coordinator for Health IT’s Beacon Communities program, implemented in 2010.

 Healthcare delivery organizations and other stakeholders participating in the 17 Beacon Communities are using federal funding to design and implement new long-term strategies to prevent and manage chronic disease, and to reduce avoidable emergency department utilization and hospital readmissions. Hospitals and physicians in these communities, including Intermountain as part of the Improving Care through Connectivity and Collaboration Beacon Community in Utah, are actively considering new provider payment programs with public and private payers with healthcare payers that can best accommodate the longterm sustainability of these iniatives.

	While in the short run new payment models should be viewed as mechanisms to lower the net costs that delivery organizations face to achieve improvements, over a longer time horizon they should make it possible for healthcare organizations that deliver high-quality, efficient care to prosper. An increasing number of healthcare delivery organizations will develop the HIT, performance measurement, and process improvement capabilities to more aggressively improve outcomes and lower per capita costs. In turn, these changes will make delivery organizations increasingly capable of participating in more advanced payment models— such as partial capitation and other forms of risk sharing&mdash;that will be uniquely enabled by HIT and the data and analytic capabilities it supports.

	The precise long-term impact of simultaneously modernizing information systems and payment models is uncertain, although directionally positive. The policy and technical issues related to implementation—for government, healthcare delivery organizations, and commercial payers&mdash;remain incredibly complex.

 Beyond the government’s direct role in promulgating new regulations to implement new HIT and payment programs and the private sector’s efforts to innovate in a changing marketplace, improvement projects like those developed at Intermountain need to be replicated in thousands of hospitals and tens of thousands of physicians’ offices nationwide—no small feat.

	Despite these challenges, the fates of HIT and payment reforms are linked. Achieving large-scale improvement in a current system whose spending is increasingly out of balance will require taking full advantage of HIT and other delivery optimization tools available in the modern age and embedding them in payment systems that consistently lower the costs that healthcare delivery organizations face in achieving better care, improved health, and greater value.

	With effective collaboration, we may be gradually building toward a new business case for performance improvement that can result in the high-value, high-quality system of care that is technically possible and that our nation deserves.

	The authors gratefully acknowledge the contribution of James Walker,

MD, chief health information officer at Geisinger Health System, for offering helpful comments and suggestions on an earlier draft of this commentary, as well as to the anonymous reviewers whose contributions improved it. This article reflects the perspectives of the authors and does not reflect any official positions or policies of Brigham and Women’s Hospital or the Office of the National Coordinator for Health Information Technology.

 Department of Health Policy and Management, Gillings School of Global Public Health, University of North Carolina at Chapel Hill, Chapel Hill, NC (AM), US Department of Health and Human Services (AM), Washington, DC; University of Cincinnati College of Medicine, Cincinnati, OH (CB); Department of Medicine, Brigham and Women’s Hospital and Harvard Medical School, Boston, MA (SHJ).

The authors (AM, CB, SHJ) 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 (AM, CB); analysis and interpretation of data (CB); drafting of the manuscript (AM); and critical revision of the manuscript for important intellectual content (AM, CB, SHJ).

Aaron McKethan, PhD, University of North Carolina at Chapel Hill, Chapel Hill, NC 27514. E-mail: amckethan@gmail.com.

	1. Leonhardt D. Making health care better. New York Times Magazine. November 3, 2009. http://www.nytimes.com/2009/11/08/magazine/ 08Healthcare-t.html. Accessed April 19, 2011.

	2. James BC, Savitz LA. How Intermountain trimmed health care costs through robust quality improvement efforts. Health Aff (Millwood). 2011;30(6):1185-1191.

	3. Blumenthal D, Tavenner M. The “meaningful use” regulation for electronic health records. N Engl J Med. 2010;363(6):501-504.

	4. Naylor MD, Aiken LH, Kurtzman ET, Olds DM, Hirschman KB. The care span: the importance of transitional care in achieving health reform. Health Aff (Millwood). 2011;30(4):746-754.

	5. Haywood TT, Kosel KC. The ACO model—a three-year financial loss? N Engl J Med. 2011;364(14):e27.

	6. Iglehart JK. Assessing an ACO prototype—Medicare’s Physician Group Practice demonstration. N Engl J Med. 2011;364(3):198-200.

	7. McKethan A, Brammer C, Fatemi P, et al. An early status report on the Beacon Communities’ plans for transformation via health information technology. Health Aff (Millwood). 2011;30(4):782-788.

	8. Ginsburg PB. Spending to save—ACOs and the Medicare Shared Savings Program [commentary]. N Engl J Med. 2011;364(22): 2085-2086.

Articles

WEB EXCLUSIVE

Recent legislative action and private sector innovation, driven by the unsustainability of the current system, may gradually create a business case for performance improvement.

Toward a Business Case for Performance Improvement

 To determine whether adding selfreported health and functional status data to a diagnostic risk-score model explains additional variance in predicting inpatient admissions and costs.

 We used data from a Health Status Questionnaire (HSQ), completed by 6407 Kaiser Permanente Northwest Medicare patients between December 2006 and October 2008. We used answers from 3 items on the HSQ: (1) General Self-rated Health score, (2) needing help with 1 or more activities of daily living, and (3) having a bothersome health condition. We calculated a DxCG relative risk score from utilization information in the year prior to the survey, using electronic medical records. We compared: (1) DxCG as the sole independent variable and (2) DxCG plus the 3 items as independent variables. We estimated area under the curve (AUC) for each model. Any inpatient admission (yes/no) and being in the top 10% of costs (in the year after survey) were the dependent variables for the first and second logistic regression models, respectively.

The 3 items explained an additional 2.8% and 4.0% of variance for inpatient admissions and top 10% of costs, respectively, in addition to the variance explained by the DxCG score alone. For DxCG alone, the AUC was 0.686 (95% confidence interval [CI] 0.663-0.710) and 0.741 (95% CI 0.719- 0.764), respectively, for inpatient admissions and top 10% of costs and improved to 0.709 (95% CI 0.687-0.730) and 0.770 (95% CI 0.749-0.790) when the 3 self-reported items were added.

 Using self-reported health information improved the predictive power of a DxCG model to forecast inpatient admissions and patient cost-tier.

	In a Medicare population, self-reported information about being in poorer health was associated with higher inpatient admissions and being in the top tier for costs. Items associated with these outcomes were: (1) lower score on the General Self-rated Health score item, (2) answering yes to “do you need help with 1 or more activities of daily living?” and, (3) answering yes to “do you have a bothersome health condition?” These items:

		Explained an additional 2.8% and 4.0% of admission and cost variance, respectively.

		Were independently predictive of future inpatient admissions and being in the top 10% cost group.

	Predictive healthcare models are valuable tools for identifying individuals at risk for adverse outcomes, including high healthcare utilization and mortality. Predictive models are typically based on diagnostic history from administrative data or sometimes on patient-reported outcomes, but rarely both. The purpose of our analyses was to determine whether combining these 2 types of data in an integrated predictive model improved upon the predictive power compared with models based on administrative data alone.

	Claims-based predictive models are the most common because of the ubiquitous availability of claims data. There are a variety of commercially available claims-based predictive models, and their validity is well documented.

 The Society of Actuaries has conducted periodic comparisons of the predictive power of these models for following-year costs.

 In these models, health diagnoses and prescriptions, in addition to prior expenses, are the variables with most predictive accuracy. Laboratory results have also been shown to add predictive power.

	Predictive models based on patient-reported outcomes are less common because such measures are not routinely collected and reported. However, patient-reported outcomes have been shown to have strong predictive value for forecasting costs, utilization, and mortality. In particular, the predictive value of a single General Self-rated Health (GSRH) question is well documented. DeSalvo (2005) concluded in a meta-analysis that people with “poor” self-rated health had double the mortality risk of people with “excellent” self-rated health, even after adjustment for key factors such as functional status, depression, and comorbidity.

 GSRH has also been shown to be responsive to change. In a study of older adults, self-reported health declined steeply prior to adverse events, such as death or stroke.

 A study of repeated administrations of 2 GSRH questions showed good reliability, reproducibility, and discriminate scale performance.

 Although the single-question GSRH measure is seldom used in clinical practice settings, it is widely used in medical and economic research.

 In an investigation of the causal relationship between self-reported health and subsequent outcomes, Jylha suggests that aspects of self-reported health may reflect important physiological dysregulations, such as increases in tissue inflammation.

	Similarly, DeSalvo (2009) notes that respondents provide composite assessments of their physical and mental health and social status when answering questions about their overall health, which helps explain the strong performance of GSRH in predicting future risk.

 The literature includes several examples of patientreported outcomes used in predictive models. DeSalvo (2009) compared predictive ability of single- and multi-item assessments of self-reported health to a comorbidity index and diagnostic history in predicting next year’s costs. They found that the single-item self-reported health question predicted costs as well as the multi-item assessment and the comorbidity index, but not as well as the diagnostic history model.

 Using data from a survey of Medicare members, Brody et al developed highly sensitive indices of frailty and advanced illness.

 The models use survey items about self-reported health and function, such as poor general health and assistance needed for medications, bathing, or dressing, and were shown to be superior to clinical judgment in predicting frailty and advanced illness.

	There are, however, few examples of predictive models that combine diagnostic history and patient-reported outcomes. Fleishman et al found that measures of health status improved prediction of subsequent medical expenditures in a model that included demographics, chronic conditions, and previous expenditures in a nationally representative sample.

 Fleishman and Cohen later found that measures of self-rated health and function improved prediction of high expenditures beyond more detailed information on medical conditions for a nationally representative sample.

 These results are consistent with previous studies based on nonrepresentative samples. In a study of Veterans Affairs (VA) beneficiaries, self-reported physical and mental health improved mortality prediction when combined with a diagnostic history model.

Similarly in a previous study of VA beneficiaries, a multidomain health assessment added predictive power to a diagnostic history model.

	In this study, we examined whether inclusion of answers to a few brief self-reported items improved the prediction of future inpatient admissions and medical care costs compared with using an administrative risk score alone to predict the same outcomes. This study is unique in that it is the first of which we are aware to conduct this analysis in a Medicare population. Further, the Medicare population represents a unique population, given the large proportion with chronic health conditions, which contributes to increased inpatient admissions and costs—as the US population ages. We selected the well-validated, single-item general health question, a question about activities of daily living,

 and a question indicating whether or not the person currently has a physical condition that bothers him or her. These 3 items were selected because they are questions that could be quickly administered during an office visit, over the phone, by mail, or the Internet and are not tailored to a specific population (eg, those with chronic conditions) and capture information not readily attainable through administrative data.

	Study subjects were members of Kaiser Permanente Northwest (KPNW) and were aged 65 years and older. To be included in the study, members needed to complete a Health Status Questionnaire (HSQ) between December 2006 and October 2008, and have 1 year of membership eligibility prior to and after the completion of the HSQ. This time frame allowed us to compute the DxCG scores. When the member had completed more than 1 HSQ in the time frame of interest, 1 HSQ was selected at random for inclusion in the study.

	During the period of study, the HSQ was mailed to new Medicare members aged 65 years and older. This population included those newly enrolled to KPNW and current members who recently became Medicare eligible. In addition, it was mailed annually to a Medicare expanded-care population. The general response rate for the HSQ is 76%.

	The HSQ is a 45-item survey that includes measures of physical function, activities of daily living, health conditions, self-care deficits, and general health. For this study we used data from 3 items. The self-reported general health measure is scored in a range of 1 (excellent) to 4 (poor) by answering the question “compared to other persons your age, would you say your health is…?” The second item asked “because of a disability or health problem, do you need or receive help from another person for any of the following everyday activities…?” The activities listed include preparing meals, shopping for groceries, doing routine household chores, managing money, doing laundry, taking medications, getting to places not within walking distance, and using the telephone. This item was scored 0 if no items were checked and if 1 or more items were checked. The third item was “Is there any physical condition, illness, or health problem that bothers you now?” and was scored 0 for “no” and 1 for “yes.”

 is based on an individual’s total predicted cost in the next year relative to the population mean and is used to assign individuals to a Diagnostic Cost Group (DCG). DxCG Risk Solutions software

 is used to calculate the DxCG relative risk score. DCG cost estimates are calculated using 1 year of inpatient and outpatient 

International Classification of Diseases, Ninth Revision, Clinical Modification 

) codes to classify patients into groups with similar cost patterns. The diagnoses are classified into mutually exclusive DCGs, based on predicting future inpatient and outpatient costs. The software algorithm for DCG classifications was developed using 3 different populations: a privately insured population, a Medicaid population, and a Medicare population. Diagnostic Cost Groups for this study were based on the Medicare-derived DCG grouping. The range of Medicare DxCG scores ranged from 0.04 (minimum) to 15.59 (maximum). Individuals with a score of 1.0 have the relative risk equivalent to the national Medicare reference population. This reference population includes any of the following (excluding those with end-stage renal disease): (1) individuals 65 years or older, (2) disabled (under age 65 years) Medicare eligible, or (3) Medicare/Medicaid dual eligible. Previous research has used the DxCG relative risk score to predict future costs and mortality estimates.

	In the year following the date of the HSQ, we used electronic medical records to determine if the subject had any inpatient hospital admissions. Actual health plan costs for each person for this same period were extracted from administrative data and presented in 2008 dollars.

	Prior to logistic regression analysis, we tested for collinearity among the self-reported health items and DxCG risk score by calculating a Pearson product-moment correlation coefficient. We used 2 sets of logistic regressions, with 2 models in each set: DxCG as the sole independent variable and DxCG plus the 3 self-reported items as the independent variables. We calculated sensitivity, specificity, receiver operating characteristic (ROC) curves, and the area under the curve (AUC) for each model. Any inpatient admission (yes/no) was the dependent variable for the first set of regression models and being in the top 10% of costs (based on the observed study population) was the dependent variable for the second set of regression models. Prior to conducting the main analyses, those with and without complete data were compared using t tests and χ

	A total of 7545 HSQs were returned between December 2006 and October 2008 from 7496 unique individuals. Of those, 6407 people met the inclusion criteria of 1 year of eligibility prior to and after the HSQ date. Of those, 5815 (90.8%) had complete data on the 3 self-report items of interest and the DxCG. Those with complete data were younger (mean = 68.55 vs 70.71, 

 <.001), more likely to be male (44.9% vs 39.6%, 

 = .016), more likely to have attended college (67.0% vs 59.7%, 

 <.001), more likely to be married (69.0% vs 59.5%, 

 <.001), and to be white (93.1% vs 88.0%, 

 <.001). There were no significant differences between those with and without complete data on the likelihood of being in the top 10% of costs (

 = .680) or of having an inpatient admission (

 the average age was 68.55 (standard deviation [SD] = 6.34) with a range of 65 to 97. The majority of the sample was female (55.3%), married (69.0%), white (93.1%), and had attended some college (67.0%). Only 1.0% of the sample reported their race as black or African American; 1.9% as Asian or Pacific Islander; 0.3% as Aleut, Eskimo, or American Indian; and 1.3% as other. When asked to rate their health compared with other persons of their age, 27.8% rated it as excellent, 55.4% as good, 14.9% as fair, and 2.0% as poor. Only 10.9% needed help with 1 or more instrumental daily activities and 40.5% had a physical condition, illness, or health problem that bothered them at the time of taking the survey. The mean DxCG relative risk score was 0.68 (SD = 0.75) with a range of 0.19 to 9.91. Slightly more than 10% (10.8%) had 1 or more inpatient stays during the year following the survey.

Comparison of DxCG-only Model and DxCG—plus–3–self-report–items Model

	We used logistic regression to test whether addition of 3 self-report items significantly improved the DxCG model’s accuracy in predicting inpatient admissions and being in the top 10% of costs. The correlations among the self-report items and DxCG ranged from 0.26 to 0.45 (results not shown), which did not indicate a high degree of multicollinearity. For predicting inpatient admissions, the 3 self-report items made a significant contribution to the model compared with using DxCG alone (χ

 <.001), accounting for an additional 2.8% of the variance in having 1 or more inpatient admissions. 

 summarizes results of the model. Higher DxCG scores, poorer self-reported health, needing help with at least 1 everyday activity, and having a physical condition that was bothersome at the time of completing the survey were all associated with having an inpatient admission. For each additional point on the DxCG, the likelihood of an inpatient admission increased by 63%. A change of 1 point on self-reported health (eg, from good to fair) increased the likelihood of an inpatient admission by 30%. Needing help with at least 1 instrumental daily activity and having a physical condition that bothered the respondent at the time of the survey increased the likelihood of an inpatient admission by 52% and 53%, respectively. 

 presents the ROC curves for the model that used DxCG alone, and for the model that used DxCG plus the 3 self-report variables. The AUC was 0.686 (

 <.001, 95% confidence interval [CI] 0.663-0.710) for the model with DxCG alone, and improved to 0.709 (

 <.001, 95% CI 0.687-0.730) when the 3 self-report items were added. 

 provides results for sensitivity, specificity, positive predictive value, and negative predictive value for both models. Sensitivity was 0.467 for the DxCG-only model and improved to 0.529 when the 3 self-report items were added. Specificity was 0.807 for the DxCGonly model and 0.741 when the 3 self-report items were added to the model.

	Similarly, analysis for the models’ ability to predict being in the top 10% of costs showed that the 3 self-report items made a significant contribution to the model compared with the model that used DxCG alone (χ

 <.001). The 3 self-report items accounted for an additional 4.0% of the variance explained in being in the top 10% cost group (Table 3). Table 2 summarizes the results of the full logistic model with DxCG and the 3 self-report items. Higher DxCG scores, poorer self-reported health, needing help with at least 1 everyday activity, and having a physical condition that bothered the participant at the time of completing the survey were all associated with having an inpatient admission. For each additional point on the DxCG, the likelihood of being in the top 10% of costs increased by 104%. A change of 1 point on the self-reported health score (eg, good to fair) increased the likelihood of being in the top 10% of costs by 30%. Needing help with at least 1 instrumental daily activity and having a bothersome physical condition increased the likelihood of being in the top 10% of costs by 78% and 82%, respectively. 

presents the ROC curves for the model using DxCG alone and for the model with DxCG plus the 3 self-report variables. The AUC was 0.741 (

 <.001, 95% CI 0.719-0.764) for the model that used DxCG alone and improved to 0.770 (

 <.001, 95% CI 0.749-0.790) when the 3 self-report items were added to the model. Sensitivity was 0.522 for the DxCG-only model and improved to 0.611 when the 3 self-report items were added to the model. Specificity was 0.821 for the DxCG-only model and 0.771 when the 3 self-report items were added to the model (see Table 3).

	This study found that self-reported items add predictive power for future healthcare costs and inpatient admissions beyond the DxCG relative risk score. Specifically, poorer self-reported health, needing help with at least 1 activity of daily living and having a bothersome physical condition added predictive power for higher costs and inpatient admissions in the 12 months after survey administration. This study is one of only a few studies to examine the additive effect of self-reported health items to predict subsequent costs and inpatient admissions over and above the DxCG relative risk measure for a Medicare population.

	Results from this study confirm findings from other studies. Several studies have compared self-reported health items with administrative data—based risk measures to predict future costs, but did not examine the additive effect of self-reported items over administrative data–based risk measures. Using data from the Medical Expenditure Panel Survey, DeSalvo et al found that a 1-item GSRH score was predictive of future healthcare costs, but less robust than models that use administrative data–based risk measures.

 Similarly, using administrative and survey data from a VA population, Maciejewski et al found the SF-36 Physical Component Survey and Mental Component Survey were predictive of future healthcare costs, but again less robust than administrative data—based risk measures.

 In addition, similar to our findings, Fleishman et al found that measures of health status improved prediction of subsequent medical expenditures beyond demographics, chronic conditions, and previous expenditures.

	We also found that the DxCG risk score and observed self-report health items were not highly correlated with each other. Similar to our findings, Wang et al found that DCG (similar to DxCG) and SF-36 physical and mental health scores were both predictive of future healthcare costs, but not highly correlated.

 The implication of these results is that DxCG and the individual self-report health items are measuring different underlying constructs and do not overlap greatly.

	Findings from our study have implications for care delivery and quality improvement initiatives. Asking questions about self-rated health status, activities of daily living, and bothersome physical conditions is easy to do and can be administered in a variety of care delivery settings. The strong predictive ability of these self-report questions are worthy of integration into enhanced care management and patient outreach efforts. This information is especially important when diagnostic history is not available, or is too expensive to collect.

	This study is not without limitations. First, the DxCG relative risk measure was the only administrative data—based risk measure analyzed in this study. This limitation is addressed in part in that previous research found the DxCG relative risk measure has greater predictive ability than other administrative data–based measures.

 Second, the study population was enrolled in an integrated health maintenance organization health delivery system, so study findings cannot be generalized to non—managed care settings. Third, the study population included few racial/ethnic minorities, again limiting the generalizability of the study findings. Fourth, we were not able to compare the characteristics of those who did not return an HSQ with those who did return the HSQ. Fifth, patient-level cost data include information where both utilization and matching general ledger information are available. Therefore, costs for services with no, or inadequate, available utilization information are not included, so costs may be understated for some patients. Last, the DxCG predictive model did not include laboratory and biometric data, as well as information on receipt of evidence-based care based on clinical condition, which are important factors that can improve predictive modeling.

	Future research is needed to evaluate the use of self-reported health items to enhance care management activities. Specifically, randomized studies are needed to compare enhanced delivery models using self-reported health items to usual care. Within this approach, more formal cost-effectiveness analysis is needed to identify potential reductions in hospital admissions through rigorous case finding and enhanced care management approaches. The primary objective would be to determine the extent to which the addition of self-reported health information enhances the ability of healthcare organizations to better manage overall costs and utilization over time.

 From Center for Health Research (NAP, DMM, AB, ES), Kaiser Permanente Northwest, Portland OR; Department of Care and Service Quality (MS), Kaiser Permanente, Oakland, CA; Decision Support Services (EMD), Kaiser Permanente, Oakland, CA.

Among a Medicare population, use of 3 self-reported health items improves predicted inpatient admissions and healthcare costs when used with risk-prediction model.

Self-Reported Health and Functional Status Information Improves Prediction of Inpatient Admissions and Costs

Concerns over rising drug costs, pharmaceutical advertising, and potential conflicts of interest have focused attention on physician prescribing behavior. We examine how broadly physicians prescribe within the 10 most prevalent therapeutic classes, the factors affecting their choices, and the impact of their prescribing behavior on patient-level outcomes.

 Retrospective study from 2005 to 2007 examining prescribers with at least 5 initial prescriptions within a class from 2005 to 2007. Medical and pharmacy claims are linked to prescriber information from 146 different health plans, reflecting 1975 to 8923 unique providers per drug class.

Primary outcomes are the number of distinct drugs in a class initially prescribed by a physician over 1- and 3-year periods, medication possession ratio, and out-of-pocket costs.

In 8 of 10 therapeutic classes, the median physician prescribes at least 3 different drugs and fewer than 1 in 6 physicians prescribe only brand drugs. Physicians prescribing only 1 or 2 drugs in a class are more likely to prescribe the most advertised drug. Physicians who prescribe fewer drugs are less likely to see patients with other comorbid conditions and varied formulary designs. Prescribing fewer drugs is associated with lower rates of medication adherence and higher out-ofpocket costs for drugs, but the effects are small and inconsistent across classes.

Physicians prescribe more broadly than commonly perceived. Though narrow prescribers are more likely to prescribe highly advertised drugs, few physicians prescribe these drugs exclusively. Narrow prescribing has modest effects on medication adherence and out-of-pocket costs in some classes.

	Physicians prescribe more broadly than commonly perceived. Although most physicians have a “favorite” drug, they are not reluctant to try new therapies. Physicians who prescribe broadly see more patients with varied comorbidities and formulary designs. Prescribing fewer drugs is associated with lower rates of medication adherence and higher out-of-pocket costs, but the effects are small and inconsistent across classes. Broad prescribing may be due to:

		 The increasing number of drugs in a class.

		 Pharmaceutical marketing, particularly direct-to-physician promotions.

		 The role of pharmacy benefit managers and third-party payers.

	In 2004, pharmaceutical firms spent more than $57 billion on marketing in the United States, roughly twice their expenditure on research and development.

 Most of this spending targeted physicians through sales representatives (detailing), sampling (provision of drugs at no cost), physician meetings, and advertisements in medical journals.

 For example, industry-sponsored promotional events increased from 120,000 in 1998 to 371,000 events in 2004.

 There has also been a significant increase in the frequency and size of federal and state penalties for illegal promotion of drugs and pricing irregularities.

	These trends have raised the concern that pharmaceutical companies might have undue influence on the prescribing behavior of physicians. In particular, there is concern that a significant percentage of physicians might be prescribing a narrow range of heavily promoted drugs, or might be exclusively prescribing branded drugs to the detriment of patient welfare. However, empirical evidence on the prescribing behavior of physicians and its consequences for patients is limited. Some studies suggest that physicians exhibit narrow prescribing behavior, particularly general practitioners, but much of this evidence is decades old.

More recent work finds that the prescribing patterns of physicians are substantially more concentrated than the aggregate market in each class, and that physicians differ in their preferred drug within a class.

	While theory suggests that habitual prescribing can be both clinically suboptimal and economically wasteful, the appropriateness of broad versus narrow prescribing is likely to depend on the composition of the drug class. Narrow prescribing patterns may be optimal when 1 drug is clearly superior to the others, or if all the drugs in the class act in a similar way. For example, prescribing only a generic or low-cost brand in a largely homogeneous class may be beneficial given that lower patient cost-sharing is associated with improved adherence.

 In addition, most generics are inherently safer than newer drugs because of their longer track record in clinical practice and known side effects.

 Alternatively, some drug classes are characterized by heterogeneous effects, where a specific drug provides therapeutic benefit to some patients and little to others, or has known side effects that are problematic for a subset of patients. If the heterogeneous benefits and side effects of these drugs are known ex-ante, a better informed physician will engage in broader prescribing behavior, taking into account the specific medical characteristics of each patient. Levine-Taub and colleagues

 found that psychiatrists have broader prescribing patterns than general practitioners within the atypical antipsychotic class, but they cannot determine how much of the difference is explained by variation in the case mix of patients seen by psychiatrists versus nonspecialists.

	Beyond the challenge of predicting a drug’s therapeutic value for a new patient, an unrelated factor further complicates the prescribing decision; plan formularies. Most modern drug classes include an array of similar products that compete for essentially the same population of patients, and health plans typically choose a small subset of these products to offer at low cost-sharing rates. In addition, direct-to-consumer advertising has emboldened patients to request specific treatments. 

 How these factors have affected physicians’ choice of drug therapies is uncertain.

	In this paper, we examined the prescribing patterns of physicians in 10 large drug classes with several similar-acting agents. We measured the number and type (generic or brand) of different drugs prescribed as initial prescriptions by each physician and the factors that affected their choices. We then examined whether broad or narrow prescribing is associated with patient-level outcomes, such as rates of medication adherence, therapeutic switching, and out-of-pocket drug spending. We know of no other study that examines the relationship between how broadly physicians prescribe and patient-level outcomes (eg, adherence, medical care use) that can proxy for clinical measures.

	We used unique data, matching prescriptions to prescribing physicians. The data included medical and pharmaceutical claims from 29 large employers in the United States from 2003 to 2007. The drug claims included information on the type of drug, drug name, National Drug Code, dosage, days supplied, and place of purchase (retail or mail order). Starting in 2005, all pharmacy claims identify the prescriber by masked Drug Enforcement Administration number; thus, from 2005 to 2007, we could observe prescriptions made by the same physician to different patients in different insurance plans. We did not have any additional information about the prescribers. To be eligible for the sample, a patient had to be at least 18 years of age, and continuously enrolled for at least 1 year before initiating therapy and for at least 6 months afterward.

	We used the IMS Advertising Database to measure the degree of drug promotion for each product. The advertising data were reported quarterly and contained expenditures on direct-to-consumer and direct-to-physician advertising for each drug, including medical journal advertisements, promotional visits to physicians, and drug samples.

	We use a common classification scheme—the 2007 

published by Thomson—to associate each drug with a therapeutic class. 

shows the 10 most common therapeutic classes (in terms of dollars spent) in our sample for 2005. These are cholesterol-reducing drugs, antidepressants, non—histamine-2 receptor antagonist (H2A) stomach drugs, antihistamines, nonsteroidal anti-inflammatory drugs (NSAIDs), opiates, beta-blockers, calcium channel blockers, angiotensin-converting enzyme (ACE) inhibitors, and antidiabetic drugs, excluding insulin. To further narrow the classes, we focused on statins within the cholesterol-reducing drugs (dropping ezetimibe, fibrates, and others), on selective serotonin reuptake inhibitors (SSRIs) and serotonin and norepinephrine reuptake inhibitors (SNRIs) within the antidepressants (keeping bupropion formulations and dropping tricyclic antidepressants), and on proton pump inhibitors (PPIs) within the non-H2A stomach drugs. In the antihistamine class, we dropped promethazine, which is prescribed primarily as an acute treatment and often used as a sedative or antiemetic rather than for allergy treatment. We defined an initial prescription as the absence of any pharmacy claim in the same therapeutic class for at least 12 months.

	As most plans assign lower copayments to generic drugs, and often charge the same copayment for all generics, narrow prescribing patterns are most likely to impact average costs in classes where brand drugs are dominant. For this reason, some analyses focused on the 5 drug classes in which more than 50% of initial prescriptions are for brand drugs: statins, SSRIs/SNRIs, PPIs, antihistamines, and calcium channel blockers. We called these the “brand-dominated” classes.

	In 3 of these classes, 1 major drug became available as a generic during the study period: simvastatin (statin, starting June 2006), sertraline (SSRI, June 2006), and fexofenadine (antihistamine, September 2005). In the calcium channel blocker class, 2 generics entered the market toward the end of our study period (2007); amlodipine in March and amlodipine/ benazepril in May. In measuring the number of drugs prescribed, we treated brand and generic formulations of a multisource product as different drugs; however, the results are not sensitive to this choice.

	We restricted the sample to physicians with at least 5 initial prescriptions within a class from 2005 to 2007. We focused on initial prescriptions because refills may reflect the prescribing decisions of other providers. This yielded a sample of 74,163 initial statin prescriptions, prescribed by 8923 unique providers. The corresponding prescription/provider counts for the other brand-dominant classes were PPIs (52,978/6621), SSRI/SNRI (46,040/5866), antihistamines (39,644/4788), and calcium channel blockers (13,633/1975). We categorized providers within each class by the number of distinct drugs prescribed as initial prescriptions over the sample period (2005-2007). For example, a doctor with 2 initial prescriptions of escitalopram, 3 initial prescriptions of sertraline, and 1 initial prescription of duloxetine is categorized as prescribing 3 drugs in the SSRI/SNRI class.

	Given that new drugs may have entered the market and additional clinical evidence may have emerged over the 3-year study period, we also categorized physicians based on the number of distinct drugs prescribed each year. This reduced our sample substantially, as two-thirds to three-fourths of physicians (depending on the class) in the 3-year sample did not have 5 initial prescriptions within a calendar year. To facilitate comparison with other studies of prescribing concentration, we also calculated the share of prescriptions for each physician’s “favorite” drug.

	We examined the use of the top-selling and most heavily promoted drugs in the class, as well as rates of generic drugs, by prescriber type. We also calculated the share of a physician’s observed prescriptions that are in the relevant therapeutic category (eg, cardiovascular drugs for statin prescriptions) as a proxy for their degree of specialization.

	To assess factors that influenced the breadth of a physician’s prescribing, we estimated a Poisson regression with the number of different drugs prescribed in a class as the dependent variable (categorized as 1, 2, 3, 4, or 5 drugs prescribed as initial prescriptions). Estimates from the Poisson regressions were used to predict the number of drugs prescribed by each physician given the characteristics of their patients and formulary designs. We then classified each physician as high, medium, or low in breadth (concentration) of prescribing based on how their actual number of drugs prescribed deviated from the predicted value. We used this classification to assess whether narrow prescribing was associated with 3 patient-level outcomes: medication adherence, therapeutic switching (changing medications within a class), and outof pocket drug costs. We measured each patient’s adherence at the class level based on the medication possession ratio (MPR) over the 6 months following the initial prescription. The MPR was expressed as a percentage, and defined as the number of days supply of a medication (ie, possession) over the 6 months following the initial prescription. Therapeutic switching rates are generally low in the 5 brand-dominated classes, ranging from 9% for statins to 17% for SSRIs/SNRIs. This underscores the importance of the initial drug choice in determining the patient’s course of treatment.

	The independent variables included age and its square, gender, and median household income (by 3-digit zip code). We also have salary information (in buckets) for 56% of patients. Since two-thirds of those with salary information fall in the “below $50,000” category, we included binary indicators for a high salary (>$50,000) and missing salary information. Since patients receiving prescriptions from specialists are more likely to adhere, we used a proxy for specialist, defined as the share of all of a physician’s observed prescriptions that were in the relevant category, for example, cardiac drugs. We also measured the complexity of formulary designs facing each physician in 2 ways. First, we counted the number of observed health plans represented by a physician’s patients. Second, we computed the number of unique pharmacy benefit designs facing each physician based on the 

of copayments for the most prescribed brand drug, the second most prescribed brand drug, and the top generic drug in the class.

	Finally, we controlled for comorbid conditions related to a drug class using a set of disease indicators identified in the medical claims based on I

nternational Classification of Diseases, Ninth Revision, Clinical Modification 

diagnoses. For example, we included binary indicators for hyperlipidemia, diabetes, hypertension, chronic heart failure,  cardiac disease, vascular disease, and stroke for statin users (full model results are available from the corresponding author). We also included quarterly expenditures on direct-to-consumer and direct-tophysician advertising for each drug, geographic identifiers, and in some models, plan formulary design and initial drug to control for plan- and drug-specific effects.

	Table 1 shows the distribution of brand and generic prescribing within each of the 10 classes. Most physicians did not prescribe brand or generic medications exclusively, with some notable exceptions. Nearly half of the physicians prescribing ACE inhibitors and NSAIDs, and 90% of physicians prescribing opiates, prescribed only generic drugs in the class. By contrast, less than 1% of physicians prescribed only generic statins or PPIs. As the share of generic prescribing in a class increased, the proportion of physicians prescribing only generics increased and the share prescribing only brands decreased. In the 5 classes where the generic share is closest to one-half (38%-61%), between 80% and 89% of physicians prescribed both brand and generic medications as initial prescriptions.

	The distribution of the number of drugs prescribed per physician is shown in 

. To put these numbers into context, we also report the number of drugs that accounted for 75% of initial prescriptions in a class, and the market share of the top-selling drug. Only a small percentage of physicians prescribed a single drug in a class, ranging from less than 1% for SSRIs/SNRIs to 15% for ACE inhibitors. In 8 of the 10 classes, the median physician prescribed 3 or 4 different drugs. This reflected broad prescribing patterns given that the median number of initial prescriptions per physician in our sample ranged from 6 to 8 in the 10 classes. The case of SSRI/SNRI antidepressants was particularly striking: 45% of physicians prescribed 5 or more different drugs in the class. Of the 1659 physicians for whom we observed 8 to 12 initial prescriptions, 72% prescribed 5 or more different drugs and less than 2% prescribed 1 or 2 drugs.

 shows the distribution of physician prescribing in the 5 brand-dominated classes. Physicians prescribing 1 or 2 drugs were more likely to prescribe the leading drug in a class, which in most cases was the most heavily promoted drug. For example, among physicians prescribing just 1 statin, 80% prescribed the market leader and most heavily promoted drug. The generic share increased with the number of initial drugs prescribed in 3 of the 5 classes, while PPIs and antihistamines exhibited a different pattern.

	In the PPI class, which had only 1 generic drug (omeprazole) during 2005 to 2007, the generic share decreased monotonically with the number of drugs prescribed (as did the share of the top brand drug), indicating that prescribers with narrow prescribing patterns in this class were split between high prescribers of the top (brand) drug and high prescribers of generic omeprazole. Perhaps due to the degree of similarity between these 2 products (esomeprazole, the top brand, and generic omeprazole), physicians generally prescribed one drug or the other. For example, among the 1229 physicians prescribing just 2 drugs in the class, only 23% prescribed both esomeprazole and omeprazole, while 46.5% prescribed the leading brand and another brand drug, and 20% prescribed generic omeprazole and another brand drug. By contrast, the leading brand and generic antihistamines have different active ingredients and most physicians prescribed both. Overall, as physicians showed broader prescribing patterns, they moved away from the most prescribed drug in a class toward generics and/or less common brands.

	Physicians treating patients with different comorbidities had broader prescribing patterns. This pattern occurred in all 5 classes and is nearly monotonic (Table 3). Further, physicians treating patients from a larger number of health plans (and formularies) were more likely to have broader prescribing habits. These results were robust to multivariate models that controlled for detailed patient and plan characteristics (results not shown; see 

	If prescribers with broader prescribing habits are better able to match a patient to their optimal drug, we might observe better adherence to medications and less switching within class. We found that broader prescribing was associated with modestly better adherence in 2 of the 5 classes (

). Patients prescribed PPIs and antihistamines by a physician in the broadest category of prescribing were 7% to 8% more likely to continue use for 6 months than a patient treated by a physician who prescribes with the most narrow prescribing patterns. However, we found no statistically significant differences for SSRIs/SNRIs and calcium channel blockers, and a small opposite effect (lower adherence) for statins. Similarly, we found little evidence to suggest that broader prescribing significantly affected switch rates or the average out-of-pocket cost per 30-day prescription.

	There is a widespread perception that physicians prescribe a narrow range of drugs within a therapeutic class. This is often attributed to 2 primary factors. The first is clinical experience, wherein physicians gain knowledge of a particular drug through experience and then prescribe it to most of their other patients who require a drug in that class. The second factor is pharmaceutical marketing. Prior work has established that detailing has a significant effect on prescribing behavior and brand loyalty, particularly among physicians with limited access to colleagues.

	Despite these perceptions, we find surprisingly broad prescribing across 10 prominent classes. While 40% to 60% of their prescriptions are for 1 drug, the median physician in our sample prescribed at least 3 different drugs for incident users in 8 of the 10 classes. These results were even more striking considering the small number of initial prescriptions per physician (median = 8), and the dominance of brand drugs in 5 of the 10 classes studied. Physicians whose patients were covered by a wider array of health plans and formularies tended to have broader prescribing habits, as did physicians who treated patients with varying comorbidities. This suggests that attempts to match specific drugs to a patient’s health condition and formulary design were important factors in deviating from their favorite drug. While physicians whose prescribing habits were narrow were more likely to prescribe highly advertised drugs, few doctors prescribed these drugs exclusively.

	Our results suggest that physician prescribing habits were less entrenched than commonly perceived. Why we observed these patterns is unclear. Broad prescribing patterns may simply reflect the increasing number of drugs in a class, many of which act in a similar way and share common side-effect profiles. Broad prescribing may also reflect the influence of pharmaceutical marketing, but with less pernicious effects. Surveys of physicians reveal that detailing is an important source of information for many providers, and that drug samples provide greater flexibility in prescribing to low-income patients.

 The widespread availability and use of drug samples may provide the clinical experience physicians depend on to assess the efficacy and benefits of new products.

	An alternative explanation for the observed breadth of prescribing is the influence of manufacturers, pharmacy benefit managers, and third-party payers. Through explicit campaigns that promote switching to “featured” products or financial incentives inherent in the formulary design, physicians and patients may be steered toward a wider array of products than in the past.

 These incentives may interact, as prior research suggests that advertising affects demand only for drugs that have preferential status on the patient’s formulary.

Physicians prescribe more broadly than commonly perceived. Although most physicians have a favorite drug, they are not reluctant to try new therapies.

Physician Prescribing Behavior and Its Impact on Patient-Level Outcomes

 Key stakeholders agree better data on patients’ language are needed to effectively address language-related barriers to timely, highquality healthcare. Our objective was to describe health plan efforts to collect language data from its members, provide language services, and improve the provision of culturally and linguistically appropriate services (CLAS).

 National surveys in 2003, 2006, and 2008.

 Surveys were administered to health plans offering commercial, Medicaid, and/or Medicare Advantage products.

 123 health plans responded to the 2008 survey (50% response rate), including 65 commercial (50%), 46 Medicaid (53%), and 12 Medicare plans (44%), representing a total enrollment of 133.8 million Americans. In 2008, 74.0% of health plans collected language data (commercial 60.0%, Medicaid 89.1%, Medicare 91.7%), which is an increase for each plan type since 2003. Health plans used direct and indirect collection methods. Nearly all health plans reported offering language services, the most common being telephonic interpreting, multilingual member materials, and access to bilingual providers. A variety of strategies for improving CLAS were cited by health plans, including improving health plan communication materials, health literacy initiatives for members, and targeted training for providers and staff.

 Health plans have made substantial progress in the collection of language data and many are offering options for language services. With the rapid growth in Medicaid participation and newly insured individuals anticipated under the Affordable Care Act, health plans may be uniquely positioned to implement and test interventions that aim to improve appropriate utilization of language services by providers and patients.

	Key stakeholders agree that better data on patients’ language are needed to effectively address language-related barriers to timely, high-quality healthcare.

		In 2008, 74.0% of health plans collected language data (commercial 60.0%, Medicaid 89.1%, Medicare 91.7%); an increase since 2003.

		Health plans cited a variety of strategies for improving the provision of culturally and linguistically appropriate services and nearly all plans reported offering language services in 2008.

		Health plans may be uniquely positioned to implement and test interventions that aim to improve appropriate utilization of language services.

	Over the past several decades, the United States has experienced rapid growth in the racial and ethnic diversity of its population. By 2050, racial and ethnic minorities are estimated to comprise 54% of the nation’s population.

 Furthermore, in 2000, approximately 47 million Americans reported speaking a language other than English at home, and 21 million reported speaking English less than “very well.”

 Several studies have documented the negative impact of language barriers on patient-provider communication, patient safety, and the overall timeliness and quality of healthcare received.

	Although expanded insurance coverage is an important step in reducing gaps in care, it does not guarantee that patients, especially persons with limited English proficiency and low functional health literacy, can effectively navigate the healthcare system to access preventive services and obtain the care they need.

 Availability of a full range of culturally and linguistically appropriate healthcare services is essential for overcoming barriers and accessing timely care. To identify gaps in care for diverse populations, develop strategies to improve patient-provider communication, and allocate needed services and resources, key stakeholders agree that better data on patients’ language are needed.

	This article reports the results of 3 nationwide surveys of health plans regarding their efforts to improve the provision of culturally and linguistically appropriate services for their members.

 Specifically, we describe how health plans collect and use language data from their members, and what types of language access services are provided to members by plans. Our goal was to describe recent progress made by health plans and to identify areas needing further improvement.

	The sources of data for this study were surveys of health plans conducted in 2003, 2006, and 2008 by America’s Health Insurance Plans Foundation (AHIPF), with support from the Robert Wood Johnson Foundation. Reported results are primarily from the most recent survey (2008), although selected questions from the prior 2 surveys were used to identify trends over time. The target population was defined as health plans offering commercial, Medicaid, and/or Medicare Advantage products.

	We developed the 2008 survey using input from an advisory group of health plans with expertise in data collection. We pilot tested the survey with 5 health plans and made revisions based on their feedback. The final survey instrument, consisting of 50 questions, had several sections. For this article, we used information on whether health plans collected data on the “primary” or “preferred” language of their members and how they collected these data. “Primary” and “preferred” language was not explicitly defined in the survey, as plans may utilize different definitions. The questions regarding how plans collected data focused on whether each used direct methods, defined as collection by the plan of self-reported data from its members; indirect methods, defined as collection of data from sources other than members; or both. We also used information on the provision of language access services for members and strategies employed by health plans for improving the provision of culturally and linguistically appropriate services (CLAS). Of note, these questions were asked of all plans regardless of whether or not they collected language data.

	America’s Health Insurance Plans Foundation used health plans’ listings in the Atlantic Information Service’s (AIS) Directory of Health Plans for 2007 to develop the sampling frame for the survey, and excluded leased preferred provider organization (PPO) networks, subsidiary companies, plans that were no longer in business or had merged; plans with unknown enrollment; and very small plans (defined as commercial plans with <6500 members, Medicaid plans with <20,000 members, and Medicare plans with <5000 members). This resulted in a final sampling frame consisting of 245 health plans, all of which were invited to participate in the survey. For each plan, AHIPF used AIS data to identify the product with the highest enrollment and asked that responses be specific to that product. Because the commercial product has the highest enrollment for most health plans, 11 multiproduct health plans were asked to respond for 1 additional product, Medicaid or Medicare, so that the distribution of product-specific enrollment in the overall sample would reflect the composition of the industry.

	The Excel-based survey was fielded via e-mail between June and October 2008. Prior to e-mailing the survey, AHIPF utilized multiple internal and external contact lists (eg, health plan staff responsible for addressing disparities within individual companies, chief medical officers, directors of quality improvement, etc) to identify the appropriate health plan representative to complete the survey. America’s Health Insurance Plans Foundation staff made several attempts to reach nonresponding plans and contacted submitters to clarify incomplete or unclear responses. On average, health plans completed the survey in approximately 1 hour.

	A total of 123 health plans responded to the 2008 survey (50% response rate), including 65 commercial (50%), 46 Medicaid (53%), and 12 Medicare plans (44%). Response rates varied by plan size: 76% for large plans (enrollment 

300,000 for Medicaid and Medicare plans); 64% for medium plans (enrollment 200,000-499,999 for commercial and 100,000-299,999 for Medicaid and Medicare plans); and 37% for small plans. Responding health plans represented a total enrollment of 133.8 million members.

	The main methodological differences between the 2008 survey and the earlier 2003 and 2006 surveys included:

	• In 2003, 83% of the health plans in the sampling frame were included in the survey sample via random selection, while in 2006 and 2008, 100% of the plans in the sampling frame were surveyed.

	• In 2003 and 2006, all health plans were asked to respond on only 1 product.

	• In 2003 and 2006, the survey was conducted via an e-mail invitation to health plans with a link to an Internet-based questionnaire.

	• The 2003 and 2006 surveys asked about collection of “primary” language only.

	• Overall response rates in 2003 and 2006 were 40% and 60%, respectively. As in 2008, large- and medium-size health plans were more likely than small plans to respond to the surveys.

	We calculated the proportions of health plans with various characteristics of interest. When appropriate, we weighted the proportions by plan enrollment. Because the responding health plans account for a large percentage of the sampling frame, we applied the finite population correction factor to obtain correct standard errors (SEs). In the Results section, all percentages are presented with the SE in parentheses.

	The 2008 study sample included large, medium, and small health plans that were geographically distributed across all 4 US census regions, and that together represent a sizable percentage of the total US health plan enrollment (

). Commercial plans that responded to the survey included health plans that offered health maintenance organization, PPO, and/or point of service products. Health plans in the study sample for 2003 and 2006 were similarly diverse in size, location, and product types.

	Approximately 74.0% (SE, 2.8%) of health plans reported collecting language data (“primary” and/or “preferred”) for their members in 2008, including 60.0% (4.3%) commercial, 89.1% (3.2%) Medicaid, and 91.7% (6.3%) Medicare plans. This was an increase from 2003, when 38.5% (5.6%) of commercial, 84.6% (5.9%) of Medicaid, and 65.6% (6.5%) of Medicare health plans reported collecting primary language data (

). Most health plans that collected language data from their members also collected information on race and ethnicity.

	Larger commercial and Medicaid plans were more likely to collect language data than medium or small health plans. For example, in 2008, 81.5% (4.0%) of large commercial plans collected these data, compared with 54.5% (10.1%) and 40.7% (7.8%) of medium and small plans, respectively. Similarly, 100% of large Medicaid plans reported collecting these data, compared with 93.3% (4.2%) and 83.3% (5.9%) of medium and small plans. When weighted by plan enrollment, 89.1% of commercial plan members, 96.8% of Medicaid plan members, and 95.3% of Medicare plan members were in health plans that collected some language data in 2008.

	Health plans used both direct and indirect methods to obtain language data on their members. In 2008, the majority of commercial plans that collected language data relied on direct methods (66.7% [5.4%]) with only 2.6% (1.9%) reporting the sole use of indirect methods, and 30.8% (5.3%) reporting use of both data collection methods. By comparison, Medicaid and Medicare plans that collected language data used only direct methods less often (14.6% [3.9%] and 55.6% [13.4%], respectively), and only indirect data collection methods more often (24.4% [4.7%] and 22.2% [11.2%], respectively). Both methods were used by 61.0% (5.3%) of Medicaid plans and 22.2% (11.2%) of Medicare plans.

	Health plans reported a variety of direct methods for collecting information on members’ language. However, the direct method most frequently reported by commercial and Medicare health plans as providing the most valuable data was having members self-identify their primary and/or preferred language during enrollment, which was used by 52.6% (5.8%) of commercial and 71.4% (14.2%) of Medicare plans that used direct methods. The direct method most frequently chosen by Medicaid plans as providing the most valuable data was obtaining language data directly from members during contacts with plan customer service agents, which was used by 90.3% (3.7%) of Medicaid plans that used direct methods. Other direct methods commonly used by commercial, Medicaid, and Medicare plans are reported in the upper panel of 

	The indirect method most frequently cited by Medicaid and Medicare plans as providing the most valuable data was linking to files from external sources such as the Centers for Medicare & Medicaid Services (CMS) and state Medicaid agencies (Table 2). This method was used by 90.9% (3.5%) and 75.0% (19.8%) of health plans that collected indirect data, respectively. By contrast, the indirect method most frequently identified by commercial plans as providing the most valuable data was the use of geocoding software or obtaining data from the Consumer Assessment of Healthcare Providers and Systems survey, or another similar patient experience survey, which was used by 72.7% (10.2%) of plans. Surname identification software and geocoding software were used only by commercial plans and Medicare plans (Table 2).

	Health plans were asked to name the top 5 primary or preferred languages, other than English, most commonly identified by members. Among plans that collected language data, the languages most frequently reported were Spanish (96.6%), Chinese (60.9%), Vietnamese (47.1%), Russian (24.1%), and Korean (21.8%). In all, 35 different languages were reported.

Health plans have made substantial progress in the collection of language data and many are offering options for language services.

December 2011

Toward a Business Case for Performance Improvement

Self-Reported Health and Functional Status Information Improves Prediction of Inpatient Admissions and Costs

Physician Prescribing Behavior and Its Impact on Patient-Level Outcomes

Collection of Language Data and Services Provided by Health Plans

Comparing Variation in Medicare and Private Insurance Spending in Texas

Cost Savings in Inpatient Oncology Through an Integrative Medicine Approach

Adjunctive Aripiprazole for Depression: Predictive Value of Early Assessment

Costs of Providing Antiplatelet Medication for Percutaneous Coronary Intervention Patients

Effect of Market Competition on Hospital Performance for Heart Failure