The impact of atrial fibrillation type on the risks of thromboembolic recurrence, mortality and major haemorrhage in patients with previous stroke: A systematic review and meta-analysis of observational studies

Abstract

Introduction

There is conflicting evidence on the impact of atrial fibrillation (AF) type, i.e. non-paroxysmal AF or paroxysmal AF, on thromboembolic recurrence. The consensus of risk equivalence is greatly based on historical evidence, focussing on initial stroke risks. We conducted a systematic review and meta-analysis to describe the impact of AF type on the risk of thromboembolic recurrence, mortality and major haemorrhage in patients with previous stroke.

Methods

We systematically searched four multidisciplinary databases from inception to December 2018. We selected observational studies investigating clinical outcomes in patients with ischaemic stroke and AF, stratified by AF type. We assessed all included studies for risk of bias using the ‘Risk of Bias In Non-randomised Studies – of Exposures’ tool. The Comprehensive Meta-Analysis Software was used to calculate odds ratios from crude event rates.

Results

After reviewing 14,127 citations, we selected 108 studies for full-text screening. We extracted data from a total of 26 studies, reporting outcomes on 23,054 patients. Overall, risk of bias was moderate. The annual incidence rates of thromboembolism in patients with non-paroxysmal AF and paroxysmal AF were 7.1% (95% confidence interval: 4.2–11.7) and 5.2% (95% confidence interval: 3.2–8.2), respectively. The odds ratio for thromboembolism in patients with non-paroxysmal AF versus paroxysmal AF was 1.47 (95% confidence interval: 1.08–1.99, p = 0.013). The annual mortality rates in patients with non-paroxysmal AF and paroxysmal AF were 20.0% (95% confidence interval: 13.2–28.0) and 10.1% (95% confidence interval: 5.4–17.3), respectively, and odds ratio was 1.90 (95% confidence interval: 1.43–2.52, p < 0.001). There was no difference in rates of major haemorrhage, odds ratio  = 1.01 (95% confidence interval: 0.61–1.69, p = 0.966).

Conclusion

In patients with prior stroke, non-paroxysmal AF is associated with significantly higher risk of thromboembolic recurrence and mortality than paroxysmal AF. Although current guidelines make no distinction between non-paroxysmal AF and paroxysmal AF for secondary stroke prevention, future guidance and risk stratification tools may need to consider this differential risk (PROSPERO ID: CRD42019118531).

Introduction

Atrial fibrillation (AF) is an atrial arrhythmia characterised by uncontrolled, rapid firing of atrial action potentials. This causes reduced cardiac output and turbulent flow, which can lead to blood coagulation and emboli.¹ Hence, AF is associated with an increased risk of stroke and consequently death.² AF is subclassified into paroxysmal AF (PAF) and non-paroxysmal AF (NPAF) forms. PAF refers to short spontaneously terminating episodes, while NPAF is persistent or permanent. PAF, NPAF and sinus rhythm are often difficult to differentiate and are recognised as non-mutually exclusive categories. Diagnoses differentiating PAF and NPAF are limited by the ability of current monitoring detection algorithms and arbitrary cut offs of AF burden. Emerging evidence suggests that even in the absence of clinical AF, atrial cardiopathy may associated with thromboembolism, thus determining cardiac thrombogenicity remains more complex than categorisation of AF burden.³ Nevertheless, guidelines suggest that categorisation of AF burden remains practical. The risk of stroke in AF patients is considered independent of AF type. Stroke-risk stratification scores that inform prescribing and management are based on sex, age and comorbidities rather than AF type.⁴ This consensus of relative risk (RR) equivalence in patients with PAF and NPAF is based on historical evidence evaluating the risk of index stroke.⁵ Despite some conflicting evidence,^6–8 the European Society of Cardiology (ESC) guidelines do not recommend that AF type should be a major factor in making decisions regarding oral anticoagulation (OAC) therapy.⁹ However, due to limited studies, the impact of AF type on outcomes following acute stroke remains unclear.

Although previous systematic review has examined the role of AF type as a risk factor for initial stroke occurrence and adverse outcomes,⁶ no systematic review has investigated the role of AF type following acute ischaemic stroke, i.e. as a risk factor for thromboembolic recurrence, mortality and major haemorrhage. The risk of ischaemic stroke is significantly higher in AF patients who have had a previous stroke than those who have not.¹⁰ Due to higher event rates, the impact of AF type on secondary prevention outcomes may be more pronounced.

Decisions to start effective AF-related stroke thromboprophylaxis following acute ischaemic stroke or intracerebral haemorrhage are rarely clear cut: patients have reluctance and own prejudices, relative contraindications and are influenced by their individual clinicians’ perceptions of risks and benefits. In addition, the current AF risk stratification tools are not perfect. A better understanding of risk factors could improve their prognostic and clinical utility. There are plausible reasons to think that AF type may be an important factor that has hitherto been ignored.

We conducted a systematic literature review and meta-analysis of observational studies to evaluate the impact of AF type on the risk of thromboembolic recurrence (stroke and systemic embolism), major haemorrhage and mortality in patients with prior stroke.

Methods

Inclusion and exclusion criteria

We conducted a systematic literature search of multidisciplinary databases: MEDLINE (OVID), EMBASE (OVID), Web of Science (Thomson Reuters) and CINAHL (EBSCO) to identify observational studies in which clinical outcome data were prospectively or retrospectively collected from inception to December 2018. We registered the study protocol on PROSPERO (CRD42019118531).

Study designs

We included prospective and retrospective cohort studies and case series that investigated patients with AF post-stroke and distinguished between AF types. We did not include studies on topics other than stroke outcomes in AF patients, unless stroke outcomes were reported. We excluded randomised control trials, as they are not representative of natural population frequencies. We also excluded review articles, commentaries, conference papers and case reports. We placed no restrictions on the basis of language; however, any foreign-language studies identified were only included when they could be translated into English.

Patients/participants

We included studies in which patients had an ischaemic stroke/transient ischaemic attack (TIA) of any form consistent with the current (International Classification of Disease (ICD-10)) World Health Organization¹⁰ definition, diagnosed in a hospital setting. We included mixed population studies (i.e. studies including patients with index and recurrent stroke) if they differentiated between AF types. We contacted the studies’ authors of mixed population studies if (1) the data of interest were not available from the original report, (2) they had been conducted within the last 10 years, and (3) they differentiated between AF types. We allowed one month for the authors to reply, with a follow-up/reminder email after two weeks. We included studies of authors who provided us with outcome data of interest within the specified time frame.

Exposures

We included studies if their definition of AF was compatible with the current ICD-10.¹⁰ The definitions of AF types in included studies had to be consistent with current American Heart Association, American College of Cardiology, ESC and Heart Rhythm Society guideline classifications of AF patterns:

1. PAF: Self-terminating, typically within 48 h, but may continue up to seven days. An AF episode terminated by cardioversion may still be considered paroxysmal if this (i.e. cardioversion) occurs within seven days.

2. NPAF:

   1. Persistent AF: AF that lasts longer than seven days, including episodes terminated by cardioversion after seven or more days.

   2. Long-standing persistent AF: Continuous AF lasting for at least a year until a rhythm control strategy is adopted.

   3. Permanent AF: AF that is accepted by patient and physician, therefore not including patients on rhythm control therapy.^9,11,12

We included studies that failed to use the current terminology of PAF and NPAF if their definitions were considered compatible, as judged by consensus of two authors (AM and AHAR). We grouped the terms ‘sustained’, ‘constant’ or ‘chronic’ AF as NPAF and ‘intermittent’ or ‘recurrent’ AF as PAF. We excluded studies failing to differentiate between AF types.

Outcomes

Our outcomes of interest were the incidence of stroke and systemic embolism, major haemorrhage and mortality:

1. Ischaemic stroke or systemic embolism diagnosed as per definition taken from the original article. This includes stroke recurrence during the treatment period and during follow-up, which was either definitely ischaemic (haemorrhage excluded by brain imaging or autopsy) or of unknown type (no brain imaging or autopsy performed);

2. Death from any cause during the scheduled follow-up period

3. Any intracerebral or major extracranial haemorrhage during the scheduled treatment period:

   1. Intracerebral haemorrhage, including symptomatic haemorrhagic transformation of the cerebral infarct, during the scheduled treatment period and during follow-up. The haemorrhage must have been confirmed by appropriate brain imaging after clinical deterioration or by autopsy

   2. The definition of major haemorrhage was taken from the original article but if none was given it was defined as any fatal bleed or bleeding severe enough to require transfusion or operation.

We included studies in which these outcomes for patients with a previous, definitely ischaemic stroke were extractable (i.e. stroke population and mixed population studies).

Study selection

The search syntax for MEDLINE was created in cooperation with a research librarian of the University of Glasgow. We adapted this search strategy for the other databases (Supplementary Table 1). All databases were accessed on 18 December 2018. We reviewed titles and abstracts using Covidence software (version 1.0, Veritas Health Innovation, Australia). We used the reference lists of retrieved articles for hand searches to identify additional relevant studies. As our emphasis was on published, peer-reviewed articles, we did not search grey literature beyond the scope of the included search engines and hand searches.

Data extraction and statistical analysis

Risk of bias assessment

All included studies underwent risk of bias assessment (Supplementary Table 2). We used the ROBINS-E (Risk of Bias In Non-randomised Studies – of Exposures) tool for non-randomised control trials.¹³ We judged all studies, with a particular emphasis on the focus of the paper, selection bias (i.e. recruitment method and exclusion and inclusion criteria), classification of AF and the acknowledgement of confounding factors and co-exposures. Any studies with serious risk of bias were excluded in sensitivity analyses.

Data extraction

We extracted the total number of patients with PAF and NPAF who had a previous ischaemic stroke, along with any data on the incidence of recurrent thromboembolic events, mortality and major haemorrhage. We converted Kaplan–Meier curves and risk ratios into crude event numbers to allow data uniformity. We also recorded follow-up period and OAC data. We stored all data on an electronic spreadsheet (Excel, version 2016 Microsoft, USA) after extraction.

Statistical analysis

The primary outcome of the meta-analyses was the incidence of recurrent thromboembolic events. The secondary outcomes were incidence of all-cause mortality and major haemorrhage. These analyses were conducted using Comprehensive Meta-Analysis software version 3 (Biostat, USA). As specified in the study protocol, we created a random-effects model to generate a pooled estimate of the summary event rates for both PAF and NPAF and performed a subgroup analysis accounting for OAC use post-stroke. The a-priori decision to use a random-effects model was made to accommodate the anticipated variation in study design and small sample size of observational studies. We calculated annual event rates by dividing the total number of events by length of follow-up (in years). Our analysis assumes that the incidence of events was constant over time. We assessed heterogeneity among studies by visual inspection of forest plots and I². We also calculated odds ratios (ORs) for the outcomes to compare NPAF and PAF event rates. This analysis only included studies that reported on both AF types. We assessed the strength of the summary data post-stroke comparing different AF types using the GRADE criteria¹⁴ (Supplementary Table 3). We visually inspected the funnel plots of outcomes for publication bias.

Results

We retrieved 14,127 references. Following deduplication, we screened 10,855 references. Finally, we included 108 studies of which 23 had the data of interest available from the original report. We contacted authors of the remaining 93 studies and received data from 3 further studies. Therefore, we extracted data from a total of 26 studies, reporting outcomes on 23,054 patients.^15–40 Figure 1 shows the process of study selection. Table 1 presents the baseline characteristics of included studies.

Figure 1.

PRISMA for study selection. The final analysis included 26 studies, reporting data from 23,054 patients. Adapted from Moher et al.⁴¹ AF: atrial fibrillation; NHS: National Health Service.

Table 1.

Baseline characteristics of all included non-randomised observational studies.

Study	Study type	Number of patients in study with history of previous stroke	Number of patients by AF type		Inclusion criteria	Comparators/exposure	OAC during follow-up (%)		Follow-up (mean or median)
			PAF	NPAF			PAF	NPAF
Al-Khalili et al.15	Retrospective cohort study	766	61	24	AF patients in Stockholm health centre treated with NOACs	Dabigatran versus Rivaroxaban versus Apixaban	100	100	395 days
Aronow et al.16	Prospective cohort study	136	–	136	Chronic AF patients ages 62 years or older	Warfarin versus Aspirin	–	50a	3 years
Azoulay et al. 17	Retrospective cohort study with case-control analysis	4643	–	4643	Chronic AF patients in the UK General Practice Research Databaseb	Warfarin versus Aspirin	–	23	3.9 years
Baturova et al.18	Prospective cohort study	336	65	44	AF patients who suffered first-ever ischaemic stroke in the Lund Stroke Register	Effect of heart rhythm and OAC use	45	45a	10 years
Britton and Gustafsson19	Prospective cohort study	288	31	61	AF patients diagnosed with brain infarction in Stockholm	AF versus sinus rhythm	–	–	10 days
Christensen et al.20	Prospective cohort study	85	18	–	Patients with prior cerebrovascular ischaemic event without prior AF diagnosis	Benefits of PAF detection by implantable loop recorder, no direct comparator	94	–	1.5 years
Friberg et al.21	Prospective cohort study	298	91	207	Patients with AF treated as inpatients in 2002 in Stockholm	AF type (paroxysmal versus permanent)	32	38	3.6 years
Grond et al.22	Prospective multicentre cohort study	1135	49	–	Survivors of stroke or TIA without known AF	Detection rates of PAF by 24-h versus 72-h Holter ECG monitoring	–	–	Hospital stay
Koga et al.23	Prospective multicentre cohort study	1192	434	758	NVAF patients with acute ischaemic stroke or TIA	Type of AF (paroxysmal versus sustained)	–	–	1.8 years
Levy et al.24	Prospective cohort study	15	1	14	Patients diagnosed with AF, none hospitalised	AF type (paroxysmal, chronic or persistent)	26	52	8.6 months
Liantinioti et al.25	Single-centre prospective cohort study	184	23	–	Cryptogenic stroke patients with no prior history of AF	Duration of PAF	85	–	3 months
Marini et al.26	Prospective cohort study	3530	−55	814	Patients with index ischaemic stroke	AF versus sinus rhythm	–	–	3.75 years
Ntaios et al.27	Prospective cohort study	811	277	534	AF patients with acute ischaemic stroke	AF types (paroxysmal, persistent and permanent)	33.9	38.8	10 years
Önundarson et al.28	Prospective cohort study	6	–	6	Men and women of Reykjavik	Chronic AF versus sinus rhythm	–	–	Unclear
Paciaroni et al.29	Prospective cohort study	2040	886	1154	Patients with acute ischaemic stroke and AF	PAF versus sustained AF	87.3	80	120 days
Palomaki et al.30	Retrospective cohort study	3256	1448	1808	AF patients with acute ischaemic stroke or TIA	PAF versus chronic AF	32.2	63.6	30 days
Petty et al.31	Retrospective cohort study	1111	129	138	All residents of Rochester who have suffered from stroke	Characteristics that could impact survival and recurrence post-stroke	–	–	Unclear
Rietbrock et al.32	Retrospective population-based cohort study	7628	–	7628	Chronic AF patients over 40 years in the UK General Practice Research Databaseb	CHADS2 risk stratification points	–	–	3.3 years
Staszewski et al.33	Prospective cohort study	178	70	108	AF patients with acute ischaemic stroke and at least 72 h of continuous ECG monitoring	AF type (paroxysmal versus permanent)	–	–	6 months
Tanaka et al.34	Retrospective cohort study	449	178	271	Patients with acute ischaemic stroke and AF	Age (aged 80 years or older versus younger than 80)	–	–	90 days
Tsivgoulis et al.35	Prospective cohort study	207	66	141	AF patients with first-ever ischaemic stroke	Oral anticoagulants versus aspirin	–	–	7.4 years
Wolf et al.36	Prospective cohort study	20	–	20	Men and women of Framingham aged 30 to 62	Chronic AF versus sinus rhythm	–	–	Unclear
Yamanouchi et al.37	Retrospective cohort study	23	5	18	NVAF patients with sustained embolic brain infarction on warfarin anticoagulation treatment	Warfarin versus no treatment (autopsy series)c
	100	100	3.8 years
Yanagisawa et al.38	Prospective cohort study	64	27	37	Elderly patients with AF receiving outpatient treatment in Nagoya	Body mass index	–	–	1.6 years
Yu et al.39	Retrospective cohort study	69	–	69	NV, persistent AF patients who survived hospital stay	Type of antithrombotic treatment	–	41.1	360 days
Zolotovskay et al.40	Prospective cohort study	661	153	354	NVAF patients with history of carotid cardio-embolic stroke without carotid artery stenosis	Type of AF (first diagnosed, paroxysmal, persistent and constant)	–	–	1 year

Note: AF: atrial fibrillation; NPAF: non-paroxysmal atrial fibrillation; ECG: electrocardiogram; PAF: paroxysmal atrial fibrillation; OAC: oral anticoagulation; TIA: transient ischaemic attack; NV: Non-valvular; NVAF: Non-valvular atrial fibrillation; NOAC: Non-vitamin K oral anticoagulant.

^aSubgroup analysis of OAC versus no OAC treatment available from original report.

^bSame database used, therefore data was not taken from these studies for the same outcomes.

^cAutopsy series was not included in meta-analysis.

Notably, two studies^17,32 were based on a single database (UK General Practice Research Database), reporting on thromboembolic recurrence in patients from 1993 to 2008 and 1987 to 2007, respectively. Thus, to prevent reporting bias, we used Rietbrock et al.³² for the analysis of thromboembolic recurrence due to its larger sample size. We extracted major haemorrhage rates from Azoulay et al., as these data were not available from Rietbrock et al.^17,32

Quality of evidence

The risk of bias was assessed for all included studies (Supplementary Table 2). All studies were from representative AF populations. Confounding factors were common potential sources of bias. Eight studies did not report baseline characteristics stratified by AF types. Three studies had significantly older patients or higher incidences of comorbidities in the NPAF group, particularly coronary artery disease, hypertension, chronic heart failure and diabetes mellitus. Selection bias was common: 13 studies excluded patients who had died after their index stroke. We classified the risk of bias in Petty et al.³¹ for reporting outcomes as serious because the RR for all-cause mortality could not be reconciled with crude event rates and must have been misreported. We therefore excluded the all-cause mortality data from Petty et al.³¹ in our analysis. Follow-up duration was adequate for most studies. However, Grond et al.²² only followed up patients with PAF for the length of hospital stay post-stroke, potentially leading to an underestimate of outcome incidences in PAF. The funnel plot for the log OR of thromboembolic recurrence indicated potential publication bias (Supplementary Figure 1). The distribution of studies was asymmetrical, smaller studies tended to show lower OR than larger studies. Visual inspection of the other funnel plots showed greater symmetry. Overall, certainty of evidence, in accordance with the GRADE criteria,¹⁴ was low (Supplementary Table 3). However, we were conservative in scoring, downgrading the evidence for being observational.

The impact of AF type on the recurrence of thromboembolism

Thromboembolic data were recorded by 18 studies, reporting on a total of 17,627 patients (Figure 2). Confounding factors were reported in 10 studies, accounting for 2535 patients. Patients with NPAF had a higher median National Institutes of Health Stroke Scale (NIHSS) as well as higher rates of ischaemic heart disease and congestive heart failure than patients with PAF. The mean reported age was 75 in patients with PAF and 77 in patients with NPAF (Supplementary Table 4). Twelve studies compared PAF and NPAF patients and 6 studies reported only on either PAF or NPAF. Three studies^23,29,35 reported a composite outcome of systemic embolism and ischaemic stroke recurrence, while the others reported ischaemic stroke recurrence alone. The pooled random-effects estimates for the risk of recurrent thromboembolism in NPAF and PAF patients were 14.1% (95% confidence interval (CI): 8.2–23.1, Figure 2(a)) and 9.0% (95% CI: 5.4–14.6, Figure 2(b)), respectively. The average follow-up times of studies reporting thromboembolic recurrence in NPAF and PAF were 721 days and 577 days, respectively. We conducted a meta-regression to address the assumption of stable risk of thromboembolic recurrence over the follow-up period. This showed a slight reduction of risk with increasing follow-up but of small magnitude (Supplementary Figure 2). Considering only the studies that reported follow-up duration, the average annual event rates of thromboembolic recurrence in NPAF and PAF were 7.1% (95% CI: 4.2–11.7) and 5.2% (95%CI: 3.2–8.2), respectively. We performed a sensitivity analysis as Aronow et al.¹⁶ appeared as an outlier. This resulted in a reduction of the estimated NPAF annual event rate, 6.4% (95%CI: 4.2–9.5). Direct comparison of thromboembolic recurrence in NPAF versus PAF showed significant difference, OR was 1.47 (95%CI: 1.08–1.99, p = 0.013, Figure 2(c)) based on 12 studies (n = 5680). Heterogeneity as measured with I² was moderate at 40.1%. The funnel plot indicated potential publication bias (Supplementary Figure 1).

Figure 2.

Thromboembolic estimated event rates for (a) non-paroxysmal atrial fibrillation and (b) paroxysmal atrial fibrillation patients and (c) the direct comparison by odds ratio of the risk of thromboembolic recurrence. CI: confidence interval.

Oral anticoagulation

Subgroup analysis comparing the incidence of thromboembolism in studies with a low proportion of patients on OAC (i.e. <50%) to studies with a high proportion of patients on OAC (i.e. >50%) showed no significant difference between NPAF and PAF (Supplementary Figure 3). The event rate estimates in NPAF and PAF were non-significantly higher in the studies reporting lower OAC use, by 3.3% (95%CI: −16.8 to 21.2) and 3.1% (95%CI: −14.3 to 25.7), respectively.

The impact of AF type on all-cause mortality post-stroke

All-cause mortality was reported in 18 studies, representing 7928 patients. Confounding factors were reported in 12 studies, accounting for 5897 patients. Patients with NPAF had higher median NIHSS and higher rates of ischaemic heart disease and congestive heart failure than patients with PAF. The mean reported age was 75 in patients with PAF and 79 in patients with NPAF (Supplementary Table 4). Fourteen studies compared NPAF and PAF and four studies reported on one AF type (Figure 3). The pooled random-effects estimates for all-cause mortality rate in NPAF and PAF were 34.5% (95%CI: 22.7–48.4, Figure 3(a)) and 16.3% (95%CI: 8.8–28.1, Figure 3(b)), respectively. The average follow-up times of studies reporting all-cause mortality in NPAF and PAF were 630 days and 584 days, respectively. We conducted a meta-regression to address the assumption of stable risk of mortality over the follow-up period. The risk of mortality in the included studies showed no association to the length of follow-up. Excluding all studies that failed to report follow-up duration, the estimated annual mortality rates in NPAF and PAF were 20.0% (95%CI: 13.2–28.0) and 10.1% (95%CI: 5.4–17.3), respectively. The OR for all-cause mortality was significant, 1.90 (95%CI: 1.43–2.52, p < 0.001, Figure 3(c)). Heterogeneity was moderate (I² = 63.0%).

Figure 3.

All-cause mortality estimated event rates for (a) non-paroxysmal atrial fibrillation and (b) paroxysmal atrial fibrillation patients and (c) the direct comparison by odds ratio of the risk of all-cause mortality. CI: confidence interval.

Baturova et al.¹⁸ was an outlier. We performed a sensitivity analysis including the three-year all-cause mortality data of the study (Supplementary Figure 4). Heterogeneity fell to I² = 59.8%, with lower event rates and OR (1.75, 95%CI: 1.35–2.28). Another outlier, as seen in Figure 3(c), was Levy et al.²⁴ We performed a sensitivity analysis; however, due to the small weight of the study (n = 15), it had little impact on the OR and did not improve the measure of heterogeneity (I² = 62.4%).

The impact of AF type on the risk of major haemorrhage post-stroke

Major haemorrhage data were reported in eight studies based on 2072 patients (Figure 4). Three studies^17,22,33 only reported intracranial haemorrhages as major haemorrhage events. The estimated rates of major haemorrhage in NPAF and PAF were 6.3% (95%CI: 2.9–13.1, Figure 4(a)) and 4.4% (95%CI: 3.0–6.3, Figure 4(b)), respectively. There was no difference in major haemorrhage risk, and OR was 1.01 (95%CI: 0.61–1.69, p = 0.966, Figure 4(c)). Heterogeneity was negligible (I² = 0.00).

Figure 4.

The estimated event rates of major haemorrhage for (a) non-paroxysmal atrial fibrillation and (b) paroxysmal atrial fibrillation and (c) the direct comparison by odds ratio of the risk of major haemorrhage. CI: confidence interval.

Discussion

The impact of AF type on the risk of the thromboembolic recurrence

Our analysis suggests that, in patients with prior stroke, NPAF is associated with a significantly higher risk of thromboembolic recurrence than PAF. Our analysis is distinct from previously conducted studies evaluating the impact of AF type on clinical outcomes,^5–8 as the vast majority of patients evaluated in these had no prior stroke. Our analysis adds to the findings of a recent meta-analysis, suggesting that NPAF is associated with a significant increase of thromboembolic risk in patients without prior stroke.⁶

AF is associated with a six-fold increase in stroke.¹ Patients with previous ischaemic stroke have an even higher risk. There are several stroke-risk stratification scores for AF patients that determine whether OAC therapy is suitable. However, none account for AF type, and current guidelines recommend that AF type should not influence decisions regarding OAC therapy.⁹ Our analysis challenges this and indicates that the current belief in the equivalence of thromboembolic risk in NPAF and PAF needs to be re-evaluated.

The potential causes for an increased observed thromboembolic recurrence risk in NPAF than PAF patients may be burden of AF, inherent differences in pathophysiology and development of AF or confounding factors. Even though confounding factors stratified by AF type and history of stroke were only reported in 10 studies, these suggest higher rates of comorbidities in patients with NPAF and are likely to have contributed to the perceived higher thromboembolic risks. Exploring the burden of AF among PAF patients may provide further insight into the causes of the differential risk.

Furthermore, AF is a progressive disease. Up to 15% of new-onset PAF patients may progress to NPAF within one year.⁴² A previous observational study reported that the progression from PAF to NPAF was associated with increased adverse events.⁴³ Even though we did not specifically look at patients who had progressed, the higher thromboembolic risk for NPAF in our analysis suggests that there may be clinical need to monitor or even prevent the progression of PAF to NPAF. The potential of catheter ablation and risk factor modification, which slow disease progression,^44,45 needs further investigation.

We found potential publication bias on inspection of the funnel plot of log OR of thromboembolic recurrence, which could have led to an overestimation of the increased risk associated with NPAF compared to PAF.

The subgroup analysis adjusting for OAC did not suggest any significant difference in thromboembolic recurrence rates in patients with NPAF and PAF. Unfortunately, due to small sample size and incomplete reporting of OAC use post-stroke, we were unable to analyse the efficacy and risks of OAC in NPAF and PAF appropriately. Our findings do not suggest that the effectiveness of OAC in reducing thromboembolic recurrence is dependent on AF type. In fact, given the smaller bleeding risk profile of direct OAC and the emerging evidence for thromboembolic risks in atrial cardiopathies in the absence of AF, more patients may benefit from OAC than are currently treated.³ Future studies investigating the effects appropriately could provide further guidance for the initiation of thromboprophylaxis post-stroke. Moreover, according to ESC guidelines,⁹ all patients included in the study should have received OAC unless contraindicated, as CHA₂DS₂-VAS_C recommends OAC in patients with previous stroke or TIA. OAC prescription was much lower than anticipated, as several studies reported that less than 50% of their patient population were receiving OAC.

The impact of AF type on the risk of mortality post-stroke

Our study implies that, in patients with prior stroke, NPAF is associated with an increased risk of all-cause death compared to PAF. The higher risk profile and comorbidities of NPAF patients, in particular higher rates of ischaemic heart disease and congestive heart failure, may have contributed to a higher mortality rate. The difference in mortality rate in NPAF versus PAF exceeded the difference of thromboembolic recurrence in our study. As the increased mortality in AF is primarily attributed to cerebrovascular events,¹ the higher mortality rate may have masked thromboembolic recurrence in NPAF patients.

Strengths and limitations

One major limitation of the current study was the restricted sample size, a result of the small number of included studies, poor reporting and missing stratification of data in mixed population studies. This led to wide CIs, particularly for estimated event rates. Unfortunately, we were unable to account for the limitations of included studies. The data available did not allow the analysis of the impact of patients’ risk profiles (i.e. CHA₂DS₂-VAS_C) on the difference in thromboembolic risks. Further studies are needed to explore the impact of AF type on the risk of thromboembolic recurrence at different stroke-risk profiles.

We only included observational studies, only determining association not causation. While they are non-randomised and result in a higher risk of confounding bias, they represent normal population frequencies and are more suitable for meta-analyses in epidemiology than randomised control trials. Our study may also be subject to recording bias and error, as retrospective studies rely on the adequacy of registry data.

We analysed both retrospective and prospective studies together. Previous analyses have demonstrated differences in outcomes between retrospectively and prospectively identified patients with AF.⁴⁶ These differences may have increased heterogeneity of results and decreased reliability of the data. Heterogeneity between studies, evaluated by I² and visual inspection of funnel plots, was moderate for the analysis of thromboembolic recurrence and mortality. We used a random-effects model to adjust for this, maintaining the robustness of results.

We used crude event rates, allowing data uniformity. However, reporting on confounding was poor; thus, we did not adjust for significant confounders. Nevertheless, CHA₂DS₂-VAS_C, HAS-BLED, NIHSS scores and risk factors for stroke were recorded when possible (Supplementary Table 4). Not adjusting for confounders, such as age, comorbidities and structural heart disease may have led to overestimation of event rates in patients with NPAF. A secondary analysis using adjusted risk ratios, where available, could reduce confounding bias. We did not conduct such an analysis as adjusted risk ratios were only reported in 3 of the 26 studies.

Selection bias, in particular survivor bias, was common in the included studies, which may have diluted true event rates. We included studies that investigated patient outcomes immediately after stroke and others who examined stroke survivors. Thus, our data do not provide insight into whether the time since index stroke influences the risk of recurrence and death. A sensitivity analysis evaluating outcomes in studies with and without survivor bias independently could determine whether it has an impact on risks of thromboembolic recurrence or mortality. However, this analysis would have limited our sample size further. Furthermore, due to large variation in follow-up time and missing longitudinal data, we would still be unable to investigate temporal trends of event rates. High mortality rates could have masked thromboembolic recurrence in both PAF and NPAF, therefore underestimating the risk of stroke recurrence. We were unable to adjust for these competing risks in the study-level meta-analysis.

All included studies evaluated AF at baseline without re-evaluating exposure status during follow-up. As PAF can progress into NPAF over time, some patients with PAF may have unknowingly progressed to NPAF. This could have led to an underestimate of the effect of NPAF.

Conclusions

In patients with prior stroke, NPAF is associated with significantly higher risks of thromboembolic recurrence and mortality compared to PAF. This suggests potential clinical need to monitor or even prevent the progression of AF. Future stratification scores may need to include this parameter to better estimate stroke recurrence risks. Nevertheless, atrial thrombogenicity remains more complex than the categorisation of AF burden. All patients with previous stroke should receive OAC therapy, and AF burden should not determine whether patients would benefit from OAC.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

Ethical approval

Not applicable.

Informed consent

Not applicable.

Guarantor

Not applicable.

Contributorship

AHAR supervised the project. AM and AHAR developed the protocol, appraised the literature and performed the statistical analysis. AM drafted the initial article. AM and AHAR were involved in reviewing and reporting of the work. All authors provided critical revision of the article for important intellectual content and approved the final version.

ORCID iDs

Alan C Cameron https://orcid.org/0000-0001-6965-1109

Azmil H Abdul-Rahim https://orcid.org/0000-0002-1318-4027

Supplemental Material

Supplemental material for this article is available online.