Contemporary trends in incident ischaemic stroke, intracranial haemorrhage, and mortality in individuals with atrial fibrillation

Abstract

Background

The prognosis for atrial fibrillation (AF) patients is based on data that is decades old. Given evolving standards of clinical practice, we sought to evaluate temporal trends in clinically important outcomes among patients with AF.

Methods and results

California's Department of Health Care Access and Information databases were used to identify adults aged ≥18 years with AF receiving hospital-based care in California. We compared three time-periods: 2005–2009, 2010–2014, and 2015–2019. International Classification of Diseases codes were used to identify chronic diseases and acute events. The outcomes were incident ischaemic stroke, intracranial haemorrhage, and overall mortality. We included 2 009 832 patients with AF (52.7% males, 70.7% Whites, and mean age of 75.0 years), divided in three cohorts: 2005–2009 (n = 738 954), 2010–2014 (n = 609 447), and 2015–2019 (n = 661 431). Each outcome became substantially less common with time: compared with 2005–2009, AF patients diagnosed in 2015–2019 experienced a 34% (adjusted hazard ratio [HR] 0.66, 95% confidence interval 0.64–0.69), 22% (HR 0.78, 0.75–0.82), and 24% (HR 0.76, 0.75–0.77) reduction in risk of incident ischaemic stroke, intracranial haemorrhage, and mortality, respectively. Between 2005–2009 and 2015–2019, patients aged ≥65 years experienced more reductions in each outcome compared with younger patients (P < 0.001 for all), and declines in each outcome were significantly lower for Hispanics and Blacks compared with white patients.

Conclusion

The risks of stroke, intracranial haemorrhage, and death have significantly declined among AF patients, although differences in the magnitude of improvement of these outcomes by demographic groups were observed. Commonly described estimates of the prognosis for AF patients should be updated to reflect contemporary care.

Graphical Abstract

Graphical Abstract.

Introduction

Atrial fibrillation (AF) is an important contributor to the burden of disease in the United States and globally.¹ Atrial fibrillation accounts for up to a third of all ischaemic strokes,² and AF-related strokes are associated with higher morbidity and mortality compared with strokes from non-AF causes.³ Furthermore, AF is associated with an increased risk of mortality.⁴

The magnitude of adverse health effects of AF was determined mostly by studies that are now decades old and no longer reflect current practice patterns. Most occurred before guidelines clearly described the need for anticoagulation and prior to the broad use of risk prediction scores (such as the CHA₂DS₂-VASc score) to provide objective, uniform guidance regarding risks vs. benefits of anticoagulation therapies.^5,6 When anticoagulation was given, warfarin was the mainstay, which carried a significantly higher risk of intracranial haemorrhage than currently available direct oral anticoagulants (DOACs). While the first DOAC was approved in 2010, prescriptions for those safer and more effective drugs began to exceed those for warfarin by 2013.⁷ Because of the convenience of DOACs, subsequent evidence also demonstrated that uptake of and compliance with these drugs are substantially higher than with warfarin.^7–9 Also in recent years, catheter ablation has become safer, more effective, and more widespread, with growing evidence, the procedure may lead to improved outcomes in AF patients.^10–12 Finally, observational studies and randomized trials have demonstrated that lifestyle factors may be modified to improve AF burden among those with the disease,^13–17 with some evidence those lessons have penetrated clinical practice.¹⁷

Whether these changes have impacted outcomes among those with AF to a meaningfully degree that can be detected in the general population remains unknown. We therefore sought to characterize modern prognostication among patients with AF and compare more recent outcomes to those observed in the past.

Methods

Data source, study design, and population

This was a cohort study using the California's Department of Health Care Access and Information (HCAI). Death information was retrieved from both the healthcare databases (for in-hospital mortality) and the California Comprehensive Death File. We included individuals aged ≥18 years with AF who received care in an ambulatory surgery unit, emergency department, or inpatient hospital between 1 January 2005 and 31 December 2019. Patients were not included based on the presence of atrial flutter. Healthcare encounters with missing data (age, sex, race, and ethnicity) were excluded. We excluded AF coded within 30 days of an open cardiac surgery as post-operative AF likely represents a different mechanism and may resolve with lower risks of AF-associated outcomes.¹⁸ Patients were divided into three mutually-exclusive groups, each corresponding to a 5-year period according to the time of first AF diagnosis in the databases: 1st January 2005–31st December 2009, 1st January 2010–31st December 2014, and 1st January 2015–31st December 2019. Patients could belong to only one the three period cohorts.

Diagnosis ascertainment

Diagnoses were identified through the International Classification of Diseases, Ninth Revision (ICD-9) code from January 2005 to September 2015 and the International Classification of Diseases, Tenth Revision (ICD-10) codes from October 2015 to December 2019 (Supplementary material online, Table S1). In addition to ICD-9 and ICD-10 codes, we used the Clinical Classification Software (CCS) codes for diseases with specific CCS codes (Supplementary material online, Table S1). Cardiac surgical procedures were identified using ICD-9 (Supplementary material online, Table S2) and ICD-10 codes (Supplementary material online, Table S3).

Study variables and outcomes

We gathered information regarding patients’ demographics, including age, sex, race, and ethnicity. We used the normalized race group for a patient based on a combination of their reported race and ethnicity. If a patient's ethnicity was ‘Hispanic,’ then the normalized race and ethnicity group was coded as Hispanic. For others, the normalized race and ethnicity group was assigned the same value as the reported race. Hence, race and ethnicity were mutually exclusive and categorized as Asian/Pacific Islander, Black, Hispanic, Native American (American Indian/Alaska Native), White, and ‘Other’. Co-morbidities of interest included hypertension, diabetes mellitus, heart failure, dyslipidaemia, coronary artery disease, peripheral artery disease, and chronic kidney disease. Dichotomous medical comorbidity variables were accumulated at each healthcare encounter and carried forward over time. The outcomes were incident ischaemic stroke, intracranial haemorrhage (non-traumatic cerebral and sub-arachnoid haemorrhage), and death.

Statistical analysis

Categorical variables are expressed as frequencies and percentages, while continuous variables are expressed as means with standard deviations. We assessed differences between groups using the Student's t-test for continuous variables and the Pearson Chi square test for categorical variables. Time to incident ischaemic stroke, intracranial haemorrhage, and death were assessed using survival analysis. Crude incidence rates of ischaemic stroke, intracranial haemorrhage, and all-cause mortality were calculated as the number with each outcome divided by the number of person-years at risk for each outcome. To compare change in incidence rates between demographic groups, we calculated the incidence rate ratio (IRR) with 95% confidence interval (CI) as the incidence rate in the period 2015–2019 divided by the incidence rate in the period 2005–2009. An IRR < 1 indicates a decline in incidence rate, and the lower the IRR, the higher the decline. The IRR were then compared by age (<65 years vs. ≥ 65 years), sex, and race and ethnicity, using the Mantel Haenszel test of homogeneity. Furthermore, we used multivariable Cox regression analysis to explore whether, compared with 2005–2009, the periods 2010–2014 and 2015–2019 were associated with a significant change in the outcomes of interest after adjustment for demographics and cardiovascular co-morbidities (hypertension, diabetes mellitus, dyslipidaemia, heart failure, chronic kidney disease, coronary artery disease, and peripheral artery disease). Covariates for inclusion in multivariable models were selected a priori as potential mediators or confounders of the outcomes based on biological plausibility and the medical literature. We performed a sensitivity analysis including the population with possible post-operative AF. We also investigated differences in the amplitude of change in the rates of adverse outcomes between 2005–2009 and 2015–2019 across demographic groups. Follow-up time was constrained to the duration of the individual cohort (e.g. patients in the 2005–2009 cohort were only followed for outcomes until the end of 2009). Risk estimates were reported as adjusted hazard ratios (HR) with 95% CI. A two-tailed P-value of ≤0.05 was considered statistically significant. All analyses were performed using Stata 17 statistical package (Statacorp, College Station, TX, USA).

Ethical considerations

This study received ethical approval from the Institutional Review Board of the University of California-San Francisco.

Results

We included 2 009 832 patients with AF. They were mostly males (52.7%) and Whites (70.7%), with a mean age was 75.0 ± 13.6 years. Compared with more recent years, patients in the 2005–2009 cohort were older, were less likely to be males, more likely to be white, and exhibited variable differences in cardiovascular comorbidities (Table 1). Throughout the 15 years of study, the incidence rates per 1000 person-years were 7.83 (95% CI 7.73–7.92) for ischaemic stroke, 3.17 (3.11–3.23) for intracranial haemorrhage, and 137.81 (95% CI 137.44–138.18) for all-cause mortality (Table 2). The incidence of ischaemic stroke, intracranial haemorrhage, and all-cause mortality each consistently diminished over time (Figure 1 and Table 3). After adjusting for age, sex, race and ethnicity, and co-morbidities, compared with 2005–2009, the periods 2010–2014 and 2015–2019 were associated with lower risks of each outcome (Table 3 and Supplementary material online, Table S4). A sensitivity analysis including patients with post-operative AF showed consistent results (Supplementary material online, Table S5).

Table 1.

Participant characteristics in three successive time periods

Variables	Total (n = 2 009 832)	2005–2009 (n = 738 954)	2010–2014 (n = 609 447)	2015–2019 (n = 661 431)	P value
Mean age (years)	75.0 ± 13.6	76.1 ± 13.2	74.6 ± 13.8	74.1 ± 13.7	<0.001
Male (%)	1 058 493 (52.7)	376 092 (50.9)	321 741 (52.8)	360 660(54.5)	<0.001
Race and ethnicity (%)					<0.001
Asian	163 356 (8.1)	51 744 (7.0)	50 686 (8.3)	60 926 (9.2)
Black	106 411 (5.3)	34 813 (4.7)	32 340 (5.3)	39 258 (5.9)
Hispanic	265 009 (13.2)	80 860 (10.9)	82 704 (13.6)	101 445 (15.3)
Native American	5 067 (0.3)	1 211 (0.2)	1 757 (0.3)	2 099 (0.3)
White	1 420 477 (70.7)	553 506 (74.9)	426 092 (69.9)	440 879 (66.7)
Other	49 512 (2.5)	16 820 (2.3)	15 868(2.6)	16 824 (2.5)
Diabetes (%)	542 448 (27.0)	182 454 (24.7)	165 980 (27.2)	194 014 (29.3)	<0.001
Hypertension (%)	1 129 989 (57.9)	365 191 (49.4)	309 681 (50.8)	455 117 (75.4)	<0.001
Heart failure (%)	558 259 (27.8)	218 980(29.6)	156 902 (25.7)	182 377 (27.6)	<0.001
Dyslipidaemia (%)	729 306 (36.3)	208 821 (28.3)	234 796 (38.5)	285 689 (43.2)	<0.001
CKD (%)	342 821 (17.1)	79 539 (10.8)	119 214 (19.6)	144 068 (21.78)	<0.001
CAD (%)	562 050 (28.0)	241 705 (32.7)	180 857 (29.7)	139 488(21.1)	<0.001
PAD (%)	130 233(6.5)	36 897 (5.0)	28 592 (4.7)	64 744 (9.8)	<0.001

CAD, coronary artery disease; CKD, chronic kidney disease; PAD, peripheral artery disease.

Table 2.

Crude incidence rates (per 1000 person-years) of ischaemic stroke, intracranial haemorrhage, and all-cause mortality among patients with atrial fibrillation

Variables	Total (n = 2 009 832)	2005–2009 (n = 738 954)	2010–2014 (n = 609 447)	2015–2019 (n = 661 431)	P value*
Ischaemic stroke
Overall	7.83 (7.73–7.92)	9.20 (9.04–9.36)	7.67 (7.50–7.83)	6.41 (6.27–6.56)	<0.001
Age (years)
<65	2.92 (2.80–3.03)	2.84 (2.65–3.04)	2.84 (2.64–3.05)	3.06 (2.87–3.27)	.17
≥65	9.31 (9.19–9.42)	10.85 (10.66–11.05)	9.25 (9.05–9.46)	7.51 (7.33–7.69)	<0.001
Sex
Male	6.15 (6.94–6.26)	6.96 (6.77–7.16)	5.96 (5.76–6.16)	5.46 (5.29–6.65)	<0.001
Female	9.73 (9.59–9.88)	11.58 (11.32–11.84)	9.63 (9.36–9.90)	7.58 (7.35–7.82)	<0.001
Race and ethnicity
Asian	8.72 (8.38–9.07)	9.55 (8.94–10.21)	8.84 (8.23–9.49)	7.91 (7.39–8.46)	<0.001
Black	9.29 (8.86–9.75)	10.47 (9.68–11.34)	8.92 (8.15–9.76)	8.54 (7.86–9.28)	<0.001
Hispanic	8.23 (7.97–8.49)	9.75 (9.26–10.27)	7.91 (7.46–8.38)	7.23 (6.85–7.64)	<0.001
Native American	6.02 (4.62–7.85)	9.58 (6.31–14.56)	4.21 (2.45–7.27)	5.33 (3.44–8.26)	.03
White	7.59 (7.49–7.70)	9.07 (8.88–9.25)	7.42 (7.23–7.61)	5.89 (5.73–6.06)	<0.001
Others	7.01 (6.49–7.57)	7.54 (6.64–8.57)	7.56 (6.61–8.66)	6.04 (5.26–6.94)	.02
Intracranial haemorrhage
Overall	3.17 (3.11–3.23)	3.34 (3.25–3.44)	3.18 (3.08–3.29)	2.97 (2.87–3.06)	<0.001
Age (years)
<65	1.71 (1.62–1.80)	1.61 (1.47–1.76)	1.63 (1.49–1.79)	1.87 (1.72–2.03)	.63
≥65	3.60 (3.53–3.67)	3.79 (3.67–3.90)	3.68 (3.55–3.81)	3.32 (3.20–3.43)	<0.001
Sex
Male	3.12 (3.04–3.20)	3.27 (3.14–3.41)	3.07 (2.93–3.21)	3.00 (2.87–3.13)	<0.001
Female	3.23 (3.15–3.31)	3.42 (3.28–3.56)	3.31 (3.16–3.47)	2.92 (2.75–3.07)	<0.001
Race and ethnicity
Asian	5.59 (5.32–5.86)	6.13 (5.66–6.65)	5.56 (5.09–6.06)	5.14 (4.73–5.58(	<0.001
Black	3.84 (3.57–4.13)	3.64 (3.19–4.14)	3.74 (3.27–4.28)	4.09 (3.65–5.60)	.74
Hispanic	3.81 (3.64–3.99)	3.85 (3.55–4.17)	3.79 (3.50–4.12)	3.79 (3.52–4.08)	.01
Native American	3.04 (2.11–4.37)	3.35 (1.67–6.69)	2.50 (1.25–5.00)	3.29 (1.91–5.67)	.80
White	2.75 (2.68–2.81)	3.00 (2.90–3.11)	2.75 (2.63–2.87)	2.42 (2.32–2.53)	<0.001
Others	3.15 (2.82–3.52)	4.54 (2.95–4.25)	3.40 (2.79–4.13)	2.59 (2.10–3.18)	.06
Mortality
Overall	137.81 (137.44–138.18)	161.04 (160.39–161.70)	132.55 (131.88–133.22)	116.37 (115.78–116.97)	<0.001
Age (years)
<65	54.04 (53.56–54.53)	58.29 (57.42–59.18)	51.85 (51.01–52.71)	52.01 (51.21–52.83)	<0.001
≥65	162.43 (161.97–162.89)	187.06 (186.27–187.86)	158.34 (157.50–159.18)	136.91 (136.17–137.65)	<0.001
Sex
Male	131.55 (131.05–132.05)	154.94 (154.04–155.84)	126.41 (125.52–127.32)	111.50 (110.72–112.29)	<0.001
Female	144.86 (144.30–145.41)	167.42 (166.46–168.38)	139.51 (138.51–140.51)	122.30 (121.40–123.21)	<0.001
Race and ethnicity
Asian	133.05 (131.75–134.35)	147.97 (145.57–150.40)	129.82 (127.52–132.16)	122.87 (120.85–124.93)	<0.001
Black	159.19 (157.41–160.99)	183.14 (179.83–186.52)	153.86 (150.67–157.11)	139.47 (139.47–145.03)	<0.001
Hispanic	125.29 (124.31–126.28)	142.55 (140.69–144.43)	122.71 (120.97–124.48)	113.23 (111.73–114.75)	<0.001
Native American	120.90 (114.15–128.05)	138.79 (124.67–154.50)	105.36 (94.75- 117.16)	122.72 (112.32–134.10)	.11
White	139.86 (139.42–140.30)	164.68 (163.92–165.45)	133.97 (133.17- 134.77)	114.12 (113.41–114.84)	<0.001
Others	119.27 (117.12–121.45)	127.88 (124.09–131.79)	115.36 (111.55–119.30)	114.46 (110.99–118.04)	<0.001

*P value is for comparison of incidence rates between all three time periods.

Figure 1.

Multivariable adjusted survival curves for incident ischaemic stroke, intracranial haemorrhage, and all-cause mortality among patients with atrial fibrillation over 3 successive 5-year time periods. Adjusted for age, sex, race and ethnicity, and co-morbidities (hypertension, diabetes, dyslipidaemia, heart failure, chronic kidney disease, coronary artery disease, and peripheral artery disease).

Table 3.

Predictors of incident stroke, intracranial haemorrhage, and all-cause mortality among patients with atrial fibrillation

	Ischaemic stroke		Intracranial haemorrhage		Mortality
Variables	HR (95% CI)	P value	HR (95% CI)	P value	HR (95% CI)	P value
Era
2005–2009	Ref	–	Ref	–	Ref	–
2010–2014	0.84 (0.82–0.86)	<0.001	0.90 (0.86–0.94)	<0.001	0.83 (0.82–0.84)	<0.001
2015–2019	0.66 (0.64–0.69)	<0.001	0.78 (0.75–0.82)	<0.001	0.76 (0.75–0.77)	<0.001
Age
<65	Ref	–	Ref	–	Ref	-
≥65	2.69 (2.58–2.81)	<0.001	1.98 (1.87–2.10)	<0.001	2.62 (2.60–2.65)	<0.001
Sex
Male	Ref	–	Ref	–	Ref	–
Female	1.45 (1.41–1.48)	<0.001	0.96 (0.93–0.99)	.03	1.03 (1.02–1.04)	<0.001
Race and ethnicity
White	Ref	–	Ref	–	Ref	–
Asian	1.12 (1.08–1.17)	<0.001	1.94 (1.84–2.05)	<0.001	0.92 (0.91–0.93)	<0.001
Black	1.27 (1.21–1.34)	<0.001	1.42 (1.31–1.53)	<0.001	1.11 (1.09–1.12)	<0.001
Hispanic	1.13 (1.09–1.17)	<0.001	1.40(1.33–1.47)	<0.001	0.91 (0.90–0.91)	<0.001
Native American	0.87 (0.66–1.14)	.31	1.17(0.81–1.69)	.41	0.93 (0.87–0.98)	.009
Others	0.97 (0.90–1.05)	.44	1.19 (1.06–1.33)	.004	0.87 (0.86–0.89)	<0.001
Co-morbidities
Diabetes	1.10 (1.07–1.13)	<0.001	1.18(1.13–1.23)	<0.001	1.11 (1.10–1.12)	<0.001
Hypertension	1.20 (1.17–1.23)	<0.001	1.20 (1.16–1.25)	<0.001	0.92 (0.91–0.92)	<0.001
Heart failure	1.37 (1.34–1.41)	<0.001	1.14 (1.09–1.19)	<0.001	1.76 (1.75–1.78)	<0.001
CKD	1.44 (1.40–1.49)	<0.001	1.52 (1.45–1.60)	<0.001	1.65 (1.63–1.66)	<0.001
Dyslipidaemia	0.90 (0.87–0.92)	<0.001	0.95 (0.91–0.99)	.009	0.73 (0.73–0.74)	<0.001
CAD	1.08 (1.05–1.11)	<0.001	0.98 (0.94–1.02)	.40	1.14 (1.13–1.15)	<0.001
PAD	1.33 (1.27–1.138)	<0.001	1.14 (1.06–1.22)	<0.001	1.33 (1.32–1.35)	<0.001

CAD, coronary artery disease; CI, confidence interval; CKD, chronic kidney disease; HR, hazard ratio; PAD, peripheral artery disease; Ref, reference category age, sex, race and ethnicity, cardiovascular co-morbidities (hypertension, diabetes, dyslipidaemia, heart failure, chronic kidney disease, coronary artery disease, and peripheral artery disease), and era were all included in the multivariable Cox regression analysis.

While declines in incident stroke and mortality were consistent across demographic subgroups, the magnitude of observed differences differed by age, sex, and race and ethnicity; differences in changes in incident intracranial haemorrhage were heterogenous (Figure 2). After adjustment for potential confounders, each racial and ethnic group except Native Americans experienced less declines in ischaemic strokes compared with white individuals; each group except Asians and Native Americans exhibited less declines in haemorrhagic stroke compared with white individuals; and white individuals had more declines in mortality compared with all other racial and ethnic groups (Table 4).

Figure 2.

Differences in temporal changes of crude incidence of ischaemic stroke, intracranial haemorrhage, and all-cause mortality by individual demographic characteristics. 95% CI, 95% confidence interval. IRR, incidence rate ratio comparing incidence rate between 2005–2009 and 2015–2019; an IRR < 1 indicates a decline in incidence rate; the lower the IRR, the higher the decline. P values are for groups comparison.

Table 4.

Differences in risk of ischaemic stroke, intracranial haemorrhage, and all-cause mortality between 2005–2009 and 2015–2019 by individual demographic characteristics

	Ischaemic stroke			Intracranial haemorrhage			All-cause mortality
Variables	HR (95% CI)	P value	P value for group difference	HR (95% CI)	P value	P value for group difference	HR (95% CI)	P value	P value for group difference
Age
<65	0.93 (0.84–1.03)	.18		0.97 (0.85–1.11)	.67		0.89 (0.87–0.91)	<0.001
≥65	0.65 (0.62–0.67)	<0.001	<0.001	0.77 (0.73–0.81)	<0.001	<0.001	0.75 (0.74–0.76)	<0.001	<0.001
Sex
Male	0.72 (0.68–0.75)	<0.001		0.79 (0.74–0.84)	<0.001		0.75 (0.74–0.76)	<0.001
Female	0.63 (0.60–0.66)	<0.001	<0.001	0.79 (0.74–0.85)	<0.001	0.19	0.77 (0.76–0.78)	<0.001	<0.001
Race and ethnicity
White	0.63 (0.61–0.66)	<0.001	–	0.76 (0.72–0.81)	<0.001	–	0.73 (0.72–0.74)	<0.001	–
Asian	0.77 (0.69–0.86)	<0.001	.001	0.73 (0.64–0.83)	<0.001	.82	0.87 (0.84–0.89)	<0.001	<0.001
Black	0.78 (0.68–0.89)	<0.001	.001	1.01 (0.83–1.23)	0.95	0.001	0.83 (0.80–0.85)	<0.001	<0.001
Hispanic	0.71 (0.65–0.77)	<0.001	.008	0.86 (0.76–0.98)	.02	.003	0.84 (0.83–0.86)	<0.001	<0.001
Native American	0.46 (0.23–0.91)	.03	.34	1.12 (0.43–2.92)	.82	.72	0.88 (0.75–1.03)	.10	.007
Others	0.79 (0.64–0.97)	.03	.03	0.76 (0.56–1.03)	.08	.58	0.92 (0.88–0.97)	.001	<0.001

CI, confidence interval; HR, hazard ratio.

Hazard ratios represents the risks for each outcome in 2015–2019 compared with 2005–2009 (baseline and reference category) in different demographic groups. A hazard ratio <1 denotes a decline in risk in 2015–2019 compared with 2005–2009.

Age, sex, race and ethnicity (each category compared with Whites), cardiovascular co-morbidities (hypertension, diabetes, dyslipidaemia, heart failure, chronic kidney disease, coronary artery disease, and peripheral artery disease), and era (2005–2009 and 2015–2019 time periods) were included in the multivariable Cox regression analysis.

For race and ethnicity, each group is compared with Whites.

Discussion

AF patients have experienced substantial declines in stroke, intracerebral haemorrhage, and death over time. However, differences by demographic categories were observed, with older patients experiencing the largest gains in avoiding these outcomes and Hispanics and Blacks consistently experiencing less improvements compared with their white counterparts.

Data regarding the incidence of ischaemic stroke in patients with AF is mostly limited to the era before 2010. Based on randomized controlled trial data, the use of warfarin for stroke prevention became part of standard recommendations for the management of patients with AF in the 1990s.¹⁹ Declines in strokes accompanied by rising use of warfarin were observed in studies focusing on trends from that time into the first decade of the millennium, although anticoagulation uptake remained suboptimal and diminishing stroke rates were not universally observed.^20–23

Direct oral anticoagulants were first introduced in the USA in 2010, the same year as the CHA₂DS₂-VASc score,²⁴ a risk score meant to identify appropriate candidates for anticoagulation that was more inclusive of the predecessor CHADS₂ score.²⁵ Since that year, DOAC use has increased in the US, and, as expected given the enhanced convenience of these drugs versus warfarin, evidence has shown increased uptake of appropriate anticoagulation in general.^7–9,26 It was not until 2014 that the CHA₂DS₂-VASc score was recommended and that DOACs were recommended as first line ahead of warfarin in official US guidelines.⁶ Prior European studies have demonstrated declines in AF-related stroke with greater use of DOACs,^27,28 but to our knowledge no study has re-examined contemporary trends in ischaemic stroke among US patients with AF. Indeed, we observed a 42% decline in the risk of stroke in these patients. Of note, the current study was not designed to determine the mechanisms of observed differences over time (e.g. the database utilized does not include medication prescriptions), but rather to describe changes in the pertinent epidemiology, and therefore other explanations may be operative. For example, prior to the year 2000, improved systolic blood pressure control had been described as a progressive phenomenon explaining reductions in ischaemic stroke in AF, a trend that may have continued.²¹ Similarly, although no randomized controlled trial has shown a reduction in stroke due to catheter ablation of AF, the procedure has both evolved and become substantially more common over the past two decades, with multiple observational studies suggesting that successful procedures may reduce the risk of stroke.^12,29,30 Another potential explanation for the decline in the incidence of ischaemic stroke in patients with AF is the increasing detection of short-episodes AF by extended cardiac monitoring with an implantable loop recorder, pacemaker, defibrillator, or implantable cardiac monitor.³¹ Such subclinical AF may be associated with a lower risk of ischaemic stroke than AF detected by more conventional means.³²

Lower rates of major bleeding, particularly intracranial haemorrhage, occur with DOACs compared with warfarin.^33–35 Hence, in parallel to a decline in incident ischaemic stroke, a decrease in the incidence of intracranial haemorrhage would be expected with a transition from warfarin to DOACs. Similarly, strategies to reduce the burden or impact of AF, such as more successful catheter ablations and lifestyle modifications, might theoretically reduce the frequency of anticoagulant use, which itself might contribute to lower risks of intracranial bleeding. Prior to multivariable adjustment, the crude incidence rate of intracranial haemorrhage in patients with AF failed to reveal any significant changes over time. However, after multivariable adjustment, we observed a statistically significant 17% decline in intracranial haemorrhage among AF patients in the most contemporary 5-year period vs. the earliest 5-year period, suggesting that comorbidities likely mediated the relationship. A prior analysis utilizing the inpatient sample concluded that the 3-year incidence of intracranial haemorrhages increased in comparing 2004–2006 with 2016–2018, a finding that coincided with a higher prevalence of hypertension and anticoagulation use.³⁶ However, the inpatient sample does not allow for a longitudinal assessment of the same patients (hence two intracranial haemorrhage events in the same patient would be considered the same as two independent patients) and is taken from only a sample of the patients experiencing inpatient hospitalization (whereas our current data enables tracking the same individual over time, includes multiple types of encounters, and is inclusive of every actual patient in three different healthcare settings in California).

Previous data on changes in overall mortality among patients with AF are especially limited. No significant change in AF-associated mortality in Olmsted County was observed between 1980 and 2000,³⁷ an era prior to many of the advancements already discussed. In the current study, mortality decreased by nearly a third when comparing the most recent time period to 2005–2009. Observational studies and randomized trials suggest that overall mortality may be lower than with DOACs than warfarin,^33–35 and once again the growing use of catheter ablation could potentially contribute to this observation.^11,12,38 A growing recognition that lifestyle factors, such as obesity and alcohol use, contribute to AF^13–17 might also motivate healthier behaviours that result in reduced recurrences and burden of AF, most recently shown to influence adverse outcomes in these patients.^39–41 Furthermore, with the advent of patient-centric monitoring devices and increased ambulatory monitoring in general, it is possible that individuals with paroxysmal AF who would have never been previously diagnosed with AF now carry this diagnosis in more contemporary periods. Hence, the observed decline in adverse outcomes could be partly explained by higher proportion of ‘lower risk’ patients with AF in the most recent period compared with 2005–2009.

We observed that several fundamental demographic characteristics were associated with substantial differences in each outcome, suggesting important public health implications when considering equitable improvements in AF-related care. For each demographic variable of interest, it is important to consider multiple possible explanations for the disparities observed, ranging from biological differences to healthcare access issues to reluctance to receive therapies when offered.

Older individuals consistently enjoyed a greater decline in each outcome than those who were younger. This is likely related to the fact that, not surprisingly and precisely as observed, older individuals experienced a higher risk for each outcome. Given the low-risk nature of the younger group, there may simply be little improvements to be had (and certainly less opportunity for improvements compared with the older individuals who began at a higher risk). It is also interesting to consider the age 65-year-old cut-off inherent to anticoagulation recommendations.²⁴ While that age threshold has persisted as conferring a higher-risk ‘point’ for both the CHADS₂ and CHA₂DS₂-VASc and during the transition from warfarin to DOACs,^6,24 some have speculated that the more convenient and lower risk DOACs might warrant consideration of broader inclusion among the younger and lower risk patients.⁴²

Differences in ischaemic stroke, intracranial haemorrhage, and death by sex were statistically significant, but relatively small in magnitude and variable in direction. The higher incidence rates of ischaemic stroke and mortality in females have been previously documented,^43,44 and yet, in contrast to the benefits of the higher risk among the older vs. younger, our study demonstrated that women also experienced less of a decline in stroke rates over time. Whereas females with AF have been less likely to receive vitamin K antagonist, more recent evidence suggests they are just as likely to receive DOACs than males,⁴⁵ which should have resulted in larger gains in improvements in stroke and intracranial haemorrhage if universally true. It is worth noting that females receive less rhythm control therapy,⁴⁶ which might ultimately translate into diminishing benefits in both stroke and intracranial haemorrhage.³⁹ It is surprising that, despite less improvements in stroke for men compared with women, men experienced more improvements in mortality over time. There are several potential explanations for these apparently discordant findings. First, again differences by sex were small and the P values were not particularly low—in these secondary analyses, we need to acknowledge that multiple hypothesis testing may lead to type 1 errors. However, the particularly small P value (of <0.001) for mortality favouring men would likely withstand even the most conservative adjustment for multiple tests, which, in the context of reduced declines in stroke and intracranial haemorrhage, suggest that other factors important to longevity are at play in AF patients and that women may have benefitted in particular in more contemporary times.

Multiple studies have demonstrated less commonly prescribed rhythm control, anticoagulation receipt, and more strokes for minority populations compared with Whites.^47–49 Once again, using the findings related to older vs. younger patients as an example, these racial disparities arguably provide more opportunity for improvements over time. To our knowledge, this is the first report to demonstrate that such disparities persist in considering improvements over time. Socioeconomic issues related to education and healthcare access for patients and potentially bias on the part of some providers may persist,⁵⁰ requiring ongoing evaluation and remediation.

Our study has several limitations. The coding of race and ethnicity was mutually exclusive, resulting in misclassification of multiracial individuals. Furthermore, race may be coded differently for the same individual and self-identified race may not fully represent genetic ancestry.⁵¹ Because outpatient data was limited only to procedures, our study population, inclusive of nearly every hospitalization, emergency department visit, and outpatient procedure in California, may not be fully representative of the entire AF population. However, we would expect stroke and intracranial haemorrhage to result in healthcare utilization captured by these databases. We were able to leverage the social security index to capture total mortality (even including deaths that occurred outside the three encounter types), which might also explain enhanced power for that particular outcome. We also acknowledge that some strokes and intracranial haemorrhages may present as death, potentially leading to misclassification in the current manuscript—however, given the severity of each condition and the nature of the datasets utilized, at least one of those outcomes was likely reliably ascertained. The use of administrative databases and diagnosis codes may lead to ascertainment bias from errors in coding and disease classification. However, prior research comparing use of these codes to chart review have revealed highly accurate test characteristics,⁵² and several of our observations within each time period (such as relationships between certain demographics and risk of each outcome) have been previously well established in other cohorts,^43,44,47,48 providing a degree of validation. Although we were able to account for multiple cardiovascular comorbidities, data regarding medication use, such as oral anticoagulation, and the degree of control of cardiovascular risk factors, was not available. However, the purpose of the current study was to describe the changing epidemiology and prognosis of adverse outcomes among AF patients and those limitations are pertinent to the mechanistic pathways that might explain the observed differences. The majority of patients were non-Hispanic white, and the findings observed may represent phenomena specific to California, potentially limiting generalizability. Finally, as this is an observational study, we cannot make confident statements regarding causality; however, again, this is predominately a description of the changing epidemiology of the most severe complications among nearly two million AF patients in the general population without the intent of identifying or describing the underlying causes of observed changes.

Conclusion

The risks of ischaemic stroke, intracranial haemorrhage, and death among AF patients have declined substantially over a recent 15-year period. These observations should be reassuring to modern AF patients, while reinforcing the ongoing importance of original research and implementation strategies to optimize evidence-based care. Differential improvements by demographic characteristics, most prominently less consistent improvements among Hispanic and Black AF patients, reveals subgroups most worthy of attention to enhance the quality of care.

Funding

NIH/NHLBI R01 HL158825-01 to Dr Marcus. Dr Marcus reports receiving funding from the NIH, PCORI, and the California Department of Cannabis Control; he is a consultant for and equity holder in InCarda.

Data Availability

The data analyzed are available with permission from the Department of Health Care Access and Information (HCAI) at https://hcai.ca.gov/data/request-data/.

Authors’ contributions

Conception and design: G.M.M. and J.J.N. Access to data: G.M.M. Data analysis and interpretation: J.J.N., G.M.M., T.A.D., and J.J.T. Manuscript drafting: J.J.N. Manuscript revision: J.J.N., T.A.D., J.J.T., and G.M.M. Supervision: G.M.M. Approval of the final manuscript: J.J.N., G.M.M., T.A.D., and J.J.T.