 To comprehensively describe the populations, interventions, and outcomes of interactive voice response system (IVRS) clinical trials.

 We identified studies using MEDLINE (1950-2008) and EMBASE (1980-2008). We also identified studies using hand searches of the Science Citation Index and the reference lists of included articles. Included were randomized and controlled clinical trials that examined the effect of an IVRS intervention on clinical end points, measures of disease control, process adherence, or quality-of-life measures. Continuous and dichotomous outcomes were meta-analyzed using mean difference and median effects methodology, respectively.

 Forty studies (n = 106,959 patients) met inclusion criteria. Of these studies, 25 used an IVRS intervention aimed at encouraging adherence with recommended tests, treatments, or behaviors; the remaining 15 used an IVRS for chronic disease management. Three studies reported clinical end points, which could not be statistically pooled. In 6 studies that reported objective clinical measures of disease control (glycosylated hemoglobin, total cholesterol, and serum glucose), the IVRS was associated with nonsignificant improvements. In 14 studies that measured objective process adherence outcomes, the median effect was 7.9% (25th-75th percentile: 2.8%, 19.5%). For the 16 studies that assessed patient-reported measures of disease control and the 11 studies that assessed patient-reported process adherence outcomes, approximately one-third of the outcomes significantly favored the IVRS group.

 IVRS interventions, which enable patients to interact with computer databases via telephone, have shown a significant benefit in adherence to various processes of care. Future IVRS studies should include clinically relevant outcomes.

A systematic review was performed to comprehensively describe the populations, interventions, and outcomes of interactive voice response system (IVRS) clinical trials.

IVRS-based interventions are feasible in many settings and can result in modest improvements in adherence to many processes of care.

We caution against the interpretation that IVRS improves outcomes, as there are currently insufficient data to support such a conclusion.

For high-quality ambulatory care, physicians and their healthcare team must have a high level of communication with patients between visits. These interactions frequently include the provision of treatment advice and monitoring of chronic disease.

 Physicians must support patients and advise them to achieve adherence with test, treatment, and behavioral recommendations. Patients who receive a timely reminder are significantly more likely to have screening tests.

 Similarly, the management of chronic diseases often requires ongoing assessments of various clinical parameters between visits (eg, glucose values in patients with diabetes or body weight in patients with heart failure).

 Patients may experience improved outcomes if their treatment is promptly modified after measurements outside of the desired treatment ranges. The capacity for physicians to fulfill such monitoring and support functions is greatly limited by various factors, including a reimbursement system that does not explicitly recognize the time required to perform them.

Information and communication technologies may effectively and efficiently facilitate intervisit management.

 One such technology, the interactive voice response system (IVRS), could be used to contact patients with reminders or to track patient-assessed parameters measured at home. The IVRS is a technology that enables patients to interact with computer databases via telephones.

 It prompts patients to provide information following a scripted dialogue. It captures responses using keypad entry or speech recognition and stores the information in a database. Patient responses may trigger the IVRS to perform other actions such as sending electronic notifications. Other information and communication technologies, such as patient-accessible Web portals and e-mail, also could be used to support intervisit management. However, IVRS may be more easily adopted because most people own and can use a telephone.

A comprehensive analysis of the utility of IVRS-based interventions is needed. Although used in industries other than healthcare for years, IVRSs only recently were adopted for use in healthcare settings.

 As a result, their effectiveness in improving care and acceptability to patients is largely unknown. Although some data suggest they are effective, negative studies also exist. Furthermore, very few published studies of IVRS interventions have used a comprehensive health technology assessment framework that evaluates processes and outcomes of care. A systematic review of IVRS interventions will identify gaps in the current evidence, investigate the utility of this technology, and inform future technology development. Although prior reviews of IVRS technologies were conducted in 2002 and 2003,

 these studies did not quantitatively examine the effects of IVRS on outcomes. Also, more than 10 IVRS randomized controlled trials (RCTs) have been published since 2003. For these reasons, we conducted a systematic review to comprehensively describe the populations, interventions, and outcomes of IVRS clinical trials.

We identified potentially pertinent citations in the MEDLINE database (1950-2008) using the search strategy in 

). We used a combination of key words related to IVRS (eg, automated, telephone). We modified the MEDLINE search to identify citations in the EMBASE database (1980-2008). We also included studies found using hand searches of the Science Citation Index and the reference lists of included articles.

We retrieved the full text of articles if the title or abstract suggested that the investigators evaluated an IVRS. We included studies published in English that examined the effect of an IVRS intervention on at least 1 of the following types of outcomes: 

, measures of disease control, process adherence, and quality-of-life measures. Clinical end points included disease-related outcomes such as death or hospitalization; 

 were objective and patient-reported markers of disease or health status such as blood pressure, glycosylated hemoglobin (A1C), or a score on a disease-specific validated scale; 

 outcomes assessed whether patients followed a targeted process of care such as screening tests, immunization protocols, or home glucose monitoring; and 

 measures included general health scores from validated questionnaires. Only RCTs and controlled clinical trials (CCTs) (ie, trials that included at least 2 groups and used a quasi-random allocation method) were included in the review. Studies that used an IVRS only to collect data or conducted a validation study were excluded.

From each study, we abstracted details about the population, the IVRS intervention, and outcomes according to an intention-to-treat approach. If a study presented data for more than 2 groups, we abstracted data for 1 intervention group and 1 control group to capture the most direct comparison (

). Where possible, we included the intervention group that received the IVRS intervention only and the control group that received no intervention. If no such groups were reported, we included the intervention group that received the IVRS intervention plus the simplest other intervention (eg, educational booklet) and the control group that received the same simple intervention. Two reviewers (NO and AJ) independently abstracted the data.

We assigned studies an overall quality score using the checklist specified by the Cochrane Effective Practice and Organization of Care group.17 The checklist, used for both RCTs and CCTs, included 3 primary criteria (concealment of allocation, blind assessment of primary outcome, completeness of follow-up) and 3 secondary criteria (balanced baseline measures, reliable outcome measures, protection against contamination). Studies were classified as high quality when they satisfied all primary criteria and did not elicit significant concerns regarding the secondary criteria, as moderate quality if 1 or 2 of the primary criteria were scored as “not clear” or “not done,” and as low quality if the 3 primary criteria were scored as not clear or not done.

Meta-analyses within the various outcome categories were limited because of heterogeneity across the outcomes. In addition, a significant number of studies captured patient-reported outcomes that were not externally validated (eg, patient-reported exercise frequency). To minimize bias, outcomes that were not externally validated were not included in our quantitative analyses. Among the studies that measured outcomes that could not be externally validated (eg, pain), heterogeneity across the outcomes prevented meta-analyses. Study data could be pooled only for measures of disease control that were clinical and dichotomous process adherence outcomes.

For the different types of measures of disease control, we calculated the overall mean difference. These calculations required the mean and standard deviation for intervention and control groups. If a study did not report standard deviations, we used reported statistics (eg, 

 value) to impute a standard deviation for both study groups.

 Analyses were performed using Review Manager version 4.2.10 (The Cochrane Collaboration, The Nordic Cochrane Centre, Copenhagen, Denmark).

For dichotomous process adherence outcomes, we calculated the overall effect estimate using median effects methodology. 20 We first calculated the median effect, defined as the absolute percentage change between the intervention and control groups, for each study. If a study presented more than 1 dichotomous process adherence outcome, we ranked the study’s effect sizes and selected the median. Individual study effects then were ranked, and the median and interquartile range were selected as the summary process adherence measure. These analyses were performed using SAS version 9.1 (SAS Institute Inc, Cary, NC).

Change scores were calculated for objective outcomes that could not be pooled and patient-reported outcomes. The change score was defined as the difference between the change from baseline to follow-up in the intervention and control groups. If a study did not report baseline measures, the change score was the difference between follow-up measures for the 2 groups. No analyses were performed on these data.

Our electronic search yielded 3018 citations (

). The full text of 165 citations was retrieved and reviewed. We excluded studies where the IVRS was not the study intervention (n = 88), the study design was not an RCT or CCT (n = 36), and the study outcomes could not be grouped into this study’s outcome categories (n = 6). An additional 5 included studies were retrieved by hand-searching.

Forty studies published between 1989 and 2008 met inclusion criteria (

 Nearly two-thirds of the studies were conducted at a community practice (n = 26). The remaining studies were conducted at specialty clinics (n = 7), health management organizations (n = 4), home care companies (n = 2), and a Veterans Affairs medical center (n = 1). The 40 studies included a total of 106,959 patients. The median sample size was 237 (25th- 75th percentile: 122-647). Participant follow-up varied from 1 month to 1 year. The mean age and sex distribution of each study are detailed in eAppendix C.

The majority of studies (n = 25) used an IVRS for achieving adherence with treatments, screening, tests, or recommended behavior modifications (eAppendix C).

21,23,26,28-31, 33-35,38,39,41-44,48-50,52,54-56,58,59

 The most common targeted behaviors were immunization (n = 10) and healthy lifestyle (n = 5). Of the 24 studies, 12 used an IVRS to deliver reminders (eg, immunization, laboratory monitoring). These IVRSs were programmed to contact patients between 1 and 7 days before the appointment. The remaining IVRSs were used to assess and reinforce behaviors. Patients in these studies were instructed to call the toll-free IVRS at an interval that ranged from daily to weekly. In 4 of the 25 studies, the IVRS was a part of a multifaceted intervention.

 By far the most common outcome among these studies was process adherence. None of the 25 studies measured clinical end points, and 6 examined a measure of disease control.

22,24,25,27,32,36,37,40,45-47,51,57,60,61

 used an IVRS for managing chronic disease (eAppendix C), most commonly diabetes (n = 3), heart failure (n = 3), and mental illness (n = 3). Of the 15 studies, 12 used an IVRS to collect clinical measurements (eg, blood pressure) or assess disease symptoms (eg, hypoglycemia for diabetes patients). Frequently, patients’ report of symptoms triggered follow-up from a healthcare professional. The other 3 studies used an IVRS that administered treatment modules or assessed patients’ needs for therapy changes. The majority of the 15 IVRSs were programmed to provide personalized feedback to participants based on their responses during the automated dialogue. In 6 of the 15 studies, the IVRS was a part of a multifaceted intervention.

 The most common outcomes among the chronic disease studies were measures of disease control and process adherence.

Seven studies were considered to be high quality,

 studies were considered to be moderate quality,

21-23,25,27,28,30,31,33-36,40,43,45,47-52,55-60

 and 6 studies were considered to be low quality.

 Randomization was clearly done in 11 studies

 It was not clear whether randomization had been done in 22 studies.

22,23,25,27,30,31,33-36,38-40,42-45,47,49-51,57

 Twenty-nine studies reported more than 80% follow-up,

 The follow-up rate was not reported in 7 studies.

 With respect to the third important quality criterion, 22 studies blindly assessed their primary outcome,

 It was not clear whether blind assessment of the primary outcome was done in 2 studies.

 We assessed study quality to further describe the included studies. Study quality was not used to conduct sensitivity analyses because the meta-analyses were already limited by study heterogeneity.

Only 3 studies reported clinical end points (

). We could not pool these data because the specific outcomes varied across the studies. For the 2 studies of heart failure patients, the IVRS was associated with improved outcomes. However, the sample sizes of the 2 studies were very small (n = 70). No difference was found between the intervention and control groups in the asthma study.

Six studies reported externally validated measures of disease status that were clinical. The specific outcomes evaluated in these studies varied (

). The outcomes included A1C, total cholesterol, and serum glucose. Overall, the IVRS interventions were associated with a nonsignificant improvement in all outcomes. The pooled mean differences (95% confidence interval [CI]) were 0.16 (−0.12, 0.43) for A1C, 5.56 (−6.33, 17.46) for total cholesterol, and 10.30 (−12.13, 32.73) for serum glucose. It is noteworthy that none of the sample sizes were larger than 140 patients.

Across all studies, the most common outcome was process adherence. Fifteen studies measured 24 objective, dichotomous process adherence outcomes. However, 1 study was excluded from the analysis because it reported only the relative change in screening rates between the intervention and control groups.

 For the 14 other studies, the median effect of the IVRS intervention was significant at 7.9% (interquartile range 2.8-19.5). As a subgroup analysis, we pooled 8 studies that measured immunization rates. Overall, patients who received the IVRS intervention had significantly higher immunization rates (odds ratio = 1.70; 95% CI = 1.41, 2.04). We also performed a sensitivity analysis to examine whether the median effect differed by process type (immunization vs other) and found no such effect (

Twenty-four studies reported a total of 74 patient-reported outcomes. Sixty percent of the outcomes were measures of disease control, which ranged from weight to scores on the obsessive-compulsive disorder scale. The remaining 40% of outcomes assessed patient process adherence. These outcomes ranged from medication adherence to physical activity. Because of the heterogeneity across all patient-reported outcomes and the susceptibility to bias of many of the outcomes, we did not analyze these outcomes. However, we did abstract study information for descriptive purposes (

). For two-thirds of the outcomes, there were no significant differences between the intervention and control groups. The other outcomes had results that favored the IVRS intervention.

We summarize quality-of-life outcomes in eAppendix E (available at www.ajmc.com). Five studies used 4 different scales to measure general health outcomes. Two studies used the 36-Item Short Form Health Survey (SF-36), 2 used the Work and Social Adjustment Scale, and the other used the General Health Status Index. Because of these differences, we did not combine the data. Of the 5 studies, 2 reported that the IVRS intervention was associated with significant improvements in general health.

Our study highlights the potential usefulness of IVRS for facilitating 2 aspects of ambulatory care: ensuring adherence with tests, treatments, and behavior change; and managing chronic disease. We identified 40 clinical trials evaluating different IVRS-based interventions, which were implemented in diverse settings for a number of medical conditions. Not surprisingly, the diversity of the studies created difficulty in making definitive conclusions about the technology’s overall effectiveness. However, the available data suggest a modest beneficial effect. We found IVRS-based interventions were associated with a 7.9% improvement in adherence with recommended treatments and tests. In addition, IVRS-based interventions were associated with a trend toward improvement in clinical measures associated with better disease outcomes in patients with diabetes, hypertension, and dyslipidemia. The impact of IVRS-based interventions on clinical end points is largely unstudied, but 1 of the 3 studies that did measure clinical end points found a reduction in hospital days for patients with heart failure.

In an era of greater accountability for physicians and health systems, any tool that can improve adherence with recommended therapies and treatments may be worthy of implementation, even in the absence of proof of an effect on outcomes. Many health plans, including Medicare, are implementing initiatives to monitor the performance of their providers.

 Some of these initiatives include pay for performance, in which providers receive financial rewards for meeting explicit standards of care or indicators. Commonly, these indicators are defined using evidence-based processes of care and surrogate outcomes. Examples of such indicators include adherence to screening guidelines in selected populations and the proportion of patients meeting lipid or hypertension targets. Rarely, if ever, do plans use patient outcomes as a measure of performance. As demonstrated in this review, IVRS can be leveraged to obtain better performance on such process indicators.

Interactive voice response systems may be used alone or as a complement to other initiatives or technologies. For example, IVRSs can be linked with components of the electronic health record, thereby expanding this promising technology’s benefit. In fact, it may be easier to implement and monitor the impact of an IVRS that is integrated with an existing information system infrastructure. Also, recent quality improvement studies have focused on different organizational approaches for chronic disease management. A common component of many of these initiatives is frequent telephone contact.

 Such calls usually involve the patient providing data on his or her health status and the practitioner providing advice. For many diseases and situations, this information can be standardized, which in turn lends itself to the use of an automated IVRS solution. It should be noted that in this approach the IVRS would be designed to improve the efficiency of the nurses and other personnel in the program, not to replace them. Thus, the IVRS will help screen patients who might be having problems, so the nurse or physician responsible for the program can focus on them, rather than spend a disproportionate amount of time on patients who are well.

The current study provides the most thorough and up-todate evaluation of the use of IVRS technology in healthcare. Our review has a number of strengths. We identified and included studies using widely accepted criteria of methodologic rigor. We classified interventions and outcomes using a predefined framework that provides a more comprehensive assessment. In some cases, this allowed us to pool individual study results to determine an overall effect. These effects are less biased because we excluded studies with patient-reported outcomes.

There also are some limitations to this review. Many different study outcomes were evaluated, and few studies evaluated the technology using both clinical and process outcomes. It is noteworthy that only 3 of the 39 studies measured clinical end points. Another predominant issue in the included studies was the common use of patient-reported outcomes. Although patient input is clearly important, it is arguably more important to use externally validated objective outcomes when performing technology assessments. Finally, although we did perform a rigorous search to identify published literature, it is possible that some studies of IVRSs are not published.

We conclude that IVRS-based interventions are feasible in many settings and can result in modest improvements in adherence to many processes of care. If provider groups are focused on improving specific processes of care, then we feel the published data support a decision to implement the technology. However, we must caution against an interpretation that the technology improves outcomes, as there are currently insufficient data to support such a conclusion. We recommend that future studies evaluate patient outcomes in addition to focusing on process measures.

From the Clinical Epidemiology Program (NO, AJ, CvW, AJF), Ottawa Health Research Institute, Ottawa, Ontario, Canada. 

Articles

The American Journal of Managed Care

A systematic review of interactive voice response system studies showed that these interventions significantly benefit adherence to various processes of care.

Interactive Voice Response Systems for Improving Delivery of Ambulatory Care

 To investigate the effects of managed care on drug formulary use and to determine whether these effects have changed over time.

 We use longitudinal data from the Community Tracking Study physician surveys and compare the effects of managed care on drug formulary use between 2000-2001 and 2004-2005.

 Ordinary least squares regression models are estimated to quantify the effects of managed care on drug formulary use. Data from 2000-2001 and 2004-2005 are pooled together, and the year trend effect is represented by a dummy variable for 2004-2005.

 We find that the coefficient measuring the effects of managed care on drug formulary use declined from 0.37 (P <.01) in 2000-2001 to 0.27 (P <.01) in 2004-2005. However, this relative decline in the effects of managed care on formulary use reflects not an absolute decline in formulary use in managed care but an increase in formulary use in the nonmanaged care sector. For the typical physician who has a mix of managed care and nonmanaged care patients, drug formulary use increased over time.

 Differences in drug formulary use between managed care and nonmanaged care declined from 2000-2001 to 2004-2005. The overall level of drug formulary use increased, reflecting an increase in the nonmanaged care sector.

Some observers have argued that managed care has relaxed restrictive practices designed to control costs in light of public backlash.

We find that the effects of managed care on drug formulary use weakened from 2000-2001 to 2004-2005.

However, nonmanaged care increased formulary use during this period, and the absolute level of formulary use under managed care remained virtually unchanged. Managed care and nonmanaged care were converging in formulary use over this period.

These results are consistent whether we study capitated managed care, noncapitated managed care, or managed care as a whole.

Despite the rapid growth of managed care during the past 2 decades, its restrictive practices have drawn increasing criticism from physicians and their patients.

 In response to growing concerns about such restrictions, the role of managed care in constraining healthcare services is believed to have declined in recent years.

If managed care has become less restrictive, this could have important implications for patient access and patterns of care. Yet, there is little quantitative evidence on these issues. Using a national physician sample,

 we found that the restrictions on the provision of physician services declined in managed care between 2000-2001 and 2004-2005. The present study seeks to provide further evidence on the evolving role of managed care with respect to the use of drug formularies. A drug formulary is a list of prescription medications covered under a health insurance plan. Many formularies have several copayment tiers, with the lowest copayments for “preferred drugs,” which are often generics or less costly brand-name drugs.

 This article investigates whether the association between physician involvement in managed care and prevalence of prescription formulary use among their patients has declined over time. 

The present study uses data from the Community Tracking Study (CTS) physician surveys for 2000-2001 and 2004-2005, maintained at the Center for Studying Health System Change. The CTS physician surveys achieve response rates of about 60% and comprise a nationally representative sample of physicians in the United States. A review of the CTS database concluded that “there was little evidence of a systematic underrepresentation among demographic and practice characteristics available for all physicians from the AMA [American Medical Association] Masterfile.”

 The surveys include approximately 12,000 physicians in 2000-2001 and 6000 physicians in 2004-2005. The 2004-2005 CTS sample survey was 50% as large as the survey in 2000-2001, which accounts for the smaller size of our 2004-2005 sample. However, the sampling method was the same in both surveys.

Physicians in the CTS surveys engaged in direct patient care for at least 20 hours/week in 60 randomly selected communities in the United States, and the survey questionnaires include information on managed care revenue, medical care management, and various other physician and practice characteristics. After excluding approximately 9% of subjects who did not respond to the prescription formulary, managed care, and key control variables, our study sample includes 11,940 physicians in 2000-2001 and 5463 physicians in 2004-2005.

To gauge the extent of physicians whose patients use prescription formularies, the CTS survey asked each physician the following question: “What percentage of your patients have prescription coverage that includes the use of a formulary?” Physician responses ranged from 0% to 100%.

Managed care involvement for physicians is measured as the percentage of their practice revenues from managed care organizations. More specifically, the CTS survey asked each physician: “What percentage, if any, of the patient care revenue received by the practice in which you work comes from all managed care combined?”

The share of revenue from managed care is correlated with the number of managed care patients. That is, a physician with a higher share of revenue from managed care has more patients in managed care than does a physician who has a lower share of revenue from managed care, other things being equal. Managed care is further disaggregated into the share of revenue from capitated managed care and noncapitated managed care, which we also examine separately.

Many managed care organizations offer point-of-service options so that patients may choose the provider and treatment style of their choice, albeit at higher cost. Although in 2004 only 16% of employees had a choice of conventional indemnity plans, 30% also had a point-of-service option.9 Therefore, there is substantial opportunity to go out of network for more traditional types of coverage. Indeed, Strunk and Reschovsky10 report that managed care revenues accounted for slightly less than half of physician practice revenues in 2001. In our sample, managed care revenues accounted for 46% of physician revenues in 2000-2001 and 44% in 2004-2005. Nevertheless, these revenue measures may underestimate the actual number of patients in managed care plans because reimbursement is typically lower in managed care.

We also control for several physician and practice characteristics that may be correlated with the percentage of patients whose prescription drug coverage includes formulary use. These variables are listed in 

We first estimate separate prescription formulary equations for 2000-2001 and 2004-2005. This may be written as: 

 indicates the percentage of patients who have formulary prescription coverage; 

, the percentage of practice revenue from managed care; 

, a vector of other explanatory variables; ε, a disturbance term; 

, the time subscript for 2000-2001 or 2004-2005; and β

, the coefficients to be estimated. We estimate the equation using ordinary least squares regression because the dependent variable is a continuous measure from 0 to 100. We then compare the coefficients measuring the effects of managed care on drug formulary use in 2000-2001 versus 2004-2005 to examine whether the relationship between managed care and formulary coverage has changed over time.

To provide a direct estimate of year trend effects on formulary coverage between 2000-2001 and 2004-2005, we next pool these data, including a dummy variable for 2004-2005 and the interaction of this year dummy variable with the managed care variable. Therefore, the pooled multivariate regression model is as follows: 

 indicates a dummy variable for 2004-2005; 

, an interaction term between managed care and the year dummy; and Θ

Table 1 gives descriptive statistics for each wave of the CTS physician surveys. From 2000-2001 to 2004-2005, we find that the percentage of patients having prescription coverage that includes the use of a formulary rose modestly from 60% to 64%. From 2000-2001 to 2004-2005, the percentage of revenue from managed care declined slightly from 46% to 44%.

 first summarizes the ordinary least squares regression results estimating data from 2000-2001 and 2004-2005 separately. The coefficients measuring the effects of managed care on drug formulary use are 0.37 in 2000-2001 and 0.27 in 2004-2005. Therefore, a 1% increase in managed care revenue will lead to a 0.37% increase in prescription formulary use among patients in 2000-2001 but only a 0.27% increase in 2004-2005. Therefore, the elasticity has been reduced by more than one-fourth from 2000-2001 to 2004-2005.

Table 2 also summarizes the ordinary least squares regression results when we pool the data from 2000-2001 and 2004-2005. We have included a dummy variable for 2004-2005 and its interaction term with the managed care variable. The coefficient of managed care is 0.37, and the coefficient of the interaction term is −0.09. Both estimates are statistically significant at the level of 1%. This robust test supports our hypothesis that the effects of managed care on the percentage of patients with formulary coverage declined over time. The effects of capitated managed care and noncapitated managed care on formulary coverage may differ, as prescription formulary coverage is believed to be more common among capitated managed care plans.

 Therefore, we also break managed care into 2 parts, capitated managed care and noncapitated managed care. Turning to separate estimates for each wave of data, we find that the coefficient of capitated managed care is 0.44 and the coefficient of noncapitated managed care is 0.32 in 2000-2001. Both coefficients (capitated managed care, 0.30; and noncapitated managed care, 0.25) decline in 2004-2005 compared with the earlier period. All these coefficients are statistically significant at the level of 1%. Because of space limitations, the full results are not presented herein but are available from the authors on request.

Has formulary coverage declined in managed care? Thus far, the results have shown that formulary coverage has declined in managed care relative to nonmanaged care. But what has happened to the absolute level of formulary use? To help answer this question, we use the estimated results in Table 2 to generate predicted values for formulary coverage in 2000-2001 and 2004-2005. When physician involvement in managed care is 100%, the mean predicted percentage of formulary coverage is virtually unchanged between 2000-2001 and 2004-2005. This indicates that the absolute level of formulary coverage did not change for managed care over this period. In contrast, when there is no managed care involvement, the mean predicted percentage of formulary coverage increases substantially from 43% in 2000-2001 to 52% in 2004-2005. Because most physicians have a mix of managed care and nonmanaged care patients, overall formulary coverage increased over time. Therefore, for physicians who receive 50% of their revenues from managed care, the mean percentage of prescription formulary coverage increased from 62% to 66%.

Some observers have argued that, in response to backlash from providers and patients, managed care has relaxed restrictive practices intended to control costs.

 This study has examined this issue in the context of prescription drug formulary use. We find that, while the effects of managed care on formulary coverage declined relative to that of nonmanaged care, the absolute level of formulary coverage in managed care did not change. Given the widespread presence of formulary use in managed care, the stability of formulary coverage over time may represent a ceiling effect. In contrast, formulary coverage became more common among nonmanaged care plans in 2004-2005 than in 2000-2001. This finding suggests that managed care and nonmanaged care plans are becoming more similar in terms of formulary coverage. Because of this convergence, changes in the mix of indemnity versus managed care plans may have little effect on the extent of prescription drug formulary use in the future. Given consumer and provider backlash against managed care, many plans are relaxing restrictions and allowing beneficiaries to go out of network.12 But others have observed that indemnity plans are assuming more features that were typical of managed care plans, including greater use of formularies:

Over time, the distinction between HMOs [health maintenance organizations] and indemnity plans has become increasingly blurred, says Larry Levitt, vice president at the Kaiser Family Foundation. “Traditional plans have introduced utilization measures that were historically found only in managed care plans,” he says…. Another familiar HMO-style feature found in traditional plans is the drug formulary—the list of medicines that a plan will pay for. When Blue Cross/ Blue Shield of New Jersey sent out its list to policymakers last year, it included a letter recommending they speak with their physicians about finding substitutions for prescriptions that weren’t on the formulary.

Some investigators have questioned the success of drug formulary use as a cost-control measure, arguing that physicians contract with multiple plans and that these plans have very different formulary lists.

 In a review of health plans in California, Shrank et al note great heterogeneity in formulary lists:

None of the branded drugs evaluated were preferred on all formularies in 2002, and 10% were not available on any of the formularies. Formulary status varied greatly across plans, and more than 60% of drugs were preferred on 2 to 4 of the 6 formularies studied. Formulary status within health plans varied between 2000 and 2002 in more than half of the plans in the drug classes evaluated.

 Moreover, formulary use may have adverse consequences for consumers. Huskamp et al

 found that, while the use of more restrictive formularies constrained spending for several drug classes, it also shifted the burden of costs onto consumers. These researchers also found that different types of formulary changes could have substantially different effects on drug utilization and costs and that changes in copayment structures have far-reaching consequences for compliance, patient costs, and possibly quality of care. Prescription formularies are an increasingly common feature of health insurance plans. Investigations to obtain more evidence on the implications of these changes for cost, access, and quality of care are important directions for further research.

Health Economics Research Group (HF), University of Miami, Coral Gables, FL; and Department of Economics (JAR) and Department of Preventive Medicine (JAR), Stony Brook University, Stony Brook, NY. Dr Fang is now with the Department of Health Systems, Management and Policy, University of Colorado Denver, Aurora.

The authors (HF, JAR) report no relationship or financial interest with any entity that would pose a conflict of interest with the subject matter of this article.

Concept and design (HF, JAR); acquisition of data (HF, JAR); analysis and interpretation of data (HF, JAR); drafting of the manuscript (HF, JAR); critical revision of the manuscript for important intellectual content (HF, JAR); statistical analysis (HF, JAR); provision of study materials or patients (HF, JAR); administrative, technical, or logistic support (HF, JAR); and supervision (HF, JAR).

Hai Fang, PhD, MPH, Department of Health Systems, Management and Policy, University of Colorado Denver, 13001 E 17th Ave, 3rd Fl, Campus Box B119, Aurora, CO 80045. E-mail:fanghaimail@gmail.com.

1. Blendon RJ, Brodie M, Benson JM, et al. 

Understanding the managed care backlash. 

 The managed care backlash and the task force in California. 

4. Draper DA, Hurley RE, Lesser CS, Strunk BC.

 Is managed care still an effective cost containment device? Forum Health Econ Policy. 2006;9(1):e1007. Frontiers in Health Policy Research article 3. http://www.bepress.com/fhep/9/1/3. Accessed May 15, 2009.

The changing effect of managed care on physician financial incentives. 

7. Huskamp HA, Deverka PA, Epstein AM, Epstein RS, McGuigan KA, Frank RG. 

The effect of incentive-based formularies on prescriptiondrug utilization and spending. 

8. Center for Studying Health System Change.

Community Tracking Study: Physician Survey Methodology Report: 2000-01 (Round Three). 

Washington, DC: Center for Studying Health System Change; May 2003:C.19-C.20. Appendices to Technical Publication No. 38. http://www.hschange.org/CONTENT/570/570a.pdf. Accessed August 30, 2008.

9. Health Research & Educational Trust (HRET) Web site.

 HRET and The Kaiser Family Foundation: 2004 Employer Health Benefits Annual Survey. http://www.hret.org/hret/publications/2004ehbs.html. Accessed September 14, 2008.

 Kinder and gentler: physicians and managed care, 1997-2001. Center for Studying Health System Change Tracking Report: results from the Community Tracking Study No. 5. November 2002. http://www.hschange.com/CONTENT/486/486.pdf. Accessed September 14, 2008.

. The effect of managed care on prescription drug costs and benefits. 

A closer look at the managed care backlash. 

Ten things your health insurer won’t tell you. http://www.badfaithinsurance.org/reference/HL/0015a.htm. Accessed January 21, 2009.

Consumers face higher costs as health plans seek to control drug spending. Center for Studying Health System Change Issue Brief: findings from HSC No. 45. November 2001. http://www.hschange.org/CONTENT/389/389.pdf. Accessed September 14, 2008.

15. Shrank WH, Ettner SL, Glassman P, Asch SM.

 A bitter pill: formulary variability and the challenge to prescribing physicians. 

This article examines the changing effects of managed care on prescription drug formulary use in the United States.

Managed Care and Physicians' Perceptions of Drug Formulary Use

 To examine costs associated with an asthma management program that reduces asthma-related health services utilization and to calculate potential return on investment (ROI) from the Medicaid managed care plan perspective.

 Clinical and economic data were obtained for 3298 ethnically diverse children with asthma (48% with persistent asthma) who resided in a poor urban community (Hartford,  Connecticut) and were enrolled in Easy Breathing, an asthma management program for pediatricians. We calculated the cost per participating child with asthma during the first 3 years (July 1998 to June 2001) relative to the difference in costs for participating and nonparticipating children calculated by applying Medicaid reimbursement rates to data on services.

Start-up costs were $28.95 per child with asthma in year 1, and operating costs averaged $10.28 in years 2 and 3. The mean reduction in costs was $36.72 per child per year in years 2 and 3. If Medicaid managed care plans had been charged an amount equal to program operating costs after year 1 ($10.28 per child with asthma per year), at-risk health plans could have incurred cost savings of approximately $26.44 per child with asthma per year. The potential ROI for years 2 and 3 was $3.58 per US dollar spent.

Easy Breathing reduced overall costs of care for urban children with asthma of varying severities. If managed care plans held at risk by Medicaid had reimbursed program operating costs for participants in Easy Breathing, they would have experienced a positive ROI.

Asthma-related costs in urban children can be reduced by simple programs that are designed to move practices toward compliance with recommendations for instituting inhaled corticosteroid therapy for children with persistent asthma and for creating a severity-appropriate, written asthma treatment plan.

Improved health is associated with savings for at-risk public payers.

Start-up and operating costs are modest ($28.95 per child with asthma during the first year and $10.28 per child with asthma in each subsequent year).

The potential return on investment in implementing Easy Breathing for an ethnically diverse group of minority children was $3.58 per US dollar spent.

The mean healthcare expenditures for children with asthma are roughly 1.5 times higher compared with those for children without asthma.

 This economic burden of asthma is disproportionately distributed among blacks and Hispanics, who have higher rates of hospitalization and emergency department (ED) visits than whites.

 These higher costs are thought to be due to the greater asthma prevalence and severity among minority populations and to the consequences of uncontrolled asthma, which potentially could be reduced with improved asthma management.

 disseminated by the National Asthma Education and Prevention Program (NAEPP), including the 2007 NAEPP Expert Panel Report 3, have not been widely adopted by primary care clinicians, and few studies have assessed the costs of guideline interventions. Of these, most have demonstrated cost savings only when used among individuals with severe disease or during times of high utilization (eg, hospitalization or ED visits).

 Easy Breathing is an NAEPP guideline—adapted asthma management program that is used by more than 325 clinicians in Connecticut. Easy Breathing participants increased their use of inhaled corticosteroid (ICS) therapy for children with persistent asthma from 38% before program implementation to 95% after implementation.

 These increases in ICS therapy were associated with decreases in hospitalizations, ED visits, and outpatient visits among urban minority children with asthma and with decreases in outpatient visits and ED visits among children cared for in private practices.

The objectives of this study were to examine the costs associated with implementation of Easy Breathing in an urban setting, the cost changes associated with a decrease in medical services utilization and an increase in pharmacy charges for children enrolled in Easy Breathing, and the potential return on investment (ROI) from a managed care perspective. In Connecticut, the Medicaid agency placed most children under capitated managed care plans in 1996. Most of the plans reimbursed providers on a fee-for-service or contact capitation basis, which meant that the managed care plans were at risk for changes in utilization and costs. Consequently, any reduction in utilization of healthcare services by  Medicaid-enrolled children with asthma resulted in direct cost savings to the Medicaid managed care plans during this period. No fees were charged to the plans by Easy Breathing, but our objective was to assess the potential ROI if fees had been assessed or paid voluntarily.

We projected financial effects for Medicaid managed care plans among children before and after enrollment in Easy Breathing. Although Easy Breathing was privately funded and not reimbursed by insurers, we modeled the potential financial effect as if the operating costs had been apportioned to the health plans. Specifically, we calculated the Medicaid payments associated with observed changes in utilization of healthcare services by Medicaid-enrolled children with asthma enrolled in Easy Breathing and subtracted the estimated changes in payments to clinicians resulting from changes in the utilization of services from the cost of the program.

 consists of the following 4 major elements: a survey, a provider assessment, an asthma treatment selection guide, and an asthma treatment plan. A fifth element, the assessment of asthma control, was added after the release of the 2007 NAEPP Expert Panel Report 

The Easy Breathing Survey contains 15 questions, 4 of which have been validated for the diagnosis of asthma.

 In addition, there are 2 questions about a previous diagnosis of and treatment for asthma, 1 family history question, and 8 demographic and environmental exposure questions. The survey is completed by the parents of children (age range, 6 months to 18 years) who present for care for any reason at any of the primary care clinics in Hartford, Connecticut, usually while the parent is waiting to be seen. Of 6 primary care clinics in Hartford, 2 are hospital-based clinics, 2 are federally funded health centers, and 2 are university-affiliated clinics. Using parent responses to the 4 asthma-validated questions, the child’s medical record, and additional history and testing as needed, the clinician determines whether the child has asthma. Clinicians consider a diagnosis of asthma when children report recurrent (≥3) episodes of wheezing, cough, or shortness of breath in response to known asthma triggers and when other diseases have been excluded.

 For children with asthma, the clinician determines asthma severity by asking 4 questions about the frequency of daytime and nocturnal symptoms, exercise impairment, and school absenteeism for asthma (provider assessment).

 Treatment is chosen using the asthma treatment selection guide, a list of severity- appropriate medications grouped by potency, with annotation of which Medicaid plans cover which drugs. A preprinted asthma treatment plan with parent instructions is generated, to which the clinician adds the prescribed medications. The asthma treatment plan tells the parent what medications to use daily, when to use the sick plan and what medications to use, and who and when to call. A new asthma treatment plan is generated every time that treatment is changed. Changes in the asthma treatment plan that occurred within 12 months of the initial Easy Breathing Survey were assumed to reflect asthma control rather than changes in severity, as changes in intrinsic disease severity usually occur slowly over time.

Clinicians and staff participated in a 2-hour training session about how to use Easy Breathing and how to create a written treatment plan. Training and remediation were provided by a physician champion who was available for consultation and by a program coordinator who made weekly visits to pick up and review forms and to provide feedback to clinicians. During the first year, a pediatric pulmonologist provided on-site consultation and program-related education for 1 to 2 hours per week.

The Easy Breathing program and this study were approved by the institutional review boards at the University of Connecticut Health Center and Connecticut Children’s Medical Center. They were also approved by the Centers for Disease Control and Prevention, with oversight deferred to Connecticut Children’s Medical Center.

Personnel costs (physician champion and program coordinator) for program start-up (year 1) and ongoing support (years 2 and 3) were based on the grant that supported the program and the actual time spent in the various activities. The budgeted costs for personnel were compared with the estimated actual annual costs, and the larger of the 2 personnel costs was used in each case. Other costs (eg, travel and forms) were taken directly from the costs reported in the grant. Research-related costs (eg, database creation and data entry) were not included.

Program costs were examined by year and were separately tabulated for the first year and for the second and third years. Start-up during year 1 was associated with higher costs. The rationale was that it would not be reasonable to charge payers for start-up costs but that it would be reasonable for operating costs during years 2 and 3 to be apportioned to the payers. All costs are expressed in 2006 US dollars.

The healthcare costs associated with asthma (

International Classification of Diseases, Ninth Revision code

 493.xx), including hospitalizations, ED visits, outpatient visits, and asthmarelated drug prescriptions, were estimated in 2 steps. First, observed utilization was calculated based on claims data and eligibility files that were obtained from Connecticut’s peer review organization, Qualidigm, Inc, for Medicaid enrollees from July 1997 to June 2001. Second, 2006 Medicaid reimbursement rates in Connecticut were used as unit costs. This approach is consistent with the payer perspective for economic evaluation. The potential ROI to plans was calculated as the ratio of reduction in healthcare expenditures during years 2 and 3 to the mean program costs during the same years.

Asthma drugs were identified using National Drug Codes and were grouped into 4 categories (bronchodilators, ICSs, nonsteroidal anti-inflammatory drugs, and oral corticosteroids). The mean cost for a drug within each category in 2006 was multiplied by the number of filled prescriptions to yield the annual cost. The 2006 prices of prescription drugs for asthma and medical services were obtained from the Connecticut Department of Social Services.

Healthcare services costs were calculated for 1 year before program implementation (July 1997 to June 1998) and for 3 years after program implementation (July 1998 to June 2001) for children who were continuously enrolled in Easy Breathing between June 1, 1998, and July 1, 2002. Each child contributed 1 month of person-time to the analysis for every month of claims data that were available at any point during the data analysis. During years 1 through 3, at any point in time there were children already enrolled in Easy Breathing and children not yet enrolled; all of these children contributed person-time and incident events such as hospitalizations to the analyses. We determined utilization by pooling the person-time and events of all children, both before and after their enrollment.

 On average, children were Medicaid eligible for at least 10 months of each year.

For each child during the preenrollment and postenrollment periods in Easy Breathing per year, we calculated the following: the total number of paid claims for each of the asthma drug classes; the total number of hospitalizations, ED visits, and outpatient visits; and the total number of hospitalizations, ED visits, and outpatient visits for which the primary discharge diagnosis was asthma. Because all children contributed prior time to both the preenrollment and postenrollment periods, the preenrollment and postenrollment utilizations for an individual child are not independent. Therefore, marginal Poisson distribution regression models that were fit using generalized estimating equations were used.

 Predicted values from these equations were used to estimate utilization. All analyses were controlled for sex, race/ethnicity, clinic site, asthma severity, calendar time,

 and aging of the cohort to account for the uneven distribution of asthma rates by these variables.

The demographics of the children enrolled in Easy Breathing between June 1, 1997, and December 31, 2001, have been reported.

 Claims data were identified for 8324 of 9339 children (89%) enrolled in Easy Breathing during this period. Of these, 3298 children were enrolled between July 1, 1997, and June 30, 2001, and had physician-confirmed asthma. Of 3298 enrolled children with asthma, 1569 (48%) were diagnosed as having persistent asthma (29% mild, 17% moderate, and 2% severe). Almost two-thirds of the children with asthma were Hispanic (primarily Puerto Rican), 22% were African American, and 5% were non-Hispanic white; the remaining 7% reported mixed or no race/ethnicity. One-third of the children were aged 6 months to 4 years, 38% were 5 to 9 years, 25% were 10 to 14 years, and the remaining children were 15 to 17 years. Thirty-four percent of the children with asthma reported exposure to environmental tobacco smoke. Claims data were available for 2841 children during year 2, of whom 2238 were enrolled in Easy Breathing, and for 2678 children during year 3, of whom 2386 were enrolled in Easy Breathing.

The total program start-up cost during year 1 divided by the number of children with physician-diagnosed asthma enrolled in the program in year 1 was $28.95. The total program costs during the second and third years of the program divided by the number of children with physician-diagnosed asthma enrolled in the program during those years were $10.28 (

). These costs included the program coordinator’s time to visit the sites and provide feedback and support and the physician champion’s time to oversee the project’s data quality.

Children with asthma who were enrolled in Easy Breathing filled more prescriptions for ICSs and fewer prescriptions for oral corticosteroids in the 12 months after enrollment than children with asthma in the 12 months before their enrollment in Easy Breathing (

). However, the frequency of filled prescriptions for ICSs (123.6 prescriptions per 100 children per year) is far less than what would have been predicted from the asthma treatment plans. The increase in filled prescriptions for ICSs among children with intermittent asthma (from 6.2 to 22.3 prescriptions per 100 children per year) reflects reclassification of children to greater disease severity in the 12 months after enrollment. Within 6 months of enrollment in Easy Breathing (ie, completion of a survey), asthma therapy (and the classification of asthma severity) increased in 3.3% of children with intermittent disease. There was no change in the number of bronchodilator prescriptions for either group. As expected, expenditures for asthma drugs were higher among children with persistent asthma than among children with intermittent asthma.

Children with persistent asthma experienced significant decreases in the number of hospitalizations, ED visits, and outpatient visits for asthma after enrollment in Easy Breathing, while utilization rates for children with intermittent asthma were low before and after enrollment in Easy Breathing (

). Consequently, after enrollment in Easy Breathing, there was a 48.5% decrease in the costs associated with medical services for children with persistent asthma but a 10.0% increase for children with intermittent asthma. 

Taking into account the difference in Medicaid reimbursements for children with intermittent and persistent asthma, the mean start-up cost (year 1) was $31.94 per child. In years 2 and 3, there was a reduction in costs to Medicaid managed care plans associated with ongoing support of the Easy Breathing program of $36.72 per child per year. Assuming that managed care plans were assessed a fee equal to the mean cost of supporting the program ($10.28), there was a net cost savings of $26.44 per child in years 2 and 3. After the initial start-up, implementing Easy Breathing would have a positive ROI for Medicaid managed care plans that were at risk. If they paid the mean operating costs of the program in years 2 and 3, the ROI would have been $3.58 to $1.00. 

Implementation of Easy Breathing has improved the health and decreased medical services utilization among urban minority children in Hartford with asthma. Children with asthma enrolled in Easy Breathing experienced a 35% decrease in overall hospitalization rates (

 <.006), a 27% decrease in asthma ED visits (

 <.01), and a 19% decrease in outpatient visits (

 This study was not subject to bias due to regression to the mean, as (unlike many other studies) our study did not enroll children after ED visits or hospitalizations or target enrollment of children with severe disease.

 Rather, any child with asthma who received care at any primary care clinic in Hartford for any reason was eligible for enrollment.

Over a 3-year period in which 3298 unique low-income Medicaid-enrolled urban children with asthma were enrolled, Easy Breathing helped to reduce direct medical care costs. The magnitude of the reduction in medical costs was such that, if a healthcare payer had been charged a fee equal to the mean operating cost of the program, the payer would have saved more than $3.50 for each US dollar spent on the program. Bencause the study was performed in multiple urban-based clinics with an ethnically diverse population, the findings are likely to be generalizable to inner-city populations, among whom the burden of asthma is especially high.

A major focus of Easy Breathing is the appropriate use of ICSs in children with persistent asthma in association with a written asthma treatment plan. Despite the demonstrated benefits of ICS use in managing asthma, ICSs are underprescribed and underused by primary care clinicians and their patients.

 Remarkably, patients with persistent asthma enrolled in Easy Breathing filled a mean of only 1.2 ICS prescriptions per year. Although we were unable to monitor free-sample dispensing, the use of over-the-counter asthma medications, or the use of nonsteroidal anti-inflammatory therapy for mild persistent asthma, it seems that patients are continuing to underuse prescribed asthma medications.36 We did not assess other elements of asthma management such as quality of life or measurements of pulmonary function, as the program staff did not directly contact participating children or their families and did not review individual medical records.

A significant increase in ICS use was observed among children with asthma in Hartford during the implementation of Easy Breathing. Before implementation, 38% of children with persistent asthma were prescribed an ICS, a percentage similar to what has been reported nationally.

 Afterward, the use of ICSs increased to 95% among children enrolled in Easy Breathing and to 42% among children not enrolled in Easy Breathing. Although these increases occurred shortly after the release of the 1997 NAEPP guidelines,

An asthma disease management program reduced medical services utilization for urban children and had a potentially positive return on investment for Medicaid managed care plans.

June 2009

Interactive Voice Response Systems for Improving Delivery of Ambulatory Care

Managed Care and Physicians' Perceptions of Drug Formulary Use

The Economic Impact of an Urban Asthma Management Program

Treatment Decisions for Complex Patients: Differences Between Primary Care Physicians and Midlevel Providers

Value Based Insurance Design: Maintaining a Focus on Health in an Era of Cost Containment

Does a Personalized Approach Improve Patient Satisfaction in Thoracic Oncology?

Improving the Outcomes of Disease Management by Tailoring Care to the Patient's Level of Activation

Is Integration in Large Medical Groups Associated With Quality?

Effect of Medication Dosing Frequency on Adherence in Chronic Diseases