Objectives: To provide physicians with evidence-based recommendationsfor care at the point of service, using an automatedsystem, and to evaluate its effectiveness in promoting prescriptionsto prevent cardiovascular events.
Study Design: Randomized controlled trial.
Methods: Patients at risk for cardiovascular events who mightbenefit from angiotensin-converting enzyme inhibitors or 3-hydroxy-3-methylglutaryl coenzyme A reductase inhibitors (statins)were identified from electronic data in a managed care organizationand randomly assigned into 2 groups. Physicians seeing outpatientsin the intervention group were faxed a sheet with pertinentpatient data, including a recommendation to prescribe the indicatedmedication. In the control group, the data sheet did not includethe recommendation. Dispensed prescriptions were comparedbetween groups.
Results: More than 4000 visits were observed for each medicationtype. Angiotensin-converting enzyme inhibitors or angiotensinII receptor blockers were dispensed in 7.1% of visits in the interventiongroup versus 5.7% in the control group (= .048) followingthe first patient-physician encounter. No significant differencewas observed for statins (intervention, 8.1% vs control, 7.7%).Data for all patient-physician encounters and both medicationswere combined in logistic regression analysis. The odds ratio was1.19 for a dispensed prescription in the intervention group and1.54 for 2 or more visits versus 1 visit.
Conclusions: An automated system that provides pertinent dataand tailored recommendations for care at the point of service modestlyincreased prescription dispensing rates. Targeting patient-providerencounters to change provider behavior is challenging;however, even small effects can produce clinically importantresults over time.
(Am J Manag Care. 2005;11:298-
Preventive care, among other services, is not deliveredoptimally in the United States.1-3 Knowledgeabout what to do often surpasses performance bya wide margin. Translating research into practice is difficult,especially for chronic disease care, in whichpatients and providers are influenced by many factorsthat make adoption of new practices difficult.4-7 All toooften, clinical trial interventions require intensiveservices that may not be practical in many healthcaresettings, and guidance on implementing these interventionsis lacking.7 Some understanding exists of thefeatures of research7 (eg, broad applicability) and innovations8(eg, simplicity) that improve the practicalapplication of research results.
Information technologies have been recommendedby the Institute of Medicine9 as a means to improvehealthcare in many ways. Clinical information systemsmay reduce barriers to preventive care and errors ofomission by providing timely information and decision-makingsupport in busy practice settings.10,11Experience suggests that some applications to supportdecision making, such as computerized reminders, aremore effective than others.12,13 Published evaluations ofcomputerized reminders often report effectiveness, dependingon the type of prompt, setting, and service targeted.13-17 Although successful interventions are not therule even in experienced hands,18 areas of demonstratedeffectiveness include drug prescribing and dosing,preventive care (generally primary prevention, includinginpatient prevention services),19 and some otheraspects of care but not diagnosis. Decision supportsystems in which patient data are matched to a computerizedalgorithm for generating patient-specific recommendationscan improve care but have not beenconvincingly demonstrated as effective in chronic diseasemanagement and secondary prevention.16,17,20
The objective of this study was to determine theeffectiveness of delivering patient-specific reminders forprescriptions to physicians at the point of service. Aclinical information system was used to generate recommendationsfor cardiovascular medications in a largepopulation of high-risk patients. Several characteristicsof the intervention contribute to potential effectivenessand generalizability. First, the system was conceived bypracticing physicians as a workable approach to problems they face in processing information and rememberingguidelines while delivering care. Second, theintervention is a form of computerized reminder, whichis a comparatively strong means of translating guidelinesinto practice. Third, patient-specific recommendationsfor care were based on convincing evidencedemonstrating reduced cardiovascular disease (CVD)events. Fourth, the system was already well establishedin a large healthcare organization, demonstrating itsfeasibility in similar settings. To our knowledge, similarreminder systems tested across a large healthcare systemhave not been reported.
In brief, an intervention group composed of physiciansseeing outpatients with diabetes mellitus at riskfor CVD events received recommendations to start anindicated medication, while a control group received norecommendations (ie, usual care). The medications recommendedwere angiotensin-converting enzymeinhibitors (ACEIs) and 3-hydroxy-3-methylglutarylcoenzyme A reductase inhibitors (statins). Rates of dispensedprescriptions were compared between studygroups. The Southern California Kaiser PermanenteInstitutional Review Board approved the study.Informed consent was waived for patients. SouthernCalifornia Kaiser Permanente Medical Group physicianswere informed of the study, were given study investigatorcontact information, and could opt out of the study.
Participants came from Southern California KaiserPermanente, which is an integrated, group-practice,prepaid health plan that provides comprehensive medicalservices to more than 3 million members in southernCalifornia. Family practice, internal medicine,cardiology, endocrinology, and nephrology physicianswere potential participants. Physicians became subjectsif they saw an eligible outpatient during the study.
Patients eligible for the ACEI recommendation weremen and women aged 55 to 80 years with no dispensationof an ACEI or angiotensin II receptor blocker (ARB)in the past 12 months and with (a) diabetes mellituswith high-density lipoprotein cholesterol levels lessthan 36 mg/dL (< 0.93 mmol/L) or low-density lipoproteincholesterol levels greater than 130 mg/dL (> 3.37mmol/L) or (b) atherosclerotic vascular disease. Menand women aged 55 to 80 years were eligible for thestatin recommendation if they had no statin or otherlipid-lowering drug dispensed in the past 6 months andhad diabetes mellitus with low-density lipoprotein cholesterollevels greater than 100 mg/dL (> 2.59 mmol/L).These target populations represented an expansion ofprevious health plan guidelines for ACEI and statin usefollowing the results of the Heart Outcomes PreventionEvaluation study,21 published in 2000, and the HeartProtection Study,22 published in 2002. The statin recommendationdepended on a shorter, 6-month periodwithout a dispensation because the medical group haddecided to identify patients more aggressively by thetime of the statin recommendation, which started 7months after the ACEI recommendation.
Classification of Diseases, Ninth Revision, Clinical
Established case-identification databases providedinformation on eligible patients, using data from 1997 tothe start of each intervention. Patients with diabetes mellituswere identified by 1 or more inpatient codes for diabetes mellitus, 2or more outpatient codes for diabetes mellitus,a hemoglobin A1c level greater than 7.5%, a record of dispensedoral hypoglycemic agents or insulin, or directentry into the system by care providers. Based on priormedical chart review, the sensitivity of the database isestimated to be at least 93% and the positive predictivevalue is about 95%. Patients with atherosclerotic vasculardisease were identified by inpatient codesindicating atherosclerotic vascular disease (eg, acutemyocardial infarction), procedure codes for coronaryartery revascularization, 2 or more outpatient codes indicating atherosclerotic vascular disease or 1 outpatientcode with 3 or more nitrate dispensationsin a 3-year period, or direct entry into the systemby care providers. This database is estimated to have apositive predictive value of about 83% based on severalvalidity investigations, including medical chart review.
The health plan has a multifaceted program for populationhealthcare management of chronic disease thatrelies heavily on chronic disease case-identificationdatabases. An important component of this program isthe care management summary sheet, which containspertinent data for the care of patients with chronic disease.Single-page sheets are faxed from a central locationthe morning of a scheduled outpatient appointmentand attached to the medical chart. Providers are targetedto receive the sheet if they are among medical specialtiesthat provide outpatient preventive care;however, some surgical specialties request the faxbecause the data summary is deemed useful.
These care management summary sheets were conceivedby physician leaders and first implemented inlate 1998 as an efficient means of providing clinicianswith up-to-date, pertinent information for decisionmaking. These physician leaders recognized that ittakes time to gather data, providers often forget tocheck for clinically recommended prescriptions, andconfusion often results when guidelines are applied inparticular situations.23 Patient-specific recommendationswere included to remind providers to take actionand to assist in decision making. Over time, experienceusing the system has led to an emphasis on a limitednumber of short and directive recommendations. Thisapproach assumes that providers know about clinicalguidelines but do not always remember to incorporatethis knowledge into practice.
The care management summary sheet identifies thepatient and lists his or her chronic disease diagnoses,with estimates of disease severity as determined byrisk algorithms (an example sheet is available from theauthor). Separate tables list the latest dispensed medicationsand pertinent laboratory data (eg, hemoglobinA1c levels) with associated dates. The dates and primarydiagnoses of the most recent hospital and emergencydepartment visits are included. A list ofrecommendations for testing, medications, advice, orreferrals (eg, to education classes) is included on eachsheet. Recommendations are derived from periodicallyupdated clinical practice guidelines. Typically, thereare 2 to 3 such recommendations per patient. Althoughthe recommendations are considered the mostimportant part of the care management summarysheet, the effectiveness of the recommendations hadnot been tested until the present study.
We tested the effectiveness of 2 new recommendationsfor care supported by a clear evidence base asindicators of the effectiveness of this approach forprompting prescriptions. The recommendations werebegun at different periods because of the natural timeframe of organizational decisions and actions.Physicians were sent care management summary sheetsfor targeted encounters with eligible patients. Thesheets included the new recommendation message ordid not include the new message but were otherwise thesame. Messages not related to the study, such as recommendationsto get a hemoglobin A1c test, were continued.Targeted encounters were those betweeneligible patients and providers, following proceduresthat were independent of the study. After the studyended, the recommendations were included on all datasheets for targeted encounters.
The message for the ACEI-targeted population was"high CVD risk: start lisinopril (target, 10-40 mg/d)."For the statin-targeted population, the message stated"high CVD risk: start lovastatin 10 mg/d" if the lastserum creatinine measurement was 2.0 mg/dL or higherand "high CVD risk: start lovastatin 40 mg/d" if thelast serum creatinine measurement was less than 2.0mg/dL or if measurement results were not available inthe past 12 months.
Whenever significant changes to the care managementsummary sheet are made, all physicians are informed bymail. Medical group leaders in local service areas areasked to notify providers by discussing these changes atlocal group practice meetings. For this study, physicianswere informed by mailed fliers that new recommendationsfor ACEIs and statins were forthcoming and that theeffect of the recommendations would be studied. Beforeand independent of these events, information on theHeart Outcomes Prevention Evaluation study21 and theHeart Protection Study22 had been presented in localservice areas by population healthcare physician "champions."In addition, all health plan providers receivehard copies of clinical practice guidelines, which areoften discussed in local and regional conferences andseminars and are also available on an intranet.
Outcomes were dispensed prescriptions of an ACEIor ARB and a statin within 2 weeks after a visit by an eligiblepatient. Prescriptions were identified by computerfrom the health plan's pharmacy database using generalproduct identifier codes and product name recognition.If an ARB was started, it was considered equivalent tostarting an ACEI. About 92% of health plan membershave a prescription benefit, and it is believed that mostmembers fill prescriptions at health plan pharmacies,which is required for the benefit.
Less than 2 weeks before the start date for each medicationrecommendation, patients whose data wouldresult in a message were randomly assigned by computerequally into the intervention and control groups. For theACEI group, patients with diabetes mellitus, atheroscleroticvascular disease, and both conditions were randomizedequally between intervention and control groups.If a patient visited a provider more than once duringthe study, the assignment was retained for each visit.Physicians could see patients in the intervention andcontrol groups. Random assignment of patients ratherthan providers was chosen because by using this methodall providers had equal access to information and comparisonsbetween the intervention and control groupsfocused on the reminder and prompting function of therecommendations, which was their primary purpose.
A minimum sample size was calculated by expectingan increase from 15% to 20% in drug dispensation rates.Using the significance level of <.05, equal-sizedgroups, a 2-sided test of proportions, and 90% powerrequired 2504 unique patient-provider encounters. Thisnumber was estimated to be achievable within 3 weeksfor the ACEI recommendation and within 4 weeks forthe statin recommendation. It was thought to be importantto continue the intervention for at least 4 weeks toallow sufficient time for targeted physicians to beexposed several times to the new recommendations andto become fully cognizant of the change. The ACEIintervention lasted 4 weeks, and the statin intervention,which occurred 7 months later, lasted 6 weeks to havesimilarly large sample sizes.
The effects of randomization were assessed with tests, Fisher exact tests, and the binomial test. The basicunit of analysis was the patient-physician dyad thatoccurred at each clinic visit. Dispensation rates for eachmedication following the first encounter of a patient witha physician were compared between the interventionand control groups using Fischer exact test with 2-sidedhypotheses. The data for both medication recommendationswere combined and similarly analyzed to assessthe effectiveness of the intervention more generally.Differences in dispensed prescriptions were examinedbetween primary care and specialty care. An homogeneityof odds ratio test was used to compare the differencesbetween primary care and specialty care in responsivenessto the intervention. A patient-level analysis thatcombined both medication recommendations and controlledfor the number of patient visits was performedusing logistic regression. Allfaxes were analyzed even if thetransmission record indicatedthat the fax was not received.Data were analyzed using SASversion 8 (SAS Institute, Cary,NC) and StatXact version 5(Hearne Scientific SoftwareLLC, Chicago, Ill).
There were 31 015 patientsrandomized into the study whowere not taking ACEIs or ARBsand 20 185 randomizedpatients not taking statins. Toplace these numbers in context,among patients who wereconsidered eligible for medications,63.9% were already taking ACEIs or ARBs at thestart of the study, and 53.2% were already taking statinsor other lipid-lowering drugs. Table 1 summarizes thecharacteristics of patients who were randomized andwho visited study providers. There were 8557 patientsand 1089 physicians who received the study intervention.No physicians asked to be excluded from participationin the study. Ninety-three percent of all faxeswere recorded as received by the targeted fax machines.
Angiotensin-Converting Enzyme Inhibitor Message
For the ACEI portion of the study, there were 4678unique patient encounters with 975 different primarycare and specialty physicians during 4 weeks starting inJuly 2002. In the intervention group, 164 (7.1%) of 2311patients were dispensed an ACEI or ARB. In the controlgroup, 134 (5.7%) of 2367 patients were dispensed themedications (Table 2). The difference, 1.4%, is justbarely significant (= .048).
For the statin message, there were 4183 uniquepatient encounters with 941 different primary care andspecialty physicians during 6 weeks starting in February2003. In the intervention group, 171 (8.1%) of 2103patients were dispensed a statin (Table 2), and in thecontrol group 160 (7.7%) of 2080 patients were dispenseda statin. The 0.4% difference between groups isnot significant (= .61).
Data for both recommended medications were combined (Table 2). The combined dispensation rate foreither type of medication was 7.6% in the interventiongroup and 6.6% in the control group (difference, 1.0%; = .08). Whether in the intervention or control group,visits with specialists were more often followed by a dispensedprescription than visits with primary care physicians(difference, 3.0%; < .001). However, there wasno difference between specialists and primary carephysicians (odds ratios [ORs], 1.16 and 1.20, respectively;= .92) in the odds of a prescription being dispensedin the intervention versus the control groups.Combined across the intervention and control groups,the overall dispensation rate was 6.7% for patients with1 visit and 10.3% for patients with 2 or more visits (difference,3.6%; < .001).
Logistic regression analysis was used to furtherexamine the effect of the intervention on dispensationrates if 1 or more visits occurred, controlling for thenumber of visits, the medication recommended, andpatient age, sex, and past medication use. The numberof visits was coded as 1 versus more than 1, becauseonly 1.4% of patients had more than 2 visits during thestudy. We controlled for known dispensations ofACEIs, ARBs, statins, and other lipid-lowering drugsup to 2 years before the date of the patient visit,excepting the dispensation-free period that promptedthe recommendation. To preserve independence,observations on patients who were in both the ACEIand the statin groups were dropped. The combinedanalysis involved 8557 patients with 1 or more physicianencounters.
The OR of dispensing an ACEI, ARB, or statin in theintervention versus the control group was 1.192 (95%confidence interval [CI], 1.01-1.40; = .04); hence, therecommendation messagesincreased the odds of a dispensedprescription by 19.2% atany given visit. The odds of adispensation increased formore than 1 visit versus 1 visit(OR, 1.538; 95% CI, 1.24-2.91[< .001]); however, an interactionterm revealed no differencein trend from 1 visit tomore than 1 visit between theintervention and controlgroups. Prior medication usewas associated with anincreased odds of a dispensation(OR, 2.25; 95% CI, 1.82-2.79 [< .001]). Aftercontrolling for prior use of themedication, the differencebetween dispensations of ACEIs or ARBs versus statinswas nonsignificant (OR, 0.87; 95% CI, 0.74-1.04). Patientage and sex were not significant predictors of prescriptiondispensations.
Quality Review of Encounters
As part of a quality review, 50 medical charts ofpatients for whom no ACEI or ARB was dispensed,despite a recommendation to do so, were reviewed. Allpatients qualified for the medication. Seventeen ofthese encounters were judged "missed opportunities,"while the remaining 33 were deemed "not inappropriate."Examples of encounters that understandably didnot result in a dispensed drug include gastrointestinalendoscopy, preoperative consultation, a pacemakercheck, an infectious disease consultation, nephrologyvisits in which an ACEI was contraindicated, a cardiologyvisit in which the patient had low blood pressurewhile taking a small dosage of β-blocker, a pulmonologyconsultation, and a documented prescription that wasnot picked up.
The fundamental question evaluated by this study iswhether a reminder to providers for evidence-basedcare that is targeted to the patient, is specificallydetailed, and is delivered at the point of outpatient servicecan change provider prescribing behavior in a largehealth plan. The system we evaluated demonstrated lessthan the expected effect. Overall, although the odds of aprescription being filled increased by 19.2% when a recommendationwas delivered, the absolute increase indispensation rates was small at 1.0% (ie, from 6.6% to7.6%) in the combined analysis. In general, cardiologists,nephrologists, and endocrinologists had higherrates of dispensed prescriptions than primary carephysicians but no difference in responsiveness to therecommendation message. Although the study was notdesigned to test if the intervention worked better orworse over time, there was no evidence that the promptbecame more "potent" with repeated member visits.
It is unclear why the implementation of the recommendationwas better for ACEIs and ARBs (1.4%) thanfor statins (0.4%). The explanation may be in partbecause of the more recent publication of the HeartProtection Study22 than the Heart Outcomes PreventionEvaluation study21 and the time it takes to change practicebehavior in response to new information. There isno difference between the drugs in copayment for memberswith a drug benefit and only small differences incharges for members without a drug benefit. Overall,statins were dispensed more frequently than ACEIs duringclinic visits, perhaps reflecting more opportunity forimprovement (eg, less history of noncompliance) associatedwith the lower proportion of eligible patientsalready taking statins. Differences between dispensationrates were reduced after adjustment for prior use ofmedications in logistic regression analysis. Prior use ofmedications increased the odds of a dispensation, possiblyindicating a lapse in compliance or reluctance touse medications, perhaps because of costs or borderlineblood pressure and cholesterol levels.
Although the quality review cannot fully explain theobserved dispensation rates, it demonstrated how oftenthis intervention targeted encounters in which the recommendedaction was not reasonably expected. If dispensationsfollowing the ACEI recommendation areadjusted by assuming that two thirds of encounters withno dispensed prescription would not normally result ina prescription, then the response becomes 18.0%.
Patients with 2 or more visits had 53.8% higher dispensationrates than patients with a single visit,although frequent visits during a short period might suggestdifficult medical problems; therefore, repeated contactwith providers is important. In our studypopulations, patients had medians of 10 and 12 visits ofany kind in a year for the statin-and ACEI-eligiblepatients, respectively, so the effects of the interventionare expected to accumulate over time. An absolute differenceof 1.0% between the intervention and controlgroups suggests that 100 patient-physician encounterswould need the intervention for 1 additional patient toget a prescription. The Heart Outcomes PreventionEvaluation study21 suggested that 15 high-risk peoplewith diabetes mellitus would have to be treated withACEIs for about 5 years to prevent 1 major cardiovascularor microvascular outcome, and the HeartProtection Study22 suggested that 10 to 14 high-riskindividuals would have to be treated with statins for 5years to prevent major vascular disease events.
The primary difference between our study and priorstudies is in the application of the intervention to anentire large healthcare system rather than to a singleclinical setting, an academic institution, or providersselected and trained to receive the intervention. Webelieve that this use of clinical information systems isreproducible in other healthcare settings. For example,the HMO Research Network24 is a consortium of 13health maintenance organizations with research capabilities,including corresponding data systems, andmore than 40 million members. However, many organizationsmay not have access to updated and comprehensivelaboratory, pharmacy, and administrative datathat are used to initiate and target a broad array of populationhealthcare recommendations.
In this study, patients were randomized rather thanphysicians; therefore, with about 4.5 study visits perprovider, each physician was approximately equallyexposed to the recommendations. With repeated exposure,physicians probably learn to recognize the need fora prescription without the recommendation. Providersmight apply that learning to patients in the controlgroup, thus boosting dispensations in the control grouprelative to the intervention group. Alternatively, someproviders may become dependent on the reminder recommendationsand not act when they are absent, thusexaggerating the difference between groups.
If we had randomized the providers in the study, thedifference measured between the intervention and controlgroups would represent not only the reminder effectof the recommendations but also a positive learningeffect. Although unlikely to occur in the short term, randomizingproviders would not have eliminated potentialbias resulting from providers in the control group learningand applying the recommendations based on theexperiences of colleagues in the intervention group. Byrandomizing patients, the difference between groupsrepresents the proportion of encounters in which equallyaware providers needed prompting to apply guidelines.The high proportion of eligible patients takingACEIs and statins before the study supports the contentionthat providers were already aware of guidelines.Physician leaders thought that it was valuable to assessthe intervention using an estimate of the remindereffect, believing it would hold for similar, establishedtherapeutic recommendations.
Although our study patient population was large andlikely to be representative of many patients at high riskfor CVD, it is possible that our study providers are moreor less responsive to similar interventions than providersin other organizations. A survey of provider perceptionsof an earlier version of the care management summarysheet from June 2002 found that 92% of respondentsreviewed the data, 56% believed that it prompted themto start a therapy, and 59% rated the overall effectivenessin helping manage patients with chronic disease as goodor excellent (vs poor or fair). We do not know detailsabout how providers use the care management summarysheet (eg, always, only if there is time, or if they seekspecific information). Data were not collected onwhether the sheet was placed on the medical chart. Theeffectiveness of the intervention is reduced by faxes notreceived or not placed on the medical chart. We did notcollect data on written prescriptions, which are a moredirect measure of provider responsiveness to the interventionthan dispensed prescriptions. The intervention'seffectiveness was not evaluated over the long term.Health outcomes were not addressed; however, the connectionbetween appropriate use of statins, ACEIs, orARBs and outcomes was the basis for the recommendationsstudied. Patients misidentified as being at highCVD risk could reduce the responsiveness of physicians.Finally, no cost analysis was done.
We believe that the results of this evaluation are generallyinformative of the effectiveness of tailored recommendationsfor medications given at the time ofroutine outpatient visits. Although the recommendationstested had a small effect at one point in time, thecumulative effect becomes important within a systemof longitudinal care with multiple points of patient contact.Moreover, the care management summary sheethas several recommendations for care in addition tothose studied and is deemed useful beyond the carerecommendations, so its evolution will continue as it isadapted for more effective use. For example, the caremanagement summary sheets are now printed on-siteat patient registration rather than being faxed the nightbefore. Systems with flexibility and the potential toevolve and improve are more likely to have continuedeffects over time and with repeated patient contact.
We thank Roger Benton, PhD, Anthony Farley, and Heather Watson,BS, for information on physicians' opinions about the care managementsummary sheet described in this article.
From the Southern California Kaiser Permanente Department of Research and Evaluation, Pasadena (SFD, RC), Department of Medicine, San Diego (JRD), Department of Family Medicine, Harbor City (VMB), Pharmacy Services, Downey (RKN), and Department of Medicine, Woodland Hills (FHZ).
This research was supported by Merck Health Management Services, Whitehouse Station, NJ.
Address correspondence to: Stephen F. Derose, MD, Southern California Kaiser Permanente Department of Research and Evaluation, 393 E Walnut Street, Pasadena, CA 91188. E-mail: firstname.lastname@example.org.
1. Schuster MA, McGlynn EA, Brook RH. How good is the quality of health carein the United States? 1998;76:517-563.
2. Stange KC, Flocke SA, Goodwin MA, Kelly RB, Zyzanski SJ. Direct observationof rates of preventive service delivery in community family practice. 2000;31:167-176.
N Engl J Med.
3. McGlynn EA, Asch SM, Adams J, et al. The quality of health care delivered toadults in the United States. 2003;348:2635-2645.
4. Bodenheimer T, Wagner EH, Grumbach K. Improving primary care for patientswith chronic illness. 2002;288:1775-1779.
5. Bodenheimer T, Wagner EH, Grumbach K. Improving primary care for patientswith chronic illness: the chronic care model, part 2. 2002;288:1909-1914.
Manag Care Q.
6. Wagner EH, Davis C, Schaefer J, Von Korff M, Austin B. A survey of leadingchronic disease management programs: are they consistent with the literature?1999;7:56-66.
7. Glasgow RE. Translating research to practice: lessons learned, areas for improvement,and future directions. 2003;26:2451-2456.
Diffusion of Innovations.
8. Rogers EM. 5th ed. New York, NY: Free Press; 2003.
Crossing the Quality Chasm: A New Health System for the 21st Century.
9. Committee on Quality Health Care in America, Institute of Medicine.Washington, DC: National Academy Press; 2001.
10. Grumbach K, Bodenheimer T. A primary care home for Americans: puttingthe house in order. JAMA. 2002;288:889-893.
N Engl J Med.
11. James BC. Making it easy to do it right. 2001;345:991-993.
12. Davis DA, Taylor-Vaisey A. Translating guidelines into practice: a systematicreview of theoretic concepts, practical experience and research evidence in theadoption of clinical practice guidelines. 1997;157:408-416.
Health Technol Assess.
13. Grimshaw JM, Thomas RE, MacLennan G, et al. Effectiveness and efficiencyof guideline dissemination and implementation strategies. 2004;8:iii-iv, 1-72.
Arch Intern Med.
14. Balas EA, Weingarten S, Garb CT, Blumenthal D, Boren SA, Brown GD.Improving preventive care by prompting physicians. 2000;160:301-308.
J Am Med Inform Assoc.
15. Shea S, DuMouchel W, Bahamonde L. A meta-analysis of 16 randomizedcontrolled trials to evaluate computer-based clinical reminder systems for preventivecare in the ambulatory setting. 1996;3:399-409.
Med J Aust.
16. Bennett JW, Glasziou PP. Computerised reminders and feedback in medicationmanagement: a systematic review of randomised controlled trials. 2003;178:217-222.
17. Hunt DL, Haynes RB, Hanna SE, Smith K. Effects of computer-based clinicaldecision support systems on physician performance and patient outcomes: asystematic review. 1998;280:1339-1346.
J Gen Intern Med.
18. Tierney WM, Overhage JM, Murray MD, et al. Effects of computerized guidelinesfor managing heart disease in primary care. 2003;18:967-976.
N Engl J Med.
19. Dexter PR, Perkins S, Overhage JM, Maharry K, Kohler RB, McDonald CJ. Acomputerized reminder system to increase the use of preventive care for hospitalizedpatients. 2001;345:965-970.
Med J Aust.
20. Eccles MP, Grimshaw JM. Selecting, presenting and delivering clinical guidelines:are there any "magic bullets"? 2004;180(suppl):S52-S54.
21. Heart Outcomes Prevention Evaluation Study Investigators. Effects oframipril on cardiovascular and microvascular outcomes in people with diabetesmellitus: results of the HOPE study and MICRO-HOPE substudy [published correctionappears in 2000;356:860]. 2000;355:253-259.
22. Heart Protection Study Collaborative Group. MRC/BHF Heart ProtectionStudy of cholesterol lowering with simvastatin in 20,536 high-risk individuals: arandomised placebo-controlled trial. 2002;360:7-22.
Br J Gen
23. Kedward J, Dakin L. A qualitative study of barriers to the use of statins and theimplementation of coronary heart disease prevention in primary care. 2003;53:684-689.
24. HMO Research Network Web site. Available at: http://hmoresearchnetwork.org/members.htm. Accessed December 15, 2003.