 To determine whether changes in health maintenanceorganization (HMO) penetration or payer mix affected inhospitalmortality and treatment patterns of patients with acutemyocardial infarction (AMI).

 Observational study using patient-level logisticregression analysis and hospital and year fixed effects of data fromthe Agency for Healthcare Research and Quality's Healthcare Costand Utilization Project Nationwide Inpatient Sample, a geographicallydiverse sample of 20% of the hospitalized patients in theUnited States.

 Discharges of patients (n = 340 064)with a primary diagnosis of acute myocardial infarction who weretreated in general medical or surgical hospitals that contributed atleast 2 years of data to the HealthCare Cost and Utilization ProjectNationwide Inpatient Sample from 1989 to 1996. In-hospital mortalityand rates of cardiac catheterization, angioplasty, or coronaryartery bypass grafting for Medicare patients or non-Medicarepatients were the main outcome measures.

 Among Medicare patients, increases in HMO penetrationwere associated with reduced odds of receiving cardiaccatheterization, angioplasty, or coronary artery bypass grafting of3% to 16%, but were not associated with any change in mortalityrisk. Increases in the number of HMOs within a metropolitan statisticalarea, our measure of HMO competition, were associatedwith small but significant increases in the odds of cardiac catheterizationand angioplasty of about 2%. There was no pattern ofchanges in cardiac procedure rates or in-hospitality mortalityamong non-Medicare patients.

 Increases in HMO penetration reduced cardiacprocedure rates by statistically significant but small amountsamong Medicare patients with AMI, without affecting mortalityrates.

The growth of managed care as a mechanism tolower healthcare costs has led to concerns thatthis may adversely affect quality of care. Relativereductions in payment may affect the quality of carebecause hospitals have fewer resources with which toprovide services. The overall effect on patient care ofdeclines in reimbursement by a particular insurer comprises2 phenomena: a direct effect on the care ofpatients with that particular insurance, who mayreceive fewer beneficial services, and an indirect orspillover effect on other patients within the same hospital,whose care is altered because of hospital-wideresource constraints.

Direct comparisons of quality of care between healthmaintenance organization (HMO) and fee-for-service(FFS) patients have produced mixed results.

However, commercial HMO penetration has beenshown to have spillover effects on the cost of careamong Medicare FFS patients, as lower costs amongMedicare patients have been observed in areas withhigh commercial HMO penetration.

 In addition, higherHMO and managed care (HMO or preferred providerorganization) penetration rates have been shown to beassociated with reduced access to care,

 The effect of HMOs on quality of care for hospitalizedpatients could be mediated by several mechanismsbesides reduced availability of resources. Growthin managed care could bring about changes in physicianpractice patterns, system-wide availability of resourcesand new technology, and changes in the mix of servicesavailable to favor more cost-effective technologies andservices.

The quality effects for non-HMO patients of theobserved spillover effects of HMOs on hospital costs areunknown. While reductions in hospital cost growthcould lead to fewer resources being available for patientcare and to worse outcomes, HMOs have greater abilityto shop for better quality providers than individuals.Two cross-sectional analyses have examined the associationbetween HMO penetration and quality of carereceived by Medicare FFS patients. Heidenreich et al

used data from 1994 to 1995 on Medicare FFS patientswith acute myocardial infarction (AMI) in a cross-sectionalanalysis of geographic areas with different levelsof HMO penetration and found that higher HMO penetrationwas associated with higher use of beneficialmedications such as &#946;-blockers and aspirin and lowerrates of coronary angiography, but was not associatedwith AMI mortality rates. Mukamel et al

 examined1990 data for Medicare patients and found a statisticallysignificant but small reduction in 30-day mortalityrisk of roughly 0.15% among patients in the highest(>24.5%) HMO market penetration group relative tothose in the lowest (<10.6%) HMO penetration group.

The cross-sectional design of the Heidenreich andMukamel studies makes inferences about causalityproblematic. Explanations such as a tendency by HMOsto locate in higher cost markets because of greaterpotential for cost savings may explain the observedassociations, as higher cost markets may also have betteroutcomes.

 In addition, both of the studies focusonly on the Medicare population, and the effect of managedcare on patient quality may have been more significantin hospitals with greater proportions of patientsyounger than 65 years, particularly if a significant portionof patients is part of a vulnerable population group,such as the uninsured. The role of HMO competition inthis context is also important to explore. Previous workhas demonstrated that, while HMO profits have beenshown to be lower in markets with more competition,HMO market penetration was not independently associatedwith profit rates.

 This suggests that competitionbetween HMOs may limit the ability of HMOs to reduceprices paid to hospitals and might also reduce thedegree of cost reductions and the potential effect onquality of care.

In this analysis, we use data from 1989 through 1996from the Agency for Healthcare Research and Quality'sHealthcare Cost and Utilization Project NationwideInpatient Sample (HCUP NIS) to examine the associationbetween changes in HMO penetration and changesin outcomes and treatment patterns for all AMI patients.Using discharges from repeated cross-sectional samplesof hospitals makes the selection bias inherent in singlecross-sectional studies less problematic and allows us tocontrol for baseline quality differences among patientsin different hospital markets to examine how outcomesand treatment patterns changed over time in associationwith changes in HMO penetration. In contrast toprevious work, we studied not only Medicare patientsbut also non-Medicare patients younger than 65 years,because the effects of price competition and reducedreimbursement may be most direct in this group.

While it is important to study Medicare and non-Medicare patients, there is some evidence that within agiven marketplace outcomes and treatment patternsmay be similar for all patients, although the price beingpaid for similar services may vary widely.

 For this reason,we focus on intertemporal changes in mortality andcardiac procedure use for all patients within each of our2 main groupings of patients (the Medicare and non-Medicare groups), as opposed to looking at changes inthese measures for patients with specific insurancetypes. In addition, we examine the role of HMO marketconcentration to determine whether the effects observedare a function of the competitiveness of HMO markets.We adjust for changes in the proportion of uninsured,Medicaid, and Medicare patients, as reimbursement forthese payers has been low relative to FFS indemnity;therefore, increases in the proportion of these low-payingpayers may also have affected the process and outcomesof care delivered within hospitals.

Discharges with AMI were chosen for this studybecause AMIs are common, the treatment objectivegenerally is to ensure survival, and the quality of careaffects the likelihood of survival.

 In addition, allpatients with recognized AMI are admitted on an emergencybasis, usually to the nearest hospital. Therefore,there is little opportunity for selection bias to affect thedecision about who gets admitted or risk-adjusted mortalityover time.

International Classification of Diseases, Ninth Revision,Clinical Modification [ICD-9-CM]

The database for this study was constructed from 3primary sources: the HCUP NIS, the Area Resource File,and the HMO Geographical Enrollment Data file (D. R.Wholey, PhD, Division of Health Services Research andPolicy, University of Minnesota, unpublished data,2001). We used data from 1989 to 1996 and includedindividuals who had a principal discharge diagnosis ofAMI (

 code 410.XX),excluding subsequent care (code 410.X2) (807 743 observations).These codes were chosen to focus on newAMI admissions, using the same approach taken by theCalifornia Hospital Outcomes Project.

 We excludedpatients who were not treated in hospitals located withinmetropolitan statistical areas (MSAs) (n = 100 701),as our HMO data were limited to patients in MSAs;patients with a coded primary diagnosis of AMI whowere discharged alive after a length of stay of less than2 days (n = 32 110), because it is likely that thesepatients did not experience an AMI; individuals youngerthan 18 years and older than 85 years (n = 47 483);patients with a length of stay greater than 30 days (n =9105); patients for whom neither discharge alive ordead was coded (n = 636); patients treated in governmentalhospitals (n = 58 210); and patients with agemissing (n = 4071), as age is an important risk adjuster.Because we were not able to determine their ultimateoutcome, we excluded individuals who were transferredto another hospital (52 591 observations). In addition,we excluded patients who went to hospitals at whichuninsurance rates, Medicare rates, or Medicaid rateswere coded as 0% for any year, on the basis that theinsurance variable was likely miscoded (n = 122 359).To use hospital fixed effects to control for unobservedheterogeneity among hospitals, we excluded dischargedpatients who were treated in hospitals for which we hadonly a single year of data (n = 40 413). The final sampleof this unbalanced panel consisted of 340 064 dischargeobservations from 457 hospitals from 1989 to 1996.

The Area Resource File was used to link the individualpatient-level data records from the NIS with MSAcodes. The HMO Geographical Enrollment Data file providedthe commercial HMO penetration and the numberof HMOs within each MSA for each year being studied.

We used an MSA-level measure of HMO penetrationand the number of HMOs within each MSA as a measureof HMO competition. We used an MSA-level measurerather than hospital-level measures because it is lesslikely to be endogenous, as HMOs might choose to contractwith better quality hospitals but would be unlikelyto make selective contracting decisions on the marketlevel based on differing levels of quality. Using an MSA-levelmeasure of HMO penetration also protects againstpotential hospital-level selection bias, which couldresult from significant changes in admission patternsof patients, with different-than-average AMI mortalityin conjunction with increases in HMO penetration.

Each discharge was grouped into 1 of 4 quartilesbased on the market-level HMO penetration for thathospital in that year: HMO penetration between 0% andless than 9.3% (quartile 1), HMO penetration between9.3% and 14.5% (quartile 2), HMO penetration between14.6% and 21.4% (quartile 3), and HMO penetrationgreater than 21.4% (quartile 4).

We calculated the percentage of uninsured,Medicare, and Medicaid patients in each hospital peryear in the sample using the NIS data.

Measurement of In-hospital Mortalityand Treatment Patterns

The principal outcome measure was death thatoccurred during the initial hospitalization, provided thelength of stay was less than or equal to 30 days. Toexamine changes in treatment patterns, we examinedcardiac catheterizations (

 codes 37.21-37.23and 88.55-88.57), coronary artery bypass graft (CABG)procedures (codes 36.11-36.19), and coronary angioplastiesor stents (codes 36.01, 36.02, 36.05, and 36.06)during the first 30 days of the initial hospitalization.

The method for risk adjustment developed byElixhauser and colleagues

 at the Agency forHealthcare Research and Quality was used, as thismethod has tested well in terms of superior discriminationvs other risk adjustment approaches for use withadministrative data.

 Variables in the risk adjustmentalgorithm include age, sex, peripheral vascular disease,paralysis, other neurological disorders, chronic pulmonarydisease, diabetes mellitus, diabetes mellituswith chronic complications, hypothyroidism, renal failure,liver disease, peptic ulcer disease, AIDS, lymphoma,metastatic cancer, solid tumor withoutmetastasis, rheumatoid arthritis, coagulopathy, obesity,weight loss, fluid and electrolyte disorders, chronicblood loss anemia, deficiency anemias, alcohol abuse,drug abuse, pyschoses, depression, and hypertension.

 We used patient-level logistic regressions with hospitaldummy variables to control for unobserved heterogeneityamong hospitals, and year dummy variables tocontrol for intertemporal changes common to all hospitalsin the sample. Hospital-level dummy variables areincluded because, even with the inclusion of many controlvariables, there may be unobservable characteristicsof areas that are not included but that arecorrelated with HMO market share and treatment patternsor in-hospital mortality.

To test for spillover effects, primarily whetherchanges in the proportion of patients with HMO insuranceaffected the outcomes and treatment patterns ofpatients with Medicare insurance, the sample was dividedinto 2 subsamples: Medicare and non-Medicarepatients. We examined treatment patterns and outcomesat the discharge level in both subsamples. Thedependent variables were mortality and 3 cardiacprocedures: catheterization, angioplasty, and coronaryartery bypass graft (CABG). In this way, by examiningthe Medicare subsample, we were able to test whetherthere was a spillover effect from changes in HMO penetration,Medicaid, or the percentage of uninsured on theoutcomes and treatment patterns of the Medicare population.We examined the non-Medicare patient subsampleseparately, as nearly all studies of HMO spillovershave examined Medicare patients only, and this allowedus to examine whether there were changes in outcomesand treatment patterns based on changes in payer mixwithin the non-Medicare population.

Patient-level logistic regression analysis was used toexamine the relationship between changes in HMO penetration,HMO market concentration, and the outcomeand treatments of interest. To control for baseline qualitydifferences among the hospitals, hospital fixed-effectsdummy variables were used. Through the use ofhospital fixed effects, the outcome (death or survival) isinterpreted as a function of within-hospital changes inpayer mix, as variation in time-invariant hospital characteristicsis accounted for by the hospital fixed-effectsdummy variables. Because we used quartiles of HMOpenetration to measure HMO penetration and becauseHMO penetration increased in every market during thestudy, the estimated regression coefficients examine theeffect of interquartile increases in HMO market shareon treatment patterns and in-hospital mortality. Patientcharacteristics such as age, sex, and clinical risk factorswere used as controls. Coronary artery bypass graft surgery,angioplasty, and cardiac catheterization weremodeled as dependent variables in a fashion similar tothat used for mortality. We also controlled for intertemporalchanges common to all hospitals in the sample byusing dummy variables for each year.

To compensate for the fact that our analyses of HMOpenetration were correlated at the MSA level, we adjustedour standard errors for clustering at the MSA level.Analyses were done using SAS software (SAS Institute,Cary, NC), with adjustment of standard errors for clusteringdone using STATA (Stata Corporation,College Station, Tex).

The hospitals included in the NISchanged from year to year based on theNIS sampling criteria. In 1989, our sampleincluded 180 hospitals, and in 1996, oursample included 238 hospitals. Our totalsample included 457 hospitals that contributeddata for at least 2 years, whichenabled the use of hospital fixed effects.Mean HMO penetration increased from18.8% in 1989 to 27.5% in 1996. Therewere no large changes in the mean numberof HMOs within each MSA. The percentageof uninsured decreased from 6.0%to 4.1%. Changes in payer mix for each ofthe other classes of payers were small(

). Overall, in-hospital mortality for AMIpatients declined from 14.6% to 9.3% during this period,while cardiac catheterization, angioplasty, and CABGrates rose substantially. Hospitals in the sample in 1989and hospitals in the sample in 1996 were not necessarilythe same hospitals.

Measurement of the HMO effect on changes in outcomesand treatments over time has the potential to bebiased if there are underlying unmeasured differencesin areas that attracted HMOs to a greater degree thanothers. For instance, HMO penetration might be morelikely to increase in areas in which the population isyounger or healthier. To investigate whether there weresystematic differences between markets in which HMOpenetration increased or did not, we comparedobserved patient characteristics in these 2 types of markets(

). There were 96 124 patients in hospitalsthat did not change their quartile of HMO penetrationduring the study. There were 243 940 patients in hospitalsthat experienced interquartile change in HMO penetration.Although there are several statisticallysignificant differences in the proportion of patients withthese comorbidities between these 2 groups, the magnitudeof the differences in all cases is small and does notfavor one group or the other.

We present results separately for the Medicarepatients and the non-Medicare patients.

. Among Medicare patients, wefound extensive evidence for spillover effects fromincreases in HMO penetration, particularly in terms ofchanges in treatment patterns. Odds ratios (ORs) thatwere significantly lower than 1 were found with eachinterquartile increase of HMO penetration for cardiaccatheterization and angioplasty and for patients in hospitalmarkets in which HMO penetration increased intothe highest quartile for CABG (OR, 0.84; 95% confidenceinterval [CI], 0.74-0.95) (

). For each ofthese procedures, the OR stayed the same or decreasedwith increasing levels of HMO penetration, such thatthe ORs were lowest in hospital markets in which HMOpenetration increased into the highest quartile. Themagnitude of these coefficients means that, for everyinterquartile increase in HMO penetration (roughly 5%-7%, depending on the quartile), the odds of a MedicareAMI patient receiving cardiac catheterization decreasedby 12% to 17%, the odds of receiving an angioplastydecreased by 7% to 12%, and the odds of receiving aCABG decreased by 3% to 16%.

There was no evidence that increases in HMO penetrationhad a significant effect on inpatient mortality, aspatients in hospitals that experienced increases fromthe first quartile to the second quartile of HMO penetrationhad a slightly increased odds of mortality (OR,1.07; 95% CI, 1.01-1.13), but patients in hospitals inwhich HMO penetration increased into the third orfourth quartile did not experience an OR of mortalitysignificantly different than 1. Increases in the numberof HMOs within an MSA, our measure of HMO competition,were associated with small but significant increasesin cardiac catheterization (OR, 1.03; 95% CI,1.01-1.04) and angioplasty (OR, 1.02; 95% CI, 1.00-1.04) rates. There were no significant effects of thechange in the number of HMOs on mortality or CABGrates.

Increases in the hospital's proportion of uninsured orMedicaid patients had no statistically significant effecton the mortality risk for Medicare patients or the oddsof receiving cardiac catheterization, angioplasty, orCABG. Increases in the percentage of Medicare patientswere associated with reductions in the odds of mortality(OR, 0.71; 95% CI, 0.53-0.96) and in the odds ofbypass surgery (OR, 0.50; 95% CI, 0.25-0.99), but hadno significant effect on the odds of cardiac catheterizationor angioplasty.

Using fixed effects to control for unobserved heterogeneityin a logistic regression analysis may result inbiased coefficient estimates in cases in which there arefew events per hospital. As a test for robustness and tomitigate the potential for bias, the regression for eachdependent variable was also run using a data subsamplelimited to patients in hospitals with 10 or more occurrencesof each dependent variable. We found no evidenceof biased coefficients in any of the specifiedregression models.

. We found less evidence ofchanges in treatment patterns and in-hospital mortalitydue to changes in HMO penetration in the non-Medicaresample. Interquartile changes in HMO penetration hadno significant effect on mortality, cardiac catheterization,or CABG rates. Increases into the third quartile ofHMO penetration were associated with a reduced oddsof angioplasty (OR, 0.90; 95% CI, 0.82-0.99), butincreases into the second or fourth quartile of HMOpenetration were not associated with ORs of angioplastythat were significantly different than 1. Increases inthe number of HMOs within an MSA, our measure ofHMO competition, were associated with a small but statisticallysignificant increase in angioplasty rates (OR,1.02; 95% CI, 1.01-1.04), but no significant change inthe odds of mortality, cardiac catheterization, or bypasssurgery (

Increases in the percentage of uninsured had no significanteffect on the odds of mortality, cardiaccatheterization, or angioplasty, but were associatedwith a significant reduction in the odds of all non-Medicare patients receiving bypass surgery (OR, 0.17;95% CI, 0.03-0.84). Increases in the percentage ofMedicaid patients were also associated with significantreductions in the odds of bypass surgery (OR, 0.15; 95%CI, 0.05-0.49), but had no significant effect on the oddsof mortality, cardiac catheterization, or angioplasty.The reduction in the odds of receiving bypass surgeryfrom a 1% increase in the percentage of uninsured orMedicaid patients is large, but given that the baselinerates of each of these is only about 4%, a 1% increaserepresents a potentially substantial additional financialburden. Changes in the percentage of Medicare patientshad no significant effects on any of these measureswithin the non-Medicare population.

There was no evidence of bias in any of the studyvariable coefficients in subsamples of the data that werelimited to patients in hospitals with 10 or more occurrencesof the dependent variable.

Our results indicate that increases in HMO penetrationfrom 1989 to 1996 significantly affected treatmentpatterns for Medicare AMI patients within the HCUPNIS, and these effects were largest in the markets inwhich HMO penetration increased into the highestquartile. The relative risk of receiving cardiac procedureswas reduced by about 10% to 15% at hospitalswith increases in HMO penetration into 1 of the 3 upperquartiles. Given the rates at which patients receivedthese procedures, the magnitude of this relativedecrease is about 1% to 2% for CABG, 2% to 3% forangioplasty, and 5% to 7% for cardiac catheterization.We found no evidence of any increased mortality riskfrom increases in HMO penetration. Increases in thenumber of HMOs, suggesting greater HMO competition,were associated with modest increases of approximately2% in the rates of cardiac catheterization (eg, 1.0%,given a baseline rate of about 50%) and angioplasty(0.5%) after AMI. We believe that these estimates areunbiased by cross-sectional differences in case-mix orhospital characteristics, because we used a measurementframework that provides estimates of within-hospitalchanges in response to different degrees of changein MSA-level HMO penetration.

In the non-Medicare population, we found little evidencethat treatment patterns or in-hospital mortalitywas affected by increases in HMO penetration.However, increases in the proportion of uninsuredpatients and Medicaid patients were associated withsignificant reductions in the odds of receiving bypasssurgery for all non-Medicare patients. Because care forsuch patients (to the degree that Medicaid reimbursementdoes not cover marginal costs) is funded out ofhospital margins, these results are not surprising.

Our results on treatment patterns and mortality inthe Medicare population are similar to those found inthe study by Heidenreich et al,

 who showed that higherrates of HMO penetration were associated with lowerrates of coronary angiography (relative risk, 0.93; 95%CI, 0.86-1.01) but no significant differences in AMImortality. We did not observe reductions in mortality,as Mukamel and colleagues

 did (a small reduction of0.15%). Reductions in cardiac procedure use are consistentwith the literature that has demonstrated reductionsin expenditures in areas with bigger increases inHMO penetration.

Given that there were no changes in mortality riskas a result of changes in HMO penetration, the ramificationsof the observed changes in treatment patternsfor quality of care are difficult to assess. From theadministrative data contained in the NIS, we cannotdetermine the appropriateness of the treatments thatindividual AMI patients received. Therefore, we do notknow whether the changes in treatment patterns are ofa beneficial or detrimental nature, as it depends onwhether the procedures received were appropriate andnecessary, or not.

An important limitation of this study is that we studiedonly patients with AMI; therefore, we cannot makeinferences about how outcomes and treatment patternsfor other conditions were affected by changes in HMOpenetration. We were limited to in-hospital measures ofmortality, but the mean length of stay for all AMIpatients decreased to a similar degree in hospitals inthe areas with the biggest and smallest increases inHMO penetration, so these measures would not havebeen biased by differential changes in length of stay. Inaddition, some investigations focused on quality of carehave found results robust to using inpatient deaths ordeaths within 30 days of admission as measures of quality.

 The HCUP NIS did not code whether patients wereHMO patients; therefore, we could not directly analyzespillover effects among non-HMO patients younger than65 years or among Medicare patients who were notHMO enrollees. However, we think this was unlikely tohave any significant effect on our results, as the percentageof Medicare HMO patients nationwide wassmall during 1989 to 1996, and Baker

 demonstratedthat overall HMO market share, as opposed to MedicareHMO market share, predicts spending among Medicarepatients in several different investigations. It would alsobe useful to estimate the managed care effects byincluding preferred provider organization data, but suchdata were not available on a national level for all MSAsat the time of the study. Furthermore, HMOs have beenthe managed care entities that have typically mostaggressively negotiated with providers, exerted influenceover treatment decisions, and attempted to affectthe practice patterns of physicians.

Articles

POLICY

The Effect of Increases in HMO Penetration and Changes in Payer Mix on In-hospital Mortality and Treatment Patterns for Acute Myocardial Infarction

 The objectives of this drug utilization review programwere (1) to increase &#946;-blocker prescribing to fee-for-servicepost-acute myocardial infarction (AMI) Medicaid patients; (2) toimprove compliance among patients who were prescribed &#946;-blockers post-AMI; and (3) to evaluate the economic implicationsof increased &#946;-blocker prescribing.

 The intervention targeted physicians ofPennsylvania Medicaid recipients who had an AMI betweenNovember 1, 1998, and November 1, 1999. Educational materialswere sent to the physicians of post-AMI patients not receiving &#946;-blockers. Preintervention and postintervention rates of &#946;-blockerprescribing in the Medicaid program within 7 and 30 days of dischargeafter an AMI hospitalization were compared. Similarly, pre andpostintervention compliance rates were compared for AMIpatients who were prescribed &#946;-blockers. Cost savings and numberof avoided deaths also were calculated.

 There was a 5.5% to 6.9% increase in &#946;-blocker prescribingafter the intervention, depending on the follow-up period.Postintervention AMI patients were 16% more likely to beprescribed a &#946;-blocker. There was an 8.3% increase in patientcompliance with &#946;-blocker therapy from preintervention to postintervention.In the first 2 years of the intervention, the estimated costsavings to the Pennsylvania Medicaid program ranged from$71 970 to $76 678, respectively. An estimated 3 deaths wereavoided.

 The intervention resulted in increased appropriateprescribing and compliance with &#946;-blockers among post-AMIpatients. There also were estimated cost savings to PennsylvaniaMedicaid as a result of reduced hospitalization, and fewer deaths.

Patients who survive an acute myocardial infarction(AMI) are at an increased risk for suddendeath, nonsudden death, and reinfarction.

 In thepast 2 decades, numerous trials have demonstrated that&#946;-blockers can improve survival and reduce the rate ofreinfarction in patients with AMI.

 However, publishedreports indicate that &#946;-blockers are underutilized.

Most educational programs targeting hospital physicianshave successfully increased &#946;-blocker prescribingduring hospitalizations for AMI.

 These examplesshow that intensive educational programs can influenceprescriber behavior and increase the proportion ofpatients who receive &#946;-blockers after an AMI. However,these interventions were directed at hospital physicians.How these results might apply to a less intensiveeducation program targeting ambulatory care physiciansis unclear.

Once AMI patients are prescribed a &#946;-blocker, therapyshould continue indefinitely, unless a complicationarises. Therefore, compliance becomes an issue. In theBeta-Blocker Heart Attack Trial, investigators foundthat patients who took &#8804;75% of their prescribed &#946;-blocker regimen were 2.6 times more likely to die withinthe first year of follow-up than more compliantpatients.

 Clinical data relative to long-term use of lifesavingdrugs in 156 survivors of definite myocardialinfarction at a government, university-affiliated teachinghospital were analyzed over a 24-month follow-upperiod. Among post-AMI patients who were dischargedhome with a &#946;-blocker prescription, only 71% remainedon their &#946;-blocker medication at 24 months.

Among intervention programs to increase the use of&#946;-blockers after AMI, economic evaluations have notbeen done to our knowledge. Bradford et al believedthat there are 3 types of costs associated with underutilizationof &#946;-blockers: (1) costs due to higher mortalityor morbidity (reduced health) attributed to suboptimalcare; (2) increased use of medical resources (eg, hospitaladmissions, physicians visits) because of less appropriatetreatment; and (3) costs of less effectivetherapies.

 Due to the paucity of literature on costimplications of &#946;-blockers, Bradford and colleaguescould only calculate the first type of costs: years of lifelost due to underutilization of &#946;-blockers in the UnitedStates and the associated dollar value of the loss.Soumerai et al estimated that &#946;-blockers could reducethe risk of readmission to hospitals for all cardiac eventsby 22%.

 However, the cost consequences of cardiacevents can vary significantly, which hinders the attachmentof a dollar value to the reduction in hospitaladmissions. Goldman et al, in a widely cited article,examined the cost of &#946;-blockers per year of life saved.

Phillips et al used a computer-simulation Markov Modelto estimate that the incremental cost per quality-adjustedlife-year gained would always be less than$11 000.

 However, the cost implications in terms oflife-years saved often are not the primary concern ofintervention programs. On the contrary, the interventionprograms are most interested in the cost savings tothe payers.

The overall goal of this educational intervention programwas to increase the prescribing of &#946;-blockers tofee-for-service Medicaid patients immediately after ahospitalization for AMI. A secondary project aim was toimprove compliance among patients who were prescribed&#946;-blockers post-AMI, but who were refillingtheir prescriptions at intervals that suggested poor compliance.The third aim was to evaluate the economicimplications of an increase in &#946;-blocker prescribing.

Educational materials were produced and distributedspecifically for this project by the Pennsylvania MedicalSociety's Center for Professional Drug Education.Physicians for the intervention were identified based onclaims data. They were put into 1 of 3 groups: (1) thosewith post-AMI patients who were noncompliant with &#946;-blocker therapy; (2) those with post-AMI patients whohad not been prescribed a &#946;-blocker; and (3) all otherphysicians with a specialty identified by the PennsylvaniaMedical Society that had the potential to treat a post-AMIpatient (eg, internal medicine, family medicine).Cardiologists were excluded from the first 2 interventiongroups because it was assumed that most patients werefollowed by their primary care physician after dischargefrom the hospital, and that cardiologists were more likelyto prescribe &#946;-blocker therapy when appropriate.

In the first group, prescribing physicians were identifiedfor those patients who had 1 or more &#946;-blockerprescriptions, but had &#946;-blocker availability rates below80%. Beta-blocker availability rates were defined as thepercentage of time when &#946;-blockers were available forthe patients after the initiation of &#946;-blocker use post-AMI. It was calculated by dividing the &#946;-blocker availabledays by episode days for all recipients with 1 ormore &#946;-blocker claims after their index AMI hospitalization.Beta-blocker available days were determined bycounting days of drug availability based on the days supplyfrom the dispensing date forward. Episode days werethe number of days from the first &#946;-blocker prescriptionafter the AMI through termination of Medicaid fee-forserviceeligibility, or the end of the study period, ifearlier. Days hospitalized were subtracted from thedenominator and patients with negative &#946;-blocker availabilityrates due to extended hospitalizations wereexcluded. Compliance was defined as having &#946;-blockeravailability rates equal to or above 80%. Patients with &#946;-blocker availability rates below 80% were considerednoncompliant.

Patient drug history profiles were created for thosepatients for whom compliance was an issue. A case-by-casereview of the profiles was conducted to excludefalse-positives, which were patients who were labeled"positive" but who were in fact compliant based onimplicit review by Pennsylvania Medical Society staff.These false-positives were probably due to administrativeor data entry errors. From 309 "positive" patients,135 were excluded.

Two intervention educational packets were developedand administered. One packet was mailed to thephysicians whose patients were identified as noncompliant.The educational materials included an introductionletter, the 

 newsletter, a Medicaidpatient report, a response form, and each patient's drughistory profile.

The introduction letter discussed the review aimsand reported the overall post-AMI &#946;-blocker prescribingand compliance rates in Pennsylvania Medicaid. The

 newsletter, which offered continuingmedical education credit, included current strategies inthe treatment of myocardial infarction, identified thefactors contributing to compliance problems amongpatients, and laid out actions that physicians need totake to increase patient compliance rates. The Medicaidpatient report listed the physician's noncompliantpatients. In the drug history profiles, &#946;-blocker prescriptionswere grouped separately and listed first. Thephysician response form was included with the educationalmaterials to assess the acceptance of the interventionby the physicians.

The other packet was mailed to physicians whosepatients were identified by the underutilization criterion.These physicians received an introduction letterand the 

 newsletter. If a physician hadpatients who met both criteria, they only received thecompliance educational packet. In total, 328 physiciansreceived "underutilization packages," 157 physiciansreceived "noncompliant packages," and 10 972 physicians(the third group of physicians) received only the

 newsletter. We were unableto determine what percentage of AMIpatients saw these physicians in the postinterventionperiod.

The timeline of the study was as follows:(1) Physicians were identified fromNovember 1, 1998, through November 31,1999. (2) The intervention packages weremailed to them on April 28, 2000. (3) Twoweeks later a follow-up certified letter wassent to each nonresponder. (4) General mailingof 

 newsletters to physiciansin the third group took place on May 1,2000.

To evaluate the impact of the educationalprogram on the prescribing of &#946;-blockerspost-AMI, we calculated the proportion ofAMI patients with a paid claim for a &#946;-blockerwithin 7 and 30 days of discharge.Medicaid recipients who were hospitalizedfor an AMI between November 1, 1998, andOctober 31, 2001, were identified throughMedicaid inpatient administrative claims.Patients were classified into 1 of 3 periodsbased on the timing of their AMI hospitalizationwith respect to our educational program:preintervention (November 1, 1998, toMarch 31, 2000); during the interventionperiod (April 1, 2000, through May 31,2000); and postintervention (June 1, 2000,through October 31, 2001). Patients wereexcluded if they were less than 21 years old,if they died during the first (index) AMI hospitalization,or if they had no prescriptionclaims during the study period.

To avoid overestimation of &#946;-blocker use,only the first AMI hospitalization within thestudy period was selected. Patients withavailable &#946;-blockers, based on prescriptionhistory immediately before the AMI hospitalization,were counted as receiving &#946;-blockers in both the 7 andthe 30-day analyses. Patients without a pharmacy claim100 days before the index AMI hospitalization wereexcluded from this analysis to limit analysis to thosewith prescription benefits. Because this exclusion hadsuch a large effect on the denominator (almost one halfof the patients were ineligible because they had no priorprescriptions), it would tend to result in overestimationof the percentage of &#946;-blocker use in the population.Because it is applied consistently over each time period,it shouldn't affect the relative difference from 1 periodto the next. Finally, because of missing pharmacy data,patients with fewer than 7 (or 30) days of continuousMedicaid eligibility, or an inpatient or psychiatric hospitalizationduring these postdischarge periods, wereexcluded from the 7-day and/or 30-day utilization calculationsas appropriate.

The economic effects of the intervention programwere evaluated in terms of (1) cost savings to thePennsylvania Medicaid fee-for-service program and (2)number of deaths avoided in the PennsylvaniaMedicaid program. The outcomes of patients who weredischarged from hospitals after AMIs were categorizedas sudden death, nonsudden death, nonfatal reinfarction,and other, as reported in the 

 (probabilitieswere based on the meta-analysis by Yusuf et al

).This figure was generated by using Data, Version 3.5,&#169;No[1988-1998] Treeage Software, Inc,&#174; of Treeage Inc,Williamstown, Massachusetts, USA.

 When estimatedfrom the perspective of the Pennsylvania Medicaid program,the economic effects of the intervention programcould result from the number of nonsudden deaths andnumber of nonfatal reinfarctions avoided. The totalcost savings of the intervention was considered to bethe sum of the reduced hospitalization costs for treatingfewer patients with nonsudden deaths and thereduced costs for treating fewer patients with nonfatalreinfarctions.

The data points were estimated as follows: The ratesof nonsudden deaths for patients who were put on &#946;-blockers (4.1%) and for those who were not (4.6%),within an average follow-up period of approximately 2years, were based on an estimate from the meta-analysisby Yusuf et al.

 From the same meta-analysis, therates of nonfatal reinfarction were estimated to be 5.7%in the &#946;-blocker group and 7.5% in the control groupwithin the same follow-up period.

 The hospitalizationcost of nonsudden AMI-related deaths was estimated tobe $21 679, which was adjusted to 2001 dollars from astudy that reported direct medical costs of treating afatal AMI.

 The hospitalization cost for a patient whosurvived an AMI was estimated to be $19 326, whichwas adjusted to 2001 dollars from a study on the costof ischemic complications.

In addition to estimating the cost savings to thePennsylvania Medicaid program, we also estimated thetotal number of deaths avoided due to the intervention.The reduction in mortality rates due to increased use of&#946;-blockers was based on the meta-analysis by Yusuf etal, which reported that &#946;-blockers could reduce themortality rate from 10.0% to 7.9% in the long term.

 Thetotal number of AMI patients considered (4914) wasbased on the total number of unique patients with AMIhospitalizations after excluding multiple AMIs.Estimations of cost savings and deaths avoided were calculatedfor the actual increases seen in &#946;-blocker use at7 days and 30 days post-AMI.

The retrospective drug utilization review programunder which this study was conducted was reviewedand deemed to be exempt by the institutional reviewboard of the University of Maryland (IRB protocol#0101412).

There were 5241 AMI hospitalizations during the 3-year study period. With exclusions for multiple AMIsduring the study period (n = 327), patients with no prescriptionclaims within 100 days before the index AMI(n = 2408), and patients without follow-up time afterdischarge (n = 1002), the project evaluation was basedon the remaining 2543 AMI patients (these categories ofpatients overlapped). As reported in 

, the majorityof the patients were classified into the preinterventionor postintervention period based on the timing oftheir AMI hospitalization with respect to our educationalprogram. They were mostly women and more than 50years of age. Comparing the preintervention and postinterventiongroups, there were no statistically significantdifferences in age and sex. However, compared with thepreintervention group, the intervention group has asmaller proportion of patients under age 50 (

In the preintervention period, 46.4% of patients withan AMI hospitalization filled a &#946;-blocker prescriptionwithin 7 days and 61.3% within 30 days of discharge(

). These percentages increased to 49.6% of AMIpatients at 7 days and 64.7% at 30 dayspostintervention. These results translatedto a population-wide 6.9% increase in &#946;-blocker prescribing immediately after anAMI hospitalization (within 7 days of thedischarge). However, these increases werenot statistically significant (&#967;

Using a multivariate proportional-hazardsmodel to adjust for age group, sex,and race, patients with their first AMI inthe postintervention period were 16%(hazard ratio = 1.16, 95% confidenceinterval = 1.05, 1.29) more likely to beprescribed a &#946;-blocker than AMI patientsbefore the intervention period (

Based on prescription refill dates, 64.1%of preintervention &#946;-blocker users had &#946;-blockers available 80% or more of the time (

).These patients were considered "compliant with therapy."Compliance increased to 69.4% of &#946;-blocker usersafter the intervention. This represents an 8.3% increasein compliance from preintervention to postinterventionamong &#946;-blocker users in Pennsylvania Medicaid. Thisincrease is statistically significant (

The projected cost savings of the intervention programare reported in 

. The difference betweenthe total costs before and after intervention was the estimateof cost savings due to the intervention. The costsavings of the intervention program during the first 2years were estimated to be in the range of $71 970 to$76 678.

Based on the changes in &#946;-blocker prescribing fromthe preintervention to the postintervention period andthe reduction in mortality rates reported in the literaturewhen patients are put on &#946;-blockers, we estimatedthat 3 deaths were avoided.

The University of Maryland, in conjunctionwith the Pennsylvania Medical Societyand Commonwealth of Pennsylvania'sDepartment of Public Welfare, Office ofMedical Assistance Programs, designed andimplemented a systemwide educational programto increase physician awareness of thecurrent treatment guidelines for AMI survivors.A program goal was to increase prescribingof &#946;-blockers immediately after anAMI hospitalization among Medicaidenrollees. This goal is closely aligned with aHealth Plan Employer Data and Information Set(HEDIS&#174;) measure of health care quality: the proportionof patients prescribed a &#946;-blocker within 7 daysafter an AMI.

 Before the educational program, 46.4% ofeligible enrollees filled a &#946;-blocker prescription within 7days and 61.3% within 30 days of hospital discharge.After the education program, the proportion of AMIpatients on &#946;-blockers increased 6.9% and 5.5% at 7days and 30 days, respectively.

Although these increases appear modest, they applyto all fee-for-service patients with prescription benefits,and thus include a large absolute number (approximately5000) of patients. Furthermore, because theintervention was programwide, materials were disseminatedto most general practitioners in Pennsylvania,and as such, are likely to have some impact on care ofAMI patients well beyond the study population. In addition,due to funding limitations, our educational interventionwas carried out in a simple and low-costmanner relative to more intensive interventions such asacademic detailing. However, our program resulted inan increase in &#946;-blocker use despite the larger geographicarea of our intervention and wider dispersion ofindependent physicians in the ambulatory care settingcompared with interventions in hospital settings.

The second program goal, to improve compliancewith &#946;-blocker therapy among &#946;-blocker users, was reasonablysuccessful. Good compliance was defined ashaving medication available for at least 80% of the daysafter initiation of &#946;-blocker use. We found an 8.3%increase in the proportion of compliant patients after thesystemwide physician education program with a personalizedmailing to physicians of noncomplaint patients.

Applying the increase in &#946;-blocker use at 30 days toall 4914 AMI patients (regardless of prior prescriptionclaims) resulted in an estimated savings of up to $76 678and 3 deaths avoided. Because only nonsudden deathsresult in consumption of healthcare, the analysis on costsavings only included nonsudden deaths. For patientswith nonsudden deaths or nonfatal reinfarction, only hospitalizationcost was included in this analysis becausehospitalization cost is the largest component of resourceuse due to AMI. As such, these estimated cost savingsshould be viewed as a lowerbound. Actual cost savingsmay be greater.

The cost savings due toincreased prescribing of &#946;-blockers do not come as asurprise; &#946;-blockers havebeen established as cost effective for secondary preventionof AMI. In the most widely cited study on thecost effectiveness of &#946;-blockers at discharge forAMI, Goldman et al reporteda cost per year of life saved in the range of $2327 to$13 571 (in 1987 dollars).

 Kuntz and Lee updated theresults by Goldman and estimated a cost per life-yearsaved to be from $3 400 to $19 800.

 Both before andafter being updated, these cost-effectiveness ratios arevery advantageous, because treatments with cost-effectivenessratios in the range of $20 000 to $30 000 per life-yearsaved are generally considered highly costeffective.

These analyses were limited to Medicaid recipientswho have prescription benefits as determined by aprescription claim before their AMI hospitalization.Because we used prescriptions as a proxy for drugcoverage, approximately one half of the AMI patientswere excluded because they did not have a prescriptionclaim within 100 days of their AMI. This wasdone to avoid underestimating &#946;-blocker use.

There are several factors inherent in the study design,and particularly the use of claims data, that would affectour estimate of the impact of the intervention. One factoris misclassification of persons who did not have anAMI as having an AMI hospitalization. Only records ofpatients with an AMI indicated as a primary diagnosis onthe hospital claim were studied. Because the diagnosis ofAMI was not validated against the medical record, itcould be miscoded, particularly in cases where a suspectedAMI was later ruled out. Also, patients in whom&#946;-blocker therapy is considered optional because theyhad a small myocardial infarction with no evidence ofcontinuing ischemia could not be identified.Furthermore, patients who took home medication fromthe hospital (eg, free samples) may not fill their prescriptionuntil that supply is exhausted. This also wouldcause us to underestimate the number of patients on &#946;-blockers within 7 days, though use of free samples isunlikely to bias the 30-day estimate.

In our analysis, we included all post-AMI patients inthe calculation of &#946;-blocker prescribing, although somepatients might not have been on &#946;-blockers due to contraindications.Determining the medical conditions ofpatients accurately by reviewing medical records wasnot possible with this study.

The evaluation of compliance depends on a numberof assumptions. First, patients' average daily dose had tobe estimated in order to calculate the number of pillsavailable. Although there are typical or standard doses,compliance may be underestimated among patients whosplit pills and overestimated among patients who take anunusually high dose. However, we assumed standarddoses. Additionally, we used drug availability as a proxyfor adherence, assuming patients took the medicationthat was available and did not refill their prescriptionuntil they used all of their available supply. An accuratedetermination of compliance typically requires directobservation of a patient's daily medication-taking behaviorand/or direct assessment of drug levels in the body.

The systemwide physician education program inPennsylvania Medicaid program increased &#946;-blockerprescribing after AMI hospitalization by increasingphysicians' awareness of the guidelines for treatment ofAMI survivors. The educational intervention also improvedpatients' compliance with &#946;-blocker therapy.These effects are likely to apply to AMI patients wellbeyond the study population. Besides clinical effects,this intervention program also led to cost savings forthe Pennsylvania Medicaid program, as well as avoidanceof a few deaths.

The authors thank Lori Walker, BS, of the School ofPharmacy's Pharmaceutical Research Computing, for her analyticand programming support.

MANAGERIAL

Impact of an Educational Intervention for Secondary Prevention of Myocardial Infarction on Medicaid Drug Use and Cost

 Clinical trials do not represent community settings,making widespread implementation of evidence-based medicineproblematic. New heart failure treatments are an example, asresults comparable to those of clinical trials have not beenobserved in the community. Alternatives to clinical trials couldprovide useful complementary information.

 To review the clinical trials and communityexperiences in heart failure management by searchingPubmed with key words "observational studies," "clinical trials,"and "heart failure," to present the preliminary results of a community-based heart failure registry as a complementary database, andto assess the potential value and limitations of the registryapproach.

 Recent advances in the treatment of heart failure led toguidelines using clinical trial evidence as the rationale for transferringnewer therapeutic technologies to the community practice setting.Implementation of such guidelines is slow, reflecting concernsover applicability of clinical trial results to the community setting.A community-based registry of &#946;-blocker treatment for heart failureshowed outcomes comparable to those of clinical trials, despitesignificant differences between physicians and their patients inthese settings.

 Registries can complement clinical trials to expeditetechnology transfer to the community setting.

Evidence-based medicine relies on results of clinicaltrials, which may be problematic as clinicaltrials often do not represent patients or careproviders from the broad community practice setting.Clinical trials are carried out by investigators who areusually based in academic centers and have a highdegree of experience, specialized training, or interest inthe particular clinical problem being investigated.Patients enrolled into these trials are selected by criteriathat tend to optimize their responsiveness to thequestion being investigated. Therefore, the results ofclinical trials may not be applicable to the communitysetting.

The transfer of technology from clinical trials to thecommunity setting occurs slowly, as the "evidence" forevidence-based medicine derives from sources not readilyapplicable to the community setting. The usualapproach for expediting the transfer of new technologyto the community setting is to expand efforts aimed ateducating about the results of clinical trials and urgingtheir rapid and broad adoption. Inherent in thisapproach is the assumption that clinical trials aloneoffer sufficient supportive evidence and rationale. Thecalls for increased educational efforts often emanatefrom the clinical trialists, who may have biased confidencein their findings. Therefore, this approach maynot consider the shortcomings of clinical trials. Anotherinterpretation of the slow transfer of technology is thatthe community practitioner has been educated but isskeptical about the applicability of clinical trials resultsto his or her practice setting. In that case, further educationabout clinical trial results would likely have littleeffect.

The objective of this article is to examine the potentialrole of alternative means, especially registries, as investigativetools for collecting valid and useful information tocomplement clinical trial data and to expedite the transferof technology from the clinical trial to the communitysetting. There exists a so-called efficacy-effectivenessgap in the applicability of clinical trial results, especiallyin the contexts of managed care, disease managementprograms, and pharmacoeconomics.

 The limitedamount of reliable effectiveness data affects healthcareproviders and health policy makers in their ability tomake well-informed decisions and formulate recommendations.

 Heart failure represents an area in which anefficacy-effectiveness gap has been recognized. Therefore,we used heart failure as a model for reviewing theroles of clinical trials and registries. The preliminaryresults of a registry of patients beginning treatment with&#946;-blockers for heart failure are reviewed as an exampleof these methods.

CLINICAL TRIALS AND COMMUNITYEXPERIENCES IN HEART FAILURE

There have been great advances in our understandingof the pathophysiology of heart failure and its management.The favorable results of new treatments, suchas vasodilators, angiotensin-converting enzyme (ACE)inhibitors, and &#946;-blockers, on clinical outcomes haveled to the development of heart failure management andtreatment guidelines that are being updated at shorterintervals.

 The latest guidelines rely on evidence-basedmedicine as the rationale for transferring these newerdiagnostic and therapeutic technologies from the realmof clinical research to the community setting of practicingcardiologists and primary care providers, neither ofwhom necessarily has specialized interest or expertisein heart failure. Therefore, the rate and extent of implementationof the latest technologies in the communitysetting appear limited. The scientific basis for and clinicalapplication of &#946;-blockers in heart failure were thesubjects of recent reviews that concluded that "[t]hescience supporting &#946;-blockers must be translated intopractice safely and rationally if the agents are to achievetheir full potential."

The concerns in heart failure are that clinical trialsare usually carried out by experts in the field andinclude patients who are predominantly younger whitemen, selected to have stable symptoms, little comorbidity,and rigidly defined criteria for heart failure (eg,reduced left ventricular ejection fraction). Large-scalecommunity observations indicate that the general populationwith heart failure contains significantly morewomen, African Americans, and older patients thanthose included in clinical trials.

 In addition, almosthalf of the patients in the community setting have normalsystolic function, with possible diastolic dysfunction.

 Furthermore, most of these patients aremanaged by primary care providers who are not cardiologistsor do not have specialized expertise or interestin heart failure. In fact, the ability to replicate theapparent improvements in the management and outcomesof heart failure outside the clinical trial settinghas been questioned, as results comparable to thosereported from clinical trials have not been observed inthe broad community, especially among groups inadequatelyrepresented in the clinical trials.

 The mostrecent heart failure management guidelines recognizethis problem and call for further investigation in sub-populationswith heart failure.

When carvedilol became the first &#946;-blocker approvedfor use in heart failure in the United States, there was areluctance among practitioners to use it in the communitysetting, because &#946;-blockers were previouslythought to be contraindicated in heart failure and it wasperceived as difficult to apply this new therapeutic technologysafely and effectively. Therefore, a &#946;-blockerheart failure registry was designed and implemented toenroll patients starting a regimen of carvedilol for heartfailure in the usual care setting and to follow them withprospective longitudinal observations.

Complete details of the registry design have beenpublished elsewhere.

 Specific objectives were to collectoutcomes data and to observe the experience usinga &#946;-blocker, carvedilol, in unselected patients with heartfailure managed by community physicians in their usualpractice without a structured protocol. There were nopatient selection criteria other than that patients mustbe adults starting a regimen of carvedilol for heart failure.The decision to prescribe carvedilol was at the discretionof the participant physician, without applyingany prespecified eligibility criteria. The physicians participatingin the registry were selected to be representativeof community practitioners in the United Statesand Canada. They included cardiologists and primarycare physicians, and special care was taken to minimizethe inclusion of physicians from academic settings orwith a special interest in heart failure. There was noprespecified schedule of visits or procedures to be followed,as patients were seen according to the usualpractice of the participant physician, who performedonly those procedures or assessments (eg, echocardiogramsand laboratory tests) per his or her usual practice.Information about patient status was recorded atbaseline, at the end of carvedilol titration, and as closeas possible to 6 and 12 months after completing titration.It is assumed, although not specified, that patientswith heart failure would be seen at least at these intervalsper standards of good clinical practice. The informationrecorded included survival status, New YorkHeart Association (NYHA) class, current medications,clinical or adverse events, and other relevant assessmentsthe physician may have made. Detailed informationon the carvedilol dose titration experience wascollected.

The registry was completed and analyzed. Characteristicsof patients and physicians in the registry andthe primary results have been preliminarily reported,comparing groups within the registry and examining registryresults relative to those of the carvedilol clinical trialsexperience in the United States.

 There were 4280patients enrolled, with 259 cardiologists enrolling 3121patients and 129 primary care physicians enrolling 1159patients. Both physician groups had prior experiencewith use of &#946;-blockers in the treatment of heart failure,but that experience was considered extensive by morecardiologists than primary care physicians. The primarycare physicians enrolled more women, AfricanAmericans, patients with diabetes mellitus or hypertension,and patients who were older. Their patients alsohad higher left ventricular ejection fractions. Thepatients of cardiologists were more likely to be receivingbackground therapy with a combination of ACE inhibitors,diuretics, and digitalis. The duration of carvediloltitration and the reported degree of difficulty with titrationwere similar in both groups. Fewer patients managedby primary care physicians achieved recommended maximaltarget maintenance doses of carvedilol (25 or 50 mgtwice daily), but they were less likely to discontinuecarvedilol. Therefore, there were significant differencesbetween physician profiles, the baseline characteristicsof their patients, and their titration experience withcarvedilol in the setting of this &#946;-blocker registry. Thesedifferences may reflect heart failure management in generalin the community and might be expected to affectthe overall results of the registry.

Important differences were also observed when thesesame types of characteristics were compared betweenthe registry and the carvedilol clinical trials experience.Physicians in the registry were less likely to have an academicaffiliation, be hospital-based, or be cardiologists.Patients in the registry were significantly older, includedmore women, had higher left ventricular ejection fractions,and were less likely to have NYHA class III or IVsymptoms. Fewer registry patients were receiving ACEinhibitors, digoxin, or diuretics at baseline comparedwith those in clinical trials in which background therapywas prespecified per protocol. Although no or little difficultywith carvedilol titration was reported in 83% ofpatients in the registry, registry patients took longer tocomplete titration than those in clinical trials. In addition,fewer registry patients achieved the maximal targetmaintenance doses of carvedilol compared with patientsin clinical trials, and more patients in the registry discontinuedcarvedilol.

Despite the observed differences within the registrybetween physician profiles, the characteristics of theirpatients, and the dosing of carvedilol, there was no significantdifference in 1-year all-cause mortality betweenpatients treated by cardiologists vs primary care physicians(6.5% vs 7.5%, 

 = .33). In addition, in both groupsof patients, hospitalizations for heart failure werereduced by 42% from the year before entering the registry.

In summary, these preliminary results suggest thatthis &#946;-blocker registry recruited physicians and patientgroups who differ significantly from each other and fromclinical investigators and patients in heart failure clinicaltrials. Despite such differences, the all-cause mortalityand heart failure hospitalization rates werecomparable between patient groups within the registryand were in the same range as those observed incarvedilol clinical trials.

 Therefore, the differences inpatient populations and their risk factors were unimportantprognosticators, or they tended to cancel eachother out. This registry provides a large prospectivedatabase evaluating a recently approved heart failuretreatment (at the time of registry development) in thecommunity setting and demonstrates the key role thatcommunity-based healthcare providers can fulfill asclinical investigators outside the traditional clinical trialssetting.

Whereas the preliminary results of the registrydescribed in the previous section appear to bolster thoseof clinical trials by confirming them in the different setting,such may not always be the case, as other community-based observations may not replicate clinical trialresults (eg, spironolactone treatment of heart failure wasmore problematic in the community setting than in clinicaltrials

). Therefore, it is imperative to evaluate theadvantages and disadvantages of clinical trials and otherapproaches for providing the most reliable and applicableinformation for patients and healthcare providers.

The intent is not to challenge the role of clinical trialsand the educational activities they generate, asthese are well-established important and desirableactivities. However, they could be complemented byother initiatives aimed at helping to "fill in the gaps"with additional data that appropriately address thecommunity setting. Therefore, alternative means ofcollecting more representative data are needed andhave been called for.

Alternative means include the use of subset analyses,targeted clinical trials, observational studies, and registries.It has become commonplace to report subsetanalyses from large clinical trials. Examples of thisinclude trial results among heart failure subpopulations,defined by ethnicity, sex, comorbidity, and age.

Whereas this information may be of some value, itremains a subset of the overall clinical trial results and,therefore, has the same limitations as clinical trials.Furthermore, the overall limitations may be magnifiedin these subset analyses because the sample sizes aresmaller, the question being asked was not prospectivelyformulated, and randomization may not be balanced forthe subset in question. Given these limitations, theresults of these subset analyses can be used, at best, togenerate hypotheses for testing in subsequent studiesand cannot be considered conclusive. Therefore, a newclinical trial targeted at a previously identified subsetcould be designed. An example of such a targeted clinicaltrial deriving from subset analyses is the recentlyannounced African American Heart Failure Trial (AHeFT),

designed to prospectively address issues recognizedduring earlier trials among small numbers ofAfrican American patients who appeared to respond differentlyto treatment.

 However, this approach is stilla clinical trial, with the associated limitations of patientand physician selectivity, as well as protocol rigidity.

Observational studies use existing databases to providelarge experiences, but are limited by their retrospectivenature and the inherent design of theparticular database. For example, health claims databasescontain coded information, and the accuracy orcompleteness of coded diagnoses may not meet standardsof practice. In addition, much demographic andcomorbidity information may be missing. Prescriptiondatabases are confined largely to types of drugs prescribed,with limited information about the patient orthe indication for the drugs. Therefore, such studiesoften include inaccuracies of diagnosis and rely onassumptions about the reasons for patient managementdecisions. These, too, are probably best used forhypothesis generation.

Registries are another means of collecting complementarydata from the community setting. Comparedwith a clinical trial, a registry is a database created froma broader source, using flexibly applied means of datacollection. Registries may be designed to acquire newinformation not yet available in any organized fashion.For example, the Coronary Artery Surgery Study

 registrywas begun with the introduction of coronary arterybypass grafting to learn more about the natural historyof medically treated coronary artery disease to evaluatethe role of the new surgical technology. Registries alsoprovide preliminary data for use in designing clinical trialsand comparing patients in a clinical trial with similarpatients observed outside the protocol setting.

Registries have limitations. They are usually uncontrolledand lack provisions to assure the quality, accuracy,and completeness of data. Furthermore, participationof investigators and patients involves some selectionprocess. Although the &#946;-blocker heart failure registrydescribed herein was designed to simulate the communitysetting and be less like the clinical trials setting, ithad to recruit physicians and patients and ended upwith a preponderance of cardiologists. Therefore, thesephysicians and their patients were selected, at least inpart. Nevertheless, this registry may reflect the naturalhistory and management of heart failure in the "realworld." That such is the case is suggested by the similarityof the registry's patient characteristics and outcomesto those reported from large databases ofcommunity patients with heart failure.

Registries constructed from clinical data are limitedby often being retrospective and examining 1 or a fewfixed points in time. For example, there is limited informationon the natural history of heart failure in thebroad population, as the observations derive largely fromlocalized databases examining retrospective data.

Registries have not been widely used in a prospectivemanner (as was done in the registry described herein) tocollect information longitudinally while the actual dataof interest are generated, as is commonly done in a clinicaltrial. There is a role for prospective collection of datafrom the community setting to assess the effect of newtreatments. To provide a more valid interpretation relativeto the clinical trials experience, such community-deriveddata should include information about thepatients and the care providers involved in that setting.Some of these kinds of limitations are exemplified by theStudies of Left Ventricular Dysfunction (SOLVD)

 multicenterregistry, which enrolled patients who were ineligiblefor the clinical trial and, thus, represented selectedpatients. In addition, the data were derived from retrospectivereview, and some registry patients entered asubstudy protocol. Finally, the care providers were thesame clinical trialists, largely heart failure experts.

That registries are subject to patient and physicianselection biases is further exemplified by some recentheart failure registries that have noted important differencesbetween their observations and results from clinicaltrials. The Management to Improve Survival inCongestive Heart Failure registry noted higher mortalityand rehospitalization rates in a community setting, alongwith reduced use of recommended medications after hospitaldischarge for heart failure, compared with thereported results of similar clinical trials.

 The registrypatients were predominantly older white women, amarked difference from most clinical trial populations.The Italian Network on Congestive Heart FailureRegistry

 recruited outpatients with heart failure from acommunity setting and also noted a higher than expectedrate of short-term heart failure decompensation, aswell as frequent unexplained withdrawal of medicationsthat are routinely recommended for heart failure. Thisregistry is limited in its generalizability, as the patientswere followed up by expert cardiologists in specializedcenters and their demographic characteristics were similarto those of patients in clinical trials. The RochesterEpidemiology Project recently compared incidence andsurvival of heart failure measured a decade apart in theOlmstead County, Minnesota, community.

 The incidenceof heart failure did not change significantly inpatient cohorts enrolled in 1981 or 1991, and the survivalrate failed to change during that decade of markedadvances in heart failure management. The patients inthis registry were typical of the community population inthat they were older, included more women, and had significantcomorbidity, but differed by coming from a predominantlywhite population of middle to uppersocioeconomic class that received much of its healthcarefrom a prestigious tertiary care provider.

Other registries and observational studies havefocused on differences in quality of care and outcomesaccording to type of healthcare providers and havenoted differences in patient characteristics, medicationuse, and use of procedures among patients managed bycardiologists or primary care physicians, while clinicaloutcomes have been mixed.

 A limitation of most ofthese other registries is their use of hospitalizedpatients to find index cases, rather than recruiting fromoutpatients, who constitute the much larger and morerepresentative population of heart failure patients.Identifying patients at hospital discharge tends toinclude a population that is at increased risk of mortalityand readmission, whereas most patients with heartfailure are ambulatory and have a lower incidence ofprior hospitalizations and lower risk of subsequent mortalityand hospitalization.

 The limitations of thisapproach have been the subject of recent comment thathas called for the establishment of community-basedregistries for longitudinal observations.

Although the methods of randomized, controlledclinical trials limit the general applicability of theirresults because of investigator and patient selectionbiases and adherence to a rigid protocol, they remainthe cornerstone for establishing proof of concept anddemonstrating efficacy within the populations studiedand the particular research environment. It is more difficultand costly to use these same methods to demonstratecomparable results in larger diverse patientpopulations managed in the usual community care setting.It is not uncommon for new treatments to fail andfor new drugs to be recalled by regulatory authoritiesshortly after their approval and widespread use in thecommunity setting. Regulatory agencies recognize theselimitations of small clinical trials and have urged inclusionof broader, more representative patient populationsand investigators during the early development ofnew therapeutic approaches.

 Therefore, a role formethods other than the randomized, controlled trialmay exist.

The heart failure registry experience described hereinreinforces the potential value and role of observationalstudies to assess experience with and outcomes of newtreatments outside the clinical trial setting. Like a clinicaltrial and unlike most registries, the data in this exampleregistry were collected in a longitudinal prospectivemanner. This should permit interpretation of the resultsin juxtaposition to those of a clinical trial, with a greaterdegree of validity than most registry-type data, which areretrospective or point estimates. As already described,registries have limitations. Without an appropriate controlgroup, a registry cannot evaluate efficacy. Therefore,we cautiously interpret the general experience of this &#946;-blocker registry as demonstrating the value of a prospectiveregistry as a tool for collecting data to complementthose obtained from more structured clinical trials.

In summary, the relatively small randomized, controlledclinical trial should probably still be used toestablish proof of concept and efficacy in the idealpatient population. Other techniques, such as prospectiveregistries, may be more suited to assessing the efficacyand broad applicability of new therapeutictechnologies to the real-world community setting.Combining the 2 methods might be a better approachand be more cost effective by limiting the clinical trials(with their high per patient costs) to small numbers ofpatients and using the low per patient costs of a registryfor the large observational studies. The "practical clinicaltrial" has been suggested as a way of combining theseapproaches to broaden the diversity of patients andinvestigators, rendering them more representative ofcommunity settings, while maintaining controlled conditionsand validity of observations.

Alternatives to clinical trials, such as registry data,can be used to fill in the gaps left by clinical trials and toprovide experience in patient and physician populationsthat more closely reflects the real world. The registryapproach also emphasizes an important role for communityphysicians as investigators, a role they can fulfillwith fewer disruptions to their usual practice than mightbe demanded by involvement in clinical trials. A registrycan provide community physicians with firsthand experiencein using a new treatment modality to confirm orallay their concerns and possible misconceptions aboutusing that new treatment. If the clinical outcomes of aregistry parallel the benefits observed in clinical trials,important new therapeutic technologies might be transferredmore widely and expeditiously to the communitysetting. Therefore, the registry experience describedherein may serve as an important model for future applicationsand merit consideration for earlier inclusion indevelopment programs. Successful implementation ofdifferent approaches to clinical investigation, such asregistries and practical clinical trials, will require governmentagencies and private industry sectors that supporthealthcare research to rethink and reprioritize theirgoals and resource allocation.

From the Zena and Michael A. Wiener Cardiovascular Institute, Mount Sinai School ofMedicine, New York, and the Weill Medical College of Cornell University, Ithaca, NY.

Dr Franciosa is a paid consultant to GlaxoSmithKline Pharmaceuticals, ResearchTriangle Park, NC.

Address correspondence to: Joseph A. Franciosa, MD, Zena and Michael A. WienerCardiovascular Institute, Mount Sinai School of Medicine, 209 West 86th Street, ApartmentM-17, New York, NY 10024. E-mail: 

 Practical clinical trials: increasing the value ofclinical research for decision making in clinical and health policy. 

 Treatment research at the crossroads: the scientific interface of clinicaltrials and effectiveness research. 

3. Weinberger M, Oddone EZ, Henderson WG, et al.

 Multisite randomized controlledtrials in health services research: scientific challenges and operationalissues. 

4. ACC/AHA Task Force on Practice Guidelines.

 Guidelines for the evaluation andmanagement of heart failure: report of the American College ofCardiology/American Heart Association Task Force on Practice Guidelines(Committee on Evaluation and Management of Heart Failure). 

5. Packer M, Cohn JN, Steering Committee and Membership of the AdvisoryCouncil to Improve Outcomes Nationwide in Heart Failure.

 Consensus recommendationsfor the management of chronic heart failure. 

 ACC/AHA guidelines for the evaluationand management of chronic heart failure in the adult: executive summary: a reportof the American College of Cardiology/American Heart Association Task Force onPractice Guidelines (Committee to revise the 1995 Guidelines for the Evaluationand Management of Heart Failure). 

 &#946;-blocker therapy in heart failure: scientificreview. 

The Potential Role of Community-based Registries to Complement the Limited Applicability of Clinical Trial Results to the Community Setting: Heart Failure as an Example

 To evaluate how well hypertension is managed inHMO patients and to assess opportunities for improvement.

 The study population included HMOmembers (age 45-84 years) who had at least 1 ambulatoryencounter with an ICD-9-CM diagnosis code of essential hypertensionduring the first 6 months of 1999. Medical records werereviewed to obtain information on blood pressure measurements,sex, age, coexisting medical conditions, smoking status, andchanges made to the antihypertensive drug regimen.

 We identified 681 members with 3347 encountersrelated to hypertension management during 1999. Overall, 74(11%) patients were at target blood pressure for all visits and 260(38%) were at target blood pressure for at least 50% of the visits;222 (33%) patients were not at target blood pressure for any visit.A history of coronary artery disease or cerebrovascular disease wasassociated with better blood pressure control (defined as being atgoal levels during at least 50% of visits), while being older (age&#8805;75) or having diabetes mellitus was associated with poorer control.Medication regimen intensifications occurred in 10% of visitswith systolic blood pressure levels of 140-149 mm Hg, comparedwith 45% of visits with levels of &#8805;180 mm Hg. Medication regimenintensifications occurred in 21% of visits with diastolic blood pressurelevels of 90-99 mm Hg and 43% of visits with levels of &#8805;100mm Hg.

July 2004 - Part 2

The Effect of Increases in HMO Penetration and Changes in Payer Mix on In-hospital Mortality and Treatment Patterns for Acute Myocardial Infarction

Impact of an Educational Intervention for Secondary Prevention of Myocardial Infarction on Medicaid Drug Use and Cost

The Potential Role of Community-based Registries to Complement the Limited Applicability of Clinical Trial Results to the Community Setting: Heart Failure as an Example

Hypertension Management: The Care Gap Between Clinical Guidelines and Clinical Practice

Can We Use Automated Data to Assess Quality of Hypertension Care?

Improving Quality of Cardiovascular Care in the Real World: How Can We Remove the Barriers?

Aligning Financial Incentives With "Get With The Guidelines" to Improve Cardiovascular Care