If you don't remember your password, you can reset it by entering your email address and clicking the Reset Password button. You will then receive an email that contains a secure link for resetting your password
If the address matches a valid account an email will be sent to __email__ with instructions for resetting your password
Department of Cardiovascular Medicine, Mayo Clinic, Rochester, MN, USADivision of Epidemiology, Mayo Clinic, Rochester, MN, USADepartment of Quantitative Health Science, Mayo Clinic, Rochester, MN, USA
To demonstrate early aging in patients with lamin A/C (LMNA) gene mutations after hypothesizing that they have a biological age older than chronological age, as such a finding impacts care.
Patient and Methods
We applied a previously trained convolutional neural network model to predict biological age by electrocardiogram (ECG) [Artificial Intelligence (AI)-ECG age] to LMNA patients evaluated by multiple ECGs from January 1, 2003, to December 31, 2019. The age gap was the difference between chronological age and AI-ECG age. Findings were compared with age-/sex-matched controls.
Thirty-one LMNA patients who had a total of 271 ECGs were studied. The median age at symptom onset was 22 years (range, <1-53 years; n=23 patients); eight patients were asymptomatic family members carrying the LMNA mutation. Cardiac involvement was detected by ECG and echocardiogram in 16 patients and consisted of ventricular arrhythmias (13), atrial fibrillation (12), and cardiomyopathy (6). Four patients required cardiac transplantation. Fourteen patients had neurological manifestations, mainly muscular dystrophy. LMNA mutation carriers, including asymptomatic carriers, were 16 years older by AI-ECG than non-LMNA carriers, suggesting accelerated biological age. Most LMNA patients had an age gap of more than 10 years, compared with controls (P<.001). Consecutive AI-ECG analysis showed accelerated aging in the LMNA group compared with controls (P<.0001). There were no significant differences in age-gap among LMNA patients based on phenotype.
AI-ECG predicted that LMNA patients have a biological age older than chronological age and accelerated aging even in the absence of cardiac abnormalities by traditional methods. Such a finding could translate into early medical intervention and serve as a disease biomarker.
Mutations in the lamin A/C (LMNA) gene cause rare genetic disorders called laminopathies, which manifest with a spectrum of phenotypes spanning from arrhythmogenic cardiomyopathy to muscular dystrophy with various phenotypes, Charcot-Marie-Tooth disease, restrictive dermatopathy, mandibuloacral dysplasia, Hutchinson-Gilford progeria syndrome, or overlapping phenotypes.
A few hundred LMNA mutations have been reported, scattered throughout the entire gene. More than 90% of patients with Hutchinson-Gilford progeria syndrome, which is characterized by early aging and premature death due to myocardial infarction and stroke, carry a heterozygous de novo mutation, LMNA (c.1824C>T, p. Gly608Gly) in exon 11. This specific LMNA variant alters normal splicing and results in a shorter lamin A called progerin, which is thought to be responsible for the underlying molecular defect in progeria.
Patients with LMNA mutations associated with cardiac phenotypes usually have a rapid disease course and carry a high risk of sudden cardiac death, mainly due to malignant ventricular arrhythmias or severe heart failure, which occurs at higher frequency and earlier age than other inherited cardiomyopathies.
Early detection of cardiac involvement and early diagnosis of laminopathy can be challenging, as patients may present with mild symptoms or be asymptomatic until sudden death. It is also known that asymptomatic patients with LMNA mutations can display abnormal cardiac findings, such as atrioventricular conduction disorders, atrial and ventricular tachyarrhythmias, reduced ventricular function, myocardial fibrosis, or late gadolinium myocardial enhancement by magnetic resonance.
We recently created a machine learning algorithm that predicts age using the 12-lead electrocardiogram (ECG) artificial intelligence (AI), also known as AI-ECG age. This biomarker has been suggested to signal physiologic age.
Additionally, our studies and those from other investigators have shown that AI-ECG age and the difference between AI-ECG age and chronological age (also known as the age gap) correlate with total and cardiovascular mortality and may possibly reflect biological aging.
In this study, we hypothesized that LMNA mutation carriers display an increased biological age, as suggested by an increased AI-ECG age gap, showing accelerated aging also in those without cardiac symptoms or abnormal cardiac findings by traditional methods, compared to age- and sex-matched controls.
Patients And Methods
The study was approved by the Mayo Institutional Review Board. Only patients who had previously agreed to include their data in a retrospective chart review research study were included. The study follows Strengthening the Reporting of Observational Studies in Epidemiology reporting guidelines.
In this cross-sectional and observational study, we retrospectively reviewed medical records from patients with LMNA mutations who underwent at least a 12-lead ECG at our institution between January 1, 2003, and December 31, 2019. We included only patients evaluated in our cardiology and neurology clinics. In asymptomatic family members carrying the LMNA mutation, ECGs were performed as part of the initial screening after detection of the familial mutation. Controls were selected from 8,898,622 consecutive patients not known to have LMNA mutations and seen at Mayo Clinic from July 1987 to 2020. Those without research authorization were excluded per Minnesota statute.
We randomly selected five age-/sex-matched ECG controls per each patient’s ECG. For control pediatric ECGs, we excluded patients with congenital heart diseases (such as ventricular septal defects, atrial septal defects, or tetralogy of Fallot) and cardiac transplants. For control adult ECGs, we did not exclude cardiac pathologies so that they would represent the general population and expected ECG age. We excluded ECGs from the original ECG age derivation set used for the ECG age algorithm creation.
We collected clinical information at first evaluation and follow-up from the electronic medical record, including age at symptom onset, symptoms at onset, clinical findings, and genetic data. Patients who underwent genetic testing and cardiac evaluation as part of family screening and had not reported symptoms at the time of genetic testing were defined as asymptomatic. The 12-lead ECG was recorded for 10 seconds and sampled at a rate of 250 Hz or 500 Hz using a GE-Marquette ECG machine (Marquette, WI, USA). ECG data were stored as 500-Hz signals in the MUSE data management system at Mayo Clinic.
Overview on the AI-ECG Model
A convolutional neural network model was previously developed using the Keras with a Tensorflow (Google, Mountain View, CA, USA) and Python backend. Detailed methods are reported elsewhere.
Briefly, a total of 774,783 unique subjects with ECG scans were used to develop the neural network: 399,750 in the training set, 99,977 in the internal validation set, and 275,056 ECGs in the holdout testing set. The training, validation, and test sets were mutually exclusive for patient identification. The convolutional neural network architecture was divided into blocks to which the ECG matrix was injected. Each block was followed by a nonlinear activation function. After the first group of blocks extracted temporal features, another spatial block was used to fuse data from all leads, and then the extracted features were used in a fully connected layer to predict age with output as a single number. The convolutional neural network was trained by inputting 10-second samples of resting, digital, standard 12-lead ECGs that were digitally stored within the Mayo Clinic digital vault and the patients’ chronological age at the time of the ECG, during the training process, and the weights of the convolutional filters were adjusted to extract meaningful and relevant features of the inputs in respect to the patients’ age. The network had a single output (age) as a continuous number. We used the previously developed AI-ECG algorithm with no additional retraining to assess ECG-age in our study population. The ECG-age was calculated to all observations available for LMNA cases and matched controls. We defined age gap as the difference between the AI-ECG predicted age minus chronological age. A positive age gap was determined when an AI-ECG–estimated age was older than the chronological age.
Patients and ECG characteristics were summarized with frequencies and percentages or means ± SDs, or median when not normally distributed. The Kolmogorov-Smirnov test was used to test the normality of continuous variables. The χ2 or Fischer exact test was used to compare categorical variables, and the Mann-Whitney U test or Kruskal–Wallis test for continuous variables, when appropriate, visualized with Violin plots across cases and controls. Correlation was assessed using Pearson or Spearman tests as appropriate. We created locally weighted error sum of squares (loess) curves for LMNA patients and controls and for prespecified study subgroups using chronological age as the time scale. Loess regression is a nonparametric approach useful for evaluating trends for nonlinear continuous data observed over time. Resulting loess splines assess data points across different time points, fitting local regression lines computed from observed data (missing values are omitted) and connecting these lines to display a smooth line. Missing data due to loss to follow-up or due to unavailable ECGs were assumed to be at random. We created locally weighted error sum of squares (Loess) curves for LMNA patients and controls and for pre-specified study subgroups. The association between LMNA AI-ECG age with an age gap of more than 10 years (ie, older biological age) was assessed with Kaplan-Meier curves that displayed chronological age as the time scale; no further multivariate analysis was performed due to limited sample size. Differences between groups were analyzed using the log-rank tests. For all tests, a two-tailed P<.05 was considered statistically significant. All statistical analyses were performed using JMP Pro software (SAS Institute, Inc, Cary, NC, USA) and R version 3.2.3 (R Foundation for Statistical Computing, Vienna, Austria). The case figure was created using Phyton package matplotlib.
All requests for raw and analyzed data and related materials, excluding programming code, will be reviewed by the Mayo Clinic legal department and Mayo Clinic Ventures to verify whether the request is subject to any intellectual property or confidentiality obligations. Any data and materials that can be shared will be released via a Material Transfer Agreement.
Demographics and Clinical Symptoms
We identified 31 genetically confirmed LMNA patients from 28 families who had 12-lead ECGs (n=271) at our institution electronically available for analysis. Table 1 and Supplemental Table 1 (available online at http://www.mayoclinicproceedings.org) summarize demographics, symptoms, neurological and cardiac features, cardiovascular risks factors, and genetic findings. Twenty-three of 31 patients were symptomatic. Presenting symptoms were cardiac (n=12), neurological (n=10), or abnormal adipose tissue distribution (n=1).
Table 1Demographics, symptoms, findings, and cardiovascular risk factors of patients with LMNA mutations
Cardiac symptoms were heterogeneous, mainly consisting of exertional dyspnea and palpitations (Table 1), and preceded the skeletal myopathy onset in four patients (2 males and 2 females). Neurological symptoms included muscle weakness due to skeletal myopathy (n=7; 3 of whom presented with hypotonia at birth and/or delayed motor development), chronic myalgia and fatigue (n=1), rhabdomyolysis (n=1), and stroke (n=1). The patient with rhabdomyolysis (patient 7), in addition to the LMNA mutation, carried a coexistent homozygous pathogenic variant in the fukutin-related protein (FKRP) gene likely responsible for the rhabdomyolysis. Three of eight patients manifesting with skeletal myopathy showed cardiac involvement at time of neurological evaluation or follow-up. Median age at symptom onset was 22 years (range, <1 to 53 years) and there was no difference between males and females (P=.6). Indeed, it was 21.5 years (range, 3 months to 48 years) in males and 26.5 years (range, 4 to 53 years) in females (P=.3). Patients presenting with neurological manifestations were significantly younger (median age, 11 years) than those presenting with cardiac symptoms (median age, 36 years) (Table 1). Among those presenting with cardiac symptoms, there was no difference in age at symptom onset between males and females (males: median age 42 years, range: 41-44 years; females: 33 years, range: 5-53 years; P=.2). Median time from development of neurological symptoms to cardiac symptoms was 2 years (range, 2 to 34 years). Conversely, median time from development of cardiac symptoms to neurological symptoms was 11.5 years (range, 5 months to 26 years). Median time from symptom onset to genetic diagnosis was 5 years (range, <1 to 39 years) in patients who manifested with myopathy, and 10 years (range, 3 months to 23 years) in patients who presented with cardiac symptoms.
Cardiac and Neurological Findings
Cardiac involvement, as indicated by ECG or echocardiographic abnormalities, was present in 51.6% of patients (n=16; 7 males) with median age of 27.5 years (range, 3 to 53 years). The most common ECG abnormalities are summarized in Tables 1 and Supplementary Table 2 (available online at http://www.mayoclinicproceedings.org). Median ECG follow-up time from symptom onset to last follow-up was 86 months (range, <1 to 367 months) and was longer in LMNA patients with cardiac manifestations. The most frequent abnormality was ventricular arrhythmias (13 patients), followed by atrial fibrillation in 12 patients, six of whom required permanent pacemaker placement. Median age of pacemaker implantation was 36 years (range, 17 to 62 years). Fifteen patients underwent implantable cardioverter-defibrillator (ICD) placement for prevention of lethal ventricular arrhythmias, including two asymptomatic individuals with family history of sudden cardiac death. All patients with cardiomyopathy had concomitant atrial fibrillation and ventricular arrhythmia. Orthotopic cardiac transplantation was performed in four patients who presented with cardiac symptoms at a median age of onset of 29.5 years (range, 19 to 37 years). Two patients died from cardiac causes during follow-up, one at the age of 56 years from ventricular arrhythmia and one at age of 45 years from end-stage congestive cardiomyopathy. Both patients had an ICD that had been implanted at age 34 and 36 years of age, respectively.
Fourteen (45.2%) patients had neurological involvement at diagnosis or during follow-up. Skeletal myopathy included limb-girdle muscular dystrophy (n=7), Emery-Dreyfuss muscular dystrophy (n=3), and congenital muscular dystrophy (n=2). Three patients had a length-dependent sensory neuropathy accompanying the myopathy; in one patient from this group the neuropathy was likely due to diabetes. One patient had peripheral neuropathy without myopathy. Three patients had an embolic ischemic stroke (patients 4 ,7, and 13). The stroke occurred at disease-onset in one patient (patient 13), in the setting of newly diagnosed atrial fibrillation in another patient previously diagnosed with LMNA muscular dystrophy (patient 7), and in one patient with progeroid phenotype (patient 4).
Correlation Between ECG Age and Chronological Age
wTo each LMNA patient’s ECG, we matched five age/sex individual controls. This resulted in 271 LMNA and 1252 distinct control ECGs (Supplementary Table 2). LMNA patients had a predicted biological age significantly older than controls (median of 22 years in LMNA vs median of 6 years in controls) (Figure 1A, Supplementary Table 2). There was no difference in AI-ECG age gap based on sex in the LMNA patients compared with controls (LMNA males: median age gap was 21 years with a range of –12 to 41 years in LMNA vs 6 years with a range of –22 to 37 years in controls [P<.001]; LMNA females: median age gap was 24 years with a range of –20 to 39 years in LMNA vs 6 years with a range of –16 to 39 years in controls [P<.001]). Most (78%) LMNA patients had a positive age gap of more than 10 years compared to controls (31%) (P<.001; Supplementary Table 2). Consecutive serial analysis of all ECGs showed “accelerated aging” in the LMNA group compared with controls (P<.0001) (Figures 1B and 2).
To capture patients with minimal-to-no cardiac involvement, we performed a subgroup analysis to compare patient’s first ECG with one ECG from each of five age-/sex-matched controls to show higher age gap in LMNA group (Figure 3). We noticed the model detected the higher age gap between the ages of 35 and 45 years. To narrow down the age at which the age gap main difference occurred, we performed a subset analysis in the LMNA group comparing patients younger and older than the age of 40 years. Patients with LMNA mutations who were older than 40 years had a significantly higher age gap (median, 70 years) than those younger than 40 years (median, 48 years; P<.001). Stratification of the LMNA cohort by sex showed no significant difference in age gap between females vs males (median, 24 years with a range of –20 to 39 years in females vs median, 21 years with a range of –12 to 41 years in males [P=.63]). We also tested the ECG-age model in LMNA patients with cardiac transplant or LMNA patients whose ECG was interpreted as normal (n=26) and compared findings with 559 control normal ECGs (Figure 4) to show higher age gap in the LMNA group. We performed another subgroup analysis limited to the eight asymptomatic LMNA individuals (total of 21 ECGs) and sex-/age-matched controls (n=115) showing significantly higher AI-ECG age gap in the LMNA patients (median age gap of 18 years with a range of 7 to 38 years vs 9 years among controls with a range of –7 to 37 years [P<.001]).
There were no significant differences in age gap in LMNA patients manifesting with cardiac disease (n=16) compared with patients with no symptoms or isolated skeletal myopathy (n=15). The median age gap was 16 years in both groups (cardiac group, range of –20 to 41 years vs noncardiac group, range of –4 to 37 years [P=.39]).
We also compared the estimated age gap in two patients with an LMNA mutation previously reported in progeroid phenotypes (patients 4 and 21, Supplementary Table 1) vs other LMNA patients with cardiomyopathy and muscular dystrophy. The median age gap was not statistically significant between these two groups, but the sample size is too small to make reliable conclusions (median 11 years with a range of 2 to 20 years in the progeroid vs median of 22 years with a range of –20 to 41 years for the latter [P=.29])
In this study, by comparing predicted AI-ECG–derived biological age in LMNA patients and controls, we show that AI-ECG can demonstrate older and accelerated ageing in individuals with mutations in a gene associated with senescence. We show that AI-ECG–predicted age is older not only in LMNA patients with cardiac symptoms and abnormal cardiac findings, but also in LMNA carriers without symptoms or abnormal cardiac findings by traditional methods. A representative example of patients with accelerated aging, despite the lack of abnormal ECG findings by traditional interpretation and lack of clinical progeroid features, is patient 5 (Figure 4C, Supplementary Table 1). Interestingly, most of our patients do not have mutations previously associated with progeria or progeroid syndrome, in which cardiac involvement has a vascular basis.
Older biological age was also predicted in patients with dilated cardiomyopathy as the main cardiac feature and no evidence of cardiovascular disease. The median age gap was not statistically significant when comparing the two LMNA patients with progeroid features (patients 4 and 21, Supplementary Table 1) to other LMNA patients with cardiomyopathy or muscular dystrophy. Based on these observations, it is very likely that the AI-ECG captured a biological abnormality related to the laminopathy itself, rather than showing early detection of cardiac or cardiovascular involvement. Further longitudinal studies and larger patient cohorts would be informative in this regard. The older AI-ECG age and accelerated aging would be in keeping with the known role of lamins in aging and perhaps related to the shorter expected life span of LMNA mutation carriers.
Whereas previous studies have shown that AI-enabled ECG age can predict cardiac disease risk in normal population and a few acquired diseases, to our knowledge, our work demonstrates for the first time that AI-enabled ECG age can be a potential biomarker in a genetic disorder caused by a mutated aging-involved protein by capturing older and accelerated aging in individuals with LMNA mutations.
Such findings would benefit from further validation in larger LMNA cohorts and complementary studies to investigate specificity by targeting other genetic disorders featured by similar phenotype. Nevertheless, our results of a positive age gap in a laminopathy patient could have clinical applications and be leveraged, for example, to inform about clinical response to the newly proposed genetic therapies. Antisense oligonucleotide and in vivo base editing therapies in LMNA disorders, for which timing of treatment is critical to optimize clinical response, could find in the AI-ECG model a noninvasive tool to monitor such response.
Potential future applications of AI-ECG could include its clinical use to support potential pathogenicity of LMNA variants of unknown significance, although future studies are needed to investigate specificity of findings.
Prior methods to assess biological age included many proposed mathematical methods with different selection criteria as biomarkers for age.
All of these models needed an input using primarily linear mathematics, whereas our machine-learning method uses deep neural networks addressing nonlinear dynamics. We believe that our proposed methodology represents a new paradigm in the assessment of biological age that is partially supported on the basic principles of prior methods.
Our results show that the raw ECG signals can identify chronological age and, more importantly, differential aging rates in a genetic disorder caused by mutations in LMNA, a gene encoding a protein known to be involved in aging.
Potential Competing Interests
The authors report no potential competing interests.
The funders had no role in the design and conduct of the study; collection, management, analysis, or interpretation of the data; preparation, review, or approval of the manuscript; or decision to submit the manuscript for publication.