Clinical Investigations

Acute Ischemic Heart Disease

Influence of cardiovascular absolute risk assessment onprescribing of antihypertensive and lipid-loweringmedications: A cluster randomized controlled trialSanjyot Vagholkar, MPH, a,b Nicholas Zwar, PhD, b Upali W. Jayasinghe, PhD, c Elizabeth Denney-Wilson, PhD, d

Anushka Patel, MD, PhD, e Terry Campbell, MD, PhD, FAHA, f and Mark F. Harris, MD c Sydney, Australia

Background Guidelines for management of hypertension and lipids recommend using cardiovascular absolute risk(CVAR) to manage patients. This randomized controlled trial investigated the impact of CVAR assessment in family practice onmanagement of cardiovascular risk, including prescription of antihypertensive and lipid-lowering medication.

Methods A cluster randomized controlled trial was conducted from 2008 to 2010 in Sydney, Australia. Family practiceswere randomized, and patients aged 45 to 69 years were invited to participate. Intervention family physicians (FP) weretrained in use of CVAR, provided with an electronic CVAR calculator, and assessed their patients' absolute risk in a dedicatedconsultation. Control practice patients received a general health check. Primary outcome analyzed was the proportion ofpatients in each group on antihypertensive and/or lipid-lowering medication at 12 months. Multilevel logistic regression wasperformed to explore variables influencing changes in pharmacologic therapy.

Results The study recruited 36 FPs from 34 practices and 1,074 patients, of which 906 (84.4%) completed 12-monthfollow-up. At 12 months, there was no significant difference between the intervention and control groups in proportion ofpatients on antihypertensives (31.2% vs 34.3%, P = .31), but control group patients were more likely to be on lipid-loweringmedications (30.2% vs 22.7%, P = .01). After multilevel analysis, this difference was not present. Intensification or reduction ofpharmacologic therapy was associated with meeting treatment targets for blood pressure and lipids but not with the CVAR orintervention group.

Conclusions Single-risk factor management remains a strong influence on FP prescribing practices. Shifting to anapproach based on CVAR will require more intensive intervention. (Am Heart J 2014;167:28-35.)

BackgroundCardiovascular disease (CVD) is the leading cause of

mortality worldwide, accounting for 12.9 million deathsin 2010.1 Prevention of CVD is dependent on managingmodifiable risk factors, 2 of which are blood pressure(BP) and lipids. Control of BP and lipids often involves

From the aGeneral Practice Unit, South Western Sydney Local Health District, Sydney,Australia, bSchool of Public Health & Community Medicine, UNSWMedicine, University ofNew SouthWales, Sydney, Australia, cCentre for Primary Health Care & Equity, Universityof New South Wales, Sydney, Australia, dFaculty of Health, University of Technology,Sydney, Australia, eThe George Institute for Global Health and The University of Sydney,Sydney, Australia, and fOffice of the Dean, UNSW Medicine, University of New SouthWales, Sydney, Australia.The trial was registered with the Australian New Zealand Clinical Trials Registry (http://www.anzctr.org.au; trial no: ACTRN12608000387325).Submitted March 25, 2013; accepted October 5, 2013.Reprint request: Sanjyot Vagholkar, MPH, GP Unit, Fairfield Hospital, PO Box 5, Fairfield,NSW 1860, Australia.E-mail: Sanjyot.Vagholkar@sswahs.nsw.gov.au0002-8703/$ - see front matter© 2014, Mosby, Inc. All rights reserved.http://dx.doi.org/10.1016/j.ahj.2013.10.002

pharmacologic treatment, medicines being a major costto the community. The annual medicines cost forhypertension alone in the United States is $21.3 billion2;the Australian government spends approximately $3billion annually on BP and lipid-lowering medicines.3

Given these costs, it is important to ensure that they aredirected at those who will benefit most.Guidelines for the management of hypertension and

lipids no longer focus on BP or lipid levels in isolation,rather they recommend assessing cardiovascular absoluterisk (CVAR) to guide management decisions.4-7 CVARtakes into account multiple cardiovascular (CV) riskfactors (age, sex, smoking, diabetes status, BP, and lipids)to provide an estimated percentage risk of an individualwith a CV event over a 5- or 10-year period. Pharmaco-logic therapy is targeted to those at high absolute risk,defined as an estimated risk greater than 15% or 20%depending on the guideline. However, management ofhigh-risk patients has been shown to be suboptimal.8,9

Systematic CVAR assessment may improve this, but todate, research exploring the impact of CVAR onprescribing has shown minimal effect,10,11 with only 2

Vagholkar et al 29American Heart JournalVolume 167, Number 1

randomized trials showing a modest increase in prescrib-ing of CV medications for high-risk patients.12,13

The use of CVAR assessment in clinical practice isincreasing but by no means routine.14-16 This Australiantrial aimed to test the impact of a program ofimplementation of CVAR assessment on family physi-cian (FP) management of CV risk, patient lifestylebehaviors, physiological parameters, new diagnoses ofCV disease, and clinical process in family practice. Thisarticle presents the findings of the impact of CVARassessment on prescribing of antihypertensive and lipid-lowering medications.

MethodsThis cluster randomized controlled trial was conducted in

urban family practice in Sydney, Australia, in 2008 to 2010. Theprotocol for this study has been previously published but issummarized below.17

RecruitmentFamily physicians. Family physicians and practices were

recruited from 4 Divisions of General Practice (geographicallybased FP organizations that support clinical practice) in Sydney.Practices were eligible if they used electronic prescribingsoftware and were not involved in other CVD research. Familyphysicians were eligible if they worked 4 or more sessions (1session = 3-4 hours) per week.

Patients. Patients were eligible if they were aged 45 to 69years, had no history of CVD, and had attended the practice inthe previous 12 months. Electronic records of participatingFPs were searched, and a list of patients who met theeligibility criteria was generated; from this, 160 patients wererandomly selected and sent letters of invitation. Patientswho had insufficient English or were cognitively impairedwere excluded.

RandomizationRandomization was conducted at the practice level to avoid

contamination between intervention and control FPs andpatients. A person (U.J.) independent of the intervention anddata collection conducted the allocation using a computerrandomization program. It was done in 2 blocks of 4 practiceswithin each division with an extra block of 2 practices from oneof the divisions. Research staff collecting practice data wereblinded to group allocation, as were patients.

InterventionThis was a multifaceted intervention in family practice

involving 3 components: training of FPs, patient consultationto assess and manage CVAR, and practice support forintervention implementation.Intervention FPs were provided with a 3-hour training

workshop covering assessment of CVAR, use of the NewZealand CV risk calculator18 (based on Framingham risk score),and current recommendations for management of CV risk basedon Australian guidelines for hypertension and lipid managementand the New Zealand calculator guidelines.6,7 The New Zealand

CV risk calculator was used because there was no Australiancalculator available at the time the study commenced.Intervention patients received a dedicated 20- to 30-minute

consultation that involved calculating the CVAR and thenproviding appropriate management based on risk level andcurrent guidelines.6,7,18 Patients in the control group had ageneral health check. At the 12-month health check, FPs wereasked to reassess the CVAR of intervention patients.

Data collectionData were collected by patient questionnaire, blood test

results (fasting lipids and glucose), and medical record audit.Questionnaires were administered at baseline and 12 months,and a medical record audit was conducted after the 12-monthhealth check.

OutcomesA full set of primary and secondary outcomes was outlined in

the protocol paper.17 This paper is focused on pharmacologicmanagement of CV risk. The primary outcomes were theproportion of patients in each group at 12 months prescribed

• Antihypertensive medication• Lipid-lowering medication• Both antihypertensive and lipid-lowering medication

Sample sizeThe sample size calculation was based on the ability to detect

a difference of 20% between the control and interventiongroups in the prescribing of antihypertensive and lipid-loweringmedication, with a design effect of 2.0 because of clustering andan expected 10% loss to follow-up. For 80% power at the 5%significance level, this gave a sample size of 16 practices (20 FPs)and 660 patients in each group.

AnalysisUnivariate analysis (χ2 and McNemar test) of the data was

conducted with SPSS software (v18; SPSS, Chicago, IL) toexamine medications prescribed within and between theintervention and control groups at baseline and 12 months.Medication data were analyzed for all patients for whom therewas both baseline and 12-month medication data to allow forwithin-group analysis. The researchers calculated CVAR scoreswhere all necessary input variables (age, sex, smoking status,BP, lipid levels, and diabetes status) were available. Forpatients where a CVAR score was calculated, pharmacologicmanagement was stratified by risk. Analysis was on anintention-to-treat basis.Data were subject to multilevel logistic regression analysis to

explore factors associated with intensification (commenced,increased dose, or number of medications) and reduction(ceased, decreased dose, or number of medications) of BP andlipid-lowering therapy. Intensification and reduction were basedon the difference between baseline medications and those at 12months, thus accounting for baseline differences. Multilevelanalysis was chosen to take account of the potential forclustering of medication changes around the individual FPs,given that prescribing is an individual FP's decision. Multilevellogistic regression models were used with 4 potential

Table I. Characteristics of FPs

Characteristics Intervention Control

FPs (practices) n = 20 (n = 18) n = 16 (n = 16)Age (y)35-54 12/19 (63.2%) 7/15 (46.7%)55+ 7/19 (36.8%) 8/15 (53.3%)

Male 12 (60.0%) 9 (56.3%)Australian/New Zealand graduate 11/19 (57.9%) 8/15 (53.3%)Solo practice 5 (25.0%) 5 (31.3%)Full-time practice 17/19 (89.5%) 11/15 (73.3%)Mean years in general practice (SD) 21.9 (7.8) 24.5 (10.8)

30 Vagholkar et alAmerican Heart Journal

January 2014

explanatory variables and intensification and reduction ofantihypertensive and lipid-lowering therapy as 4 independentoutcome variables. This was adjusted for clustering of patients(level 1) within FPs (level 2). In the model, the variance betweenpractices was added to that at FP level and the model accountedfor clustering within FP and practices.19 This analysis alsoprovided an estimate of provider variance. All multilevel analysiswas conducted with MLwiN (version 2.02) software (Centre forMutilevel Modelling, University of Bristol, UK).For patients on CV medication at baseline, analysis was

conducted to determine the risk level after imputation for theeffect of medication. This was based on previous estimates of a20% risk reduction for either BP or lipid-lowering agents with amultiplicative effect if on both.20

EthicsThis study was approved by the University of New South

Wales Human Research Ethics committee (HREC No.08005). AllFPs and patients gave informed, written consent to participate inthis study.

This study was funded by a grant from the National Healthand Medical Research Council of Australia. The authors aresolely responsible for the design and conduct of this study, allstudy analyses, the drafting and editing of the manuscript, andits final contents.

ResultsThe study recruited 36 FPs from 34 practices. Table I

summarizes the characteristics of the FPs. The genderprofile (58.2% male) was similar to national figuresshowing that 56% of Australian FPs are male. However,the FPs were slightly older (44% N55 years vs 35%nationally) and more had graduated overseas (40% vs 29%nationally).21

From these, FPs' patient lists 1,404 patients respondedto the invitation and 1,074 were recruited and consentedat baseline. Figure 1 shows patient recruitment andfollow-up through the study period. The patients lost tofollow-up or who withdrew differed from those whocompleted the study in the following ways: more wereemployed (71.4% vs 60.1%), fewer had been attending

the FP for N6 years (51.8% vs 62.1%), and fewer were onan antihypertensive (21.4% vs 29.7%).Baseline characteristics of the enrolled patients are

shown in Table II. Most patients (87.3%) were at low CVrisk, 28.4% were on an antihypertensive, and 23.7% wereon a lipid-lowering medication. Smoking rates were lessthan found in national general practice data (11.7% vs17.8%), whereas the prevalence of diabetes (11.8% vs12.6%) and overweight/obese patients (68.5% vs 69.2%)was similar.21 There were no significant differencesbetween the intervention and control groups apart from agreater proportion of women in the control group and alower mean low-density lipoprotein (LDL) cholesterolwith more reaching LDL target.Paired medication data were available for 906 patients.

In the univariate analysis of proportions of patients onmedications (Table III), significantly more patients in thecontrol group were on lipid-lowering medications at 12months; after multilevel analysis accounting for baselinemedications this difference was not present. Within thecontrol group, there was a significant increase inpatients on antihypertensives (30.9%-34.3% P = .03),but there was no significant difference compared withthe intervention group at 12 months. Other within-group analysis showed no significant changes before andafter the intervention. There were changes in pharma-cologic management, with more intensification thanreduction of therapy, but no significant differencesbetween the groups.Absolute risk was calculated for those patients who had

a complete set of input data, resulting in 685 patients inthe risk-stratified data. The between- and within-groupanalysis of these patients showed no significant differ-ences in the 12-month medication proportions andchanges in therapy (Figure 2). There were somedifferences between these patients and the 389 patientsfor whom CVAR could not be calculated. The 389patients were younger (55.2 years vs 57.1 years), hadmore men (46.3% vs 39.3%), had a greater rates ofdiabetes (15.7% vs 9.7%) and smoking (16.4% vs 9.2%),and therefore potentially had a higher CVAR.Similar proportions of low-risk patients commenced

antihypertensives in the intervention and controlgroups (5.2% vs 4.9%, P = .89). More low-risk controlgroup patients commenced lipid-lowering medication,but the difference was not significant (8.0% vs 4.5%,P = .09). There were no significant differences betweenthe intervention and control groups in the number oflow-risk patients who ceased medications (antihyper-tensive 2.3% vs 3.8%, P = .28; lipid-lowering 4.2% vs3.8%, P = .80).Multilevel analysis, taking account of FP clustering,

was performed to explore predictors of intensificationand reduction of pharmacologic therapy, as shown inTable IV. This analysis adjusted for baseline differ-ences in medications and confounding factors. In

Figure 1

Invited n = 5354Responded N = 1404 (26.2%)

Excluded n = 330

12 month medication analysis n = 475

Patients from 20 GPs in 18 practices

Lost to follow-up or withdrew: n = 92 (16.2%)

Intervention: n = 567Patients from 20 GPs in 18 practices

Lost to follow-up or withdrew: n = 76 (15.0%)

Control: n = 507Patients from 16 GPs in 16 practices

12 month medication analysis n = 431

Patients from 15 GPs in 15 practices

Allocation

Analysis

Follow-up

Enrollment

Patients from 36 GPs in 34 practices

Practices randomised

n = 1074 (20.1%) enrolled

Patient recruitment and follow-up.

Vagholkar et al 31American Heart JournalVolume 167, Number 1

these models, age, gender, smoking, and diabetesstatus were not included as independent variables toavoid multicollinearity (a state of very high intercor-relations among independent variables that might giveinvalid results) because CVAR takes these variablesinto account. The results show that CVAR and groupallocation did not predict changes in pharmacologictherapy. Not meeting treatment targets predictedintensification and reduction of therapy, with patientsnot meeting BP target more likely to intensify therapyand those not meeting lipid target less likely toreduce therapy. There was no significant FP variancein these models.Imputation of CV risk for patients who were low risk

at baseline and on medication showed that 188 of 228remained low risk after medication was taken intoaccount. The remaining 40 shifted to moderate risk. Ofmoderate risk patients, 27 of 40 shifted to high riskafter imputation.

DiscussionThis is the first Australian trial evaluating the impact of

CVAR assessment in clinical practice on prescribingbehavior. The main finding of the univariate analysis wasthat there was no significant difference between groupsin the proportion of patients on antihypertensivemedication at 12 months and although there was adifference in lipid-lowering medications, this was notpresent when confounding factors, including baselinedifferences, were accounted for in multilevel analysis.There were changes in prescribing in both groups asshown by the numbers who intensified and reducedtherapy, but again no significant differences betweengroups. These findings were consistent with previousresearch exploring CVAR assessment and prescribing thathas shown minimal impact. Two systematic reviewsincluded a total of 6 studies,10,11 of which only 2 showedan improvement in prescribing of CV medications amonghigh-risk patients.12,13 Both these studies had larger

Table II. Baseline characteristics of patients

Intervention(n = 567)

Control(n = 507)

Age (y), mean (SD) 56.2 (6.6) 56.6 (6.9)Female 312 (55.0%) 312 (61.5%)University education 140/540 (25.9%) 150/492 (30.5%)Employed 355/544 (65.3%) 310/497 (62.4%)Anglo-Celtic 287/536 (53.5%) 287/489 (58.7%)Attended practice N6 y 352/543 (64.8%) 298/489 (60.9%)CV risk factorsSystolic BP (mmHg) mean (SD) 126.5 (16.3) 128.3 (15.1)BP b140/90 mmHg 253/338 (74.9%) 243/335 (72.5%)LDL cholesterol (mmol/L),mean (SD)

3.3 (0.9) 3.1 (0.8)

LDL b2.5 mmol/L 79/442 (17.9%) 108/435 (24.8%)Smoker ⁎ 64/550 (11.6%) 59/501 (11.8%)Diabetes ⁎ 58/471 (12.3%) 48/426 (11.3%)BMI category ⁎

Overweight/Obese 346/519 (66.7%) 324/461 (70.3%)Absolute risk †

CVAR (%), mean (SD) 5.0 (4.0) 5.4 (4.4)Low risk (b10%) 319/359 (88.9%) 279/326 (85.6%)Moderate risk (10–15%) 29/359 (8.1%) 37/326 (11.3%)High risk (≥15%) 11/359 (3.1%) 10/326 (3.1%)

MedicationsAntihypertensive (AH) 159 (28.0%) 146 (28.8%)Lipid-lowering (LL) 123 (21.7%) 130 (25.6%)Both AH and LL 68 (12.0%) 63 (12.4%)

⁎Self-reported status smoking, diabetes, weight, height (BMI calculated).†Absolute risk data available 685 patients.

Table III. Univariate analysis antihypertensive and lipid-loweringmedication

MedicationIntervention(n = 475)

Control(n = 431) P

AntihypertensiveBaseline 136 (28.6%) 133 (30.9%) .4612 mo 148 (31.2%) 148 (34.3%) .31Intensification⁎ 56 (11.8%) 46 (10.7%) .60Reduction† 32 (6.7%) 25 (5.8%) .56

Lipid-loweringBaseline 101 (21.3%) 120 (27.8%) .02‡

12 mo 108 (22.7%) 130 (30.2%) .01‡

Intensification⁎ 37 (7.8%) 41 (9.5%) .36Reduction† 24 (5.1%) 26 (6.0%) .52

BothBaseline 59 (12.4%) 60 (13.9%) .5012 mo 63 (13.3%) 69 (16.0%) .24

⁎ Intensification = commenced, increased dose or increased no. of medications.†Reduction = ceased, decreased dose or decreased no. of medications.‡Bold = statistically significant at P b .05.

32 Vagholkar et alAmerican Heart Journal

January 2014

proportions of high-risk patients than did our study. Thestudy findings reflect FPs' views that they find CVAR auseful tool for patient education rather than a guide toprescribing decisions.16 A US survey found a high level ofawareness of risk calculators among doctors, but their useto guide prescribing decisions was still limited.14

The population in this study had few moderate- andhigh-risk patients. Because moderate-risk patients oftenprogress to pharmacologic therapy, it is difficult tocomment on the appropriateness of prescribing in thisgroup. The few high-risk patients in this sample were notall on both antihypertensive and lipid-lowering therapy at12 months. Among moderate- and high-risk patients,intensification of therapy was not significantly higher inthe intervention group. Previous work has identified thatthere are treatment gaps in primary prevention of CVD,with a substantial number of high-risk patients not onpharmacologic treatment.22-24

The multilevel analysis supported single-risk factormanagement as the determinant for prescribing changesrather than CVAR. Treatment to BP and lipid targets wasthe only factor in our model that was associated withchanges in pharmacotherapy. This is in keeping withresearch showing that doctors do not routinely targetprescribing according to risk level.9,23,24 The influence oftreatment to BP and lipid targets in determining therapychallenges the implementation of CVAR guidelines inprimary care. The emphasis on single-risk factor targetsremains potent despite the potential risks of overtreat-ment of single-risk factors.Various barriers to using CVAR assessment have been

identified, including time constraints and concerns aboutoversimplification of risk.25,26 The decision-makingprocesses of FPs are also influenced by a range of factorssuch as prior experience, pharmaceutical influences, andpatient wishes.27,28 Furthermore, doctors often feel thatthey can assess risk intuitively without calculating CVAR,although their accuracy has been shown to besuboptimal.15,29

At baseline, about a quarter of patients were on anantihypertensive or lipid-lowering medication, similar tolow-risk populations in other studies.9,24 A study explor-ing use of CV medications with risk level 10 years earlierfound that a substantial proportion of people onmedications were at low risk before their commence-ment.30 We conducted an imputation subanalysis giventhat it was not possible to know patients' risk levelsbefore the commencement of medication. The analysisshowed that 188 of 228 patients who were onmedications and low risk at baseline remained low riskafter imputation for medication effect. It is likely thatthese patients were commenced on medication based onsingle-risk factor management. Use of CVAR assessmentmight limit prescribing in this group of patients.

LimitationsOur study had several limitations. Control group

patients received a health check from their FP, and thismight have prompted assessment and management of CVrisk factors. It is also possible that some of the control FPsused CVAR calculation as part of their routine clinical

Figure 2

A) LOW RISK

P = .08P = .51

P = .21

P = .31P = .80

P = .11

P = .05

0

10

20

30

40

50

60

70

80

Antihyp

erte

nsive (

AH)

Lipid

-lower

ing (L

L)

AH + L

L

Inte

nsifica

tion A

H*

Inte

nsifica

tion L

L*

Reduct

ion A

H†

Reduct

ion L

L†

Per

cen

t o

f p

atie

nts

(%

)

Intervention n = 309

Control n = 264

B) MODERATE RISK

P = .45

P = 1.00P = .06

P = 1.00

P = 1.00

P = .74P = .07

0

10

20

30

40

50

60

70

80

Antihyp

erte

nsive (

AH)

Lipid

-lower

ing (L

L)

AH + L

L

Inte

nsifica

tion A

H*

Inte

nsifica

tion L

L*

Reduct

ion A

H†

Reduct

ion L

L†

Per

cen

t o

f p

atie

nts

(%

)

Intervention n = 28

Control n = 35

C) HIGH RISK

P = 1.00

P = 1.00

P = 1.00 P = 1.00

P = 1.00

P = 1.00

P = 1.00

0

10

20

30

40

50

60

70

80

Antihyp

erte

nsive (

AH)

Lipid

-lower

ing (L

L)

AH + L

L

Inte

nsifica

tion A

H*

Inte

nsifica

tion L

L*

Reduct

ion A

H†

Reduct

ion L

L†

Per

cen

t o

f p

atie

nts

(%

)

Intervention n = 10

Control n = 10

Antihypertensive and lipid-lowering medication and changes in prescribing at 12 months stratified by risk level.

Vagholkar et al 33American Heart JournalVolume 167, Number 1

Table IV. Multilevel logistic regression analysis for assessment of intensification and reduction of pharmacologic therapy

Explanatory variable

Antihypertensive therapy (n = 610), OR(95% CI)

Lipid-lowering therapy (n = 641), OR(95% CI)

Intensification⁎ Reduction† Intensification⁎ Reduction†

GroupControl 1.0 (reference) 1.0 (reference) 1.0 (reference) 1.0 (reference)Intervention 1.23 (0.71-2.13) 0.77 (0.38-1.54) 0.69 (0.37-1.30) 0.95 (0.36-2.52)

Absolute riskLow 1.0 (reference) 1.0 (reference) 1.0 (reference) 1.0 (reference)Moderate 0.87 (0.37-2.04) 1.41 (0.46-4.32) 1.54 (0.67-3.52) 2.01 (0.66-6.08)High 2.45 (0.87-6.85) 1.07 (0.13-8.68) 2.19 (0.60-8.06) 2.81 (0.54-14.55)

TargetsMeeting 1.0 (reference) 1.0 (reference) 1.0 (reference) 1.0 (reference)Not meetingBP ≥ 140/90 mmHg 4.25§ (2.47-7.31) 0.68 (0.28-1.65)LDL cholesterol ≥2.5 mmol/L 0.81 (0.42-1.56) 0.37§

(0.16-0.84)EducationUniversity 1.00 (reference) 1.0 (reference) 1.0 (reference) 1.0 (reference)School only 1.14 (0.75-2.64) 1.17 (0.53-2.57) 1.08 (0.58-2.03) 1.64 (0.60-4.47)

Between provider variance‡ (SE) 0.034 (0.129) 0.00 (0.00) 0.137 (0.179) 0.418 (0.425)ICC 0.01 0.00 0.04 0.11Design effect 1.39 1.0 2.56 3.29

⁎ Intensification = commenced, increased dose or increased no. of medications.†Reduction = ceased, decreased dose or decreased no. of medications.‡No significant provider (FP) variance.§ Bold data = statistically significant.

34 Vagholkar et alAmerican Heart Journal

January 2014

practice. Both factors could have diluted the effect ofthe intervention.Complete data for CVAR calculation were available for

only a subset of the sample, reducing the power of thestratified analysis. The small number of high-risk patientslimited our ability to draw conclusions about the role ofCVAR assessment in improving pharmacologic treatmentgaps. A previous Australian study in family practiceshowed rates of high-risk patients close to 15%; however,it involved an older cohort (mean age 68 years).9 Anotherstudy exploring vascular risk in an Australian family practicepopulation found a similar profile of low CV risk31;therefore, the risk profile of patients in our study appearsto be consistent with the Australian family practicesetting. Given our study involved inviting patients toparticipate, it may have self-selected those who wereproactive about their health and therefore likely to be atlower risk.The study sample was underpowered for the primary

medication outcomes. The assumption of ability to show a20% difference in prescribing was in retrospect unrealisticgiven the unexpectedly small number of high-risk patientsin the final sample. Recruiting a larger number of moderateand high-risk patients would require a much larger samplesize, potentially 3 to 4 times the sample in this study.Targeting higher-risk patients is an alternative option;however, it risks selection bias and limits the generalizabil-ity of the findings in family practice.

The actual design effects computed for antihyperten-sive therapy based on the calculated intracluster corre-lation coefficient (ICC) in the multilevel model (Table IV)were lower than that used in the power calculation,whereas for lipid-lowering therapy, they were higher.The multilevel analysis therefore had an adequate samplesize for predictors of changes in antihypertensive therapybut not for predictors of lipid-lowering therapy. The ICCsfrom the multilevel analysis provide a reference point forfuture research.

ConclusionThe CVAR intervention in this study had no significant

impact on prescribing of antihypertensive and lipid-lowering medications. Shifting FP management to onebased on CVAR assessment for primary prevention of CVDis likely to require more intensivemeasures, given the long-standing practice of treating risk factors in isolation. NewAustralian management guidelines for CVAR were intro-duced in May 2012,32 and research to explore their impacton clinical practice will be important.

AcknowledgementsThe investigators thank the participating practices, FPs,

patients, and Divisions of General Practice. Thanks toIqbal Hasan and Heike Schütze for data collection and

Vagholkar et al 35American Heart JournalVolume 167, Number 1

Charmaine Rodricks for assistance with the preparationof the manuscript.

References1. Lozano R, Naghavi M, Foreman K, et al. Global mortality from 235

causes of death for 20 age groups in 1990 and 2010: a systematicanalysis for the Global Burden of Disease Study 2010. Lancet 2012;380:2095-128.

2. Roger VL, Go AS, Lloyd-Jones DM, et al. Heart disease and strokestatistics—2012 update: a report from the american heartassociation. Circulation 2012;125:e2-220.

3. Australian Institute of Health & Welfare. Australia's health 2012Australia's health series no 13. Cat. no. AUS 156. Canberra: AIHW.2012.

4. Chobanian AV, Bakris GL, Black HR, et al. The Seventh Report of theJoint National Committee on Prevention, Detection, Evaluation andTreatment of High Blood Pressure: the JNC7 report. JAMA 2003;289:2560-72.

5. Third Report of the National Cholesterol Education Program (NCEP)Expert Panal on Detection, Evaluation and Treatment of High BloodCholesterl in Adults (Adult Treatment PaneI III) 2002 http://www.nhlbi.nih.gov/guidelines/cholesterol/atp3_rpt.htm (Accessed29/9/2013).

6. National Heart Foundation of Australia (National Blood Pressure andVascular Disease Advisory Committee). Guide to management ofblood pressure 2008. Updated December 2010. http://www.heartfoundation.org.au/SiteCollectionDocuments/HypertensionGuidelines2008to2010Update.pdf (Accessed29/9/2013).

7. Position Statement on Lipid Management. National Heart Foundationof Australia and the Cardiac Society of Australia and New Zealand.Heart Lung and Circulation 2005;14:275-91.

8. Ma J, Sehgal NL, Ayanian JZ, et al. National trends in statin use bycoronary heart disease risk category. PLoS Med 2005;2:e123.

9. Heeley EL, Peiris DP, Patel A, et al. Cardiovascular risk perception andevidence-practice gaps in Australian general practice (the AusHEARTstudy). Med J Aust 2010;192:254-9.

10. Brindle P, Beswick A, Fahey T, et al. Accuracy and impact of riskassessment in the primary prevention of cardiovascular disease: asystematic review. Heart 2006;92:1752-9.

11. Sheridan SL, Crespo E. Does the routine use of global coronary heartdisease risk scores translate into clinical benefits or harms? Asystematic review of the literature. BMC Health Services Research2008;8(60).

12. Hall LML, Jung RT, Leese GP. Controlled trial of effect of documentedcardiovascular risk scores on prescribing. BMJ 2003;326:251-2.

13. Montgomery AA, Fahey T, Peters TJ, et al. Evaluation of computerbased clinical decision support system and risk chart for managementof hypertension in primary care: randomised controlled trial. BMJ2000;320:686-90.

14. Shillinglaw B, Viera AJ, Edwards T, et al. Use of global coronaryheart disease risk assessment in practice: a cross sectional survey of asample of US physicians. BMC Health Services Research 2012;12(20).

15. Gupta R, Stocks NP, Broadbent J. Cardiovascular risk assessment inAustralian general practice. Aust Fam Physician 2009;38:364-8.

16. Torley D, Zwar NA, Comino EJ, et al. General practitioners views ofabsolute cardiovascular risk and its role in primary prevention. AustFam Physician 2005;34:503-5.

17. Wan Q, Harris MF, Zwar N, et al. Study protocol for a randomisedcontrolled trial: the feasibility and impact of cardiovascular absoluterisk assessment in Australian general practice. Am Heart J 2009;157:436-41.

18. New Zealand cardiovascular risk charts. www.health.govt.nz/publication/new-zealand-cardiovascular-risk-charts. Accessed Au-gust 29, 2013.

19. Moerbeek M. The consequences of ignoring a level of nesting inmultilevel analysis. Multivariate Behavioural Research 2004;39:129-49.

20. PILL Collaborative Group. An international randomisedplacebo-controlled trial of a four-component combination pill(“Polypill”) in people with raised cardiovascular risk.PLoS ONE 2011;6(5):e19857, http://dx.doi.org/10.1371/journal.pone.0019857.

21. Britt H, Miller GC, Henderson J, et al. Genral practice activity inAustralia 2009-10. General practice series no. 27. Cat. no. GEP 27.Canberra: AIHW. 2010.

22. Doroodchi H, Abdolrasulna M, Foster JA, et al. Knowledge andattitudes of primary care physicians in the management of patients atrisk of cardiovascular events. BMC Family Practice 2008;9(42).

23. Webster RJ, Heeley EL, Peiris DP, et al. Gaps in cardiovasculardisease risk management in Australian general practice. Med J Aust2009;191:324-9.

24. Rafter N, Connor J, Hall J, et al. Cardiovascular medications inprimary care: treatment gaps and targeting by absolute risk. The NewZealand Medical Journal 2005;118(1223).

25. Hobbs FDR, Jukema JW, Da Silva PM, et al. Barriers to cardiovasculardisease risk scoring and primary prevention in Europe. Q J Med2010;103:727-39.

26. Eichler K, Zoller M, Tschudi P, et al. Barriers to apply cardiovascularprediction rules in primary care: a postal survey. BMC Family Practice2007;8(1).

27. Bastholm Rahmner P, Gustafsson LL, Lassson J, et al. Variations inunderstanding the drug-prescribing process: a qualitative studyamong Swedish GPs. Fam Pract 2009;26:121-7.

28. Oriel-Zerbe C, Abholz HH. Primary prevention of cardiovasculardiseases by lipid-lowering treatment in German general practice:results from GPs ignoring guidelines and risk calculators.European Journal of General Practice 2007;13:27-34.

29. Cacoub P, Tocque-Le Gousse E, Fabry C, et al. Application in generalpractice of treatment guidelines for patients with dyslipidaemia: theRESPECT study. Archives of Cardiovascular Disease 2008;101:715-21.

30. Brekke M, Straand J. Does present use of cardiovascular medicationreflect elevated cardiovascular risk scores estimated ten years ago? Apopulation based longitudinal study. BMC Public Health 2011;11(144).

31. Harris MF, Fanaian M, Jayasinghe UW, et al. A cluster randomizedcontrolled trial of vascular risk factor management in generalpractice. Med J Aust 2012;197:387-93.

32. National Vascular Disease Prevention Alliance. Guidelines for themanagement of absolute cardiovascular disease risk 2012. http://strokefoundation.com.au/site/media/AbsoluteCVD_GL_webready.pdf. Accessed August 29, 2013).