TRAJECTORIES OF RISK FACTORS AND RISK OF NEW-ONSET ATRIAL FIBRILLATION IN THE FRAMINGHAM HEART STUDY

Abstract

The associations of long-term patterns of risk factors and the risk of incident atrial fibrillation are incompletely characterized. Among 4351 Framingham Study participants (mean age 50±11 years at baseline examination, 57% women) from the Original and Offspring cohorts we defined longitudinal patterns, referred to as trajectories, of atrial fibrillation risk factors and a composite atrial fibrillation risk score using approximately 16-years of data. We used Cox proportional hazards models to examine the association of trajectories to 15-year risk of atrial fibrillation. During follow-up, 719 participants developed atrial fibrillation. Five distinct trajectory groups were identified for systolic blood pressure: groups 1 and 2 (normotensive throughout), group 3 (prehypertensive), group 4 (hypertensive initially with decreasing blood pressure), and group 5 (hypertensive and increasing blood pressure). In multivariable-adjusted analyses, compared to group 1, groups 4 (hazard ratio, 2.05; 95% confidence interval, 1.24 to 3.37) and 5 (hazard ratio, 1.95; 95% confidence interval, 1.08 to 3.49) were associated with incident atrial fibrillation. Three trajectory groups were identified for antihypertensive treatment. Compared with the group with no treatment throughout, the other two groups were associated with increased risk of incident atrial fibrillation. Distinct trajectories for diastolic blood pressure, smoking, diabetes, and the composite risk score were not associated with increased 15-year risk of atrial fibrillation. Longitudinal trajectories may distinguish how exposures related to atrial fibrillation contribute towards prospective atrial fibrillation risk. Distinct trajectory groups with persistently elevated systolic blood pressure and longer antihypertensive treatment are associated with increased risk of incident atrial fibrillation.

Introduction

Over the past few decades a wide variety of risk factors for atrial fibrillation (AF) have been identified. Established risk factors have been incorporated into risk models from the Framingham Heart Study (FHS),¹ Atherosclerosis Risk in Communities,² Women’s Genome Health Study³ and Cohorts for Heart and Aging Research in Genomic Epidemiology (CHARGE)-AF consortium.⁴ Previous studies generally have evaluated risk factors for AF, such as blood pressure (BP) or body mass index, at a single time point. However, mapping the longitudinal trajectory of risk factors may reflect the cumulative exposure of covariates that contribute to the autonomic, neural, electrical, and structural changes promoting the development of AF. The longitudinal pattern of risk factors for AF has not been rigorously studied.

The aim of the present study was to characterize the longitudinal patterns, we refer to as trajectories, of risk factors of AF and the composite risk score from the CHARGE-AF consortium, using multiple decades of data available from the FHS. The CHARGE-AF risk score has been validated in two independent studies.^5,6 We hypothesized that risk factors and CHARGE-AF risk score trajectories are associated with increased 15-year risk of incident AF.

Methods

Subjects

The FHS is a prospective community-based study that has been described previously.^7,8 Briefly, 5209 participants were recruited in 1948 into the Original cohort and underwent systematic biennial examinations. In 1971, 5124 children of the Original cohort and their spouses were enrolled into the Offspring cohort and underwent routine examinations every 4–8 years. All participants gave informed consent. The study protocol was approved by the Institutional Review Board of the Boston University Medical Center.

Clinical variables

Participant’s medical history, physical examination, blood tests, and 12-lead electrocardiogram were obtained routinely at each FHS follow-up examination. In addition, records including electrocardiograms from outpatient visits or hospitalizations between examinations were routinely retrieved and reviewed. Participants were defined to have AF if AF or atrial flutter was confirmed by a FHS cardiologist on review of electrocardiograms, as detailed previously.⁸

Weight was measured in kilograms and height in meters. Current smoking (yes/no) was defined as the self-reported use of ≥1 cigarette/day within the year prior to the FHS examination. Smoking pack-years were calculated by multiplying the number of years of smoking by average number of cigarettes smoked per day and then divided by 20. Two seated systolic and diastolic BPs were measured after a 5-minute rest period, and the average of each was taken as the FHS BP. Hypertension was defined as the presence of systolic BP ≥140mmHg and/or diastolic BP ≥90mmHg, or self-reported use of anti-hypertensive medications. Pulse pressure was defined as diastolic BP subtracted from systolic BP. Diabetes mellitus was diagnosed if the participant reported use of insulin or an oral hypoglycemic agent, or had a fasting glucose level ≥126 mg/dl, or a random blood glucose ≥200mg/dl at the FHS examination. Myocardial infarction and heart failure were diagnosed by review of hospital records and physician reports, and adjudicated by three FHS cardiologists.⁹

Trajectory Modeling and Statistical Analyses

We modeled 16-year longitudinal patterns of AF risk factors and the CHARGE-AF risk score as trajectories using the PROC TRAJ command in SAS.^10,11 Trajectory analysis identifies latent patterns of longitudinal data by employing a semiparametric, group-based modeling strategy in which each group represents individuals with similar trajectories. We limited the number of trajectory groups to ≤ 5, using the Bayesian Information Criterion to assess best model fit. The Original and Offspring cohorts were pooled for trajectory analyses.

The CHARGE-AF simple risk model includes age, race, height, weight, current smoking status, systolic and diastolic BP, hypertension treatment, diabetes mellitus, history of myocardial infarction, and history of heart failure.⁴ We performed trajectory analyses of risk factors from the composite risk model including weight, current smoking status, systolic and diastolic BP, hypertension treatment, and diabetes mellitus. Trajectory modeling was not performed for age and for risk factors that would not change – race, height, history of myocardial infarction and history of heart failure. We also performed trajectory analysis on the presence or absence of hypertension. In secondary analysis, we performed trajectory modeling on pulse pressure as a hypothesis-generating exercise.

In addition, we performed trajectory analysis on the CHARGE-AF risk score. Age was excluded from the CHARGE-AF risk score in our trajectory modeling because aging is the same for all participants and would increase the risk score over the 16-year trajectory modeling period. As described by the CHARGE-AF consortium,⁴ the risk score was calculated as the weighted sum of the remaining risk factors, specifically: $\begin{array}{l}\frac{\left[0.248\times \mathit{\text{height}}\left(cm\right)\right]}{10}+\frac{\left[0.115\times \mathit{\text{weight}}\left(kg\right)\right]}{15}+\frac{\left[0.197\times \mathit{SBP}\left(\mathit{mmHg}\right)\right]}{20}-\frac{\left[0.101\times \mathit{DBP}\left(\mathit{mmHg}\right)\right]}{10}\\ +\left[0.359\times \mathit{\text{current smoker}}\right]+\left[0.349\times \mathit{\text{hypertension treatment}}\right]+\\ \left[0.237\times \mathit{\text{current diabetes}}\right]+\left[0.701\times \mathit{\text{prevalent heart failure}}\right]+\left[0.496\times \mathit{\text{prevalent MI}}\right]\end{array}$; all dichotomous variables were coded as 1 for true, and 0 for false.

For the Original cohort, participants (n= 2363) were included in trajectory modeling using data from examinations 13, 14, 16, 18 and examination 20 (1972–1990). Offspring cohort participants (n=3945) were included from examination cycles 2–6 (1979–1998).

The hinge examination was defined as the final examination for trajectory modeling, and the examination from which the follow-up for AF began (Figure 1). The hinge was examination cycle 20 for the Original cohort and examination cycle 6 for the Offspring cohort.

Figure 1.

Trajectory modeling is based on the trajectory phase if participants attended at least 3 examination cycles over 16 years. The hinge examination was the last examination within the trajectory phase. If participants attended the hinge examination, they were included in the follow-up phase. The follow-up phase began at the hinge examination for 15-year follow-up to identify AF events.

FHS = Framingham Heart Study; AF = atrial fibrillation

The examination cycles included were chosen to maximize the number of eligible individuals to identify relatively rare trajectory groups, to account for changes in ascertainment of clinical factors experienced over the decades, and to allow for 15-year subsequent follow-up data for incident AF.¹⁴ Similarly, to avoid significant differences in measurement scale and precision, examination cycles from the Original and Offspring cohorts are from similar time periods.

Hinge examination participants (n=4810) were excluded if they were ≤45 years of age at the hinge examination (n=229), or developed AF (n=230) during the 16-year trajectory modeling period (i.e. prevalent or interim AF between Original cohort examinations 13–20 and Offspring examinations 2–6). For a trait to be modeled for trajectories, we required that each individual had at least 3 non-missing measurements across the 5 examinations. Hence, the number of individuals used for different traits could be slightly different.

For the time-to-AF analysis participants had to have attended the hinge examination. We used Cox proportional hazards regression models to examine the association of trajectories to 15-year risk of AF. The assumption of proportionality of hazards was met using a Kolmogorov-type supremum test using 1000 simulated patterns. All analyses were stratified by cohort status (generation). We used likelihood ratio tests to evaluate the overall significance of trajectory groups. Multivariable models were adjusted for the following factors at the hinge examination: age, sex, height, weight, current smoking status, systolic and diastolic BP, hypertension treatment, diabetes mellitus, history of myocardial infarction, and history of heart failure.⁴ We performed analyses using SAS version 9.3 (SAS Institute, Cary, NC) and considered 2-sided P<0.05 statistically significant.

Results

Of the 4351 individuals included in the study, the mean age at the first trajectory examination was 50±11 years and at the hinge examination was 65±11 years; and 57% were women (Table 1). During median follow-up of 15 years, 719 participants developed AF (338 women).

Table 1.

Characteristics of the study sample at first and hinge examinations, n=4351

Examination Trajectory phase	First Beginning	Hinge End
Age, years	50±11	65±11
Women	2477 (57)	2477 (57)
Height, cm	166±10	166±10
Weight, kg	72±15	76±17
Current smoker	1205 (30)	597 (14)
Systolic blood pressure, mmHg	125±18	134±21
Diastolic blood pressure, mmHg	78±10	75±10
Antihypertensive treatment	486 (12)	1485 (34)
Diabetes mellitus	90 (2)	403 (9)
Myocardial infarction	33 (1)	180 (4)
Heart failure	9 (0.2)	57 (1)

Values are presented as mean ± standard deviation for continuous traits, or number (%) for binary traits.

Blood Pressure Trajectories

Five trajectory patterns for systolic BP were identified (Figure 2A and Table 2): normotensive (groups 1 and 2), prehypertensive with increasing BP (group 3), hypertensive initially with decreasing BP (group 4), and hypertensive with increasing BP (group 5). Hypertension treatment increased for all five groups during follow-up (Table S1). At the hinge examination, participants in groups 4 and 5 were older than group 1 (Table S2). In age- and sex-adjusted analysis, compared to group 1 there was progressively increasing risk of developing AF in 15-year follow-up from group 3 to group 5 (Table 2). In multivariable-adjusted models, there continued to be statistically significant increased risks of incident AF in groups 4 (hazard ratio [HR], 2.05; 95% confidence interval [CI], 1.24 to 3.37; p=0.005), and 5 (HR, 1.95; 95% CI, 1.08 to 3.49; p<0.03).

Figure 2.

Trajectory groups were identified for systolic (A) and diastolic (B) blood pressure. Trajectory groups identified when hypertension or not (C) and hypertension treatment or not (D) were defined at each examination cycle.

—Observed mean; - - - Trajectory

Table 2.

Fifteen-year risk of atrial fibrillation in blood pressure and antihypertension treatment trajectory groups

Group*	Events/Participants	Age- and sex-adjusted			Multivariable-adjusted†
		HR (95% CI)	P-value	LRT p-value	HR (95% CI)	P-value	LRT p-value
Systolic BP
1	37/599	Reference		<0.0001	Reference		0.02‡
2	227/1775	1.36 (0.96–1.94)	0.09		1.19 (0.82–1.72)	0.37
3	304/1489	1.77 (1.24–2.53)	0.002		1.43 (0.94–2.18)	0.09
4	78/228	2.95 (1.95–4.47)	<0.0001		2.05 (1.24–3.37)	0.005
5	73/259	2.37 (1.56–3.61)	<0.0001		1.95 (1.08–3.49)	0.03
Diastolic BP
1	62/537	Reference		0.003	Reference		0.61‡
2	274/1823	1.19 (0.90–1.57)	0.22		1.10 (0.81–1.50)	0.54
3	178/1044	1.28 (0.96–1.72)	0.09		1.20 (0.81–1.77)	0.37
4	121/537	1.64 (1.20–2.23)	0.002		1.20 (0.85–1.71)	0.30
5	84/409	1.62 (1.16–2.27)	0.005		1.40 (0.88–2.21)	0.16
Hypertension treatment
1	322/2723	Reference		<0.0001	Reference		0.0001‡
2	203/941	1.56 (1.31–1.87)	<0.0001		1.48 (1.08–2.03)	0.02
3	194/687	2.05 (1.71–2.47)	<0.0001		1.97 (1.42–2.73)	<0.0001

BP = blood pressure; HR = hazard ratio; CI = confidence interval; LRT = likelihood ratio test; Hazard ratios are for 15-year risk of atrial fibrillation in each trajectory group compared to the reference group.

^*See Figure 3 for the trajectories patterns of each group.

^†Adjusted for hinge examination age, sex, height, weight, current smoking status, systolic and diastolic blood pressures, hypertension treatment, diabetes, history of myocardial infarction, history of heart failure.

^‡Likelihood ratio test p-value adjusted for age, sex and hinge examination clinical factor was <0.0001 for systolic BP, 0.0005 for diastolic BP and <0.0001 for hypertension treatment.

Diastolic BP trajectory analysis identified five distinct groups (Figure 2B). Group 1 had the lowest diastolic BP and group 5 had the highest diastolic BP throughout the trajectory phase. In age- and sex-adjusted analysis, groups 4 and 5 had a higher risk of incident AF compared to group 1 (Table 2). After multivariable-adjustment, there were no significant differences in the risk of AF between the trajectory groups.

When participants were divided into hypertensive or not at each examination, three trajectory groups were identified (Figure 2C). Trajectory groups in which participants were initially normotensive but then became hypertensive (group 2), or were persistently hypertensive (group 3) were significantly more likely to develop AF compared with participants who were normotensive throughout (Table S3).

Hypertension treatment had three distinct trajectories (Figure 2D): no treatment throughout (group 1), no treatment initially with increasing number of participants being treated (group 2), and majority treated throughout (group 3). Group 3 did not have 100% treated throughout as participants may have fluctuated on and off treatment. In age- and sex-adjusted and multivariable-adjusted analyses, groups 2 and 3 were significantly more likely to develop AF in 15-year follow-up compared to group 1 (Table 2). The systolic and diastolic BP characteristics in each group at each examination cycle are shown in Table S4.

In secondary analyses, we examined trends in pulse pressure in relation to future AF risk. Five trajectory groups were identified: group 1 had the lowest pulse pressures and was stable, with groups 2 to 5 demonstrating progressively increasing pulse pressures (Figure S1). The mean pulse pressure was progressively higher from groups 1 to 5 at the initial examination and the trend persisted throughout the trajectory analysis period except for group 4. In age- and sex-adjusted and multivariable-adjusted models, compared with group 1, groups 2 to 5 demonstrated increasing 15-year risk of incident AF (Table S5).

Other Risk Model Trajectories

Trajectory analysis for body weight did not identify any distinct groups separate from the single time point weight measurements (Figure S2).

Trajectories for smoking identified four groups: group 1 (never smokers), groups 2 and 3 participants started and stopped smoking during the trajectory phase, and group 4 were persistent smokers (Figure 3). In age- and sex-adjusted analyses, groups 2, 3, and 4 did not show significantly increased risk of AF development compared with never-smokers (Table 3). In secondary analyses, per 10-pack year increment in smoking was associated with an increased risk of AF but there was no significant difference after multivariable-adjustment (Table S6).

Figure 3.

Trajectory groups were identified for smoking status defined at each examination cycle.

Table 3.

Fifteen-year risk of atrial fibrillation in smoking and diabetes trajectory groups

Group*	Events/Participants	Age- and sex-adjusted			Multivariable-adjusted†
		HR (95% CI)	P-value	LRT p-value	HR (95% CI)	P-value	LRT p-value
Smoking trajectory groups
1	501/2966	Reference		0.12	Reference		0.16‡
2	48/280	1.24 (0.92–1.68)	0.15		1.13 (0.82–1.54)	0.46
3	67/384	1.24 (0.96–1.60)	0.10		1.23 (0.95–1.60)	0.12
4	96/690	1.20 (0.96–1.50)	0.10		0.77 (0.51–1.17)	0.22
Diabetes trajectory groups
1	601/3870	Reference		<0.0001	Reference		0.56‡
2	66/293	1.52 (1.18–1.97)	0.01		0.92 (0.45–1.86)	0.87
3	39/137	1.95 (1.40–2.71)	<0.0001		1.14 (0.56–2.35)	0.29

HR = hazard ratio; CI = confidence interval; LRT = likelihood ratio test; Hazard ratios are for 15-year risk of atrial fibrillation in each trajectory group using group 1 as reference.

^*See Figure 4 for trajectory patterns for each smoking group, and Figure 5 for trajectory patterns for diabetes groups.

^†Adjusted for hinge examination age, sex, height, weight, current smoking status, systolic and diastolic blood pressures, hypertension treatment, diabetes, history of myocardial infarction, history of heart failure.

^‡Likelihood ratio test p-value adjusted for age and sex: and hinge examination smoking status was 0.09; and hinge examination diabetes status was 0.31.

Trajectory analysis for diabetes mellitus identified three groups (Figure 4): no diabetes throughout (group 1), initially no diabetes but subsequently developed diabetes during trajectory phase (group 2), and diabetes almost throughout (group 3). In age- and sex-adjusted analyses, group 2 (HR, 1.52; 95% CI, 1.18 to 1.97; P=0.01) and group 3 (HR, 1.95, 95% CI, 1.40 to 2.71; P<0.0001) were significantly more likely to develop AF compared with group 1. After multivariable-adjustment, no significant difference was present (Table 3).

Figure 4.

Trajectory groups were identified for presence or absence of diabetes defined at each examination cycle.

CHARGE-AF Risk Score Trajectories

Trajectory modeling of CHARGE-AF risk score identified five trajectory groups (Figure 5). Each trajectory group showed an upward trend in the CHARGE-AF risk score during the trajectory phase. During the trajectory phase there was an increase in the mean systolic BP, antihypertensive treatment, diabetes mellitus, myocardial infarction and heart failure (Table 1) explaining the increase in the CHARGE-AF risk scores. Compared to hinge examination risk score, the use of trajectory modeling did not aid in predicting the 15-year risk of AF.

Figure 5.

Trajectory groups from CHARGE-AF simple risk score calculated at each examination – components of the score are age, race, height, weight, systolic blood pressure, diastolic blood pressure, current smoking, antihypertensive treatment, diabetes, heart failure, and myocardial infarction. For scoring all components were included except for age.

Discussion

In our community-based cohort study we identified trajectories of risk factors used in a validated composite AF risk model, the CHARGE-AF model.^5,6 We found that trajectories of systolic BP and antihypertensive treatment were associated with 15-year risk of incident AF. We identified five different trajectories for systolic BP that demonstrated the long-term patterns of systolic BP in FHS. Groups 4 and 5 were persistently hypertensive during the trajectory phase and had an approximate 2-fold increased 15-year risk of incident AF. Although the finding was not statistically significant, participants with prehypertension throughout (group 3) showed a trend towards increased risk of incident AF. Even though the mean systolic BP at the hinge examination for groups 3 and 4 were similar, the risk of AF was significantly higher for group 4. Additionally, although groups 4 and 5 had very different systolic BP at the hinge examination, their risk of AF was similar and may reflect long-term systolic BP exposure. Our findings suggest the prognostic value of longitudinal systolic BP data. Similarly, prolonged hypertension and antihypertensive treatment were associated with increased risk of AF.

Although not a CHARGE-AF risk score component, we investigated if pulse pressure trajectory groups were predictive of AF, increased pulse pressure has been reported as a risk factor for incident AF.^12,13 As the mean pulse pressure increased in each trajectory group, the risk of incident AF was higher, suggesting that the cumulative exposure of higher pulse pressure was associated with AF development.

Previous studies have demonstrated that individuals with higher systolic BP, pulse pressure, or on hypertension treatment have increased risk of AF.^4,8,13,14 Higher systolic BP and pulse pressure have been associated ventricular remodeling and arterial stiffness,^12,15 and adverse outcomes.¹⁶ In addition, higher systolic BP, pulse pressure, hypertension, and antihypertensive treatment are associated with higher left atrial diameter.^17,18 Both left atrial enlargement and left ventricular structural changes are related to increased risk of future AF.¹⁹ Therefore, the association of increased risk of incident AF in distinct trajectory groups for systolic BP, pulse pressure, hypertension, and hypertension treatment seen in our study may be related to the long term structural remodeling of the atria and ventricles.

In comparison, diastolic BP trajectory groups in our study were not helpful predictors after multivariable adjustment. Although the lack of an association did not support our hypothesis that diastolic BP trajectories may increase risk of incident AF, our finding may be limited given Groups 1–4 had a mean diastolic BP <90mmHg throughout the trajectory phase, and group 5 also had a diastolic BP of <90 mmHg for part of the trajectory phase. It is generally accepted that diastolic BP ≥90mmHg is abnormal²⁰ and should be targeted for treatment, and thus the lower mean diastolic BP in our study may have impaired our ability to find an association. In addition, advancing age is associated with increasing arterial stiffness and decreasing diastolic BP. Hence, relations between diastolic BP and outcomes may be complex in older adults.

Diabetes mellitus⁸ and smoking²¹ have been identified as risk factors for the development of AF. However, unique trajectory groups with longer duration of smoking or diagnosis of diabetes mellitus did not have increased risk of incident AF over participants who were never smokers or free from diabetes mellitus. Our finding is consistent with recent studies demonstrating smoking is a not a major attributable factor for AF,^22,23 and diabetes has a small contribution.²² Alternatively, our modeling of diabetes mellitus as a dichotomous variable may limit our study, and future studies may consider using hemoglobin A1c values over time.

Using the CHARGE-AF risk scores we identified five trajectory groups. As seen in Figure 5, the groups progressed similarly in their CHARGE-AF risk score over the trajectory phase. For most individuals the overall AF risk score steadily increased regardless of the baseline risk level. As a result, the risk score group variable is highly correlated with the one-time risk score. Therefore, trajectory modeling did not help improve identification of higher AF risk participants over using the single time point hinge examination score. The restriction to five trajectory groups may have precluded the identification of small risk groups with dramatic AF risk changes over time.

Our study has several implications. The primary prevention of AF is minimally addressed in current AF guidelines.^24,25 However, as the worldwide prevalence and burden of AF increases,²⁶ it is increasingly important to develop strategies to prevent the onset of AF. Risk factors measured at a single time point have been used to develop AF risk models such as CHARGE-AF,⁴ Atherosclerosis Risk in Communities,² and Women’s Genome Health Study.³ Risk models ignore longitudinal patterns of covariates and their cumulative contributions towards increased AF risk. Our findings suggest that AF prevention strategies may target addressing long-term systolic blood pressure and pulse pressures. Timely intervention may prevent the emergence of long-term atrial and ventricular remodeling^17,19 that promote AF development. However, in our study different trajectory groups of hypertension treatment were also associated with AF and thus further investigation is required to evaluate the best strategy to prevent AF. Primary care providers and other clinicians who often have access to longitudinal clinical data of patients may play a central role in such strategies.

Second, we were able to demonstrate that individuals may have different trends of several important clinical characteristics including systolic and diastolic BP, pulse pressure, smoking, and diabetes. Single time point measurements of risk factors may incompletely predict the risk of AF. Single measures do not identify longitudinal risk factor patterns and may fail to recognize the total exposure of risk factors for adverse outcomes. In fact, in our study the trajectories of individual risk factors often crossed those of other groups, for example, systolic BP in group 4 crossed that in group 5 over time. Measurement of systolic BP at the crossing point would have led to grouping together of individuals in both groups 4 and 5 despite their different systolic BP trajectories.

Thirdly, identifying unique trajectory groups raises several questions that require further investigation. Additional studies may investigate if each risk factor trajectory group has unique omic properties including genomics and proteomics. Further, studies will need to investigate if the burden of AF, success of treatment (such as ablation), and adverse outcomes associated with AF may be related to individuals’ trajectory groups.

Our study has several strengths including the use of a moderate-sized community-based sample, rigorous ascertainment of clinical risk factors, and long-term follow-up. There are also certain limitations of our study to consider. Participants in FHS are mainly middle-aged to older, white adults with a mean age of 50 years at the initial trajectory examination, which may limit generalizability. In our study, we did not map exposures at younger ages, which may identify additional trajectory groups. Individuals who were excluded from our study because they had prevalent or interim AF during the trajectory phase, may have constituted additional trajectory groups, with an even higher risk of AF. Our findings also may not represent other ethnicities and geographical locations both in terms of trajectory groups identified and the association with risk of AF. Additionally, we used PROC TRAJ in SAS to identify trajectory groups. The classification into individual groups was based on the highest probability but may lead to misclassification, especially in participants attending fewer examination cycles. To reduce the effect of the problem, we required that participants attend at least 3 examination cycles. Although our study size is moderate, there may be risk factor trajectories associated with AF that were too rare in our sample to be identified and could be seen in larger datasets. Participants may have undiagnosed AF that was not detected during hospitalizations, outpatient visits, and FHS examination cycles. With closer interaction with healthcare and more frequent electrocardiograms over the past few decades, missed AF cases were likely to be more common earlier in the study. We tried to minimize confounding by adjusting for major AF risk factors⁴ but cannot rule out residual confounding by other risk factors such as chronic kidney disease, specific blood pressure and diabetes medications, or unknown risk factors. In addition, we acknowledge that some covariates may be misclassified at each examination. For example, diabetes might be present before or between examinations but not at the time of a specific examination. We had insufficient power to differentiate between atrial flutter and atrial fibrillation and between specific AF patterns (paroxysmal, persistent, etc.), which may have distinguishing trajectories.

Perspectives.

In a community-based sample, we identified unique trajectory groups for major AF risk factors, which illustrated longitudinal patterns that may be observed in individuals. In addition, we demonstrated that long-term trajectories of systolic BP, pulse pressure, hypertension, and hypertension treatment were associated with a higher risk of AF development. Future research can investigate how individuals on a high-risk trajectory may be identified early to target AF risk prevention. Our research provides indirect support for the importance of a lifecourse approach to blood pressure management to prevent AF.

Novelty and Significance.

What is new?

• Previous studies have identified risk factors for AF by using single time point measurements, which may fail to recognize the total exposure of individuals of covariates. We identified longitudinal risk factor patterns or trajectories for systolic and diastolic blood pressure, pulse pressure, hypertension, antihypertensive treatment, smoking, and diabetes in the Framingham Heart Study.

• Longitudinal patterns of AF risk factors may reflect the cumulative exposure of covariates that promote the development of AF. In our study, trajectories of systolic BP, pulse pressure, hypertension, and antihypertensive treatment were associated with 15-year risk of incident AF.

What is relevant?

• We were able to demonstrate that individuals have different long-term trajectories of systolic and diastolic BP, pulse pressure, hypertension, and hypertension treatment.

• Trajectories of systolic BP, pulse pressure, and hypertension treatment is associated with increased 15-year risk of incident AF. With the increasing burden of AF, AF prevention is important. Our findings suggest that clinicians may identify longitudinal patterns of risk factors to target AF prevention strategies to groups that are high risk.

• Each trajectory group of systolic and diastolic BP, pulse pressure, hypertension, and hypertension treatment may have unique properties that require further investigation.

Summary.

In our community-based cohort, we identified trajectory groups that demonstrate longitudinal patterns seen in individuals for systolic and diastolic BP, pulse pressure, hypertension, hypertension treatment, smoking, and diabetes. In addition, we showed that trajectories of systolic BP, pulse pressure, hypertension, and hypertension treatment were associated with increased 15-year risk of incident AF. Our data provide support for managing long-term blood pressure trends in AF prevention strategies. Further research is required to identify if high-risk trajectories for blood pressure can be identified early for targeted management.

Abbreviations

AF

atrial fibrillation

BP

blood pressure

CHARGE

Cohorts for Heart and Aging Research in Genomic Epidemiology

FHS

Framingham Heart Study