Claire M. Spettell, PhD

The dramatic rise in healthcare expenditures calls for innovative and scalable strategies to achieve measurable, near-term improvements in health. Our objective was to determine whether a remotely delivered behavioral health intervention could improve medical health, reduce hospital admissions, and lower cost of care for individuals with a recent cardiovascular event.

This retrospective observational cohort study included members of a commercial health plan referred to participate in AbilTo’s Cardiac Health Program. AbilTo is a national provider of telehealth, behavioral change programs for high risk medical populations.

The program is an 8-week behavioral health intervention delivered by a licensed clinical social worker and a behavioral coach via phone or secure video.

Among the 201 intervention and 180 comparison subjects, the study found that program participants had significantly fewer all-cause hospital admissions in 6 months (293 per 1000 persons/year vs 493 per 1000 persons/year in the comparison group) resulting in an adjusted percent reduction of 31% (

= .03), and significantly fewer total hospital days (1455 days per 1000 persons/year vs 3933 per 1000 persons/year) with an adjusted percent decline of 48% (

= .01). This resulted in an overall savings in the cost of care even after accounting for total program costs.

Successful patient engagement in a national, remotely delivered behavioral health intervention can reduce medical utilization in a targeted cardiac population. A restored focus on tackling barriers to behavior change in order to improve medical health is an effective, achievable population health strategy for reducing health costs in the United States.

		Inadequately addressed behavioral health issues commonly accompany medical conditions and account for worse medical outcomes and greater healthcare utilization.

		This study demonstrates that a remotely delivered behavioral health intervention targeted to individuals with high-risk medical conditions can indeed reduce medical resource utilization and lower healthcare costs within 6 months.

		This study confirms that a collaborative strategy to bridge the gaps between behavioral health and medical health serves as an effective and achievable population health strategy to improve quality and lower cost of care.

Despite tremendous progress in improving morbidity and mortality in patients with high-risk medical conditions such as cardiovascular disease, healthcare expenditures continue to rise at a dramatic pace.

 Innovative population health strategies that focus not only on disease but also on sustainable improvements in health and well-being are sorely needed. The anticipated benefits include happier, healthier patients, ultimately leading to lower healthcare costs. Unfortunately, existing strategies to improve health have not adequately focused on addressing the behavioral determinants of health, which when adequately treated, may lead to tangible optimization of medical health and reductions in medical utilization.

For many individuals with medical conditions such as cardiovascular disease, concomitant behavioral health issues—such as depression, stress, and anxiety&mdash;are common, and pose substantial challenges to recovery from medical illness.

Even in those individuals who do not meet the clinical criteria for behavioral health concerns, inadequate resiliency to cope with the challenges posed in the face of a medical or life event can significantly impact health. In patients with a recent cardiovascular event (eg, myocardial infarction [MI], coronary artery bypass surgery, or congestive heart failure), major depression is known to affect as many as 1 in 4 individuals and can lead to adverse cardiac outcomes, greater all-cause mortality, and significantly greater healthcare utilization.

 Addressing these behavioral health issues and helping patients to develop the life skills needed to overcome barriers to self-care and self-management are necessary prerequisites to improving medical health and lowering healthcare costs.

Aetna, a national health benefits company, enhanced its care management programs in 2011 by collaborating with AbilTo, a network of behavioral health providers, to provide structured, condition-specific behavioral health programs to its members identified with specific medical conditions. We hypothesized that an intervention that successfully engages patients to address the behavioral health issues that commonly accompany high-risk medical conditions, such as cardiovascular disease, would lead to improved health resource utilization and lower healthcare costs. Accordingly, we conducted a retrospective study to assess the impact of AbilTo’s Cardiac Health program, a remotely delivered behavioral health intervention, on all-cause hospital readmissions and total hospital days during the 6-month follow-up period in a commercially insured population of individuals with cardiovascular disease.

A retrospective, observational study design was used to compare individuals who had completed at least 7 weeks of the 8-week AbilTo program with those who had completed the initial assessment and 2 or fewer weeks of the program. Individuals with partial program completion (ending anytime between week 3 and week 6) were not included in this analysis.

Commercially insured patients were included in the study if they met the following inclusion criteria: 1) referral from Aetna to AbilTo’s Cardiac Health Program based on evidence of a recent cardiovascular event; 2) completion of an initial consultation with an AbilTo therapist; and 3) availability of continuous enrollment with Aetna 6 months prior to and 6 months post AbilTo program intake. This approach was selected so that baseline behavioral health symptom scores were available for all individuals, as these symptoms can independently affect the utilization and cost outcome measures. All participants had Aetna as their primary healthcare benefits provider. Cardiovascular events were defined on the basis of hospitalization or outpatient claims submitted with a principal diagnosis code from the 

International Classification of Diseases, Ninth Revision, Clinical Modification 

) for myocardial infarction (410.xx), intermediate coronary syndrome (411.xx), or cardiac dysrhythmia (427.xx), or with a principal procedural code from Current Procedural Terminology-10 (CPT-10) for coronary artery bypass surgery, valve surgery, coronary stenting, or angioplasty.

Between September 2011 and May 2013, 552 participants were referred to AbilTo’s Cardiac Health program and completed an initial consultation. Of these, 251 individuals completed 7 weeks or more, 241 completed 2 weeks or less of the program, and 60 participants were excluded on the basis of partial program completion (completing between 3 and 6 weeks). After applying the requirement for 6 months pre-intake and 6 months post intake eligibility for Aetna medical benefits, there were 201 individuals remaining in the intervention group and 180 individuals in the comparison group.

AbilTo’s programs are based on widely accepted behavior change tools, including cognitive behavioral therapy (CBT), acceptance and commitment therapy (ACT), mindfulness, motivational interviewing, and other related, evidence-based best practice approaches. All protocols are crafted by AbilTo’s clinical team in collaboration with an advisory group consisting of psychiatrists, psychologists, and other medical professionals. All programs are delivered remotely by telephone or secure video, and the care is administered by a specially trained provider team consisting of a behavioral health provider (a licensed clinical social worker [LCSW] or equivalent) and a behavioral coach. All providers receive Association of Social Work Board—certified training administered specifically by AbilTo in order to ensure delivery of AbilTo’s best practices treatment protocols.

The programs are all 8 weeks in duration and consist of 16 sessions in total. Program participants undergo a one-time clinical intake session (“initial consultation”) with an LCSW followed by separate weekly one-on-one sessions with both an LCSW and a behavioral coach for a total of 8 weeks. Providers participate in case conferences under the guidance of an LCSW clinical supervisor during the course of each program to review the participant’s progress. The clinical supervisor also reviews all case notes on a weekly basis to ensure high quality and adherence to the treatment protocols. A proprietary content management system ensures consistent delivery of program content and provides a secure platform in compliance with the Health Insurance Portability and Accountability Act (HIPAA) to allow sharing of clinical notes among the LCSW, behavioral coach, and clinical supervisor.

The primary outcome measures were all-cause hospital admissions and total hospital days in the 6-month period from the date of initial consultation. Additional outcome measures included: total emergency department (ED) visits; outpatient visits, including behavioral health (BH) visits; and cardiac-specific hospital readmissions and hospital days. Outcome measures were derived from Aetna medical claims data that included facility and professional services. Claims for AbilTo sessions were adjudicated as behavioral health visits by the participants’ health plan, and were included in the overall BH outcome metric.

Severity of depression, anxiety, and stress were evaluated using the Depression Anxiety Stress Scale 21 (DASS-21), a scale to measure these behavioral health dimensions that has been widely validated in multiple clinical populations.

 The DASS-21 was administered by an LCSW at baseline (for both groups) and at program completion (intervention group only). Baseline demographics and presence of clinical conditions were collected from Aetna’s administrative databases. Scores representing each participant’s risk of future healthcare usage were calculated using Ingenix Episode Risk Group software.

The descriptive analyses of baseline differences used t tests for continuous variables and χ

 tests for categorical variables. Multivariable logistic regression was used to test the odds of binary outcomes—such as likelihood of an inpatient admission or ED visit&mdash;controlling for demographic and baseline differences between the groups. Poisson or negative binomial multivariable regression was used to test the differences in count data such as inpatient admissions, ED visits, and office visits between the groups. All analyses were conducted using SAS 9.2 software (SAS Institute Inc, Cary, North Carolina).

Participant identification for the AbilTo program intervention was made on the basis of a triggering cardiac event. The participants included in this study were primarily referred by nurses evaluating individuals as part of care management programs led by the health plan. Participants were additionally identified by targeted outreach to at-risk eligible individuals known to have a recent cardiovascular event identified on the basis of relevant 

 or CPT-10 codes as described above. Of these, 552 individuals were referred for participation, completed an initial consultation, and were judged eligible for this analysis. As shown in 

, of the 552 individuals referred who completed an initial consultation, 394 (71%) enrolled (ie, completed week 1), and of these 394, 242 (61%) completed the program. To ensure completeness of data for analysis, individuals were required to have been enrolled in the health plan for the 6-month pre and post periods. Therefore, we were left with 201 individuals in the intervention group and 180 individuals in the comparison group for the final study population.

The baseline characteristics were very well balanced between the 2 study groups (

). The average age was 56 years in both groups, and a similar proportion of both groups were male (70% in the intervention group and 67% in the comparison group). Although there were slightly more individuals in the intervention group from the northeast region and fewer from midwest (

= .04), the groups were well balanced with respect to the proportion of participants residing in rural, suburban, and urban community settings. The prevalence of baseline comorbid clinical conditions was similar in the 2 groups, including rates of diabetes, hypertension, and hyperlipidemia (Table 1). There were no significant differences in average baseline maximal DASS-21 scores for depression, anxiety, and stress (Table 1), and more than 60% of participants in each group fell into the normal subclinical depression range on the DASS-21 scale (

). In addition, there was a nonsignificant trend toward more individuals in the comparison group (43.3%) falling in the normal range on all 3 dimensions of the DASS-21 scale compared with the intervention group (32.2%). There was no significant difference in the prospective risk scores between the 2 groups (6.39 ± 5.2 in the intervention group vs 6.85 ± 6.2 in the control group; 

= .43), and baseline medical utilization (in the 6-month period prior to clinical intake) showed no significant differences between the 2 groups with similar rates of pre-period total medical expenditures, total inpatient expenditures, inpatient admissions, cardiac-specific inpatient admissions, ED visits, and total outpatient services (

). The only difference noted was higher utilization of outpatient behavioral health services in the comparison group at baseline (1544 per 1000 persons per year vs 842 per 1000 persons per year in the intervention group; 

Individuals in the AbilTo intervention group had significant reductions in the severity of all components of the DASS-21 score, including depression, anxiety, and stress (

). During the 6-month follow-up period, the intervention group had 38% fewer total hospital admissions (386 admissions per 1000 persons per year vs 622 per 1000 persons per year in the comparison group) (

). After multivariable adjustment for demographic variables, baseline risk score, and baseline depression score, as well as pre-period medical utilization, the intervention group had a statistically significantly 31% fewer hospital admissions (

= .03) during the 6-month followup period (

). Interestingly, a similar proportion of individuals in each group were hospitalized during the 6-month period (15% in the intervention group vs 21% in the comparison group; 

= .16) toward more individuals in the comparison group with multiple (2 or more) admissions (7.2%) compared with fewer multiple admissions in the intervention group (3.5%).

Individuals in the AbilTo intervention group had 63% fewer total inpatient hospital days in the 6-month follow-up period (1455 days per 1000 persons per year in the intervention group compared with 3933 days per 1000 persons per year in the comparison group). These data were statistically significant with an adjusted reduction of 48% (

= .01) even after accounting for demographics, baseline risk, baseline depression scores, and pre-period medical utilization.

There was no significant difference in utilization of ED services in the 2 groups, although there was a trend toward fewer ED visits in the intervention group (505 per 1000 persons per year vs 689 per 1000 persons per year in the comparison group; adjusted 

= .40). The intervention group had significantly more behavioral health visits—almost entirely accounted for by participation in the AbilTo program itself (19,713 visits per 1000 persons per year vs 2822 per 1000 persons per year in the comparison group; adjusted 

<.0001). Excluding behavioral health visits, there were no differences in utilization of all outpatient services (34,634 visits per 1000 persons per year vs 31,167 per 1000 persons per year; adjusted 

 = .19), nor were there any differences in cardiac-specific outpatient visits (

To determine the potential cost savings derived from program participation, we calculated the potential cost savings attributable to the significant reduction in total days in the hospital. To estimate the potential savings, we first determined the average total cost of an inpatient hospital day (including facility, professional, and ancillary charges) for the comparison group in the 6-month follow-up period using claims data for this population. Individuals in the comparison group averaged 1.88 inpatient hospital days in the 6-month follow-up period, and the average cost was $4500 per hospital day. Using the adjusted percent difference in total inpatient hospital days of 48% between the comparison group and the AbilTo intervention group, each individual completing the AbilTo program would be expected to avoid an average of 0.95 inpatient hospital days. Thus, the 202 individuals in the intervention group were estimated to have saved 191.9 hospital days. Applying the average cost per hospital day of $4500, we estimated that the individuals who fully participated in the AbilTo program saved $864,000 in the 6-month follow-up period. Comparing this cost saving with the estimated total program cost demonstrated an overall cost savings as early as 6 months.

We demonstrated that an 8-week remotely delivered behavioral change intervention was associated with cost savings, driven by an adjusted 48% reduction in total inpatient days and a 31% reduction in all-cause hospital admissions in the 6-month follow-up period. These substantial reductions in healthcare utilization and associated cost savings were attributable to the delivery of a high-quality behavioral health program for this high-risk group of patients with cardiovascular disease. This study shows that focused targeting of patients with highrisk clinical conditions, coupled with highly successful engagement strategies, can lead not only to meaningful behavioral health improvements, but also to improved medical outcomes and lower healthcare expenditures.

It has been long recognized that behavioral health issues can be both a cause and a consequence of medical disease.

In individuals with cardiovascular disease, comorbid behavioral health concerns are common—affecting up to 25% of patients

7—and result in poorer adherence to medications and lifestyle recommendations as well as worse overall clinical outcomes, including increased hospital readmissions and higher mortality.

 Taken together, inadequate management of behavioral health issues can lead to unnecessarily greater medical utilization and as much as a doubling in the cost of care.

It stands to reason that a program that successfully influences behavioral health could have a profound impact on overall health, medical utilization, and total health expenditures. However, prior studies to assess the effect of behavioral health interventions in cardiovascular disease have met with mixed results. For example, the Enhancing Recovery in Coronary Heart Disease (ENRICHD) trial investigated the effects of CBT, with or without pharmacologic intervention, on post MI patients with depression, and found no difference in the primary end point of event-free survival at an average follow-up of 29 months.

On the other hand, a follow-up evaluation of the ENRICHD study data showed that the intervention led to reduced late-term mortality with the benefits corresponding to the degree of improvement in depression.

Other studies using a purely pharmacologic intervention for depression have not shown an impact on cardiovascular outcomes or mortality.

 A few recent studies have employed a remote or Internet-based approach to reach patients with cardiovascular disease and have demonstrated improvements in depression, adherence, and quality of life.

Our study and our intervention differ from the existing literature in several important ways. First, we did not design our intervention to focus only on improvements in behavioral health or cardiac outcomes. Instead, our goal was to demonstrate that successful targeting and engagement of high-risk cardiac patients in a behavioral health intervention would impact 2 specific outcomes: 1) medical utilization, and 2) healthcare expenditures. The impact of treating behavioral health on these critical components of the healthcare continuum has not previously been well studied in cardiovascular disease.

Articles

Successful patient engagement in a nationally available, remotely delivered behavioral health intervention can significantly improve medical outcomes and lower healthcare costs.

Leveraging Remote Behavioral Health Interventions to Improve Medical Outcomes and Reduce Costs

Evaluation of quality of care across retail clinics in a geographically diverse population has not been undertaken to date. We sought to evaluate and compare the quality of care for otitis media, pharyngitis, and urinary tract infection received in retail medical clinics in CVS pharmacies (“MinuteClinics” [MCs]), ambulatory care facilities (ACFs), and emergency departments (EDs).

We used 14 measures constructed from RAND Corporation’s Quality Assurance Tools and guidelines from the American Academy of Pediatrics, the American Academy of Family Physicians, and the Infectious Diseases Society of America. Our cohort was drawn from Aetna medical and prescription claims, 2009-2012. Members were matched on visit date, condition, and propensity score. Generalized estimating equations were used to compare quality across clinic type, overall, and by index condition.

We matched 75,886 episodes of care, of which 20,153 were eligible for at least 1 quality measure. MCs performed better than EDs and ACFs in 7 measures. In a multivariable model, MCs performed better than ACFs and EDs across all quality measures ([OR 0.42; 95% CI, 0.40-0.45; 

 <.0001; ACF vs MC] [OR 0.29; 95% CI, 0.27-0.31; 

 <.0001; ED vs MC]). Results for each condition were significant at 

Quality of care for these conditions based on widely accepted objective measures was superior in MinuteClinics compared with ACFs and EDs.

		These findings are consistent with previous studies demonstrating that quality of care is not compromised, and even appears superior, in retail clinics for specific acute conditions.

		When taken together with evidence suggesting that retail clinics are more cost-effective and even cost saving to patients, these results underscore the potential of retail clinics for offering care of higher quality and lower cost at a time of primary care shortages.

		More extensive studies exploring costs for chronic disease management and the role of retail clinics in improving care coordination and population health management are warranted.

Retail clinics have become increasingly prevalent in the United States. As of September 2014, there are about 1760 retail clinics,

 a number that by some estimates will grow to 2800, with 10.8 million annual visits made to them, by 2015.

 Retail clinics aim to treat a limited range of clinical conditions that are routine family medical problems, and thus can serve as an alternative to care in physicians’ offices, ambulatory/urgent care facilities (ACFs), or emergency departments (EDs).

 These clinics typically rely on nurse practitioners and physician assistants who deliver care based on strict adherence to clinical practice guidelines.

With increasing calls for healthcare delivery systems that can reduce the cost of care, offer more convenient primary care options to patients, and address primary care shortages, retail clinics may offer an attractive option.

 Recent studies suggest that retail clinics deliver care at substantially lower cost than ACFs and EDs, and have been associated with lower cost of care to the patient.

 Overall patient satisfaction is high, with walk-in care convenience and fixed, transparent pricing cited as 2 main attractions of the retail clinic model.

At the same time, some have expressed concern that the retail clinic model of care may not provide care equivalent in quality to what is available in other settings.

 Professional physician organizations, such as the American Academy of Pediatrics (AAP) and the American Academy of Family Physicians (AAFP), have expressed ongoing concern about retail clinics, including questions about overall quality of care and the use of tests for the purpose of diagnosis without appropriate follow-up.

 Notwithstanding these concerns, evidence regarding quality of care for acute conditions in particular, which account for the majority of retail clinic services, has suggested that the quality of care is high. A study by Jacoby et al evaluating retail clinic performance on 2 Healthcare Effectiveness Data and Information Set (HEDIS) measures—the appropriate testing of children with strep throat and appropriate treatment of children with upper respiratory tract infection&mdash;concluded that retail clinics ranked above the 90th percentile for the former measure and between the 50th and 90th percentiles for the latter.

A study by Rohrer et al evaluating return visits for children treated for ear infection found a lower rate of return visits within 2 weeks in retail clinics compared with standard office clinics,

 while a different analysis of retail clinics in Minnesota by Mehrotra et al examined 14 performance measures for care of 3 common conditions treated at retail clinics and found that the quality of care was superior to that received at ACFs and EDs.

 Nonetheless, evidence in this domain remains sparse. As the prevalence of these clinics grows, a contemporary evaluation of quality of care for acute conditions in a geographically diverse population is essential. We applied the performance measures for urinary tract infection (UTI), strep throat (pharyngitis), and ear infection (otitis media; OM) used in analysis of retail clinics in Minnesota by Mehrotra et al to a nationally representative cohort to compare the quality of care in retail clinics with that in ACFs and EDs.

This study was approved by the institutional review board of Partners Healthcare System, Boston, Massachusetts, and Harvard University, Cambridge, Massachusetts.

We used nationwide medical and prescription claims from Aetna, a large healthcare benefits company, with dates of service from January 1, 2009, through December 31, 2012. The study population included commercial members with a visit to a MinuteClinic (the retail medical clinics in CVS pharmacies, with locations in 25 states and the District of Columbia during the study period), an ACF facility, or an ED for OM, pharyngitis, or UTI during this period. ACF visits were defined using Aetna’s definition of all visits to outpatient clinics including non-ED urgent care clinics, same-day surgeries, and same-day ambulatory hospital encounters. The analysis was restricted to the first episode of care for 1 of the 3 index conditions in the study period; if a member had episodes of care for more than 1 index condition, only the earliest episode and corresponding index condition were considered. Clinic type was identified using place of service codes, including specific codes for retail clinics, and conditions were identified using 

International Classification of Disease, Ninth Revision, Clinical Modification 

) codes for OM (381, 382, 388.7), pharyngitis (034, 462, 463), and UTI (599, 595). Members were required to be continuously enrolled for the 6 months prior to and 60 days after their index episode.

We matched members with MinuteClinic episodes of care to those from ACFs and EDs in a 1:1:1 fashion on the following criteria: month of index visit, index condition, and a propensity score. The propensity score was constructed from the following baseline characteristics: age, gender, tertile of household income (per zip code from 2000 US Census data), index condition, number of prescription medications and medical visits in the 6 months prior to the index visit, and Charlson Comorbidity Index score using ICD-9-CM codes from the 6 months prior to the index visit. Matching was operationalized using a validated SAS macro with a 0.05 caliper for the propensity score.

 We sampled without replacement, meaning an individual is a member of only 1 matched set.

Some zip codes in the 2000 US Census do not have income data available, and so members living in these zip codes were grouped into the second income tertile. A multiple imputation (N = 10) of these missing values was conducted as a sensitivity analysis.

Quality of care was assessed using 14 measures constructed from RAND Corporation’s Quality Assurance (QA) Tools as well as guidelines from the AAP, the AAFP, and the Infectious Diseases Society of America, measures also employed in the analysis by Mehrotra et al of retail clinics in Minnesota (

The RAND QA Tools were developed through comprehensive evaluation of research literature and clinical guidelines, and are considered a gold standard for measurement of whether patients are receiving evidence-based healthcare including appropriate diagnosis, treatment, and follow-up.

Each measure is a binary marker of quality of clinical care (eg, throat culture or rapid strep test ordered on day of initial visit or day after in diagnosis of pharyngitis) and was oriented such that a higher score (ie, 1) indicates higher quality. There were 5 measures assessing quality of care for OM, and 6 and 3 measures for pharyngitis and UTI, respectively.

We evaluated whether performance on each quality measure differed between visits to MinuteClinics and ACFs or EDs using χ2 tests. To compare overall quality across the 3 clinic types, we used a generalized estimating equation with a binomial distribution and a logit link. Because members were typically eligible for more than 1 quality measure and up to as many as 5 measures, this model accounted for correlation across member responses using an exchangeable covariance matrix. We included the quality measure in the model as a categorical variable to account for differing mean success rates across measures. We compared this analysis with a multivariable analysis that also adjusted for the following baseline characteristics: gender, age, income, number of medications, and number of medical visits in the 6 months prior to the visit.

We excluded quality measure 11 from all models, as did Mehrotra et al, because from a clinical perspective, the measure does not discriminate high versus low quality of care as distinctly as the other 13 measures. We used SAS, version 9.3 (SAS Institute, Cary, North Carolina) to conduct all analyses.

We identified 1,231,866 eligible members with an index episode for 1 of the 3 conditions occurring at 1 of the clinic types during the 3-year study period (

). The majority of episodes (91%) were ACF visits, with 6% and 2% occurring at EDs and MinuteClinics, respectively. The majority of cases seen in MinuteClinics were for pharyngitis (61% vs 51% and 37% in ACFs and EDs, respectively), whereas the majority of cases seen in EDs were for UTI (40% vs 19.3% and 9.4% for ACFs and MinuteClinics, respectively).

Members seen in MinuteClinics tended to have lower Charlson Comorbidity Index scores and fewer physician visits in the 6 months prior to index. Children under 5 years of age comprised 3.4% of cases seen at MinuteClinics, compared with 10.6% and 13.7% in EDs and ACFs, respectively. Members seen in EDs generally lived in lower- income zip codes, followed by ACFs (second lowest) and MinuteClinics (highest).

After matching on propensity score, index condition, and index month, matches were found for 25,334 members (87% of eligible MinuteClinic episodes). The matched cohort was predominantly female (61%) and the majority (54%) were aged between 18 and 44 years. Most episodes were for pharyngitis (58%), while three-fourths of members had had at least 1 doctor’s visit in the 6 months prior (74%), and about half had 1 to 5 unique medication fills in the 6 months prior (51%).

There were 20,153 members from the matched cohort who were eligible for at least 1 quality measure; the number for which they were eligible ranged from 1 to 5. MinuteClinics had higher success rates than both EDs and ACFs in 7 of 14 measures (

). Differences in success rates across sites of care were statistically significant (

 = .05) in 10 measures, including all pharyngitis and UTI measures and 1 OM measure. Pooled, unadjusted success rates by index condition are presented in the 

MinuteClinics performed better than ACFs and EDs across all quality measures ([OR 0.43; 95% CI, 0.40-0.45, comparing ACFs with MinuteClinics; OR 0.29; 95% CI, 0.27-0.31 comparing EDs with MinuteClinics]) with both results significant at P <.0001 (

). Results were very similar after adjustment for baseline covariates ([OR 0.42; 95% CI, 0.40-0.45, comparing ACFs with MinuteClinics; OR 0.29; 95% CI, 0.27-0.31, comparing EDs with MinuteClinics]).

Results stratified by each of the conditions individually were also similar. The difference in quality between MinuteClinics and other clinic types was most pronounced for UTI ([OR 0.31; 95% CI, 0.27-0.35, comparing ACFs with MinuteClinics; OR 0.28; 95% CI, 0.25-0.32, comparing EDs with MinuteClinics]), and least pronounced for OM ([OR 0.82; 95% CI, 0.75-0.89, comparing ACFs with MinuteClinics; OR 0.73; 95% CI, 0.67-0.80, comparing EDs with MinuteClinics]). Results across each index condition remained statistically significant at 

Nearly 5% of income values were missing in the cohort prior to matching. After multiply imputing (N = 10) these values, results from the models adjusting for quality measure number were very similar ([OR 0.45; 95% CI, 0.42-0.48, comparing ACFs with MinuteClinics; OR 0.30; 95% CI, 0.29-0.32; comparing EDs with MinuteClinics]).

The results from this analysis suggest that overall quality of care for OM, pharyngitis, and UTI at MinuteClinics is superior to that delivered at ACFs or EDs

 based on widely accepted objective measures. Whereas ACFs and EDs performed better on several individual quality measures, overall performance at the index condition level and across all measures was consistently higher in MinuteClinics. These results did not differ after adjustment for baseline covariates.

This study is the largest to date evaluating the qualityof care for acute conditions at retail clinics compared with other care settings, and corroborates previous findings regarding quality of care.

In particular, the success rates by quality measure and clinic type (Table 2) were similar to those of Mehrotra et al.

However, Mehrotra et al did not observe statistically significant differences by site of care in regression analysis, which may in part be due to the fact that the sample from that prior analysis was less than 1/15 the size of ours. Moreover, we account for correlation within success rates by individuals who are eligible for more than 1 quality measure, as well as response trends within a quality measure (eg, 1 measure may have a higher success rate than another), providing more robust results. By matching on numerous demographic and clinical covariates, we also addressed concerns about residual confounding by baseline characteristics.

It is notable that MinuteClinics performed worse than other settings for UTI indicators that focused on high-risk patients and long-term antibiotic therapy to treat complicated UTIs (quality measures 12 and 14). MinuteClinic guidelines indicate that such patients should be referred to more acute settings; until 2014, MinuteClinics did not even offer urine cultures on the premises. The low prevalence of visits for these measures suggests that patients are selecting appropriate levels of care for this condition.

We did not include quality measure 11 in our multivariate analyses, consistent with the exclusion in the study conducted by Mehrotra et al. The indicator was not deemed clinically appropriate as designed, as it rewards antibiotic prescribing even in the absence of documented strep infection. In contrast, in the setting of an OM diagnosis, antibiotic treatment is clinically appropriate (measure 5).

The 3 conditions we examined, along with several upper respiratory conditions, swimmer’s ear, and pinkeye, comprise nearly 90% of all MinuteClinic acute care visits.

 When taken with evidence that retail clinics offer care at lower cost than other places of service and also lower the patient and payer’s total cost of care, the results of our study suggest that strategies for treatment of these conditions may be both economically advantageous and cost-effective.

Our study did not examine MinuteClinic preventive care service visits, about half of which are for influenza vaccination.

More research is needed to understand the quality of preventive care services offered at retail clinics as well as the integration of these services into a larger primary care medical home strategy. In particular, ACFs may provide better education or preventive services that reduce the rates of subsequent presentation.

Since 2010, retail clinics have begun to broaden their array of services, including care for chronic disease management. While early findings suggest that there is no detrimental effect of retail clinic care on preventive care and diabetes management,

 more research is needed to evaluate the quality of care in expanded domains and to examine the extent to which retail clinic care can be integrated into preventative care services offered in other settings.

Some research has shown that more affluent people have greater access to retail clinics,

and that retail chain pharmacies tend to locate in wealthier neighborhoods. 30 Conversely, other studies have observed that retail clinics serve a large segment of people with weak or no ties to a primary care physician.

 In our analysis, members seen at MinuteClinics generally lived in more affluent areas than those seen at EDs and ACFs, but by matching on income we ensured that this potential confounder did not bias our results. Our analysis examined only MinuteClinic retail clinics, offered by a single pharmacy chain. While MinuteClinics represent more than half the nation’s retail clinics, the extent to which these findings can be generalized to other retail clinics is not known.

Performance measures are inherently limited in their ability to capture the full spectrum of clinically appropriate care. While the RAND QA Tools and other similar measure sets reflect the current standard of measurement for appropriate diagnosis, treatment, and follow-up for acute conditions,

 the use of any claims-based quality indicators is limited in its ability to fully account for illness severity and capture clinical reasoning that would justify diagnostic and treatment behavior outside clinical guidelines. Matching on Charlson Comorbidity Index score as well as other patient characteristics, such as age associated with illness severity, may account for many of the severity differences that exist across these care settings. If anything, we expect that patients with more severe illness would be more likely to receive guideline-adherent care, which would bias our results toward the null under the assumption that EDs and ACFs differentially treat patients with more severe illness. Additionally, the quality measures in this analysis focus on same-day care and so are not subject to any bias from follow-up care that may have been received differentially at another care setting. Mehrotra et al found that fewer than 20% of episodes in their analysis included any follow-up visits.

Finally, to capture clinical reasoning that would justify treatment outside of these quality indicators, we would need data typically only available in electronic medical records, which were outside the scope of this analysis but may be an area of future research.

Overall, these findings are consistent with previous studies that demonstrate that quality of care is not compromised, and even appears superior, in retail clinics for specific acute conditions. When taken together with evidence suggesting that retail clinics are more cost-effective and even cost saving to patients, these results underscore the promise of retail clinics in offering care of higher quality and lower cost at a time of primary care shortages.

 In a healthcare system that sees 177 million ED and 577 million physician office visits annually, there is significant potential for retail clinics to play an increasingly important role in reducing the burden on our primary care channels.

In light of a transforming healthcare system, more extensive studies exploring costs for chronic disease management and the role of retail clinics in improving care coordination and population health management are warranted.

CVS Caremark, Woonsocket, RI (WHS, OSM, AS, TAB); Division of Pharmacoepidemiology and Pharmacoeconomics, Department of Medicine, Brigham and Women’s Hospital and Harvard Medical School, Boston, MA (AAK, AYT, NKC); Aetna Informatics, Blue Bell, PA (CMS).

 This work was supported by a research grant from CVS Caremark to Brigham and Women’s Hospital.

Quality of care for 3 conditions, based on widely accepted objective measures, was superior in MinuteClinics compared with ambulatory care facilities or emergency departments.

Quality of Care at Retail Clinics for 3 Common Conditions

 To assess the impact of supplementing nurse case management with Internet-connected telemonitoring on clinical outcomes in an elderly heart failure (HF) population.

 Randomized clinical trial allocating high-risk HF subjects to the telehealth system with case management (THCM) versus case management (CM) alone.

Medicare Advantage members with HF and recent hospitalization were randomized to THCM or CM. Consenting participants received THCM for 6 months or the health plan’s usual Medicare specialized CM as long as deemed necessary. The primary outcome measure was a composite of all-cause hospitalization, ED visit, or death. Because only a subset of those who were randomized consented and participated, the strategies were evaluated as-treated rather than by intention-to-treat, with multivariate adjustment in logistic regression models for confounding introduced by postrandomization exclusions.

The treated sample size was 164 subjects in THCM and 152 in CM. Treatment arms did not differ significantly for the primary composite outcome (45% for THCM vs 40% for CM, relative risk 1.18; P = .22). Telehealth alerts prompted frequent telephonic contact, increasing THCM case managers’ workload. The participant population overall had 42% fewer inpatient days during the intervention period compared with the previous year.

 Despite effective implementation of an Internet-based telehealth intervention in an elderly HF population, there was no discernible impact on overall morbidity or mortality. Notably, limited postrandomization participation decreased power to detect a difference. The reduction in days hospitalized for trial participants overall underscores the benefit of CM for such HF patients.

	Medicare Advantage members with heart failure and recent hospitalization were randomized to home telemonitoring with case management or case management alone.

		In 6 months, there was no significant difference in rates of hospital admission, emergency department visits, or death between the 2 groups.

		Overall, telemonitoring increased the case manager activity required per participant, and participant satisfaction and adherence were high.

		A 42% reduction in inpatient days during the intervention period compared with the previous year suggested positive overall impact of case management.

Heart failure (HF) is a major public health problem in the United States. It represents the most common cause for hospitalization among elderly Americans1 and was the nation’s most expensive medical condition to treat in 2005, costing $76 billion. While recent advances in medical treatments for HF have improved patient outcomes, overall morbidity and mortality remain high,

 with approximately 30% to 40% of patients requiring readmission within 6 months of hospitalization.

We conducted a randomized controlled clinical trial to explore the impact of supplementing nurse case management (CM) with the Intel Health Guide System (HGS) in an elderly population with HF and a high risk for rehospitalization. The goal was to assess impact on outcomes and quality of life, and also to assess the feasibility of utilizing an Internet-based telemonitoring technology to facilitate CM in an elderly HF population with multiple comorbidities.

The study design was a prospective, randomized, active controlled study, in which members were allocated to the telehealth system with CM (THCM) or to CM alone.

Aetna Medicare Advantage members with medical and pharmacy benefits were identified through analysis of claims and eligibility information. Criteria for invitation to the study were residence in New Jersey, New York, or Pennsylvania; medical claims for chronic HF in medical claims in the past 3 years; inpatient admission or 2 or more emergency department (ED) visits for any cause within the 6 months before identification; and high risk for a subsequent admission or ED visit based on the health plan’s proprietary predictive model.

Exclusion criteria included claims or utilization records indicating a terminal condition, end-stage renal disease, dementia, use of a ventricular assist device, active listing for cardiac transplantation, completed heart transplant, or requirement for chronic or ongoing intravenous HF medication. Members managed in the health plan’s CM program within the 30 days prior to randomization were also excluded.

Recruitment, consent procedures, and study design were approved by Western Institutional Review Board (Olympia, WA, Protocol 20081787). Oral and written consent to telehealth and the study were sought after randomization for THCM. For those assigned to CM, consent was implied by their oral agreement to participate in CM. Participation began for both groups as telehealth systems were deployed between December 2008 and June 2009.

All participants in both arms of the study received Aetna Medicare’s current standard CM services, which are designed to meet the frequently complex needs of older adults. These services began upon enrollment. For CM, case closure occurred when member goals were deemed met. For THCM, CM continued throughout the monitoring period. Thus, in some cases THCM members received CM longer than members in CM alone. Nonparticipants among both randomized groups continued to be eligible for identification and outreach for CM. Case managers were not directly responsible for patient care, but worked closely with the members, physicians, subspecialty providers,  and others to facilitate care to manage the complex cardiac, oncologic, psychiatric, social, other medical, and/or end-of-life needs of the participants. Identified needs were addressed on scheduled calls with the member (typically 2-3 times per week initially, with frequency decreasing as issues resolved). Common issues covered included health education, safety and emergency measures, medication regimen, and care coordination needs.

Telehealth Monitoring With Case Management Group

In addition to the CM intervention outlined above, the THCM subjects received home monitoring of weight and blood pressure. They also answered questions about adherence and their health regimen through their use of the HGS.

The Intel HGS is a US Food and Drug Administration—approved system comprising a touch screen, a remote patient management unit placed in the member’s home, and a case manager user interface accessed via secure Internet link through a broadband connection. A wired blood pressure monitor and a wireless weight scale transmitted patient data to the CM team. The system was also programmed to ask members about their health status, activities, and medication adherence, and to offer and show educational videos. HGS telemonitoring sessions were administered weekdays at a time the participant selected, according to the member-customized rotocol the case manager had entered. Alerts to the case manager occurred when measurements or response violated threshold values set in alignment with the Seventh report of the Joint National Committee on Prevention, Detection, Evaluation, and Treatment of High Blood Pressure blood pressure guidelines,

 and the physician’s care plan. Participants in THCM transmitted their weight and blood pressure as often as every weekday. Periodically, they were asked to input responses about medication adherence, exercise, and any recent ED visits or hospital stays. The HGS-mediated data flow between the participants and case managers catalyzed frequent CM interactions between the participant, his or her case manager, and the participant’s physician. Educational content was also available to participants on demand  through the HGS. Case managers were instructed to check the Web-based clinical user interface for alerts on each member’s condition and to respond with care coordination assistance as needed. Case managers modified members’ care protocols and thresholds as needed to follow the protocol and the physician’s care plans.

Outcome Measures and Statistical Analyses

The primary outcome measure was admission to an acute care hospital, an ED visit, or death during the measurement period, as determined through claims and Medicare eligibility data.

Secondary variables reflecting healthcare utilization were also measured using claims data. Measures included total hospital inpatient days, ED visits, primary care visits, specialist visits, prescriptions filled for selected types of cardiac medication, and the subset of hospital admission and ED visits for cardiovascular cause. Acute hospital inpatient days per member per year for both groups during the intervention period were also compared with the groups’ rate of hospital days in the year leading up to their randomization date.

Participant experience was measured telephonically before and after the intervention period, using the Quality Control Inc SF-12 Health Survey (SF-12) (http://www.sf-36.org/tools/sf12.shtml). For THCM, a member experience and satisfaction survey (5-point Likert scale with responses ranging from strongly disagree to strongly agree) was administered directly via HGS at the end of the study. Descriptive measures included the proportion of telehealth participants completing the planned 6 months’ monitoring and program adherence. Adherence was measured as proportion of the scheduled weekday protocol-based telehealth sessions completed.

Based on a proprietary predictive model showing an anticipated event rate of 49% in 6 months for the control arm, we calculated that a sample size of 138  patients per arm would provide 80% power to detect a statistically significant decrease of 30% in the primary composite outcome with THCM. Calculations were conducted  using SAS version 9.1.3 (SAS Institute Inc, Cary, NC) using an alpha level of .05 for a 1-tailed test of significance.

 Baseline characteristics were expressed as means and percentages, and compared between the 2 groups using 

2 tests and independent t tests. The analysis of the primary outcome was initially conducted according to intention-to-treat for the randomized population. However, since only a fraction of the randomized population ultimately enrolled in the study, the data are presented for the as-treated cohort to better assess the effect of the intervention as utilized in our population. These as-treated analyses compared outcomes for THCM members who consented and had HGS installed with those of members who began the CM intervention during the same time period.

The odds ratio for the primary end point was assessed using logistic regression models adjusted for baseline differences between the 2 groups that persisted after randomization and subsequent dropouts. The multivariate regression analyses also included numerous covariates to control for any differences between the groups, including age, state of residence, presence of various comorbid conditions, and prior cardiac events including coronary artery bypass surgery. Date of randomization was included as a covariate to account for temporal differences. Sensitivity analyses were performed on secondary outcomes capping outliers’ results at the 99th percentile;  since no difference was seen, only the raw data are shown. Poisson regression analyses using generalized estimating equations were used to compare rates of events (number of hospital admissions, number of ED visits, number of acute inpatient hospital days).

Baseline Demographic, Clinical, and Utilization Characteristics

We initially identified 2314 members who met the study eligibility criteria. A total of 114 were excluded before randomization due to loss of health plan coverage or missing contact information. The remaining 2200 were randomized to THCM or CM. In the interval between randomization and study enrollment, there was considerable attrition in  the study cohort. The 

 details the specific reasons for nonparticipation among those initially randomized, such as CM for anotherreason, lack of desire to participate, or lack of success reaching members. Ultimately, 164 members consented and received THCM, and 152 were placed into the CM arm of the study.

The randomized groups were similar in demographics, prior healthcare utilization, and clinical conditions. Average age was 78.1 years, and the average individual had  numerous comorbid conditions. The treated groups included a higher proportion of black/African American members than the nonparticipating group (24.4% of THCM and 20.4% of CM participants vs 16% of nonparticipants; relative risk 1.5 and 1.3, respectively, P <.05), but the groups were statistically similar with respect to other demographic and clinical aspects. Detailed baseline characteristics for the treated groups are shown in

. Treated members in the THCM arm of the study were younger than those  in CM (75.8 vs 77.7 years, P = .015), were more likely to have had coronary bypass surgery (11.1% vs 3.9%, P = .019), and were more likely to be taking angiotensin receptor antagonists (31.1% vs 20.4%, P = .03). The burden of comorbid medical illness was high and was similar between groups, with a median of at least 5 comorbid conditions per person.

Process Results: Uptake, Adherence, Alert Activity, and Interventions

Eighty-five percent of the THCM group (140 participants) who had an HGS installed continued to use it for the entire 6-month monitoring period. A total of 24 (15%) discontinued use sooner: 

 expressed frustration with or disinterest in using the device; 3 cited poor health; 4 died; and 5 left the health plan. More than 15,000 device sessions were logged among those assigned to THCM, reflecting a mean of 180 sessions per member over the 6-month study duration. All data were transmitted to the members’ case managers for evaluation and disposition. Average adherence to requested daily biomonitoring sessions was 74%, and only 7 members had measured adherence lower than 25%. Beyond the assigned daily sessions, there were many ad hoc measurements as participants elected to use the HGS to measure weight or blood pressure or review educational materials.

During the 6-month monitoring period, the 164 THCM members generated a total of 29,612 threshold violations. On average, there were more than 50 alerts per  member during the first month of monitoring and 27 to 30 alerts permember per month thereafter. Blood pressure alerts were the most common threshold violation (35% of the total). Weight thresholds and health assessment responses triggered 24% and 28% of the alerts, respectively. Thirteen percent were false-positives generated by failed peripheral device readings or transmissions. The alerts gave immediate feedback to the member through the device and also were viewed and managed by the case managers through the Web-based portal. Case managers facilitated 324 medication adjustments for 90 members in the THCM group. Telehealth events increased the CM resources required to support THCM. The average number of case manager contacts with each member was 9.4 per month over a 5-month period for THCM, compared with 3.5 per month over a 2-month period for the CM group.

The primary study outcomes are summarized in 

. For the primary outcome of hospitalization, ED visit, or death, no significant difference was found between the THCM and CM groups, whether analyzed as intention-to-treat or as-treated. Of the 12 deaths, 8 were from a non-HF cause and/or occurred in the context of hospice care. The death rate during the 6-month intervention was under 4% for each group.

 summarizes the secondary outcome results. The THCM and CM groups did not differ significantly in measures of hospital days, cardiovascular admissions, primary care visits, and annualized months of prescriptions filled for diuretics or angiotensin-converting enzyme inhibitors. Carndiology visits and angiotensin receptor antagonist prescription purchases were more frequent for THCM than for CM members (P <.05). Although the SF-12 physical and mental component scores were numerically lower for THCM than for CM participants at the start of the study (NS), the change in these scores from the beginning to the end of the study did not differ significantly between the groups.

An Internet-based telehealth intervention for elderly heart failure patients found no discernible incremental impact on morbidity or mortality compared with case management alone.

Claire M. Spettell, PhD

Articles

Leveraging Remote Behavioral Health Interventions to Improve Medical Outcomes and Reduce Costs

Quality of Care at Retail Clinics for 3 Common Conditions

Telemonitoring With Case Management for Seniors With Heart Failure