This study was designed as a prospective cohort study conducted at the Arrhythmia Department, Cho Ray Hospital, from December 2024 to June 2026. We followed a group of patients with non-atrial fibrillation (non-AF) arrhythmias to evaluate the incidence of new-onset AF and associated factors over time, with a specific focus on the relationship with baseline NT-proBNP levels.

The study population consisted of patients diagnosed with cardiac arrhythmia and treated at the department. Specific inclusion criteria were:

Patients were excluded if they had conditions that could significantly influence the study results or complicate follow-up, including:

The sample size was calculated based on the primary objective of predicting the incidence of AF, using survival analysis. The formula for the minimum sample size in a survival study, based on the hazard ratio (HR), was applied. Based on a study by Staerk et al. (2020) [13], we used an HR for NT-proBNP of 1.73. The prevalence of AF in patients with arrhythmia was referenced from a study by Lindberg et al. [14], which was 36.8%.

Substituting these values into the formula, we calculated an initial sample size of 160 patients. Accounting for an estimated 30% patient dropout rate during follow-up, we recruited a minimum of 207 patients.

At the time of hospital admission, baseline clinical information was collected from all patients, including:

Data were analyzed using R Studio software.

The study initially collected data from 285 patients. After excluding 41 patients with a pre-existing history of atrial fibrillation (AF) and 12 patients with severe renal impairment (eGFR < 30 ml/min/m²), a total of 232 patients without baseline AF were eligible for participation. The median follow-up duration was 12 months, with an interquartile range (IQR) of 6.75 to 13 months.

The study group comprised 124 males (53.4%) and 108 females (46.6%). The mean age for the entire cohort was 63.7 ± 14.0 years, with a median of 67. Notably, females had a significantly higher mean age than males (67.5 ± 14.4 years vs. 60.4 ± 16.7 years). Other anthropometric indices reflected the biological differences between the sexes: males were taller (1.63 ± 0.05 m) and heavier (61.4 ± 10.2 kg) than females (1.55 ± 0.04 m and 53.3 ± 9.19 kg, respectively). The mean BMI for the cohort was 22.5 ± 3.34 kg/m², which falls within the normal range. The mean blood pressure was also stable, with a systolic pressure of 129 ± 19.6 mmHg and a diastolic pressure of 75.1 ± 10.2 mmHg. However, the mean blood pressure tended to be slightly higher in females, which could be related to their higher mean age. (Table 1)

Regarding medical history, the most prevalent cardiovascular risk factors were hypertension (59.0%), dyslipidemia (40.5%), and diabetes mellitus (22.0%). Heart failure (18.0%), coronary artery disease (13.4%), and chronic kidney disease (10.3%) were also common. The presenting symptoms for this patient cohort were highly varied, with syncope (59.9%) and dizziness (55.6%) being the most common, each affecting more than half of the patients. (Fig. 1, Table 2)

The distribution of NT-proBNP levels in the study cohort was heterogeneous, with most patients having low concentrations while a smaller subgroup had very high levels. The median NT-proBNP value was 34.5 pmol/L (IQR: 7.2–157). (Fig. 2)

Most other laboratory values were within normal limits. The median Creatinine was 0.87 mg/dL and the median eGFR was 83.9 mL/min/1.73m², reflecting an acceptably preserved renal function. Blood lipid levels (Cholesterol, LDL, HDL, Triglyceride) were also within the average range. On echocardiography, the median left atrial (LA) size was 31 mm and the median ejection fraction (EF) was 65%, indicating that the overall cardiac function of the study cohort was preserved. (Table 3)

After a median follow-up period of 12 months, 16 cases (6.9%) of new-onset atrial fibrillation developed among the 232 baseline patients.

The Kaplan-Meier curve demonstrated a gradual increase in the cumulative incidence of AF over time, reaching nearly 10% after 12 months. The 95% confidence intervals were narrow initially but widened progressively, indicating that the precision of the estimate decreased as the number of remaining patients diminished. The patient count declined significantly over time: from 232 at baseline to just 19 at 17 months, due to events or loss to follow-up. (Fig. 3)

Initial analysis showed that baseline NT-proBNP levels were slightly higher in the group that developed AF (median 36.0 pmol/L, IQR: 7.47–162) compared to the group that did not (median 34.7 pmol/L, IQR: 7.22–157). However, this difference was not statistically significant (p = 0.327) by the Wilcoxon rank sum test. An analysis of Kaplan-Meier curves across NT-proBNP quartiles also revealed no significant difference between the groups (p = 0.3). (Fig. 4)

A multivariate Cox proportional hazards regression model was constructed to identify independent predictors for the incidence of AF. The results revealed that only three factors were statistically associated with AF risk:

In contrast, other variables in the model, such as NT-proBNP, age, sex, hypertension, heart failure, eGFR, and other indices, were not statistically significant (p > 0.05) in predicting AF. Although a history of myocardial infarction had a high HR (5.1160), its p-value was 0.0942, failing to reach statistical significance. (Table 4)

This study was conducted with the initial hypothesis that high plasma NT-proBNP levels at hospital admission could be an independent predictor for the onset of atrial fibrillation (AF) in patients with non-AF arrhythmias. The primary objective was to evaluate the role of NT-proBNP in predicting new-onset AF. After excluding cases with baseline AF and severe renal impairment, 232 non-AF patients were included in the analysis and followed for a median duration of 12 months.

Although NT-proBNP levels tended to be higher in the group that developed AF, the admission NT-proBNP concentration did not demonstrate a statistically significant, independent predictive role for new-onset AF in the multivariable Cox model (HR = 0.9995; p = 0.717). Instead, left atrial (LA) size, a history of diabetes mellitus, and a history of stroke were the three strongest statistically significant factors for predicting new-onset AF. Specifically, a history of stroke increased the risk of AF by more than 10 times, while diabetes increased it by nearly 6 times, and each millimeter increase in LA size raised the risk by 27%.

Our findings differ significantly from many previous studies that have demonstrated a strong predictive role of NT-proBNP for new-onset AF [1, 6, 15–19]. For instance, studies in the general population or in heart failure patients often found an independent association between NT-proBNP and AF risk (Schnabel et al., 2010; Li et al., 2018) [20, 21]. This discrepancy can be explained by several factors.

First, our study population consisted of patients with pre-existing non-AF arrhythmias who were admitted with a diverse spectrum of symptoms, with syncope (59.9%) and dizziness (55.6%) being the most common, rather than typical palpitations. This may suggest a more complex underlying pathophysiology, where the baseline arrhythmias were already severe enough to cause hemodynamic disturbances, and NT-proBNP may be reflecting myocardial wall stress from these arrhythmias rather than directly predicting the conversion to AF.

Second, the median follow-up period of our study was 12 months, whereas some other studies with longer follow-up durations may have been better able to detect a clearer association.

Nevertheless, our results reinforce the role of left atrial (LA) size as a strong and consistent risk factor for AF, which has been widely established in the literature [5, 22, 23]. Left atrial enlargement is a marker of mechanical and electrical atrial remodeling, which is a prerequisite for the development of AF [22]. Similarly, the association between diabetes and AF risk is well-documented, through mechanisms such as systemic inflammation, myocardial fibrosis, and microvascular disease [24–26]. The finding of a history of stroke as a very strong predictor of AF (HR > 10) is particularly noteworthy. This could suggest that patients with a prior stroke may have had undiagnosed paroxysmal AF or other severe underlying risk factors that make them more susceptible to AF.

The lack of statistical significance for NT-proBNP in the multivariable model may be due to:

Conversely, the prominence of left atrial size is physiologically sound. LA enlargement is a consequence of chronic elevated atrial filling pressure and a direct marker of atrial remodeling, which increases the risk of developing re-entry circuits and ectopic foci, leading to AF [28, 29]. Similarly, diabetes induces diabetic cardiomyopathy, including atrial fibrosis and endothelial dysfunction, creating a favorable substrate for AF [24, 30–33]. A history of stroke, with a high HR, may reflect a patient group with pre-existing cerebrovascular damage often accompanied by severe underlying cardiovascular conditions (including undiagnosed paroxysmal AF) or a high thrombotic risk.

The results of this study provide a nuanced understanding of AF predictors in a specific and clinically relevant patient population. While NT-proBNP was not an independent predictor in this cohort, the findings underscore that a careful evaluation of classical clinical factors and left atrial size remains central to AF risk stratification.

For patients with non-AF arrhythmias who are admitted to the hospital, particularly with symptoms like syncope and dizziness, meticulous screening for a history of diabetes, stroke, and an echocardiographic assessment of LA size are critically important. Patients with these risk factors should be monitored more aggressively for AF, which may include extended Holter ECG or more advanced cardiac rhythm monitoring devices. This has significant implications for personalizing risk management and stroke prevention strategies related to AF.

This study possesses several notable strengths that contribute to its scientific value. First, the prospective cohort design allows us to establish a temporal relationship between predictors and outcomes, significantly minimizing recall bias compared to retrospective designs. Second, the focus on a specific study population—patients with pre-existing non-AF arrhythmias—fills a crucial knowledge gap in a field often overlooked by larger studies. Third, comprehensive data collection, including clinical information, standardized laboratory tests (including NT-proBNP), and detailed echocardiographic indices, was performed. Finally, the median follow-up of 12 months is a reasonable timeframe to observe new-onset AF events, allowing for a dynamic assessment of the disease over time.

However, this research also has several limitations that should be acknowledged. The most significant drawback is the small number of events; despite a total of 232 patients, only 16 new-onset AF cases were recorded. This low event count may reduce the statistical power of the analysis, making it difficult to detect statistically significant associations for variables with small or moderate effects, especially in a complex multivariable analysis, which may be the primary reason why NT-proBNP did not achieve statistical significance in our model. Another limitation is the intermittent follow-up and a notable rate of loss to follow-up; the substantial decrease in the number of study participants over time (down to only 19 patients at 17 months) may introduce bias into the survival estimates and reduce the reliability of the results in the later stages of the study. Furthermore, the detection of AF relied solely on ECG and Holter ECG, which may have missed short or asymptomatic episodes of paroxysmal AF. The use of more prolonged monitoring devices like implantable loop recorders could improve detection but was not feasible in the context of this study. Finally, this is a single-center study, which may limit the generalizability of the results to more diverse patient populations or other medical centers with different clinical characteristics and management protocols.

Our study concludes that baseline NT-proBNP is not an independent predictor for new-onset AF in patients with non-AF arrhythmias. However, traditional clinical factors play a highly significant role. Specifically, left atrial size, a history of diabetes mellitus, and a history of stroke were identified as strong, independent predictors. These findings underscore the necessity of a comprehensive assessment of classical risk factors for AF screening and management, particularly in high-risk patients.