Article Text

Download PDFPDF

Original research article
N-terminal pro B-type natriuretic peptide in systematic screening for atrial fibrillation
  1. Emma Svennberg1,
  2. Peter Henriksson1,
  3. Johan Engdahl1,
  4. Ziad Hijazi2,
  5. Faris Al-Khalili1,
  6. Leif Friberg1,
  7. Viveka Frykman1
  1. 1 Department of Clinical Sciences, Cardiology Unit, Karolinska Institutet, Danderyd's University Hospital, Stockholm, Sweden
  2. 2 Department of Medical Sciences and Cardiology, Uppsala Clinical Research Center, Uppsala University, Uppsala, Sweden
  1. Correspondence to Dr Emma Svennberg, Department of Clinical Sciences, Cardiology Unit, Karolinska Institutet, Danderyd University Hospital, Hjärtkliniken plan 2, hus 18, Stockholm, SE-182 88, Sweden; emma.svennberg{at}


Objective Screening for atrial fibrillation (AF) in individuals aged 65 and above is recommended by the European Society of Cardiology. Increased levels of the biomarker N-terminal pro B-type natriuretic peptide (NT-proBNP) has in cohort studies been associated with incident AF.

The aim of this study was to assess whether NT-proBNP could be useful for AF detection in systematic screening.

Methods The Strokestop study entailed 7173 Swedish residents aged 75/76 that were screened for AF using twice daily intermittent ECG recordings during 2 weeks. In a substudy of 886 participants, the last 815 consecutive participants and 71 individuals with newly detected AF, levels of NT-proBNP were determined.

Results Participants with newly detected AF (n=96) had a median NT-proBNP of 330 ng/L (IQR 121;634). In individuals without AF (n=742), median NT-proBNP was 171 ng/L (IQR 95;283), p<0.001. The CHA2DS2-VASc parameters did not differ significantly between individuals with newly detected AF and without AF nor between newly detected AF in the NT-proBNP cohort compared with the cohort where NT-proBNP was not assessed. Using an NT-proBNP cut-off of ≥125 ng/L in a non-acute setting yielded a negative predictive value of 92%, meaning that 35% fewer participants would need to be screened when applied to systematic AF screening. Adding weight to NT-proBNP further reduced participants needed to be screened with a preserved sensitivity.

Conclusions NT-proBNP was increased in individuals with newly detected AF. Prospective studies could clarify if NT-proBNP can be used to correctly select individuals that benefit most from AF screening.

Clinical trials Identifier: NCT01593553.

  • Atrial Fibrillation
  • Stroke
  • Epidemiology

Statistics from

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.


Atrial fibrillation (AF) can be difficult to diagnose, as it might be intermittent and asymptomatic.1 Regardless of symptomatology individuals with AF have a fivefold increased risk of ischaemic stroke.2 3 The risk of stroke is assessed using a stroke risk score with the acronym CHA2DS2-VASc. Oral anticoagulant (OAC) therapy reduces the risk of stroke by at least 64% and is recommended for all except for those at a truly low risk and those whose perceived bleeding risk outweighs the benefits of treatment.4

Early identification of AF could be beneficial as a stroke-preventive measure, since initiation of protective OAC treatment can be made. Opportunistic screening for AF using pulse–palpation is recommended above age 65, according to the latest European Society of Cardiology guidelines,4 and has a detection rate of new AF of 1.6%.5

Single time-point screening has in a meta-analysis identified new AF in 1.4% of >65-year olds.6 In the systematic screening Strokestop study, 7173 individuals aged 75/76 years in two Swedish regions were screened for silent AF using intermittent ECG recordings for 2 weeks. This yielded 3% individuals with newly detected AF and increased AF prevalence by more than 30%.7

N-terminal pro B-type natriuretic peptide (NT-proBNP) is an established biomarker, clinically used as a marker for heart failure. More recently, NT-proBNP has emerged as an independent predictor for incipient AF in several large cohort studies.8 9

The aim of this study was to evaluate the usefulness of NT-proBNP in a systemic screening programme cohort; to establish whether there is a difference in levels of NT-proBNP between individuals with screening detected AF compared with participants in which AF was not detected; and to assess whether NT-proBNP could be a useful predictor of whom would benefit from screening for AF.


The design of the Strokestop study has been published previously.10 In short, all individuals (n=28 768) aged 75 or 76 residing in two Swedish regions were randomised in a 1:1 fashion to a control group or invited to screening for AF. In the screening group, 7173 individuals (54%) attended screening. They were asked to fill out a health questionnaire detailing cardiovascular comorbidities, prior AF diagnosis and if they were on treatment with OAC or a platelet inhibitor. In the Stockholm region, participants were asked to report self-estimated height and weight.

Individuals without known AF were equipped with a handheld one-lead device ( for intermittent 30 s ECG recordings twice daily during 2 weeks. This device has been validated in prior studies and has been shown to be superior to 24 hours Holter recordings for detection of AF.11 Newly detected AF was diagnosed if there was at least one registration of 30 s of AF (irregular rhythm lacking p-waves) or at least two episodes of AF lasting 10–29 s. If intermittent ECG recordings were difficult to interpret, additional long-term ECG recordings were performed. All individuals with newly detected AF and patients with AF not treated with OACs were offered a structured follow-up by a cardiologist to ensure adequate treatment.4

In the last 815 consecutive participants attending the Strokestop study, NT-proBNP was analysed bedside prior to ECG screening using a point-of-care analyser (Cobas 232 point-of-care) by Roche.12 In addition, 71 individuals with newly detected AF had NT-proBNP taken at cardiology follow-up (Immulite 2000 XPI).

Participants with NT-proBNP levels exceeding 9000 ng/L were referred to a cardiologist for further work-up. The investigators interpreting the ECGs were blinded to NT-proBNP levels.


All statistical tests were two sided and p<0.05 was used to identify statistically significant results. For all continuous variables, visual inspection of histograms and the Shapiro-Wilk's test was calculated to assess deviation from normal distribution. Normally distributed continuous data are described as mean +/- SD NT-proBNP data were not normally distributed and are reported as median and IQR. CHA2DS2-VASc score was regarded as ordinal data. For NT-proBNP and CHA2DS2-VASc, non-parametric tests were used, Mann-Whitney U and Kruskal-Wallis tests, respectively, as appropriate. For normally distributed data, x2, Fisher's exact and student t-tests were used to compare differences between groups.

A univariate and multivariable analysis was performed using binary logistic regression and the results are presented as ORs with 95% CIs. For all categorical variables, the ORs show a change from a reference category (no present illness and male gender). For the continuous variables, height and weight, one unit change (1 cm or 1 kg) was used. For NT-proBNP, quartiles were used. The discriminative ability of the model was estimated as the area under the curve (AUC).

The data were split 50/50 into two randomised groups for derivation and validation, respectively, using frequency matching based on AF diagnosis. In the derivation group sensitivity, specificity and predictive values for different cut-offs of NT-proBNP were studied to find a clinically relevant cut-off value. In addition, receiver operating characteristic regression (ROC) curves were assessed. Youden's index can be used to address performance of a dichotomous test, and this was calculated to see whether it could be of used to find a clinically helpful cut-off for NT-proBNP. The suggested cut-off for NT-proBNP was subsequently verified in the validation group.

All analyses were performed with IBM SPSS Statistics V.22 software.


The study complies with the Declaration of Helsinki, and the protocol was approved by the regional ethics committee (DNR 2011–1363–31/3). Informed consent was obtained from all participants in the screening programme. identifier: NCT01593553.


Comparison between groups with regards to NT-proBNP assessment

A total of 886 individuals had NT-proBNP analysed (NT-proBNP cohort) of the 7173 participants. Individuals with newly detected AF in the NT-proBNP cohort did not differ significantly compared with the group with newly detected AF who did not have NT-proBNP assessed. The group without detected AF had higher prevalence of prior vascular disease (p<0.001) and stroke/transient ischaemic attack (TIA) (p=0.023) in the NT-proBNP cohort (table 1). CHA2DS2-VASc scores did not differ between the groups.

Table 1

Baseline characteristics of participants from the entire cohort divided per assessment of NT-proBNP and per diagnosis. Data from the STROKESTOP study7

NT-proBNP cohort

Individuals with newly detected AF (n=96) had a NT-proBNP median of 330 ng/L (IQR 121–634), whereas individuals where AF was not detected (n=742) had a NT-proBNP median of 171 ng/L (IQR 95–283), p<0.001. Participants with AF diagnosed prior to the screening programme, n=48, had a significantly higher NT-proBNP median of 472 ng/L (IQR 187– 1170), compared with newly detected AF, p=0.016 and compared with individuals where AF was not detected, p<0.001 (table 2).

The clinical characteristics included in the CHA2DS2-VASc score did not differ significantly between the groups with newly detected AF compared with those without detectable AF in the group where NT-proBNP was assessed (table 2). Individuals with a prior diagnosis of AF had a median CHA2DS2-VASc score that did not differ from individuals with newly detected AF and individuals without AF. There were significantly more participants with heart failure (p<0.001) and a significantly lower proportion of women (p=0.012) in the group with a previous diagnosis of AF than in the group without AF. No other clinical parameters differed.

Table 2

Baseline characteristics in participants per diagnosis

Information about height and body weight was obtained from 478 individuals. Individuals with newly detected AF were heavier than individuals without AF (83.3 kg vs 75.7 kg, p=0.002) (table 2).

There were no significant gender differences in NT-proBNP levels regardless of AF status (see online supplementary table 1).

Supplementary Table 1

NT-proBNP cohort with exclusion of heart failure

After exclusion of participants with known heart failure, n=29, NT-proBNP median in individuals with newly detected AF was 328 ng/L (IQR 118–632) and in known AF 421 ng/L (IQR 153–979), both remained increased as compared with individuals where AF was not detected (166 ng/L (IQR 94–276)), p<0.001 for both comparisons.

NT-proBNP and type of AF

The majority of the 96 individuals with newly detected AF had paroxysmal AF detected using intermittent ECGs (n=61) or other types of long-term ECGs (n=16). One patient had paroxysmal AF diagnosed during a hospital visit (n=1). Eighteen individuals were in AF on the first visit; these were more likely to have a persistent arrhythmia.

Participants with paroxysmal AF, n=78, had significantly higher NT-proBNP levels (median 227 ng/L (IQR 93–452) than participants without detection of AF n=742 median 171 (IQR 95–283), p=0.035.

Individuals with AF detected at first medical contact (n=18) had a higher NT-proBNP level than individuals with paroxysmal AF (median 847 (IQR 504– 1220) vs 227 ng/L (IQR 93–452), p<0.001 and individuals with known AF, n=48 (median 472 ng/L (IQR 184–1186)), p=0.05, figure 1.

Figure 1

Box-plot depicting median NT-proBNP in ng/L for AF subtypes. AF, atrial fibrillation; NT-proBNP, N-terminal pro B-type natriuretic peptide.

Participants with newly detected AF who had five or fewer AF episodes on their intermittent ECG recordings had lower median levels of NT-proBNP, 321 ng/L (IQR 113–592) than those with six or more episodes, 551 ng/L (IQR 298–2506), p=0.037.

NT-proBNP as a predictor for AF

Binary logistic regression unadjusted ORs for the likelihood of detecting new AF were non-significant for all clinical parameters included in the CHA2DS2-VASc score, whereas NT-proBNP was significantly associated with AF (OR per quartile 1.53 (1.24–1.88)). This association remained significant after adjustments (table 3). Adding height and weight in a second step of the multivariable analysis showed that for every increase of 1 kg in weight the OR was 1.03 (95% CI 1.01 to 1.05, p=0.004) and for height (1 cm increase) OR was 1.04 (1.0–1.09, p=0.08). The AUC for discrimination of newly detected AF for NT-proBNP alone was 0.64 (CI 0.57 to 0.71). Adding weight to NT-proBNP increased the AUC to 0.71 (CI 0.64 to 0.77) (figure 2).

Figure 2

ROC curve depicting detection of new AF as a function of weight and NT-proBNP. AF, atrial fibrillation; NT-proBNP, N-terminal pro B-type natriuretic peptide; ROC, receiver operating characteristic.

Table 3

Univariate and multivariable analysis using binary logistic regression for newly detected AF. For all categorical variables, the ODs show a change from a reference category (no present illness and male gender). For the continuous variables, height and weight, one unit change (1 cm or 1 kg) was used and for NT-proBNP quartiles

Using weight, height, history of hypertension, history of diabetes, heart failure and vascular disease, we made a model based on the Cohorts for Heart and Ageing Research in Genomic Epidemiology (CHARGE)-AF consortium's risk score for AF.13 Only weight was significantly associated with the detection of new AF, p=0.002. The AUC for this risk score was 0.64 (CI 0.58 to 0.71). Adding NT-proBNP to the score increased the AUC of the adjusted CHARGE-AF score to 0.71 (CI 0.65 to 0.77).

In the derivation group, ROC curves were used to determine a cut-off level for NT-proBNP for AF screening. A sensitivity of at least 75% was warranted, and the value of 125 ng/L was considered as an optimal level in this non-acute setting. In the validation group NT-proBNP using a cut-off of 125 ng/L resulted in a sensitivity of 75.5% and a specificity of 37% with a negative predictive value of 92%. Fewer individuals would have to undergo prolonged intermittent ECG screening if a cut-off of 125 ng/L was used. In our validation group, 147/415 (35%) would not have had to undergo prolonged screening. If Youden's index was used instead to find the optimal cut-off for NT-proBNP, it resulted in a NT-proBNP value of 290 ng/Ll. This cut-off (290 ng/L) had a sensitivity of 56% and specificity of 76%.

NT-proBNP levels are lower in obese individuals.14 This suggests that a weight adjusted NT-proBNP could be used to improve screening, as depicted in figure 3. Using weight adjusted NT-proBNP levels sensitivity in the whole cohort was 75%, specificity 45% and negative predictive value 90%. Using a cut-off based on weight, fewer individuals (288/495, 58%) would need to go through AF screening with intermittent ECG recordings.

Figure 3

Model of weight-adjusted NT-proBNP cut-off for AF screening using intermittent ECG recordings. AF, atrial fibrillation; NT-proBNP, N-terminal pro B-type natriuretic peptide. PPV, positive predictive value; NPV, negative predictive value.


This is the first systemic screening study for AF where a biomarker was added to the screening protocol to evaluate who might benefit from screening. Participants in the screening programme for AF who had newly detected AF had a significantly higher NT-proBNP than individuals without AF. An NT-proBNP cut-off of 125 ng/L in the non-acute setting provided a reasonable balance between specificity and sensitivity.

In cohort studies, NT-proBNP has been shown to be a marker of incipient AF,8 9 15 and in patients with ischaemic stroke/TIA higher levels of NT-proBNP was a predictor of subsequent detection of AF.16 In one pilot study NT-proBNP levels were higher in participants where AF was detected using 7-day Holter monitoring;17 These results are in accordance with the results in our study.

In the ARISTOTLE and RELY studies, patients with AF and a high level of NT-proBNP had an increased risk of stroke,18 19 although these studies only encompassed OAC treated individuals with AF. NT-proBNP has in a meta-analysis been associated with cardioembolic stroke.20 If screening for AF would be performed only in individuals with a raised level of NT-proBNP, some individuals with AF might be missed, but these might be individuals with a lower risk of stroke. This is in accordance with a proposed new biomarker-based risk score for stroke in patients with AF, where NT-proBNP is used to assess stroke risk.21 It might hence not be useful to screen individuals with a low NT-proBNP as their risk of stroke, should they have AF, could be low. Further studies are however needed to clarify the prognosis of these individuals with low NT-proBNP.

None of the clinical characteristics in the CHA2DS2-VASc score differed between the group with newly detected AF and the group without AF, which is at odds with prior studies.7 NT-proBNP levels have been reported to be higher in women than in men in a previous study,22 but in our study, we found no such difference between the genders. Our study did however only include 96 participants with newly detected AF, and the value of subgroup analyses could be limited due to a reduced power.

In subjects with higher BMI we detected more AF, which confirms the findings in earlier studies.23 Prior studies have shown an inverse relation between obesity and NT-proBNP concentrations.14 It could be assumed that some of the effect of AF on NT-proBNP has been diminished by the fact that participants with new AF weighed more, so the true difference in NT-proBNP levels between the groups could be even greater.

NT-proBNP and type of AF

Participants with arrhythmias discovered on the first ECG, ‘index ECG’ are likely to have persistent AF, and had significantly higher levels of NT-proBNP compared with individuals with paroxysmal AF episodes. These findings are in accordance with findings from the ARISTOTLE study.19 Participants with more episodes of AF during intermittent ECG recordings had higher NT-proBNP levels than participants with few episodes. The exact AF burden was not possible to assess, but participants with several registered AF episodes were more likely to have a high AF burden compared with participants with fewer registered AF episodes.

NT-proBNP as a predictor for development of AF

In this study, we propose using a NT-proBNP level of 125 ng/L as a cut-off for participants who could benefit from further screening with intermittent ECG. This cut-off was derived from visual inspection of the ROC curve, and from the clinical reasoning that sensitivity in this respect is more important than specificity as a first step in AF screening. Youden's index resulted in a higher NT-proBNP value of 290 ng/L as the optimal cut-off but did not prioritise sensitivity (56%), whereas specificity was higher at 76%. Hence, the use of Youden's index to find the optimal cut-off was deemed inappropriate from a clinical perspective. Using only NT-proBNP to screen for AF would yield a too low specificity to merit a life-long OAC therapy, and a confirmatory test with higher specificity for AF detection, like intermittent ECG registration,24 would be needed regardless. We therefore chose a cut-off with a sensitivity of 75.5%, even though specificity became low at 37%. With this cut-off, at most a quarter of new AF cases could be missed, probably in individuals with few episodes of AF where NT-proBNP may have been normalised by the time of screening.

This could potentially reduce the cost-effectiveness of screening; however, the stroke risk in these individuals has not been studied prospectively and needs to be assessed in future studies. However, using a NT-proBNP cut-off of 125 ng/L would mean that 35% fewer patients have to undergo prolonged screening using intermittent ECG thus reducing the cost of ECG equipment and ECG interpretation. The cost of a point-of-care blood test would be added to the expenditures. In total, the cost reduction was estimated to 800 SEK (=94 USD) per patient not needing to pass prolonged screening, which is a cost reduction of 35%.

In the European Society of Cardiology's Heart Failure guidelines, an NT-proBNP cut-off of 125 ng/L is used in the non-acute setting to rule out heart failure.25 Individuals with newly detected AF might have high NT-proBNP due to a strain on the heart from arrhythmic episodes or as a sign of cardiac pathology making a patient more prone to arrhythmia.

Risk scores for prediction of AF have been evaluated previously in several cohorts.13 26 The AUC for NT-proBNP was 0.64, which statistically is considered as low, although still similar to guideline recommended risk scores such as the CHA2DS2-VASc.21 27 However, if body weight was added to NT-proBNP, the AUC increased substantially.

The AUC of the modified CHARGE-AF model was 0.64 in our cohort compared with the published results of 0.76513 however, we did not use age in our model since all participants were of a similar age, which might explain the divergent results. We showed that NT-proBNP increased the AUC of the CHARGE-AF score.13 This increase in AUC using NT-proBNP could help discern subjects who are more likely to develop AF and thus will be more likely to benefit from participation in a screening programme.

Limitations and strengths

There are several limitations to our study. Using intermittent ECG recordings will only record the patient's heart rhythm <1 ‰ of the time, making underdiagnoses of AF likely. In prior studies using intermittent ECG recordings, this method found significantly more AF compared with 24 hours Holter monitoring.11 It is probable that more AF would be found using a continuous method but the benefit of ease of use, cost and patient comfort must also be considered in screening, and for these aspects intermittent ECG recordings has an advantage. Screening for AF using intermittent ECG recordings has also been shown to be cost-effective in a recent study.28

The sampling of NT-proBNP in only a subset of the entire cohort might introduce selection bias, however, as shown in table 1, there was no significant difference between the groups with new AF. There might also be potential confounders such as renal failure, as individuals with chronic renal failure have an increased risk of AF29 and higher NT-proBNP30 which we have not taken into account. Comorbidities were self-reported by the participants, which can cause misclassification bias.

A strength of our study is that all participants are 75/76 years old, hence, age differences should not influence levels of NT-proBNP as it might otherwise.

NT-proBNP is available as a point of care analysis and is already widely used in clinical practice for heart failure, so the clinical application is already in place and the method could easily be applied for AF screening. However, the proposed cut-off point of NT-proBNP of 125 ng/L needs to be validated externally before implementation in clinical practice. To evaluate this a new study, Strokestop2, has recently been initiated.


NT-proBNP levels were higher in participants with newly detected AF when screened by intermittent ECG as compared with participants where AF was not detected. Increased levels of NT-proBNP were found in participants with repeated episodes of AF and participants with AF at the index visit. Prospective studies are needed to clarify whether a cut-off based on NT-proBNP could be used to determine which individuals would benefit most from AF screening.

Key questions

What is already known about this subject ?

  • The cardiac biomarker N-terminal pro B-type natriuretic peptide (NT-proBNP) was associated with an increased risk of atrial fibrillation (AF) in several cohort studies.

What does this study add ?

  • This is the first time that NT-proBNP has been evaluated in a systemic screening programme for AF. Patients with newly detected AF had significantly higher levels of NT-proBNP compared with individuals where AF was not detected. Increased levels of NT-proBNP might thus indicate an increased risk of undetected AF. A NT-proBNP level of 125 ng/L in a non-acute setting yielded a balanced sensitivity-specificity cut-off.

How might this impact on clinical practise ?

  • Patients at risk of cardioembolic events might benefit from systematic screening for AF if they have an NT-proBNP level above 125 ng/L. These findings need to be confirmed in prospective studies.


The authors thank the entire KTA Prim staff, in particular: Maria Englund, Anna Hollander, Berit Källberg, Katarina Risbecker, Viktor Rollfelt, Katarina Fägerskiöld, Olga Beltzikoff and Pia Jaensson. At SHC Catrine Lindström. In Halland Lisbeth Andersson, Christine Palm, Eva Mellberg, Ida Hyltbäck, Anna Magnusson, Paola Lian, Emma Sandgren and Sofia Sandgren.



  • Contributor ES has been responsible for the design, initial draft and revisions of the manuscript, as well as the data collection and part of the analysis. All coauthors have been involved in the design of the study, have been critically reviewing the manuscript, analysed the data (mainly PH) and have been part of this revised manuscript. All authors have given final approval of the submitted version.

  • Funding This work was supported by Stockholm County Council, The Swedish Heart and Lung

    Foundation, King Gustav V and Queen Victoria's Free Masons’ Foundation, the Klebergska

    Foundation, Carl Bennet AB and the Tornspiran Foundation. Roche provided the point-of-

    care analysis apparatus but had no input into the scientific contents of the study.

  • Competing interests None declared.

  • Provenance and peer review Not commissioned; externally peer reviewed.