Risk factor heterogeneity across cardiovascular disease outcomes in treatment-eligible hypertensive patients

Authors: Olivia Gästgivars, Urban Wiklund, Mattias Brunström

Abstract

Objective:

Prediction of specific cardiovascular outcomes may facilitate precision medicine in hypertension. We aimed to explore risk factor heterogeneity for acute coronary syndrome (ACS), stroke and heart failure in treatment-eligible hypertensive patients.

Methods:

In this population-based cohort study in Västerbotten, Sweden, middle-aged patients with treatment-eligible hypertension were included. Association between traditional cardiovascular risk factors and ACS, stroke and heart failure was studied. Multinomial logistic regression was performed to determine standardized odds ratios.

Results:

Over a 10-year period, there were 1190 cardiovascular events in 12 238 participants. Risk factor associations were mainly of the same direction but differed in magnitude across cardiovascular outcomes. Total cholesterol was significantly associated with ACS, but not with heart failure and stroke. Smoking and male sex were more strongly associated with ACS than with heart failure and stroke. BMI and socioeconomic factors were significantly associated with heart failure, but not with ACS and stroke. Diabetes and DBP had stronger associations with stroke compared to ACS and heart failure.

Conclusion:

We found clinically important differences in the association between different risk factors and ACS, stroke and heart failure in treatment-eligible hypertensive patients. The current model was insufficient to guide treatment decisions but should stimulate further research into risk factor heterogeneity between cardiovascular disease outcomes.

INTRODUCTION

All first-line antihypertensive drug classes effectively reduce the risk of cardiovascular disease (CVD) and death [1]. However, differences between drug classes in their preventive effect for myocardial infarction, stroke and heart failure may exist [2,3]. For example, calcium channel blockers have been shown to prevent stroke more effectively compared to other agents, whereas diuretics are superior in preventing heart failure. These differences are not considered in guidelines or clinical practice, partly because risks for specific CVD outcomes are unknown.

Current cardiovascular risk assessment tools, such as SCORE2, can predict who may benefit from preventive treatment, but they cannot discriminate which CVD outcome is most likely [4–6]. Risk can be estimated by traditional, modifiable risk factors that account for the majority of CVD cases and are common to ischemic heart disease, stroke, and heart failure [7,8]. However, risk factor profiles differ slightly across CVD outcomes. Among traditional risk factors, hypertension is the most important for stroke and heart failure, while abnormal lipid profile is the leading risk factor for acute myocardial infarction [9–11]. These differences suggest that it may be possible to identify which specific type of CVD an individual is most likely to develop. This, in turn, would enable personalized antihypertensive treatment, with the potential to enhance primary prevention.

The aim of this study was to investigate differences in the association between common risk factors and acute coronary syndrome (ACS), stroke or heart failure in treatment-eligible hypertensive patients, with the ultimate goal to be able to guide personalized antihypertensive treatment.

MATERIALS AND METHODS

Data source

In this cohort study, data from local databases and national registers in Sweden were used. Using the personal identification number of each resident, data from the Swedish Stroke Prevention Study (SSPS) and Västerbotten Intervention Programme (VIP) were combined with the National Patient Register (NPR), the Swedish Cause of Death Register (SCDR) and information on socioeconomic status from Statistics Sweden. The study was approved by the Swedish Ethical Review Authority (dnr 2025–04985–02).

SSPS was a study evaluating a primary care intervention in Västerbotten county. In that study, a cohort of all blood pressures recorded in primary care from the implementation of the electronic medical record system (1992–1993) to December 2011 was created. The SSPS database also contains data on hypertension and diabetes diagnoses [12].

VIP is a health intervention program in Västerbotten County, continuously ongoing since 1985. VIP invites all residents turning 40, 50 and 60 each year for a health examination, assessing cardiovascular risk factors from clinical measurements, blood samples and questionnaires, including lipid profile, glucose measurements, BMI and smoking status. The participation rate was lower in the early years (around 50–60%) but increased to 66–67% during 2005–2010, with similar socioeconomic status among participants and nonparticipants [13–16].

The NPR covers all inpatient care since 1987, with diagnosis codes from hospital admissions and visits. Diagnoses are typically registered at the discharge date from inpatient care. The validity of the NPR is high for diagnoses heart failure, stroke, and myocardial infarction [17]. The SCDR contains cause of death data from 1952 and onwards, classified according to the ICD (International Classification of Diseases) system. Causes of death and data on diagnoses from inpatient care were available until 31 December 2019.

Study sample

Individuals were included when they first were identified with hypertension, to resemble the clinical scenario when antihypertensive treatment would be initiated. Treatment-eligible hypertensive individuals were identified from SSPS by either having two consecutive blood pressure recordings of at least 140 mmHg SBP or at least 90 mmHg DBP, or by having a hypertension diagnosis and at least one prior blood pressure value of at least 140 mmHg SBP or at least 90 mmHg DBP. Participants meeting the inclusion criteria after 31 December 2009 were excluded, as they had less than 10 years of possible follow-up time.

Individuals were excluded if they had less than two blood pressure values before the entry of the study. Individuals with no VIP-examination within 5 years prior to the entry of the study were excluded, to avoid outdated risk factor data. Individuals were also excluded if prior to the study entry they had a diagnosis of ischemic heart disease (ICD-8/9 code 410,414; ICD-10 code I20-I25), heart failure (ICD-8/9 code 402,404, 428; ICD-10 code I11, I13, I50), stroke (ICD-8/9 code 431, 434, 436; ICD-10 code I61, I63 excluding I63.6, I64), transient ischemic attack (ICD-8/9 code 435; ICD-10 code G45.3, G45.9) or had undergone a prior procedure of percutaneous coronary intervention (procedure codes 3080, FNG00, FNG02, FNG05; ICD-10 code Z95.5), coronary artery bypass grafting (procedure codes 3066–3068, 3105, 3127, FNA-FNE; ICD-10 code Z95.1) or valve replacement (ICD-10 codes Z95.2-Z95.4), as these diagnoses could potentially influence treatment decisions and future outcomes.

Risk factors

Initially, a wide set of variables were included from which multiple models were run to minimize collinearity and optimize explanatory value (Supplementary Methods). Using a stepwise approach, the impact of individual risk factors on the model were examined, removing them one by one. First, in groups of risk factors with presumed collinearity (blood pressure, glucose and socioeconomic related, respectively) and then in combined models (Supplementary Methods). Model selection was based on minimizing the Akaike Information Criterion (AIC), with the exception that excluding fasting glucose from the final model yielded slightly higher AIC but improved the explanatory value of diabetes diagnosis.

In the final model, age, sex, mean SBP and mean DBP were included from the SSPS database. Mean SBP and DBP were calculated from all recorded blood pressure values preceding date of inclusion. BMI (kg/m^2^), total cholesterol (mmol/l) and smoking status (defined as smoker, nonsmoker, ex-smoker, occasional smoker, previously occasional smoker) were included from VIP, and the diagnoses atrial fibrillation (ICD-8/9 code 427D, ICD-10 code I48), hypertension and diabetes before the date of inclusion were included from NPR and SSPS. Income and educational level were gathered from Statistics Sweden.

As the rate of missing data was low (3.3%), we excluded participants with missing values in any of the risk factors, and hence a complete case analysis was performed.

Outcomes

The outcomes were first fatal or nonfatal ACS (ICD-8/9 410, 411B; ICD-10 I20.0, I21), heart failure (ICD-8/9 code 402,404, 428; ICD-10 code I11, I13, I50) or stroke (ICD-8/9 code 431, 434, 436; ICD-10 code I61, I63 excluding I63.6, I64), or the absence of these events after 10 years. The event was regarded as ACS in cases where ACS and heart failure were registered at the same date, as it was probable that heart failure developed from ACS. Cases where stroke and ACS or heart failure were registered at the same date were excluded, as we could not ascertain which event occurred first.

Statistical analysis

Baseline characteristics were reported as frequencies and percentages for categorical variables, and as means and standard deviations (SDs) for continuous variables.

To explore risk factors for a specific cardiovascular event, multinomial logistic regression was performed. The event-free outcome was used as reference category, and ACS, stroke and heart failure were considered alternative outcomes.

Standardized odds ratios (ORs) with 95% confidence intervals (CIs) were derived for each risk factor for each outcome, expressing ORs per SD for continuous variables. Estimated response probabilities were calculated for each individual, for each outcome.

AUC was calculated to assess the model's ability to distinguish between the cardiovascular outcomes and the event-free outcome. To estimate confidence intervals for AUC values, five-fold cross validation with stratified subsampling was repeated 200 times. In each loop, data were randomly divided in five subsets of equal size, using stratified subsampling (same proportion of individuals with different events in each subset as in the total cohort). In the five-fold cross validation, five iterations were performed where different combinations of four of the subsets (80%) were used to determine new multinominal logistic regression models.

Sensitivity analyses were performed stratified on sex and by restricting the analysis to individuals who experienced a cardiovascular event, using ACS as the reference category.

Analyses were performed in IBM SPSS Statistics version 28.0.1.1 and R Statistical Software version 4.4.3.

RESULTS

Of the 12 667 individuals fulfilling the inclusion criteria, 429 (3.4%) were excluded because of missing risk factor data or because of concurrent stroke and ACS and/or heart failure (Supplementary Figure 1). Finally, 12 238 participants, of which 5889 (48%) men, were analyzed. The mean age at inclusion was 54.6 years (SD 7.7) and mean blood pressure was 148.9/90.0 mmHg (SD 14.7/8.2) (Table 1). During the 10-year follow-up period, 1190 first cardiovascular events occurred, of which 594 cases were ACS (of which 51 fatal), 204 cases heart failure (10 fatal), 366 cases stroke (five fatal). There were 26 cases with concurrent ACS and heart failure, classified as ACS.

Association between risk factors and cardiovascular disease outcomes

Several differences were observed in the association between risk factors and outcomes (Fig. 1, Supplementary Table 1). Higher total cholesterol implied significantly increased odds for ACS (standardized OR 1.33, 95% CI 1.23–1.44), while not significantly associated with stroke or heart failure. Active smoking and male sex were more strongly associated with ACS, although these risk factors were significantly associated with increased risk for heart failure and stroke as well.

For heart failure, higher BMI and lower education level were significantly associated with increased risk, while these risk factors were not significantly associated with ACS and stroke. Mean SBP was significantly associated with higher odds for all outcomes, but mean DBP presented with contrary odds for stroke (standardized OR 1.17, 95% CI [1.03–1.34]) compared to ACS and heart failure (0.83 [0.74–0.92] and 0.78 [0.66–0.93], respectively). Diabetes conferred the highest risk for stroke (OR 2.13, 95% CI [1.41–3.21]) and ACS (1.84, [1.34–2.52]) but was not significantly associated with heart failure. Age was associated with increased risk for all outcomes, but associations were stronger for stroke and heart failure.

The direction and magnitude of associations remained essentially unchanged in separate analyses for men and women, however some associations lost significance in analyses in women because of attenuated statistical power (417 events, compared to 773 events in men) (Supplementary Tables 2 and 3). In sensitivity analysis of people with cardiovascular events only, associations changed expectedly as reference point was changed to ACS but yielded similar interpretations as in the main analysis (Supplementary Table 4).

Explanatory value

The fitted model was significantly better at explaining event status compared to the intercept model (χ^2^ = 790.59; P < 0.001). AUC indicated moderate ability to discriminate from event-free outcome with values of 0.74 (95% CI 0.73–0.75) for ACS, 0.73 (0.72–0.75) for heart failure and 0.69 (0.68–0.71) for stroke (Fig. 2). The discrimination between specific CVD outcomes was however poor (Supplementary Table 5, Supplementary Figure 2).

DISCUSSION

This study aimed to explore differences in risk factor profiles for ACS, stroke and heart failure in treatment-eligible hypertensive patients. Risk factor associations were mainly of the same direction, but with important differences in magnitude across CVD outcomes. Sex, smoking, and total cholesterol stood out as most important for ACS, while BMI and socioeconomic factors appeared to be most important for heart failure. Diabetes and DBP was distinguished for stroke.

This is mainly in line with results from previous studies, exploring risk factor associations for individual outcomes, although the literature on risk factor heterogeneity across CVD outcomes is sparse in general, and nonexisting for treatment-eligible hypertension in particular. A study comparing risk factors for coronary events and ischemic stroke, in a population-based, middle-aged Swedish cohort, found stronger associations between coronary events and smoking, lipid markers and male sex, while age was preferentially associated with ischemic stroke [18]. A study assessing age-dependent associations between risk factors and subtype of CVD, showed overall stronger associations for both dyslipidemia and smoking with myocardial infarction, as compared with stroke and heart failure, in a Japanese population [19]. BMI has previously been shown to more closely relate to heart failure compared to other cardiovascular events [20,21]. This is also coherent with the recent proposition that obesity acts as a principal driver of heart failure with preserved ejection fraction through imbalanced adipokine signaling, potentially even overriding the impact of hypertension [22–24].

We found higher SBP to be associated with increased risk for all outcomes, while an inverse relationship for DBP was shown for heart failure and ACS in this treatment-eligible hypertensive population. Increase in SBP and DBP has consistently been shown to be associated with a higher risk for all vascular outcomes, especially stroke [11,25,26]. The decreased risk with higher DBP values shown here is likely to reflect increased risk with higher pulse pressure in this particular group. Pulse pressure is a recognized marker of arterial stiffness, associated with cardiovascular events and mortality [27–29].

The strengths of this study include a large representative sample from primary care, comparing multiple risk factors readily available in clinical practice and their association with the three most important CVD outcomes. However, there are some limitations. First, some recognized risk factors, such as alcohol use, physical activity and renal function, were not included in the analysis. Second, the sample is limited to a quite homogenous middle-aged population in northern Sweden, which may compromise generalizability to other ethnic and age groups. The generalizability might also be impaired due to selection bias in social and health factors, as data comes from a voluntary health examination program (i.e., VIP). However, previous comparisons of VIP participants to nonparticipants conclude small socioeconomic differences [15,16]. Third, cases of incident heart failure might be underestimated, due to diagnoses being measured by hospitalization or death, not detecting heart failure patients managed in the outpatient clinic.

While the rationale behind the study was to enable tailored antihypertensive treatment for treatment-eligible hypertensive patients, through identifying the most likely CVD outcome, our model provides limited clinical usefulness. However, the evident differences in risk factor patterns between the outcomes found suggest a potential for accurate prediction of specific CVD outcomes, which should be explored in future research. Ambale-Venkatesh et al.[30] found machine learning methods to improve prediction of subtypes of CVD. Machine learning and incorporation of nontraditional risk factors might provide ways forward in this question.

With suboptimal discrimination of individual CVD outcomes, it could be argued to primarily target treatment against the most common outcome, which here was ACS, but this will differ across populations. For example, heart failure is more prevalent in black people, while stroke burden is highest in east and central Asia [31,32]. In an older population, heart failure and stroke might be more common [33]. It should also be recognized that guidelines recommend two-drug combination as first-line therapy in most patients, which allows for simultaneous targeting of multiple outcomes, and with greater BP reduction being associated with improved prognosis in general [34–38].

CONCLUSION

We found differing risk factor patterns for ACS, heart failure and stroke in treatment-eligible hypertensive patients. This might enable future studies to predict individual CVD outcomes, which ultimately could help tailor antihypertensive treatment for individual patients.

ACKNOWLEDGEMENTS

This study was designed and performed in collaboration with Artificial Intelligence for Medicine in Northern Sweden (AIM North).

Financial support was provided through a regional agreement between Umeå University and Västerbotten county council (ALF, grant number RV-1010705, RV-1005575), the Swedish Heart and Lung Foundation (grant number 20241373) and the Heart Foundation of Northern Sweden. The funders did not take part in any aspect of the study.

Conflicts of interest

M.B. has received modest honoraria from AstraZeneca, Medtronic and Amarin and is executive officer for the European Society of Hypertension. The other authors report nothing to declare.

Supplementary Material