Impact of physician interpretation of at-home ECG on clinically-detected AF

1. Introduction

Current guidelines state that definite diagnosis of Atrial fibrillation (AF) in screen-positive cases is established only after physician reviews the single-lead ECG recording of > 30 s and confirms that it shows AF. [1], [2], so a definitive diagnosis of AF requires a physician's diagnosis using a 12-lead electrocardiogram (ECG), Holter ECG, or patch ECG etc. One reason for this is the risk of false-positive results for possible AF [3]. However, the impact of a physician's reading of possible AF (true positive or false positive) on the future detection of clinically-detected AF was not known.

2. Methods

The multicenter observational trial design has previously been described [4]. In brief, patients were selected from six cardiovascular centers in Japan. Patients with persistent AF scheduled for ablation between May 2019 and December 2020 were enrolled and 94 patients were included in the final analysis set. All patients underwent extensive encircling pulmonary vein isolation (PVI) and were instructed to record at least one ECG daily after discharge. ECG measurements were made using a blood pressure monitor with ECG (The Complete; Omron Healthcare Co., Ltd., Kyoto, Japan) [3]. Patients were followed up for 1 year at the outpatient clinic, with 12-lead ECG and 24-hour Holter ECGs performed every 3 months until after 12 months. The Complete was not available in Japan during the study. Therefore, physicians made all clinical decisions based on their usual practice, and physicians and patients were blinded to Complete results.

“Clinically-detected AF” was defined as AF documented by a 12-lead ECG or 24-hour Holter ECG in the clinical practice during the one-year follow-up. “Possible AF” was defined as AF documented by an ECG rhythm strip of the Complete, and it has duration of ≥ 30 s. The definition of “ physician confirmed AF” was based on two trained cardiologists (TN and SS) who read the ECGs of “Possible AF” in a blinded fashion and classified those with AF (true positive) as “ physician confirmed AF” and those without AF (false positive) as “not AF (Fig. 1).

Fig. 1.

Physician interpretation of the Complete.

The sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV) were calculated by comparing the proportion of clinically-detected AF vs Possible AF and clinically-detected AF vs physician confirmed AF, and significant differences were calculated by McNemar's test. Data were presented as number (%) or percentage (95 % confidence interval [CI]). The odds ratios (OR) for AF recurrence were calculated using multivariable logistic regression analyses. The explanatory variables were sex, age, congestive heart failure, hypertension, diabetes mellites, stroke, vascular disease, body mass index, duration of AF, number of ablation, antiarrhythmic agents use. All analyses were carried out using R (version 4.3.0; April 21, 2023).

3. Results

The mean age was 64.9 ± 8.8 years, 18.1 % of the patients were female, 50.0 % had hypertension, 44.7 % had long-term persistent AF (≥1 year), and the mean CHADS2 score was 1.0 ± 0.9. A median follow-up was 12.0 months (range 7.1–12.3 months), and the mean number of ECG measurements during the follow-up were 326.0 ± 300.6 times in the Complete. At 12 months, 18 AF recurrences were detected clinically. Table 1 shows that a large proportion of patients with recurrent clinically-detected AF have possible AF on at-home ECG (n = 15) (sensitivity 83 %). However, the specificity was low (37 %), increasing the risk of false positives, with a PPV of only 24 %. On the other hand, in terms of physician confirmed AF, specificity improved to 80 % and PPV increased to 50 % while sensitivity remained at 83 %. NPV did not change much, from 90 % to 95 %, with or without physician confirmation. Logistic regression analysis of recurrent clinically-detected AF showed that stroke (OR; 1.34, 95 % CI; 1.09–1.64, p < 0.01) were significantly associated, while possible AF was not (OR;1.15, 95 %CI; 0.96–1.38, p = 0.12). On the other hand, when the explanatory variable was changed from Possible AF to physician confirmed AF, physician confirmed AF was significantly associated with recurrent clinically-detected AF (OR;1.48, 95 %CI; 1.26–1.75, p < 0.01). No other significantly associated factors were found.

Table 1.

Association of “Possible AF” vs ”clinically-detected AF” and “physician confirmed AF” vs “clinically-detected AF”.

variable	Possible AF			Physician confirmed AF
	(+)	(-)	P-value	(+)	(-)	P-value
Clinically-detected AF
(+)	15 (16)	3 (3)		15 (16)	3 (3)
(-)	48 (51)	28 (30)		15 (16)	61 (65)
			<0.01*			<0.01**
Sensitivity	83 (64–94)			83 (64–94)
Specificity	37 (32–39)			80 (76–83)
Positive predictive value	24 (18–27)			50 (38–56)
Negative predictive value	90 (79–97)			95 (90–98)

Data are presented as number (%) or percentage (95% confidence interval). PPV, positive predictive value; NPV, negative predictive value.

*Significant difference between possible AF and clinically-detected AF based on Fisher exact test.

**Significant difference between physician confirmed AF and clinically-detected AF based on Fisher exact test.

4. Discussion

We showed for the first time the significance of physician assessment of at-home ECGs in the post-ablation diagnostic workflow. While Possible AF was less associated with the incidence of future clinically-detected AF due in part to the presence of false positives, physician assessment of Possible AF reduced false positives and was associated with the incidence of future clinically-detected AF. This has led to physician confirmed AF predicting future clinically-detected AF. If more than one year has passed since the ablation, outpatient follow-up is less frequent. However, AF recurrence is increasing every year [5]. Screening with at-home ECG may particularly be valuable in these cases. However, the accuracy and reliability need to be carefully considered. The accuracy of the technology itself needs to be improved in order to achieve reduced false positives. Machine learning and artificial intelligence (AI) could then be used to assist in the ECG analysis process. The introduction of telemedicine, where patients measure ECGs at home and share the data with physicians in remote areas, could also reduce the burden on physicians and make medical treatment more efficient, as they can check the data when necessary and intervene as required. A high negative predictive value also means a high ability to judge non-recurrent patients as 'not recurrent'. This may therefore reduce unnecessary additional testing for patients who continue to show negative results with at-home ECGs, allowing healthcare resources to be focused on higher-risk patients. In conclusion, patients whose physicians assessed 'possible AF' as AF had an approximately 1.5 times higher risk of recurrence of clinically-detected AF within one year.

5. Source of funding

The Omron Healthcare company provided Complete device and an iPhone (Apple Inc., CA, USA) for this study but was not involved in the study design, analysis, or drafting of the manuscript.

6. Disclosures

Dr. K Senoo had university research contracts with the Omron Healthcare company. The other authors have declared no conflicts of interest.

7. Ethics approval statement

The study was approved by the Medical Ethics Review Committee of the Kyoto Prefectural University of Medicine (approval number: ERB-C-1347).

CRediT authorship contribution statement

Keitaro Senoo: Writing – review & editing, Writing – original draft, Investigation, Funding acquisition, Data curation, Conceptualization. Arito Yukawa: Project administration, Data curation. Hibiki Iwakoshi: Project administration, Data curation. Tetsuro Nishimura: Project administration, Data curation. Satoshi Shimoo: Supervision, Formal analysis. Keiji Inoue: Project administration, Data curation. Tomohiko Sakatani: Project administration, Data curation. Ken Kakita: Project administration, Data curation. Tetsuhisa Hattori: Project administration, Data curation. Hiroki Kitajima: Project administration, Data curation. Kentaro Nakai: Project administration, Data curation. Suguru Nishiuchi: Project administration, Data curation. Mitsuko Nakata: Formal analysis. Satoshi Teramukai: Supervision, Formal analysis. Hirokazu Shiraishi: Supervision. Satoaki Matoba: Supervision, Conceptualization.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.