Impact of provision of cardiovascular disease risk estimates to healthcare professionals and patients: a systematic review

Objective To systematically review whether the provision of information on cardiovascular disease (CVD) risk to healthcare professionals and patients impacts their decision-making, behaviour and ultimately patient health. Design A systematic review. Data sources An electronic literature search of MEDLINE and PubMed from 01/01/2004 to 01/06/2013 with no language restriction and manual screening of reference lists of systematic reviews on similar topics and all included papers. Eligibility criteria for selecting studies (1) Primary research published in a peer-reviewed journal; (2) inclusion of participants with no history of CVD; (3) intervention strategy consisted of provision of a CVD risk model estimate to either professionals or patients; and (4) the only difference between the intervention group and control group (or the only intervention in the case of before-after studies) was the provision of a CVD risk model estimate. Results After duplicates were removed, the initial electronic search identified 9671 papers. We screened 196 papers at title and abstract level and included 17 studies. The heterogeneity of the studies limited the analysis, but together they showed that provision of risk information to patients improved the accuracy of risk perception without decreasing quality of life or increasing anxiety, but had little effect on lifestyle. Providing risk information to physicians increased prescribing of lipid-lowering and blood pressure medication, with greatest effects in those with CVD risk >20% (relative risk for change in prescribing 2.13 (1.02 to 4.63) and 2.38 (1.11 to 5.10) respectively). Overall, there was a trend towards reductions in cholesterol and blood pressure and a statistically significant reduction in modelled CVD risk (−0.39% (−0.71 to −0.07)) after, on average, 12 months. Conclusions There seems evidence that providing CVD risk model estimates to professionals and patients improves perceived CVD risk and medical prescribing, with little evidence of harm on psychological well-being.

Eligibility criteria for selecting studies: (1) Primary research published in a peer-reviewed journal; (2) inclusion of participants with no history of CVD; (3) intervention strategy consisted of provision of a CVD risk model estimate to either professionals or patients; and (4) the only difference between the intervention group and control group (or the only intervention in the case of before-after studies) was the provision of a CVD risk model estimate.
Results: After duplicates were removed, the initial electronic search identified 9671 papers. We screened 196 papers at title and abstract level and included 17 studies. The heterogeneity of the studies limited the analysis, but together they showed that provision of risk information to patients improved the accuracy of risk perception without decreasing quality of life or increasing anxiety, but had little effect on lifestyle. Providing risk information to physicians increased prescribing of lipid-lowering and blood pressure medication, with greatest effects in those with CVD risk >20% (relative risk for change in prescribing 2.13 (1.02 to 4.63) and 2.38 (1.11 to 5.10) respectively). Overall, there was a trend towards reductions in cholesterol and blood pressure and a statistically significant reduction in modelled CVD risk (−0.39% (−0.71 to −0.07)) after, on average, 12 months.
Conclusions: There seems evidence that providing CVD risk model estimates to professionals and patients improves perceived CVD risk and medical prescribing, with little evidence of harm on psychological wellbeing.

BACKGROUND
Even though there have been advances in diagnosis, treatment and prevention of cardiovascular disease (CVD) in recent decades, CVD still remains the single largest cause of death worldwide. 1 In 2011, 3 in every 10 deaths were caused by CVD, 2 and it is estimated that by 2030, 23.3 million people will die annually due to CVD. 3 This has led to increasing focus on affordable effective preventive strategies. These include collective approaches targeting the wider underlying determinants of CVD in an attempt to shift the entire population distribution of CVD risk factors, and approaches that focus on identification of individuals at high risk. An integral part of the latter approach is the use of CVD prognostic models or risk scores, such as the Framingham risk score, 4 QRISK, 5 ASSIGN, 6 SCORE, 7 PROCAM 8 and Reynolds 9 which share a core set of Strengths and limitations of this study ▪ This systematic review is the first to address the impact of provision of cardiovascular disease risk model estimates on patients' or physicians' behaviour or health outcomes. ▪ The use of broad inclusion criteria and the systematic search of multiple databases allowed us to include studies in which assessment of the impact of provision of a risk score alone was not the primary outcome. ▪ Despite this, the small number and heterogeneity of included studies limit the strength of conclusions that can be made. Most report only shortterm changes and those that address behaviour change use mostly self-reported measures and are underpowered to detect small changes that may be clinically important at the population level.
established risk factors (age, sex, smoking, blood pressure and total cholesterol) among other risk factors (eg, Townsend score, family history). These scores enable estimation of an individual's risk of developing CVD, and so have the potential to help physicians with decisions regarding initiation, type and intensity of treatment (eg, cholesterol-lowering treatment and blood pressure management), to facilitate an informed discussion between physician and patient regarding lifestyle changes and pharmacological treatment, to improve risk perception of both physicians and patients, and to motivate individuals to improve their health-related behaviours, with the ultimate goal to prevent CVD events. They also provide an opportunity to prioritise individuals with the highest CVD risk and so allocate resources efficiently. Such risk models have been incorporated into many major clinical guidelines for routine practice [10][11][12][13][14] and the UK National Health Service (NHS) Health Checks programme which aims to assess CVD risk for all those aged 40-74 years without pre-existing CVD. Despite this strong advocacy of the use of such CVD risk models, relatively little is known about the benefits and harms of provision of CVD risk model estimates to patients, and whether their use by physicians actually translates into improved behavioural and clinical outcomes. Previous groups have reviewed randomised clinical trials of the effectiveness of healthcare professionals using a CVD risk model or score to aid primary prevention, 15 the effectiveness of the use of CVD risk model when combined with lifestyle interventions in the prevention of CVD, 16 the effects of providing individuals with global CVD risk information with or without tailored interventions, 17 and the effects of providing CVD risk model estimates on physician knowledge of global CHD risk. 18 These systematic reviews all included studies in which the provision of a risk model estimate was part of a multifactorial intervention. To our knowledge, no recent systematic review has comprehensively addressed the specific impact of provision of a CVD risk model estimate to either practitioners or patients.
The purpose of this review was, therefore, to assess whether provision of a CVD risk model estimate to either patients or practitioners, as opposed to other simultaneous or subsequent interventions, such as lifestyle advice or exercise programmes, impacts patient or practitioner behaviour or health outcomes.

METHODS
We performed a systematic literature review following an a priori established study protocol (available on request). Reporting was according to the PRISMA statement. 19 Search strategy As part of a larger systematic review on CVD risk scores, we performed an electronic literature search of MEDLINE and PubMed from 01/01/2004 to 01/06/ 2013 with no language restriction. The search strategy is described in full in online supplementary appendix 1. Briefly, it included terms for CVD, coronary heart disease, hypertension, hyperlipidaemia, stroke or cerebrovascular disease in combination with terms for risk assessment, prediction, score or decision support, and named risk scores. We also reviewed the reference lists of systematic reviews on this topic [15][16][17][18] for studies published prior to 2004 and manually screened the reference lists of all included papers.

Study selection
We included studies that met the following criteria: (1) primary (randomised and non-randomised) studies published in a peer-reviewed journal; (2) inclusion of participants with no history of CVD; (3) intervention strategy consisted of provision of a CVD risk model estimate to either physicians or patients (ie, not just providing a means by which physicians or patients could calculate CVD risk score); and (4) the only difference between the intervention group and control group (or the only intervention in the case of before-after studies) was the provision of a CVD risk model estimate. Observational and qualitative studies, studies calculating CVD scores for the secondary prevention of CVD or including both primary and secondary prevention where it was not possible to separate out the primary prevention group, and conference abstracts, editorials, commentaries, letters and reviews were excluded.
We selected studies in a three-stage process. In the first stage, titles of all studies identified from the electronic search were screened in duplicate by six reviewers involved in a large systematic review on CVD risk prediction led by ES and KGMM to identify all studies that described the application of a risk model into clinical practice or focused on risk-based management. In the second stage, this process was repeated with seven reviewers based on abstract. In the third stage, we combined those studies identified from stage 2 with studies from systematic reviews on similar topics, [15][16][17][18] and two researchers ( JU-S+SG/BS) independently screened the titles and abstracts against the inclusion and exclusion criteria. For studies where a definite decision to reject could not be made based on title and abstract alone, we obtained the full paper for detailed assessment. Two reviewers ( JU-S and BS) then independently assessed the full-text articles for the possibility of inclusion in the review. We excluded papers identified by both researchers as not meeting the inclusion criteria. Any disagreements were resolved by discussion, and a final decision was made at consensus meetings by JU-S, BS and SG.
Data extraction JU-S and BS independently extracted data from all studies included in the review using a standardised data extraction form to reduce bias. The data extracted included: (1) study characteristics (research question, risk model or score used, study design, study setting, intervention, duration of follow-up, outcomes measured); (2) selection of participants (inclusion criteria, method of recruitment/randomisation); (3) participant characteristics (sample size, age, gender, comorbidity, level of CVD risk); and (4) measured outcome(s). Reviewers were not blinded to publication details. We requested additional unpublished data from the authors of papers in which it was mentioned that additional data were available or additional data were required to meet the inclusion criteria or for clarification of results.
Quality assessment JU-S and BS conducted quality assessment at the same time as data extraction. Since our review included studies with different designs, we used a checklist based on the Critical Appraisal Skills Programme guidelines for cohort studies and randomised controlled trials (RCTs; available from http://www.casp-uk.net/#! casp-tools-checklists/c18f8) as an initial framework and then classified each study as high, medium or low quality. No studies were excluded based on quality assessment alone.

Data synthesis and statistical analysis
For analysis, we grouped the measured outcomes into those relating to risk perception, changes in health-related behaviour, intermediate outcomes (eg, blood pressure, cholesterol levels), modelled cardiovascular risk, medication prescribing, anxiety and psychological well-being, and contact with healthcare professionals after provision of risk information. For data on continuous outcomes, where possible, we expressed results as the difference in the mean change between groups. Where standardised mean changes were presented in the studies, we used the SD of the control group to convert data to non-standardised changes. Where this was not possible, we presented the results as mean±SD. For data on binary variables, such as a change in prescribing or meeting targets, we presented data as ORs or relative risk and 95% CIs. Where possible we combined data from different studies using random effects meta-analysis, but due to variations in study design and reporting, we were only able to do this for a small number of outcomes. We analysed all data according to the different outcomes and the recipient of the CVD risk score ( physician or patient). For outcomes with data from three or more studies, we assessed the heterogeneity between studies using the I 2 statistic. We did not perform formal tests of heterogeneity for outcomes with data from less than three studies. All analyses were conducted using statistical software package STATA/SE V.12. Significance was set at p<0.05.

RESULTS
After duplicates were removed, the initial electronic search identified 9671 papers (figure 1). One hundred and fifty-nine of these were identified as possible inclusion papers during stage 1 and 2 of the screening. When the papers from existing systematic reviews were added to this number, we screened 196 papers at title and abstract level against the inclusion criteria. We excluded 162 and a further 17 after full-text assessment. The most common reasons for exclusion were that we were unable to isolate data on primary prevention or the effect of giving risk information alone (figure 1). The analysis is, therefore, based on 17 studies. [20][21][22][23][24][25][26][27][28][29][30][31][32][33][34][35][36] A summary of the characteristics of those 17 studies is shown in table 1. They showed considerable heterogeneity in terms of size, setting, risk score used, duration of follow-up and outcomes measured. Seven (2 RCTs and 5 before-and-after studies) measured changes in understanding or risk, risk perception, psychological wellbeing or anxiety with 10 (9 RCTs and 1 before-and-after study) reporting changes in risk factors, prescribing or calculated risk. Eleven provided risk information to patients alone, three to physicians and three to both patients and physicians. Study quality assessment is summarised in table 2. Overall, it was variable with only one study being judged as high quality. All nine RCTs reporting changes in risk factors, prescribing or calculated risk were judged as medium or high quality. The effect sizes between the different studies are, therefore, more likely to reflect between-study heterogeneity for a number of study characteristics, in particular differences in the patient populations and delivery of risk information, than overall quality of the studies. Table 3 shows additional details about the participants in each of the studies, including the inclusion criteria, methods of recruitment and randomisation, and the baseline CVD risk of participants. Of the nine RCTs reporting changes in risk factors, prescribing or calculated risk, five included only high-risk patients (CVD risk estimate ≥20%) or those with uncontrolled hypertension or untreated hyperlipidaemia. 28 29 33-35 The other four 25 31 32 36 included both high-risk and low-risk participants, with the proportion of each well balanced between the intervention and control groups.

Risk perception
Two before-and-after studies reported the immediate effects of receiving risk information. In one, patients tended to initially overestimate their risk and giving risk information resulted in a significant reduction in perceived risk (eg, mean perceived 10-year risk of CVD fell from 48% to 20%, n=95, p<0.001). 20 The other reported a significant increase in the proportion with accurate risk perception from 34% to 74% ( p<0.0001) following risk information mostly due to a reduction in the rate of underestimation (60% to 18% ( p<0.0001)), while the rate of overestimation was low and did not change (7% to 8%; p=0.82; n=146). 26 However, Price et al showed in their RCT that the risk perception after 4 weeks did not differ between those who received the risk estimate only (mean 33.7, SD 18.9) and those who did not (mean 34.8, SD 19.6; p=0.87). Similar pattern of results were present when the comparison was made between those who received risk estimate + lifestyle advice (mean 35.6, SD 18.6) and those who received lifestyle advice only (mean 41.8, SD 20.7, p=0.14).
Three studies also reported changes in risk perception at different time intervals after receipt of risk information. In two, patients tended to initially overestimate their risk and giving risk information resulted in a sustained significant reduction in perceived risk after 2-4 weeks (mean 18% from 32%, n=37, p<0.00 001) 27 and 6 weeks (eg, mean perceived 10-year risk of CVD was 26% after 6 weeks compared with 48% prior to risk information, n=95, p<0.001). 20 Asimakopoulou et al 20 additionally showed that, controlling for actual risk, those patients given their 10-year risk reported consistently higher risk estimates than those given 1-year or 5-year risk estimates, while the 1-year and 5-year groups' estimates were not significantly different from each other. The third study reported change in perceived risk 7-12 weeks after provision of risk information in one of four formats (CDC, RISKO, Arizona Heart Institute or Medical age). 21 Forty per cent were initially accurate (42% underestimating their risk and 18% overestimating risk) and the majority of respondents did not change their perceived risk. Those who received feedback that they were above average risk were more likely to increase their perceived risk, but all other groups were equally likely to increase their perceived risk as decrease with 10.4% of those who were told they were below average risk increasing their perceived risk and 11.9% of those who were told their risk was higher, decreasing their perceived risk.

Changes in health-related behaviour Diet
One before-and-after study in general practice 30 and one RCT of factory workers 25 included self-reported changes in diet. The provision of risk information led to no statistically significant changes in reported total fat intake or unsaturated fat intake at 6 months 30 or the percentage who increased fruit and vegetable or fibre consumption or reduced fat at 5 months. 25 In the latter study, although not statistically significant, the effects of  provision of risk information to different groups ranged from a relative reduction of 10.6% to a relative increase of 47.4% in fibre intake. Only one study used an objective measure of changes in diet ( plasma vitamin C) and also showed no significant effect of risk information among those with CVD risk >20% (standardised difference in means −0.079 (−0.37 to 0.21), p=0.589) at 1 month. 29 Smoking cessation Four studies reported on smoking cessation. 25 29 30 36 All showed no significant effect of risk communication irrespective of differences in study design (before-and-after study or RCT), duration of follow-up (1 to 12 months) or outcome measure (self-report or objectively measured cotinine): one found no significant difference in the percentage of smokers at 3 months with the mean difference adjusted for baseline and patients nested within the same physician (−0.8%, p=0.64, control n=89, intervention n=202); 36 another measured smoking status among participants with CVD risk >20% using cotinine and also found no significant difference (standardised difference in means −0.53 (−1.23 to 0.17), p=0.136) at 1 month; 29 the third, a before-and-after study, showed a small but non-significant increase in smoking (10.0% to 11.4% prerisk and postrisk information); 30 and the fourth showed mixed results with more factory workers stopping smoking at 5 months in the group receiving risk information and health education than health education alone (9.6% vs 6.0%) but less stopping smoking in the group receiving risk information and feedback on cholesterol and health education than those receiving feedback on cholesterol and health education alone (3.6% vs 8.2%). 25 Physical activity Two studies reported on intention to increase physical activity. Both showed no effect with one before-and-after study including 11.4% of participants with CVD risk >20% showing no change in the number of participants in the action or maintenance stages of change for exercise (n=145/298 prerisk information and n=151/298 postrisk information) 30 and the other RCT showing no significant between-group differences in intention to increase physical activity (no data given). 29 Only one study assessed change in physical activity and reported no significant difference in mean accelerometer counts at 1 month among participants with CVD risk >20% (standardised difference in means 0.086 (−0.20 to 0.37), p=0.559). 29

Alcohol consumption
Two RCTs showed no difference in self-reported alcohol consumption. One including only participants with CVD risk >20% reported no difference in mean alcohol intake at 1 month (standardised difference in means= −0.033 (−0.36 to 0.29), p=0.84) 29 and the other no change in the proportion of factory workers who

Changes in intermediate measures Cholesterol
Four RCTs reported the between-group difference in mean change in total cholesterol or low-density lipoprotein cholesterol (LDL). Two were cluster randomised studies in which physicians in the intervention group received risk information about their patients. 31 36 After adjusting for lipid-lowering or antihypertensive medication at baseline 31 and baseline characteristics and patients nested within the same physician, 36 neither showed a difference in cholesterol after 3 months ( figure 2). In the two other trials, risk information was given to participants. In one, factory workers were randomised to receive either health education±risk information or health education and feedback on cholesterol±risk information. There was no significant difference in mean change in plasma cholesterol at 5 months ( pooled difference in mean change −0.05 mmol/L (−0.13 to 0.03, p=0.208)) or 12 months ( pooled difference in mean change −0.025 mmol/L (−0.103 to 0.025, p=0.529)). 25 The other trial was the only one to include participants with CVD risk >20% and showed a significant difference in mean change in LDL after 1 month 29 with sensitivity analysis excluding participants who had a change in drug treatment also showing a net 7.8% reduction in LDL ( p<0.001). 29 Combining the data for 5 months from Hanlon et al with Price et al gave a nonsignificant decrease, and pooling both with the other two studies showed a trend towards lower cholesterol but this was not significant ( p=0.069, I 2 =69.9%; figure 2). No significant differences were found with highdensity lipoprotein cholesterol (HDL) or total cholesterol to HDL ratio after adjusting for baseline risk, 36 or with triglycerides 29 or the number of patients who had had their LDL repeated and in whom it was 30 mg/dL or lower than baseline at 9 months (OR 0.99 (0.56 to 1.74)) 28 among those with CVD risk >20% or with raised LDL. One RCT of participants with untreated hyperlipidaemia did, however, find that patients were more likely to reach lipid targets if they received risk information (OR 1.26 (1.04 to 1.53), n=1163 (intervention) and 1193 (control)) and there was a significant interaction ( p=0.04) between being given a risk profile and the age gap (estimated cardiovascular risk age minus actual age) with the OR for reaching lipid targets in individuals who were reassured that they were at low risk 0.92 (0.64 to 1.31) compared with 1.69 (1.21 to 2.36) for those in the highest age gap quintile. 34 Blood pressure Five RCTs reported the difference in mean change in blood pressure between patients with and without risk information. 28  information to physicians. 31 33 36 All three showed a nonsignificant difference in mean change in systolic blood pressure (SBP) or diastolic blood pressure (DBP). Pooling the data from Bucher et al 31 which is adjusted for lipid-lowering medication or antihypertensive medication and Lowensteyn et al 36 adjusted for group differences at baseline and patients nested within the same physician gave a non-significant reduction in both SBP and DBP ( figure 3A,B). It was not possible to pool the data from Hanon et al 33 because they reported only the pooled blood pressure preintervention and postintervention (intervention group (n=556); preintervention SBP 167, SD 13; and postintervention SBP 140, SD 11; and control group (n=712) SBP 166, SD 12, and 140, SD 10 preintervention and postintervention, respectively). In the two other trials, risk information was provided to patients. 28 29 Both included only participants with CVD risk >20% or raised LDL, and together they showed a significant reduction in SBP in those patients that received risk information after 1 and 9 months (figure 3A). Persell et al 28 additionally showed no significant effect on DBP (figure 3B), and combining all five studies together gave non-significant differences in both SBP and DBP ( figure 3A, B).

Weight/BMI
Two studies reported changes in weight and found no significant difference in mean weight (standardised difference in means 0.065 (−0.22 to 0.35), p=0.66) or body fat percentage (standardised difference in means 0.063 (−0.23 to 0.35), p=0.67) among participants with CVD risk >20% at 1 month 29 or mean change in body mass index (BMI) at 3 months adjusted for baseline and   Changes in modelled cardiovascular risk Five RCTs reported changes in modelled CVD risk. In three, risk information was provided to physicians. 31 34 36 Bucher et al 31 reported difference in mean change in 10-year Framingham risk after 12-18 months in HIV patients after adjusting for lipid-lowering or antihypertensive medication; Lowensteyn et al 36 reported the difference in mean change in 8-year coronary risk after 3-6 months adjusting for baseline and patients nested within the same physician; and Grover et al 34 reported the difference in mean change in 10-year risk of CVD after 12 months adjusted for baseline. Together, they showed a statistically significant reduction in modelled risk ( figure 4).
In the other two RCTs, risk information was provided to patients. 25 29 One found no significant difference in modelled risk at 1 month among those with CVD risk >20% (standardised difference in change in means −0.155±0.146, p=0.239). 29 The other showed a nonsignificant increase at 5 months (difference in mean change −0.154 (−0.373 to 0.066), p=0.171 on a scale 1-100 where 1 is highest risk) and non-significant decrease at 12 months (difference in mean change 0.167 (−0.116 to 0.450), p=0.248) among factory workers. 25 Combining the data from Hanlon et al at 12 months with the three studies providing risk information to physicians gave a significant reduction in risk score ( figure 4).

Changes in prescribing
Lipid-lowering medication Four RCTs reported changes in lipid-lowering medication. One in which risk scores were provided to physicians blinded to the trial showed a 42% increase in the probability of having a change in lipid-lowering medication among all patients, but this was not statistically significant ( p=0.29). In the same study, patients with a CVD risk >20% were twice as likely to have their medication changed when physicians were presented with the risk score (relative risk (RR) 2.32 (1.01 to 5.29), p=0.03). 32 The other three trials reported the difference in the number of new prescriptions for lipid-lowering medication. Two gave risk information to physicians treating patients with HIV, 31 or patients with CVD risk >20% or raised LDL, 28 and one to patients with a CVD risk >20%. 29 When pooled there was RR of 1.35 not achieving statistical significance ( p=0.08, I 2 =0%; figure 5), but when only those studies including participants with CVD risk >20% or raised LDL are pooled, there is a significant increase in initiation of lipid medication (RR 1.83 (1.13 to 2.98)) which increases to 2.11 (1.27 to 3.49) when only the two studies providing risk information to physicians are included.
Blood pressure-lowering medication Three RCTs reported changes in blood pressure-lowering medication. As with lipid-lowering medication, Hall et al 32 showed no difference in the probability of a change in medication among all patients (RR 1.52 (0.86 to 2.69), p=0.146) but patients with a CVD risk >20% who received risk information were twice as likely to have their medication changed (RR 2.38 (1.11 to 5.10),  p=0.0189). The other two included only participants with uncontrolled hypertension or CVD >20% or raised LDL and reported no difference in the percentage of patients requiring dual therapy after 8 weeks (46%, n=712 control vs 41%, n=556 risk information), 33 the percentage of patients with uncontrolled hypertension at baseline who had an increase in number of antihypertensive drug classes (OR 2.89 (0.70 to 11.9) p=0.14) 28 or the median number of antihypertensive drug classes used ( p=0.45). 28 One study including only participants whose blood pressure was above currently recommended targets reported the mean blood pressure threshold for intensifying treatment. 35 They found no significant difference in the mean blood pressure thresholds for the whole study population (difference in means 2.8 mm Hg (−0.5 to 6.0) for systolic BP and 1.1 mm Hg (−0.9 to 3.2) for diastolic BP) or when stratified by age gap (the difference between their CVD risk age and chronological age).

Glucose-lowering medication
Three RCTs found no effect of risk information on either change in medication in all participants (RR 1.17 (0.89 to 1.53), p=0.273) or those with CVD risk >20% (RR 1.26 (0.87 to 1.84), p=0.219), 32 or initiation of medication following provision of risk information to physicians caring for patients with HIV (RR 1.33 (0.30 to 5.96), p=0.704) 31 or patients with CVD risk >20% (RR 6.78 (0.36 to 129.43)). 29 Psychological well-being and anxiety Psychological well-being following receipt of risk information was assessed in four studies. Three showed no difference. In the first, a before-after study, there was no significant difference in the 12-item General Health Questionnaire (GHQ) score after 6 months in either those at low/moderate risk or those at high risk and no significant difference between the groups (n=146 in low/moderate group and 116 in high-risk group, p=0.80). 22 A subsequent RCT also showed no significant difference in GHQ-12 after 1 year ( p=0.603) or 5 years ( p=0.727) between those receiving risk information (n=802, 745) and controls (n=390, 381). 23 The third, also an RCT, 29 measured change in general health after 1 month using the EQ-5D-3L and showed no between-group differences between those who received the risk estimate only (mean change 0.05, SD 0.12) and those who did not (mean change 0.01, SD 0.08; p=0.06). A similar pattern of results was seen when the comparison was made between those who received risk estimate + lifestyle advice (mean change −0.003, SD 0.08) and those who received lifestyle advice only (mean change 0.02, SD 0.11; p=0.442). The fourth used the GHQ-28 and showed a significant ( p<0.005) reduction in mean scores from 7.21, SD 5.18 to 6.59, SD 5.7 in high-risk participants after 10 days. 24 Three studies also explored changes in anxiety. Two were before-and-after studies and used the Spielberger state anxiety inventory. As with psychological well-being in the same study, a significant reduction in mean anxiety from 32.3, SD 8.72 to 28.1, SD 8.60 was seen in high-risk participants after 10 days ( p<0.005, n=1676), 24 but this was not seen in the other study including all participants after 6 months (median (IQR) prerisk 35 (26.7 to 43.3) and postrisk 33 (23.3 to 43.3)). 30 The third, an RCT, reported no significant between-group difference in change of anxiety at 1 month measured by the six-item Spielberger state anxiety inventory when the comparison was made between those who received risk estimate only (mean change −0.45, SD 2.87) and control group (mean change −0.63, SD 2.61; p=0.707). A similar pattern of results was seen when the comparison was made between those who received risk estimate +lifestyle advice (mean change −0.27, SD 2.87) and those who received lifestyle advice only (mean change −0.64, SD 3.21; p=0.324). The same study also found no difference in worry about future risk of heart attacks, measured using the adapted Lerman breast cancer worry scale, or in self-regulation. 29 Contact with healthcare professionals Two studies reported the effect on healthcare usage of giving risk information. In one, 36 the design of the study encouraged control patients to see their physician in order to receive their risk score, and so follow-up was higher in the control group. There was, however, a significant difference between low-risk and high-risk individuals ( p for interaction 0.026) with high-risk individuals being more likely to be followed up than low-risk individuals.
A further study in the USA including only participants with CVD risk >20% or with raised LDL showed no significant difference in healthcare usage at 9 months between those receiving risk information (n=218) and controls (n=217) with no difference in the percentage with any follow-up visit ( p=0.28), number of follow-up visits ( p=0.14), number of telephone ( p=0.75) or email ( p=0.96) contacts. 28

Principal findings
To our knowledge, this is the first systematic review to address specifically whether the provision of CVD risk model estimates alone impacts patient or practitioner behaviour or health outcomes. Despite the widespread adoption of risk scores in guidelines, [10][11][12][13][14] only 17 studies were identified and they are heterogeneous in terms of size, design, baseline level of risk and choice of outcomes. They do, however, show that providing patients with risk information changes risk perception and increases the accuracy of perceived risk without decreasing quality of life or increasing anxiety. While there is no current evidence that this translates into changes in lifestyle, small reductions in cholesterol, blood pressure and modelled CVD risk are seen consistently. These may be mediated through changes in prescribing: providing risk information to physicians leads to statistically significant changes in prescribing of both lipid-lowering and blood pressure medications, with these effects greater in those at higher risk. However, trends towards small reductions in cholesterol and blood pressure were also seen in studies in which risk information was only provided to patients. Only one study included in our systematic review was judged as high quality and most of the studies relied on self-reported measures of health-related behaviour. Consequently, there is a need for further research with adequately powered trials and objective outcomes to better understand the impact on behaviour of provision of a CVD risk estimate to physicians and individuals.

Strengths and weaknesses
The main strengths of this review are the use of broad inclusion criteria and the systematic search of multiple databases in addition to the inclusion of studies identified in previous systematic reviews on related topics. This allowed us to include studies in which assessment of the impact of provision of a risk score alone was not the primary outcome. While this reduces publication bias, the literature on CVD risk is diverse and rapidly expanding, and so it remains possible that there are additional studies of relevance that we were not able to identify. Furthermore, additional studies may have been published since June 2013, the date of the literature search. We are not aware, however, of any papers that would alter the overall conclusions of this review.
The other main limitation, as with most systematic reviews, is the extent and quality of the published data. Given the interest in CVD prevention and widespread use of CVD risk scores, it is both surprising and concerning that so few studies concerning their impact on care have been published. Additionally, of the 17 studies identified, most report only short-term changes (<6 months) and those that address behaviour change use mostly self-reported measures and are underpowered to detect small changes that may be clinically important at the population level. The small number and heterogeneity of the studies also made combining results difficult. Not only did they report different outcomes at different time periods with participants of varying baseline risk, but many of the studies adjusted for different baseline variables without reporting unadjusted changes or provided insufficient data to allow us to calculate effects adjusted for baseline risk. Data on the SD of outcomes were also not available for most studies, meaning it was not possible to calculate standardised changes in outcomes. A further limitation was that the small number of studies differed in whether the risk model estimates were presented to patients or physicians. This meant that in most cases there were only one or two studies presenting risk model estimates to the same group and reporting the same outcome. To provide the greatest interpretation of this limited data, where we were able to synthesise the data, in addition to presenting these separately, we also presented an overall summary estimate. Although this means these estimates need to be interpreted with caution, they probably reflect real life as in routine clinical practice this distinction is often not as clear cut: physicians will often discuss risk estimates with patients and patients will often ask physicians for advice on interpreting estimates.

Implications for clinicians and policymakers
Despite these limitations, the results from this review are of relevance to the large number of clinicians worldwide who use CVD risk information with patients regularly in their practice, and policymakers involved in designing and implementing strategies for the prevention of CVD, including the recent debate about the NHS programme of CVD risk reduction in the UK.
The finding that providing patients with risk information changes risk perception and increases the accuracy of perceived risk is consistent with previous reviews which have shown that global risk information with accompanying education or counselling increases the accuracy of perceived risk. 17 However, even immediately after being provided with risk information, 1 in 4 participants still had an inaccurate perceived risk 26 and 1 in 10 changed their perceived risk in the opposite direction to the feedback they received. 21 Such challenges to the communication of risk are well known, 46 and there is scope for further work, but this highlights the need for clinicians to remain aware of the limitations of current methods.
With this inaccuracy in risk perception along with existing experience from other areas about the challenges of behaviour change, it is perhaps not surprising that there is no evidence that simply providing patients with a number leads to statistically significant changes in habitual environmentally cued behaviours such as diet, smoking, physical activity or alcohol intake. Where there were differences between the groups, most were in favour of providing risk information but the studies were generally of poor quality, underpowered and imprecise with most relying on self-reported information. While we are, therefore, unable to rule out the possibility of a small potentially clinically important effect, we can say that it appears unlikely that providing risk information will result in harm through false reassurance and the adoption of unhealthy behaviours. There was also either no change or an improvement in psychological wellbeing, anxiety, worry about future risk of heart attacks, or self-regulation in these studies. This is an important finding as screening programmes based on risk assessment such as the NHS Health Checks in the UK have the potential to cause harm and a key decision when considering implementation is the extent of that harm. Our review suggests that such screening programmes are safe when it comes to psychological well-being.
The effect of the provision of risk information on intermediate measures of CVD and modelled CVD risk was more consistent with studies consistently showing reductions in cholesterol, SBP and DBP, and modelled CVD risk. While the effect sizes are small, they may be of clinical significance at the population level and those studies including only high-risk participants showed significant reductions in cholesterol and SBP. It is possible these effects are mediated through changes in prescribing: although only on the border of statistical significance, patients were 1.4-1.5 times more likely to have a change of lipid-lowering or blood pressure medication when risk information was provided to their physicians and this increased to over twice as likely when only those with a CVD risk >20% were included. This difference in prescribing is not unexpected. Guidelines for prescribing lipid-lowering and blood pressure medication are based on assessment of CVD risk. The increased prescribing for those at high risk therefore likely reflects a greater number of physicians following existing guidance and the smaller effect in those studies including low-risk participants, the effect of including participants in whom no change in treatment would be expected.
Trends towards small reductions in cholesterol and blood pressure were also seen in studies in which risk information was only provided to patients. This highlights the central role patients have in making decisions about their treatment and the impact their risk perception and views about preventive medicine have on the outcomes: while some may be risk averse and start medication at low-risk levels, we know from existing literature that many patients are reluctant to start medication with 5% of people on the streets of London stating that they would not take a statin even if it gave them another 5 years of life, 47 and people responding to a US-based internet survey prepared to pay an average of $1445 (£948; €1265) to avoid taking one pill a day for CVD prevention. 48 The impact of provision of risk information is, therefore, complex and is likely to reflect a combination of factors from initial risk perception, peer comparison, beliefs about the disease, and to-date unmeasured effects such as medication adherence.

Unanswered questions and future research
While this review shows that provision of risk information to patients and physicians improves accuracy of risk perception, increases prescribing and reduces levels of CVD risk factors and modelled CVD risk without causing changes in psychological well-being or anxiety, no studies included actual CVD events as an outcome, and so we are unable to comment on the effect of risk information on CVD morbidity or mortality. Additionally, while there is some suggestion that the effects may be greater in those at higher risk, with such a small number of studies that could simply reflect regression to the mean. There is, therefore, a need for further research with adequately powered trials with objective outcomes and longer follow-up to understand how best to communicate risk information to increase understanding, enhance shared decision-making and encourage behaviour change.
We are also unable to comment on the costeffectiveness of provision of risk information and whether risk information delivered directly to patients is comparable to physician-led risk assessment. In the current age of increasing demand for healthcare and rising costs of treatment, ways of stratifying the population to enable delivery of interventions to those most likely to benefit are of increasing interest, but with such small changes observed in these studies, the question remains as to whether greater benefit could be derived from investment in population-wide prevention strategies 49 rather than screening and individual risk assessment.