A portable prototype magnetometer to differentiate ischemic and non-ischemic heart disease in patients with chest pain

Abstract

Background

Magnetocardiography (MCG) is a non-invasive technique used to measure and map cardiac magnetic fields. We describe the predictive performance of a portable prototype magnetometer designed for use in acute and routine clinical settings. We assessed the predictive ability of the measurements derived from the magnetometer for the ruling-out of healthy subjects and patients whose chest pain has a non-ischemic origin from those with ischemic heart disease (IHD).

Methods

MCG data were analyzed from a technical performance study, a pilot clinical study, and a young healthy reference group. Participants were grouped to enable differentiation of those with IHD versus non-IHD versus controls: Group A (70 IHD patients); Group B (69 controls); Group C (37 young healthy volunteers). Scans were recorded in an unshielded room. Between-group differences were explored using analysis of variance. The ability of 10 candidate MCG predictors to predict normal/abnormal cases was analyzed using logistic regression. Predictive performance was internally validated using repeated five-fold cross-validation.

Results

Three MCG predictors showed a significant difference between patients and age-matched controls (P<0.001); eight predictors showed a significant difference between patients and young healthy volunteers (P<0.001). Logistic regression comparing patients with controls yielded a specificity of 35.0%, sensitivity of 95.4%, and negative predictive value for the ruling-out of IHD of 97.8% (area under the curve 0.78).

Conclusion

This analysis represents a preliminary indication that the portable magnetometer can help rule-out healthy subjects and patients whose chest pain has a non-ischemic origin from those with IHD.

About the Authors:

Shima Ghasemi-Roudsari

Roles Conceptualization, Methodology, Visualization, Writing – original draft

Current address: Creavo Medical Technologies Ltd., Coventry, United Kingdom

Affiliation: Department of Physics and Astronomy, University of Leeds, Leeds, United Kingdom

Abbas Al-Shimary

Roles Formal analysis, Methodology, Software, Validation, Visualization, Writing – original draft

* E-mail: [email protected]

Current address: Creavo Medical Technologies Ltd., Coventry, United Kingdom

Affiliation: Department of Physics and Astronomy, University of Leeds, Leeds, United Kingdom

ORCID logo http://orcid.org/0000-0002-5608-8174

Benjamin Varcoe

Roles Conceptualization, Data curation, Formal analysis, Project administration, Supervision, Visualization, Writing – review & editing

Affiliation: Department of Physics and Astronomy, University of Leeds, Leeds, United Kingdom

Rowena Byrom

Roles Investigation

Affiliation: Division of Cardiovascular and Diabetes Research, Leeds Institute of Cardiovascular and Metabolic Medicine, University of Leeds, Leeds, United Kingdom

Lorraine Kearney

Roles Investigation

Affiliation: Division of Cardiovascular and Diabetes Research, Leeds Institute of Cardiovascular and Metabolic Medicine, University of Leeds, Leeds, United Kingdom

Mark Kearney

Roles Conceptualization, Data curation, Investigation, Project administration, Supervision, Visualization, Writing – review & editing

Affiliation: Division of Cardiovascular and Diabetes Research, Leeds Institute of Cardiovascular and Metabolic Medicine, University of Leeds, Leeds, United Kingdom

Introduction

Distinguishing between acute coronary syndrome (ACS) and non-cardiac chest pain is a major challenge in the emergency department (ED) [1]. Current diagnostic tests for patients presenting with chest pain are time-consuming and involve an electrocardiogram (ECG), as well as serial blood biomarker testing [2]. Yet, almost 75% of patients with chest pain of a non-cardiac origin are triaged through this pathway [3–5]. The ability to rule-out ACS earlier would greatly improve patient care and hospital resource utilization [5]. Furthermore, current diagnostics fail to detect up to 2% of non-ST-elevated myocardial infarction (NSTEMI) or unstable angina patients, who are inappropriately discharged from hospital [6,7].

Electrical activity in the heart simultaneously produces a magnetic field [8], which can be detected non-invasively using magnetocardiography (MCG) [9]. MCG is emission-free and of very low risk to the patient [10]. Since magnetic fields are not attenuated by bodily tissue or fluids, MCG is more sensitive than an ECG to very weak cardiac signals [11–13]; MCG can detect ischemia-induced deviations in depolarization and repolarization with greater accuracy than an ECG [14]. Substantial improvements in the diagnosis of ischemic events, such as NSTEMI, which are difficult to detect by an ECG, have already been demonstrated with MCG [13,15].

Superconducting Quantum Interference Device magnetometers have been widely used in cardiology and neurology research [12,13]. However, they require separation from sources of background interference, liquid helium cooling, highly trained operators, and have high running costs. A portable prototype magnetometer was recently designed to overcome the limitations of existing technology [16]. Its novel coil configuration allows high sensitivity detection with negligible background magnetic field interference [16]. The magnetometer can be used at the patient’s bedside, making it suitable for use in acute and routine clinical settings. A technical performance study investigated the predictive performance and reliability of candidate predictors obtained from a broad spectrum of ischemic heart disease (IHD) patients and age-matched healthy volunteers (S1 Text). The results showed an overall good discrimination performance, with generally good to excellent reliability for the derived measurements.

In the current study, we assessed predictive ability of the measurements derived from the magnetometer for the rule-out of healthy subjects and patients whose chest pain has a non-ischemic origin from those with IHD.

Materials and methods

Data collection

MCG data were collected from two studies. A technical performance study was conducted in patients (≥25 years) with suspected IHD and in healthy age-matched volunteers (S1 Text). A subgroup of patients with NSTEMI within 3 days of enrollment was also recruited. Patients were recruited from cardiology clinics at Leeds General Infirmary. Patients were identified from medical records and approached by the Clinical Investigator during a clinic appointment or on the hospital ward if the patient was an in-patient. Healthy volunteers were recruited by poster advertising in the cardiology clinic.

A pilot clinical study was conducted in NSTEMI patients (≥18 years) requiring admission for chest pain (troponin ≥50 ng/L 12 h post-onset of chest pain), and in a control group of non-IHD (NIHD) patients experiencing chest pain (ClinicalTrials.gov, number: NCT02359773, available at: https://www.clinicaltrials.gov/ct2/show/NCT02359773). Patients with ST-elevated myocardial infarction, hemodynamic instability, or revascularization were excluded from this cohort. NIHD patients (≥18 years) were eligible if they had a referral due to chest pain in the last 8 weeks and a negative magnetic resonance imaging scan, Myoview, or a stress echocardiogram within 4 weeks of the first MCG scan. Patients with symptoms of ACS or troponin ≥50 ng/L were excluded from this cohort.

Patients with a pacemaker or pregnant/lactating women were excluded. Additional exclusion criteria included: internal cardiac defibrillator or active implantable device; metal implants in the torso; and any comorbidity that prevented patients from being scanned. These studies were approved by the relevant Institutional Review Board and all clinical investigations were conducted according to the principles expressed in the Declaration of Helsinki. Written informed consent was obtained prior to any study-related procedures. The technical performance study was approved by the National Research Ethics Service (NRES) Committee Yorkshire & The Humber—Leeds Central, with Ethics Certificate number 12/YH/0562. The pilot clinical study was registered (protocol NCT02359773) on ClinicalTrials.gov., and approved by the NRES Committee Yorkshire & The Humber—Leeds Central, with Ethics Certificate number 14/YH/1222.

An additional data set was collected from young (<30 years) healthy volunteers as a reference standard for a normal scan. Young healthy volunteers were recruited by poster advertising at the University of Leeds.

Data were grouped to allow differentiation of patients from controls: Group A (IHD patients), Group B (controls), and Group C (young healthy volunteers).

MCG recordings

Scans were carried out for 10 min in a supine position in an unshielded room. Participants removed upper garments and wore a hospital gown to minimize potential interference. For NSTEMI patients, the scan was conducted within 48 h of the onset of chest pain and before coronary angioplasty or other surgical intervention; for NIHD patients, the scan was completed within 4 weeks of a negative functional test for IHD.

MCG signals were baseline-corrected and averaged, centering on the R wave peak, to increase the signal-to-noise ratio. A custom Python program was used to extract candidate predictor estimates from the MCG data. The signal processing software also performed virtual gradiometry and frequency domain filtering [16]. Ten MCG predictors were measured from the ventricular depolarization phase (S1 Table; S1 Fig) [15,17,18].

Statistical analysis

Analysis of variance was used to compare the candidate MCG predictors between patients and controls. Regularized logistic regression modeling [19] was employed to study the relationship between the response variable (control vs patient) and the 10 candidate MCG predictors. Model 1 was fitted on all available data with both Group B and Group C as controls. Model 2 included Group B only as a control, and Model 3 included Group C only as a control. The patient group (Group A) was the same across all three models.

Several measures of predictive performance were reported [19]. The discriminatory ability of the models to distinguish patients and controls was evaluated using the area under the receiver operator characteristics curve (AUC). The AUC varies between 0.5 and 1.0, with a higher value indicating better performance [20].

The ability of the models to rule-out subjects was assessed via sensitivity, specificity, and negative predictive value (NPV) using a cut-off on the receiver operator characteristics at which sensitivity exceeded 90%. These are defined in the confusion matrix given in Table 1.

[Figure omitted. See PDF.]

Table 1. Confusion matrix for a binary classifier.

https://doi.org/10.1371/journal.pone.0191241.t001

The NPV is not intrinsic to the test and depend also on the prevalence of disease. Assuming IHD prevalence of 15%, NPV is calculated using the following formula based on Bayes' theorem:

The models were internally validated using 10 times repeated 5-fold cross validation with stratified sampling [20], whereby the model was developed on 80% of the data and tested on the remaining 20%. The average performance was calculated over five repetitions and the revised average value was viewed as a more accurate estimate of the model performance.

Statistical analysis was performed using R version 3.2.2 (R Core Team) statistical language. Functions from the glmnet library for logistic regression and cross validation (glmnet and cv.glmnet) were used.

Results

Participants

Between August 2013 and November 2015, 60 patients and 60 healthy volunteers were enrolled in the technical performance study; an additional three patients were recruited to replace scan malfunctions. The mean participant age was 64.2 years (± 10.79) and 37% were female (S2 Table). A total of 55/63 patients and 51/60 healthy volunteers were eligible for analysis (Fig 1). Three patients had all MCG scans lost due to device malfunction, two patients had scans excluded due to low signal-to-noise ratio, and three patient’s scans were all missing. Six healthy volunteers were found to have a history of IHD, and three were missing all their MCG scans.

[Figure omitted. See PDF.]

Fig 1. CONSORT Diagram: Technical performance study.

Participant flow through the technical performance study. Data were analyzed for 55/63 patients and 51/60 healthy controls.

https://doi.org/10.1371/journal.pone.0191241.g001

Between March 2015 and May 2015, 21 NSTEMI patients and 21 NIHD patients were recruited to the pilot clinical study. The mean patient age was 59.0 years (± 12.71) and 44% were female. A significant between-group difference was noted in gender distribution: 70% of the NIHD patients were female versus 19% of the NSTEMI patients (P = 0.002; S3 Table]). Patients with NSTEMI were significantly older than those with NIHD (mean difference –15.5 years; P<0.0001). Overall, 15/21 NSTEMI patients and 18/21 NIHD patients were eligible for analysis (Fig 2). Three NSTEMI patients and two NIHD patients had scans excluded due to low signal-to-noise ratio. A further three NSTEMI patients and one NIHD patient were excluded from the analysis due to missing or corrupted MCG data.

[Figure omitted. See PDF.]

Fig 2. CONSORT Diagram: Pilot clinical study.

Participant flow through the pilot clinical study. Data were analyzed for 15/21 patients and 18/21 healthy controls.

https://doi.org/10.1371/journal.pone.0191241.g002

Combined data set

Group A included 70 patients (55 patients from the technical performance study and 15 patients from the pilot clinical study). Group B included 69 controls (51 subjects from the technical performance study with no IHD as confirmed by an ECG and 18 subjects from the pilot clinical study with non-ischemic chest pain and no IHD as confirmed by Myoview or a stress echocardiogram). Group C consisted of 37 young healthy volunteers.

Candidate MCG predictors

Of the 10 candidate MCG predictors, three showed a significant difference between patients and controls (QR_peak and RS_peak, both P<0.01; RS_MMR, P<0.05; Table 2). Apart from QR_interval and RS_interval, all of the remaining predictors showed a significant difference between patients and young healthy volunteers. Fig 3 shows a representative histogram of the RS_peak predictors.

[Figure omitted. See PDF.]

Fig 3. Histogram of the RS_peak predictors.

A representative histogram of the RS_peak predictors for study participants enrolled in Group A, Group B, and Group C.

https://doi.org/10.1371/journal.pone.0191241.g003

[Figure omitted. See PDF.]

Table 2. Average value for each predictor for patient and control groups.

https://doi.org/10.1371/journal.pone.0191241.t002

Model performance

All logistic regression models showed respectable rule-out ability (Table 3). Model 1 yielded an AUC of 0.82, specificity of 33.0%, sensitivity of 98.6%, and NPV for the ruling-out of patients of 99.3%. A near perfect separation between the groups was achieved in Model 3 comparing patients with young healthy volunteers, with an AUC of 0.96 and specificity of 78.4% (sensitivity and NPV both 100%). Separation between the groups was lower in Model 2, with an AUC of 0.75, specificity of 20.3%, and sensitivity of 94.3% (NPV 95.2%) due to the absence of young healthy volunteers.

[Figure omitted. See PDF.]

Table 3. Apparent and cross-validation performance of the logistic regression models.

https://doi.org/10.1371/journal.pone.0191241.t003

We noted optimism in the apparent performance of the models compared with the cross-validated performance, which was similar across all models (Table 3).

Discussion

This study investigated the clinical performance of a portable prototype magnetometer to effectively rule-out patients whose chest pain has a non-ischemic origin from those with IHD. In an ED and an outpatient setting, use of the magnetometer would have the potential to reduce or negate the need for further tests in a subset of these non-ischemic patients.

Based on logistic regression Model 1, using an a priori cut-off of 0.2, the magnetometer ruled-out 35.0% of the control group with 97.7% NPV. When utilized as part of the clinical risk stratification, the ability to identify these patients could improve both outcomes and hospital resource utilization.

Although almost 30% of patients presenting with chest pain are aged <50 years, the admission rate for this age group is the lowest amongst all ages [21], highlighting the particular need to differentiate IHD from NIHD and non-cardiac chest pain in this group. Our results showed remarkable discriminating capability of MCG between IHD patients and young healthy volunteers (Model 3), which is in-line with previous studies. Lim et al identified high discriminating capability of MCG predictors between young control and patient groups [15]. In another study, Kandori et al. reported a tendency of the maximum magnetic fields at the QRS complex and T wave to decrease with age [17]. These observations are consistent with an age-related decline in heart function.

The most significant MCG predictors in our analysis were ventricular depolarization predictors. Researchers who focus on depolarization measures report diagnostic performance characteristics of MCG that is substantially equivalent to, if not better than, those who do not [22].

A limitation of our analysis is that we did not consider the effect of gender on MCG. The heterogeneous nature of the IHD patients in the technical performance study could also be a limitation. In the subgroup of NSTEMI patients, MCG scans were recorded up to 3 days after the index event, which could have negatively impacted the sensitivity and NPVs observed. Enrollment of NSTEMI patients to the pilot clinical study relied exclusively on a raised troponin level, the assumption being that troponin is 100% sensitive and specific, which it is not [23,24]. Similarly, recruitment of NIHD patients relied on a normal Myoview or a stress echocardiogram result, neither of which have 100% sensitivity or specificity. Due to the nature of the pilot clinical study it was not possible to implement consecutive recruitment, meaning that selection bias could not be excluded. The healthy young volunteers were recruited from a university campus with an average age <30 years, and thus they did not represent a typical young person presenting to the ED. Finally, the total number of participants analyzed in the study was small. Accurate estimation of the internal validity of a predictive model is problematic when using such a small sample size [20].

In summary, the portable prototype magnetometer fills an unmet need for a rule-out test in triaging chest pain patients to reduce the number of patients with a non-ischemic cause of chest pain who go through screening and follow-up, which could ultimately improve patient care and outcomes, and reduce resource utilization.

Supporting information

[Figure omitted. See PDF.]

S1 Fig. Fifteen averaged overlapped MCG waveforms recorded from a healthy subject.

(A) Averaged MCG signal with the ECG superimposed. (B) MFMs at the peaks of the QR and RS section with angle notation. The pole distance and the MFM angle measurements are also visible. ECG: electrocardiogram; MCG: magnetocardiography; MFM: magnetic field map.

https://doi.org/10.1371/journal.pone.0191241.s001

(JPG)

S1 Table. Definitions of candidate MCG predictors measured from the ventricular depolarization.

MCG = magnetocardiography; MFM = magnetic field map.

https://doi.org/10.1371/journal.pone.0191241.s002

(DOCX)

S2 Table. Baseline demographic characteristics of patients enrolled in the technical performance study.

https://doi.org/10.1371/journal.pone.0191241.s003

(DOCX)

S3 Table. Baseline demographic characteristics of patients enrolled in the pilot clinical study.

https://doi.org/10.1371/journal.pone.0191241.s004

(DOCX)

S4 Table. Confusion matrix for Model 1.

https://doi.org/10.1371/journal.pone.0191241.s005

(DOCX)

S5 Table. Confusion matrix for Model 1 with cross validation.

https://doi.org/10.1371/journal.pone.0191241.s006

(DOCX)

S6 Table. Confusion matrix for Model 2.

https://doi.org/10.1371/journal.pone.0191241.s007

(DOCX)

S7 Table. Confusion matrix for Model 2 with cross validation.

https://doi.org/10.1371/journal.pone.0191241.s008

(DOCX)

S8 Table. Confusion matrix for Model 3.

https://doi.org/10.1371/journal.pone.0191241.s009

(DOCX)

S9 Table. Confusion matrix for Model 3 with cross validation.

https://doi.org/10.1371/journal.pone.0191241.s010

(DOCX)

S1 Text. Unpublished reference.

https://doi.org/10.1371/journal.pone.0191241.s011

(DOCX)

Acknowledgments

Mark Kearney is British Heart Foundation Chair of Cardiovascular and Diabetes Research at the University of Leeds, Leeds, United Kingdom. This study was performed in a National Institute for Health Research (NIHR) clinical research facility. Medical writing assistance with the preparation of this manuscript was provided by Fiona Fernando of Bollin Strategies Ltd., United Kingdom.

Citation: Ghasemi-Roudsari S, Al-Shimary A, Varcoe B, Byrom R, Kearney L, Kearney M (2018) A portable prototype magnetometer to differentiate ischemic and non-ischemic heart disease in patients with chest pain. PLoS ONE 13(1): e0191241. https://doi.org/10.1371/journal.pone.0191241

References

1. Mehta SR, Granger CB, Boden WE, Steg PG, Bassand JP, Faxon DP, et al. Early versus delayed invasive intervention in acute coronary syndromes. N Engl J Med 2009; 360: 2165–2175. pmid:19458363

2. Amsterdam EA, Wenger NK, Brindis RG, Casey DE Jr, Ganiats TG, Holmes DR Jr, et al. 2014 AHA/ACC guideline for the management of patients with non-ST-elevation acute coronary syndromes: a report of the American College of Cardiology/American Heart Association Task Force on Practice Guidelines. J Am Coll Cardiol 2014; 64: e139–e228. pmid:25260718

3. Rohacek M, Bertolotti A, Grützmüller N, Simmen U, Marty H, Zimmermann H, et al. The challenge of triaging chest pain patients: the bernese university hospital experience. Emerg Med Int 2012; 2012: 975614. pmid:22114740

4. Backus BE, Six AJ, Kelder JC, Bosschaert MA, Mast EG, Mosterd A, et al. A prospective validation of the HEART score for chest pain patients at the emergency department. Int J Cardiol 2013; 168: 2153–2158. pmid:23465250

5. Six AJ, Backus BE, Kingma A, Kaandorp SI. Consumption of diagnostic procedures and other cardiology care in chest pain patients after presentation at the emergency department. Neth Heart J 2012; 20: 499–504. pmid:23090421

6. Lindahl B. Acute coronary syndrome–the present and future role of biomarkers. Clin Chem Lab Med 2013; 51: 1699–1706. pmid:23525876

7. Pope JH, Aufderheide TP, Ruthazer R, Woolard RH, Feldman JA, Beshansky JR, et al. Missed diagnoses of acute cardiac ischemia in the emergency department. N Engl J Med 2000; 342: 1163–1170. pmid:10770981

8. Yamada S, Yamaguchi I. Magnetocardiograms in clinical medicine: unique information on cardiac ischemia, arrhythmias, and fetal diagnosis. Intern Med 2005; 44: 1–19. pmid:15704657

9. Kim D, Kim K, Lee YH, Ahn H. Detection of atrial arrhythmia in superconducting quantum interference device magnetocardiography; preliminary result of a totally-noninvasive localization method for atrial current mapping. Interact Cardiovasc Thorac Surg 2007; 6: 274–279. pmid:17669841

10. Malmivuo J, Plonsey R. Bioelectromagnetism. New York: Oxford University Press; 1995.

11. Fenici R, Brisinda D, Venuti A, Sorbo AR. Thirty years of clinical magnetocardiography at the Catholic University of Rome: diagnostic value and new perspectives for the treatment of cardiac arrhythmias. Int J Cardiol 2013; 168: 5113–5115. pmid:23958422

12. Fenici R, Brisinda D, Meloni AM. Clinical application of magnetocardiography. Expert Rev Mol Diagn 2005; 5: 291–313. pmid:15934809

13. Agarwal R, Saini A, Alyousef T, Umscheid CA. Magnetocardiography for the diagnosis of coronary artery disease: a systematic review and meta-analysis. Ann Noninvasive Electrocardiol 2012; 17: 291–298. pmid:23094875

14. Hänninen H, Takala P, Mäkijärvi M, Montonen J, Pesola K, Korhonen P, et al. Detection and localization of myocardial ischemia by multichannel magnetocardiography. Int J Bioelectromagnetism 2000; 2.

15. Lim HK, Kwon H, Chung N, Ko YG, Kim JM, Kim IS, et al. Usefulness of magnetocardiogram to detect unstable angina pectoris and non-ST elevation myocardial infarction. Am J Cardiol 2009; 103: 448–454. pmid:19195500

16. Mooney TW, Ghasemi-Roudsari S, Reade Banham E, Symonds C, Pawloski N, Varcoe BTH. A portable diagnostic device for cardiac magnetic field mapping. Biomed Phys Eng Express 2017; In press.

17. Kandori A, Ogata K, Watanabe Y, Takuma N, Tanaka K, Murakami M, et al. Space-time database for standardization of adult magnetocardiogram-making standard MCG parameters. Pacing Clin Electrophysiol 2008; 31: 422–431. pmid:18373760

18. Fenici R, Brisinda D, Meloni AM. Effects of filtering on computer-aided analysis for detection of chronic ischemic heart disease with unshielded rest magnetocardiographic mapping. Neurol Clin Neurophysiol 2004; 2004: 7. pmid:16012666

19. Harrell FE Jr. Binary logistic regression. In: Regression Modeling Strategies, Springer Series in Statistics. Switzerland: Springer International Publishing; 2015. pp. 219–274.

20. Steyerberg EW, Harrell FE Jr, Borsboom GJ, Eijkemans MJ, Vergouwe Y, Habbema JD. Internal validation of predictive models: efficiency of some procedures for logistic regression analysis. J Clin Epidemiol 2001; 54: 774–781. pmid:11470385

21. Ekelund U, Akbarzadeh M, Khoshnood A, Björk J, Ohlsson M. Likelihood of acute coronary syndrome in emergency department chest pain patients varies with time of presentation. BMC Res Notes 2012; 5: 420. pmid:22871081

22. Alday EA, Ni H, Zhang C, Colman MA, Gan Z, Zhang H. Comparison of electric- and magnetic-cardiograms produced by myocardial ischemia in models of the human ventricle and torso. PLoS One 2016; 11: e0160999. pmid:27556808

23. Xu RY, Zhu XF, Yang Y, Ye P. High-sensitive cardiac troponin T. J Geriatr Cardiol 2013; 10: 102–109. pmid:23610580

24. Zymliński R, Sokolski M, Siwolowski P, Biegus J, Nawrocka S, Jankowska EA, et al. Elevated troponin I level assessed by a new high-sensitive assay and the risk of poor outcomes in patients with acute heart failure. Int J Cardiol 2017; 230: 646–652. pmid:28069251