Diagnostic performance of two commercially available, PPG-based smartphone applications to detect atrial fibrillation

Abstract

Aims

Atrial fibrillation is the most common sustained cardiac arrhythmia, associated with considerable morbidity and mortality. Early detection is essential for optimal management. While wearable smart devices have shown promise, smartphone applications using photoplethysmography have also emerged as potential screening tools, but their diagnostic accuracy has not been adequately validated. This study aimed to compare the diagnostic performance of two commercially available photoplethysmography-based smartphone applications.

Methods and results

In this prospective multicentre study, consecutive patients from three cardiology centres underwent simultaneous recordings with the FibriCheck® and Preventicus Heartbeats® applications and a 12-lead electrocardiogram during hospital admission or outpatient evaluation. Electrocardiograms were independently interpreted by blinded cardiologists. Diagnostic accuracy, rate of inconclusive recordings, technical performance, features, and pricing were compared between the two applications. A total of 206 patients (median age 69.5 years, 64.1% male) with 216 electrocardiograms were analysed; atrial fibrillation was present in 43.1%. Successful photoplethysmography recordings were obtained in 97.7% of FibriCheck (mean attempts 1.2 ± 0.7) and 87.5% of Preventicus Heartbeats (1.6 ± 1.3) measurements. Sensitivity and specificity were 89.0% and 100.0% for FibriCheck, and 86.4% and 99.0% for Preventicus Heartbeats. Inconclusive reading rates (where the algorithm indicated a suspicion of mild arrhythmia) were 8.5% and 11.1%, respectively.

Conclusion

Both smartphone applications demonstrated high and comparable diagnostic accuracy for atrial fibrillation detection in a real-world population. These tools represent a viable, non-invasive option for early rhythm screening and follow-up, potentially improving patient outcomes, although further refinement is needed to reduce inconclusive recordings.

Graphical Abstract

Graphical Abstract.

What’s new?

• This is the first prospective, multicentre head-to-head study directly comparing the diagnostic accuracy, technical performance, and subscription models of two commercially available, photoplethysmography-based (PPG) smartphone applications for atrial fibrillation detection.

• Both applications demonstrated comparable and high diagnostic accuracy in a real-world clinical population, highlighting the potential of non-invasive PPG monitoring as a feasible and accessible alternative to conventional methods.

• Our findings provide novel evidence on the clinical relevance of inconclusive PPG recordings, showing that post-hoc physician review can enhance diagnostic yield, and thus inform future refinement of algorithmic design.

Introduction

Atrial fibrillation (AF) is the most common sustained cardiac arrhythmia and a major contributor to stroke, heart failure, and cardiovascular mortality.^1,2 Its lifetime risk is estimated to be approximately 25% after the age of 40 years,³ and its prevalence is projected to rise substantially in the coming decades.^1,4 Early and accurate detection of AF is essential for initiating timely therapeutic interventions and preventing associated complications.^2,5–7

Conventional monitoring tools, such as 12-lead electrocardiograms (ECGs) and Holter monitors, remain effective but are inherently limited by their intermittent recording and the lack of continuous availability. These limitations are particularly relevant in the case of paroxysmal arrhythmias, where the use of these methods may not be sufficient to detect transient episodes of AF. In recent years, technological advancements have led to the development of alternative, user-friendly solutions for AF detection, aiming to improve accessibility and long-term monitoring.⁸ Mobile health (mHealth) technologies developed for AF detection can be broadly classified into two main types: ECG-based and non-ECG-based.⁹ ECG-based devices, including smartwatches, have demonstrated high accuracy and clinical utility in identifying AF.^10–12 However, these devices are often costly and may have limited accessibility for the general population.

Non-ECG-based technologies for AF detection rely mainly on photoplethysmography (PPG), a non-invasive optical method that estimates heart rhythm from blood volume changes in the microvasculature.¹³ PPG-based diagnostic tools offer a more accessible alternative, requiring only a smartphone equipped with a light source and a camera compatible with specific applications.^13–16 These tools are more affordable, widely accessible, and some have received endorsement by the European Heart Rhythm Association (EHRA).⁹ However, real-world clinical evidence remains limited, and no head-to-head comparisons between these applications have yet been reported.

This study aims to evaluate and compare the diagnostic accuracy of two commercially available PPG-based smartphone applications for AF detection in a real-world clinical setting.

Methods

Study design and patient population

This study evaluated two mHealth smartphone applications, the FibriCheck® (Qompium NV, Hasselt, Belgium) and the Preventicus Heartbeats® (Preventicus GmbH, Jena, Germany), both listed by the 2022 EHRA position paper as commercially available [conformité européenne (CE) labelled] heart rhythm monitoring applications with prior use in clinical settings and peer-reviewed publications.⁹

This prospective, multicentre, non-randomized, open-label diagnostic study was conducted at three tertiary cardiology centres in Hungary: Cardiology Center—University of Szeged, Heart Institute—University of Pecs, and Gottsegen National Cardiovascular Center. The study enrolled consecutive patients from August 2023 to August 2024. Patients were screened during outpatient visits or upon hospital admission; the latter was often associated with various cardiac procedures, such as catheter ablation, electrical cardioversion, or invasive coronary angiography.

Eligible participants were patients aged 18 years or older who underwent a 12-lead ECG and consented to take part in the study. Exclusion criteria included age under 18 and the presence of a cardiac implantable electronic device (CIED), as the evaluated technology has not been validated for rhythm analysis of CIED patients, and AF with regular pacing may pose challenges for PPG-based detection. A minority of patients were screened twice (e.g. before and after electrical cardioversion). These patients were included only once in the descriptive analysis; however, each recording was analysed separately to evaluate the diagnostic performance of the applications.

Additionally, other relevant clinical data were obtained via comprehensive, detailed medical history interviews and review of electronic medical records.

The study was conducted in accordance with the principles of the Declaration of Helsinki and was approved by the Hungarian National Ethics Committee (ETT-TUKEB, No.: BM/24322-1/2023).

Study protocol

As part of the routine outpatient or hospital care, a standardized 12-lead ECG was obtained from each patient using an ECG recorder with standard settings (sweep speed: 25 mm/s, amplitude: 10 mm/mV). After acquiring the ECG, a test smartphone was handed to the patient for the PPG recording using both applications under investigation. The recording was initiated by the attending physician. PPG recordings were acquired immediately before or after the reference 12-lead ECG, typically within a few minutes, although the exact interval was not formally standardized. Patients were instructed to hold the phone steady for one minute during each recording session. The order of the applications was alternated between participants to avoid any learning effect. If the initial recording attempt failed, up to four additional attempts were allowed (maximum five attempts per application). Unsuccessful recordings were mostly due to low signal quality, or if excessive noise (e.g. hand movement) was detected by the application. After 5 unsuccessful attempts, the recording was deemed unsuccessful. Once successful recordings were obtained, the data were anonymized using unique patient identifiers. Both raw recordings and automatic analyses were subsequently saved in the corresponding application.

Study objectives

The primary objective of this study was to evaluate and directly compare the diagnostic accuracy of two commercially available, PPG-based smartphone applications in detecting AF, using cardiologist-interpreted 12-lead ECG as the reference standard.

Secondary objectives were to identify independent predictors of technically unsuccessful tests (defined as the inability to obtain a valid recording after five attempts), to quantify the proportion of inconclusive results, and to assess their diagnostic yield through post-hoc physician review. In addition, a comparative analysis of application features, subscription models, and pricing was performed based on information available from official product websites and mobile application stores (i.e. iOS App Store and Google Play Store).^17–22

ECG interpretation

All 12-lead ECGs corresponding to the PPG recordings were independently evaluated by two experienced cardiologists who were blinded to the PPG results. The consensual interpretations were used as the gold standard reference for determining the presence of AF or any other arrhythmia.

PPG recording analysis

While both applications classify results into three categories, the specific labels differ slightly. FibriCheck categorizes results as ‘regular rhythm’, ‘possibly irregular’, or ‘possible atrial fibrillation’, each accompanied by a colour code. A ‘regular heart rhythm’ (green) indicates sinus rhythm; ‘possibly irregular’ (yellow) suggests frequent irregular beats, typically due to ectopic or extra beats; and ‘possible atrial fibrillation’ (red) denotes absolute arrhythmia, most often AF. In comparison, Preventicus Heartbeats classifies recordings as ‘regular heart rhythm’ (green), ‘irregular heart rhythm’ (yellow), or ‘extremely irregular heart rhythm’ (red). Notably, while FibriCheck reports the presence of mild arrhythmia without further specification, Preventicus Heartbeats offers more detailed interpretations for intermediate (‘yellow’) results, such as frequent extrasystoles or regular tachycardia/bradycardia. In contrast, FibriCheck provides visual outputs, including the raw PPG signal, tachogram, and Lorenz plot, to support clinical interpretation. Figure 1 summarizes the classification of categories and visual outputs provided by each application.

Figure 1.

Examples of the three categories provided by each application.

Inconclusive results were defined as intermediate findings generated by the application despite a technically successful recording. These correspond to the ‘yellow’ category (examples shown in Figure 1) and were excluded from the primary diagnostic accuracy analysis to ensure a clear head-to-head comparison of the two algorithms. A post-hoc physician review was subsequently conducted to evaluate whether the clinical diagnosis could be reliably established using only PPG data. In this process, two blinded cardiologists independently evaluated the recordings. Concordant diagnoses were classified as correct and verified against the corresponding 12-lead ECG.

Statistical analysis

Data collection and anonymized storage were performed in Microsoft Excel (Microsoft Corporation, Redmond, WA, USA). For statistical analysis, IBM SPSS programme’s 26.0 version (International Business Machines Corporation, Armonk, NY, USA), DATAtab: Online Statistics Calculator (DATAtab e.U. Graz, Austria), and the Diagnostic Test Evaluation Calculator by MedCalc Software Ltd. (Acacialaan, 22, 8400, Ostend, Belgium) were used. A P-value <0.05 was considered statistically significant.

Baseline patient characteristics were summarized using descriptive statistics. Continuous variables were expressed as mean ± standard deviation (SD) or median with interquartile range (IQR), depending on the distribution, and categorical variables were reported as absolute numbers and percentages. ‘N’ refers either to the number of patients/ECGs/PPG recordings included in the analysis, as appropriate.

Standard diagnostic accuracy metrics—including sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), positive likelihood ratio (PLR), and negative likelihood ratio (NLR)—were calculated for each application, using the cardiologist’s interpretation of the standard 12-lead ECG as reference standard. Recordings that had been labelled as inconclusive were excluded from diagnostic accuracy analyses, since in these cases, the algorithm does not provide a clear diagnosis, prompting the user to further evaluate heart rhythm or seek professional help. Screening performance metrics (success rate, unsuccessful screening rate, inconclusive rate, mean number of attempts, and first-attempt success rate) were also analysed. The proportion of inconclusive results was calculated for each application, and their diagnostic yield was determined after a blinded cardiologist's re-evaluation.

Independent predictors of unsuccessful tests were identified using multivariate logistic regression. Variables with univariate association (P-value <0.1) were entered into the multivariate model, including demographic, clinical, echocardiographic, and laboratory parameters.

For consistency, results are presented in alphabetical order throughout the manuscript, starting with FibriCheck, followed by Preventicus Heartbeats.

Our study workflow is presented in Figure 2.

Figure 2.

Study workflow. Consecutive patients underwent 12-lead ECG and simultaneous PPG recordings using FibriCheck and Preventicus Heartbeats. AF, atrial fibrillation; ECG, electrocardiogram; PPG, photoplethysmography; N, number of patients, recordings, or ECGs included in the analysis, as appropriate.

Results

Demographic data

A total of 206 patients were included in the final analysis; one patient was excluded due to incomplete clinical information and unsuccessful PPG recordings with both applications (Figure 2). The cohort comprised predominantly male patients (n = 132, 64.1%), with a median age of 69.5 years (IQR: 62–75). AF was present in the medical history of 121 patients (58.7%), and 72 patients (35.0%) had a history of heart failure.

At the time of screening, the majority of patients (n = 176, 85.9%) were receiving beta-blocker therapy. The median left atrial diameter was 58 mm (IQR: 52–64), while the median left ventricular ejection fraction was 57.5% (IQR: 50–63).

Comprehensive demographic and clinical characteristics of the study population are presented in Table 1.

Table 1.

Baseline demographic and clinical characteristics

	n = 206
Male sex, n (%)	132 (64.1%)
Age, years, median (IQR)	69.5 (62–75)
Weight, kg, median (IQR)a	86 (73.5–100)
Height, cm, median (IQR)b	172 (164.3 -178)
BMI, kg/m2, median (IQR)c	29.2 (26.0–32.7)
Hypertension, n (%)	182 (88.4%)
Hyperlipidaemia, n (%)	128 (62.1%)
Atrial fibrillation in medical history, n (%)	121 (58.7%)
Coronary artery disease, n (%)	100 (48.5%)
Heart failure, n (%)	72 (35.0%)
Valvular disease, n (%)d	71 (34.6%)
Diabetes mellitus, n (%)	65 (31.6%)
Peripheral artery disease, n (%)	48 (23.3%)
Stroke/TIA, n (%)	13 (6.3%)
Deep vein thrombosis, n (%)	10 (4.9%)
Systemic embolization, n (%)	5 (2.4%)
Current smokers, n (%)e	59 (31.7%)
Left atrial diameter, mm, median (IQR)f	58 (52–64)
Left ventricular ejection fraction, %, median (IQR)g	57.5 (50–63)
End-diastolic diameter, mm, median (IQR)h	51 (47–56)
End-systolic diameter, mm, median (IQR)i	35 (30–39)
Intraventricular septum thickness, mm, median (IQR)j	11 (10–13)
LV lateral/free wall thickness, mm, median (IQR)k	10 (10–12)
eGFR, mL/min/1.73 m2, median (IQR)l	69 (53–81)
Haemoglobin, G/L, median (IQR)m	142 (126–151)
Pertinent medication
Beta-blocker, n (%)n	176 (85.9%)
ACEi/ARB/ARNi, n (%)o	161 (78.9%)
Oral anticoagulant, n (%)	132 (64.1%)
Statin, n (%)	132 (64.1%)
MRA, n (%)p	68 (33.2%)
Thrombocyte aggregation inhibitor, n (%)	77 (37.4%)
Anti-diabetic drug, n (%)	60 (29.1%)
SGLT2i, n (%)	33 (16%)
GLP-1 RA, n (%)	14 (6.8%)
Ivabradine, n (%)	1 (0.5%)

n = total number of patients included in the analysis; IQR, interquartile range; BMI, body mass index; TIA, transient ischaemic attack; LV, left ventricle; eGFR, estimated glomerular filtration rate; ACEi, angiotension receptor inhibitor; ARB, angiotension receptor blocker; ARNi, angiotensin receptor/neprilysin inhibitor; MRA, minerolacorticoid receptor antagonist; SGLT2i, sodium-glucose transporter type 2 inhibitor; GLP1-RA, glucose-like peptide-1 receptor antagonist.

^aAvailable information for 159 patients;

^bAvailable information for 158 patients;

^cAvailable information for 158 patients;

^dAvailable information for 205 patients;

^eAvailable information for 186 patients;

^fAvailable information for 190 patients;

^gAvailable information for 202 patients;

^hAvailable information for 186 patients;

ⁱAvailable information for 165 patients;

^jAvailable information for 144 patients;

^kAvailable information for 135 patients;

^lAvailable information for 173 patients;

^mAvailable information for 173 patients;

ⁿAvailable information for 205 patients;

^oAvailable information for 204 patients;

^pAvailable information for 205 patients.

Technical success rates of the PPG screenings

A total of 216 ECGs and their corresponding PPG recordings were analysed. Recordings were obtained on multiple smartphone models, including iPhone 14 Pro (n = 112), iPhone 13 Pro (n = 44), Nokia X20 (n = 32), Huawei Mate 20 (n = 24), and Xiaomi Mi 9T Pro (n = 4). FibriCheck was selected as the initial application in 53.2% of screenings, whereas Preventicus Heartbeats was used first in 46.8% of the cases. Inter-observer agreement for the ECG reference standard was 100%, with no discrepancies between the two cardiologists.

A greater portion of successful first-attempt recordings was achieved with FibriCheck compared to Preventicus Heartbeats (90.3% vs. 77.8%). The mean number of attempts required per patient to obtain a technically successful recording was lower with FibriCheck (1.2 ± 0.7) than with Preventicus Heartbeats (1.6 ± 1.3). Overall, 211 (97.7%) successful recordings were achieved with FibriCheck, and 189 (87.5%) with Preventicus Heartbeats. Detailed screening performance metrics are presented in Table 2.

Table 2.

Technical success rates and performance metrics of the PPG screening

	FibriCheck	Preventicus heartbeats
Application first used, n (%)	115 (53.2%)	101 (46.8%)
Successful screening on first attempt, n (%)	195 (90.3%)	168 (77.8%)
Successful screening rate, n (%)	211 (97.7%)	189 (87.5%)
Number of attempts, mean (SD)	1.2 (±0.7)	1.6 (±1.3)

n, total number of patients included in the analysis.

Recordings classified as unsuccessful (after repeated failed attempts as described in the Methods section) were further analysed for predictors using univariate and multivariate logistic regression of the baseline patient characteristics. A history of atrial fibrillation emerged as the only independent predictor of unsuccessful recordings (odds ratio (OR): 8.08, 95% confidence interval (CI): 1.6–40.76; P = 0.011), indicating a significantly higher likelihood of technical failure in this subgroup (Table 3).

Table 3.

Predictors of unsuccessful PPG recordings (by uni- and multivariate logistic regression)

Variable	Univariate analysis			Multivariate analysis
	Odds ratio	95% CI	P-value	Odds ratio	95% CI	P-value
Sex (male)	0.46	0.18–1.19	0.109
Age (/1 year)	1	0.97–1.04	0.867
First application used	0.83	0.37–1.86	0.658
Weight (/1 kg)	1	0.98–1.02	0.933
Height (/1 cm)	1.05	1.00–1.11	0.073	1.03	0.98–1.09	0.260
Body mass index (/1 kg/m2)	0.97	0.89–1.06	0.459
Hypertension	0.78	0.22–2.78	0.705
Hyperlipidaemia	2.04	0.91–4.53	0.082	3.59	0.97–13.37	0.057
Atrial fibrillation in medical history	3.41	1.24–9.35	0.017	8.08	1.6–40.76	0.011
Coronary artery disease	0.92	0.42–2.05	0.845
Heart failure	1.6	0.72–3.57	0.249
Valvular disease	2.4	1.08–5.36	0.032	1.27	0.29–5.64	0.753
Diabetes mellitus	3.1	1.38–6.96	0.006	2.26	0.65–7.88	0.201
Peripheral arterial disease	0.89	0.34–2.34	0.817
Stroke/TIA	0.5	0.06–3.97	0.511
Deep vein thrombosis	3.1	0.75–12.78	0.117
Systemic embolism	0	0 - ∞	0.988
Current smokers	1.14	0.5–2.61	0.752
Left atrial diameter (/1 mm)	1.04	0.99–1.09	0.112
LV ejection fraction (/1%)	0.95	0.93–0.98	0.001	0.98	0.92–1.04	0.518
LV end diastolic diameter (/1 mm)	1.06	1.01–1.11	0.025	1.01	0.92–1.12	0.792
LV end systolic diameter (/1 mm)	1.05	1.01–1.1	0.008	1.02	0.93–1.13	0.645
Interventricular septum (/1 mm)	0.89	0.69–1.15	0.356
LV lateral wall thickness (/1 mm)	1.07	0.78–1.45	0.691
eGFR (/1 mL/min/1.73 m2)	1	0.98–1.02	0.994
Haemoglobin (/1 G/L)	1.01	1–1.03	0.089	1.01	0.99–1.04	0.392
Beta blocker	0.46	0.1–2.05	0.307
ACEi/ARB/ARNi	0.76	0.27–2.11	0.593
Statin	1.02	0.45–2.34	0.963
Mineralocorticoid receptor antagonist	1.8	0.81–4.03	0.150
Thrombocyte-aggregation inhibitor	0.51	0.21–1.27	0.148
Anti-diabetic drug	1.86	0.83–4.19	0.134
SGLT2i	2.29	0.92–5.7	0.076	1.27	0.23–6.89	0.784

Statistically significant variables in the multivariate analysis are highlighted in bold.

PPG, photoplethysmography; CI, confidence interval; TIA, transient ischaemic attack; eGFR, estimated glomerular filtration rate; ACEi, angiotensin receptor inhibitor; ARB, angiotensin receptor blocker; ARNi, angiotensin receptor/neprilysin inhibitor; SGLT2i, sodium-glucose transporter type 2 inhibitor; LV, left ventricle/left ventricular.

Diagnostic accuracy metrics

Among the 216 analysed ECGs, AF was present in 43.1% and sinus rhythm in 56.9%, providing the reference standard for evaluating application performance.

Both PPG-based applications demonstrated high diagnostic accuracy. Sensitivity was 89.0% (95% CI: 80.2–94.9%) for FibriCheck, and 86.4% (95% CI: 75.7–93.6%) for Preventicus Heartbeats, while specificity was similarly high for both applications: 100.0% (95% CI: 96.7–100.0%) for FibriCheck and 99.0% (95% CI: 94.7–100.0%) for Preventicus Heartbeats. Additional diagnostic metrics are provided in Table 4 and Figure 3, confirming comparable overall diagnostic accuracy between the two applications. Similar results were obtained when only recordings in which both applications produced technically successful outputs were included (n = 161) (see Supplementary material online, Table S1). The diagnostic accuracy metrics of the intention-to-diagnose analysis, including inconclusive and technically unsuccessful recordings, are summarized in Supplementary material online, Table S2.

Table 4.

Diagnostic performance of FibriCheck and preventicus heartbeats applications to detect atrial fibrillation

	FibriCheck	95% CI	Preventicus heartbeats	95% CI
Sensitivity	89.0%	80.2–94.9%	86.4%	75.7–93.6%
Specificity	100.0%	96.7–100.0%	99.0%	94.7–100.0%
Positive predictive value	100.0%	95.1–100.0%	98.3%	89.0–99.8%
Negative predictive value	92.5%	86.9–95.8%	91.8%	85.9–95.4%
Positive likelihood ratio	−a	—	88.1	12.5–620.9
Negative likelihood ratio	0.1	0.1–0.2	0.1	0.1–0.3

CI, confidence interval

^aSince there were no false positives in this analysis, PLR could not be determined.

Figure 3.

Diagnostic accuracy of FibriCheck and Preventicus Heartbeats applications for atrial fibrillation detection. Sensitivity, specificity, and predictive values are shown with 95% confidence intervals (CIs) based on cardiologist-interpreted 12-lead ECG as the reference standard. Error bars represent 95% confidence intervals. ECG, electrocardiogram.

Because predictive values depend strongly on disease prevalence, we performed a Bayesian adjustment using our observed sensitivity and specificity values to estimate PPV and NPV across AF prevalences representative of community screening (around 4%).²³ FibriCheck’s diagnostic metrics remained consistently high (PPV = 100.0%, same as before and NPV 99.5% vs. 92.5% previously). However, under these lower-prevalence conditions, the modelled PPV decreased from 98.3 to 78.6% with Preventicus Heartbeats, while NPV remained consistently high (99.4% vs. 91.8% previously). The full prevalence-adjusted values for both applications are provided in Supplementary material online, Table S3.

An exploratory subgroup analysis was performed to assess whether diagnostic performance differed by the history of AF. Sensitivity was somewhat higher in participants with a history of AF (FibriCheck 91.0%; Preventicus Heartbeats 87.3%) compared with those without prior AF (FibriCheck 50.0%; Preventicus Heartbeats 66.7%), whereas specificity remained high across all groups (>98% for both applications). We present these results in Supplementary material online, Table S4.

Re-evaluation of inconclusive recordings

Both applications classified a subset of technically successful recordings as intermediate findings (‘yellow’ category), which were considered inconclusive in this study. Inconclusive results occurred in 18 cases (8.5%) with FibriCheck and 21 cases (11.1%) with Preventicus Heartbeats. All such recordings were re-evaluated by two independent, blinded cardiologists, as detailed in the Methods section.

Correct rhythm identification was achieved in 8 of 18 inconclusive FibriCheck recordings (44.4%) and in 11 of 21 Preventicus Heartbeats recordings (52.4%). For FibriCheck, inconclusive recordings (based on corresponding 12-lead ECGs), comprised sinus rhythm (n = 5), sinus rhythm with frequent premature ventricular complexes (PVCs; n = 7)), AF (n = 5), and one atypical atrial flutter (n = 1). Post-hoc PPG review enabled correct identification in 8 cases (2 sinus rhythm, 1 sinus rhythm with PVCs, 1 atrial flutter, and 4 AF). In the Preventicus Heartbeats group, inconclusive tracings showed sinus rhythm (n = 6), sinus rhythm with frequent PVCs (n = 9), AF (n = 5), and atrial flutter (n = 1). Of these, 11 were correctly classified on post-hoc review (6 sinus rhythm with PVCs, 3 sinus rhythm, and 2 AF).

A comprehensive summary of inconclusive findings is provided in Table 5.

Table 5.

Re-evaluation of inconclusive readings

	FibriCheck	Preventicus heartbeats
	n = 211	n = 189
Inconclusive recordings, n (%)	18 (8.5%)	21 (11.1%)
Successful rhythm identification of inconclusive recordings, n (%)	8 (44.4%)	11 (52.4%)

n, total number of patients or recordings included in the analysis.

Comparison of features and pricing options

Both applications are commercially available smartphone tools that operate on subscription-based models. While the exact pricing and access routes differ between the two vendors, both offer monthly and annual plans that allow unlimited use of the rhythm analysis features. Preventicus additionally provides an optional teleconsultation service, whereas FibriCheck includes expert review within its premium plans. These practical aspects may be relevant when selecting an application for clinical use or patient self-monitoring, particularly in settings where cost or provider-supported access influences uptake. A concise overview of pricing structures and access options is provided in Table 6.

Table 6.

Comparison of the two applications’ features and pricing

Option	FibriCheck (EUR)	Preventicus heartbeats (EUR)
Free version	—	€0
Monthly fee	€6.99 (Essential)	€6.99
	€24.99 (Premium)
Annual fee	€49.99 (Essential)	€33.99
	€129.99 (Premium)
Access via providera	€10 (7 days)	€0 (1 year)
	€15 (14 days)
	€25 (30 days)
Telecare services	Included in Premium	€35.99 (1 report)
		€44.99 (5 reports)

^aOptions may be available via healthcare providers, depending on local arrangements.

All prices are listed in euros (EUR) and include applicable taxes. Pricing reflects the available options in Germany as of June 2025.

Discussion

Main findings

In this prospective multicentre study, we conducted the first head-to-head clinical comparison of two commercially available, PPG-based smartphone applications for AF detection in a real-world population. Both applications demonstrated high diagnostic accuracy; FibriCheck achieved slightly higher diagnostic accuracy and superior technical success rates. Inconclusive recordings occurred in 8–11% of cases, and post-hoc physician review allowed correct classification in approximately half of these.

The study cohort had a median age of 69.5 years and included patients admitted for a variety of clinical reasons, thereby reflecting everyday practice. Screening was integrated into routine clinical care, underscoring the practicality of PPG-based applications as convenient, non-invasive tools for AF monitoring.^9,10,24,25

Unsuccessful recordings

The rate of unsuccessful recordings was relatively high, especially with Preventicus Heartbeats (2.3% vs. 12.5%). A history of AF was the only independent predictor of technical failure (OR 8.08, 95% CI: 1.6–40.76; P = 0.011), although the higher rate of unsuccessful recordings was most likely due to technological differences between the two applications. These findings suggest that targeted patient education—particularly in individuals with AF—together with repeated use of the applications may reduce failure rates. In addition, the longer (5-min) screening protocol of Preventicus Heartbeats warrants further evaluation as a potential means to improve recording success.

Importantly, all recordings in our study were performed with direct physician assistance and using test devices. Whether elderly patients, who are often less experienced with digital technologies, would be able to perform unsupervised screenings at home remains uncertain. This underlines the importance of assessing digital health literacy prior to large-scale implementation, as highlighted in the recent EHRA practical guide on digital devices.⁹

Diagnostic accuracy

Sensitivity was slightly higher with FibriCheck (89.0% vs. 86.4%), whereas specificity was nearly identical (100.0% vs. 99.0%). Most discrepancies were attributable to false negatives (n = 9 for both), while false positives were rare (n = 1 for Preventicus Heartbeats and n = 0 for FibriCheck). The clinical relevance of these minor differences remains uncertain. Notably, the sensitivity and specificity values in our study were somewhat lower than those reported in previous PPG trials requiring daily repeated measurements,¹⁵ suggesting that user familiarity and repeated use may improve diagnostic performance.

Because technical success rates differed between the two applications, direct comparison of diagnostic accuracy requires caution. To allow like-for-like comparison, we evaluated diagnostic accuracy within the subset of recordings in which both applications produced technically successful outputs. Accuracy estimates in this paired dataset were similar to those in the primary analysis, although the reduced sample size resulted in wider confidence intervals (see Supplementary material online, Table S1).

To account for the potential inflation of accuracy metrics when inconclusive and technically unsuccessful recordings are excluded, we also performed an intention-to-diagnose analysis in which all such outputs were classified as diagnostic failures. Under this conservative approach, diagnostic accuracy metrics proved to be lower than in our primary analyses (see Supplementary material online, Table S2).

When interpreting these findings, it is important to consider that several characteristics of our study population and recording environment may influence generalizability. The prevalence of atrial fibrillation was substantially higher than in community-level screening,²³ and most participants were recruited in a hospital setting. This higher prevalence may influence the observed diagnostic accuracy metrics²⁶ and diagnostic accuracy, especially positive and negative predictive values derived from this population, may not directly extrapolate to low-prevalence screening settings. Therefore, our results may be most applicable to secondary-care or higher-risk groups rather than to unselected population screening.

Furthermore, all recordings were obtained under direct supervision, which likely yielded better signal quality than unsupervised home use. These factors may partially inflate observed diagnostic performance and should be taken into account when extrapolating our results to real-world, community-based AF screening.

Post-screening analysis of tracings

Inconclusive recordings occurred in 8.5% of FibriCheck and 11.1% of Preventicus Heartbeats screenings. Upon blinded physician review, correct rhythm classification was achieved in 44.4% and 52.4% of these cases, respectively. Preventicus Heartbeats additionally provides algorithm-based descriptions of mildly arrhythmic (‘yellow’) tracings—such as frequent extrasystoles or regular tachycardia—which FibriCheck does not (Figure 1).

When compared with the BASEL Wearable Study, the proportion of inconclusive tracings with PPG-based smartphone applications (8–11%) was not higher than that reported for several smartwatches (17–26%).¹⁰ However, while inconclusive smartwatch recordings could be successfully adjudicated in 95% of cases, the yield of physician reassessment in our study remained substantially lower, at around 50%. This underscores the need for further refinement of PPG algorithms and highlights an important difference between ECG- and non-ECG-based technologies in terms of post-hoc interpretability.

The operational definitions of ‘inconclusive’ also differ between the two applications (Figure 1), which complicates the comparison of their inconclusive output rates. Moreover, the fact that only approximately half of inconclusive tracings could be assigned a rhythm diagnosis during post-hoc physician adjudication indicates that a substantial proportion of PPG signals are inherently uninterpretable due to insufficient waveform quality. This limitation appears independent of the specific algorithm used and reflects a fundamental constraint of PPG-based rhythm assessment in certain recording conditions.

Potential role in AF management

The use of smartphone applications for AF detection has the potential to significantly impact the management of patients with paroxysmal AF. Consistent with prior studies demonstrating high sensitivity of smartwatch-based algorithms compared to insertable cardiac monitors (ICM),^14,24 our results support the feasibility of PPG-based applications for long-term, non-invasive rhythm surveillance. Such tools may reduce the need for frequent ambulatory visits, improve resource allocation, and enhance patient engagement.

Recent data also suggest the feasibility of repeated, on-demand smartphone monitoring after AF ablation,^27,28 which may be especially relevant in patients with an intermediate thromboembolic risk profile (CHA₂DS₂-VA = 1). According to the latest European Heart Rhythm Association (EHRA)/Heart Rhythm Society (HRS)/Asia-Pacific Heart Rhythm Association (APHRS)/Latin-America Heart Rhythm Association (LAHRS) consensus statement on catheter and surgical ablation of atrial fibrillation, anticoagulation discontinuation may be considered 12 months post-ablation in the absence of documented recurrence, provided that reliable long-term rhythm monitoring is available.²⁹ This approach, however, slightly contrasts with the 2024 European Society of Cardiology (ESC) guideline on the management of atrial fibrillation, which recommends continuation of anticoagulation according to the patient’s CHA₂DS₂-VA score and not the perceived success of the ablation procedure.² The findings of a pilot study support this previous notion, demonstrating that targeted anticoagulation for AF based on continuous rhythm assessment with an insertable cardiac monitor can be effectively managed.¹⁴ By extension, PPG-based smartphone applications may offer a more accessible alternative. The ongoing prospective randomized REACT-AF trial, evaluating intermittent vs. continuous administration of direct oral anticoagulants (DOACs) guided by smartwatch-detected AF episodes, may further clarify the clinical value of such an approach.³⁰ Beyond anticoagulation management, early and consistent detection of paroxysmal AF episodes could also prevent subsequent complications such as worsening or acute decompensated heart failure.² Moreover, this benefit extends beyond paroxysmal AF and may also support patients with permanent AF by allowing closer rhythm surveillance and facilitating more effective rate-control management.

The recent EHRA practical guide also emphasizes the growing acceptance role of mobile health technologies in AF care.⁹ Integrating PPG-based smartphone applications like FibriCheck and Preventicus Heartbeats into routine practice could support a more proactive, guideline-aligned approach to screening and follow-up. Compared with smartwatches, these applications are more widely accessible, as they require only a smartphone camera rather than dedicated hardware.^10,17,18 This lower cost and broader availability may reduce barriers to implementation and facilitate large-scale rhythm monitoring in both clinical and home environments. However, this technology is not only useful as a stand-alone tool but also as a complementary approach. As demonstrated in the Apple Heart Study, the Apple Watch combines passive arrhythmia detection via photoplethysmography with an on-demand single-lead ECG. When irregular rhythms are detected through PPG, the users receive a notification, which can prompt them to record a confirmatory ECG. This dual approach enhances the opportunity for atrial fibrillation screening.^31,32

Limitations

Despite their potential for widespread clinical use, PPG-based smartphone applications have several technological limitations. Most importantly, they cannot reliably differentiate normofrequent regular tachycardias from sinus rhythm, since the technology is based solely on heart rate variability. Furthermore, although a post-hoc physician review of inconclusive tracings is possible, our findings indicate that such interpretation may be challenging and time-consuming. Additional barriers include the predominance of AF in elderly patients—who often have limited digital literacy and access to smartphones—as well as the risk of false positive results in populations with low AF prevalence, emphasizing the need of careful integration of PPG results into the overall diagnostic work-up and clinical decision making.

Some limitations of our study should also be highlighted. First, all screenings were performed under direct assistance from medical staff and using standardized test devices. These supervised conditions likely resulted in higher signal quality than would be expected during unsupervised home use, where lighting, motion artefacts, and variability in user technique are common. Therefore, the diagnostic performance observed in this study may overestimate real-world accuracy. The time interval between PPG and ECG recordings was not strictly standardized and, although typically only a few minutes, theoretically allows for rhythm changes—particularly in patients with paroxysmal atrial fibrillation. Multiple smartphone models were used, and hardware differences—particularly cameras and LEDs—may affect PPG signal. Another important limitation is the potential for spectrum bias. Because most participants were recruited during planned electrophysiology or cardiology procedures, many had structural heart disease, rhythm irregularities, or prior arrhythmia history. These factors may modify PPG waveform characteristics and differ substantially from those seen in community screening users. Therefore, the diagnostic accuracy observed in our mainly hospital-based cohort may not fully reflect real-world algorithm performance in unselected, lower-risk populations. Also, no learning phase was included, and initial attempts were also counted, potentially inflating the number of unsuccessful recordings. Moreover, patients with cardiac implantable devices were excluded, limiting the generalizability of our findings to this population. Notably, not all participants could be evaluated in a direct head-to-head comparison, as screening was unsuccessful with only one of the applications in certain cases. This may have introduced a degree of verification bias, limiting the precision of within-subject performance estimates. Finally, although our population also included patients with arrhythmias other than AF (e.g. regular tachycardias and ventricular and supraventricular extrasystoles), in cohorts with a higher prevalence of these conditions, the diagnostic efficacy of the tools demonstrated in our study may be reduced.

Conclusions

In this prospective head-to-head, multicentre study, we demonstrated that both commercially available, PPG-based smartphone applications—FibriCheck and Preventicus Heartbeats—achieved comparable and high diagnostic accuracy for detecting AF in a real-world clinical population. FibriCheck showed higher technical success and had a slightly higher diagnostic accuracy, while Preventicus Heartbeats had a somewhat higher diagnostic yield upon re-evaluation of inconclusive recordings.

Although these tools represent accessible and non-invasive options for rhythm monitoring, challenges remain, particularly regarding unsuccessful recordings and the limited interpretability of inconclusive tracings. Further algorithmic refinement, patient education, and clarification of their role within clinical workflows are warranted. Overall, PPG-based smartphone applications hold promise for routine AF screening and follow-up, with the potential to improve patient engagement and healthcare efficiency. However, our results primarily reflect secondary-care and higher-risk populations. Extrapolation to population-level AF screening should therefore be made with caution and warrants evaluation in larger, unselected cohorts.

Supplementary material

Supplementary material is available at European Heart Journal – Digital Health.

Author contributions

Mihály Dániel Szőnyi (MD (Conceptualization [supporting]; Data curation [lead]; Formal analysis [lead]; Investigation [equal]; Methodology [supporting]; Project administration [equal]; Resources [equal]; Supervision [supporting]; Validation [equal]; Visualization [equal]; Writing—original draft [lead]; Writing—review & editing [equal])), Flora Diana Gausz (MD (Conceptualization [supporting]; Data curation [equal]; Formal analysis [supporting]; Investigation [equal]; Methodology [supporting]; Project administration [equal]; Resources [supporting]; Supervision [supporting]; Validation [supporting]; Visualization [supporting]; Writing—original draft [supporting]; Writing—review & editing [equal])), Botond Bocz (MD (Data curation [equal]; Formal analysis [supporting]; Investigation [equal]; Project administration [supporting]; Validation [supporting]; Visualization [supporting]; Writing—original draft [supporting]; Writing—review & editing [equal])), David Pilecky (MD (Conceptualization [supporting]; Data curation [supporting]; Investigation [supporting]; Methodology [supporting]; Supervision [equal]; Validation [supporting]; Writing—original draft [supporting]; Writing—review & editing [equal])), Balazs Muk (MD (Conceptualization [supporting]; Investigation [supporting]; Methodology [supporting]; Supervision [equal]; Validation [supporting]; Visualization [supporting]; Writing—original draft [supporting]; Writing—review & editing [equal])), Fanni Banfi-Bacsardi (MD (Data curation [supporting]; Formal analysis [supporting]; Investigation [supporting]; Methodology [supporting]; Software [supporting]; Supervision [equal]; Validation [supporting]; Writing—original draft [supporting]; Writing—review & editing [equal])), Csaba Foldesi (MD (Conceptualization [supporting]; Data curation [supporting]; Investigation [supporting]; Methodology [supporting]; Project administration [supporting]; Resources [supporting]; Supervision [equal]; Validation [supporting]; Writing—original draft [supporting]; Writing—review & editing [equal])), Peter Andreka (MD (Conceptualization [supporting]; Investigation [supporting]; Project administration [supporting]; Supervision [supporting]; Writing—original draft [supporting]; Writing—review & editing [equal])), Tamas Szili-Torok (MD (Conceptualization [supporting]; Funding acquisition [supporting]; Investigation [supporting]; Methodology [supporting]; Project administration [supporting]; Resources [supporting]; Supervision [equal]; Validation [supporting]; Writing—original draft [supporting]; Writing—review & editing [equal])), Peter Kupo (MD (Conceptualization [lead]; Data curation [lead]; Formal analysis [supporting]; Funding acquisition [supporting]; Investigation [lead]; Methodology [lead]; Project administration [lead]; Resources [equal]; Supervision [lead]; Validation [equal]; Visualization [supporting]; Writing—original draft [lead]; Writing—review & editing [equal])), and Mate Vamos (MD (Conceptualization [lead]; Data curation [equal]; Formal analysis [equal]; Funding acquisition [lead]; Investigation [lead]; Methodology [lead]; Project administration [lead]; Resources [lead]; Supervision [lead]; Validation [lead]; Visualization [equal]; Writing—original draft [lead]; Writing—review & editing [equal]))

Funding

The authors declare that they have not received additional funding for the conduct of this study. However, open access publication of this article was supported by the University of Szeged (Grant ID: 8202).

T.S.-T. reports an educational activity/advisory relationship with Biotronik, and until 2022, he was in educational/advisory/product development relationships with Ablacon, Inc., Abbott, Acutus Medical, BiosenseWebster, and Stereotaxis.

P.K. reports educational activities and research grants on behalf of Abbott and Biosense Webster, and educational activities supported by Boston Scientific/TwinMed, Novo Nordisk, and Pfizer, outside the submitted work.

Data availability

All data used in this study are available by reasonable request from the corresponding author.