Content uploaded by Maarten van Smeden
Author content
All content in this area was uploaded by Maarten van Smeden on Jan 07, 2022
Content may be subject to copyright.
...............................................................................................................................................................................
The year in cardiovascular medicine 2021:
digital health and innovation
Panos E. Vardas
1,2
*, Folkert W. Asselbergs
3,4
, Maarten van Smeden
5
, and
Paul Friedman
6
1
Heart Sector, Hygeia Hospitals Group, HHG, 5, Erithrou Stavrou, Marousi, Athens 15123, Greece;
2
European Heart Agency, ESC, Brussels, Belgium;
3
Department of Cardiology,
Division of Heart & Lungs, University Medical Center Utrecht, Utrecht University, Utrecht, The Netherlands;
4
Health Data Research UK and Institute of Health Informatics,
University College London, London, UK;
5
Julius Center for Health Sciences and Primary Care, University Medical Center Utrecht, Utrecht University, Utrecht, The Netherlands;
and
6
Department of Cardiovascular Medicine, Mayo Clinic, Rochester, MN, USA
Received 13 October 2021; revised 15 November 2021; accepted 23 November 2021
Graphical Abstract Digital tool development in cardiovascular medicine
----------------------------------------------------------------------------------------------------------------------------------------------------------
This article presents some of the most important developments in the field of digital medicine that have appeared over the last 12 months
and are related to cardiovascular medicine. The article consists of three main sections, as follows: (i) artificial intelligence-enabled cardio-
vascular diagnostic tools, techniques, and methodologies, (ii) big data and prognostic models for cardiovascular risk protection, and (iii)
wearable devices in cardiovascular risk assessment, cardiovascular disease prevention, diagnosis, and management. To conclude the article,
the authors present a brief further prospective on this new domain, highlighting existing gaps that are specifically related to artificial intelli-
gence technologies, such as explainability, cost-effectiveness, and, of course, the importance of proper regulatory oversight for each clinical
implementation.
Keywords AI-ECG •AI-wearables •Digital health •Cardiovascular medicine •Big data •Machine learning
* Corresponding author. Tel: +30 210 6867691/+30 210 686 7937, Fax: +30 210 6869926, Email: pvardas@hygeia.gr
© The Author(s) 2022. Published by Oxford University Press on behalf of European Society of Cardiology. All rights reserved. For permissions, please e-mail:
journals.permissions@oup.com
European Heart Journal (2021) 00,1–10
doi:10.1093/eurheartj/ehab874
SPECIAL ARTICLE
Downloaded from https://academic.oup.com/eurheartj/advance-article/doi/10.1093/eurheartj/ehab874/6487290 by guest on 07 January 2022
Introduction
Digital health, a broad-spectrum concept that has received a signif-
icant boost as a result of the COVID-19 pandemic, is growing ex-
ponentially, flexing its muscles with scientific breakthroughs and
associated publications, while also driving trends and develop-
ments in industry.
For cardiovascular medicine in particular, during the last year, an
impressive number of authoritative new publications have con-
firmed previous research findings and proposed new innovative
ideas and practices related to the diagnostic and therapeutic
management of cardiovascular diseases, with the promise of
ground-breaking developments during the coming years, for
both cardiovascular sciences and care.
In the year 2021, as in the years immediately preceding, the field
of digital health has been flooded with publications referring to the
diverse applications of artificial intelligence (AI), from supervised
to unsupervised learning, focusing mainly on the diagnostic capabil-
ities of this impressive new technology.
Furthermore, the role of machine learning algorithms in the de-
velopment of clinical prognostic models for risk assessment and
early warning systems represents a rapidly evolving field that
may be expected to have a catalytic effect by improving the predic-
tion of medium- and long-term clinical outcomes.
Indeed, the prospects seem to be excellent.
Nonetheless, some questions still remain. Apart from the in silico
design and development, the explainability of the machine learning
algorithms and their validation methodology need to be more so-
lidly confirmed in well-designed longitudinal studies, as well as in
clinical practice before these algorithms find their way into
the guidelines.
Beyond the field of AI—though often closely connected with
it—developments in wearable devices have commandeered a
significant part of the recent scientific literature, highlighting emer-
ging new possibilities for the fuller monitoring and treatment of
cardiovascular diseases and their related risk factors.
The technological developments in wearables—especially as
they expand to cover not only the needs of fitness but also those
of diagnosis and monitoring of cardiovascular diseases—will ob-
viously require more substantial regulation to ensure device relia-
bility, backed by well-organized studies that will highlight their
cost-effectiveness so that insurance companies may be persuaded
they should be reimbursable.
Artificial intelligence-enabled
cardiovascular diagnostic tools,
techniques, and methodologies
A new era in electrocardiogram analysis
The application of AI to the electrocardiogram (ECG) has seen sig-
nificant advances recently and has developed in the following two
broad categories: (i) tools to automate ECG interpretation, ex-
tending human capabilities via massive scalability, important as mo-
bile form factors permit signal acquisition and (ii) algorithms to
identify conditions not visible to human readers by training
networks to identify multiple, complex, non-linear patterns in
the ECG signal to find occult disease (confirmed using other tests
such as imaging), or impending disease. In contrast to automation
tools in which a human overread provides a gold standard, algo-
rithms identifying occult or future conditions require additional pa-
tient information.
Several groups have used large, labelled data sets to train neural
networks to accurately apply diagnostic codes to single-lead and
multiple-lead ECGs. Hannun et al.
1
used 91 232 single-lead ECGs
from a wearable patch to train a network to provide 12 rhythm
classes and found that the network outperformed the average car-
diologist’s read. Subsequently, two mega trials using smart watches
based on photoplethysmography technology enrolled 419 297 and
246 541 patients to screen for atrial fibrillation (AF) in under 9
months.
2,3
These trials confirmed the ability to massively enrol subjects and
acquire data, at the cost of high rates of early dropout and a low
yield of disease (,0.5% in both studies), and with limited clinical
characterization of the study subjects. Ongoing trials will assess
these tools in the context of patients selected for arrhythmia
risk. Finally, there have been recent reports of interesting research
that aimed to develop and validate an AI-enabled ECG algorithm
capable of comprehensive 12-lead ECG analysis comparable to
that of practising cardiologists.
4
Furthermore, the AI-ECG has identified occult and manifest
cardiac conditions, including ventricular dysfunction,
5
peripartum
cardiomyopathy,
6
amyloid heart disease,
7
and pulmonary hyperten-
sion,
8
as well as non-cardiac conditions such as hyperkalaemia and
cirrhosis.
9,10
In addition, special algorithms have been used for the
early diagnosis of valvular diseases such as asymptomatic or oligo-
symptomatic severe aortic stenosis and mitral regurgitation,
11–13
left ventricular hypertrophy,
14,15
myocardial infarction,
16,17
and a
number of other conditions. Common findings in these studies
include a strong clinical performance [area under the curve
(AUC) often .0.90] and detection of disease months to years
ahead of the clinical diagnosis.
The significance of these findings remains to be evaluated,
taking into account the scalability of electrocardiography, and
hence the contribution of AI to its further and more
substantial utilization.
The ECG is an ever-present diagnostic tool that has served
medical practitioners for more than a century. With the support
of deep learning AI techniques, it is clearly entering a new era, in
which it may prove to be a powerful detector of subclinical and
clinical cardiac diseases, going beyond the boundaries of human
observation. There can be no doubt that when the previous cap-
abilities of the ECG are combined with the evolving features of
wearable devices such as smartphones, the chances of a much
broader and pluralistic diagnostic process will increase rapidly.
Artificial intelligence-electrocardiogram
and clinical trials
Clinical trials are essential to demonstrate the ability of novel digit-
al tools like the AI-ECG to improve human health. Factors to con-
sider in evaluating the quality of AI-ECG studies are listed in
Table 1. A framework for the assessment of how well AI-ECG
2P.E. Vardas et al.
Downloaded from https://academic.oup.com/eurheartj/advance-article/doi/10.1093/eurheartj/ehab874/6487290 by guest on 07 January 2022
clinical trials can predict meaningful outcomes, based on whether
the trials are single-centre or multicentre, prospective or retro-
spective, is shown in Table 2. It is likely that Level 3 or higher would
be required for regulatory approval, allowing for variation in spe-
cific tests and regional differences. There is a pressing need for
additional clinical trials to assess AI-ECG tools. A search of clinical-
trials.gov on 8 October 2021, for trials utilizing the terms ‘artificial
intelligence’and ‘ECG’returned 27 studies, with only 5 completed.
The first AI-ECG prospective trial published, the Eagle study,
18
demonstrated how digital, pragmatic trials can effectively and ra-
pidly enrol subjects, and how the AI-ECG can positively impact
clinical practice. It randomized 120 primary care teams from
45 clinics or hospitals in Minnesota and Wisconsin to an interven-
tion arm (clinicians have access to AI-ECG results screening for left
ventricular dysfunction when routinely ordering a clinical ECG) or
a control arm (no AI results). Despite the development of the pan-
demic, .22 000 patients were enrolled in 8 months, and the
AI-ECG increased the diagnosis in the overall cohort [odds ratio
(OR) 1.32, P=0.007]. The test performance (AUC =0.92)
matched that of the initial retrospective cohort (0.93).
19
Interestingly, the overall utilization of echocardiography was simi-
lar in both groups but in the intervention group, more echocardio-
grams were ordered for patients with a positive AI-ECG (38.1%
control vs. 49.6% intervention, P,0.001), suggesting that the
AI-ECG did not lead to more echocardiograms, but to better se-
lection of patients to undergo imaging.
Cardiovascular imaging
Imaging has been the frontrunner in the application of AI in health-
care, because of the repetitive nature of imaging processing and
evaluation. Artificial intelligence may improve imaging quality—
and thereby scan and dose time—and assist in segmentation, pro-
cessing, and analysis.
20
Furthermore, most data are retrieved from
a single standardized data source, making it more accessible for
large-scale analyses. During the pandemic, critics were pointing
out that, despite massive efforts, AI had no impact on the care of
COVID-19 patients, while simple straightforward randomized con-
trolled trials did save lives.
21
However, this clearly shows only one
side of the coin. The pandemic led to a greater burden on radiology
resources, as computerized tomography (CT) scans were carried
out routinely in all patients. Artificial intelligence is key in all parts
of the imaging pipeline, including acquisition, processing, and ana-
lyses.
22,23
Furthermore, a plethora of papers have been published
during the pandemic, showing the prognostic value of calcium score
measurements in COVID-19 chest CT scans.
Those measurements can be automated using deep learning,
24
providing clinicians with information, not only about the pulmon-
ary status of COVID-19 patients, but also their cardiovascular
risks.
25
Artificial intelligence will enable automated analyses of
routine chest CT examinations for opportunistic cardiovascular
screening, allowing early preventive treatment. All these develop-
ments, together with the notable Food and Drug Administration
clearance of a new technology to identify strokes on brain CT
scans enabled by AI, hold out the prospect of a bright future in
medical diagnostics.
26,27
Retinal photography to detect
cardiovascular disease
Another imaging application that can determine risk across a wide
range of diseases is retinal photography. Retinal photography is a
non-invasive imaging modality that aides in the diagnosis and treat-
ment of major eye diseases, butcan also provide information on the
human vasculature and therefore cardiovascular disease. Prior
manually coded studies have shown that retinal vascular abnormal-
ities are predictive for cardiovascular disease.
28
Deep learning can
extend this knowledge through the automation and detection of
more subtle signs that are not clearly visible to the human eye.
Several large-scale studies have been published recently, focusing
on the predictive value of features extracted from retinal photo-
graphs. Studies have shown that deep learning algorithms can pre-
dict levels of biomarkers such as haemoglobin to detect
Table 1 Factors to consider in evaluating artificial
intelligence-electrocardiogram studies
1. Data label accuracy: robustness of data labels used for training
and testing
a. Proxy labels (EMR report of ‘chest pain’) vs. gold standard
labels (physician-described angina, troponin levels, serial
ECGs)
b. Number of subjects for whom labels available
c. Absence in labels of false distractors (e.g. all ECGs from
patients with condition taken at one hospital, using an
acquisition system different than that used in controls, so
that network may identify differences in ECG machines
rather than disease)
2. Risk of bias: cohort creation and controls
a. Controls not identical to cases in all conditions except the
desired AI differentiator, most commonly in demographics
(example: using adult controls for paediatric ECGs with
WPW to train a network)
b. Controls and cases taken from public data sets (difficult to
know details regarding the absence/presence of
conditions, poor phenotyping)
c. Use of only subsets of larger data sets, introducing
potential bias—need for racial, ethnic, and geographic
diversity in data sets (example: initial face recognition AI,
trained using only Caucasians, mislabelled African
Americans as primates).
d. Inappropriate exclusion of data at the patient or signal
feature level will bias results (examples: exclusion of
signals on the basis of artefact of those same exclusions
will not be used in real-world implementation; or
exclusion of patients with hypertension when creating an
AI-ECG screen for hypertension)
e. Temporal shifts—training using data acquired in the
remote past and application to recent data sets
f. Commercial interest and backgrounds of engineers
creating AI tools (potential bias)
3. Overfitting/lack of generalizability
a. Overly complex AI-ECG network with a small number of
samples (the results are not generalizable to other
populations)
b. Most data sets for AI-ECG training number in the tens of
thousands or more, although exceptions exist
The year in cardiovascular medicine 2021 3
Downloaded from https://academic.oup.com/eurheartj/advance-article/doi/10.1093/eurheartj/ehab874/6487290 by guest on 07 January 2022
anaemia,
29
as well as age, sex, body composition, and creatinine le-
vels,
30
although external validation is warranted before this can be
widely adopted in population screening. Another interesting study
investigated the predictive capability of deep-learning-enabled cor-
onary artery calcium (CAC) scores derived from retinal scan data.
31
Computerized tomography scans and retinal measurements were
performed on the same day and the score derived from retinal
images showed an AUC of 0.74 for predicting CAC .0.
Although higher than other single risk factors, such as age, sex,
and cholesterol, the added predictive value in the multivariable clin-
ical model was limited (AUC from 0.782 to 0.784). However, the
CAC score derived from retinal scans showed a similar perfor-
mance in predicting cardiovascular outcomes to CAC measured
by CT scan (both AUC 0.71). Furthermore, the authors showed
in the UK Biobank that this retinal-based CAC score could improve
risk stratification in those with borderline or intermediate risk.
However, this method has certain disadvantages. Home-based
tests are not yet available, and images with poor quality were ex-
cluded in the reported analyses, which is likely to limit the external
validity. Real-world data are necessary to estimate the added value
in population screening, and the development of mobile applica-
tions for self-tests is needed
32
before implementation on a large
scale. These deep learning applications are, however, already use-
ful in those who already undergo regular retinal scans, such as dia-
betic patients, to screen for retinopathy.
33
To end this section, at least a brief mention should be made of
the diagnostic capability and cost-effectiveness of the combined
imaging approach, where the use of AI and magnetic resonance
imaging yields the atheroma index of the coronary arteries or per-
ipheral vessels as a byproduct of the primary diagnostic evaluation
of other organs.
34
Automation of imaging processing
While the application of AI in cardiovascular imaging for clinical
decision-making is still in its infancy, the use of AI to automate im-
aging processing in other fields, such as ophthalmology as dis-
cussed above, oncology, and dermatology, has already matured.
However, several promising studies using different imaging mod-
alities have recently been published and have shown that cardi-
ology is able to catch up with the other disease domains. A
large international collaborative study showed that the coeffi-
cient of variation in measuring left ventricular wall thickness by
cardiovascular magnetic resonance was significantly lower for
machine learning in comparison to human experts.
35
This study
involved a cohort of patients with hypertrophic cardiomyopathy,
where variations in wall thickness measurements directly impact
clinical decision-making by affecting the calculation of sudden
death risk and thereby the indication for preventive implementa-
tion of an implantable cardioverter defibrillator (ICD).
Another recent example of automation is the International
Society for Heart and Lung Transplantation’s grading of endomyo-
cardial biopsies in heart transplant patients.
36
The authors
compared histological grading performed by expert pathologists
with a computer-assisted automated pipeline and showed similar
performance of the Computer-Assisted Cardiac Histologic
Evaluation (CACHE) grader in comparison to the pathologist
........................................................................................................................................................................
Table 2 Proposed categories of clinical trials to assess the artificial intelligence-electrocardiogram
Study
category
Description of
population used to
test an AI-ECG
network
Study design Strengths Limitations
1 Public data set Retrospective Inexpensive, rapid Unreliable phenotyping, high risk of
bias, limited clinical utility
2 Single centre: same
hospital/clinic used to
acquire data, but
different patients
Retrospective Rich data sets to phenotype patients,
rapid, relatively inexpensive, robust
proof-of-concept approach
Risk of bias, under-representing
important populations
3 Multicentre: different
hospital systems used
to test AI, than one
used to create
Retrospective Lower risk of bias, potential for greater
diversity among subjects, test types,
potential to rapidly and meaningfully
assess tests
Need to confirm labels assessed in
systematic, similar manner across
sites (example: assessment of EF
by echo)
4 Single centre: same
hospital used to test
AI, different patients
Prospective Assesses AI, impact on workflow,
adoption by clinicians, clinical impact
Greater technical infrastructure
required, more expensive,
greater time requirement
5 Multicentre Prospective (may use
retrospective ECGs
to prospectively enrol
patients)
Prospective trial but with accelerated
enrolment, by screening large data set
of stored ECGs; potential for portal/
email study invitations and pragmatic
design, statistical robust, potential to
minimize bias
Greater technical requirements,
time, expense
4P.E. Vardas et al.
Downloaded from https://academic.oup.com/eurheartj/advance-article/doi/10.1093/eurheartj/ehab874/6487290 by guest on 07 January 2022
(Figure 1). Moreover, they showed only limited attenuation of the
performance when it was applied to an external validation data set,
indicating good generalizability across different scanning and tissue
preparation protocols. International collaborative efforts in the
field of transplant research have been hampered by variations in
grading by individual centres, which increase the noise-to-signal ra-
tio in the detection of biologically meaningful results when data
sets from individual centres are merged. CACHE-enabled auto-
mated grading can play an instrumental role in advancing the field
of transplant research.
Finally, AI will increasingly be applied in the field of echocardiog-
raphy. Prior studies have shown that AI can identify different echo
views, segment cardiac structures, estimate ejection fraction,
37,38
and diagnose diseases such as cardiac amyloidosis.
39
Recently, a
Figure 1 An overview of the ‘Computer-Assisted Cardiac Histologic Evaluation (CACHE)-Grader’multicentre validation experiment. The
CACHE-Grader’s performance was compared with both the grade of record and independent pathologists performing re-grading, demonstrat-
ing non-inferiority to expert pathologists, generalizability to external data sets, and excellent sensitivity and negative predictive value.
Reproduced by permission from Peyster et al.
36
The year in cardiovascular medicine 2021 5
Downloaded from https://academic.oup.com/eurheartj/advance-article/doi/10.1093/eurheartj/ehab874/6487290 by guest on 07 January 2022
study from Stanford also showed that deep-learning algorithms are
able to detect pacemaker or ICD leads and, interestingly, are able
to predict age, sex, height, and weight based on echo images.
40
Furthermore, they used gradient-based sensitivity mapping meth-
ods to highlight the regions of interest for human interpretation.
Visualization methods to unlock the so-called black box algorithms
are essential if healthcare professionals are to fully adopt the re-
sults generated by AI models. These algorithms will support un-
trained professionals with the interpretation of echocardiograms
when cardiological expertise is of limited availability. A recent
study showed that deep learning can even help untrained nurses
to perform limited echocardiograms for standard evaluation of
the left and right ventricular size and pericardial effusion, enabling
the use of echocardiograms in non-cardiological settings, such as
primary care, COVID wards, or remote areas.
41
However, before
its widespread implementation, additional studies regarding safety
and generalizability are warranted.
Big data and prognostic models
for cardiovascular risk prediction
Machine learning for risk prediction
Clinical risk prediction modelling based on machine learning has
been an active field of research. During the first months of the pan-
demic, hundreds of such models were developed.
42
Clinical pre-
diction models are commonly developed to inform physicians
about the probability of a certain disease being present (diagnosis),
or to predict a certain health state in the future (prognosis), for in-
dividual patients, and to use that knowledge in the care of those
patients.
43
By applying machine learning techniques that can use
complex data relationships between predictors and outcome
without the need for the modeller to pre-specify them, the ex-
pectation is that the accuracy of predictions will improve com-
pared with traditional risk prediction modelling approaches, and
that its application will be less labour-intensive at the bedside.
Improvements in predictive accuracy are, however, not guaran-
teed.
44
For instance, a study that developed machine learning
models to predict the risk of death after acute myocardial infarc-
tion (AMI) found that machine learning models were not uniformly
superior to a traditional logistic regression approach in a cohort of
755 402 AMI patients.
45
In fact, of the three models used, two
were superior to the logistic regression model for risk stratifica-
tion. In addition, those two models were much better calibrated
across patient groups based on age, sex, race, and mortality risk,
and thus better suited for risk prediction. In contrast, the third
model, based on a neural network, was found to be inferior to
the logistic regression model used in the study. There may be prag-
matic reasons for this inferiority, but they are probably related to
the methodology used and in particular the sample sizes of each of
the study’s populations.
Nonetheless, in other settings, machine learning approaches
have yielded promising results. One such study developed models
to predict the risk of death, myocardial infarction, and major
bleeding after an acute coronary syndrome (ACS). The
machine-learning-based models were developed from a cohort
with 19 826 adult ACS patients and were shown to predict the
risk with high AUCs on external validation, at 1 year (AUCs:
0.81–0.92) and 2 years (AUCs: 0.84–0.93).
46
Early warning systems
Early warning systems are prognostic predictive models that aim to
inform physicians about important future health outcomes. Often,
these early warning systems are used to monitor patients and to
update these predictions over time. For instance, to predict circu-
latory failure in patients admitted to intensive care, a machine
learning model was developed that made a new prediction for
every patient every 5 min.
47
The early warning systems developed
were shown to yield high AUCs, between 0.88 and 0.94. However,
these models also produced two to three alarms per patient per
day. This may result in the so-called alarm fatigue, which can
lead to inadequate responses and may even impact patient
safety.
48
Hence, for these early warning systems and other risk
prediction models used to guide clinical decisions, it is essential
to ensure safety and effectiveness in improving patient outcomes,
for instance, through a randomized controlled trial (RCT) compar-
ing the early warning system to standard of care. One such RCT
evaluated a machine-learning-based early warning system for
pending intraoperative hypotension.
49
This early warning system
updates every 20 s the probability of a hypotensive event in the
next 15 min (warning when estimated probability .85%) based
on the arterial pressure waveform.
50
In an RCT with 60 adult
elective non-cardiac surgery patients, the early warning system,
in combination with a haemodynamic diagnostic guidance and
treatment protocol, reduced the median total time of hypotension
per patient from 32.7 min under standard of care to 8 min.
Big data: representativeness
and algorithmic fairness
Access to large and diverse databases with electronic health records
creates important new research opportunities. Such large databases
include the Clinical Practice Research Datalink (CPRD), with highly
detailed data from .5 million individuals representative of the UK
population. Using the CPRD data, one interesting study developed
and validated several machine-learning-based risk prediction models
for predicting the risk of familial hypercholesterolaemia in primary
care patients.
51
These prediction models were shown to have
high AUCs of around 0.89. The large scale and representativeness
of large databases also allows for studying specific groups that
may otherwise be difficult to study. For instance, one study com-
pared cardiovascular disease incidences and outcomes in homeless
individuals using a linkage between CPRD, hospital episode statistics,
and the Office of National Statistics for mortality data.
52
This study
showed that homeless individuals have a 1.8 times higher risk of de-
veloping cardiovascular disease and are 1.6 times more likely to die
within 1 year after cardiovascular disease diagnosis, compared with
similar individuals who are not homeless. Finally, large and diverse
databases, where minority groups are also well represented, are es-
sential to ensure that the algorithms developed are fair,
53
i.e. do not
systematically disadvantage certain groups of individuals. This re-
quires evaluation of the performance of the algorithms in important
subgroups. For instance, a recent study on atherosclerotic cardio-
vascular disease risk prediction showed a comparable performance
6P.E. Vardas et al.
Downloaded from https://academic.oup.com/eurheartj/advance-article/doi/10.1093/eurheartj/ehab874/6487290 by guest on 07 January 2022
of existing pooled cohort equations and newly developed
machine-learning-based models in Asian and Hispanic subgroups,
for which the performance was so far uncertain.
54
Wearable devices in cardiovascular
risk assessment, cardiovascular
disease prevention, diagnosis,
and management
Wearables in atrial fibrillation risk
assessment and management
The role of physical activity as a modifiable risk factor for the de-
velopment of AF was studied recently in a well-organized prospec-
tive study,
55
which included 93 669 participants from the UK
Biobank prospective cohort, without a prevalent history of AF,
who wore a wrist-based triaxial accelerometer for 1 week. The
sensor captured acceleration at 100 Hz with a dynamic range of
+8 g. The primary outcome of the study was incident AF.
According to the findings of the study, greater accelerometer-
derived physical activity is associated with a lower risk of incident
AF and stroke, after adjustment for clinical risk factors (Figure 2).
Wearable sensors may enable both objective assessment of physi-
cal activity and modification of AF risk through targeted feedback.
The authors consider that future preventive efforts to reduce AF
risk may be most effective if they target adherence to objective ac-
tivity thresholds.
Another study
56
that aimed to investigate the association be-
tween changes in physical activity and the onset of AF reported si-
milar findings. A total of 1410 participants from the general
population were studied (46.2% women, mean age 74.7 +4.1
Figure 2 Cumulative risks of atrial fibrillation (upper panel) and stroke (lower panel) stratified by adherence to physical activity recommenda-
tions, as validated by accelerometer-derived physical activity. Reproduced by permission from Khurshid et al.
55
The year in cardiovascular medicine 2021 7
Downloaded from https://academic.oup.com/eurheartj/advance-article/doi/10.1093/eurheartj/ehab874/6487290 by guest on 07 January 2022
years), with risk factors but with no prior AF diagnosis, who under-
went continuous monitoring for AF episodes along with daily ac-
celerometric assessment of physical activity, using an implantable
loop recorder, over an average period of 3.5 years.
According to the findings of the study, intra-individual changes in
physical activity were associated with the onset of AF episodes, as
detected by continuous monitoring, in a high-risk population. For
each person, a 1 h decrease in daily physical activity during the pre-
vious week increased the odds of AF onset the next day by 25%,
while the strongest association was seen in the group with the
lowest activity overall.
Apart from these two recent and revealing studies of the rela-
tionship between a person’s physical activity and the occurrence
of AF, a significant number of ongoing or recently published studies
have evaluated the capabilities of wearables, focusing on the rela-
tionship between the individual clinical outcome and the burden of
recorded episodes of clinical or subclinical AF.
57
Wearables in heart failure assessment
and management
Heart failure (HF), a fast-growing disease internationally, also has a
long-standing affinity with wearable technology, since the patho-
physiology of the disease and its clinical consequences require
close and continuous long-term monitoring. Indeed, wearables of-
fer a unique opportunity to assess patients’status and a number of
indicators closely, outside the classical settings. In patients with HF,
data from consumer wearables, such as physical activity step count
or heart rate, but also more intense monitoring of such factors as
pulmonary artery pressure or fluid retention, have long been the
target of these evolving devices.
When we look at the findings and messages of the most recent
relevant studies, those of the LINK-HF multicentre study by Stehlik
et al.,
58
which evaluated the accuracy of non-invasive remote mon-
itoring in predicting rehospitalization for HF, were quite revealing.
This was a study of 100 patients with HF, aged 68.4 +10.2 years
(only 2% female). The investigators showed that multivariate phy-
siological telemetry from a wearable sensor, in combination with
machine learning analytics, can accomplish accurate early detec-
tion of impending rehospitalization with a predictive accuracy
comparable to that of implantable devices. The authors emphasize,
however, that the clinical efficacy and generalizability of this low-
cost non-invasive approach to rehospitalization mitigation still
needs further testing.
Looking at the issues more broadly, apart from the use of mod-
ern electronic technology for continuous haemodynamic monitor-
ing in HF patients, it has become clear that such technology can
and should be used for education and support in these patients’
therapeutic management.
59
The EPIC-HF study (Electronically Delivered Patient-Activation
Tool for Intensification of Medications for Chronic Heart Failure
with Reduced Ejection Fraction) evaluated patients from a diverse
health system who had HF and reduced ejection fraction,
randomizing them to usual care vs. patient activation tools. The
tools—a 3 min video and a one-page checklist—encouraged
patients to work collaboratively with their clinicians to ‘make
one positive change’in their HF medication.
The findings were clear. A patient activation tool delivered elec-
tronically before the cardiology clinic visit enhanced clinicians’in-
tensification of guideline-directed medical therapies.
ST-segment elevation myocardial
infarction
The vast majority of wearable devices currently offer single-lead
ECG recording, which allows the detection of AF and, more rarely,
other arrhythmias to a satisfactory extent. However, such ECG re-
cordings cannot reliably detect ST/T changes due to regional myo-
cardial ischaemia. Nevertheless, a good many expectations have
been invested in this possibility, as ECG recording by wearables,
backed by telemonitoring to detect the early signs of myocardial
ischaemia, could limit its often destructive effects.
Muhlestein et al.,
60
in their relatively recent publication, re-
viewed the feasibility of combining serial smartphone single-lead
recordings to create a virtual 12-lead ECG capable of reliably diag-
nosing ST-elevation myocardial infarction. The study included
200 subjects (mean age 60 years, 43% female).
For all interpretable pairs of smartphone ECGs, compared with
standard 12-lead ECGs (n=190), the sensitivity, specificity, and
positive and negative predictive values for ST-segment elevation
myocardial infarction (STEMI) or STEMI equivalent (left bundle
branch block) achieved by the smartphone were 0.89, 0.84, 0.70,
and 0.95, respectively. The authors concluded that a 12-lead
equivalent ECG constructed from multiple serial single-lead re-
cordings from a smartphone can identify STEMI with a good cor-
relation to a standard 12-lead ECG.
Similar to the previous study, a prospective study
61
also investi-
gated the feasibility and accuracy of a smartwatch in recording
multiple electrocardiographic leads and detecting ST-segment
changes associated with ACS, compared with a standard 12-lead
ECG. A commercially available smartwatch was used in 100 parti-
cipants. The watch was placed in different body positions to obtain
nine bipolar ECG tracings (corresponding to Einthoven leads, II
and III, and precordial leads, V1–V6), which were compared with
a simultaneous standard 12-lead ECG.
To a significant extent, there was an agreement between the
findings of the smartwatch tracings and the standard ECGs for
the identification of a normal ECG, ST-segment changes, and no
ST-segment elevation.
The findings of the two previous studies give cause for optimism
that, in the near future, the technical difficulties will be overcome,
so that the recording of wearable devices will gain sufficient relia-
bility for the recording of ischaemic changes on the ECG.
Conclusions
Digital health stands poised to transform cardiovascular medicine,
much as echocardiographic imaging has upended stethoscope-
based auscultation for diagnosis. Work published in 2021 has
advanced this hope, and engaged an ever-widening group of stake-
holders, critical to ensure proper evaluation of this important
technology that may touch so many lives. Digital health’s great
promise in no small measure stems from its ability to endow ex-
tant medical tests (ECG, fundoscopy, and imaging) and electronic
8P.E. Vardas et al.
Downloaded from https://academic.oup.com/eurheartj/advance-article/doi/10.1093/eurheartj/ehab874/6487290 by guest on 07 January 2022
health record data, which are known to practitioners and inte-
grated into workflows, with new superpowers, and to draw mas-
sively scalable data from wearables into the fold. This integration
will accelerate adoption and impact care.
Before the promise of digital health can bear fruit to improve
human health, a major gap must be addressed—the paucity of clin-
ical trials to address outcomes. The ‘black box’issue and lack of
explainability are widely discussed concerns that may not be
solved in the short term, but may be mitigated or overcome
with robust evidence from prospective clinical trials. Data manage-
ment processes to prevent overwhelming an already taxed health-
care system are mandatory. Further development of novel hybrid
regulatory strategies, recognizing software as a medical device
coupled to consumer hardware, are pre-requisites to exponential-
ly driving data availability. With broad input from clinicians, indus-
try, regulators, and patients; attention to privacy and human rights;
diligent testing, validation and oversight; and prospective trial data,
digital health promises an exciting and healthy future, as opposed
to a brave new world.
Acknowledgements
The authors thank the following for their contributions to the pre-
paration of the manuscript: Dr Anastasia Xintarakou, Mr Philip
Lees, and Ms Alexandra Kourlampa. Thanks also to Professors
E. Peyster and S. Khurshid for permission to use the figures.
Conflict of interest: P.E.V. has received teaching fees from
Servier International and consultancy fees from Hygeia Hospitals
Group, HHG, and ESC/European Heart Agency, Brussels. P.F. is
the co-inventor of AI-ECG algorithms, and he and Mayo Clinic
may benefit from their commercialization. F.W.A. is supported
by UCL Hospitals NIHR Biomedical Research Center. M.v.S. has
nothing to declare.
References
1. Hannun AY, Rajpurkar P, Haghpanahi M, Tison GH, Bourn C, Turakhia MP, et al.
Cardiologist-level arrhythmia detection and classification in ambulatory electro-
cardiograms using a deep neural network. Nat Med 2019;25:65–69.
2. Perez MV, Mahaffey KW, Hedlin H, Rumsfeld JS, Garcia A, Ferris T, et al.
Large-scale assessment of a smartwatch to identify atrial fibrillation. N Engl J
Med 2019;381:1909–1917.
3. Guo Y, Wang H, Zhang H, Liu T, Liang Z, Xia Y, et al. Mobile photoplethysmo-
graphic technology to detect atrial fibrillation. J Am Coll Cardiol 2019;74:
2365–2375.
4. Kashou AH, Ko W-Y, Attia ZI, Cohen MS, Friedman PA, Noseworthy PA. A com-
prehensive artificial intelligence–enabled electrocardiogram interpretation pro-
gram. Cardiovasc Digit Health J 2020;1:62–70.
5. Adedinsewo D, Carter RE, Attia Z, Johnson P, Kashou AH, Dugan JL, et al.
Artificial intelligence-enabled ECG algorithm to identify patients with left ventri-
cular systolic dysfunction presenting to the emergency department with dyspnea.
Circ Arrhythmia Electrophysiol 2020;13:e008437.
6. Adedinsewo DA, Johnson PW, Douglass EJ, Attia IZ, Phillips SD, Goswami RM,
et al. Detecting cardiomyopathies in pregnancy and the postpartum period with
an electrocardiogram-based deep learning model. Eur Heart J - Digit Health
2021:ztab078.
7. Grogan M, Lopez-Jimenez F, Cohen-Shelly M, Dispenzieri A, Attia ZI, Abou
Ezzedine OF, et al. Artificial intelligence–enhanced electrocardiogram for the
early detection of cardiac amyloidosis. Mayo Clin Proc 2021;96:2768–2778.
8. Kwon J-M, Kim K-H, Medina-Inojosa J, Jeon K-H, Park J, Oh B-H. Artificial intelli-
gence for early prediction of pulmonary hypertension using electrocardiography. J
Heart Lung Transplant 2020;39:805–814.
9. Kwon J-M, Jung M-S, Kim K-H, Jo Y-Y, Shin J-H, Cho Y-H, et al. Artificial intelli-
gence for detecting electrolyte imbalance using electrocardiography. Ann
Noninvasive Electrocardiol 2021;26:e12839.
10. Galloway CD, Valys AV, Shreibati JB, Treiman DL, Petterson FL, Gundotra VP,
et al. Development and validation of a deep-learning model to screen for hyper-
kalemia from the electrocardiogram. JAMA Cardiol 2019;4:428–436.
11. Kwon J-M, Lee SY, Jeon K-H, Lee Y, Kim K-H, Park J, et al. Deep learning–based
algorithm for detecting aortic stenosis using electrocardiography. J Am Heart Assoc
2020;9:e014717.
12. Cohen-Shelly M, Attia ZI, Friedman PA, Ito S, Essayagh BA, Ko W-Y, et al.
Electrocardiogram screening for aortic valve stenosis using artificial intelligence.
Eur Heart J 2021;42:2885–2896.
13. Kwon J-M, Kim K-H, Akkus Z, Jeon K-H, Park J, Oh B-H, et al. Artificial intelligence
for detecting mitral regurgitation using electrocardiography. J Electrocardiol 2020;
59:151–157.
14. Attia ZI, Kapa S, Noseworthy PA, Lopez-Jimenez F, Friedman PA. Artificial intelli-
gence ECG to detect left ventricular dysfunction in COVID-19: a case series.
Mayo Clin Proc 2020;95:2464–2466.
15. Kwon J-M, Jeon K-H, Kim HM, Kim MJ, Lim SM, Kim K-H, et al. Comparing the
performance of artificial intelligence and conventional diagnosis criteria for de-
tecting left ventricular hypertrophy using electrocardiography. Europace 2020;
22:412–419.
16. Cho Y, Kwon J-M, Kim K-H, Medina-Inojosa JR, Jeon K-H, Cho S, et al. Artificial
intelligence algorithm for detecting myocardial infarction using six-lead electro-
cardiography. Sci Rep 2020;10:20495.
17. Makimoto H, Höckmann M, Lin T, Glöckner D, Gerguri S, Clasen L, et al.
Performance of a convolutional neural network derived from an ECG database
in recognizing myocardial infarction. Sci Rep 2020;10:8445.
18. Yao X, Rushlow DR, Inselman JW, McCoy RG, Thacher TD, Behnken EM, et al.
Artificial intelligence–enabled electrocardiograms for identification of patients
with low ejection fraction: a pragmatic, randomized clinical trial. Nat Med 2021;
27:815–819.
19. Attia ZI, Kapa S, Lopez-Jimenez F, McKie PM, Ladewig DJ, Satam G, et al.
Screening for cardiac contractile dysfunction using an artificial intelligence–en-
abled electrocardiogram. Nat Med 2019;25:70–74.
20. Antoniades C, Oikonomou EK. Artificial intelligence in cardiovascular imaging—
principles, expectations, and limitations. Eur Heart J 2021:ehab678.
21. RECOVERY Collaborative Group; Horby P, Lim WS, Emberson JR, Mafham M,
Bell JL, et al. Dexamethasone in hospitalized patients with Covid-19. N Engl J
Med 2021;384:693–704.
22. Shamout FE, Shen Y, Wu N, Kaku A, Park J, Makino T, et al. An artificial intelli-
gence system for predicting the deterioration of COVID-19 patients in the emer-
gency department. npj Digit Med 2021;4:80.
23. Shi F, Wang J, Shi J, Wu Z, Wang Q, Tang Z, et al. Review of artificial intelligence
techniques in imaging data acquisition, segmentation, and diagnosis for
COVID-19. IEEE Rev Biomed Eng 2021;14:4–15.
24. Eng D, Chute C, Khandwala N, Rajpurkar P, Long J, Shleifer S, et al. Automated
coronary calcium scoring using deep learning with multicenter external validation.
npj Digit Med 2021;4:88.
25. Zeleznik R, Foldyna B, Eslami P, Weiss J, Alexander I, Taron J, et al. Deep convo-
lutional neural networks to predict cardiovascular risk from computed tomogra-
phy. Nat Commun 2021;12:715.
26. Oren O, Gersh BJ, Bhatt DL. Artificial intelligence in medical imaging: switching
from radiographic pathological data to clinically meaningful endpoints. Lancet
Digit Health 2020;2:e486–e488.
27. Chilamkurthy S, Ghosh R, Tanamala S, Biviji M, Campeau NG, Venugopal VK, et al.
Deep learning algorithms for detection of critical findings in head CT scans: a ret-
rospective study. Lancet 2018;392:2388–2396.
28. Seidelmann SB, Claggett B, Bravo PE, Gupta A, Farhad H, Klein BE, et al. Retinal
vessel calibers in predicting long-term cardiovascular outcomes: the atheroscler-
osis risk in communities study. Circulation 2016;134:1328–1338.
29. Mitani A, Huang A, Venugopalan S, Corrado GS, Peng L, Webster DR, et al.
Detection of anaemia from retinal fundus images via deep learning. Nat Biomed
Eng 2020;4:18–27.
30. Rim TH, Lee G, Kim Y, Tham Y-C, Lee CJ, Baik SJ, et al. Prediction of systemic
biomarkers from retinal photographs: development and validation of
deep-learning algorithms. Lancet Digit Health 2020;2:e526–e536.
31. Rim TH, Lee CJ, Tham Y-C, Cheung N, Yu M, Lee G, et al. Deep-learning-based
cardiovascular risk stratification using coronary artery calcium scores predicted
from retinal photographs. Lancet Digit Health 2021;3:e306–e316.
32. Tran T, Huang NT, Montezuma SR. Smartphone funduscopy - How to use smart-
phone to take fundus photographs. https://eyewiki.aao.org/Smartphone_
Funduscopy_-_How_to_Use_Smartphone_to_Take_Fundus_Photographs (2
March 2018, date last accessed).
33. Bora A, Balasubramanian S, Babenko B, Virmani S, Venugopalan S, Mitani A, et al.
Predicting the risk of developing diabetic retinopathy using deep learning. Lancet
Digit Health 2021;3:e10–e19.
The year in cardiovascular medicine 2021 9
Downloaded from https://academic.oup.com/eurheartj/advance-article/doi/10.1093/eurheartj/ehab874/6487290 by guest on 07 January 2022
34. Canton G, Hippe DS, Chen L, Waterton JC, Liu W, Watase H, et al.
Atherosclerotic burden and remodeling patterns of the popliteal artery as de-
tected in the magnetic resonance imaging osteoarthritis initiative data set. JAm
Heart Assoc 2021;10:e18408.
35. Augusto JB, Davies RH, Bhuva AN, Knott KD, Seraphim A, Alfarih M, et al.
Diagnosis and risk stratification in hypertrophic cardiomyopathy using machine
learning wall thickness measurement: a comparison with human test-retest per-
formance. Lancet Digit Health 2021;3:e20–e28.
36. Peyster EG, Arabyarmohammadi S, Janowczyk A, Azarianpour-Esfahani S, Sekulic
M, Cassol C, et al. An automated computational image analysis pipeline for histo-
logical grading of cardiac allograft rejection. Eur Heart J 2021;42:2356–2369.
37. Zhang J, Gajjala S, Agrawal P, Tison GH, Hallock LA, Beussink-Nelson L, et al. Fully
automated echocardiogram interpretation in clinical practice: feasibility and diag-
nostic accuracy. Circulation 2018;138:1623–1635.
38. Madani A, Ong JR, Tibrewal A, Mofrad MRK. Deep echocardiography:
data-efficient supervised and semi-supervised deep learning towards automated
diagnosis of cardiac disease. npj Digit Med 2018;1:59.
39. Goto S, Mahara K, Beussink-Nelson L, Ikura H, Katsumata Y, Endo J, et al. Artificial
intelligence-enabled fully automated detection of cardiac amyloidosis using elec-
trocardiograms and echocardiograms. Nat Commun 2021;12:2726.
40. Ghorbani A, Ouyang D, Abid A, He B, Chen JH, Harrington RA, et al. Deep learn-
ing interpretation of echocardiograms. npj Digit Med 2020;3:10.
41. Narang A, Bae R, Hong H, Thomas Y, Surette S, Cadieu C, et al. Utility of a
deep-learning algorithm to guide novices to acquire echocardiograms for limited
diagnostic use. JAMA Cardiol 2021;6:624–632.
42. Wynants L, Van Calster B, Collins GS, Riley RD, Heinze G, Schuit E, et al.
Prediction models for diagnosis and prognosis of covid-19: systematic review
and critical appraisal. BMJ 2020;369:m1328.
43. van Smeden M, Reitsma JB, Riley RD, Collins GS, Moons KGM. Clinical prediction
models: diagnosis versus prognosis. J Clin Epidemiol 2021;132:142–145.
44. Christodoulou E, Ma J, Collins GS, Steyerberg EW, Verbakel JY, Van Calster B,
et al. A systematic review shows no performance benefit of machine learning
over logistic regression for clinical prediction models. J Clin Epidemiol 2019;110:
12–22.
45. Khera R, Haimovich J, Hurley NC, McNamara R, Spertus JA, Desai N, et al. Use of
machine learning models to predict death after acute myocardial infarction. JAMA
Cardiol 2021;6:633–641.
46. D’Ascenzo F, De Filippo O, Gallone G, Mittone G, Deriu MA, Iannaccone M, et al.
Machine learning-based prediction of adverse events following an acute coronary
syndrome (PRAISE): a modelling study of pooled datasets. Lancet 2021;397:
199–207.
47. Hyland SL, Faltys M, Hüser M, Lyu X, Gumbsch T, Esteban C, et al. Early predic-
tion of circulatory failure in the intensive care unit using machine learning. Nat
Med 2020;26:364–373.
48. Wilken M, Hüske-Kraus D, Klausen A, Koch C, Schlauch W, Röhrig R. Alarm fa-
tigue: causes and effects. Stud Health Technol Inform 2017;243:107–111.
49. Wijnberge M, Geerts BF, Hol L, Lemmers N, Mulder MP, Berge P, et al. Effect of a
machine learning-derived early warning system for intraoperative hypotension vs
standard care on depth and duration of intraoperative hypotension during elect-
ive noncardiac surgery: the HYPE randomized clinical trial. JAMA 2020;323:
1052–1060.
50. Hatib F, Jian Z, Buddi S, Lee C, Settels J, Sibert K, et al. Machine-learning algorithm
to predict hypotension based on high-fidelity arterial pressure waveform analysis.
Anesthesiology 2018;129:663–674.
51. Akyea RK, Qureshi N, Kai J, Weng SF. Performance and clinical utility of super-
vised machine-learning approaches in detecting familial hypercholesterolaemia
in primary care. npj Digit Med 2020;3:142.
52. Nanjo A, Evans H, Direk K, Hayward AC, Story A, Banerjee A. Prevalence, inci-
dence, and outcomes across cardiovascular diseases in homeless individuals using
national linked electronic health records. Eur Heart J 2020;41:4011–4020.
53. Mhasawade V, Zhao Y, Chunara R. Machine learning and algorithmic fairness in
public and population health. Nat Mach Intell 2021;3:659–666.
54. Ward A, Sarraju A, Chung S, Li J, Harrington R, Heidenreich P, et al. Machine
learning and atherosclerotic cardiovascular disease risk prediction in a multi-
ethnic population. npj Digit Med 2020;3:125.
55. Khurshid S, Weng L-C, Al-Alusi MA, Halford JL, Haimovich JS, Benjamin EJ, et al.
Accelerometer-derived physical activity and risk of atrial fibrillation. Eur Heart J
2021;42:2472–2483.
56. Bonnesen MP, Frodi DM, Haugan KJ, Kronborg C, Graff C, Højberg S, et al.
Day-to-day measurement of physical activity and risk of atrial fibrillation. Eur
Heart J 2021;42:3979–3988.
57. Bonnesen MP, Diederichsen SZ, Isaksen JL, Frederiksen KS, Hasselbalch SG,
Haugan KJ, et al. Atrial fibrillation burden and cognitive decline in elderly patients
undergoing continuous monitoring. Am Heart J 2021;242:15–23.
58. Stehlik J, Schmalfuss C, Bozkurt B, Nativi-Nicolau J, Wohlfahrt P, Wegerich S,
et al. Continuous wearable monitoring analytics predict heart failure hospitaliza-
tion: the LINK-HF multicenter study. Circ Heart Fail 2020;13:e006513.
59. Allen LA, Venechuk G, McIlvennan CK, Page RL, Knoepke CE, Helmkamp LJ, et al.
An electronically delivered patient-activation tool for intensification of medica-
tions for chronic heart failure with reduced ejection fraction: the EPIC-HF trial.
Circulation 2021;143:427–437.
60. Muhlestein JB, Anderson JL, Bethea CF, Severance HW, Mentz RJ, Barsness GW,
et al. Feasibility of combining serial smartphone single-lead electrocardiograms for
the diagnosis of ST-elevation myocardial infarction: smartphone ECG for STEMI
Diagnosis. Am Heart J 2020;221:125–135.
61. Spaccarotella CAM, Polimeni A, Migliarino S, Principe E, Curcio A, Mongiardo A,
et al. Multichannel electrocardiograms obtained by a smartwatch for the diagnosis
of ST-segment changes. JAMA Cardiol 2020;5:1176–1180.
10 P.E. Vardas et al.
Downloaded from https://academic.oup.com/eurheartj/advance-article/doi/10.1093/eurheartj/ehab874/6487290 by guest on 07 January 2022