Stroke is the third leading cause of death globally, with a significant socioeconomic impact.^1,2 Ischemic stroke accounts for 87% of these cases, and its etiology is linked to atherothrombotic and cardioembolic events.³ Despite adequate investigation, the etiology is not properly clarified in 25% of these cases, which are referred to as cryptogenic strokes. In such cases, a knowledge gap exists regarding optimal secondary prevention strategies, resulting in an average recurrence rate of 3-6% of new events per year.^4,5 Emerging evidence suggests that a portion of cryptogenic stroke cases are caused by cardioembolic or atheroembolic mechanisms.⁶ In this scenario, atrial cardiomyopathy (AC) may play a causal role in cardioembolic strokes.⁷

AC is characterized by structural and functional endocardial changes driven by several genetic and environmental stimuli, including age and cardiovascular risk factors.^8,9 Traditionally, it is considered a marker or a consequence of other cardiac conditions, as a precursor of atrial fibrillation (AF). However, the paradigm of this chain of events has been evolving in light of evidence that AC may be related to ischemic events, in spite of the presence of AF.⁷ AC disrupts homeostasis, which triggers fibroproliferative signalling pathways and leads to various cardiomyocyte alterations, resulting in endothelial dysfunction, inflammation, hypocontractility, and the release of procoagulant factors, ultimately contributing to thrombogenesis.⁸ While AF enhances this cascade, as it accelerates atrial remodelling and increases blood stasis, its presence is not essential for the occurrence of thromboembolic events.

The 12-lead electrocardiogram (ECG) is a simple, non-invasive, and widely available diagnostic tool that may be useful in the detection of AC through an increase in P-wave duration and an increase of P-wave terminal force in lead V1 (PTFV1) - a product of duration and amplitude of the terminal P-wave negative phase in V1.^10,11 Observational studies in structured databases suggest a link between AC in ECG and adverse outcomes, such as ischemic stroke.^12-14 In this context, we aimed to assess the association between cardiovascular events and death in an electronic Brazilian cohort (Central Illustration).

The CODE-BH database comprised 474,764 ECGs. After excluding those that met exclusion criteria and those that were second or subsequent ECGs from the same patient, the final sample included 245,588 patients (Figure 1).

Figure 1

The baseline characteristics of the study participants are presented in Table 1. Male sex, hypertension, diabetes mellitus, dyslipidemia, and LVH were significantly associated with an increase in both P-wave parameters, whereas smoking was associated only with prolonged P-wave duration, and CAD was not significantly associated with P-wave abnormalities.

Table 1

	Total	P-wave duration		p-value	PTFV1		p-value
	(n=245.588)	< 120 ms (n=180.946)	> 120 ms (n=64.642)		≤ 4.000 µv × ms (n=220.846)	> 4.000 µv × ms (n=24.742)
Age (years), mean (SD)	59.54	58.9 (11.7)	61.2 (11.8)	<0.001	59.6 (11.7)	59.4 (11.7)	0.07
Male sex, n (%)	89 639 (36.5)	61 662 (34.1)	27 867 (43.1)	<0.001	79 460 (36)	10 069 (40.7)	<0.001
Hypertension, n (%)	93 418 (38)	66 018 (36.5)	27 400 (43.1)	<0.001	82 816 (37.5)	10 602 (42.9)	<0.001
Diabetes mellitus, n (%)	39 702 (16.2)	28 363 (15.7)	11 339 (17.5)	<0.001	35 253 (16)	4 449 (18.0)	<0.001
Smoking, n (%)	19 584 (8)	15 144 (8.4)	4 440 (6.9)	<0.001	17 583 (8)	2 001 (8.1)	0.49
Dyslipidemia, n (%)	23 214 (9.5)	16 535 (9.1)	6 679 (10.3)	<0.001	20 639 (9.3)	2 575 (10.4)	<0.001
Left Ventricle Hypertrophy, n (%)	4 172 (17)	2 687 (1.5)	1 485 (2.3)	<0.001	3 508 (1.6)	664 (2.7)	<0.001
Arterial coronary disease, n (%)	1 016 (0.4)	711 (0.4)	305 (0.5)	7,000	899 (0.4)	117 (0.5)	0.13

SD: standard deviation; PTFV1: P-wave terminal force in V1 lead.

The mean follow-up duration for patients in the final cohort was 3.5 years (standard deviation, 2.3). Mortality occurred in 10,980 patients (4.5%), hospitalization in 80,718 patients (32.9%), cardiovascular mortality in 1,753 patients (0.7%), acute coronary syndrome (ACS) in 1,016 patients (0.4%), and incident AF in 316 patients (0.1%). The primary outcome of hospitalization or death due to stroke occurred in 1,860 patients (0.8%). The occurrence rate of primary and secondary outcomes is presented in Table 2. P-wave duration >120 ms and P-terminal force in V1 (PTFV1) >4,000 µV·ms, analyzed as categorical variables, were associated with the primary outcome of death or hospitalization due to stroke in unadjusted analyses, with the association persisting after incremental adjustments for sociodemographic variables, cardiovascular risk factors, and left ventricular hypertrophy (HR 1.24; 95% CI, 1.12–1.36 and HR 1.20; 95% CI, 1.05–1.38, respectively).

Table 2

	P-wave duration		p-value	PTFV1		p-value
Outcome	< 120 ms (n=180.946)	> 120 ms (n=64.642)		≤ 4.000 µv × ms (n=220.846)	> 4.000 µv × ms (n=24.742)
Hospitalization or death due to stroke, n (%)	1216 (0.7)	646 (1.0)	<0.001	1619 (0.7)	241 (1.0)	<0.001
All-cause mortality, n (%)	7703 (4.3)	3287 (5.1)	<0.001	9684 (4.4)	1302 (5.3)	<0.001
Cardiovascular mortality, n (%)	1179 (0.7)	575 (0.9)	<0.001	1522 (0.7)	232 (0.7)	<0.001
Acute Coronary Syndrome, n (%)	711 (0.4)	305 (0.5)	0.007	899 (0.4)	117 (0.5)	0.13
Incident Atrial Fibrillation, n (%)	199 (0.1)	117 (0.2)	<0.001	279 (0.1)	37 (0.1)	0.33

PTFV1: P-wave terminal force in V1 lead.

Analysis of the association between ECG findings and primary and secondary outcomes was performed with a Cox multivariate regression model, and the Proportional Hazards assumption was assessed with the Schoenfeld residuals test, which indicated a significant global violation of the Proportional Hazards (PH) assumption in the P-wave model for hospitalization or death due to stroke (p=0.0008) and all-cause Mortality (p=0.0002). In the stroke model, this violation was attributed to time-varying effects of age (p=0.0008) and arterial hypertension (p=0.0007). For all-cause mortality, non-proportionality was driven by the P-duration category (p=0.0002) and sex (p=0.0032). For cardiovascular mortality, acute coronary syndrome (ACS), and incident AF, the global test showed a marginal trend toward violation (p=0.064), consistently associated with PH violations related to the P-duration category (p=0.023) and arterial hypertension (p=0.040). In the PTFV1 model, a highly consistent pattern of PH assumption violation was observed across all five cardiovascular endpoints. Although the global test for each endpoint showed only a marginal trend toward violation (p = 0.064), two covariates consistently exhibited significant individual violations across all models: P-wave duration category (p = 0.023) and arterial hypertension (p = 0.040). No other covariates, including age, sex, or other comorbidities, showed significant evidence of non-proportionality in the PTFV1 analysis.

Multivariate regression analysis performed as described showed that adjusted prolonged P-wave duration was significantly associated with increased risks of cardiovascular mortality and incident AF, while adjusted elevated PTFV1 was also associated with higher risk of all-cause mortality and cardiovascular mortality (Tables 3 and 4).

Table 3

Outcome	Model 1* HR (CI 95%)	p value (M1)	Model 2† HR (CI 95%)	p value (M2)	Model 3‡ HR (CI 95%)	p-value (M3)	Model 4§ HR (CI 95%)	p-value (M4)
Hospitalization or death due to stroke	1.45 (1.32 – 1.6)	<0.001	1.25 (1.14 – 1.39)	<0.001	1.24 (1.13 – 1.37)	<0.001	1.24 (1.12 – 1.36)	<0.001
All-cause mortality	1.17 (1.12 – 1.22)	<0.001	0.98 (0.94 – 1.02)	0.38	0.97 (0.95 – 1.03)	0.51	0.98 (0.94 – 1.02)	0.34
Cardiovascular mortality	1.33 (1.21 – 1.47)	<0.001	1.12 (1.01 – 1.24)	0.02	1.12 (1.02 – 1.24)	0.02	1.11 ( 1.01 – 1.23)	0.04
Acute Coronary Syndrome	1.2 (1.05 – 1.37)	0.008	1.06 (0.93 – 1.21)	0.4	1.04 (0.91 – 1.19)	0.56	1.04 (0.91 – 1.19)	0.62
Incident Atrial Fibrillation	1.6 (1.28 – 2.02)	<0.001	1.35 (1.08 – 1.70)	0.01	1.35 (1.08 – 1.70)	0.01	1.34 (1.06 – 1.68)	0.01

HR: hazard ratio; CI: confidence interval; *unadjusted; ^† adjusted for sex and age; ^‡ model 2 + adjusted for hypertension, diabetes mellitus, smoking, and dyslipidemia; ^§ model 3 + adjusted for left ventricular hypertrophy.

Table 4

Outcome	Model 1* HR (CI 95%)	p-value (M1)	Model 2† HR (CI 95%)	p-value (M2)	Model 3‡ HR (CI 95%)	p-value (M3)	Model 4§ HR (CI 95%)	p-value (M4)
Hospitalization or death due to stroke	1.24 (1.09 – 1.42)	0.01	1.23 (1.08 – 1.41)	0.002	1.22 (1.06 – 1.4)	0.004	1.2 (1.05 – 1.38)	0.01
All-cause mortality	1.13 (1.07 – 1.2)	<0.001	1.12 (1.06 – 1.19)	<0.001	1.12 (1.06 –1.19)	<0.001	1.11 (1.05 – 1.17)	0.001
Cardiovascular mortality	1.28 (1.18 – 1.47)	<0.001	1.28 (1.11 – 1.49)	<0.001	1.27 (1.11 –1.46)	0.001	1.25 (1.09 – 1.43)	0.002
Acute Coronary Syndrome	1.16 (0.96 – 1.41)	0,13	1.12 (0.92 – 1.36)	0.25	1.08 (0.89 –1.31)	0.45	1.07 (0.88 – 1.29)	0.52
Incident Atrial Fibrillation	1.1 (0.78 – 1.55)	0.58	1.1 (0.78 – 1.55)	0.58	1.1 (0.78 –1.55)	0.58	1.08 (0.76 – 1.52)	0.67

HR: hazard ratio; CI: confidence interval; *unadjusted; ^†adjusted for sex and age; ^‡model 2 + adjusted for hypertension, diabetes mellitus, smoking, and dyslipidemia; ^§model 3 + adjusted for left ventricular hypertrophy.

Survival analysis curves demonstrated a significant divergence in event rates among patients stratified by the presence or absence of AC, with a more pronounced divergence observed from the fourth year of follow-up onward (Figures 2, 3, and 4).

Figure 2

Figure 3

Figure 4

In this cohort study, we identified a significant, independent association between electrocardiographic markers of AC (increase in P-wave duration and PTFV1) and an increased risk of stroke-related death and hospitalization. P-wave duration>120 ms was also associated with cardiovascular mortality and incident AF, while elevated PTFV1 was linked to all-cause mortality and cardiovascular mortality. When compared to cohort studies with similar designs, the present research features the largest sample and is the first to be conducted in a Brazilian population. We expected that a shorter mean follow-up time when compared with other cohorts (3.5 years vs CHS: 12,9 years; MESA: 8,5 years; ARIC: 22 years)^13,14,20 would reduce the chance of events. This limitation, however, has been overcome by the larger sample size.

Our results corroborate prior evidence linking AC to adverse outcomes.^13,21,22 In this context, ischemic cerebrovascular events stand out, having shown a significant association with P-wave alterations, which translates to underlying atrial myocardial pathology. These findings validate the hypothesis that AC may be associated with thromboembolic events through a direct, independent pathway, not only mediated by the presence of AF or other cardiomyopathies described as a risk factor for thrombogenesis. Furthermore, P-wave alterations were also associated with outcomes other than thromboembolic, hence suggesting that AC may be a marker of cardiovascular risk and mortality. AC remains incompletely understood, highlighting the need for further research due to its potential to impact clinical practice as a predictor of cardiovascular risk.

AF is a part of the AC spectrum and plays a role as a mediator or enhancer of thromboembolic events in a dose-dependent fashion. Besides contributing to blood stasis, AC may be related to activation and amplification of already existing atrial endothelial dysfunction, hence increasing inflammation, platelet activation, and the coagulation cascade, favoring prothrombotic states.²³ The question of whether AC without AF is a sufficient cause of clinically relevant thrombogenesis remains unsolved. However, available evidence and our findings suggest this possibility, which may pave the way for a shift in the approach to preventing and investigating ischemic neurological events, and towards broader AC detection strategies via ECG and other biomarkers, beyond AF alone.^7,24

Several methods have been studied to improve the detection of AC. Tridimensional echocardiography with atrial strain measurements is an evolving technique, able to detect both atrial fibrosis and systolic dysfunction through analysis of segmental longitudinal shortening and stretching of the atrial myocardium.²⁵ Magnetic cardiac resonance may provide information on atrial morphology, left atrial (LA) function through phasic variations in volume, contraction, and segmental shortening, and LA emptying fraction,²⁰ as well as quantitative detection of atrial fibrosis through late-enhancement with gadolinium.²⁶ Computerized cardiac tomography can perform tridimensional analysis of atrial morphology and LA appendix,²⁶ and specific LA morphologic types have been associated with increased risk of ischemic stroke.²⁴ The use of such methods has shown association with relevant outcomes, e.g., AF, mortality, cardiovascular events, and ischemic stroke.^20,24,26,27 However, ECG remains the initial, widely available tool for cardiomyopathy investigation, and the recognition of its electrocardiographic patterns should trigger an extensive diagnostic approach, mainly in cryptogenic ischemic stroke investigation.

The use of artificial intelligence (AI) in the detection and prediction of AF has already been described.^22,28,29 It is reasonable to speculate that P-wave alterations may explain the results obtained with these models for the prediction of AF. The use of AI in ECG interpretation seems to be a promising tool to improve prediction models of thromboembolic events in patients with cryptogenic stroke. AC assessment methods still require further exploration in order to allow thrombotic risk prediction and clinical applicability.

This study, to the best of our knowledge, is the first to assess AC using the ECG in a large cohort of Brazilian patients. Our findings align with and corroborate existing evidence and enable the generation of new hypotheses regarding the association between AC and adverse outcomes, as well as the development of new risk prediction tools. However, our results are not yet sufficient to support new anticoagulation strategies, as they are subject to several limitations that must be addressed. First, this work is a retrospective cohort study relying on self-reported clinical data, which introduces the possibility of confounding bias. Specific exposures and risk factors may not have been captured in the questionnaire, which could affect the comprehensiveness of the data. Besides, our sample was drawn from a primary care setting, and patients who were not in sinus rhythm were excluded, which may limit the generalizability of our findings and constrain their extrapolation to populations with a presumably higher cardiovascular risk. Incident AF may also have been underestimated due to reliance on single ECG assessments, potentially hindering the detection of paroxysmal AF.

Outcome data were obtained from public health system information systems in Belo Horizonte. This aspect of the study limits our ability to separate strokes caused by atherothrombotic from cardioembolic mechanisms, since AC may play a significant role in cardioembolic strokes. This differentiation was hindered by the absence of information on the underlying mechanisms of strokes in the database from which this data was extracted, as well as the inability to adjust for extracoronary atherosclerosis, which may introduce misclassification bias. Also, this data is subject to errors in documentation, while records of hospital admissions are limited to public hospitals. Furthermore, outcomes that occurred outside Belo Horizonte, such as hospital admissions and deaths, were not included in these records, potentially underestimating the total number of events. Additionally, our data identification process relies on high-probability linkage (over 95% of matches) that links patients across databases by name, sex, and age. However, this strategy introduces the risk of pairs’ loss due to misspellings or incorrect pairings based on homonyms among patients with the same birth date. Furthermore, the violation of the proportional hazards assumption for a few covariates, as described in the Methods section, may limit the interpretability of the hazard ratios reported in our analysis. Finally, the University of Glasgow software was used for all electrocardiographic analyses. Although it is an automated tool with the potential for inaccuracies compared to manual measurements by cardiologists, its ease of use and reproducibility across various clinical settings facilitate its incorporation into clinical practice.

Electrocardiographic markers of AC (P-wave duration >120 ms and PTFV1>4000 µV × ms) adjusted to demographic data and cardiovascular risk factors are associated with increased risk of stroke-related death and hospitalization. Prolonged P-wave duration is also linked to cardiovascular mortality and incident AF, while elevated PTFV1 was also associated with all-cause mortality and cardiovascular mortality. To our knowledge, this study included the largest sample size among studies with a similar design performed to date, which strengthens the hypothesis of an independent association between AC and adverse clinical outcomes. Further research is needed to clarify the understanding of AC as a pathological entity and to identify biomarkers able to predict thromboembolic events. Such achievements could enhance risk stratification tools and anticoagulation strategies for high-risk individuals, hence reducing mortality and the burden of disease.