 Creating the value proposition for innovations in personalized genomic medicine requires generation of evidence-based demonstrations of clinical utility and cost-effectiveness.

 To assess economic studies of genomic testing for women with breast cancer and to understand the value of genomic testing for multiple stakeholders.

A structured review of the literature was conducted to identify and synthesize available evidence regarding economic analyses of genomic testing for breast cancer. A search was conducted using PubMed and Google Scholar for articles published between January 1, 2005, and December 31, 2010. The search was then expanded to include articles as far back as 1981. In  addition, snowball methodology was used to identify and include additional articles based on frequency of author publication and frequency of citation in the literature.

Of the articles reviewed, a subset of 9 articles describing specific economic analysis studies were included in a more in-depth, side-by-side comparison. This review of the literature on the economics of genomic testing for women with breast cancer found that most of the economic evidence relied on modeling rather than clinical trial data.

 Facilitating the diffusion of new technology will require more data to satisfy the payer, provider, and societal perspectives. Conversely, willingness by payers and clinicians to consider economic modeling data as part of their evaluation of new technologies can help facilitate the diffusion of newly developed genomic tests.

Many studies document the scientific basis for genomic testing for breast cancer, but few tools allow payers to assess the value of these genomic tests.

		Economic research on genomic testing for breast cancer has not involved randomized controlled trials or other direct trial data.

		Research on the economics of genomic testing will consist primarily of modeling studies for the foreseeable future.

		Payers should use economic models as the best available evidence for which genomic tests should be reimbursed.

		Payers should demand more funding for high-quality prospective trials of genomic tests with an economic evaluation.

In women, breast cancer accounts for nearly 1 in 4 cancers. It is the most common cancer in women aside from skin cancer. It has the second-greatest mortality rate for women, exceeded only by lung cancer.

 Treatment pathways for breast cancer recurrence  after primary surgical and radiation therapy are complex and are based on multiple known risk factors. Women are offered adjuvant chemotherapy if the risk of recurrence is above a certain threshold.

While the benefit of adjuvant chemotherapy is well established in patients with early-stage breast cancer that is estrogen receptor  ER) positive and lymph node negative, 65% of women diagnosed with invasive breast cancer have lymph node—negative disease; of these, 15% are expected to die or have distant metastasis in 10 years.

 Currently, novel genomic evidence about the diversity of patients is challenging scientists to develop and validate more personalized approaches to treatment.

Advances in genomic diagnostics and personalized treatment have the potential to improve health outcomes, reduce mortality, and  increase quality of life for cancer patients. Genomic tests analyze the genetic profiles of patients’ tumors by generating a “recurrence score” based on a particular algorithm. The recurrence score predicts the likelihood of cancer recurrence and informs decision making.

Several types of genomic tests are available today in the United States for breast cancer, including Oncotype DX, MammaPrint, and CancerTYPE ID. Oncotype DX is a 21-gene assay test that gives patients with stage 1 or 2 ER-positive, lymph node-negative breast cancer a recurrence score between 0 and 100.

 MammaPrint is a 70-gene assay to determine the risk of cancer recurrence for ER-positive or ER-negative, lymph node—negative patients with early-stage breast cancer (stages 1 and 2).

CancerTYPE ID measures a 92-gene expression taken from a tumor biopsy to determine 30 different types of tumors. A component of the test,  Breast Cancer Index, can be ordered if the tumor is the suspected primary source.6 CancerTYPE ID is not as widely used as the other 2 tests in breast cancer.

The total cost of breast cancer includes the financial burdens for the patients, their families, and society. The National Cancer Institute estimates the national annual financial cost of breast cancer care at $13.9 billion.

 Breast cancer care represents the largest portion of all cancer care expenditures. As the healthcare system works to find ways to pay for genomic technology and healthcare costs continue to trend upward, the potential cost savings and productivity gains associated with advances in genomic diagnostics and personalized treatment that avoid unnecessary and ineffective treatments are appealing.

Economic analysis provides a framework for assessing the value of clinical outcomes, as well as for determining how that value might differ by population.

 This literature review improves upon previously published literature reviews of genomic testing for breast cancer by focusing on the economic evidence. We are not aware of other structured reviews on this topic that utilize the economic perspective. Our key findings are that research in this area has been limited and is likely to consist primarily of modeling studies for the foreseeable future. Additional funding, better outcomes data, and regulation are identified as barriers to adoption of this new technology.

An economic evaluation prioritizes the efficient use of scarce resources. As a result, economic evaluation leads to the test that best balances the trade-offs among available alternatives rather than simply selecting the cheapest or most valid test.

While data from randomized controlled trials are a gold standard, an economic analysis uses models to combine costs and outcomes. The combination of both types of data in a model that allows for the evaluation of the cost-effectiveness of a particular type of technology results in information that maximizes efficiency.

This framework guided the snowball methodology, described in further detail in the

). In order to appear in our literature review, a study had to cover the clinical area of breast cancer, with the scientific application of a genomic test. Further, the analysis had to include a cost-effectiveness analysis broadly defined. A costeffectiveness analysis uses 2 basic building blocks—economic cost and clinical effectiveness&mdash;to assess the value of a test, drug, or other healthcare product. Value comes from calculating the incremental improvement derived from a particular test over existing technology.

This conceptual framework resulted in a review of 9 studies as described in the section titled Published Studies on Economic Outcomes. Due to the small number of studies available for each topic area of the review, we present studies on an individual basis rather than presenting summary results across studies.

Several studies have been published utilizing economic modeling methods to analyze the benefits and costs of genomic testing for breast cancer. The costs of genomic testing are 2-fold: the cost of the genomic test itself and the costs associated with  false-negative or false-positive results. For a complete listing of cost variables, please see the Ishikawa Fishbone Analysis

 depicted in the Figure. The Jefferson Population Health Continuing Professional Education Collaborative created the diagram based on the references cited in the article. Thus, the Figure is an original work based on the source material from this literature review.

The direct cost of genomic tests is clear and varies by test and payer. The evidence report commissioned by the Evaluation of Genomic Application in Practice and Prevention Working Group and the Agency for Healthcare Research and Quality identified the 3 genomic tests for women with breast cancer that are clinically available in the United States (MammaPrint, Oncotype DX, and CancerTYPE ID).

 The cost is generally regarded as $3460, with a range of $1960 to $4860. The cost for a particular patient depends on factors such as insurance coverage policy and regional variability.

Economic modeling studies have focused primarily on the direct costs associated with breast cancer treatment and recurrence, while failing to fully address indirect costs. For patients, these indirect costs play a significant role in treatment decision making. The false-negative cost mainly includes the cost of nonfatal recurrences and mortality during recurrence or chemotherapy, and is estimated at $51,000.

 Medical costs and related nonmedical costs of unnecessary chemotherapy are high: treatment costs varied from $10,000 to $23,000 in a prior study that was included in our side-byside comparison.

 In another study included in our comparison, adverse events related to chemotherapy were found to cost more than $2000 per patient.

 In a third study included in our comparison, Bacchi and colleagues

 stated that the 18% difference in cost between use and nonuse of an assay could be attributed to the cost of medications for prophylaxis and treatment of side effects of chemotherapy.

There is limited mention of indirect costs (eg, productivity loss) in any of the published studies we reviewed. There have been attempts to estimate the burden of side effects in terms of quality of life or other patient-reported outcome measures. However, there is no systematic study that translates indirect costs into monetary terms. Data used in a payer perspective study that we reviewed focused on the avoidance of costs of long-term adjuvant chemotherapy such as infertility and second primary tumors.

 in 2011 that was included in our comparison also did not translate indirect costs into monetary terms.

The most significant benefit expected from tailoring treatment using genomic testing is a reduction in adverse drug effects for subpopulations that may not benefit from chemotherapy. The adverse effects of chemotherapy include nausea, vomiting, alopecia, fatigue, vasomotor symptoms, pain, and risk of infection.

 found that 85% of patients receiving chemotherapy experienced nausea and vomiting and suffered distress, 36% of patients lost their hair, and 39% of patients developed mucosal ulceration. They also reported that depression and anxiety occurred in patients receiving chemotherapy, although psychiatric morbidity was present for some of these patients even prior to invasive surgery.

A major benefit of genomic testing is extended quality-adjusted life-years (QALYs). Hornberger and colleagues

 analyzed the preliminary QALYs for using 6-month chemotherapy treatment for “low-risk” patients. Their analysis, included in our comparison, found that if the utility of chemotherapy was set at 0.5 for 6 months of treatment, then chemotherapy was of no benefit for those low-risk patients, as no gain in QALYs resulted. However, survivors who received chemotherapy tended to have greater preference for undergo ing chemotherapy than survivors who did not receive chemotherapy.

 This and other previous studies have shown there is ambivalence and inconsistency in patients’ reports on how they view the value of chemotherapy. The literature does not clearly provide information on reduction of QALYs for the breast cancer patients who were distant recurrence free after primary surgical and radiation therapy but who were offered chemotherapy according to current breast cancer treatment practice.

The economic evaluation of genomic testing requires a valuation of whether the benefits of early diagnosis and treatment outweigh the costs of the test. This framework can include comparing direct and indirect costs, as well as clinical, financial, and humanistic benefits. Which costs and benefits  are included depends on the study’s perspective. Using decision analysis, one can choose a particular test over other diagnostic steps or interventions.

Modeling-Based Cost-Effectiveness Analysis

For the decision analysis of genomic testing, we included 3 studies that compared clinical and economic outcomes of genomic testing—guided adjuvant treatment versus guidelinebased adjuvant therapy in our comparison.

 We also included in our comparison 2 studies of different adjuvant treatment strategies: chemotherapy treatment for all patients (or no chemotherapy) or adjuvant therapy strategies guided by genomic testing.

We included 5 cost-effectiveness studies of Oncotype DX, all of which concluded that treatment guided by genomic testing is cost-effective compared with conventional treatment selected according to guidelines. In 1 study, Lyman and colleagues

examined the incremental cost-effectiveness ratio associated with 3 treatment strategies: tamoxifen alone, chemotherapy followed by tamoxifen, and recurrence score—guided treatment. They found that genomic testing is costeffective relative to the other strategies.

 applied recurrence score—guided recommendations to patients of a managed care organization in Israel. They found that treatment pathways were changed from the initial recommendation (according to traditional prognostic pathways) in 40% of patients and that the incremental cost-effectiveness ratio for using Oncotype DX was $10,770 per QALY.

That is well below the international norms of willingness to pay used in Israel and the threshold of $50,000 per QALY gained commonly used in the United States context (

One study included in our comparison discussed cost-effectiveness analyses of a 70-gene expression-profiling test (MammaPrint). This study, performed in 2005, found that MammaPrint had lower costs and worse outcomes, in terms of lower life expectancy and QALYs, than the National Institutes of Health clinical guidelines for treatment of breast cancer. Specifically, the study found that use of the test “resulted in an absolute 5% decrease in the proportion of cases of distant recurrence prevented, 0.21 fewer QALYs, and a cost savings of $2882.

A more recent study performed in 2010, also in our comparison, found that MammaPrint had a favorable  incremental cost-effectiveness ratio of $7000 per QALY and $10,000 per life-year saved compared with Adjuvant! Online software  guidance.

 The contradictory findings from these 2 studies are primarily due to the control group being used—National Institutes of  Health guidelines in one case and decision software in the other case. The findings may also be related to the timing of these  studies, one of which was published 5 years later than the other.

Two international studies included in our comparison conducted generic cost-effectiveness analyses of genomic testing rather than selecting a specific marketed product. One European study used a different type of genomic test (189-gene expression profiling) not usually available in the US market and applied cost-minimization analysis.

 They found that compared with conventional adjuvant chemotherapy, genomic testing is cost-effective only if the test cost is less than €2090 ($2720).

 A Brazilian study focused on the financial impact of using genomic testing (21-gene expression profiling) by taking into account the medical costs and excluding  utility or QALY measures.

 conducted a Web-based survey among medical oncologists in Brazil and  compared the costs associated with individual decisions for treating hypothetical patients with the costs associated with 21-gene  expression assay—guided decisions. They found that for a hypothetical cohort of 100 patients with access to the test, $79,400  would be saved in direct medical costs.

Articles

REVIEW ARTICLE

All the economic studies of genomic tests for breast cancer rely on modeling rather than randomized controlled trials or other direct trial data.

Economics of Genomic Testing for Women With Breast Cancer

 To investigate differences before and after rollout of electronic prescribing (e-prescribing) in (1) patients’ primary adherence to newly prescribed medications, (2) patients' understanding of how to use their medications, and (3) multiple pharmacy use.

 Postvisit interviews and follow-up phone calls were done with 344 patients at an academic general internal medicine clinic.

Patient interviews and follow-up phone calls were done (1) before e-prescribing, (2) 1 to 6 months after e-prescribing, and (3) 12 to 18 months after e-prescribing.

 Overall, rates of abandoned prescriptions were 6.9% before e-prescribing, 10.6% 1 to 6 months after e-prescribing, and 2.5% 12 to 18 months after e-prescribing (

 = .07). There was a reduction in awareness of the indication for a newly prescribed medicine among patients after e-prescribing (95.4%, 97.9%, and 89.8%, respectively; 

 = .03). There was a decrease in patients’ ability to demonstrate proper use of their new medicine after e-prescribing (69.0% before e-prescribing, 67.1% 1-6 months after e-prescribing, 51.9% 12 -18 months after e-prescribing; 

 = .02). There was an increasing trend in the percentage of patients using multiple pharmacies after e-prescribing was implemented.

We found both potential benefits and unexpected consequences as a result of the rollout of electronic prescribing. Adaptation to  e-prescribing might be improved with outreachand education, including explicitly informing patients of the change during the first months of rollout. Tangible prescription information for reminder purposes only may also be beneficial.

Postvisit interviews and follow-up phone calls among 344 patients (recruited from an academic general internal medicine clinic) were conducted to determine if and when patients picked up their new prescription and their understanding of it.

		Initially, rates of abandoned prescriptions increased after e-prescribing, but they later resolved to rates below baseline.

		There were decreases in patients’ ability to demonstrate both proper use of and knowledge about their medication.

		There was an increasing trend of multiple pharmacy use after implementation of eprescribing.

		These results suggest the need for improved outreach and education during e-prescribing rollout.

Between 2009 and 2011, there was a 72% increase in electronic prescribing (e-prescribing), from 191 million to 326 million e-prescribed orders.

 From a quality and safety perspective, e-prescribing has been thought to have the potential to improve patient care by improving clinic efficiency, preventing medication errors, and even improving regimen adherence.

 Yet to date there is  limited evidence on the impact of e-prescribing on the patient experience in primary care, including adherence-related concerns.

Despite the promise of e-prescribing to improve healthcare quality, a possible consequence could be primary nonadherence (ie,  e-prescriptions would actually negatively impact the timely retrieval and purchase of a new prescription). The hypothesis follows that despite the expedited order from prescriber to pharmacy, e-prescribing removes the known tangible reminder to fill a prescription by eliminating the paper prescription. Additionally, it is plausible that the nature of physicianpatient communication during a medical encounter on a newly prescribed medicine could change with greater efficiency, which may impact essential patient understanding of what a prescribed medicine is for (indication) and proper daily dosing. Beyond adherence concerns, the process of e-prescribing could lead to issues with medication reconciliation, as it requires patients to identify the pharmacy that they would like the order to be directed to; multiple pharmacies can be entered per patient (eg, pharmacy near work vs pharmacy closer to home). Multiple pharmacy use has been associated with lower compliance, higher risk of potentially dangerous or inappropriate drug combinations, and higher costs of pharmaceutical services.

In 2009, our team was conducting baseline interviews as part of a clinical trial evaluating an electronic health record strategy to  promote safe, appropriate medication use. Six months later, the clinic implemented e-prescribing for the first time, allowing for a  natural experiment to take place. While it was not the original intention of the study, we identified a unique opportunity to leverage extensive data collection to explore several critical research questions. We were able to investigate differences before and up to 18 months after e-prescribing was implemented in the clinic in the following outcomes: (1) patients’ primary adherence to newly  prescribed medications as determined by rate of prescription abandonment (unfilled prescriptions) and delays in filling a prescription; (2) medication understanding  as determined by patient understanding of a new prescription medication’s indication and demonstrated proper use (number of pills per dose, times taken per day, total number of pills to be taken daily); and (3) multiple pharmacy use.

We conducted a cross-sectional evaluation examining the impact of e-prescribing implementation within 1 primary care clinic, with 3 waves of patient interviews. Specifically, 1 baseline assessment was conducted during the 6 months prior to the implementation of e-prescribing (before e-prescribing), and 2 posttest assessments were performed, the first during the 6 months after implementation of e-prescribing (e-prescribing interval 1) and the second 12 to 18 months after implementation of eprescribing (e-prescribing interval 2). 

Adult patients (N = 344) receiving care at 1 academic general internal medicine ambulatory care clinic were recruited between September 2009 and March 2011. Three cohorts of patients were recruited: 144 patients before e-prescribing, 127 patients during the first 6 months after e-prescribing, and 73 patients 12 to 18 months after e-prescribing implementation. Individuals were eligible if they (1) were 18 years  or older, (2) were established patients at the clinic, (3) had an appointment with their physician on the day of recruitment, and (4) received a new order for a prescription medication during their visit. Patients receiving only orders for refills (ie, no orders for a new prescription medication) were not eligible to participate, nor was anyone with a moderate to severe visual, hearing, or cognitive impairment as determined by clinical staff or the interviewer at the time of recruitment. If patients received both an order for a refill and an order for a new prescription, they were eligible to participate. Patients were also ineligible if they had participated in an earlier interview wave for this study. The Northwestern University Institutional Review Board approved the study prior to its initiation.

Trained research interviewers working with clinic physicians and staff identified eligible patients upon their medical encounter at discharge. Specifically, clerical staff provided patients at check-in and discharge with a flyer that described the study in some detail as well as eligibility requirements. Staff directed interested patients to the available research staff who were waiting on site. Those who consented to participate then completed a brief, interviewer-assisted survey that included a literacy assessment. Interviewers notified patients they would receive  a follow-up phone call with additional questions about their prescription, such as if and when they filled the medication and how they were taking it.

Patients provided information regarding age, sex, marital status, race, education, income, type of insurance, number of medications currently prescribed, and number of comorbid conditions. Each patient also completed the Rapid Estimate of Adult Literacy in Medicine to assess literacy. The encounter discharge summary was reviewed to obtain the name, dose, and frequency information for the newly prescribed medication. Follow-up phone interviews occurred from as early as 6 days to as long as 2 weeks after the initial interview. During these follow-up phone interviews, appropriate medication knowledge (indication, side effects) and proper use (number of pills, number of times per day, time of day) were assessed. Additionally, patients were asked if they had filled their prescription, and if not, the reason for not obtaining the medication and the number of pharmacies used.

Data were analyzed using SAS version 9.2 (SAS Institute Inc, Cary, North Carolina). Descriptive statistics were calculated for each variable. Chi-square tests were used to evaluate the association between sociodemographic characteristics and e-prescribing period according to the time point that patients were recruited into the study (before e-prescribing [n = 144], e-prescribing interval 1 [n = 123], or e-prescribing interval 2 [n = 73]). Differences in primary adherence and multiple pharmacy use were also examined using the x

 <.05. Differences in medication knowledge (indication and proper use) by eprescribing period were examined as an additional exploratory outcome.

 presents the sociodemographic characteristics of subjects stratified by e-prescribing period. A total of 428 new medications were prescribed among the 344 patients. Specifiimplemented (29.9% in e-prescribing interval 1 and 26.0% in e-prescribing interval 2 vs 13.2% before e-prescribing; 

 = .003). Patients recruited after implementation of e-prescribing were more likely to have limited literacy (

 = .05). While not statistically significant, more patients in e-prescribing interval 2 had 3 or more chronic conditions compared with the 2 other time frames (37.0% e-prescribing interval 2, 33.9% e-prescribing interval 1, and 24.3% before   e-prescribing; 

= .12). More patients recruited before e-prescribing had private health insurance (76.4% before e-prescribing, 62.2% during e-prescribing interval 1, 66.7% during e-prescribing interval 2; 

 = . 09) and lower numbers of total prescription  medications compared with patients recruited after the implementation of e-prescribing (

Overall rates of primary nonadherence varied between recruitment periods (

). Nonadherence rates as measured by abandoned prescriptions were 6.9% before e-prescribing, 10.6% during e-prescribing interval 1, and 2.5% during eprescribing interval 2 (

= . 07). Patient-reported reasons for primary nonadherence included the medication being too expensive or a lack of prescription insurance coverage (20.7%), opting for an over-the-counter medication instead (17.3%), stating that they wanted to wait to see if they felt better (24.2%), and concerns about side effects (3.4%), as well as miscellaneous other concerns (34.4%) such as not liking samples given to them or not wanting to be on too many medications. Nonsignificant trends in delays in filling a prescription were also noted between periods, with 12.3% of patients delaying their time to fill a prescription before e-prescribingcompared with 8.5% during e-prescribing interval 1 and 6.4% during e-prescribing interval 2 (

Medications were classified into 5 categories: cardiovascular medications (20.1%), antibiotics (16.8%), analgesics/sedative hypnotics (14.5%), over-the-counter medications such as those for colds/allergies (12.2%), and “other” medications (36.4%). The other category included prescriptions that had too small a number to be included in a separate category (<5% of prescriptions). This category included antidepressants, hormones, anticonvulsants, and muscularskeletal, antimalarial, gastrointestinal, diabetic, respiratory, triptan, eye and ear, and dermatology medications. There were minor differences in medication category by e-prescribing period, as there were slightly fewer over-thecounter medications prescribed and slightly more analgesic/sedative hypnotics in e-prescribing interval 2. Prescriptions were also classified as acute or chronic medications. There were no differences in acute versus chronic medications by e-prescribing period (36.4% were chronic medications before e-prescribing, 30.0% during e-prescribing interval 1, and 41.3% during e-prescribing interval 2; 

 = .19). No differences in adherence rates were found by medication type or by chronic versus acute.

An increasing but nonsignificant trend was noted in the percentage of patients using multiple pharmacies after e-prescribing was implemented (20.0% before e-prescribing, 26.5% during e-prescribing interval 1, and 30.1% during e-prescribing interval 2; P = . 23). Interestingly, among patients with adequate literacy skills, rates of prescription abandonment did not change from before e-prescribing to e-prescribing interval 1, but decreased approximately 50% by e-prescribing interval  2 (8.7% vs 8.9% vs 3.1%). Among patients with limited literacy skills, 0% of prescriptions were abandoned before eprescribing, but 14.6% of prescriptions were abandoned during e-prescribing interval 1. The prescription abandonment rate returned to 0% during e-prescribing interval 2. No other trends were found by literacy or age group.

Exploratory analyses also involved determining patients’ understanding of physician instructions, including knowledge of a medication’s indication and proper dosing. Rates of proper medication understanding varied between recruitment periods (Table 2). There was a reduction in patients’ awareness of the indication for a newly prescribed medicine after e-prescribing (95.4% before e-prescribing, 97.9% during e-prescribing interval 1, and 89.8% during e-prescribing interval 2; 

 = .03). More notable was the decrease in patients’ ability to demonstrate proper use of their new medicine after e-prescribing (69.0% before e-prescribing, 67.1%  during e-prescribing interval 1, and 51.9% during e-prescribing interval 2; 

Our study sought to add to the currently limited available evidence around the implementation of e-prescribing and its potential to improve patient safety and outcome without compromising healthcare quality.

 We documented an initial increase in abandoned prescriptions immediately following the implementation of e-prescribing. However, with time, these issues were resolved and even improved to rates lower than baseline. Other differences that were found between the before and after e-prescribing periods included slight increases in multipharmacy use and poorer patient understanding of a newly prescribed medication’s indication and  instructions for use. There a few possible explanations for these findings.

First, the immediate increase in abandoned prescriptions could reflect the process of adaptation to a significant change to primary  care practice, for both providers and patients. In a prior study, one-third of patients surveyed were unaware of e-prescribing and 12%  did not know a prescription had been sent.

 Without an explicit orientation to the new process,  patient confusion might be expected. A previous study examining pharmacy record data found that prescriptions delivered electronically were 1.64 times as likely to be abandoned compared with those that were not electronic (

 However, this is likely due to higher transmission rates, as it is difficult to accurately determine rates from handwritten prescriptions that often are never delivered to the pharmacy. In another investigation using more precise insurance claims information, e-prescriptions transmitted directly to the pharmacy were more likely to be purchased than those given directly to the patient.

 There may be a learning curve after implementation of e-prescribing, and our single-site study ultimately demonstrated resolution of any problems with time.

Second, while there appear to be benefits to e-prescribing over time in terms of primary adherence, we also found unexpected consequences. The poorer knowledge demonstrated by patients of their prescription’s indication for use and dosing instructions could be a spurious finding. Yet it might also suggest that this new automated prescribing process improved a physician’s efficiency at the cost of some elements of spoken patient communication. Another possible explanation might be that patients relied on the tangible, handwritten prescription they previously received from their doctor to obtain the indication and instructions for use. From our data, we do not have the ability to answer this question with any certainty.

Third, though e-prescribing was associated with a slight increase in use of multiple pharmacies, this increase was not statistically significant. Compared with before e-prescribing, there was a 50% relative increase in multiple pharmacy use during e-prescribing interval 2. This would seem understandable, though, as e-prescribing allows and even encourages use of multiple pharmacies by entering as many pharmacies as are used by a patient into the electronic record system. That may be a potentially problematic health system trend, as patients increasingly have multiple prescribers and more complex medication regimens that will be dispersed further across practices and pharmacies that do not adequately communicate with one another.

 Future studies might investigate use of multiple pharmacies as a risk for medication reconciliation and patient safety.

There are several limitations to this study. Results were based on patient self-report; therefore, rates of nonadherence were unrealistically low. There may have been social desirability bias in reporting adherence to new prescription medications. We did not have access to claims data that could have monitored primary adherence more objectively. In our study, new prescriptions were  classified into types of medications prescribed and acute versus chronic. Primary adherence did not differ by either classification. This is surprising, as a previous study assessing 82,245 new e-prescriptions found that primary nonadherence was frequent for medications treating chronic conditions.

 However, pharmacy claims data were leveraged, allowing adherence to be assessed more objectively. 

Furthermore, selection bias may have occurred, as it is possible that patients willing to enroll in a research study could differ from  those unwilling to participate. Patients consenting to participate were aware that they would be receiving followup phone calls  regarding their medicines, so results presented may vastly underestimate rates of nonadherence. Additionally, results were based  on a relatively small sample of patients. Only new prescriptions were assessed, so adherence measures for other medications  taken were not quantified. Finally, this study was a natural experiment examining the implementation of e-prescribing at 1 clinic.  Since this was not the original intention of the study, it was not adequately powered to detect these small differences in primary  adherence or medication understanding in study outcomes, and we did not pursue multivariable modeling. Clearly, our findings are  not meant to be generalizable but rather to help inform future investigations. 

While we did not find that e-prescribing ultimately increased the number of unfilled or delayed new prescriptions, our findings have some specific implications. Adaptation to the new clinic protocol and subsequent patient behavior of filling a prescription with e-prescribing might be improved with explicit outreach and education. Physician orientation should include basic communication training to ensure proper discussions with patients about new prescriptions. If e-prescribing brings with it a higher rate of multiple pharmacy use, educational campaigns may be necessary to help patients carefully keep track of their prescription regimen or to encourage patients to consolidate their regimens to a single, known pharmacy. One last recommendation might be to consider how a clinical practice might better disseminate tangible prescription information—as part of the after-visit summary, through a patient Web portal, or as a stand-alone print prescription for reminder purposes only. As more practices adopt e-prescribing, we should continue to explore the many opportunities to improve medication use and adherence. 

 From Health Literacy and Learning Program (ARB, JW, MS, ALR, MSW), Division of General Internal Medicine, Feinberg School of Medicine, Northwestern University, Chicago, IL; Division of General Internal Medicine (ADF), Mount Sinai School of Medicine, New York, NY; Division of Pharmacoepidemiology and Pharmacoeconomics (WHS), Department of Medicine, Brigham and Women’s Hospital and Harvard Medical School, Boston, MA.

This research was funded by grants 1R18HS017220-0 1 and R21CA132771 from the Agency for Healthcare Research and Quality.

Dr Shrank reports that he has received grants from CVS Caremark, Aetna, and the National Association of Chain Drugs Stores. The other authors (ARB, JRW, MS, ADF, ALR, MSW) 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 (ADF, WHS, MSW); acquisition of data (JRW, ALR, MSW); analysis and interpretation of data (ARB, MS, ADF, WHS, ALR, MSW); drafting of the manuscript (ARB, JRW, MS, ALR, MSW); critical revision of the manuscript for important intellectual content (ARB, MS, ADF, WHS); statistical analysis (ARB, MSW); provision of study materials or patients (JRW); obtaining funding (MSW); and administrative, technical, or logistic support (JRW).

 Ashley R. Bergeron, MPH, Division of General Internal Medicine, Feinberg School of Medicine, Northwestern University, 750 N Lake Shore Dr, 10th Fl, Chicago, IL 60611. E-mail: a-bergeron@northwestern.edu.1. SureScripts. National progress report on e-prescribing. http://www.surescripts.com. Published 2010. Accessed July 20, 2011.

2. Abramson EL, Malhotra S, Fischer K, et al. Transitioning between electronic health records: effects on ambulatory prescribing safety. J Gen Intern Med. 2011;26(8):868-874.

3. Fischer MA, Vogeli C, Stedman M, Ferris T, Brookhart MA, Weissman JS. Effect of electronic prescribing with formulary decision support on medication use and cost. Arch Intern Med. 2008;168(22):2433-2439.

4. van Doormaal JE, van den Bemt PM, Zaal RJ, et al. The influence that electronic prescribing has on medication errors and preventable adverse drug events: an interrupted time-series study. J Am Med Inform Assoc. 2009;16(6):816-825.

TRENDS FROM THE FIELD

Both potential benefits and unexpected consequences were found as a result of the rollout of electronic prescribing.

Impact of Electronic Prescribing on Medication Use in Ambulatory Care

To determine the extent to which health plans were collecting and using data on the race, ethnicity, and language proficiency of network physicians and other network providers.

Survey of health plans conducted in 2010, with results compared with data from similar surveys conducted in 2006 and 2008.

Surveys were sent by e-mail to representatives of 250 health plans identified through the Atlantic Information Service’s Directory of Health Plans: 2010. Basic descriptive statistics were used to characterize response patterns to individual items in the 2010 survey and to compare responses to matching items in the 2006 and 2008 surveys.

Approximately half of responding plans reported collecting data on race/ethnicity of providers. This proportion was not significantly different from that of previous years’ surveys, and several plans that had been collecting data on providers’ race/ethnicity in 2006 had discontinued  the practice by 2010. Nearly all plans reported collecting data about languages spoken for 1 or more types of employees or providers. Plans were  more likely in 2010 than in 2006 to verify language proficiency of staff and share this information with plan members.

 Even though health plans and other healthcare organizations were more likely in 2010 than in the past to collect data on member/patient race, ethnicity, and language preferences, collection and use of data on providers’ race/ethnicity have not expanded in scope since 2006. Collection and use of data on providers’ proficiency in languages other than English were more common. 

National surveys of health plans were conducted to determine the extent to which health plans were collecting and using data on the race, ethnicity, and language proficiency of network physicians and other network providers.

Plans were more likely in 2010 than in 2006 to verify language proficiency of staff and share this information with plan members.

Collection and use of data on providers’ race/ethnicity have not expanded in scope since 2006.

Collection and use of data on providers’ proficiency in languages other than English were more common.

Since 1998, a variety of public and private policy initiatives have encouraged health insurance plans to collect data on the race and ethnicity of their members,

 and health plans have also been either encouraged or required to collect data on language needs of members who are not proficient in English.

 The data are used to identify disparities in care, identify targets for quality improvement initiatives, and set priorities for translation and interpretation services. The overall goal is to ensure that access to care and quality of care do not vary as a function of race, ethnicity, or primary language, and to improve healthcare quality for all plan members.

In a closely related set of policy and health plan initiatives, data on the race, ethnicity, and language proficiency of network  physicians have also been collected.

 It is presumed, with some empirical justification,

 that physicians who share significant elements of background and culture with patients will find it easier to communicate essential healthcare messages, to better understand patients’ feelings and experiences, and possibly to provide better quality care.

Health plans that have information on the race, ethnicity, and language proficiency of providers can use this information to provide members with the opportunity to voluntarily select providers through a variety of specific mechanisms. Plans can use the information as a foundation for initiatives to increase the racial, ethnic, and linguistic diversity of physicians for inclusion in their provider networks so that there can be network physicians available for members who have linguistic needs or cultural preferences. More specifically, provider information available to members at the time of plan enrollment can include information on race, ethnicity, and language to enable members desiring a physician who can understand their cultural beliefs and language needs to make a selection using information provided by the plan.

Collection of data from physicians or other providers about race, ethnicity, and language poses some unique challenges to health  plans. Some providers may be reluctant to provide such data out of concern that patients may actually select against, rather than for, a specific characteristic.

 Providers may also fear that plans might use the information to exclude, rather than include, them in network contracts.

Providers with moderate fluency in a non-English language may indicate that they speak that language, but actually do not have sufficient proficiency to carry out complex medical discussions with patients.

 Some providers,like some  patients or plan  members, may not accept or understand the specific race/ethnicity categories and may find it difficult to answer such questions, even if otherwise willing.

 For all these reasons, health plans’ collection of data on the race, ethnicity, and language fluency of providers has been more complex and more challenging than collection of data on the race, ethnicity, and language needs of members.

The Patient Protection and Affordable Care Act (ACA) of 2010 created an additional element of context for the collection of race, ethnicity, and language data on providers, as it included a requirement that health plans certified for participation in federal and state insurance exchanges include “essential community providers” in order to serve low-income or medically underserved individuals (Section 1311). Although the legislation and associated regulations did not specify race, ethnicity, and language competence as defining characteristics of essential community providers, the law encourages health plans to improve healthcare access for a population of patients who generally access care in medically underserved areas.

Since 2003, America’s Health Insurance Plans Foundation (AHIPF), with funding support from the Robert Wood Johnson Foundation (RWJF), has conducted periodic national surveys of health plans’ collection and use of race, ethnicity, and language data.

 The early surveys in the series focused on collection of data on race, ethnicity, and language preferences of members, but surveys since 2006 have included questions on collection of such data from network providers. In this article, we summarize findings from the 2010 AHIPF-RWJF survey on health plans’ collection of race, ethnicity, and language data from providers and compare responses from 2010 with those  of earlier surveys that also included questions designed to collect race, ethnicity, and language data from network providers and health plan staff.

The sampling frame for the 2010 survey was developed using listings in the Atlantic Information Service’s 

 Leased networks, plans that were no longer in business or had merged, plans with unknown enrollment, and very small plans (enrollment <6500 for commercial, <20,000 for Medicaid, and <5000 for Medicare plans) were excluded, and local/regional subsidiaries of large national companies were collapsed into their respective corporate entities. The final list consisted of 250 plans (both members and nonmembers of America’s Health Insurance Plans [AHIP]), all of which were surveyed. Plans were asked to respond to the survey questions for the product (commercial, Medicare, Medicaid) with the largest enrollment. However, because most health plans had the largest enrollment in a commercial product, 11 multiproduct plans were also asked to respond for 1 additional product, Medicare or Medicaid, so that the distribution of product-specific enrollment in the sample reflected the composition of the industry. Data sampling for previous surveys was described in a previous publication.

Prior to e-mailing the survey to plans, AHIP utilized multiple internal and external contact lists, information from previous survey responders, and company websites to identify the appropriate plan representative to complete the survey. The AHIP staff also sent invitations to companies in the sample to notify the potential respondents about the upcoming survey and ask to confirm their participation.

The surveys were fielded by e-mailing an Excel-based questionnaire, the same approach that had been used in the 2008 survey. (In 2006, the survey was fielded by e-mailing an invitation to a web-based survey.) The AHIP staff sent several reminders to  nonresponding plans. All submissions were checked for accuracy and consistency, and several plans were contacted again and asked to clarify their submissions or provide supplementary information. Of 250 plans, 127 (51%) responded to the 2010 survey, including 59 commercial (48% of eligible plans), 51 Medicaid (59%), and 17 Medicare plans (43%). The responding plans represent a total enrollment of 170 million individuals.

We calculated the proportions of health plans with specific characteristics of interest. The statistical analysis was done by using SPSS 16.0 (SPSS Inc, Chicago, Illinois). We conducted the statistical testing for the difference between binomial proportions by calculating the z scores and the corresponding 

 values. Because our sample included the large share of the population, the finite population correction was included in the significance testing. Because our sample included a large share of the population (>5% of operating health plans in the United States), the finite population correction factor was included in the siginificance testing. This technique takes into account the actual size of the population from which the sample has drawn.

Race/Ethnicity Data of Providers and Staff

Slightly fewer than half of the survey respondents in 2010 reported collecting data on race or ethnicity on any or all of the physicians or other clinicians in the provider network. The proportion of plans collecting these data did not vary markedly by plan type: 41% of commercial plans, 47% of Medicare plans, and 49% of Medicaid plans (45% overall). When weighted by plan membership size, these proportions rose to 53% in Medicaid plans, 74% in Medicare plans, and 58% in commercial plans, or 58% overall. This increase due to weighting indicates that larger plans were more likely to report collecting these data than smaller plans.

Overall, health plans in 2010 reported having race and ethnicity data on 47.3% of their network providers (unweighted data), or 12.8% when weighted by enrollment. The distinct difference in proportions in the unweighted versus weighted analyses suggests that smaller plans were more successful in obtaining race/ethnicity information on providers than were larger plans. Medicaid health plans were more likely to successfully obtain these data from providers than commercial and Medicare plans (37.7% vs 11.7% and 4.2%, weighted data). 

 summarizes the reasons given by plans reporting the collection of race/ethnicity data on providers for doing that data collection. Assessing diversity of provider network and ensuring member access to “minority physicians” were the 2 reasons most frequently cited. (Plans could indicate more than 1 reason for data collection).

Approximately half of the plans that collected this information made it available to enrollees, and the proportion of plans making information available did not vary significantly by plan type—46% for commercial plans, 50% for Medicare plans, and 52% for Medicaid plans (49% overall). Those plans that did make information available to enrollees reported doing it through a health plan provider directory (89% of plans making information available to enrollees), through a health plan website (64%), or through interactions with customer service representatives (60%).

 shows the extent to which plans reported collection of data on language spoken for various types of plan employees or network providers. Because responding plans had the option of reporting on collection of these data for “all health plan staff” and/or specific staff in the framing of the response options, it is difficult to know whether collecting data on some specific classes of staff is truly done less frequently. Some plans outsource some or all of their behavioral care and/or disease care management, but the survey did not include questions on the extent of outsourcing practices.

, responses to a question about testing for language proficiency of various types of plan employees or network providers are presented. In general, plans did not report doing formal testing of language proficiency for network providers and most types of plan staff. Customer service staff was the group most likely to be formally tested for language proficiency.

As was the case for information on race/ethnicity, information on languages spoken by plan staff and network providers was conveyed to members primarily through the plan website, the provider directory, other printed member materials, and interactions with customer service representatives.

The proportion of health plans collecting data on race/ethnicity of providers did not change significantly from 2006 to 2010 (45.4% in 2006, 39.3% in 2008, and 45.2% in 2010).

In spite of the fact that the proportion of plans reporting collection of provider race/ethnicity data was essentially the same in 2006 and 2010, closer examination of the survey data suggested that some plans that had been collecting providers’ race/ethnicity data in 2006 abandoned the practice in subsequent years. There was a statistically significant decline, for example, in the data collection of race/ethnicity of providers for Medicaid plans: 77% collected data in 2006 but only 46% collected data in 2008 (z score = —4.62, 

shows the specific pattern of changes in plans’ collection and sharing of race/ethnicity data on providers among a subset of plans that responded both to the 2006 and 2010 surveys (n = 67). Among plans that had been collecting data in 2006, nearly as many stopped by 2010 (n = 15) as continued (n = 20). Perhaps more striking is the observation that, of the 21 plans (shown in first and second columns of Table 4, second row) that were sharing data on provider’s race/ethnicity with plan members in 2006, many more had stopped doing that by 2010 (n = 15) than continued (n = 6). We were not able to determine through this survey why such efforts were suspended.

For those plans that did share information on provider race/ethnicity with members, information on the media used by plans to share that information were available in the 2008  and 2010 surveys. The pattern of media through which plans made this information available to members was similar in the 2 surveys; the most statistically significant change over the 2 years was in providers’ race/ethnicity being more likely to be available through customer services representatives in 2010 than in 2008 (z score = 4.29, 

Collection of data on languages spoken by providers and staff also showed a modest increase from 2006 to 2010. Increases from 2006 to 2010 were consistent across types of other plan staff. In most subcategories of staff, the percentage of plans reporting collection of language data increased 10%  to 20% across the 4 years. Collection of data on languages spoken for “all health insurance plan staff ” increased by 10% (28% to 38%, z score = 3.00, P 

Finally, the percentage of plans that reported testing or verifying language proficiency increased approximately 10% for most major types of plan employees (for example, the percentage of customer service staff tested was 29% in 2008 and 40% in 2010; z score = 2.84, P 

01). The notable exception was the category of providers compared to the network providers’ office staff: 5% each for providers and providers’ office staff in 2008 and only 3% for providers and 5% for providers’ office staff in 2010.

Nearly half of responding plans in 2010 reported collecting data on race/ethnicity of providers, but this activity was no more frequent in 2010 than it had been in 2006, and a number of plans that had been collecting race/ethnicity data on providers in 2006 did not continue doing that in 2008 or 2010.

The apparent decrease from 2006 to 2008 (albeit not statistically significant) was somewhat surprising, given that the collection of data on race, ethnicity, and language needs of health plan members has increased steadily for the past 10 or more years.

Collection of data about providers may be a more sensitive issue. Although the collection and publication of data on provider characteristics would seem to be an essential step to ensuring a diverse and accessible provider network for a diverse health plan membership, concerns about possible exclusions from provider networks or patients choosing against providers of specific backgrounds have raised significant challenges for health plans. Indeed, we found that 15 plans that were collecting data on providers in 2006 were not doing so in 2010. Collecting and providing these data continue to be optional activities for both health plans and providers. Based on personal conversations with health plans that collect these data (Rita Carreón, director, clinical strategies and healthcare equity, America’s Health Insurance Plans, February 8-9, 2012), plans cited limitations in capturing sufficient data on their network providers among other factors such as trust and providers’ concerns related to the use of the data.  Collection of data on languages spoken was more common than collection of data on race/ethnicity, probably because of existing accreditation or purchaser requirements to be able to provide services for patients who have limited English proficiency or who prefer communication in languages other than English, as well as the expressed need by plan members for interaction in languages other than English.

 From 2006 to 2010, health plans expanded their collection of language data on network providers and health plan staff, were more likely to test or verify language proficiency of staff, and were more likely to share this information with plan members.

The study findings clearly have limitations. Although the response rate was reasonable for a survey of this type, we did not have data for the complete universe of operating health plans. It is possible that the nonresponding plans were less engaged in issues related to race, ethnicity, and language data collection, so the activity reported here might overstate the level of activity for health plans in general. One might speculate that health plans serving areas with little racial, ethnic, or linguistic diversity would be less likely to invest in data collection and incorporation of data into programs and member materials; however, qualitative data from follow-up interviews on data collection issues related to plan members’ race/ethnicity did not show such a relatiopnship.

The sample size of plans did not permit a more detailed analysis of the relationship between service area demographics and activities in the area of race, ethnicity, and language data collection. We also were not able to gather information as to why a subset of health plans stopped the collection and sharing of data on physician’ race and ethnicity between 2006 and 2010.

The National Committee for Quality Assurance’s (NCQA’s) Multicultural Healthcare Distinction Program, which includes aspects of Culturally and Linguistically Appropriate Services, has been in place since 2010.

 Although this program is voluntary and therefore cannot enforce compliance through regular accreditation requirements, the NCQA has clearly sent a signal to plans working with it that collection of race, ethnicity, and language information on plan members, staff, and providers is an important, valued activity. It is not surprising, then, to see more plans reporting activity in this general area in 2010 than in 2008. Specific mention of race/ethnicity and language data collection in the ACA

 and in the second stage of the Health Information Technology for Economic and Clinical Health meaningful use standards for health information technology,

 although not directed explicitly to health plans, has reinforced the message from payers and accrediting bodies that such data collection is expected and essential. Such requirements will presumably encourage physicians and other providers to report this information. In their survey responses about uses of data, health plans indicated that they recognize demographic diversity among their members and are seeking to meet diverse needs. In some cases, the data are used in the design of programs (eg, wellness and disease management) and in the provision of information about benefits and other plan features in languages other than English. In other cases, information about race/ethnicity and languages spoken by providers that is shared with plan members presumably supports choices being made by members about   where to seek care, which will be critical under the new health insurance exchanges that began accepting new enrollment October 1, 2013. The diverse makeup of healthcare providers and clinicians also allows plans to draw on their expertise, via advisory councils, in improving care for racial and ethnic populations and in developing tools and written/verbal messages to consumers. Through this variety of mechanisms, it appears that plans are expanding the ways in which they can provide culturally and linguistically appropriate services to members. 

There is room for continued progress in expanding data collection efforts and increasing the availability of this information to consumers. As our nation’s population continues to become more diverse, the healthcare system will need to meet the increasingly complex cultural and language needs of patients. Although additional research may be needed to further understand the complexities of collecting data on provider networks’ race and ethnicity (eg, the nature of providers’ concerns, attitudes, and beliefs regarding provision of these data), there continues to be general consensus that improving the diversity of the US healthcare workforce is important for health plans and other key stakeholders to help improve access and quality for patients of all racial and ethnic backgrounds.

 With public programs now reinforcing the need for such data on practicing physicians and other clinicians through meaningful use requirements, physicians are seeing a concerted message on the importance of reporting such information. A broader discussion within the healthcare stakeholder community on the experiences in collecting and sharing providers’ race, ethnicity, and language data would be a critical step toward increasing patient-provider racial and language concordance.

 From Henry Ford Health System (DRN), Center for Health Policy and Health Services Research, Detroit, MI; America’s Health Insurance Plans (RC, GV), Washington, DC.

Ms Carreón and Mr Veselovskiy report employment with America’s Health Insurance Plans. Dr Nerenz reports 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 (RC, DN, GV); acquisition of data (RC, GV); analysis and interpretation of data (RC, DN, GV); drafting of the manuscript (RC, DN, GV); critical revision of the manuscript for important intellectual content (RC, DN, GV); statistical analysis (GV); obtaining funding (RC); administrative, technical, or logistic support (RC); and supervision (RC, DN).

Robert Wood Johnson Foundation to the America’s Health Insurance Plans Foundation (grant #62295).

Rita Carreón, BS, Director, Clinical Strategies and Healthcare Equity, America’s Health Insurance Plans, 601 Pennsylvania Ave, NW, Ste 500, South Bldg, Washington, DC 20004. E-mail: rcarreon@ahip.org.1. Ulmer C, McFadden B, Nerenz DR, eds. Race, Ethnicity, and Language Data: Standardization for Health Care Quality Improvement.Washington, DC: National Academies Press; 2009.

2. Brach C, Fraser I, Paez K. Crossing the language chasm. Health Aff (Millwood). 2005;24(2):424-434.

3. Tang G, Lanza O, Rodriguez FM, Chang A. The Kaiser Permanente clinician cultural and linguistic assessment initiative: research and development in patient-provider language concordance. Am J Public Health. 2011;101(2):205-208.

4. Cooper LA, Roter DL, Johnson RL, Ford DE, Steinwachs DM, Powe NR. Patient-centered communication, ratings of care, and concordance of patient and physician race. Ann Intern Med. 2003;139(11):907-915.

5. Meghani SH, Brooks JM, Gipson-Jones T, et al. Patient-provider race-concordance: does it matter in improving minority patients’ health outcomes? Ethnicity Health. 2009;14(1):107-130.

WEB EXCLUSIVE

Health plans collect and use data on network providers' proficiency in languages other than English more commonly than race and ethnicity data on providers.

Collection of Data on Race/Ethnicity and Language Proficiency of Providers

To assess the association between Dietary Diversity Scores (DDSs, range: 0-6 points) and medical service utilization and expenditures.

The Elderly Nutrition and Health Survey in Taiwan (1999-2000) provided a 24-hour dietary recall for DDSs. National Health Insurance claims were linked for 1650 eligible elders. Generalized linear models were used to appraise the association between DDS and annual medical utilization and expenditures.

Those with a higher DDS had lower medical service utilization and expenditures for emergencies and hospitalization. After adjustment for potential confounders, emergency and hospitalization expenditures for elders with a DDS of 6 were lower than those with a DDS of 3 or lower. However, for preventive care and dental services, the highest DDS of 6 predicted greater utilization (0.25 and 0.5 times) and expenditure (270 and 420 Taiwanese new dollars). Findings remained unchanged when those who died in the first year or had any medical utilizations and expenditures1 year prior to death were excluded.

Greater dietary diversity is associated with lower emergency and hospitalization utilization and expenditures, but not lower use of ambulatory services. There is a need for health services to develop a nutrition policy for nutritionally disadvantaged groups.

Healthy eating may minimize healthcare costs.

		The National Nutrition and Health Surveys linked to the National Health Insurance system database in Taiwan provide an opportunity to address this issue.

		We found that those whose diets were more favorable to health, as judged by the Dietary Diversity Score (DDS), had greater expenditures on preventive services and dental care, but lower medical service utilization and expenditures for emergencies and hospitalization.

		Because an increased DDS costs more in Taiwan, health and nutrition policy needs to focus on the socioeconomically disadvantaged if cost containment is to be fully realized.

In the formulation of health policy, economic factors have an inescapable role, but the medical costs of diet-related and nutritionally related diseases are rarely given attention. Yet the opportunities to reduce the burden and costs of this disease subset are considerable.

Greater dietary diversity is associated with lower emergency and hospitalization utilization and expenditures, and identifies a policy direction for nutritionally disadvantaged groups.

December 2013

Economics of Genomic Testing for Women With Breast Cancer

Impact of Electronic Prescribing on Medication Use in Ambulatory Care

Collection of Data on Race/Ethnicity and Language Proficiency of Providers

Dietary Diversity Predicts Type of Medical Expenditure in Elders

Cost Utility of Hub-and-Spoke Telestroke Networks From Societal Perspective

Generic Initiation and Antidepressant Therapy Adherence Under Medicare Part D

Characteristics of Low-Severity Emergency Department Use Among CHIP Enrollees

Medication Utilization and Adherence in a Health Savings Account-Eligible Plan

Implementing Effective Care Management in the Patient-Centered Medical Home