To determine whether early symptom improvement with adjunctive aripiprazole in major depressive disorder (MDD) predicts overall symptom remission.

Post hoc pooled analysis of 3 randomized, double-blind studies evaluating efficacy, safety, and tolerability of adjunctive aripiprazole or placebo with standard antidepressant therapy (ADT) in inadequate responders to a prospective 8-week ADT and at least 1 historical ADT.

 A multivariate logistic regression model was developed to determine factors predicting remission most strongly at the end point. Remission was defined as a Montgomery-Åsberg Depression Rating Scale (MADRS) total score of 10 or less at end point.

Early improvement in depression symptoms was the most significant predictor of remission. In adjunctive aripiprazole and placebo groups, improvement of 20% or more in MADRS total score (week 2) was a significant predictor of remission. At week 2, high sensitivity and high negative predictive values (NPVs) were reported for remission in both treatment arms. In the adjunctive aripiprazole arm, early improvement predicted later MADRS remission with high sensitivity (88.0%) and a high NPV (91.5%). Positive predictive value was moderate in both the adjunctive aripiprazole (45.4%) and placebo (37.5%) arms; specificity was 55.0% with adjunctive aripiprazole and 71.5% with placebo.

 Week 2 was a clinically meaningful time point to identify early improvers, and lack of improvement early in treatment was a highly significant predictor of lack of later remission. Early assessment of changes in symptoms could prove useful in clinical practice and more appropriately target healthcare costs.

	Because the economic cost of caring for patients with major depressive disorder is high, early identification of treatment responders and nonresponders may help clinicians make more timely decisions for treatment adjustment during the course of treatment.

		 Adjunctive aripiprazole provides early (week 2) improvement in depression symptoms in a significantly greater proportion of patients than adjunctive placebo.

		 Early improvement was shown to be a significant predictor of later remission; lack of improvement was associated with later nonremission.

		 Objective rating of depression symptoms using a validated scale, especially within 2 weeks of a treatment change, should become routine in follow-up care.

	It is estimated that antidepressants are 50% to 70% effective for major depressive disorder (MDD), and their delayed onset of effect has been identified as a treatment challenge.

Historically, the delay in onset of efficacy for traditional antidepressants has been thought to be as long as 4 to 6 weeks, although recent work has challenged this hypothesis.

 The latency in the onset of therapeutic effect potentially prolongs the impairments associated with depression and leaves a patient more vulnerable to an increased risk of suicide and likelihood of prematurely discontinuing treatment.

Indeed, it is estimated that only 25% to 50% of patients continue with their antidepressant regimen for the length of time recommended by guidelines.

 This is an important consideration, as the economic cost of depression is significant and inadequately treated depression costs even more,

 a price further increased by latency in the onset of antidepressant therapy (ADT).

	Personalizing patient care and optimizing outcomes are becoming increasingly important; both patients and physicians will benefit economically in terms of cost savings and in terms of wellness and quality of life.

 With the need to optimize patient care comes the desire to know early in the course of treatment whether a patient will eventually respond to treatment, and evidence is emerging that early response to antidepressant medication can predict later response/remission with high sensitivity.

Clinical markers of efficacy can aid in decision making about a patient’s treatment program but can also shift the risk-benefit profile of a treatment strategy for MDD. For example, if early onset of efficacy is of prognostic value for later response/remission, this may encourage patients to work through any tolerability issues, and patients who experience early improvement may be more compliant with their medication.

Moreover, patients who do not respond to a given therapy over a sufficient period of time will likely require treatment adjustments, including dose increases or medication switches. Early understanding of when to stay the course and when to adjust treatment will assist in clinical practice.

	Current practice guidelines recommend that antidepressant efficacy should be assessed after as little as 2 weeks to determine the level of response, and measurement-based tools (such as the Clinical Global Impression scale) should be used to do so.

 A recent review of consensus guidelines defines the goals of treating depression as (1) achievement of symptomatic remission and (2) achievement of full functional recovery.

 However, it is known that up to 70% of patients with MDD do not achieve remission after an adequate course of at least 1 ADT.

 Importantly, patients who respond to treatment have lower associated costs per patient and a higher quality of life than nonresponders.

 Thus, if after 2 to 4 weeks a patient with MDD has not responded to treatment, or achieved only a partial response to treatment, a change in dose, a switch to a new drug, or augmentation therapy may be warranted.

	Among the augmentation options, evidence from large, well-controlled clinical trials supports the use of atypical antipsychotics to augment antidepressant monotherapy in patients who have unresolved symptoms. The efficacy and tolerability of aripiprazole as adjunctive therapy to ADT have been demonstrated in 3 large, randomized, double-blind, placebo-controlled trials.

 However, the predictive value of early symptom assessment in patients with MDD receiving adjunctive medication to standard ADT has not previously been evaluated. Here, we present a post hoc analysis of pooled data from 3 studies of adjunctive aripiprazole to evaluate the predictive value of early symptom assessment in patients with MDD who had an inadequate response to standard ADT.

	This analysis used pooled data from 3 identically designed, similarly sized, multicenter, randomized, double-blind studies conducted in the United States evaluating the efficacy, safety, and tolerability of aripiprazole versus placebo as adjunctive treatment to standard ADT in patients who had an inadequate response to a prospective 8-week trial of the same antidepressant agent and at least 1 historical ADT.

	Study participants eligible for enrollment in any of the 3 studies met 

Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition, Text Revision

) criteria for a major depressive episode

 that had lasted 8 weeks or more prior to inclusion without an adequate response. The Mini- International Neuropsychiatric Interview (MINI)

 was conducted during the screening phase to confirm the patient’s diagnosis of MDD and to rule out exclusionary comorbid psychiatric diagnoses. An inadequate response to antidepressant treatment was defined as a less than 50% reduction in depressive symptoms severity, as assessed by the Massachusetts General Hospital Antidepressant Treatment Response Questionnaire, within the current depressive episode. Patients were characterized by having an inadequate response to at least 1, but no more than 3, adequate antidepressant trials (duration of >6 weeks at an adequate dose as specified in the Antidepressant Treatment Response Questionnaire).

	Eligible patients with MDD (17-item Hamilton Depression Scale [HAM-D17] total score 

18) entered an 8-week prospective ADT phase and a 6-week randomized phase. During the prospective treatment phase, patients received 8 weeks of therapy with escitalopram (10 or 20 mg/day), fluoxetine (20 or 40 mg/day), paroxetine controlled release (37.5 or 50 mg/ day), sertraline (100 or 150 mg/day), or venlafaxine extended release (150 or 225 mg/day), per investigator choice, and a single-blind, adjunctive placebo. Patients with an inadequate response (<50% reduction in HAM-D17 total score; HAMD17 

14 and Clinical Global Impression Improvement score of

3 at the end of the ADT phase) were randomized 1:1 to continue the same ADT (no dose adjustment was permitted) plus either adjunctive placebo or aripiprazole (2-20 mg/day) for 6 weeks; patients in the third study were also required to have a Clinical Global Impression Improvement score of 3 or more at study week 6. The pooled demographic characteristics of the population across the 3 studies have been published previously

; patients were, on average, approximately 45 years of age, and mostly female and white.

	The primary efficacy end point used in these studies was the mean change from end of the prospective treatment phase (baseline) to the end of the randomized, double-blind phase in the clinician-rated Montgomery-Åsberg Depression Rating Scale (MADRS) total score.

 The subject-rated Inventory of Depressive Symptomatology (IDS) was included as a secondary efficacy outcome.

	This post hoc analysis used pooled data from these 3 studies to evaluate the predictive value of early improvement with adjunctive aripiprazole to predict overall symptom remission. Early MADRS improvement was defined as a 20% or more reduction from baseline in MADRS total score at week 2 (observed case data) of the randomized, double-blind adjunctive treatment phase. Data at week 3 were also analyzed. Early IDS improvement was defined as a 20% or more reduction from baseline in IDS total score at week 2 of the randomized, double-blind, adjunctive treatment phase. Remission was defined as an MADRS total score of 10 or less at the end point.

	Both univariate and multivariate logistic regression models were developed to determine which factors predicted remission most strongly at the study end point. The following variables were included in both models: sex, baseline MADRS score (median split), body weight (median split), duration of current episode (median split), antidepressant (serotonin-norepinephrine reuptake inhibitors vs selective serotonin reuptake inhibitors), number of prior ADT failures, and time and early improvement group for MADRS and IDS (univariate only). Separate multivariate models were developed for early improvement at week 2 and week 3; the univariate analysis also included treatment (adjunctive aripiprazole/placebo) and study as additional variables. The variables were selected based on prior knowledge of previous studies that explored predictors of response in patients with depression.

shows how sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) were calculated. Analyses were conducted using the lastobservation- carried-forward method on the efficacy sample, which comprised all patients randomized to treatment who received at least 1 dose of double-blind study medication and had at least 1 post-randomization efficacy evaluation. Receiver operating characteristic (ROC) curves with area under the curve (AUC) values were also calculated for response and remission at end point using early improvement at weeks 2 and 3 as predictors.

Subject Disposition and Baseline Demographics

	In total, 1092 patients had an inadequate response following completion of the 8-week prospective treatment of ADT and were randomized to double-blind treatment with adjunctive aripiprazole (n = 550) or adjunctive placebo (n= 538) (4 randomly assigned patients did not receive doubleblind study medication).

The completion rate was high in both treatment groups (adjunctive aripiprazole, 85.3%; adjunctive placebo, 87.2%) and discontinuation due to adverse events was low in both groups (adjunctive aripiprazole, 4.4%; adjunctive placebo, 1.7%). 

 shows the baseline demographic characteristics, psychiatric history, and disease severity of patients who were early MADRS improvers and those who were early MADRS nonimprovers.

	The mean adjunctive aripiprazole dose across all ADTs on the last day of treatment was 11.0 (2-20) mg/day, and was similar between ADT groups, although no formal statistical analysis was performed (escitalopram 11.2 mg/day; fluoxetine 9.9 mg/day; paroxetine 9.8 mg/day; sertraline 12.2 mg/day; venlafaxine extended release 10.8 mg/day). Among early improvers, as defined at week 2, the mean (standard deviation [SD]) dose of aripiprazole was 7.6 (2.8) mg/day at week 2 and 9.6 (4.9) mg/day at week 6. Among week 2 nonimprovers, the mean (SD) dose of aripiprazole was 8.2 (2.5) mg/day at week 2 and 11.7 (6.2) mg/day at week 6.

	At the end point, the mean reduction in MADRS total score was significantly greater in patients receiving adjunctive aripiprazole than in patients receiving placebo (−9.7 vs −6.6; P <.001). Improvement (

20% reduction in MADRS total score) was seen in significantly more patients in the adjunctive aripiprazole treatment group than the adjunctive placebo group at all time points (

). Remission was achieved by significantly more patients receiving adjunctive aripiprazole than by patients receiving placebo from week 2 onward (

	Of the demographic and clinical factors assessed as potential predictors of remission, an early improvement in depression symptoms (

20% in MADRS total score), using week 2 criteria, was a significant predictor of remission with aripiprazole in the multivariate analysis (

). Furthermore, more patients who responded at week 2 achieved remission at week 6 in the aripiprazole group (65.7%) compared with the placebo group (34.3%). A lower baseline (ie, less ill patients) MADRS score (less than or equal to the median vs greater than the median) and IDS score were also significant predictors of remissionin the week 2 multivariate model (Figure 2). Similar results were observed using the week 3 criteria for early improvement: early improvement in depression symptoms (

20% in MADRS total score) (odds ratio [OR] 18.9; 95% confidence interval [CI] 8.0-44.7), lower baseline MADRS score (OR 2.4; 95% CI 1.5- 3.9), and IDS score (OR 1.7; 95% CI 1.1-2.8) were significant predictors of remission.

	The univariate analysis produced similar results; patients with an improvement of 20% or more in MADRS total score at week 2 were 10 times more likely to achieve remission, and patients with a 20% or more improvement in IDS score at week 2 were approximately 5 times more likely to achieve remission, than those without this improvement.

	The predictive value of nonimprovement at week 2 for later MADRS remission is shown in 

, where NPVs exceed 90%. The PPV was moderate in both the adjunctive aripiprazole (45.4%) and adjunctive placebo (37.5%) arms, and specificity was 55.0% with adjunctive aripiprazole and 71.5% with adjunctive placebo. The predictive value of early improvement at week 3 for MADRS remission was similar to that at week 2, thus making week 2 an early and clinically useful time point for assessment. At week 3, early MADRS improvement predicted later MADRS remission with high sensitivity (96.0%) and a high NPV (96.2%) (data not shown).

	Predictive response was also assessed in patients who were deemed to be early improvers or nonimprovers based on the patient-rated IDS total score at week 2. In the adjunctive aripiprazole and placebo arms, early IDS improvement predicted later MADRS remission with moderate sensitivity (63.8% and 61.7%, respectively). Slightly higher NPV was reported for remission in the adjunctive placebo arm (91.3%) than in the adjunctive aripiprazole arm (82.7%). The PPV was moderate in both the adjunctive aripiprazole (50.8%) and adjunctive placebo (39.7%) arms, and specificity was 73.7% with adjunctive aripiprazole and 81.1% with adjunctive placebo.

	The ROC curves for early improvement at week 2 and the AUC for early improvement at week 2 predicting later remission with adjunctive aripiprazole or placebo were both 0.81. The AUCs for early improvement on the clinician-rated MADRS scale at week 2 predicting later response with adjunctive aripiprazole was 0.81, whereas with adjunctive placebo it was 0.82. The AUC for early improvement at week 3 predicting later response in patients receiving adjunctive aripiprazole was 0.85, whereas with adjunctive placebo it was 0.86. The AUC for early improvement at week 3 predicting later remission in patients receiving adjunctive aripiprazole was 0.83; with adjunctive placebo it was 0.85. A comparison of the area under the ROC curves of weeks 2 and 3 showed a statistically significant difference infavor of week 3; however, themagnitude of the difference in the areas makes this finding of low clinical significance.

	This study demonstrates that among patients with MDD who receive adjunctive aripiprazole as augmentation to ADT, week 2 is a clinically meaningful time point to identify early improvers (

20% decrease in MADRS total score from baseline), who were nearly 10 times more likely to achieve remission. Based on a high NPV for remission, failure to improve early with adjunctive treatment also was predictive of failure to improve later in treatment. Thus, the data strongly support the value of early assessment to inform clinical decisions in the treatment of MDD. Clinicians should reassess a patient’s condition within 2 weeks of starting a new ADT regimen and consider adjusting the overall treatment regimen (eg, further dose titration of ADT and/or augmenting agent) or switching treatment in patients who do not show some improvement. A heightened level of follow-up and vigilance is also warranted. Similar PPVs and NPVs were observed in the adjunctive aripiprazole and adjunctive placebo groups. These data suggest that the NPV can detect patients who do not improve at week 2 and predict later nonresponse if the same treatment is continued to week 6, irrespective of the type of therapy they are receiving.

	The findings reported here with adjunctive treatment are similar to those in antidepressant monotherapy studies that used the criterion of a 20% score reduction as an early indicator of improvement rated by the HAM-D17 rather than MADRS scale.

 The 20% or more reduction in symptom score is often chosen, as it can be reliably measured and translates into clinically meaningful change in the severity of depression.

Monotherapy studies with mirtazapine and paroxetine have shown that early improvement (

20% reduction in HAM-D17) predicted stable response (

50% reduction in HAM-D17 at week 4 or 6) and stable remission (HAM-D17 score of 

20% reduction in HAM-D17) during the course of treatment with a range of ADTs with different pharmacologic profiles is also useful in predicting stable response.

 Nonresponse (<20% in HAM-D17 score) to fluoxetine as early as week 2 has been shown to predict a poor 8-week outcome.

 Additionally, the absence of early improvement (

25% reduction in MADRS total score) with fluoxetine/olanzapine combination treatment was nonresponse or nonremission in patients with treatment- resistant depression.

 Similarly, in a pooled analysis of 10 randomized, doubleblind, placebo-controlled trials, the absence of early improvement appeared to be a highly reliable predictor of eventual nonresponse and nonremission (NPV of 74% and 82%, respectively).

 The value of a 2-week assessment is underscored by the fact that patients who achieved an early 20% improvement in their MADRS scores were nearly 10 times more likely to achieve remission. Perhaps more striking, the vast majority of patients (>90%) who had not experienced an early 20% symptomatic improvement by week 2 did not achieve remission. Taken together, regardless of current treatment, the data support the clinical value of early assessment using either the clinician-rated MADRS or the patient-rated IDS scale.

	Interestingly, the early improvement in symptomatology as a predictor of remission was observed as a function of the change in both MADRS and IDS scores, which are clinician-and patient-rated scales, respectively. This observation is in agreement with data from Reimherr and colleagues, who recently reported a cross-correlation analysis that also found that improvements in depressive symptomatology could be reliably identified by clinicians and patients alike.

Articles

CLINICAL

In treatment of depression with adjunctive aripiprazole, early improvement (week 2) was a significant predictor of remission, and lack of improvement was associated with nonremission.

Adjunctive Aripiprazole for Depression: Predictive Value of Early Assessment

 To evaluate the cost impact of an integrative medicine intervention on an inpatient oncology service.

 This study used nonrandomized, nonequivalent groups. A baseline sample of inpatient oncology patients at Beth Israel Medical Center admitted to the medical oncology unit before implementation of the Urban Zen Initiative were compared with patients admitted after the Urban Zen Initiative was in place.

 The Urban Zen Initiative incorporated yoga therapy, holistic nursing techniques, and a “healing environment” into routine inpatient oncology care. Length of stay and medication use data were extracted from Beth Israel’s decision support electronic database. We compared length of stay, total medication costs, and costs of as-needed medications for both groups: the baseline sample of inpatient oncology patients and patients exposed to the Urban Zen healing environment initiative.

We had complete cost data on 85 patients in our baseline group and 72 in our intervention group. We found no difference in length of stay between the 2 groups. We found a significant decrease in use of antiemetic, anxiolytic, and hypnotic medication costs as well as a decrease in total medication costs in the Urban Zen sample compared with the baseline group.

 An integrative medicine approach including yoga therapy, holistic nursing, and a healing environment in the inpatient setting can decrease use of medications, resulting in substantial cost savings for hospitals in the care of oncology patients.

	An integrative medicine approach incorporating yoga, holistic nursing, and a “healing environment” added to the inpatient care of oncology patients can significantly reduce hospital costs.

		 Significant cost savings were realized on as-needed medications including antiemetics and anxiolytics in intervention patients compared with controls.

		 Overall medication costs were reduced as well, resulting in an overall cost savings of $156 per hospital day for patients in the intervention group.

		There was no difference in length of stay between the 2 groups.

	The Urban Zen (UZ) Initiative at Beth Israel Medical Center, New York, was a pilot project evaluating the impact of a multifaceted “optimal healing environment”

intervention—incorporating yoga therapy, holistic nursing practices, a patient navigator, and a renovated physical environment&mdash;on quality of life and cost outcomes for inpatients on a medical oncology floor. Our findings regarding quality of life and patient experience, which have been published elsewhere,

 showed significant decreases in anxiety, fatigue, and depression in the intervention group compared with controls, as well as improved energy and decreased pain. In this article, we present a summary of our findings on cost outcomes.

	Although a fair amount of research has been published on the impact of integrative/complementary medicine interventions on costs of care in the workplace and outpatient settings, to date very little research exists on the question of potential cost savings in the inpatient setting.

 What research exists has primarily focused on cost savings in the perioperative setting. For example, Montgomery et al found that a brief hypnosis intervention produced a substantial cost savings in patients undergoing excisional breast biopsy.

 The savings were primarily from reduced medication use and consequently reduced operating room and recovery time per patient; the authors hypothesized that this may have been due to reduced anxiety induced by the hypnosis intervention. Other studies have looked at the impact of nutritional supplementation, music therapy, and mind-body interventions on length of stay following surgical procedures and have found a modest benefit.

	Because implementation of this type of intervention in the inpatient setting does require potentially substantial initial investment by the hospital, and because maintaining these types of services does entail additional ongoing cost, it is critical to generate data regarding the potential cost savings that can result from this type of approach. To date such data are not available, especially regarding medical rather than surgical admissions; this study sought to determine what, if any, potential for cost savings could result from an integrative, “healing-oriented” intervention on an inpatient medical unit.

	The clinical intervention which made up the UZ Initiative took place on 9 Dazian, a 24-bed teaching service at Beth Israel Medical Center delivering the full spectrum of inpatient medical oncology care, and included 5 components:

	• Conversion of the patient lounge into a healing sanctuary for meditation, yoga practice, and quiet visiting.

	• Lighting and color changes on the floor that deemphasized standard hospital ambience.

	• Renovated staff lounge to provide a quiet space for the staff.

	2. Holistic nursing training for the nursing staff. These skills were imparted during a required series of workshops attended by the entire nursing staff on 9 Dazian, including nursing aides as well as staff nurses and nurse managers.

	• Relaxation therapies including breathwork and healing touch skills.

	• Imagery and visualization techniques, specifically to be used to address symptoms of pain, anxiety, insomnia, and nausea.

	• Aromatherapy protocol using specific essential oils such as lavender and ylang-ylang to address common inpatient complaints including insomnia and anxiety; nurses were taught a protocol for administering aromatherapy via hand massage, which is easily applied to debilitated hospitalized patients.

	3. Yoga therapists on the unit to work with patients using breathing and yoga techniques. These were registered yoga teachers, generally with at least 5 years of yoga experience, who were interested in gaining more experience in the hospital setting.

	• Yoga practitioners were specifically trained in restorative yoga techniques developed through the UZ Integrative Therapist Training Program, a 500-hour interdisciplinary course designed to prepare yoga therapists for care of hospitalized patients.

	• Yoga protocol was specifically adapted for use in the hospital setting and was designed to address common symptoms including pain, anxiety, nausea, insomnia, and constipation.

	• Other than the full-time yoga coordinator, the therapists worked on a volunteer basis.

	4. “Patient navigator” position to facilitate a smooth and efficient process of care.

	• This full-time staff position, available Monday-Friday, was fully integrated with nursing functions on the unit; the navigator typically has either a social work or a health education background.

	• Navigator greets the patient on arrival on the unit and then visits each patient 1 to 2 times daily to assess for any issues regarding the process of care that might need attention, as well as to assist with general emotional support.

	5. Audiovisual materials demonstrating yoga and relaxation techniques for patients to use in bed.

	Every patient admitted to the unit once the intervention was in place was offered access to all of these services, regardless of their willingness to participate in our study. As part of the admission process, the admitting nurse described the services to each patient, as did the patient navigator. Referral from the physician was not required for patients to access the UZ services; however, a series of educational experiential sessions were held for physician staff on the unit, in which they were exposed to the yoga and holistic nursing techniques used in the protocol. The goal of this training was to ensure that physicians were comfortable with the therapies to be offered to their patients through the UZ project.

	Because our funder required that all patients admitted to the unit be offered the UZ services, and because there is no comparable medical oncology unit at Beth Israel that could have served as a control, we chose a nonrandomized, “before/after” design for this study. Using a quasi-experimental nonequivalent- groups design, we compared a control group consisting of patients admitted to 9 Dazian prior to the implementation of the UZ intervention (baseline/control group) with a treatment group consisting of patients who were admitted to 9 Dazian after the UZ intervention was implemented. Inclusion/exclusion criteria were age 18 to 85 years; admission as an inpatient to Oncology Services on 9 Dazian at Beth Israel Medical Center; Karnofsky score of 60 or better; life expectancy of 6 months or more; and English speaking.

	The goal of the component of the project reported here was to examine whether this type of intervention can shorten length of stay and reduce medication costs. Medication cost data were extracted from the hospital’s decision support database, which contains both cost and charge data for all hospitalizations.

	All descriptive data are presented as mean ± standard deviation in the case of normally distributed variables (eg, age), median (minimum, maximum) in the case of skewed variables (eg, length of stay), and frequency (percentage) in the case of categorical variables (eg, sex). In order to correct for skewness, medication cost data were normalized using a log transform prior to significance testing, as recommended by Bland and Altman.

 Mean ± standard deviation costs in terms of actual dollars were computed using the antilog of the mean and standard deviation based on the log transformed values. Simple comparisons between the groups were carried out using 

 tests in the case of normally distributed variables and 

 tests in the case of categorical variables. Interactions between treatment group and demographic factors in terms of cost were tested for significance using analysis of variance. All quantitative analyses were done using SAS version 9.1 (SAS, Inc, Cary, North Carolina).

	A total of 89 patients were originally enrolled in the baseline sample, and 74 patients were enrolled in the UZ sample. Length-of-stay data were available for all patients, while complete medication cost data were available for 85 patients in the control group and 72 in the intervention group. Overall, there were no significant differences between the 2 groups in age, sex, other demographic factors, type of cancer, or reason for admission (

	The overall length of stay was virtually the same in both groups, with a median (minimum, maximum) of 3 (1, 34) days in the baseline group and a median (minimum, maximum) of 3 (1, 37) days in the UZ group (

 = .20). The differences between the baseline control group and the UZ intervention group in terms of medication costs are shown in

. Total mean medication costs were significantly higher for the baseline group ($889) than for the UZ group ($420), for a cost savings on average of $469  per patient. In terms of medication categories of specific interest, there were significant differences in favor of the UZ group with regard to antinausea medications (

	Analysis of variance was used to look at possible interactions between background characteristics such as age and sex on one hand and the costs of different types of medication in the baseline group versus the UZ group. No significant differences were found.

	Although patient-centered outcomes such as anxiety, depression, and pain experience are critical in evaluating the impact of an intervention, ultimately the widespread adoption of an integrative medicine approach, especially in the inpatient setting, will require clear evidence that these interventions are cost-effective, or at minimum cost neutral. A small but growing number of studies document the clinical effectiveness of several of the specific interventions that comprised our “optimal healing environment” and their impact on patient care in the inpatient setting. For example, the presence of a patient navigator on an inpatient oncology floor has been shown in at least 2 studies to decrease patient distress and improve quality of life and patient satisfaction.

 Similarly, recent studies of yoga therapy modified for inpatient use have been shown to decrease stress and anxiety levels in pediatric cancer patients

 and “physical environment” interventions

 in the inpatient setting have also been described. However, despite this growing literature very few studies have examined the impact of a multifaceted inpatient intervention such as the UZ project, and even fewer have addressed the issue of cost-effectiveness.

	Length of stay is considered the gold standard marker for cost-effectiveness in the inpatient setting. We were not able to demonstrate an impact of the UZ program on length of stay; this may have been in part because the medical oncology floor where the program was implemented has a relatively short average length of stay compared with many inpatient oncology units. Because the Beth Israel Cancer Center sees a high proportion of head and neck cancers, roughly 50% of the admissions to this floor are specifically for placement of a percutaneous gastric tube and initiation of chemotherapy. These are standardized procedures that generally require a 2- to 3-day stay. Change in symptoms or pain does not generally determine the time of discharge, as it might on a floor where more of the admissions were unscheduled and the diversity of diagnoses greater. We are planning to study a similar protocol on a surgical oncology floor, where length of stay is longer at baseline and where time to discharge is more dependent on individual patient recovery factors, which are potentially more responsive to a UZ type of intervention.

	We were able to demonstrate a significant decrease in medication costs in the UZ group compared with controls, on the order of $469 per patient. Although this decrease may seem modest in light of the overall costs of a given hospital stay, if we extrapolate this savings of approximately $156 per patient per day to a total of 6264 patient days per year (24 beds × 261 days, since services are not available on weekends), the total savings to the hospital would be $977,184 annually. Even if we assume that only half the patients on the floor choose to actively utilize the UZ services—a conservative estimate based on the participation we saw during our study&mdash;the cost savings would amount to $488,592 annually.

	The largest cost of the UZ intervention was the one time expense of physical space renovation, which was approximately $300,000. The other one time costs were for holistic nursing training for the floor staff—which cost approximately $40,000, including missed work time for nursing staff and salary support for the holistic nurse trainers&mdash;and production of audiovisual materials ($15,000). These investments do not have to be repeated, and thus can be amortized over the life of the program. The ongoing costs of the program consist of salary support for the patient navigator ($92,000 annually including fringe benefits) and the yoga coordinator ($117,000 annually including fringe benefits). Thus, the total costs of the UZ program including both one time ($355,000) and ongoing ($209,000 per year) costs were approximately $564,000. Even utilizing the most conservative estimate of cost savings ($488,592/year), by the second year of the program, with the start-up costs included, the hospital realizes a savings of $204,184 based on decreased medication costs. This annual savings rises to $279,592 in the third year, with the initial investment in space and training completed. These ongoing cost savings then continue for the remainder of the life of the program.

An approach including yoga, holistic nursing, and a "healing environment" can decrease medication use, resulting in substantial cost savings in care of inpatient oncology patients.

Cost Savings in Inpatient Oncology Through an Integrative Medicine Approach

 To investigate whether market competition is a potential driver of hospital performance on the key evidence-based Joint Commission heart-failure (HF) quality indicators of angiotensin-converting enzyme inhibitor/angiotensin receptor blocker prescribed, left ventricular function assessment, smoking-cessation counseling, and discharge instructions.

 Hospital performance data for HF was obtained from The Joint Commission’s ORYX program from 2003 to 2006. The performance data were linked with hospital characteristics from the American Hospital Association Annual Survey and area-level sociodemographic information from the Area Resource File. Healthcare markets

	were defined as hospital referral regions (HRRs) and market competition intensity was defined by the Herfindahl-Hirschman Index. Hospital-level and HRR-level ordinary least squares fixed effects regression models were used to estimate the relationship between market competition and performance.

A paired comparison indicated that there was a significant change in the mean hospital-level performance over time on all of the HF quality indicators. From the multivariate analyses, hospitals in the least competitive markets (Quintile 5) performed slightly better (2.9%) than the most competitive markets (Quintile 1) for left ventricular

 <.01). At the HRR level, however, the least competitive markets (Quintile 5) performed moderately worse (5.1%) on the discharge-instructions quality indicator compared with the most competitive markets (Quintile 1) (

Market competition intensity was associated with only small differences in hospital performance. The level of market competitiveness may produce only marginal incremental benefits to inpatient HF care.

	This study empirically examines whether market competition is a potential driver of hospital performance on The Joint Commission heart-failure quality indicators.

		Given the growth of performance measures as a way to stimulate quality improvement efforts, this is a timely and policy-relevant article.

		 This study adds to the existing literature by examining whether quality improvements made by hospitals based on public performance reports are occurring through the mechanism of market competition.

		 We include data from all patients 18 years and older who were treated for heart failure at a Joint Commission—accredited hospital and were eligible for the performance measure.

The significant chasm between the quality of care that heart failure (HF) patients 

 receive has been widely documented. Previous studies have reported on the extensive variation in the treatment and management of HF patients in hospitals across the country.

Because of the substantial geographic practice variations and underuse of appropriate HF therapies, payers and accrediting bodies have begun to measure hospital performance as a way to stimulate quality-improvement efforts. The Joint Commission now requires that hospitals submit information on their performance for the core conditions of heart attacks, HF, pneumonia, and others as part of the ORYX program, which is publicly reported on Quality Check (http://www.qualitycheck.org).

ORYX was first developed by The Joint Commission in 1997 as a way to integrate performance and outcomes measures into a continuous accreditation process.

The ORYX quality indicators are aligned with the Centers for Medicare & Medicaid Services (CMS) Hospital Compare performance measures. Since the ORYX program was initiated, there have been substantial improvements in hospital quality, although there remains a wide variation in performance across individual hospitals and states.

Despite the enormous progress that has been made to narrow the quality gap, the underlying motivation for hospitals to increase their compliance with the standardized quality indicators and the reasons for the wide heterogeneity in performance across hospitals and geographic areas is unclear. Previous research suggests that hospitals might be motivated to act on publicly reported performance data due to market competition, professional standards, and/or to preserve or enhance their reputation.

In this study, we sought to empirically test whether market competition is a potential driver of hospital performance on the key evidence- based Joint Commission HF quality indicators. We focus specifically on HF because it is one of the core conditions that are measured by The Joint Commission and it is a common and costly chronic condition.

Pathways to Performance on Quality Indicators

There are 3 possible pathways through which hospitals might be motivated to act on publicly reported quality indicators.4 Hospitals may be motivated to improve their performance on quality indicators due to market forces because they would like to hold on to or increase their market share of patients. Hospitals’ awareness of quality deficits might also be enough to stimulate quality improvement efforts because of professional standards. Lastly, hospitals may be motivated to improve quality because they are concerned about protecting or enhancing their public image, since consumers may form certain opinions about a hospital.

 In this study, we focus on market competition as a potential driver of hospital performance. We hypothesize that if hospitals compete on quality, then we expect hospitals in competitive markets will provide a higher level of quality and hospitals in concentrated (less competitive) markets will provide a lower level of quality.

The Joint Commission accounts for more than 3000 hospitals that represent approximately 80% of hospitals in the United States and comprises more than 90% of all acutecare hospital beds.

 Quarterly data from The Joint Commission’s ORYX hospital performance measurement program for HF from 2003 to 2006 was used. The 4 evidence-based HF quality indicators examined were (1) angiotensin-converting enzyme inhibitor (ACEI) or angiotensin receptor blocker (ARB) prescribed at discharge, (2) left ventricular (LV) function assessment, (3) smoking-cessation counseling, and (4) discharge instructions. Only yearly data points for the ACEI or ARB quality indicator in 2003 and 2004 were available, with the exception of Quarters 1 and 2 of 2003, since the definition for this quality indicator changed in 2005 to include ARB. A total of 3011 non-Federal, shortstay, Joint Commission—accredited acute-care hospitals were used as the primary units of analysis over 16 quarters (n = 48,176), with a secondary analysis that aggregated hospitals to 306 hospital referral regions (HRRs) over 16 quarters (n = 4896). Patients 18 years and older with LV systolic dysfunction HF defined by

 International Classification of Diseases, Ninth Revision, Clinical Modification 

diagnosis codes 402.01, 402.11, 402.91, 404.01, 404.03, 404.11, 404.13, 404.91, 404.93, and 428.XX, who were admitted to Joint Commission— accredited hospitals and who met the reporting requirements, comprised the data elements for the individual HF quality indicators.

The ORYX data were linked with hospital characteristics data from the American Hospital Association Annual Survey and sociodemographic and market-level characteristics from the Area Resource File to serve as control variables. Market share was derived from the Medicare Provider and Analysis Review (MedPAR) file based on the total number of Medicare HF patients.

As part of our secondary analysis, we aggregated hospitals to the HRR level to mitigate the potential issue of the small number of cases at a particular hospital that may adversely impact a hospital’s performance, as well as the endogeneity problem of patient selection of hospitals based on perceived or other unobservable characteristics that may be correlated with quality of care. A fixed-effects approach was used to examine within unit changes over time so that the results would not be affected by any heterogeneous differences across HRRs that might be constant over time.

We used HRRs to construct the measure of the healthcare market according to the crosswalk methodology of the 

. HRRs are a naturally occurring healthcare market and they represent a geographic area where a significant proportion of medical care is provided by a referral hospital(s) serving an entire region.

A quarterly measure of market competition was constructed based on the Herfindahl-Hirschman Index (HHI). The HHI is a standard measure of market competition and it has been widely applied for hospitals in HRRs. The HHI measure was created using Medicare HF patient volume derived from the MedPAR data file. Hospital market share was calculated as the total number of Medicare HF patients at a hospital divided by the total number of Medicare HF patients within an HRR, and it was scaled by 100.

The HHI was then determined by taking the sum of the square of market shares for Medicare HF patients for all hospitals within an HRR.

The HHI ranges from 0, which represents an infinitesimally small number of competitors in a market, to 10,000, which represents a monopoly. In our analyses, however, we used quintiles of HHI where Quintile 1 represents the most competitive markets and Quintile 5 represents the least competitive markets.

Hospital-level performance on the quality indicators was used in the primary analyses. For the HRR-level analyses, the weighted average of performance based on the total number of eligible HF patients at each hospital for the quality indicators (ACEI or ARB prescribed at discharge, LV function assessment, smoking-cessation counseling, and discharge instructions) was calculated. By using the HRR-level weighted average of hospital performance, hospitals in a market that had a higher volume of HF patients had their performance score weighted more highly than hospitals that had a lower volume of HF patients.

We first examined descriptive statistics on the hospitallevel performance and market intensity data. We then used a paired t test to assess the change in the mean hospital-level performance on the HF quality indicators from Quarter 1 of 2003 to Quarter 4 of 2006. We specified hospital-level ordinary least squares (OLS) fixed-effects regression models with Quintile 1 (most competitive markets) serving as our reference group to estimate the relationship between market competition and performance and account for the time invariant—omitted variables. We lagged market competition by one quarter in our models because we would reasonably expect a delayed response between the time when the HF performance data are made publicly available and the time when the effect of market competition occurs. Additionally, the temporal sequence of the lagged model would make it less likely for an external shock in the future to affect past performance.

A set of time indicator variables for the quarter of the study was included in the regression models, with the first quarter serving as the reference. We controlled for a number of hospital characteristics including: bed size, teaching status, profit status, presence of a cardiac intensive-care unit, percentage Medicare, percentage Medicaid, and number of nurses to 1000 patient-days. Market-level sociodemographic variables of gender, race, age, per capita income, unemployment rate, number of physicians, and health maintenance organization penetration rate were included. We also included in our models the volume of HF patients and total admissions to control for the quality-volume relationship. We adjusted for the hospital market share of patients from the first quarter to control for the potential shifting of market shares over time. The regression models were specified for hospital-level fixed effects and clustering by HRR to take into account the correlation of the HRR-level independent variables. The multivariate analyses also specified for robust standard errors.

We also explored if the rate of adherence to the HF quality indicators varied over time by market competition intensity whereby competitive markets might conform

more quickly to the HF quality indicators because of market forces and competitive pressure. Therefore, a time by lagged market-competition interaction variable was later included in the regression models.

The models were repeated as HRR-level fixed effects to account for the potential endogeneity problem of patient selection of hospitals based on perceived quality of care. In the HRR-level analyses, the market-weighted average of the covariates was included for each HRR.

 shows the change in average hospital performance for the HF quality indicators, and average number of eligible patients by quality indicator per hospital. The largest increase in average hospital performance over the study period was observed for the behavioral measures of smoking-cessation counseling (45%) and discharge instructions (38%), while smaller progress was made on the clinical measures of ACEI/ARB prescribed at discharge and on LV function assessment.

 shows the distribution of markets by the level of market competition intensity at the first and last quarters of the study. The distribution shifted slightly over time as markets became slightly less competitive.

A paired t-test was conducted to examine the changes in average performance over time. The paired comparisons for each of the quality indicators were found to be highly significant (

 <.001), indicating that there was a significant change in the mean hospital-level performance (Table 1).

POLICY

This study empirically examines whether market competition is a potential driver of hospital performance on the key evidence-based Joint Commission heart-failure quality indicators.

Effect of Market Competition on Hospital Performance for Heart Failure

To propose a model in which insurers work with hospitals to provide a discharge supply of antiplatelet medication to patients receiving stents and to examine the cost implications of this strategy.

 A decision tree was modeled using data from previously published research. The study adopts an insurer’s perspective.

Data on patient delays in filling antiplatelet prescriptions and rates of associated adverse events were taken from published research. The costs of adverse events (death or acute myocardial infarction [AMI]) are taken from Healthcare Cost and Utilization Project estimates of hospital costs for diagnosis-related groups associated with AMI.

In the base case, expected costs totaled $1782 when stent implantation patients were provided with a discharge supply of medication and $1857 under the current standard of care, a difference of $75. Insurers can supply up to 60 days of medication without increasing total costs. The strategy of offering a discharge supply of medication is cost saving under a range of estimated rehospitalization costs and medication costs. However, this result is dependent on the ability of a discharge supply of medication to reduce rates of death or AMI.

Providing discharge supplies of antiplatelet medication resulted in lower overall costs for insurers in most of the cases modeled.

	Managed care plans should consider waiving copayments on antiplatelet medications for percutaneous coronary intervention (PCI) patients and collaborating with hospitals to ensure that patients receive a supply of these medications upon discharge.

		 Research has shown that failure to begin antiplatelet therapy promptly after PCI puts patients at increased risk for acute myocardial infarction or death, but many patients delay filling prescriptions for these medications.

		 In most of the scenarios modeled here, the expected costs of these complications to insurers exceed the cost of providing copay-free discharge supplies of medication to PCI patients.

Developments in antiplatelet therapy for patients undergoing cardiac stent implantation have reduced the incidence of thrombosis in patients receiving cardiac stents.

However, thrombosis is still an all-too-common occurrence with grave consequences for patient health. Sixty-four percent of patients with stentrelated thrombosis experience either an acute myocardial infarction (AMI) or death.

 Revascularization is costly also. Patients requiring revascularization within a year of percutaneous coronary intervention (PCI) incur almost $25,000 more in medical expenses than those that do not.

 A great deal of research has been devoted to describing costs and outcomes associated with antiplatelet therapies for stent patients. However, much less information is available on the cost implications of adherence to these therapies.

Payers may have an opportunity to reduce their medical losses by designing benefits that encourage continuity of antiplatelet therapy during the period immediately following hospital discharge. One study of 3 large US managed care organizations found that 16.3% of patients undergoing stent implantation delayed filling their antiplatelet prescription. The median delay between a PCI patient’s discharge from the hospital and the time the patient first fills the antiplatelet prescription was 3 days.

 Past studies of unusually restrictive prior authorization programs in Canada have found median delays in filling antiplatelet prescriptions of 9 days.

 Given antiplatelet medications’ mechanisms of action, it is not surprising that a gap in therapy after hospital discharge is associated with an increased risk of poor outcomes. Even among a population of patients receiving drug-eluting stents, patients with a delay of 1 day or more in filling an antiplatelet prescription after discharge are significantly more likely to experience death or a myocardial infarction.

 Furthermore, higher copayments for medications are associated with an increased risk of rehospitalization in acute coronary syndrome patients receiving stents, as well as higher average follow-up costs of care.

There is reason to believe that the way in which payers choose to administer benefits may impact beneficiaries’ health outcomes as well as the total costs of treatment. If this is the case, payers may have an opportunity to reduce medical losses and improve the quality of cardiac care through changes in benefit design. Insurers could develop partnerships with hospitals to fund hospitals’ provision of discharge supplies of medication to patients receiving cardiac stents. This study examines the cost implications of such a strategy.

A decision model, using a payer’s perspective, was developed to evaluate this strategy. The 2 alternatives are (1) the current state of care, in which patients are responsible for filling their prescriptions for antiplatelet medications and insurers charge a copayment for those medications; and (2) an alternative strategy in which an insurer waives the copayment for a small discharge supply of antiplatelet medication that is provided to a patient by the hospital. The decision tree was modeled using TreeAge Pro software (TreeAge Software Inc, Williamstown, Massachusetts) and is presented in 

. Probabilities of patient adherence (defined here as a patient filling a prescription on the day of discharge) and adverse events (defined here as the occurrence of death or AMI) as well as the costs of adverse events were estimated based on published studies. Model parameters are provided in 

. Sensitivity analysis and Monte Carlo simulation were conducted to evaluate how robust conclusions were to changes in the model assumptions.

In the base case, 7.9% of patients filling a prescription for an antiplatelet medication immediately following discharge were estimated to experience death or AMI compared with 14.2% of patients who had at least a 1-day delay.

Similarly, the base case estimates that 83.65% of patients will fill a prescription without delay.

 This estimate may overstate the true percentage of patients who fill antiplatelet prescriptions upon discharge. Other studies have found results ranging from 64%

 The 83.65% estimate for patient adherence is conservative. If the true rate of adherence is lower than 83.65%, the model’s estimates of the costs associated with standard treatment will underestimate the true cost of the standard treatment. In that case, the savings afforded by the alternative strategy (in which patients are provided discharge medication) will be higher than estimated here. The model assumed that rates of death or AMI under the alternative strategy would be equal to adverse event rates for the group of patients filling prescriptions upon discharge. The model also assumed that under the alternative strategy 7 days of medication is provided by the hospital patient is completely adherent to therapy throughout the rest of the year. This supply is enough to cover the 3- to 4-day average delay in filling reported by some studies. 

 Like other assumptions, this estimate was relaxed in sensitivity analysis.

The model compared the expected cost of rehospitalization for AMI with the higher medication costs insurers would face in supplying patients with discharge supplies of antiplatelet medication without charging copayments. Patients who did not experience death or AMI incurred medication costs for 12 months per accepted guidelines.

Costs related to hospitalizations were estimated to average $8111 based on hospital costs reported by the Healthcare Cost and Utilization Project for diagnosis-related groups (DRGs) 280-285.

An insurance benefit that provides percutaneous coronary intervention patients with a discharge supply of antiplatelet medication could result in cost savings for insurers.

December 2011

Adjunctive Aripiprazole for Depression: Predictive Value of Early Assessment

Cost Savings in Inpatient Oncology Through an Integrative Medicine Approach

Effect of Market Competition on Hospital Performance for Heart Failure

Costs of Providing Antiplatelet Medication for Percutaneous Coronary Intervention Patients

Comparing Variation in Medicare and Private Insurance Spending in Texas

Collection of Language Data and Services Provided by Health Plans

Physician Prescribing Behavior and Its Impact on Patient-Level Outcomes

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

Toward a Business Case for Performance Improvement