Introduction

Modifying health behaviour is a complex process, and assessing the effectiveness of behaviour change techniques remains challenging1,2,3. Despite decades of public health initiatives and clinical efforts aimed at preventing cardiovascular disease (CVD), its prevalence continues to be alarmingly high4. In Estonia, CVD remains a major health concern, with mortality rates exceeding the EU average and a high burden of risk factors such as hypertension, obesity, and smoking5,6. The impact of CVD extends beyond individual health, often leading to hospitalization, long-term disability, and a decline in overall quality of life7,8. Additionally, the economic burden associated with CVD is substantial, placing strain on healthcare systems and national economies due to the costs of treatment, rehabilitation, and lost productivity9,10.

Health behaviour change models such as the Health Belief Model, Transtheoretical Model, and Theory of Planned Behaviour emphasize how individuals’ perceptions, readiness, and intentions shape their actions11,12,13. Delivering these models relies on primary care workers to communicate CVD risk and provide lifestyle counselling aimed at reducing risk factors and improving patients’ self-perceived risk14,15,16. CVD prevention relies on risk factor reduction and awareness, as major risks like hypertension, obesity, and diabetes stem from poor diet and physical inactivity17. While some evidence supports the positive impact of lifestyle interventions on cardiovascular risk factor management, real-world implementation often faces challenges in achieving sustained change18,19,20.

Coronary artery disease (CAD) has a substantial genetic component, with heritability estimates between 40% and 60% in Northern European populations21,22. As genome-wide association studies (GWAS) have expanded, the polygenic architecture of CAD has become increasingly evident, accounting for over one-third of its heritability23,24. However, the emerging field of polygenic risk scores (PRS) remains relatively understudied in real-life clinical trials25,26. PRS holds potential to improve chronic disease prevention and management, offering insights into how lifestyle factors interact with genetic risk to influence health outcomes27. A significant part of PRS-based risk can be mitigated by managing other risk factors, emphasizing the importance of health behaviour change28.

Several studies have begun exploring the link between PRS and health behaviour change, investigating whether genetic risk knowledge can motivate individuals to adopt healthier lifestyles29,30. These studies suggest that personalized risk data may serve as a powerful catalyst for change, but its effectiveness may depend on age31. Our prior research in Estonia found that revealing high CAD PRS to older adults did not significantly improve health behaviours6. With this study, we aim to examine whether younger individuals are more responsive to genetic risk disclosure, potentially identifying an optimal age window for PRS-based interventions to prevent CVD in high-risk groups.

Polygenic risk scores, including coronary artery disease PRS (CAD PRS), can help identify younger individuals who may benefit most from more aggressive lifestyle modifications, encouraging early intervention and health behaviour change32,33,34. Increasingly accessible PRS testing may guide preventive decisions in primary care, offering public health benefits if supported by evidence and training33,35,36,37. Assessing the clinical utility of CAD PRS requires pragmatic trials to evaluate their role in CVD risk prediction, guide preventive interventions, and determine their value in clinical decision-making35,38,39.

This study aimed to assess whether including high coronary artery disease polygenic risk scores (top 20%) in cardiovascular disease risk assessment and communication affects the body mass index of young (25–44 years) overweight adults (BMI 25.0–29.99 kg/m2).

Methods

Study design

This study was a parallel-group, double-blind, randomised controlled trial (RCT). Participants were randomly assigned in a 1:1 allocation ratio to either the intervention or control arm. A second low-PRS control group was matched to the intervention arm by age group and sex to ensure that the control group participants’ PRS remained masked from the family physicians. The intervention and one control group included individuals in the highest CAD PRS quintile (top 20%), while the matched control group was in the lowest PRS quintile. The outcome analyses in this article focus solely on comparisons between the high-PRS groups.

Eligibility criteria

Study participants were selected from Estonian Biobank (EstBB) cohort participants, including healthy overweight individuals aged 25–44 years (BMI 25.0–29.99 kg/m2) whose CAD PRS score had been pre-calculated by the EstBB following the methodology described by Inouye, et al. (2018)40.

The exclusion criteria were an underlying diagnosis of ischemic heart disease (ICD codes I20-25), stroke or transient ischemic attack (I60-64, I69, G45), peripheral vascular occlusion (I65-66, I67.2, I70, I73.9), or diabetes mellitus (E10-14). Individuals currently using cholesterol-lowering medications or those with comorbid physical or mental illnesses that would hinder their ability to provide consent or participate fully in the study were also excluded.

Interventions

Study interventions were delivered by each participant’s family physician in primary care centres across Estonia. All family physicians received detailed intervention guidelines and three training sessions covering risk assessment, lifestyle counselling, dyslipidaemia management, and hypertension treatment based on estimated risk levels. The intervention entailed family physicians assessing CVD risk, sharing it with study participants, and providing counselling (Fig. 1).

Fig. 1
Fig. 1The alternative text for this image may have been generated using AI.
Full size image

A graphical representation of the intervention approach.

At the first study visit, baseline measurements were collected from study participants, and their CVD risk was calculated using a risk assessment tool41. The CVD risk model was developed using Estonian Biobank data from gene donors aged at least 25 years at recruitment with genotyping data available (n = 202 282) as of January 1, 2022. The model predictors were a combination of cardiovascular risk assessment parameters (age, sex, height, weight, smoking status, total cholesterol, HDL cholesterol, systolic BP) and PRS for coronary artery disease (Fig. 2). The cardiovascular risk assessment parameters were primarily based on the SCORE-2 framework, with two key modifications42. First, we used total cholesterol and HDL cholesterol rather than non-HDL cholesterol. Second, we incorporated height and weight into the risk model since BMI was the primary outcome measure of this study. This modification allowed us to assess CVD risk in a younger population than the SCORE-2 model typically addresses.

At baseline, only intervention arm participants were shown their risk trajectory with PRS included. The risk model produced two graphs: one showing the participant’s current risk trajectory, potential risk reduction through lifestyle changes, and average risk for peers of the same age and sex. The second graph (intervention arm only) presented the participant’s genetic risk relative to the general population. Family physicians reviewed these graphs with participants, offered health counselling, and prescribed lipid-lowering or antihypertensive medications when indicated.

Fig. 2
Fig. 2The alternative text for this image may have been generated using AI.
Full size image

Risk assessment and communication to study participants.

The second study visit, six months after the first, was a remote telemedicine session with family nurses via phone. During the call, the nurse offered counselling and collected self-reported data on participants’ weight and waist circumference.

The third study visit was conducted at the end of the 12-month follow-up, during which updated risk assessments were performed for all participants. For the control group, polygenic risk was now included in the risk assessment, unblinding both physicians and participants to their high PRS status. Changes in overall CVD risk were reviewed, and additional counselling or treatment was offered as needed.

Outcomes

The primary outcome measure was mean BMI (kg/m2) at 12 months, comparing the intervention and control groups. Secondary outcome measures included the difference in mean systolic and diastolic blood pressure (mmHg), mean total cholesterol (mmol/L), mean LDL-and HDL-cholesterol (mmol/L), mean blood glucose (mmol/L), mean waist circumference (cm), and 10-year CVD risk (percentage points) at 12 months. Smoking-related outcomes included the difference in the proportion of current smokers and the mean number of cigarettes consumed among daily smokers between groups at 12 months. Physical activity was evaluated by comparing mean MET values between groups at 12 months. Dietary outcomes included the difference in the proportion of individuals consuming at least the recommended daily amount of fruit (> 200 g) and vegetables (> 300 g). The study also compared the proportion of participants in each study arm for whom family physicians prescribed lipid-lowering or antihypertensive medications during the study period.

Self-reported measures

Online surveys were administered via REDCap at 0, 6, and 12 months to participants for recording their smoking status, alcohol consumption, diet (fruit and vegetable consumption), and physical activity43.

Physical activity was evaluated by calculating SQUASH (Short Questionnaire to Assess Health-enhancing Physical Activity) scores based on online participant surveys administered via REDCap. SQUASH is a validated tool used in research to estimate activity in different domains of daily life, making it suitable for calculating Metabolic Equivalent of Task (MET) scores to quantify energy expenditure44.

Harms

Potential adverse effects related to participation (e.g., increased anxiety, stress from risk communication) were monitored throughout the study via participant self-report and routine follow-up.

Sample size

The sampling frame included all Estonian Biobank (EstBB) cohort participants (202,282 as of 1 January 2022). Eligible participants were aged 25–44, overweight (BMI 25–29.99), undiagnosed with cardiovascular disease or diabetes, and belonging to either the top or bottom quintile for CAD PRS. Preliminary data identified ~ 2,000 individuals in each quintile meeting these criteria. From these, 1300 high-PRS individuals were selected for the intervention and first control group, and ~ 650 low-PRS individuals for a second control group, anticipating a 70% response rate and a final sample of ~ 1,500 (500 intervention, 1,000 controls). Power calculations indicated that this sample size would detect an expected BMI difference of 0.4 kg/m2 one year after the start of the study, a type I error of 0.05, and a study power of 0.8. This is deliberately conservative compared to typical RCTs showing at least 2 kg weight loss45,46. We anticipated smaller effects given our pragmatic primary care setting, with less intensive support and a comprehensive CVD focus rather than weight-loss alone. Sampling also considered family physician practice lists, prioritising those with more eligible participants.

Randomisation

Sequence generation

The study employed simple randomisation using a computer-generated algorithm. A statistician from the EstBB not involved in participant recruitment carried out the randomisation. Randomisation was performed only for the high-PRS intervention and control group. The secondary control group with the lowest quintile PRS was matched to the intervention group based on 5-year age-group and sex.

Blinding

Control group participants and family physicians were blinded to the CAD PRS of control arm participants until the end of the trial. This was done by matching a secondary control group with the lowest quintile PRS to the intervention group based on 5-year age-group and sex. Incorporating a secondary control group for blinding purposes was done to ensure that family physicians were masked to the PRS (high or low) of attending control group participants, thus minimizing any potential influence on the study participants through counselling or other interventions.

Statistical methods

A p-value < 0.05 was considered statistically significant. The primary analyses followed a per-protocol approach, including only high-PRS participants who consented and completed the intervention as planned at all time points. Continuous baseline variables were described using means and standard deviations and compared by t-tests. In case of a skewed distribution, median and interquartile ranges were presented, categorical variables were expressed as frequencies and relative frequencies, and compared by chi-square tests. Cardiovascular risk factors were compared between the two groups at baseline, 6 months, and 12 months. Outcome assessments for visits 2 and 3 were analysed using ANCOVA, adjusting to baseline values to correct for potential baseline imbalances and to increase the precision. A Wilcoxon rank-sum test was used for 10-year CVD risk comparison. Within-group differences from baseline were assessed using paired t-tests. The number of prescriptions purchased per follow-up year was compared with of a negative-binomial model adjusted to age and sex, IRR (incidence rate ratio), and 95% confidence intervals are presented. The analysis was performed using Stata 14.2 and R (version 2024.12.0 + 467).

Results

Recruitment and participant flow

We invited 2864 eligible individuals to participate (Fig. 3) during April 2022- August 2023. Of those, 1286 were excluded (1087 declined to participate, and in 199 cases their GP declined to participate). A total of 1020 participants were randomised into an intervention (n = 510) and a control group (n = 510). The drop-out rate before the first trial visit was 9.0% in the intervention arm and 12.0% in the control arm, resulting in baseline data being analysed for intervention (n = 464) and control group (n = 449) participants. Pregnant individuals at the baseline visit were omitted from the data analysis (n = 20). A higher proportion of younger individuals were lost between randomisation and the first study visit. However, the intervention and control groups remain comparable, with both groups consisting of slightly older participants than those who initially consented to join the trial. The overall dropout rate from randomisation to the last trial visit was 18.2% for the intervention arm and 18.8% for the control arm. The last participant completed follow-up in November 2024.

Fig. 3
Fig. 3The alternative text for this image may have been generated using AI.
Full size image

Participant flowchart. The flowchart presents the analysis for the primary outcome, with secondary outcome analyses involving a varying number of observations. FP family physician.

Baseline data

The baseline characteristics of the high-PRS intervention and control arm participants are presented in Table 1. The mean (SD) age of the study population was 37.7 (5.4) years, and 45.5% were women. The mean BMI was 27.7 (2.2) kg/m2. The study population had low behavioural risk impact, with minimal smoking and moderate alcohol consumption, with over half reporting no or very infrequent consumption, and only 1.8% drinking daily. There were no significant intergroup differences except for smoking status (p = 0.043).

Table 1 Baseline sample characteristics.
Full size table

Changes in BMI and cardiovascular risk factors

Regarding BMI, no significant differences were observed between the intervention and control groups at 6 months (p = 0.713) or 12 months (adj. Δ = 0.11 kg/m2, 95% CI = − 0.09 to 0.30, p = 0.291) (Table 2). Waist circumference increased slightly in both groups at the 6-month mark, but by the end of the trial, the difference between the groups was not statistically significant (p = 0.143).

There were no statistically significant differences between groups at 12 months in systolic blood pressure (0.60 mmHg, 95% CI − 1.97 to 0.77, p = 0.393), or other parameters of interest. Systolic blood pressure slightly decreased in both groups throughout the trial, but the change was not statistically significant (intervention group: p = 0.496, control group: p = 0.724).

Baseline 10-year CVD risk calculations differed and were not directly comparable, as the intervention group’s score at baseline included both traditional risk factors and PRS (median 2.1, IQR = 1.1–3.7), while the control group’s score was based only on traditional factors (median 1.8, IQR = 0.9–3.2). At the 12 months visit, the risk scores were calculated for both groups using a combination of traditional risk factors and PRS, making them comparable. The median risk in the intervention group rose to 2.2 (SD = 1.2–3.8) and in the control group to 2.1 (SD = 1.2–3.9) mainly due to the population age increase. No statistically significant differences were observed between the study groups’ combined risk scores at the end of the trial.

Gender- and education-stratified analyses for BMI showed no significant differences between groups. Similarly, BMI analysis in high CVD-risk individuals (Q3) found no significant variation (Table S1). Comparisons between the intervention and low-PRS control arm have been provided in Table S2.

Table 2 Outcome analysis of the primary and secondary outcomes.
Full size table

Preventive treatment patterns

Family physicians prescribed anti-hypertensive (C03–C09) or lipid-lowering (C10) medications to 15.5% of participants in the intervention group and 19.2% in the control group (Table 3). The number of prescriptions for these medications was similar across groups, indicating no significant effect of the intervention on medication prescribing behaviour.

In negative-binomial regression models adjusted for age and sex, the incidence rate ratio (IRR) of package purchases in the intervention group compared to the control group was 0.93 (95% CI: 0.73–1.18) for lipid-lowering drugs and 0.97 (95% CI: 0.79–1.21) for anti-hypertensive drugs, indicating no significant effect of the intervention on medication purchasing behaviour.

We also compared the number of prescriptions issued in the first versus last 90 days of follow-up between intervention and control arms using a negative-binomial model with an offset for person-years and found no significant differences between arms.

Table 3 Comparison of prescribed and purchased lipid-lowering and anti-hypertensive medication during the study period.
Full size table

Self-reported CVD risk factor outcomes

Self-reported smoking behaviour remained largely unchanged over 12 months, with a non-significant 1.9% decline in daily smoking in the intervention arm (McNemar’s χ2 = 0.64, p = 0.40) and a 0.8% decline in controls (χ2 = 0, p = 0.999), and no between-group difference in cigarettes per day (t = − 0.30, p = 0.76). Alcohol consumption saw a modest net increase in abstinence (2.3%), driven by significantly more participants quitting than starting drinking (χ2 = 6.09, p = 0.014), with similar patterns present in both study arms. Physical activity (MET score) between-group differences at 12 months were non-significant (W = 75 974, p = 0.35). Dietary analyses of recommended intake for vegetables and fruit did not reveal between-group differences. A detailed overview of these self-reported outcomes, along with illustrative figures and full statistical results, has been provided in the Supplementary Materials.

Discussion

In this randomised controlled trial of 1020 young overweight individuals, we found no evidence that communicating high CAD polygenic risk would result in behaviour changes sufficient to reduce cardiovascular risk factors over 12 months. However, our findings provide valuable insights that may help inform the development and evaluation of alternative approaches for incorporating polygenic risk into CVD primary prevention.

Knowledge about polygenic risk can help identify individuals at higher relative risk of early CVD31. However, genetic risk may be perceived as predetermined rather than a motivator for behaviour change and in a generally healthy population like ours, with slight overweight, good education, low risk factors, minimal alcohol and tobacco use, and mildly elevated LDL, such information may be less effective in driving further lifestyle improvements47,48,49. Additionally, the young age of our participants may have contributed to limited intervention effectiveness. Older adults demonstrate greater weight loss in lifestyle interventions, motivated by health decline and comorbidity awareness45,50. In contrast, young adults are primarily driven by appearance and social factors rather than long-term health concerns51.

Another consideration is the one-on-one counselling used in this trial, reflecting routine primary care. More frequent contact and alternative delivery methods may be needed to support behaviour change in this age group52,53,54. Studies show that weight loss interventions achieve better long-term results when supported by ongoing personal contact, with recommendations for at least quarterly or preferably monthly visits55,56.

In pragmatic trials, using participants’ own family physicians to deliver interventions has been shown to potentially reduce the impact on behaviour change, as such trials reflect routine care with minimal alteration to usual practice, which can weaken intervention delivery57,58. While we achieved much lower attrition than similar lifestyle RCTs, younger participants typically demonstrate higher dropout rates in weight loss trials, and retention difficulties may have still contributed to the trial being underpowered59,60,61.

In addition, the novelty of polygenic risk — not yet widely adopted in clinical practice or fully covered in medical education until recently — may further complicate its integration into risk communication between physicians and patients62. As a result, many physicians may find it challenging to interpret, even after receiving training on the subject63,64.

Trial participation may increase social desirability, resulting in over-reporting of behaviours, as heightened risk awareness can introduce bias into self-reported outcomes65,66. This may have led to more accurate self-reporting of behavioural risk factors such as smoking prevalence, slightly contributing to the rise in 10-year CVD risk. However, this increase was likely primarily driven by aging during the trial, as age is a major determinant of cardiovascular risk. Assessing how participants perceived their own weight could have been informative, as those viewing themselves as having normal weight may employ smoking as a weight management tool—a consideration particularly relevant given our cohort was only slightly overweight rather than obese67. Although the study population was generally inactive, with average MET scores below 3 (the threshold for health-protective activity), trial participation may have influenced the control group’s increased activity levels, possibly due to their unawareness of polygenic risk or a Hawthorne effect, where behaviour changes simply from being observed68,69,70.

Although no significant difference in prescribing lipid-lowering or blood pressure medications was observed between study arms, the fact that physicians initiated preventive treatment in 25–44-year-olds—who typically would not qualify under current guidelines—represents an important finding. This suggests that participation in the trial increased clinical attention to early cardiovascular risk, even without a PRS-driven effect.

This pragmatic, real-life trial offers valuable insights, yet its one-year duration may not capture long-term behaviour changes that develop gradually or occur outside the structured RCT framework. In this context, a realist evaluation could provide valuable insights, focusing not just on outcomes but on the underlying mechanisms and contextual factors behind the intervention delivery71. This approach may help identify whether adjustments in delivery methods, target populations, or other factors are needed to improve strategies for integrating genetic risk into CVD primary prevention and to evaluate its long-term cost effectiveness72,73.

Our findings indicate that merely knowing one’s polygenic risk for coronary artery disease is not sufficient to drive significant health behaviour change. However, previous research suggests value in equipping primary care professionals with more precise risk assessments, allowing them to proactively initiate preventive treatments such as lipid-lowering or blood pressure medications. This approach may prove more effective than relying solely on individual motivation for lifestyle modifications74.

Strengths and limitations

This is the first clinical trial to test CAD PRS risk communication in a real-life primary care setting with younger individuals. Until now, it was unclear whether young people with high polygenic risk would feel compelled to adopt healthier behaviours, but this randomised trial strongly suggests that risk communication alone, even using novel indicators like PRS, is insufficient to drive short-term behaviour change and likely not enough for long-term change either. This finding highlights the need to further explore and test various primary prevention strategies for high-PRS individuals.

A key limitation of the study was participant drop-out between randomisation and the first study visit, primarily due to a slow randomisation process that delayed family physicians in scheduling appointments. Although the combined risk assessment model underwent extensive pre-implementation testing, we have not published the specific model used in this trial. However, a similar model utilizing the same clinical parameters with an updated PRS has recently been published75. Challenges in risk communication may have impacted the intervention’s effectiveness, as family physicians had to navigate both new software and the relatively recent concept of polygenic risk when explaining cardiovascular risk to patients.

Conclusion

This trial showed that communicating high CAD PRS alone is not enough to prompt meaningful behaviour change in a young population. Our results highlight that polygenic risk behaves like any other cardiovascular risk factor and should be considered within the broader context of an individual’s overall risk profile.