 There is growing evidence that practice-based care management can improve clinical quality and reduce costly healthcare utilization.

To explore how a disparate group of patient-centered medical homes (PCMHs) embedded care management in their team care environment to identify best practices.

 A positive deviance approach was used to contrast care management implementation in practices having the greatest and least improvement on clinical measures of diabetes, the initial target disease for a multipayer-supported statewide initiative involving 25 National Committee on Quality Assurance—recognized PCMH practices participating in a regional learning collaborative.

 Practices were ranked according to their average absolute percentage point increase from baseline to 18 months on 3 diabetes quality measures. Semistructured interviews were conducted with 136 individuals in 21 of the 25 practices. Interview data were analyzed using grounded theory with NVivo 9.0 software. To develop hypotheses related to care management best practices, we compared and contrasted emerging themes across clinical performance tertiles.

 Practices with the greatest diabetes improvement described (1) more patient-centered care manager duties, (2) better use of the  electronic medical record (EMR) for messaging and patient tracking, and (3) stronger integration of the care manager into the care team compared with practices with the least diabetes improvement.

 PCMHs may want to ensure that care managers are available to meet with patients during visits, support patient self-management, fully leverage the EMR for team messaging and patient tracking, and ensure integration into the care team with office huddles and ongoing communication.

This study used a positive deviance approach to explore how 25 patient-centered medical homes (PCMHs) embedded care management within the PCMH team.

		Practices with the greatest reported improvement on diabetes quality measures described more patient-centered care manager duties, more effective electronic capabilities, and stronger integration of the care manager into the team than practices with the least improvement on those measures.

		Results suggest PCMHs should ensure care managers are available to meet with patients during visits, fully leverage the electronic medical record for messaging and patient tracking, and maintain ongoing communication with providers and other team members.

The patient-centered medical home (PCMH) is a new and increasingly widespread1 model of healthcare delivery that shows significant promise for improving patient care. The PCMH emphasizes team-based care, coordinated and integrated care, and whole-person care,

and has been associated with improved measures of quality care and cost reduction.

 The process of becoming a PCMH involves practice transformation often centered on the development of a care management plan and infrastructure.

Care management involves more intensely caring for high-risk patients through the establishment and monitoring of care plans, more frequent follow-up visits, regular outreach between office visits to assess health status, extensive support for disease management  and self-care, tracking and coordination of specialty and other services, and linkages with community resources.

Care management has traditionally been conducted by insurer-based nurses providing telephonic outreach to patients identified as  either high cost or high risk by claims-based predictive modeling software. However, this method has provided inconsistent care improvement results.

Some successful models of care management include community-based care managers,

 health plan care managers embedded in primary care practices,

 and health system—based nurse teams working with primary care practices.

This study examined the development of care management within 25 heterogeneous primary care practices in southeastern  Pennsylvania implementing the PCMH focused initially on improving diabetes care. Diabetes is a common chronic disease used as a starting point for many PCMH initiatives.

A recent review described team-based care and care management as critical components in improving the care of patients with chronic conditions such as diabetes.

 With care management and team-based care both representing key elements of the PCMH

 and growing evidence that practice-based care management is highly effective in improving clinical quality and reducing costly healthcare utilization,

 it is important to better understand the implementation of care management in primary care practices. Although care management is an important addition to primary care, there is tremendous variation in the definition and implementation of the role at the practice level, making the implementation of care management an important research topic. This is one of the first studies to explore how a disparate group of unaffiliated primary  care practices embedded care management within the team care environment of a PCMH. We used a positive deviance approach   contrasting care management implementation in higher- and lower-performing practices to identify a collection of potential best practices synthesized from individual higherperforming practices.

The practices studied were part of the first regional rollout of a statewide, multipayer PCMH initiative consisting of regional learning collaborative meetings, practice facilitation support, and monthly clinical data and narrative reports describing PCMH and care management implementation. All 25 practices were recognized PCMHs by the National Committee on Quality Assurance (NCQA), and 6 regional payers provided pro rata payments to the practices to support PCMH and care management implementation. Practices were expected to take an all-payer approach to population management, including planned chronic and preventive care for all patients and, specific to this study, care management for the highest risk patients. Using a positive deviance method, performed by calculating high and low performance on standard measures of diabetes management and developing hypotheses related to the description of top-performing practices,

 we analyzed and characterized care management implementation in the PCMH setting. We aimed to identify best practices for primary care sites seeking to develop embedded care management services.

This mixed-methods study involved (1) rank-ordering the sites based on practice-reported diabetes data to determine the highest and lowest performing practices and (2) analyzing qualitative data collected from interviews to contrast care management implementation in high- and low-performing practices.

The highest and lowest performing practices were identified using practice-reported diabetes data, the initial clinical focus of the statewide initiative. The 25 practices participating in the collaborative were ranked according to their improvement from baseline to 18 months across 3 diabetes performance measures most closely associated with minimizing morbidity and mortality: the percentage of diabetes patients (1) whose latest glycated hemoglobin (A1C) result was less than 7%, (2) whose blood pressure was less than 130/80 mm Hg, and (3) whose low-density lipoprotein cholesterol was less than 100 mg/dL. The resulting improvement index was calculated as the arithmetic mean of the absolute percentage improvement in the 3 clinical diabetes measures. Practices were divided into performance tertiles based on their calculated improvement index. The improvement indexes were statistically significantly different between performance tertiles (1-way analysis of variance 

Semistructured interviews were conducted with 136 individuals, including clinicians (n = 56), practice managers (n = 15), care managers (n = 13), and other staff (n = 52), in 21 of the 25 practices. Interviews were framed by interview guides with extra questions related to finances for practice leaders and office administrators. Interviews were conducted by 2 teams of 2 trained researchers, with 1 person asking questions and the other taking notes. Both teams followed the same semistructured interview guide and recorded notes from the interviews that were used to assess and ensure inter-observer consistency within and across the interviewer teams. In addition, members of the 2 interviewer teams each observed 1 of the other team’s on-site interview sessions to identify and address any differences in interviewer style or delivery of the questions. Members of both interviewer teams also participated in weekly team meetings to review discrepancies and reach consensus. Most interviews were conducted on-site, during office hours, in private locations. Additional interviews were conducted through focus groups or phone calls if key personnel were not available in person. Participants were not compensated for their interview time.

Participants were asked to describe their  understanding of the PCMH and their experiences with implementing the PCMH, including their role, level of adoption across the practice, key leadership, accountability, surprises, and lessons learned. Interviews ranged from 15 to 120 minutes and were audio-recorded and professionally transcribed.

Transcripts were entered into NVivo version 9, a software package for qualitative analysis.

 Using grounded-theory methods, individual interviews were analyzed for themes and patterns.

 Grounded theory is a methodology that involves iterative development of theories about what is occurring in the data as they are collected.

 The process develops themes that emerge  “from the ground” based on responses to the open-ended questions developed for this study.

 Broad codes reflecting stakeholder  responses to questions about the PCMH were created by a multidisciplinary team of investigators from primary care and  communication that coded and analyzed the transcripts. Discrepancies in coding were resolved by group consensus. Keyword  searches were run of all interviews identifying use of the terms “care manager,” “care management,” and “high risk.” Topics  developed from care manager interviews were used to create further nodes and searches identifying care management mentioned in  relation to payment incentives, team-based care, hiring and firing, technology, and roles and responsibilities. In order to develop hypotheses around best practices related to care management, emergent themes were compared between higher- and  lowerperforming tertile practices. Emerging themes clustered in 3 clear categories that correlated with tertiles of practice performance measures.

Practices included in the study included 4 solo/partner, 8 small, 10 medium, and 3 large practices (

). These included private practices, residencies, systems, and federally qualified health centers. Within the first year, 8 practices had level 3 PCMH   recognition from the NCQA, 3 practices had level 2, and 14 practices had level 1. No direct correlation was seen between diabetes   care improvement measured by the improvement index and any practice demographics, including NCQA recognition levels. The   study was not powered to detect differences between diabetes improvement and the educational background of care managers,   including registered nurses, social workers, and medical assistants.

We determined the mean baseline for all 3 practice tertiles in regard to the percentage of patients achieving A1C levels less than 7%, blood pressure less than 130 mm Hg, and low-density lipoprotein cholesterol less than 100 mg/dL. The most improved practices reported 41.8%, 40.8%, and 36.4%, respectively. The middle-performing practices reported 38.9%, 40.7%, and 38.0%, respectively. The least improved practices reported 44.4%, 42.9%, and 44.8%, respectively.

Comparing interviews of care managers in the upper and lower tertiles, 3 topics were most salient for implementing successful care management. Practices with the greatest improvement indices described (1) patient-centered care manager duties, (2) using the electronic medical record (EMR) system to its fullest patient-tracking capabilities, and (3) stronger integration of the care manager in the care team as evidenced by extensive information sharing. In contrast, the practices with the lowest improvement indices described more administrative care manager duties, little EMR use, and minimal integration of care management and information sharing.

Care managers in the upper-tertile practices described performing duties best characterized as patient centered, including focusing specifically on diabetic and high-risk patients, following up with patients after visits, providing self-management health coaching and patient education, and providing care for their own roster of patients (

). A care manager from a practice in the upper tertile described her duties:

		I do all the phone calls with the patients. I remind [them of] their appointments … all the high-risk patients … I follow the dialysis roster. I look for the missing things, like they need labs, they need appointments, and I bring them back.

Care managers in the lower-tertile practices did not describe having a roster of high-risk patients and described doing mostly telephonic follow-up with patients, managing other staff, and delegating patient follow-up to others, suggesting they serve in more of an administrative role instead  of providing direct patient care. A lower-tertile care manager described her position as:

		… Overseeing the nurse practitioners, and overseeing the medical assistants, to make sure that everything flows, and that the clinical side of everything goes smoothly.

Care managers at upper-tertile practices reported using the EMR for tracking patients in a diabetes registry and coordinating their care, developing smart forms, and documenting care management services in the EMR (

). An uppertertile care manager praised the EMR:

		With this EMR, it’s so much easier to track patients that have been missed out of the system.... So I just think the EMR makes it a little easier because, once they’re in there, you can’t take them out of there until you do something with their appointment.

	Care managers at lower-tertile practices reported not using the EMR for patient tracking, not owning an EMR, or not having time to use the EMR during patient visits. They also reported not being able to develop a way to use the EMR for tracking patients via a registry or inputting forms for specific follow-up. A lower-tertile care manager expressed frustration with the EMR:

		The small health maintenance came embedded in the EMR, and we didn’t like it … we’re not using it. It’s not really helpful. It’s not doing anything for us.

	Overall, care managers in the upper-tertile practices benefited from earlier work in their practices to develop registry capabilities within their EMRs, including patient tracking systems, use of structured data fields, and standardized documentation processes.

	Many care managers in upper-tertile practices described meeting with providers to discuss patients on a regular basis, holding quick huddles with providers and other staff nearly every day, and having ongoing discussions about patient care in the EMR (

). Less structured communication in the form of ad hoc huddles was also described as useful among upper-tertile care managers  to allow for consultations on new or difficult cases while the patient was still in the office. A care manager in an upper-tertile practice described the proactive nature of their regular meetings:

		We do the weekly touch base, and I do daily huddles with the MAs in the back, and with the report cards and everything, and then they have operation meetings which is on a weekly basis.

	Some of the care managers in lower-tertile practices reported far less regular or frequent meetings that focused more on work flow issues and less on patient care issues. Few mentioned  huddles or their EMR as effective methods for discussing patient care.

		We need to set aside that time and say, “OK, this person is coming in ... this is what we need to do. That person is coming in … that’s what we need to do.” We haven’t really done that. I hate to say that we are flying by the seat of our pants, but that is kind of the feeling that I’m getting.

Articles

CLINICAL

A positive deviance approach was used to identify best practices in embedding care management in patient-centered medical home team-based care processes.

Implementing Effective Care Management in the Patient-Centered Medical Home

To evaluate the impact of a consumerdirected health plan with a health savings account (CDHP-HSA) on utilization of and adherence to medications among individuals with chronic disease.

 Pre-post comparison study with matched control group (difference-in-differences analysis).

 Data on workers and dependents with 1 or more of 5 chronic conditions—hypertension, dyslipidemia, diabetes, asthma/chronic obstructive pulmonary disease (COPD), and depression&mdash;were obtained from an employer that fully replaced its  preferred provider organizations (PPOs) with a CDHP-HSA in 2007. A control group of participants from an employer that maintained its PPO throughout the 3-year study period (2006-2008) was created by matching on preperiod (2006) individual characteristics. Difference-in-differences estimates of the impact of the CDHP-HSA were derived by chronic condition for number of prescriptions,  proportion of days covered (PDC), and an indicator for a PDC of 0.80 or higher.

 During the first year after implementation, enrollees with hypertension, dyslipidemia, and diabetes had significantly less medication utilization (by 1-2 prescriptions) and lower adherence rates (by 0.05-0.09 in PDC; 0.04-0.13 in the proportion adherent). These reductions abated, yet remained, after 2 years among hypertension and dyslipidemia patients. The PDC was significantly lower in patients with depression by 0.07 and 0.05 after 1 and 2 years under the new plan, respectively. No statistically significant impacts were detected on enrollees with asthma/COPD.

A CDHP-HSA full replacement was associated with reduced adherence for 4 of 5 conditions. If this reduced adherence is sustained, it could adversely impact productivity and medical costs.

We examined the impact of plan design on medication utilization and adherence among individuals with chronic disease employed by a company that adopted a consumer-directed health plan (CDHP) with a health savings account (HSA) for all workers.

		During the first year after CDHP-HSA implementation, enrollees with hypertension, dyslipidemia, and diabetes had significantly less medication utilization and lower adherence rates. These reductions abated, yet remained, after 2 years among hypertension and dyslipidemia patients.

		Adherence was significantly lower in patients with depression after 2 years.

		There were no statistically significant impacts on enrollees with asthma/chronic obstructive pulmonary disease.

Medication is important in the management of noncommunicable chronic diseases, which affect nearly one-half of adults and cause approximately 70% of deaths in the United States.

 Prescription drugs accounted for 12% of healthcare spending in 2012, more than double the level of 30 years ago (5%).

 In general, this shift toward greater use of pharmacotherapy has provided net societal benefits.

 For example, medical cost offsets in Medicare A and B have been documented as a result of adding drug coverage under Medicare Part D.

Prior work has found that medication adherence produced substantial savings as a result of reductions in hospitalization and emergency department use, and it is thus a matter of great importance to policy makers, insurance plan sponsors, physicians, and patients.

Despite clinical and economic benefits, only about half of patients take medications for their chronic conditions as recommended by their physicians.

 Moreover, as much as one-third of initial prescriptions go unfilled.

 For example, studies have found that more than 25% of patients with coronary artery disease discontinued drug therapy within 6 months of initiation,

 and adherence among patients receiving statins fell from nearly 80% within the first 3 months of treatment to only 25% after 5 years.

 Overall, adherence rates across a number of therapeutic classes have been reported at between 28% and 66% after 6 months, and 18% to 54% after 1 year.

 Nonadherence has been estimated to cost the US healthcare system between $100 billion and $289 billion,

and has spawned new plan designs such as value-based insurance design to address this challenge.

 Public efforts to raise awareness of  the adverse effects of nonadherence have also been initiated, such as the “Script Your Future” campaign of the National Consumers League.

Medication adherence is known to be affected by out-of-pocket cost to patients.

 For this reason, it is important to understand medication adherence in populations enrolled in relatively new types of health plans that combine potentially high out-of-pocket costs as a result of high deductibles with tax-preferred savings accounts (consumer-directed health plans; CDHPs). First introduced in 2001 with health reimbursement arrangements (HRAs), savings account—based high-deductible plans were extended by the Medicare Modernization Act of 2003, which authorized high-deductible health plans with health savings accounts (HSAs). Both types of CDHPs have grown steadily over the past decade such that by 2011, 23% of employers had offered either an HRA- or HSA-eligible CDHP—covering about 21 million individuals or about 12% of the privately insured market.

 About 13.5 million individuals were in a CDHP with an HSA account by January 2012.

 Importantly, as of 2012, 8% of large employers had completely replaced their healthcare coverage with only a CDHP.

Much of the existing literature on the impact of CDHPs on use of health services and costs focuses on HRAs. These plans have been in existence longer, and data on HSAs are not as readily available. Moreover, several studies were limited to examining the impact of CDHP on healthcare spending.

 Peer-reviewed articles about the impact on prescription drug use of adopting a CDHP have largely concentrated on the number of prescription drug fills, generic and brand use, and mail order use by CDHP enrollees compared  ith non-CDHP enrollees.

 Sometimes, populations with specific diseases were examined.

We identified only 2 studies that report on the impact of a CDHP on medication adherence. An early study examining 1 year of data after the adoption of an HRA found that 7% of those enrolled in the highest deductible CDHP and taking medication to treat hypertension in late 2003 were no longer persistent with therapy in 2004, though adherence was unchanged among individuals who continued to take medications after moving to the CDHP.

 A more recent study using data from 2005 and 2006 found that after enrolling in a CDHP (both HRAs and HSAs were examined), individuals were less likely to refill prescriptions for cardiac conditions and elevated cholesterol. The CDHP members with asthma, cardiac conditions, and high cholesterol also had reduced medication adherence and persistence with medications.

Given that CDHPs alter out-of-pocket cost for prescription drugs by subjecting them to the high deductible, one might expect this plan design to impact medication adherence. The magnitude and duration of the effect are unclear, however, because an individual with a CDHP plan may have money in an HRA or HSA, and individuals’ use of these funds will influence out-of-pocket costs for prescription drugs over  ime. The account type, employer contribution level, account ownership, and rollover provisions will further complicate the issue.

This study evaluates the impact of adopting an HSA-eligible CDHP (CDHP-HSA) on medication adherence for individuals with chronic disease. Data come from a large manufacturer that replaced all of its existing health insurance options with a CDHP-HSA. Pre-post changes in medication adherence were derived from pharmacy claims and were compared with changes in adherence in a matched control group of a second employer that did not alter its healthcare coverage.

On January 1, 2007, a large Midwestern manufacturer fully replaced its existing preferred provider organization (PPO) health insurance plans with CDHP plans with an HSA. All active employees and their dependents were transitioned to the new plan and were given a choice between 2 annual deductible levels: $1250 individual/$2150 family or $2150 individual/$4300 family. The employer contributed the same amount to the HSA regardless of deductible level, though contributions were higher for those with family coverage.

To evaluate the impact of this plan design change on medication utilization and adherence, integrated pharmacy and medical administrative claims data, as well as enrollment information, were used. A control group was constructed, as described below, using data on another larger employer’s workers, who were consistently covered under traditional PPO plans. To be included in the analysis, subjects were required to be continuously eligible for benefits during the 3-year study period (January 1, 2006, through December 31, 2008) and aged at least 18 years as of January 1, 2006, but less than 65 years as of December 31, 2008. Furthermore, individuals must have had 1 or more of 5 chronic conditions: hypertension, dyslipidemia, diabetes, asthma/chronic obstructive pulmonary disease (COPD), and depression. These conditions were chosen because they are highly prevalent, costly, and routinely managed with pharmacotherapy.

 Patients were defined as having a condition if they had at least 1 inpatient or 2 outpatient claims with an International Classification of Diseases, Clinical Modification Ninth Revision diagnosis code for the illness and at least 1 prescription drug fill indicated for the condition in 2006. 

 shows the classifications utilized. These inclusion criteria were met by 1023 CDHP-HSA enrollees with hypertension, 1184 with dyslipidemia, 314 with diabetes, 169 with asthma/COPD, and 430 with depression. 

A control group was created using data from another larger national employer. We attempted to pair each member of the CDHP group with 1 individual from the comparison pool using coarsened exact matching, a technique that reduces dimensionality through the binning of variables used in the  match process.

 Specifically, for each chronic condition, subjects were exactly matched on baseline (2006) values of the following variables with specified cut points (in parentheses): sex; age (25, 35, 45, and 55 years); geographic region; Charlson Comorbidity Index score (0, 1, 2, 3)

; proportion of days covered (PDC; 0.2, 0.4, 0.6, 0.8); and an indicator for whether or not the first prescription occurred after April 1, 2006 (ie, a proxy measure for a new user because the patient apparently did not have medication on hand at the beginning of 2006). Using these parameters, matches were obtained for a majority (68%-92%) of study subjects. Final sample sizes for the CDHP and control groups were 937 with hypertension, 1057 with dyslipidemia, 226 with diabetes, 115 with asthma/COPD, and 347 with depression.

Three annual measures of medication utilization and adherence were used as dependent variables: (1) the number of 30-day adjusted prescriptions filled for the condition; (2) the PDC for the condition, which represents the fraction of days in the period that the patient had at least 1 drug for the condition on hand; and 3) a flag indicating a PDC of 0.80 or higher, a commonly used threshold for adherence.5 Proportion of days covered is now used, for example, by the Centers for Medicare & Medicaid Services as a quality measure component of the Star Ratings calculation for stand-alone prescription drug plans, as well as Medicare Advantage Plans.

In addition to univariate and bivariate analyses, multivariate models were estimated using a difference-in-differences design. Specifically, for each of the 3 dependent measures, a model was specified, which included the following as independent variables: age, sex, region, Charlson Comorbidity Index score, year dummies, a CDHP group indicator, and a flag for whether or not the individual was the policy holder. Additionally, 2 interaction terms for CDHP group in 2007 and CDHP group in 2008 were entered into the equations to identify the CDHP effects in the post implementation period. Given its count properties, number of prescriptions was specified using a negative binomial model. The PDC model was estimated using ordinary least squares, and adherent was cast as a probit. In all models, standard errors were clustered by individual. Finally, marginal effect estimates of the CDHP impact in 2007 and 2008 were derived, taken at the mean of the regressors, using the delta method. All analyses were conducted using Stata release MP 12.1.35

Characteristics of the CDHP and control groups for each of the 5 chronic diseases are shown in 

. The average age was 52 years for individuals with hypertension, dyslipidemia, and diabetes, and 43 to 44 years for individuals with asthma/COPD and individuals with depression. Those with hypertension, dyslipidemia, and diabetes were more likely to be male and the policy holder compared with members with asthma/COPD and depression, who were more likely to be female and a dependent. Members were overwhelmingly from the Midwest. As reflected in the Charlson Comorbidity Index score, diabetic patients were the least healthy, followed by asthma/COPD patients.

Baseline adherence as measured by the average PDC ranged from 0.59 for asthma/COPD to 0.91 for hypertension. Using the commonly used threshold of 0.80, the percentage of patients who were adherent prior to CDHP adoption was 88% for hypertension, 75% for dyslipidemia, 84% for diabetes, 39% for asthma/COPD, and 70% for depression. As expected from the coarsened exact matching process, no statistically significant differences were observed in baseline mean values between the CDHP and control groups.

Prescription Drug Utilization and Adherence by Year/Group

 presents mean levels of the number of prescriptions, PDC, and proportion adherent (PDC 

0.80) for each of the 5 chronic diseases by year and group. By design of the matching process, baseline (2006) values were not statistically significant across the CDHP and control groups except for number of depression prescriptions—a point we will return to. With most of the chronic diseases, the 3 outcome measures decreased from 2006 to 2007 for both groups, although the 2007 levels were significantly (

 <.05) lower in the CDHP group for all conditions except asthma/COPD. For example, the percentage of hypertension patients who were adherent fell from 88% to 73% for CDHP enrollees compared with a drop from 88% to 78% among control group members, representing a statistically significant (

 <.05) difference of 5 percentage points. By 2008, the CDHP hypertension and depression groups had lower drug utilization and PDC than their control counterparts. Moreover, 2008 levels of all 3 measures for dyslipidemia

 presents estimates of the impact of the CDHP-HSA on prescription drug utilization and adherence in 2007 and 2008. After adjusting for the characteristics in Table 2 using the difference-in-differences multivariate models described above, the differential changes in the mean values of the measures observed in the bivariate analysis (Table 3) were confirmed. Specifically, adoption of the CDHP-HSA was associated with decreases in the number of prescriptions filled, the PDC, and the percentage of individuals who were adherent to their medications in the first year for hypertension, dyslipidemia, and diabetes. The magnitudes of the reductions for these 3 chronic diseases in 2007 ranged from —1.2 to –2.4 prescriptions, –0.05 to –0.09 PDC, and –0.04 to –0.13 in the proportion of patients who were adherent. No significant effects were detected in the asthma/COPD analyses. In patients with depression, only the PDC significantly (

 <.05) declined, which may suggest a greater rate of first- and secondfill drop-offs (ie, the lower end of the PDC distribution) because the proportion adherent (ie, the higher end of the PDC distribution) was not significantly affected.

The reductions observed in the first year after adoption of the CDHP-HSA appear to have diminished in its second year (2008). Dyslipidemia and hypertension were the only chronic conditions in which changes after CDHP adoption persisted for 2 or more measures of drug utilization and adherence for 2 years. The PDC among CDHP enrollees with depression also remained significantly lower after 2 years. We explicitly tested whether the changes in the measures from 2007 to 2008 were statistically significant, and year-over-year improvements did emerge after CDHP-HSA implementation (Table 4, righthand column). Specifically, the number of prescriptions filled for diabetes and hypertension significantly increased. More over, the proportion adherent among individuals with dyslipidemia and diabetes was 0.03 and 0.07 higher, respectively, in 2008 than in 2007, and the PDC for hypertension patients was also slightly higher.

This study evaluated the impact of a full-replacement CDHP-HSA on utilization of and adherence to medications for 5 chronic conditions. In the first year under the new plan, the number of prescriptions filled, the PDC, and the proportion of patients who were adherent declined for all conditions except asthma/COPD, although for depression, only the drop in PDC was statistically significant. These results are consistent with existing literature indicating that increased patient cost-sharing is associated with decreased health services utilization, 36 specifically prescription drug consumption.

 Except for asthma/COPD, these effects emerged across our fairly heterogeneous set of conditions, a finding also consistent with prior work.

During the second year of the new plan, CDHP-HSA effects persisted in individuals with hypertension, dyslipidemia, and depression, but levels were not as low due to some significant increases in utilization and adherence from year 1 to year 2, particularly among patients with diabetes. It is possible that second-year improvements were due to members learning about the importance of medication adherence and the  parameters of their new CDHP. Alternatively, beneficiaries may have had residual funds in their HSAs—rolled over from the first year&mdash;that they used to purchase more prescription drugs.

These findings have important policy implications. Notwithstanding the presence of HSAs and employer contributions, medication utilization and adherence declined when high deductibles were imposed. If these reduced levels of medication adherence for chronic conditions are sustained, it is likely that they will increase medical costs and adversely impact worker productivity.

 This cost-shift impact on adherence may be mitigated by designing CDHPs with first-dollar coverage for chronic disease medications using HRAs. Regulatory change is required for first-dollar coverage for HSA-associated CDHPs, as exemptions to the deductible are governed by law. Finally, it appears prudent for plan sponsors to provide education and ongoing support to encourage appropriate use of account funds so that prescription drug use for chronic conditions is prioritized by members, even when budgets are stressed.

Our study has several strengths, the first of which is the use of a matched comparison group of employees of another company that did not alter health benefits. This group allowed for statistical control of some common medication utilization and adherence confounders such as demographics, general health status, and underlying secular trends in prescription drug supply and demand. Moreover, because the study employer fully replaced existing coverage for all employees, our results are robust to selection bias, which often plagues analysis of enrollees in CDHPs introduced as alternatives to other plans. Our study also extends the CDHP literature by reporting on experience under an HSA, as opposed to the more commonly studied HRA-linked CDHP. Finally, the examination of 2 years of data after CDHP-HSA adoption is key because we found some increases and differential effects in the second year.

Our study also has several limitations. Because our study cohort was a single manufacturing employer concentrated in the Midwest, results may not be generalizable to the broader population of national employment-based health plans implementing CDHP-HSAs. Furthermore, although a control group was utilized in the analysis, estimated impacts might still be biased if unobserved characteristics correlated with CDHP were also associated with medication utilization and adherence. For example, if despite being paired with individuals similar with respect to demographics, disease burden, and baseline utilization/adherence rates, socioeconomic status was significantly different for CDHP members compared with control group participants, then our estimates could remain biased, given this factor’s influence on health-related behavior. Participantlevel income data were not available for analysis, nor was the pool of members used in the matching process large enough to include geo-linked sociodemographic variables. 

Similarly, although we matched on Charlson Comorbidity Index score, it is still possible that CDHP and control group members had different comorbidities at baseline, including but not limited to the diseases under study. If true, results might be biased if patients chose to decrease prescription drug utilization for some conditions (eg, asymptomatic ones) over others. We conducted an exploratory re-analysis of the multivariate models on subsamples of individuals with more than 1 of the 5 chronic diseases. While results were qualitatively similar, differences in effect sizes suggest that individuals may make trade-offs among their concomitant therapies when faced with plan design changes that alter out-of-pocket costs. On a related note, depression has been identified as a risk factor for medication nonadherence in patients with other comorbidities.

 Compared with the CDHP cohort, the depression control group had 12% more 30-day adjusted prescriptions at baseline, yet was nearly identical in terms of PDC. This indicates that the 2 groups differed slightly in terms of either their concomitant use or switching of antidepressants, which may suggest a difference in depression severity. Future research should delve more deeply into how patients with greater disease burden respond to increased cost sharing.

Another limitation is that we interpreted the medication utilization and adherence effects as being due to the newly adopted CDHP-HSA itself, or more specifically, the high deductible. It is possible that the plan’s newness itself caused these disruptions. Prior research has found that preventive services not subject to the deductible were associated with lower use of those services

However, a portion of our control group also likely changed plans during the study period. Therefore, plan change would not be a plausible explanation for persistent second-year impacts. Finally, we did not possess data on HSA balances and could not address key questions related to account dynamics such as contributions, withdrawals, and rollovers, and their association if any with medication utilization by account holders and their enrolled dependents.

Broader implementation of CDHPs is likely as healthcare costs continue their upward climb.

 To ensure that these plans deliver on their promise of wiser consumption of health services while supporting chronic care needs including medication, better insight into longer-term behaviors under CDHPs is needed. Studying lengthier time periods before and after CDHP adoption, and unraveling the influences of socioeconomic and multimorbidity profiles, are good candidates for future research. Collection of biometric data—including side effect profiles&mdash;would be highly valuable to allow researchers to tease apart the causes of medication nonadherence.

 From Employee Benefit Research Institute (PF), Washington, DC; IBM Corporation (M-JS), Southbury, CT; RxEconomics LLC (MCR), Hunt Valley, MD.

WEB EXCLUSIVE

A consumer-directed health plan with a health savings account was associated with reduced adherence for 4 of 5 conditions.

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

 To describe patient characteristics among those utilizing the emergency department (ED) for low-severity conditions (ie, conditions potentially treatable or manageable in a primary care setting).

A pooled cross-sectional study of administrative claims for ED visits among enrollees in Alabama’s Children’s Health Insurance Program (CHIP), ALL Kids, from January 1, 1999, through December 31, 2010.

Severity of visit was categorized based on primary diagnosis code using an established claims-based algorithm. Logistic regression was used to identify patient characteristics that predicted low-severity ED visits relative to high-severity visits.

Of a total of 141,709 qualifying ED visits, 97,961 (69%) were classified as low severity, 33,941 (24%) as intermediate severity, and 9807 (7%) as high severity. Based on absolute risk differences, we found that among children utilizing the ED, low-severity visits were more likely than high-severity visits among children who were noncompliant with recommended well-child care (1.2 percentage points, 95% confidence interval [CI], 0.4-1.9); children who were nonurban residents (urban vs isolated: 1.6 percentage points, 95% CI, 1.0-2.2; urban vs small rural: 1.1 percentage points, 95% CI, 0.5-1.7); children without chronic disease (10.3 percentage points, 95% CI, 9.9-10.7) and children whose ED visits were on Sunday versus weekdays (0.9 percentage point, 95% CI, 0.6-1.3), and on Saturday versus weekdays (1.2 percentage points; 95% CI, 0.8-1.6).

 Our results suggest that improving access to primary care on weekends and in rural areas are potential ways to improve the efficient use of ED services.

We investigated whether emergency department (ED) use for low-severity conditions among low-income children was associated with modifiable characteristics using claims data from a Children’s Health Insurance Program that is similar to private insurance. We found that among children having a visit to the ED:

		Older age, male sex, African American race, presence of chronic disease, and compliance with recommended well-child care were associated with decreased likelihood of low-severity visits (relative to high-severity visits).

		Living in an isolated rural town and visits on weekend days were associated with increased likelihood of low-severity visits (relative to high-severity visits).

From 1997 through 2007, emergency department (ED) visits in the United States increased by 23%, or 11% per capita.

Many of these ED visits were for low-severity conditions and potentially could have been avoided with the provision of effective primary care.

These visits increase ED overcrowding, wait times, and financial strain on public health insurance programs.

Use of ED services for low-severity conditions is often interpreted as a marker of a failing primary care delivery system. The reliance  on ED care may result from limited access to other, less intensive care settings,

 either because of geography, insurance coverage, or other psychosocial factors.

 Low-severity ED visits often result from the inability of patients to obtain swift “reassurance” or advice in the primary care setting.

Reducing barriers to primary care could reduce low-severity ED utilization and downstream costs through improved prevention and management. For instance, children with greater continuity of primary care have fewer hospitalizations and ED visits.

 Furthermore, access to high-quality primary care, as defined by Consumer assessment of Healthcare Providers and Systems surveys, is associated with lower ED utilization for low-severity conditions among Medicaid children.

There is a sizable literature examining the individual characteristics associated with low-severity ED visits. For example, previous research among pediatric populations suggests that low-severity ED visits are more common among nonwhites,

 individuals with low socioeconomic status,

 and those lacking access to primary care.

 Although research has shown that adult Medicaid enrollees are more  likely to use the ED for low-severity conditions,

 less is known about the predictors of low-severity ED utilization among the publicly insured, especially children.

This study investigates the characteristics of enrollees in ALL Kids (Alabama’s Children’s Health Insurance Program [CHIP]) who  utilized ED services for low-severity conditions. An improved understanding of the determinants of low-severity ED visits holds the key to better designing a system that targets unnecessary utilization of an expensive and scarce resource. Given prior literature, predictors of limited access to primary care (eg, nonwhite race, rural residence, weekend visit) were hypothesized to be associated with children’s visits to the ED for low-severity conditions.

This study utilized a pooled cross-sectional design. The study population comprised enrollees in ALL Kids, Alabama’s CHIP. ALL Kids covers legal Alabama residents younger than 19 years with family incomes from 101% to 200% of the federal poverty level (FPL). In 2010, eligibility was expanded to children living in families with incomes of 300% of the FPL. Alabama is 1 of 15 states with a free-standing CHIP that operates independently of the Alabama Medicaid program and provides enrollees with access to a more extensive private insurance provider network.

 Children enrolled in ALL Kids benefit from full medical, pharmaceutical, and dental coverage from the Blue Cross Blue Shield of Alabama (BCBSAL) preferred provider network. Enrollees pay an annual premium and copayments for various services. The unit of analysis in this study was claims from ED visits from 1999 to 2010, during which 299,906 unique children were enrolled in the ALL Kids program. This study received approval from the University of Alabama at Birmingham Institutional Review Board for Human Use.

Revenue center codes 450 through 459 were used to identify ED visits from medical claims. ED visits resulting in hospitalization were excluded. The ED physician’s primary 

International Classification of Disease, Ninth Revision, Clinical Modification

) diagnosis code was then used to classify each visit by severity based on an algorithm developed at New York University (NYU) by Billings and colleagues.

 The original NYU algorithm assigned every ED visit a probability of belonging to each of 4 severity categories: (1) nonemergent, (2) emergent/primary care treatable, (3) emergent/ED care needed, and (4) emergent/ED care needed, not preventable/avoidable. Injuries and conditions requiring specialty care (eg, those related to drugs, alcohol, or mental health) were excluded from the 4 categories. Some 

diagnosis codes were not included in the original algorithm, and ED visits for these conditions were excluded from our analysis.

We used the method developed by Wharam and colleagues

 to convert the NYU severity category probabilities into a single-dimensional measure of severity. This approach used the sum of the probabilities for the 2 categories in which ED care is needed to assign visits to 1 of 3 severity categories: high severity (>75% probability that ED care was needed); intermediate severity (25%-75% probability that ED care was needed); and low severity (<25% probability that ED care was needed). The severity category groupings utilized in this study have been associated with subsequent hospitalizations and death.

The BCBSAL claims data and enrollment files provided information regarding enrollees’ demographic characteristics including age, sex, zip code of residence, and cost-sharing group that corresponds to their FPL: low-fee group (101%-150% of FPL); fee group (151%-200% of FPL); and those whose family income is 200% to 300% of FPL. A federally exempt group, primarily children of Native American descent, is considered the no-fee group. Data obtained from BCBSAL contained adjudicated claims, fewer than 3% of which contained missing or problematic data such as nonunique identifiers; these data were excluded. Other data collected by ALL Kids during the application process were used to determine enrollee’s race and were merged with the BCBSAL enrollee data. Rural status was determined from the zip code of residence, which was converted to Rural Urban Commuting Area (RUCA) codes using the zip code approximation method and subsequently categorized into 4 levels: urban focused, large rural city/town focused, small rural town focused, and isolated small rural town focused.

 Chronic disease status was determined by the presence of any condition that might involve extensive ongoing management, including human immunodeficiency virus infection, cancer, anemia or coagulation disorders, diabetes, cystic fibrosis, schizophrenic disorders, developmental disorders, infantile cerebral palsy, epilepsy, muscular dystrophy, heart and circulatory diseases or defects, asthma, rheumatoid arthritis, spina bifida, birth trauma, spinal cord injury, and hearing loss.

As a proxy for access to primary care, the proportion of children in each zip code receiving well-child care consistent with the Healthcare Effectiveness Data and Information Set (HEDIS) recommendation in the previous year was calculated using data from the complete ALL Kids enrollment and claims  files. The HEDIS criteria are defined for the first 15 months of life, age 3 to 6 years, and age 12 years and older.

 To simplify, the current analysis defined appropriate well-child care as 6 well-child visits within the first 15 months of life and at least 1 visit for children aged 2 years to 19 years. This area-level measure captured access within the geographical area and did not suffer from the endogeniety of individual-specific adherence to the HEDIS guidelines. Quartiles based on the proportion of children in each zip code meeting the measure were defined as fewer than 22.3% for very low, 22.4% to 33.5% for low, 33.6% to 44.1% for intermediate, and 44.2% or more for high.

All analyses were performed using STATA/MP release 12.1 for Windows (StataCorp LP, College Station, Texas). The bivariate associations between patient characteristics and ED use were assessed using cross-tabs (all covariates were categorical) with statistical significance determined using a 

 statistic. Next, we used multivariate logistic regression to identify the independent effect of individual characteristics on the use of ED services for low-severity versus high-severity conditions. These regressions allowed us to observe differences in patient characteristics between the low-severity and the high-severity ED populations.

We present odds ratios and 95% confidence intervals (CIs) from the logistic regressions for each covariate. However, given the well-known problems of misinterpreting odds ratios when the outcome is not “rare,”

 we also calculated and present marginal effects—also known as absolute risk differences&mdash;associated with that covariate (henceforth, marginal effects will be referred to as risk differences). Essentially, risk differences capture the impact of the covariate on the predicted probability of the outcome compared with the reference category. The magnitude is in the natural metric, percentage points, and represents the slope of the logistic curve evaluated at the sample mean. Standard errors for the risk differences were calculated from the standard errors for the underlying beta coefficients using the delta method. We estimated Huber/White heteroscedasticity-consistent standard errors using a variance-covariance matrix that allows for correlated errors (clustering) among individuals with repeated visits.

A total of 251,480 visits to the ED were identified during the study period. Mental health, alcohol, and drugrelated visits (n = 1028) represented 0.4% of claims and were excluded. Visits for which the diagnosis was unclassifiable by the algorithm totaled 7906 (3.1%) and were excluded. There were 100,584 ED visits for injuries (40%), a heterogeneous group ranging from minor to life-threatening conditions that were not categorized by the Billings and colleagues

algorithm and were therefore excluded. Visits with missing covariates (n = 253) were excluded. Of the remaining 141,709 visits, 97,961 were classified as low severity (69%), 33,941 as intermediate severity (24%), and 9807 as high severity (7%). The 141,709 severity-classified ED visits occurred among 73,530 unique children (59.4% with a single visit and 20.7% with 2 visits); 95% of our data came from children having 5 or fewer ED visits over a span of 12 years. Our main analysis was restricted to low- and high-severity visits (n = 107,768), which occurred among 61,179 children (63.5% had 1 visit, 20% had 2 visits). The vast majority (95%) of children had 4 or fewer visits.

The most common diagnoses for each severity category are shown in

. Low-severity visits were primarily diagnoses related to upper respiratory infections or other cold-like  conditions. Abdominal pain was the primary diagnosis among 18% of intermediate-severity claims, followed by a diverse mix of other conditions. High-severity visits had several asthma-related diagnoses—in all, more than 50% of these visits had an asthma-related diagnosis.

displays the characteristics of ED visits stratified by severity. Low-severity visits occurred more frequently in children aged 1 year and younger, 2 to 3 years, and 4 to 5 years; Caucasians; children with small rural and isolated residences; children who visited the ED on Sundays and Saturdays; and children who lived in areas with low utilization of appropriate well-child care. Low-severity visits occurred less often among children aged 12 to 19 years, males, African Americans, urban residents, children receiving management for a chronic disease, and children living in areas of high utilization of wellchild care.

 presents results from multivariate logistic models that investigated factors associated with low-severity visits. The reported risk differences represent percentage point differences in the fraction of low-severity ED patients who had a given characteristic (vs the referent) relative to high-severity ED patients. For example, the table presents evidence that low-severity ED patients were more likely to be younger than high-severity ED patients. The risk differences indicate that compared with children aged 0 to 1 year, children aged 2 to 3 years were 0.7 percentage points (95% CI, —1.4 to –0.1) less likely to have had a low-severity visit; children aged 4 to 5 years were 1.2 percentage points less likely (95% CI, –1.9 to –0.5); children aged 6 to 11 years were 2.7 percentage points less likely (95% CI, –3.4to –2.1); and children aged 12 to 19 years were 3.7 percentage points less likely (95% CI, –4.3 to –3.0). Low-severity ED patients were less likely to be male (1.4 percentage points; 95% CI, –1.8 to –1.1); to be African American (–1.5 percentage points; 95% CI, –1.9 to –1.1); to have a previously diagnosed chronic disease (10.3 percentage points; 95% CI, –10.7 to –9.9); and to live in an area with high compliance with appropriate well-child care (1.2 percentage points; 95% CI, –0.4 to –1.9). In contrast, lowseverity ED patients were more likely to be from isolated areas (1.6 percentage points; 95% CI, 1.0-2.2), and have visits on  Sunday (0.9 percentage point; 95% CI, 0.6-1.3) and Saturday(1.2 percentage points; 95% CI, 0.1-1.6). Although not presented here, results from the logistic regression model comparing intermediate- versus high-severity visits are included in the 

Using a pooled cross-sectional study of ED claims, we identified characteristics of low-severity ED patients in the Alabama CHIP, ALL Kids. Overall, we found that the majority of non—injury related ED visits were for low-severity conditions. Even though we cannot determine how many of these visits were unavoidable, some of the patterns we identified suggest that barriers to accessing other less resource-intensive sources of healthcare may contribute to use of the ED for low-severity visits. For example, we observed more use of the ED for lowseverity visits relative to high-severity visits on weekends and for rural patients; both findings may represent decreased access to primary care. In addition, children living in zip codes with the highest proportions of appropriate well-child visit rates (a proxy measure of access to primary care) had a lower likelihood of visiting the ED for a low-severity relative to a highseverity condition in the 12 years of data that we examined. Access to primary care is known to reduce low-severity ED utilization. Previous research has found that shortcomings in access to primary care

 or deficiencies in the quality of primary care

 contribute to low-severity ED use. Moreover, Halfon and colleagues

 previously found that primary care physician supply within a county influenced appropriate ED use behavior. The results of the current study, which focused on lowincome children, are consistent with this body of literature.

In our study, children with chronic diseases who utilized the ED were less likely to do so for low-severity visits than children without a chronic disease. This might be either because children with chronic diseases utilized the ED for reasons related to the underlying condition and as a result the visits were classified as more severe in nature, or because the increased contact with the healthcare system stemming from their chronic disease made them better consumers of appropriate resources. Children with chronic illnesses are known to use the ED with greater frequency than other children,

but studies have not evaluated the severity of those visits. The primary chronic disease observed among children in ALL Kids was asthma. Although asthmatic episodes can be prevented through improved disease management,

children experiencing symptoms need ED care and this is considered a high-severity visit by the Billings and colleagues

 algorithm and Wharam and colleagues method.

Nonmodifiable characteristics such as sex and race were associated with differential patterns of ED use. Although some prior  studies have suggested that African Americans are more likely to use the ED for low-severity care,

 In the  current study, African American children who utilized the ED were less likely to do so for a low-severity ED visits compared with  children of other races. We also found hat female patients had a larger fraction of low-severity ED visits, which is consistent with previous work that identified females as using the ED for routine care more frequently.

 Lastly, we observed younger age groups  were associated with a larger share of low-severity ED visits than high-severity ED visits. Infants younger than 12 months of age  have the highest annual ED visit rate nationally,

 potentially due to the tendency of parents to err on the side of caution with newborns. 

The findings of this study should be interpreted in light of some limitations. First, claims data have inherent limitations for research or clinical purposes given that they are collected for administrative reasons.

Second, ED claims in our study were categorized based on primary diagnosis only, and the algorithm for classifying visit severity based on diagnosis codes may have failed to capture all aspects of a child’s ED visit. Despite this, the methodologies of Billings and colleagues

 are reproducible and have been validated. If physicians make arbitrary decisions as to which diagnosis is listed first or second, then  misclassification of severity could have occurred. Third, the algorithms to classify ED visits may have limitations when applied exclusively to children, given that they were developed in a multiuse ED. Lastly, the results of this study should be generalized with caution to populations outside Alabama or CHIPs in general, especially when administered differently from ALL Kids.

Several factors are associated with the use of the ED for low-severity conditions among Alabama children enrolled in CHIP. However, some modifiable factors are reflective of limited utilization of, and possibly access to, quality primary care. Low-severity ED utilization may be reduced when primary care providers offer evening and weekend hours.

 Moreover, retail health clinics and urgent care centers, which utilize fewer resources than EDs, offer the potential to more efficiently treat low-severity conditions during off-hours.

 Policy makers may be able to reduce low-severity ED visits by encouraging expanded access to times and  locations that primary care services are provided.

 From Department of Health Care Organization and Policy (JB, DJB, MAM, NM), University of Alabama at Birmingham, Birmingham, AL; Alabama Department of Public Health (CC), Bureau of Children’s Health Insurance, Montgomery, AL.

 This work was supported by the Alabama Department of Public Health, Bureau of Children’s Health Insurance. The funding agency maintains the data used for this study, but the analysis and interpretation are the authors’ own.

 The authors (JB, DJB, BS, MAM, CC, NM) 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 (JB, DB, MAM, NM); acquisition of data (JB, NM); analysis and interpretation of data (JB, MAM,NM); drafting of the manuscript (JB); critical revision of the manuscript for important intellectual content (MAM, DB, BS, CC, NM); statistical analysis (JB); provision of study materials or patients (CC); obtaining funding (MAM, NM); administrative, technical, or logistic support (CC); and supervision (NM).

 Justin Blackburn, PhD, 1665 University Blvd, Rm 330, Birmingham, AL 35294. E-mail: jblackburn@uab.edu.1. Niska R, Bhuiya F, Xu J. National Hospital Ambulatory Medical Care Survey: 2007 emergency department summary. Natl Health Stat Report. 2010;(26):1-31.

2. Gordon JA, Billings J, Asplin BR, Rhodes KV. Safety net research in emergency medicine: proceedings of the Academic Emergency Medicine Consensus Conference on “The Unraveling Safety Net.” Acad Emerg Med. 2001;8(11):1024-1029.

3. Johnson WG, Rimsza ME. The effects of access to pediatric care and insurance coverage on emergency department utilization. Pediatrics. 2004;113(3, pt 1):483-487.

4. Weinick RM, Burns RM, Mehrotra A. Many emergency department visits could be managed at urgent care centers and retail clinics. Health Aff (Millwood). 2010;29(9):1630-1636.

5. Grumbach K, Keane D, Bindman A. Primary care and public emergency department overcrowding. Am J Public Health. 1993;83(3): 372-378.

6. Lowe RA, Localio AR, Schwarz DF, et al. Association between primary care practice characteristics and emergency department use in a Medicaid managed care organization. Med Care. 2005;43(8):792-800.

7. Billings J, Parikh N, Mijanovich T. Emergency department use in New York City: a substitute for primary care? Issue Brief (Commonw Fund). 2000;433:1-5.

Barriers to less resource-intensive settings may contribute to use of the emergency department for low-severity conditions.

December 2013

Implementing Effective Care Management in the Patient-Centered Medical Home

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

Characteristics of Low-Severity Emergency Department Use Among CHIP Enrollees

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

Impact of Electronic Prescribing on Medication Use in Ambulatory Care

Economics of Genomic Testing for Women With Breast Cancer