Contributions of Biological Aging to Longitudinal Incidence and Dynamic Progression of Atrial Fibrillation: A Prospective Cohort Study

This study utilized data from the UK Biobank, which is a large-scale prospective cohort study comprising around 500,000 participants aged 40–69 years, who were recruited from 22 UK assessment centers between 2006 and 2010. Ethical approval was granted by the North West Multicenter Research Ethics Committee (REC reference: 21/NW/0157), and written informed consent was obtained from all participants. This study’s project approval number is 170605.

Of the initial 502,175 participants, we excluded those with a history of AF or AF-related complications at baseline (n = 124,198). We also excluded participants whose AF-related complications predated AF onset (n = 24,388). For analyses involving KDM-BA and PhenoAge, we excluded participants with missing data on these measures (n = 353,589) leaving 260,198 participants for analysis. For telomere length analyses, participants with missing telomere data (n = 13,986) were excluded, resulting in 339,603 participants (Fig. 1).

Biological ageing was assessed using two approaches: clinical traits-based biological age (KDM-BA and PhenoAge) [9, 10], and molecular-level telomere length [11]. KDM-BA was derived from forced expiratory volume in one second, systolic blood pressure, and seven blood biomarkers: albumin, alkaline phosphatase, blood urea nitrogen, creatinine, C-reactive protein, glycated haemoglobin, and total cholesterol. PhenoAge was calculated using nine blood biomarkers, including albumin, alkaline phosphatase, creatinine, C-reactive protein, glucose, mean cell volume, red cell distribution width, white blood cell count, and lymphocyte proportion. To quantify the deviation between biological and chronological age, we regressed KDM-BA and PhenoAge on chronological age using natural splines with three degrees of freedom. The algorithms and R codes for these measures are available in the “BioAge” R package and in prior publications.

Telomere length was measured using a quantitative polymerase chain reaction (qPCR) assay to quantify DNA extracted from leukocytes at baseline, and was expressed as the telomere-to-single copy gene ratio (T/S ratio). The measurements were adjusted for technical parameters, log-transformed, and Z-standardized. Detailed quality control procedures have been described previously [11].

Participants without a history of AF or AF-related complications at baseline were followed from recruitment until they were lost to follow-up, died, or October 31, 2022, whichever occurred first. Outcomes of interest included incident AF, AF-related complications (HF, MI, cerebral infarction, dementia, and other arterial embolic diseases), and all-cause mortality. These outcomes were identified through linkage with death registries, primary care records, and hospital inpatient data, using diagnostic codes from the International Classification of Diseases, 9th (ICD-9) and 10th (ICD-10) revisions. Detailed diagnostic codes are provided in Table 1. The validation steps of outcome events were that atrial fibrillation occurred first, followed by complications of atrial fibrillation.

Systemic inflammation was assessed using the neutrophil-to-lymphocyte ratio (NLR) and the systemic inflammation response index (SIRI). The NLR was calculated by dividing the neutrophil count by the lymphocyte count, while the SIRI was calculated by multiplying the neutrophil count by the monocyte count and dividing this sum by the lymphocyte count. Both measures were log-transformed to address skewed distributions.

Baseline covariates were collected via questionnaires and interviews. Demographic factors included age, sex, ethnicity, education level, and the Townsend deprivation index. Lifestyle factors included body mass index (BMI), categorized as underweight (BMI <18.5 kg/m²), normal weight (BMI 18.5–24.9 kg/m²), overweight (BMI 25–29.9 kg/m²), or obese (BMI ≥30 kg/m²); dietary pattern (healthy or unhealthy, based on DASH diet score); smoking status (never, previous, or current); and alcohol consumption (never, previous, or current).

Missing covariates were imputed using multiple imputation by chained equations, with all covariates in the model. Baseline characteristics were summarized as means (standard deviations) or frequencies (percentages) for continuous and categorical variables, respectively, across all participants and subgroups with AF or AF-related complications.

Cox proportional hazards models were employed to estimate the association between biological ageing markers and the incidence of AF, AF-related complications, and mortality. To assess the progression from a healthy state to AF and subsequent complications and death, MSMs were employed using the “mstate” R package. Five transitions were modeled: (1) baseline to AF, (2) baseline to death, (3) AF to complications, (4) AF to death, and (5) complications to death. For participants entering multiple states on the same date, the entry date of the prior state was set as 0.5 days earlier. A more detailed MSM was constructed to examine specific complications (HF, MI, stroke, dementia) involving 11 transitions. In our multi-state models, death was included as an absorbing state with dedicated transitions (e.g., baseline → death), so competing risks of death for intermediate events such as AF or complications were explicitly accounted for. Each transition had an independent baseline hazard, estimated nonparametrically, under a Markov assumption conditional on the current state and covariates. The proportional hazards assumption was assessed for each transition using Schoenfeld residuals, with no major violations detected (Supplementary Table 1,2,3). Model 1 was adjusted for age and sex, while Model 2 was further adjusted for ethnicity, the Townsend deprivation index, education level, BMI, dietary pattern, smoking status, and alcohol consumption.

Counterfactual mediation analysis was performed to investigate the mediating role of systemic inflammation (as measured by NLR and SIRI) in the relationship between biological ageing markers and AF-related outcomes. This method estimated the direct, indirect, and total effects, as well as the proportion mediated. The “CMAverse” R package was used to calculate hazard ratios (HRs) and 95% confidence intervals (CIs) for counterfactual effects with 1000 bootstrapped samples.

The Sensitivity analyses included: (a) excluding participants lacking covariates; (b) excluding participants with events within the first two years of follow-up; (c) excluding those with a history of cancer at baseline; (d) testing alternative age-scaling methods for biological age; (e) using a semi-Markov specification (time since entry into the current state as the time scale); (f) setting a prior state’s entry time to1–2 days for same-day events. ±

All statistical analyses were performed using R software (version 4.4.2, R Foundation for Statistical Computing, Vienna, Austria). Statistical significance was defined as p < 0.05.

Among the 260,198 participants analyzed for KDM-BA and PhenoAge, the mean age was 55.47 years, with 57.08% being women (Table 2). Participants were predominantly white (94.51%), with 35.15% having received a college/university education. The mean BMI was 26.86 kg/m², with most participants being never smokers (57.69%) and current alcohol drinkers (93.11%). Similar characteristics were observed in the telomere length cohort (n = 339,603) (Supplementary Table 4). Notably, participants who developed AF or its complications were more likely to be male and less educated, and to have a higher BMI.

During the follow-up period, 9.51% of the KDM-BA/PhenoAge cohort developed AF, with 17.59% of these cases progressing to complications. The most common complication was HF (9.47%), followed by MI (4.21%), dementia (1.47%), and stroke (0.97%) (Figs. 2,3). Of those with complications, 8.20% subsequently died. Mortality patterns revealed 7.26% deaths from baseline, 16.91% deaths after AF diagnosis, and 8.20% deaths after complications. Similar trends were seen in the telomere length cohort, where 9.67% developed AF, 17.85% of AF cases progressed to complications, and 10.83% died after complications.

The three biological ageing markers showed significant associations with disease progression (Table 3). Compared with the lowest quartile (Q1), the highest quartile (Q4) of both KDM-BA and PhenoAge was associated with increased risks of AF (HR: 1.16, 95% CI: 1.10–1.21; HR: 1.34, 95% CI: 1.29–1.40), complications (HR: 1.25, 95% CI: 1.11–1.39; HR: 1.25, 95% CI: 1.14–1.37), and death (HR: 1.71, 95% CI: 1.62–1.82; HR: 1.87, 95% CI: 1.78–1.97). In contrast, the Q4 of telomere length showed protective effects against AF (HR: 0.76, 95% CI: 0.74–0.78), complications (HR: 0.55, 95% CI: 0.51–0.59), and death (HR: 0.90, 95% CI: 0.87–0.92).

In the multistate models, all three biological ageing markers showed distinct patterns in different disease transitions (Table 4). For KDM-BA, the Q4 was significantly associated with increased risks of transitions from baseline to AF (HR: 1.09, 95% CI: 1.04–1.14), baseline to death (HR: 1.10, 95% CI: 1.04–1.16), and AF to complication (HR: 1.75, 95% CI: 1.58–1.94). Similar patterns were observed for PhenoAge, with the Q4 showing elevated risks in transitions from baseline to AF (HR: 1.30, 95% CI: 1.25–1.35), baseline to death (HR: 1.11, 95% CI: 1.06–1.16), and AF to complications (HR: 1.52, 95% CI: 1.37–1.68). In contrast, the Q4 of telomere length showed protective effects against transitions from baseline to AF (HR: 0.83, 95% CI: 0.80–0.86), baseline to death (HR: 0.91, 95% CI: 0.88–0.94), and AF to complication (HR: 0.78, 95% CI: 0.72–0.84).

Further analysis of specific complications revealed distinct patterns of biological ageing markers in different transitions (Table 5, Supplementary Table 5). KDM-BA and PhenoAge showed significant associations with increased risks of transitions from baseline to AF (HR: 1.09, 95% CI: 1.04–1.14 for KDM-BA Q4; HR: 1.30, 95% CI: 1.25–1.35 for PhenoAge Q4), baseline to death (HR: 1.10, 95% CI: 1.04–1.16 for KDM-BA Q4; HR: 1.11, 95% CI: 1.06–1.16 for PhenoAge Q4), and AF to HF (HR: 1.89, 95% CI: 1.65–2.17 for KDM-BA Q4; HR: 1.60, 95% CI: 1.40–1.83 for PhenoAge Q4). Additionally, PhenoAge was associated with increased risks of transitions from AF to other arterial embolism diseases (HR: 1.49, 95% CI: 1.25–1.79 for Q4), and MI to death (HR: 2.53, 95% CI: 1.44–4.45 for Q4). In contrast, telomere length demonstrated protective effects against the progression from baseline to AF (HR: 0.83, 95% CI: 0.80–0.86 for Q4), baseline to death (HR: 0.91, 95% CI: 0.88–0.94 for Q4), and AF to specific complications, especially in reducing the risk of transitions to MI (HR: 0.57, 95% CI: 0.49–0.67), dementia (HR: 0.62, 95% CI: 0.47–0.81), and other arterial embolism diseases (HR: 0.84, 95% CI: 0.73–0.96).

Mediation analyses were performed to explore whether systemic inflammation mediated the associations between biological ageing markers and AF-related outcomes (Fig. 4, Supplementary Table 6). For both KDM-BA and PhenoAge, SIRI showed stronger mediating effects compared to NLR across all transitions. Specifically, SIRI mediated 29.95% and 28.27% of the total effects of KDM-BA and PhenoAge on AF incidence, respectively. For progression to complications, SIRI mediated 16.25% of the KDM-BA effect and 32.92% of the PhenoAge effect. The mediating role of SIRI was also evident in mortality transitions, accounting for 21.27% of the total effects for both KDM-BA and PhenoAge.

Sensitivity analyses confirmed the robustness of the findings. Firstly, the analysis results after excluding the missing covariates were consistent (Supplementary Table 7), and the basic characteristics of the included and excluded populations seemed to have no difference (Supplementary Tables 8,9). Excluding events within the first two years of follow-up, participants with baseline cancer history, or employing alternative scales for biological ageing markers did not alter the primary results (Supplementary Tables 10–12). MSM results were re-estimated using a semi-Markov specification (time since entry into the current state as the time scale), yielding consistent estimates. Additionally, the analysis results were consistent using a semi-Markov specification with time-since-entry, and setting a prior state’s entry time to “±1–2 days” for same-day events (Supplementary Tables 13–17).

However, several limitations should be acknowledged. First, as an observational study, causality cannot be established between biological ageing and AF progression. Second, baseline measurements of biological ageing markers and inflammatory indicators do not capture their dynamic changes over time. Third, although the proportion of systemic inflammatory mediators is as high as 30%, the interpretation of the results requires caution, and other mechanisms warrant further investigation. Fourth, the predominantly White and relatively healthier UK Biobank may lead to a bias in the health volunteers, thereby limiting the generalizability of the findings to other ethnic groups or less healthy populations. Fifth, we primarily identified new-onset AF and its complications through hospitalization and cause-of-death data. AF cases diagnosed in the community may have been missed. Such a delayed/latent AF diagnosis may introduce bias into state transitions and transition timing. Finally, despite comprehensive covariate adjustments, residual confounding remains plausible for unmeasured factors such as genetic predisposition, medication use (particularly anticoagulants, antihypertensives), and detailed cardiovascular risk factors. Future research should address these limitations by using repeated measurements, expanding to diverse and less healthy populations, and employing comprehensive data sources to capture all AF cases, thus enhancing robustness and generalizability.

The data are available from the UK Biobank, but there are restrictions on their availability. Researchers who wish to access the UK Biobank database will need to apply for access through the following link: https://www.ukbiobank.ac.uk/enable-your-research/↗.