Machine Learning to Predict the Likelihood of Acute Myocardial Infarction

Diagnostic accuracy of a machine learning algorithm using point-of-care high-sensitivity cardiac troponin I for rapid rule-out of myocardial infarction: a retrospective study

BACKGROUND

Point-of-care (POC) high-sensitivity cardiac troponin (hs-cTn) assays have been shown to provide similar analytical precision despite substantially shorter turnaround times compared with laboratory-based hs-cTn assays. We applied the previously developed machine learning based personalised Artificial Intelligence in Suspected Myocardial Infarction Study (ARTEMIS) algorithm, which can predict the individual probability of myocardial infarction, with a single POC hs-cTn measurement, and compared its diagnostic performance with standard-of-care pathways for rapid rule-out of myocardial infarction.

METHODS

We retrospectively analysed pooled data from consecutive patients of two prospective observational cohorts in geographically distinct regions (the Safe Emergency Department Discharge Rate cohort from the USA and the Suspected Acute Myocardial Infarction in Emergency cohort from Australia) who presented to the emergency department with suspected myocardial infarction. Patients with ST-segment elevation myocardial infarction were excluded. Safety and efficacy of direct rule-out of myocardial infarction by the ARTEMIS algorithm (at a pre-specified probability threshold of <0·5%) were compared with the European Society of Cardiology (ESC)-recommended and the American College of Cardiology (ACC)-recommended 0 h pathways using a single POC high-sensitivity cardiac troponin I (hs-cTnI) measurement (Siemens Atellica VTLi as investigational assay). The primary diagnostic outcome was an adjudicated index diagnosis of type 1 or type 2 myocardial infarction according to the Fourth Universal Definition of Myocardial Infarction. The safety outcome was a composite of incident myocardial infarction and cardiovascular death (follow-up events) at 30 days. Additional analyses were performed for type I myocardial infarction only (secondary diagnostic outcome), and for each cohort separately. Subgroup analyses were performed for age (<65 years vs ≥65 years), sex, symptom onset (≤3 h vs >3 h), estimated glomerular filtration rate (<60 mL/min per 1·73 m2vs ≥60 mL/min per 1·73 m2), and absence or presence of arterial hypertension, diabetes, a history of coronary artery disease, myocardial infarction, or heart failure, smoking, and ischaemic electrocardiogram signs.

FINDINGS

Among 2560 patients (1075 [42%] women, median age 58 years [IQR 48·0-69·0]), prevalence of myocardial infarction was 6·5% (166/2560). The ARTEMIS-POC algorithm classified 899 patients (35·1%) as suitable for rapid rule-out with a negative predictive value of 99·96% (95% CI 99·64-99·96) and a sensitivity of 99·68% (97·21-99·70). For type I myocardial infarction only, negative predictive value and sensitivity were both 100%. Proportions of missed index myocardial infarction (0·05% [0·04-0·42]) and follow-up events at 30 days (0·07% [95% CI 0·06-0·59]) were low. While maintaining high safety, the ARTEMIS-POC algorithm identified more than twice as many patients as eligible for direct rule-out compared with guideline-recommended ESC 0 h (15·2%) and ACC 0 h (13·8%) pathways. Superior efficacy persisted across all clinically relevant subgroups.

INTERPRETATION

The patient-tailored, medical decision support ARTEMIS-POC algorithm applied with a single POC hs-cTnI measurement allows for very rapid, safe, and more efficient direct rule-out of myocardial infarction than guideline-recommended pathways. It has the potential to expedite the safe discharge of low-risk patients from the emergency department including early presenters with symptom onset less than 3 h at the time of admission and might open new opportunities for the triage of patients with suspected myocardial infarction even in ambulatory, preclinical, or geographically isolated care settings.

FUNDING

The German Center for Cardiovascular Research (DZHK).

Sex-specific high-sensitivity troponin T cut-points have similar safety but lower efficacy than overall cut-points in a multisite U.S. cohort

BACKGROUND

Data comparing the performance of sex-specific to overall (non-sex-specific) high-sensitivity cardiac troponin (hs-cTn) cut-points for diagnosing acute coronary syndrome (ACS) are limited. This study aims to compare the safety and efficacy of sex-specific versus overall 99th percentile high-sensitivity cardiac troponin T (hs-cTnT) cut-points.

METHODS

We conducted a secondary analysis of the STOP-CP cohort, which prospectively enrolled emergency department patients ≥ 21 years old with symptoms suggestive of ACS without ST-elevation on initial electrocardiogram across eight U.S. sites (January 25, 2017-September 6, 2018). Participants with both 0- and 1-h hs-cTnT measures less than or equal to the 99th percentile (sex-specific 22 ng/L for males, 14 ng/L for females; overall 19 ng/L) were classified into the rule-out group. The safety outcome was adjudicated cardiac death or myocardial infarction (MI) at 30 days. Efficacy was defined as the proportion classified to the rule-out group. McNemar's test and a generalized score statistic were used to compare rule-out and 30-day cardiac death or MI rates between strategies. Net reclassification improvement (NRI) index was used to further compare performance.

RESULTS

This analysis included 1430 patients, of whom 45.8% (655/1430) were female; the mean ± SD age was 57.6 ± 12.8 years. At 30 days, cardiac death or MI occurred in 12.8% (183/1430). The rule-out rate was lower using sex-specific versus overall cut-points (70.6% [1010/1430] vs. 72.5% [1037/1430]; p = 0.003). Among rule-out patients, the 30-day cardiac death or MI rates were similar for sex-specific (2.4% [24/1010]) vs. overall (2.3% [24/1037]) strategies (p = 0.79). Among patients with cardiac death or MI, sex-specific versus overall cut-points correctly reclassified three females and incorrectly reclassified three males. The sex-specific strategy resulted in a net of 27 patients being incorrectly reclassified into the rule-in group. This led to an NRI of -2.2% (95% CI -5.1% to 0.8%).

CONCLUSIONS

Sex-specific hs-cTnT cut-points resulted in fewer patients being ruled out without an improvement in safety compared to the overall cut-point strategy.

Safety and feasibility of triage and rapid discharge of patients with chest pain from emergency room: a pragmatic, randomised non-inferiority control trial of the European Society of Cardiology (ESC) 0-1 hour pathway vs. conventional 0-3 hour accelerated diagnostic protocol

Patients presenting with chest pain represent a significant proportion of Emergency Department (ED) attendances but only a minority, typically 10%, have a final diagnosis of myocardial infarction (MI). Prompt discharge of patients without MI will alleviate ED overcrowding as well as improve patient satisfaction and reduce exposure to risk of hospital acquired infections such as Covid 19. The measurement of cardiac troponin (cTn) by a high sensitivity method is recommended by the National Institute for health and Care Excellence (NICE) for rapid categorisation of patients presenting with chest pain. Strategies proposed include measurement on admission and one hour from admission (ESC 0-1-hour pathway, the recent guideline approved pathway which has not been implemented widely), and measurement on admission and three hours from admission (0-3-hour pathway, which is conventional and widely adopted). The primary objective of this study is twofold: firstly, to assess the safety, feasibility, and impact of implementing the ESC (European Society of Cardiology) 0-1-hour pathway in clinical practice by reference to the more established ESC 0-3-hour protocol. The principal outcome measure will be the safety of the ESC 0-1-hour protocol. However, there are concerns that the time from sample draw to result availability (typically around 60 minutes) will impact on the feasibility of the ESC 0-1-hour pathway. Secondly, therefore, our goal is to evaluate whether measurement of high sensitivity troponin by a bedside analyser (point of care testing, POCT), which will produce results in 15 minutes is a feasible alternative to laboratory testing. We will compare the results produced by POCT with the laboratory results in the context of the ESC 0-1 hour and 0-3-hour pathway, as a nested controlled study in the context of a randomised controlled trial. (clinicaltrials.gov: NCT05322395).

Ambient Mass Spectrometry and Machine Learning-Based Diagnosis System for Acute Coronary Syndrome

Aims: The purpose of this study is to establish a novel diagnosis system in early acute coronary syndrome (ACS) using probe electrospray ionization-mass spectrometry (PESI-MS) and machine learning (ML) and to validate the diagnostic accuracy. Methods: A total of 32 serum samples derived from 16 ACS patients and 16 control patients were analyzed by PESI-MS. The acquired mass spectrum dataset was subsequently analyzed by partial least squares (PLS) regression to find the relationship between the two groups. A support vector machine, an ML method, was applied to the dataset to construct the diagnostic algorithm. Results: Control and ACS groups were separated into the two clusters in the PLS plot, indicating ACS patients differed from the control in the profile of serum composition obtained by PESI-MS. The sensitivity, specificity, and accuracy of our diagnostic system were all 93.8%, and the area under the receiver operating characteristic curve showed 0.965 (95% CI: 0.84-1). Conclusion: The PESI-MS and ML-based diagnosis system are likely an optimal solution to assist physicians in ACS diagnosis with its remarkably predictive accuracy.

External validation of the myocardial-ischaemic-injury-index machine learning algorithm for the early diagnosis of myocardial infarction: a multicentre cohort study

BACKGROUND

The myocardial-ischaemic-injury-index (MI3) is a novel machine learning algorithm for the early diagnosis of type 1 non-ST-segment elevation myocardial infarction (NSTEMI). The performance of MI3, both when using early serial blood draws (eg, at 1 h or 2 h) and in direct comparison with guideline-recommended algorithms, remains unknown. Our aim was to externally validate MI3 and compare its performance with that of the European Society of Cardiology (ESC) 0/1h-algorithm.

METHODS

In this secondary analysis of a multicentre international diagnostic cohort study, adult patients (age >18 years) presenting to the emergency department with symptoms suggestive of myocardial infarction were prospectively enrolled from April 21, 2006, to Feb 27, 2019 in 12 centres from five European countries (Switzerland, Spain, Italy, Poland, and Czech Republic). Patients were excluded if they presented with ST-segment-elevation myocardial infarction, did not have at least two serial high-sensitivity cardiac troponin I (hs-cTnI) measurements, or if the final diagnosis remained unclear. The final diagnosis was centrally adjudicated by two independent cardiologists using all available medical records, including serial hs-cTnI measurements and cardiac imaging. The primary outcome was type 1 NSTEMI. The performance of MI3 was directly compared with that of the ESC 0/1h-algorithm.

FINDINGS

Among 6487 patients, (median age 61·0 years [IQR 49·0-73·0]; 2122 [33%] female and 4365 [67%] male), 882 (13·6%) patients had type 1 NSTEMI. The median time difference between the first and second hs-cTnI measurement was 60·0 mins (IQR 57·0-70·0). MI3 performance was very good, with an area under the receiver-operating-characteristic curve of 0·961 (95% CI 0·957 to 0·965) and a good overall calibration (intercept -0·09 [-0·2 to 0·02]; slope 1·02 [0·97 to 1·08]). The originally defined MI3 score of less than 1·6 identified 4186 (64·5%) patients as low probability of having a type 1 NSTEMI (sensitivity 99·1% [95% CI 98·2 to 99·5]; negative predictive value [NPV] 99·8% [95% CI 99·6 to 99·9]) and an MI3 score of 49·7 or more identified 915 (14·1%) patients as high probability of having a type 1 NSTEMI (specificity 95·0% [94·3 to 95·5]; positive predictive value [PPV] 69·1% [66·0-72·0]). The sensitivity and NPV of the ESC 0/1h-algorithm were higher than that of MI3 (difference for sensitivity 0·88% [0·19 to 1·60], p=0·0082; difference for NPV 0·18% [0·05 to 0·32], p=0·016), and the rule-out efficacy was higher for MI3 (11% difference, p<0·0001). Specificity and PPV for MI3 were superior (difference for specificity 3·80% [3·24 to 4·36], p<0·0001; difference for PPV 7·84% [5·86 to 9·97], p<0·0001), and the rule-in efficacy was higher for the ESC 0/1h-algorithm (5·4% difference, p<0·0001).

INTERPRETATION

MI3 performs very well in diagnosing type 1 NSTEMI, demonstrating comparability to the ESC 0/1h-algorithm in an emergency department setting when using early serial blood draws.

FUNDING

Swiss National Science Foundation, Swiss Heart Foundation, the EU, the University Hospital Basel, the University of Basel, Abbott, Beckman Coulter, Roche, Idorsia, Ortho Clinical Diagnostics, Quidel, Siemens, and Singulex.

Symptoms and coronary risk factors predictive of adverse cardiac events in chest pain patients in an Asian emergency department: the need for a local prediction score

INTRODUCTION

Clinical assessment is pivotal in diagnosing acute coronary syndrome. Our study aimed to identify clinical characteristics predictive of major adverse cardiac events (MACE) in an Asian population and to derive a risk score for MACE.

METHODS

Patients presenting to the emergency department (ED) with chest pain and non-diagnostic 12-lead electrocardiograms were recruited. Clinical history was recorded in a predesigned template. Random glucose and direct low-density lipoprotein measurements were taken, in addition to serial troponin. We derived the age, coronary risk factors (CRF), sex and symptoms (ACSS) risk score based on multivariate analysis results, considering age, CRF, sex and symptoms and classifying patients into very low, low, moderate and high risk for MACE. Comparison was made with the ED Assessment of Chest Pain Score (EDACS) and the history, electrocardiogram, age, risk factors, troponin (HEART) score. We also modified the HEART score with the CRF that we had identified. The outcomes were 30-day and 1-year MACE.

RESULTS

There were a total of 1689 patients, with 172 (10.2%) and 200 (11.8%) having 30-day and 1-year MACE, respectively. Symptoms predictive of MACE included central chest pain, radiation to the jaw/neck, associated diaphoresis, and symptoms aggravated by exertion and relieved by glyceryl trinitrate. The ACSS score had an area under the curve of 0.769 (95% confidence interval [CI]: 0.735-0.803) and 0.760 (95% CI: 0.727-0.793) for 30-day and 1-year MACE, respectively, outperforming EDACS. Those in the very-low-risk and low-risk groups had <1% risk of 30-day MACE.

CONCLUSION

The ACSS risk score shows potential for use in the local ED or primary care setting, potentially reducing unnecessary cardiac investigations and admission.

Contrast-Enhanced Magnetic Resonance Imaging Based T1 Mapping and Extracellular Volume Fractions Are Associated with Peripheral Artery Disease

BACKGROUND

Extracellular volume fraction (ECV), measured with contrast-enhanced magnetic resonance imaging (CE-MRI), has been utilized to study myocardial fibrosis, but its role in peripheral artery disease (PAD) remains unknown. We hypothesized that T1 mapping and ECV differ between PAD patients and matched controls.

METHODS AND RESULTS

A total of 37 individuals (18 PAD patients and 19 matched controls) underwent 3.0T CE-MRI. Skeletal calf muscle T1 mapping was performed before and after gadolinium contrast with a motion-corrected modified look-locker inversion recovery (MOLLI) pulse sequence. T1 values were calculated with a three-parameter Levenberg-Marquardt curve fitting algorithm. ECV and T1 maps were quantified in five calf muscle compartments (anterior [AM], lateral [LM], and deep posterior [DM] muscle groups; soleus [SM] and gastrocnemius [GM] muscles). Averaged peak blood pool T1 values were obtained from the posterior and anterior tibialis and peroneal arteries. T1 values and ECV are heterogeneous across calf muscle compartments. Native peak T1 values of the AM, LM, and DM were significantly higher in PAD patients compared to controls (all p < 0.028). ECVs of the AM and SM were significantly higher in PAD patients compared to controls (AM: 26.4% (21.2, 31.6) vs. 17.3% (10.2, 25.1), p = 0.046; SM: 22.7% (19.5, 27.8) vs. 13.8% (10.2, 19.1), p = 0.020).

CONCLUSIONS

Native peak T1 values across all five calf muscle compartments, and ECV fractions of the anterior muscle group and the soleus muscle were significantly elevated in PAD patients compared with matched controls. Non-invasive T1 mapping and ECV quantification may be of interest for the study of PAD.

From pixels to prognosis: Imaging biomarkers for discrimination and outcome prediction of pulmonary embolism : Original Research Article

PURPOSE

Recent advancements in medical imaging have transformed diagnostic assessments, offering exciting possibilities for extracting biomarker-based information. This study aims to investigate the capabilities of a machine learning classifier that incorporates dual-energy computed tomography (DECT) radiomics. The primary focus is on discerning and predicting outcomes related to pulmonary embolism (PE).

METHODS

The study included 131 participants who underwent pulmonary artery DECT angiography between January 2015 and March 2022. Among them, 104 patients received the final diagnosis of PE and 27 patients served as a control group. A total of 107 radiomic features were extracted for every case based on DECT imaging. The dataset was divided into training and test sets for model development and validation. Stepwise feature reduction identified the most relevant features, which were used to train a gradient-boosted tree model. Receiver operating characteristics analysis and Cox regression tests assessed the association of texture features with overall survival.

RESULTS

The trained machine learning classifier achieved a classification accuracy of 0.94 for identifying patients with acute PE with an area under the receiver operating characteristic curve of 0.91. Radiomics features could be valuable for predicting outcomes in patients with PE, demonstrating strong prognostic capabilities in survival prediction (c-index, 0.991 [0.979-1.00], p = 0.0001) with a median follow-up of 130 days (IQR, 38-720). Notably, the inclusion of clinical or DECT parameters did not enhance predictive performance.

CONCLUSION

In conclusion, our study underscores the promising potential of leveraging radiomics on DECT imaging for the identification of patients with acute PE and predicting their outcomes. This approach has the potential to improve clinical decision-making and patient management, offering efficiencies in time and resources by utilizing existing DECT imaging without the need for an additional scoring system.

[Digitalization and clinical decision tools]

Digitalization in cardiovascular emergencies is rapidly evolving, analogous to the development in medicine, driven by the increasingly broader availability of digital structures and improved networks, electronic health records and the interconnectivity of systems. The potential use of digital health in patients with acute chest pain starts even in the prehospital phase with the transmission of a digital electrocardiogram (ECG) as well as telemedical support and digital emergency management, which facilitate optimization of the rescue pathways and reduce critical time intervals. The increasing dissemination and acceptance of guideline apps and clinical decision support tools as well as integrated calculators and electronic scores are anticipated to improve guideline adherence, translating into a better quality of treatment and improved outcomes. Implementation of artificial intelligence to support image analysis and also the prediction of coronary artery stenosis requiring interventional treatment or impending cardiovascular events, such as heart attacks or death, have an enormous potential especially as conventional instruments frequently yield suboptimal results; however, there are barriers to the rapid dissemination of corresponding decision aids, such as the regulatory rules related to approval as a medical product, data protection issues and other legal liability aspects, which must be considered.

Machine Learning for Myocardial Infarction Compared With Guideline-Recommended Diagnostic Pathways

BACKGROUND

Collaboration for the Diagnosis and Evaluation of Acute Coronary Syndrome (CoDE-ACS) is a validated clinical decision support tool that uses machine learning with or without serial cardiac troponin measurements at a flexible time point to calculate the probability of myocardial infarction (MI). How CoDE-ACS performs at different time points for serial measurement and compares with guideline-recommended diagnostic pathways that rely on fixed thresholds and time points is uncertain.

METHODS

Patients with possible MI without ST-segment-elevation were enrolled at 12 sites in 5 countries and underwent serial high-sensitivity cardiac troponin I concentration measurement at 0, 1, and 2 hours. Diagnostic performance of the CoDE-ACS model at each time point was determined for index type 1 MI and the effectiveness of previously validated low- and high-probability scores compared with guideline-recommended European Society of Cardiology (ESC) 0/1-hour, ESC 0/2-hour, and High-STEACS (High-Sensitivity Troponin in the Evaluation of Patients With Suspected Acute Coronary Syndrome) pathways.

RESULTS

In total, 4105 patients (mean age, 61 years [interquartile range, 50-74]; 32% women) were included, among whom 575 (14%) had type 1 MI. At presentation, CoDE-ACS identified 56% of patients as low probability, with a negative predictive value and sensitivity of 99.7% (95% CI, 99.5%-99.9%) and 99.0% (98.6%-99.2%), ruling out more patients than the ESC 0-hour and High-STEACS (25% and 35%) pathways. Incorporating a second cardiac troponin measurement, CoDE-ACS identified 65% or 68% of patients as low probability at 1 or 2 hours, for an identical negative predictive value of 99.7% (99.5%-99.9%); 19% or 18% as high probability, with a positive predictive value of 64.9% (63.5%-66.4%) and 68.8% (67.3%-70.1%); and 16% or 14% as intermediate probability. In comparison, after serial measurements, the ESC 0/1-hour, ESC 0/2-hour, and High-STEACS pathways identified 49%, 53%, and 71% of patients as low risk, with a negative predictive value of 100% (99.9%-100%), 100% (99.9%-100%), and 99.7% (99.5%-99.8%); and 20%, 19%, or 29% as high risk, with a positive predictive value of 61.5% (60.0%-63.0%), 65.8% (64.3%-67.2%), and 48.3% (46.8%-49.8%), resulting in 31%, 28%, or 0%, who require further observation in the emergency department, respectively.

CONCLUSIONS

CoDE-ACS performs consistently irrespective of the timing of serial cardiac troponin measurement, identifying more patients as low probability with comparable performance to guideline-recommended pathways for MI. Whether care guided by probabilities can improve the early diagnosis of MI requires prospective evaluation.

REGISTRATION

URL: https://www.clinicaltrials.gov; Unique identifier: NCT00470587.

Imaging-based risk stratification of patients with pulmonary embolism based on dual-energy CT-derived radiomics

BACKGROUND

Technological progress in the acquisition of medical images and the extraction of underlying quantitative imaging data has introduced exciting prospects for the diagnostic assessment of a wide range of conditions. This study aims to investigate the diagnostic utility of a machine learning classifier based on dual-energy computed tomography (DECT) radiomics for classifying pulmonary embolism (PE) severity and assessing the risk for early death.

METHODS

Patients who underwent CT pulmonary angiogram (CTPA) between January 2015 and March 2022 were considered for inclusion in this study. Based on DECT imaging, 107 radiomic features were extracted for each patient using standardized image processing. After dividing the dataset into training and test sets, stepwise feature reduction based on reproducibility, variable importance and correlation analyses were performed to select the most relevant features; these were used to train and validate the gradient-boosted tree models.

RESULTS

The trained machine learning classifier achieved a classification accuracy of .90 for identifying high-risk PE patients with an area under the receiver operating characteristic curve of .59. This CT-based radiomics signature showed good diagnostic accuracy for risk stratification in individuals presenting with central PE, particularly within higher risk groups.

CONCLUSION

Models utilizing DECT-derived radiomics features can accurately stratify patients with pulmonary embolism into established clinical risk scores. This approach holds the potential to enhance patient management and optimize patient flow by assisting in the clinical decision-making process. It also offers the advantage of saving time and resources by leveraging existing imaging to eliminate the necessity for manual clinical scoring.

Serial troponin-I and long-term outcomes in subjects with suspected acute coronary syndrome

AIMS

It is unclear how serial high-sensitivity troponin-I (hsTnI) concentrations affect long-term prognosis in individuals with suspected acute coronary syndrome (ACS).

METHODS AND RESULTS

Subjects who underwent two hsTnI measurements (Siemens TnI Flex® Reagent) separated by 1-7 h, during a first-time hospitalization for myocardial infarction, unstable angina, observation for suspected myocardial infarction, or chest pain from 2012 through 2019, were identified through Danish national registries. Individuals were stratified per their hsTnI concentration pattern (normal, rising, persistently elevated, or falling) and the magnitude of hsTnI concentration change (<20%, >20-50%, or >50% in either direction). We calculated absolute and relative mortality risks standardized to the distributions of risk factors for the entire study population. A total of 20 609 individuals were included of whom 2.3% had died at 30 days, and an additional 4.7% had died at 365 days. The standardized risk of death was highest among persons with a persistently elevated hsTnI concentration (0-30 days: 8.0%, 31-365 days: 11.1%) and lowest among those with two normal hsTnI concentrations (0-30 days: 0.5%, 31-365 days: 2.6%). In neither case did relative hsTnI concentration changes between measurements clearly affect mortality risk. Among persons with a rising hsTnI concentration pattern, 30-day mortality was higher in subjects with a >50% rise compared with those with a less pronounced rise (2.2% vs. <0.1%).

CONCLUSION

Among individuals with suspected ACS, those with a persistently elevated hsTnI concentration consistently had the highest risk of death. In subjects with two normal hsTnI concentrations, mortality was very low and not affected by the magnitude of change between measurements.

A machine learning model for diagnosing acute pulmonary embolism and comparison with Wells score, revised Geneva score, and Years algorithm

BACKGROUND

Acute pulmonary embolism (APE) is a fatal cardiovascular disease, yet missed diagnosis and misdiagnosis often occur due to non-specific symptoms and signs. A simple, objective technique will help clinicians make a quick and precise diagnosis. In population studies, machine learning (ML) plays a critical role in characterizing cardiovascular risks, predicting outcomes, and identifying biomarkers. This work sought to develop an ML model for helping APE diagnosis and compare it against current clinical probability assessment models.

METHODS

This is a single-center retrospective study. Patients with suspected APE were continuously enrolled and randomly divided into two groups including training and testing sets. A total of 8 ML models, including random forest (RF), Naïve Bayes, decision tree, K-nearest neighbors, logistic regression, multi-layer perceptron, support vector machine, and gradient boosting decision tree were developed based on the training set to diagnose APE. Thereafter, the model with the best diagnostic performance was selected and evaluated against the current clinical assessment strategies, including the Wells score, revised Geneva score, and Years algorithm. Eventually, the ML model was internally validated to assess the diagnostic performance using receiver operating characteristic (ROC) analysis.

RESULTS

The ML models were constructed using eight clinical features, including D-dimer, cardiac troponin T (cTNT), arterial oxygen saturation, heart rate, chest pain, lower limb pain, hemoptysis, and chronic heart failure. Among eight ML models, the RF model achieved the best performance with the highest area under the curve (AUC) (AUC = 0.774). Compared to the current clinical assessment strategies, the RF model outperformed the Wells score ( P = 0.030) and was not inferior to any other clinical probability assessment strategy. The AUC of the RF model for diagnosing APE onset in internal validation set was 0.726.

CONCLUSIONS

Based on RF algorithm, a novel prediction model was finally constructed for APE diagnosis. When compared to the current clinical assessment strategies, the RF model achieved better diagnostic efficacy and accuracy. Therefore, the ML algorithm can be a useful tool in assisting with the diagnosis of APE.

Integrating Phenotypic Information of Obstructive Sleep Apnea and Deep Representation of Sleep-Event Sequences for Cardiovascular Risk Prediction

Background

Advances in mobile, wearable and machine learning (ML) technologies for gathering and analyzing long-term health data have opened up new possibilities for predicting and preventing cardiovascular diseases (CVDs). Meanwhile, the association between obstructive sleep apnea (OSA) and CV risk has been well-recognized. This study seeks to explore effective strategies of incorporating OSA phenotypic information and overnight physiological information for precise CV risk prediction in the general population.

Methods

1,874 participants without a history of CVDs from the MESA dataset were included for the 5-year CV risk prediction. Four OSA phenotypes were first identified by the K-mean clustering based on static polysomnographic (PSG) features. Then several phenotype-agnostic and phenotype-specific ML models, along with deep learning (DL) models that integrate deep representations of overnight sleep-event feature sequences, were built for CV risk prediction. Finally, feature importance analysis was conducted by calculating SHapley Additive exPlanations (SHAP) values for all features across the four phenotypes to provide model interpretability.

Results

All ML models showed improved performance after incorporating the OSA phenotypic information. The DL model trained with the proposed phenotype-contrastive training strategy performed the best, achieving an area under the Receiver Operating Characteristic (ROC) curve of 0.877. Moreover, PSG and FOOD FREQUENCY features were recognized as significant CV risk factors across all phenotypes, with each phenotype emphasizing unique features.

Conclusion

Models that are aware of OSA phenotypes are preferred, and lifestyle factors should be a greater focus for precise CV prevention and risk management in the general population.

Early Diagnosis of Cardiovascular Diseases in the Era of Artificial Intelligence: An In-Depth Review

Cardiovascular diseases (CVDs) are significant health issues that result in high death rates globally. Early detection of cardiovascular events may lower the occurrence of acute myocardial infarction and reduce death rates in people with CVDs. Traditional data analysis is inadequate for managing multidimensional data related to the risk prediction of CVDs, heart attacks, medical image interpretations, therapeutic decision-making, and disease prognosis due to the complex pathological mechanisms and multiple factors involved. Artificial intelligence (AI) is a technology that utilizes advanced computer algorithms to extract information from large databases, and it has been integrated into the medical industry. AI methods have shown the ability to speed up the advancement of diagnosing and treating CVDs such as heart failure, atrial fibrillation, valvular heart disease, hypertrophic cardiomyopathy, congenital heart disease, and more. In clinical settings, AI has shown usefulness in diagnosing cardiovascular illness, improving the efficiency of supporting tools, stratifying and categorizing diseases, and predicting outcomes. Advanced AI algorithms have been intricately designed to analyze intricate relationships within extensive healthcare data, enabling them to tackle more intricate jobs compared to conventional approaches.

Adopting artificial intelligence in cardiovascular medicine: a scoping review

Recent years have witnessed significant transformations in cardiovascular medicine, driven by the rapid evolution of artificial intelligence (AI). This scoping review was conducted to capture the breadth of AI applications within cardiovascular science. Employing a structured approach, we sourced relevant articles from PubMed, with an emphasis on journals encompassing general cardiology and digital medicine. We applied filters to highlight cardiovascular articles published in journals focusing on general internal medicine, cardiology and digital medicine, thereby identifying the prevailing trends in the field. Following a comprehensive full-text screening, a total of 140 studies were identified. Over the preceding 5 years, cardiovascular medicine's interplay with AI has seen an over tenfold augmentation. This expansive growth encompasses multiple cardiovascular subspecialties, including but not limited to, general cardiology, ischemic heart disease, heart failure, and arrhythmia. Deep learning emerged as the predominant methodology. The majority of AI endeavors in this domain have been channeled toward enhancing diagnostic and prognostic capabilities, utilizing resources such as hospital datasets, electrocardiograms, and echocardiography. A significant uptrend was observed in AI's application for omics data analysis. However, a clear gap persists in AI's full-scale integration into the clinical decision-making framework. AI, particularly deep learning, has demonstrated robust applications across cardiovascular subspecialties, indicating its transformative potential in this field. As we continue on this trajectory, ensuring the alignment of technological progress with medical ethics becomes crucial. The abundant digital health data today further accentuates the need for meticulous systematic reviews, tailoring them to each cardiovascular subspecialty.

Clinical Biochemistry of Serum Troponin

Accurate measurement and interpretation of serum levels of troponin (Tn) is a central part of the clinical workup of a patient presenting with chest pain suspicious for acute coronary syndrome (ACS). Knowledge of the molecular characteristics of the troponin complex and test characteristics of troponin measurement assays allows for a deeper understanding of causes of false positive and false negative test results in myocardial injury. In this review, we discuss the molecular structure and functions of the constituent proteins of the troponin complex (TnT, TnC, and TnI); review the different isoforms of Tn and where they are from; survey the evolution of clinical Tn assays, ranging from first-generation to high-sensitivity (hs); provide a primer on statistical interpretation of assay results based on different clinical settings; and discuss potential causes of false results. We also summarize the advances in technologies that may lead to the development of future Tn assays, including the development of point of care assays and wearable Tn sensors for real-time continuous measurement.

Establishing optimal cutoff values for high-sensitivity cardiac troponin algorithms in risk stratification of acute myocardial infarction

Acute myocardial infarction (AMI) is a leading cause of mortality globally, highlighting the need for timely and accurate diagnostic strategies. Cardiac troponin has been the biomarker of choice for detecting myocardial injury. A dynamic change in concentrations supports the diagnosis of AMI in the setting of evidence of acute myocardial ischemia. The new generation of high-sensitivity cardiac troponin (hs-cTn) assays has significantly improved analytical sensitivity but at the expense of decreased clinical specificity. As a result, sophisticated algorithms are required to differentiate AMI from non-AMI patients. Establishing optimal hs-cTn cutoffs for these algorithms to rule out and rule in AMI has been the subject of intensive investigations. These efforts have evolved from examining the utility of the hs-cTn 99th percentile upper reference limit, comparing the percentage versus absolute delta thresholds, and evaluating the performance of an early European Society of Cardiology-recommended 3 h algorithm, to the development of accelerated 1 h and 2 h algorithms that combine the admission hs-cTn concentrations and absolute delta cutoffs to rule out and rule in AMI. Specific cutoffs for individual confounding factors such as sex, age, and renal insufficiency have also been investigated. At the same time, concerns such as whether the small delta thresholds exceed the analytical and biological variations of hs-cTn assays and whether the algorithms developed in European study populations fit all other patient cohorts have been raised. In addition, the accelerated algorithms leave a substantial number of patients in a non-diagnostic observation zone. How to properly diagnose patients falling in this zone and those presenting with elevated baseline hs-cTn concentrations due to the presence of confounding factors or comorbidities remain open questions. Here we discuss the developments described above, focusing on criteria and underlying considerations for establishing optimal cutoffs. In-depth analyses are provided on the influence of biological variation, analytical imprecision, local AMI rate, and the timing of presentation on the performance metrics of the accelerated hs-cTn algorithms. Developing diagnostic strategies for patients who remain in the observation zone and those presenting with confounding factors are also reviewed.

Development of machine learning-based models to predict 10-year risk of cardiovascular disease: a prospective cohort study

BACKGROUND

Previous prediction algorithms for cardiovascular diseases (CVD) were established using risk factors retrieved largely based on empirical clinical knowledge. This study sought to identify predictors among a comprehensive variable space, and then employ machine learning (ML) algorithms to develop a novel CVD risk prediction model.

METHODS

From a longitudinal population-based cohort of UK Biobank, this study included 473 611 CVD-free participants aged between 37 and 73 years old. We implemented an ML-based data-driven pipeline to identify predictors from 645 candidate variables covering a comprehensive range of health-related factors and assessed multiple ML classifiers to establish a risk prediction model on 10-year incident CVD. The model was validated through a leave-one-center-out cross-validation.

RESULTS

During a median follow-up of 12.2 years, 31 466 participants developed CVD within 10 years after baseline visits. A novel UK Biobank CVD risk prediction (UKCRP) model was established that comprised 10 predictors including age, sex, medication of cholesterol and blood pressure, cholesterol ratio (total/high-density lipoprotein), systolic blood pressure, previous angina or heart disease, number of medications taken, cystatin C, chest pain and pack-years of smoking. Our model obtained satisfied discriminative performance with an area under the receiver operating characteristic curve (AUC) of 0.762±0.010 that outperformed multiple existing clinical models, and it was well-calibrated with a Brier Score of 0.057±0.006. Further, the UKCRP can obtain comparable performance for myocardial infarction (AUC 0.774±0.011) and ischaemic stroke (AUC 0.730±0.020), but inferior performance for haemorrhagic stroke (AUC 0.644±0.026).

CONCLUSION

ML-based classification models can learn expressive representations from potential high-risked CVD participants who may benefit from earlier clinical decisions.

ROMIAE (Rule-Out Acute Myocardial Infarction Using Artificial Intelligence Electrocardiogram Analysis) trial study protocol: a prospective multicenter observational study for validation of a deep learning-based 12-lead electrocardiogram analysis model for detecting acute myocardial infarction in patients visiting the emergency department

OBJECTIVE

Based on the development of artificial intelligence (AI), an emerging number of methods have achieved outstanding performances in the diagnosis of acute myocardial infarction (AMI) using an electrocardiogram (ECG). However, AI-ECG analysis using a multicenter prospective design for detecting AMI has yet to be conducted. This prospective multicenter observational study aims to validate an AI-ECG model for detecting AMI in patients visiting the emergency department.

METHODS

Approximately 9,000 adult patients with chest pain and/or equivalent symptoms of AMI will be enrolled in 18 emergency medical centers in Korea. The AI-ECG analysis algorithm we developed and validated will be used in this study. The primary endpoint is the diagnosis of AMI on the day of visiting the emergency center, and the secondary endpoint is a 30-day major adverse cardiac event. From March 2022, patient registration has begun at centers approved by the institutional review board.

DISCUSSION

This is the first prospective study designed to identify the efficacy of an AI-based 12-lead ECG analysis algorithm for diagnosing AMI in emergency departments across multiple centers. This study may provide insights into the utility of deep learning in detecting AMI on electrocardiograms in emergency departments. Trial registration ClinicalTrials.gov identifier: NCT05435391. Registered on June 28, 2022.

Derivation and external validation of machine-learning models for risk stratification in chest pain with normal troponin

AIMS

Risk stratification of patients with chest pain and a high-sensitivity cardiac troponin T (hs-cTnT) concentration

METHODS AND RESULTS

Four machine-learning-based models and one logistic regression (LR) model were trained on 4075 patients (single-centre Spanish cohort) and externally validated on 3609 patients (international prospective Advantageous Predictors of Acute Coronary syndromes Evaluation cohort). Models were compared with GRACE and HEART scores and a single undetectable hs-cTnT-based strategy (u-cTn; hs-cTnT < 5 ng/L and time from symptoms onset >180 min). Probability thresholds for safe discharge were derived in the derivation cohort. The endpoint occurred in 105 (2.6%) patients in the training set and 98 (2.7%) in the external validation set. Gradient boosting full (GBf) showed the best discrimination (area under the curve = 0.808). Calibration was good for the reduced neural network and LR models. Gradient boosting full identified the highest proportion of patients for safe discharge (36.7 vs. 23.4 vs. 27.2%; GBf vs. LR vs. u-cTn, respectively) with similar safety (missed endpoint per 1000 patients: 2.2 vs. 3.5 vs. 3.1, respectively). All derived models were superior to the HEART and GRACE scores (P < 0.001).

CONCLUSION

Machine-learning and LR prediction models were superior to the HEART, GRACE, and u-cTn for risk stratification of patients with chest pain and a baseline hs-cTnT

CLINICAL TRIAL REGISTRATION

ClinicalTrials.gov number, NCT00470587, https://clinicaltrials.gov/ct2/show/NCT00470587.

Collapse

Key Words

• Machine learning
• Myocardial infarction
• Prediction
• Troponin

Collapse

MESH Headings

• Humans
• Troponin
• Prospective Studies
• Biomarkers
• Troponin T
• Chest Pain/diagnosis
• Chest Pain/etiology
• Risk Assessment

Collapse

Grants

• Swiss National Science Foundation

Collapse

Affiliation(s)

Agustín Fernández-Cisnal Cardiology Department, Hospital Clínico Universitario de València, Instituto de Investigación Sanitaria (INCLIVA), University of València, Centro de Investigación Biomédica en Red Enfermedades Cardiovaculares (CIBERCV), València, Spain
Pedro Lopez-Ayala Cardiovascular Research Institute Basel (CRIB) and Department of Cardiology, University Heart Center Basel, University Hospital Basel, University of Basel, Basel, Switzerland
Ernesto Valero Cardiology Department, Hospital Clínico Universitario de València, Instituto de Investigación Sanitaria (INCLIVA), University of València, Centro de Investigación Biomédica en Red Enfermedades Cardiovaculares (CIBERCV), València, Spain
Luca Koechlin Cardiovascular Research Institute Basel (CRIB) and Department of Cardiology, University Heart Center Basel, University Hospital Basel, University of Basel, Basel, Switzerland
Arturo Catarralá Clinical Biochemistry Department, Hospital Clínico Universitario de València, Instituto de Investigación Sanitaria (INCLIVA), València 46010, Spain
Jasper Boeddinghaus Cardiovascular Research Institute Basel (CRIB) and Department of Cardiology, University Heart Center Basel, University Hospital Basel, University of Basel, Basel, Switzerland
José Noceda Emergency Department, Hospital Clínico Universitario de València, Instituto de Investigación Sanitaria (INCLIVA), València 46010, Spain
Thomas Nestelberger Cardiovascular Research Institute Basel (CRIB) and Department of Cardiology, University Heart Center Basel, University Hospital Basel, University of Basel, Basel, Switzerland
Òscar Miró Emergency Department, Hospital Clinic, Barcelona, Catalonia, Spain
Núñez Julio Cardiology Department, Hospital Clínico Universitario de València, Instituto de Investigación Sanitaria (INCLIVA), University of València, Centro de Investigación Biomédica en Red Enfermedades Cardiovaculares (CIBERCV), València, Spain
Christian Mueller Cardiovascular Research Institute Basel (CRIB) and Department of Cardiology, University Heart Center Basel, University Hospital Basel, University of Basel, Basel, Switzerland
Juan Sanchis Cardiology Department, Hospital Clínico Universitario de València, Instituto de Investigación Sanitaria (INCLIVA), University of València, Centro de Investigación Biomédica en Red Enfermedades Cardiovaculares (CIBERCV), València, Spain

Collapse

Circulating Cardiovascular Biomarkers in Cancer Therapeutics-Related Cardiotoxicity: Review of Critical Challenges, Solutions, and Future Directions

Cardiotoxicity is a growing concern in the oncology population. Transthoracic echocardiography and multigated acquisition scans have been used for surveillance but are relatively insensitive and resource intensive. Innovative imaging techniques are constrained by cost and availability. More sensitive, cost-effective cardiotoxicity surveillance strategies are needed. Circulating cardiovascular biomarkers could provide a sensitive, low-cost solution. Biomarkers such as troponins, natriuretic peptides (NPs), novel upstream signals of oxidative stress, inflammation, and fibrosis as well as panomic technologies have shown substantial promise, and guidelines recommend baseline measurement of troponins and NPs in all patients receiving potential cardiotoxins. Nonetheless, supporting evidence has been hampered by several limitations. Previous reviews have provided valuable perspectives on biomarkers in cancer populations, but important analytic aspects remain to be examined in depth. This review provides comprehensive assessment of critical challenges and solutions in this field, with focus on analytical issues relating to biomarker measurement and interpretation. Examination of evidence pertaining to common and serious forms of cardiotoxicity reveals that improved study designs incorporating larger, more diverse populations, registry-based approaches, and refinement of current definitions are key. Further efforts to harmonize biomarker methodologies including centralized biobanking and analyses, novel decision limits, and head-to-head comparisons are needed. Multimarker algorithms incorporating machine learning may allow rapid, personalized risk assessment. These improvements will not only augment the predictive value of circulating biomarkers in cardiotoxicity but may elucidate both direct and indirect relationships between cardiovascular disease and cancer, allowing biomarkers a greater role in the development and success of novel anticancer therapies.

Artificial Intelligence, Computational Simulations, and Extended Reality in Cardiovascular Interventions

Artificial intelligence, computational simulations, and extended reality, among other 21st century computational technologies, are changing the health care system. To collectively highlight the most recent advances and benefits of artificial intelligence, computational simulations, and extended reality in cardiovascular therapies, we coined the abbreviation AISER. The review particularly focuses on the following applications of AISER: 1) preprocedural planning and clinical decision making; 2) virtual clinical trials, and cardiovascular device research, development, and regulatory approval; and 3) education and training of interventional health care professionals and medical technology innovators. We also discuss the obstacles and constraints associated with the application of AISER technologies, as well as the proposed solutions. Interventional health care professionals, computer scientists, biomedical engineers, experts in bioinformatics and visualization, the device industry, ethics committees, and regulatory agencies are expected to streamline the use of AISER technologies in cardiovascular interventions and medicine in general.

Developing and validating clinical features-based machine learning algorithms to predict influenza infection in influenza-like illness patients

BACKGROUND

Seasonal influenza poses a significant risk, and patients can benefit from early diagnosis and treatment. However, underdiagnosis and undertreatment remain widespread. We developed and compared clinical feature-based machine learning (ML) algorithms that can accurately predict influenza infection in emergency departments (EDs) among patients with influenza-like illness (ILI).

MATERIAL AND METHODS

We conducted a prospective cohort study in five EDs in the US and Taiwan from 2015 to 2020. Adult patients visiting the EDs with symptoms of ILI were recruited and tested by real-time RT-PCR for influenza. We evaluated seven ML algorithms and compared their results with previously developed clinical prediction models.

RESULTS

Out of the 2189 enrolled patients, 1104 tested positive for influenza. The eXtreme Gradient Boosting achieved superior performance with an area under the receiver operating characteristic curve of 0.82 (95% confidence interval [CI] = 0.79-0.85), with a sensitivity of 0.92 (95% CI = 0.88-0.95), specificity of 0.89 (95% CI = 0.86-0.92), and accuracy of 0.72 (95% CI = 0.69-0.76) in the testing set over cut-offs of 0.4, 0.6 and 0.5, respectively. These results were superior to those of previously proposed clinical prediction models. The model interpretation revealed that body temperature, cough, rhinorrhea, and exposure history were positively associated with and the days of illness and influenza vaccine were negatively associated with influenza infection. We also found the week of the influenza season, pulse rate, and oxygen saturation to be associated with influenza infection.

CONCLUSIONS

The clinical feature-based ML model outperformed conventional models for predicting influenza infection.

Prediction of the Fatal Acute Complications of Myocardial Infarction via Machine Learning Algorithms

Background

Myocardial infarction (MI) is a major cause of death, particularly during the first year. The avoidance of potentially fatal outcomes requires expeditious preventative steps. Machine learning (ML) is a subfield of artificial intelligence science that detects the underlying patterns of available big data for modeling them. This study aimed to establish an ML model with numerous features to predict the fatal complications of MI during the first 72 hours of hospital admission.

Methods

We applied an MI complications database that contains the demographic and clinical records of patients during the 3 days of admission based on 2 output classes: dead due to the known complications of MI and alive. We utilized the recursive feature elimination (RFE) method to apply feature selection. Thus, after applying this method, we reduced the number of features to 50. The performance of 4 common ML classifier algorithms, namely logistic regression, support vector machine, random forest, and extreme gradient boosting (XGBoost), was evaluated using 8 classification metrics (sensitivity, specificity, precision, false-positive rate, false-negative rate, accuracy, F1-score, and AUC).

Results

In this study of 1699 patients with confirmed MI, 15.94% experienced fatal complications, and the rest remained alive. The XGBoost model achieved more desirable results based on the accuracy and F1-score metrics and distinguished patients with fatal complications from surviving ones (AUC=78.65%, sensitivity=94.35%, accuracy=91.47%, and F1-score=95.14%). Cardiogenic shock was the most significant feature influencing the prediction of the XGBoost algorithm.

Conclusion

XGBoost algorithms can be a promising model for predicting fatal complications following MI.

Personalized diagnosis in suspected myocardial infarction

BACKGROUND

In suspected myocardial infarction (MI), guidelines recommend using high-sensitivity cardiac troponin (hs-cTn)-based approaches. These require fixed assay-specific thresholds and timepoints, without directly integrating clinical information. Using machine-learning techniques including hs-cTn and clinical routine variables, we aimed to build a digital tool to directly estimate the individual probability of MI, allowing for numerous hs-cTn assays.

METHODS

In 2,575 patients presenting to the emergency department with suspected MI, two ensembles of machine-learning models using single or serial concentrations of six different hs-cTn assays were derived to estimate the individual MI probability (ARTEMIS model). Discriminative performance of the models was assessed using area under the receiver operating characteristic curve (AUC) and logLoss. Model performance was validated in an external cohort with 1688 patients and tested for global generalizability in 13 international cohorts with 23,411 patients.

RESULTS

Eleven routinely available variables including age, sex, cardiovascular risk factors, electrocardiography, and hs-cTn were included in the ARTEMIS models. In the validation and generalization cohorts, excellent discriminative performance was confirmed, superior to hs-cTn only. For the serial hs-cTn measurement model, AUC ranged from 0.92 to 0.98. Good calibration was observed. Using a single hs-cTn measurement, the ARTEMIS model allowed direct rule-out of MI with very high and similar safety but up to tripled efficiency compared to the guideline-recommended strategy.

CONCLUSION

We developed and validated diagnostic models to accurately estimate the individual probability of MI, which allow for variable hs-cTn use and flexible timing of resampling. Their digital application may provide rapid, safe and efficient personalized patient care.

TRIAL REGISTRATION NUMBERS

Data of following cohorts were used for this project: BACC ( www.

CLINICALTRIALS

gov ; NCT02355457), stenoCardia ( www.

CLINICALTRIALS

gov ; NCT03227159), ADAPT-BSN ( www.australianclinicaltrials.gov.au ; ACTRN12611001069943), IMPACT ( www.australianclinicaltrials.gov.au , ACTRN12611000206921), ADAPT-RCT ( www.anzctr.org.au ; ANZCTR12610000766011), EDACS-RCT ( www.anzctr.org.au ; ANZCTR12613000745741); DROP-ACS ( https://www.umin.ac.jp , UMIN000030668); High-STEACS ( www.

CLINICALTRIALS

gov ; NCT01852123), LUND ( www.

CLINICALTRIALS

gov ; NCT05484544), RAPID-CPU ( www.

CLINICALTRIALS

gov ; NCT03111862), ROMI ( www.

CLINICALTRIALS

gov ; NCT01994577), SAMIE ( https://anzctr.org.au ; ACTRN12621000053820), SEIGE and SAFETY ( www.

CLINICALTRIALS

gov ; NCT04772157), STOP-CP ( www.

CLINICALTRIALS

gov ; NCT02984436), UTROPIA ( www.

CLINICALTRIALS

gov ; NCT02060760).

Machine learning algorithms to predict major adverse cardiovascular events in patients with diabetes

BACKGROUND

Major Adverse Cardiovascular Events (MACE) are common complications of type 2 diabetes mellitus (T2DM) that include myocardial infarction (MI), stroke, and heart failure (HF). The objective of the current study was to predict MACE among T2DM patients.

METHODS

Type 2 diabetes mellitus patients above 18 years old were recruited for the study from the All of Us Research Program. Eligible participants were those who took sodium-glucose cotransporter 2 inhibitors. Different Machine learning algorithms: including RandomForest (RF), XGBoost, logistic regression (LR), and weighted ensemble model (WEM) were employed. Clinical attributes, electrolytes and biomarkers were explored in predicting MACE. The feature importance was determined using mean decrease accuracy.

RESULTS

Overall, 9, 059 subjects were included in the analyses, of which 5197 (57.4%) were females. The XGBoost Model demonstrated a prediction accuracy of 0.80 [0.78-0.82], which is higher as compared to the RF 0.78[0.76-0.80], the LR model 0.65 [0.62-0.67], and the WEM 0.75 [0.73-0.76], respectively. The classification accuracy of the models for stroke was more than 95%, which was higher than prediction accuracy for MI (∼85%), and HF (∼80%). Phosphate, blood urea nitrogen and troponin levels were the major predictors of MACE.

CONCLUSION

The ML models had shown acceptable performance in predicting MACE in T2DM patients, except the LR model. Phosphate, blood urea nitrogen, and other electrolytes were important predictors of MACE, which is consistent between the individual components of MACE, such as stroke, MI, and HF. These parameters can be calibrated as prognostic parameters of MACE events in T2DM patients.

The Clinical Validation of a Common Analytical Change Criteria for Cardiac Troponin for Ruling in an Acute Cardiovascular Outcome in Patients Presenting with Ischemic Chest Pain Symptoms

Serial cardiac troponin (cTn) testing on patients with symptoms suggestive of acute coronary syndrome (ACS) is primarily to identify those patients with evolving myocardial injury. With the improved analytical performance of the high-sensitivity cTn (hs-cTn) assays, different change criteria have been proposed that are mostly assay dependent. Here, we developed and compared a new Common Change Criteria (3C for the combined criteria of >3 ng/L, >30%, or >15% based on the initial cTn concentration of <10 ng/L, 10 to 100 ng/L, or >100 ng/L, respectively) method, versus the 2 h assay-dependent absolute change criteria endorsed by the European Society of Cardiology (ESC), versus the common relative >20% change criterion. These different analytical change criteria were evaluated in 855 emergency department (ED) patients with symptoms of ACS and who had two samples collected 3 h apart. The cTn concentrations were measured with four different assays (Abbott hs-cTnI, Roche hs-cTnT, Ortho cTnI-ES, and Ortho hs-cTnI). The outcomes evaluated were myocardial infarction (MI) and a composite outcome (MI, unstable angina, ventricular arrhythmia, heart failure, or cardiovascular death) within 7 days of ED presentation. The combined change criteria (3C) method yielded higher specificities (range: 93.9 to 97.2%) as compared to the >20% criterion (range: 42.3 to 88.1%) for all four assays for MI. The 3C method only yielded a higher specificity estimate for MI for the cTnI-ES assay (95.9%) versus the absolute change criteria (71.7%). Similar estimates were obtained for the composite outcome. There was also substantial agreement between hs-cTnT and the different cTnI assays for MI with the 3C method, with the percent agreement being ≥95%. The Common Change Criteria (3C) method combining both absolute and different percent changes may be used with cTnI, hs-cTnT, and different hs-cTnI assays to yield similar high-specificity (rule-in) estimates for adverse cardiovascular events for patients presenting to the ED with ACS symptoms.

A Systematic Review: Do the Use of Machine Learning, Deep Learning, and Artificial Intelligence Improve Patient Outcomes in Acute Myocardial Ischemia Compared to Clinician-Only Approaches?

Cardiovascular diseases (CVDs) present a significant global health challenge and remain a primary cause of death. Early detection and intervention are crucial for improved outcomes in acute coronary syndrome (ACS), particularly acute myocardial infarction (AMI) cases. Artificial intelligence (AI) can detect heart disease early by analyzing patient information and electrocardiogram (ECG) data, providing invaluable insights into this critical health issue. However, the imbalanced nature of ECG and patient data presents challenges for traditional machine learning (ML) algorithms in performing unbiasedly. Investigators have proposed various data-level and algorithm-level solutions to overcome these challenges. In this study, we used a systematic literature review (SLR) approach to give an overview of the current literature and to highlight the difficulties of utilizing ML, deep learning (DL), and AI algorithms in predicting, diagnosing, and prognosis of heart diseases. We reviewed 181 articles from reputable journals published between 2013 and June 15, 2023, focusing on eight selected papers for in-depth analysis. The analysis considered factors such as heart disease type, algorithms used, applications, and proposed solutions and compared the benefits of algorithms combined with clinicians versus clinicians alone. This systematic review revealed that the current ML-based diagnostic approaches face several open problems and issues when implementing ML, DL, and AI in real-life settings. Although these algorithms show higher sensitivities, specificities, and accuracies in detecting heart disease, we must address the ethical concerns while implementing these models into clinical practice. The transparency of how these algorithms operate remains a challenge. Nevertheless, further exploration and research in ML, DL, and AI are necessary to overcome these challenges and fully harness their potential to improve health outcomes for patients with AMI.

Using logistic regression models to investigate the effects of high-sensitivity cardiac troponin T confounders on ruling in acute myocardial infarction

OBJECTIVES

Confounding factors, including sex, age, and renal dysfunction, affect high-sensitivity cardiac troponin T (hs-cTnT) concentrations and the acute myocardial infarction (AMI) diagnosis. This study assessed the effects of these confounders through logistic regression models and evaluated the diagnostic performance of an optimized, integrated prediction model.

METHODS

This retrospective study included a primary derivation cohort of 18,022 emergency department (ED) patients at a US medical center and a validation cohort of 890 ED patients at a Canadian medical center. Hs-cTnT was measured with 0/3 h sampling. The primary outcome was index AMI diagnosis. Logistic regression models were optimized to predict AMI using delta hs-cTnT and its confounders as covariates. The diagnostic performance of model cutoffs was compared to that of the hs-cTnT delta thresholds. Serial logistic regressions were carried out to evaluate the relationship between covariates.

RESULTS

The area under the curve of the best-fitted model was 0.95. The model achieved a 90.0% diagnostic accuracy in the validation cohort. The optimal model cutoff yielded comparable performance (90.5% accuracy) to the optimal sex-specific delta thresholds (90.3% accuracy), with 95.8% agreement between the two diagnostic methods. Serial logistic regressions revealed that delta hs-cTnT played a more predominant role in AMI prediction than its confounders, among which sex is more predictive of AMI (total effect coefficient 1.04) than age (total effect coefficient 0.05) and eGFR (total effect coefficient -0.008).

CONCLUSIONS

The integrated prediction model incorporating confounding factors does not outperform hs-cTnT delta thresholds. Sex-specific hs-cTnT delta thresholds remain to provide the highest diagnostic accuracy.

Machine learning algorithm can provide assistance for the diagnosis of non-ST-segment elevation myocardial infarction

INTRODUCTION

Our aim was to use the constructed machine learning (ML) models as auxiliary diagnostic tools to improve the diagnostic accuracy of non-ST-elevation myocardial infarction (NSTEMI).

MATERIALS AND METHODS

A total of 2878 patients were included in this retrospective study, including 1409 patients with NSTEMI and 1469 patients with unstable angina pectoris. The clinical and biochemical characteristics of the patients were used to construct the initial attribute set. SelectKBest algorithm was used to determine the most important features. A feature engineering method was applied to create new features correlated strongly to train ML models and obtain promising results. Based on the experimental dataset, the ML models of extreme gradient boosting, support vector machine, random forest, naïve Bayesian, gradient boosting machines and logistic regression were constructed. Each model was verified by test set data, and the diagnostic performance of each model was comprehensively evaluated.

RESULTS

The six ML models based on the training set all play an auxiliary role in the diagnosis of NSTEMI. Although all models taken for comparison performed differences, the extreme gradient boosting ML model performed the best in terms of accuracy rate (0.95±0.014), precision rate (0.94±0.011), recall rate (0.98±0.003) and F-1 score (0.96±0.007) in NSTEMI.

CONCLUSIONS

The ML model constructed based on clinical data can be used as an auxiliary tool to improve the accuracy of NSTEMI diagnosis. According to our comprehensive evaluation, the performance of the extreme gradient boosting model was the best.

Machine learning for diagnosis of myocardial infarction using cardiac troponin concentrations

Although guidelines recommend fixed cardiac troponin thresholds for the diagnosis of myocardial infarction, troponin concentrations are influenced by age, sex, comorbidities and time from symptom onset. To improve diagnosis, we developed machine learning models that integrate cardiac troponin concentrations at presentation or on serial testing with clinical features and compute the Collaboration for the Diagnosis and Evaluation of Acute Coronary Syndrome (CoDE-ACS) score (0-100) that corresponds to an individual's probability of myocardial infarction. The models were trained on data from 10,038 patients (48% women), and their performance was externally validated using data from 10,286 patients (35% women) from seven cohorts. CoDE-ACS had excellent discrimination for myocardial infarction (area under curve, 0.953; 95% confidence interval, 0.947-0.958), performed well across subgroups and identified more patients at presentation as low probability of having myocardial infarction than fixed cardiac troponin thresholds (61 versus 27%) with a similar negative predictive value and fewer as high probability of having myocardial infarction (10 versus 16%) with a greater positive predictive value. Patients identified as having a low probability of myocardial infarction had a lower rate of cardiac death than those with intermediate or high probability 30 days (0.1 versus 0.5 and 1.8%) and 1 year (0.3 versus 2.8 and 4.2%; P < 0.001 for both) from patient presentation. CoDE-ACS used as a clinical decision support system has the potential to reduce hospital admissions and have major benefits for patients and health care providers.

Machine learning prediction of mortality in Acute Myocardial Infarction

BACKGROUND

Acute Myocardial Infarction (AMI) is the leading cause of death in Portugal and globally. The present investigation created a model based on machine learning for predictive analysis of mortality in patients with AMI upon admission, using different variables to analyse their impact on predictive models.

METHODS

Three experiments were built for mortality in AMI in a Portuguese hospital between 2013 and 2015 using various machine learning techniques. The three experiments differed in the number and type of variables used. We used a discharged patients' episodes database, including administrative data, laboratory data, and cardiac and physiologic test results, whose primary diagnosis was AMI.

RESULTS

Results show that for Experiment 1, Stochastic Gradient Descent was more suitable than the other classification models, with a classification accuracy of 80%, a recall of 77%, and a discriminatory capacity with an AUC of 79%. Adding new variables to the models increased AUC in Experiment 2 to 81% for the Support Vector Machine method. In Experiment 3, we obtained an AUC, in Stochastic Gradient Descent, of 88% and a recall of 80%. These results were obtained when applying feature selection and the SMOTE technique to overcome imbalanced data.

CONCLUSIONS

Our results show that the introduction of new variables, namely laboratory data, impacts the performance of the methods, reinforcing the premise that no single approach is adapted to all situations regarding AMI mortality prediction. Instead, they must be selected, considering the context and the information available. Integrating Artificial Intelligence (AI) and machine learning with clinical decision-making can transform care, making clinical practice more efficient, faster, personalised, and effective. AI emerges as an alternative to traditional models since it has the potential to explore large amounts of information automatically and systematically.

Coronary low-attenuation plaque and high-sensitivity cardiac troponin

OBJECTIVE

In patients with acute chest pain who have had myocardial infarction excluded, plasma cardiac troponin I concentrations ≥5 ng/L are associated with risk of future adverse cardiovascular events. We aim to evaluate the association between cardiac troponin and coronary plaque composition in such patients.

METHODS

In a prespecified secondary analysis of a prospective cohort study, blinded quantitative plaque analysis was performed on 242 CT coronary angiograms of patients with acute chest pain in whom myocardial infarction was excluded. Patients were stratified by peak plasma cardiac troponin I concentration ≥5 ng/L or <5 ng/L. Associations were assessed using univariable and multivariable logistic regression analyses.

RESULTS

The cohort was predominantly middle-aged (62±12 years) men (69%). Patients with plasma cardiac troponin I concentration ≥5 ng/L (n=161) had a higher total (median 33% (IQR 0-47) vs 0% (IQR 0-33)), non-calcified (27% (IQR 0-37) vs 0% (IQR 0-28)), calcified (2% (IQR 0-8) vs 0% (IQR 0-3)) and low-attenuation (1% (IQR 0-3) vs 0% (IQR 0-1)) coronary plaque burden compared with those with concentrations <5 ng/L (n=81; p≤0.001 for all). Low-attenuation plaque burden was independently associated with plasma cardiac troponin I concentration ≥5 ng/L after adjustment for clinical characteristics (adjusted OR per doubling 1.62 (95% CI 1.17 to 2.32), p=0.005) or presence of any visible coronary artery disease (adjusted OR per doubling 1.57 (95% CI 1.07 to 2.37), p=0.026).

CONCLUSION

In patients with acute chest pain but without myocardial infarction, plasma cardiac troponin I concentrations ≥5 ng/L are associated with greater burden of low-attenuation coronary plaque.

Deep learning augmented ECG analysis to identify biomarker-defined myocardial injury

Chest pain is a common clinical complaint for which myocardial injury is the primary concern and is associated with significant morbidity and mortality. To aid providers' decision-making, we aimed to analyze the electrocardiogram (ECG) using a deep convolutional neural network (CNN) to predict serum troponin I (TnI) from ECGs. We developed a CNN using 64,728 ECGs from 32,479 patients who underwent ECG within 2 h prior to a serum TnI laboratory result at the University of California, San Francisco (UCSF). In our primary analysis, we classified patients into groups of TnI < 0.02 or ≥ 0.02 µg/L using 12-lead ECGs. This was repeated with an alternative threshold of 1.0 µg/L and with single-lead ECG inputs. We also performed multiclass prediction for a set of serum troponin ranges. Finally, we tested the CNN in a cohort of patients selected for coronary angiography, including 3038 ECGs from 672 patients. Cohort patients were 49.0% female, 42.8% white, and 59.3% (19,283) never had a positive TnI value (≥ 0.02 µg/L). CNNs accurately predicted elevated TnI, both at a threshold of 0.02 µg/L (AUC = 0.783, 95% CI 0.780-0.786) and at a threshold of 1.0 µg/L (AUC = 0.802, 0.795-0.809). Models using single-lead ECG data achieved significantly lower accuracy, with AUCs ranging from 0.740 to 0.773 with variation by lead. Accuracy of the multi-class model was lower for intermediate TnI value-ranges. Our models performed similarly on the cohort of patients who underwent coronary angiography. Biomarker-defined myocardial injury can be predicted by CNNs from 12-lead and single-lead ECGs.

Serial troponin-T and long-term outcomes in suspected acute coronary syndrome

BACKGROUND

Long-term prognostic implications of serial high-sensitivity troponin concentrations in subjects with suspected acute coronary syndrome are unknown.

METHODS AND RESULTS

Individuals with a first diagnosis of myocardial infarction, unstable angina, observation for suspected myocardial infarction, or chest pain from 2012 through 2019 who underwent two high-sensitivity troponin-T (hsTnT) measurements 1-7 h apart were identified through Danish national registries. Absolute and relative risks for death at days 0-30 and 31-365, stratified for whether subjects had normal or elevated hsTnT concentrations, and whether these concentrations changed by <20%, > 20 to 50%, or >50% in either direction from first to second measurement, were calculated through multivariable logistic regression with average treatment effect modeling. Of the 28 902 individuals included, 2.8% had died at 30 days, whereas 4.9% of those who had survived the first 30 days died between days 31-365. The standardized risk of death was highest among subjects with two elevated hsTnT concentrations (0-30 days: 4.3%, 31-365 days: 7.2%). In this group, mortality was significantly higher in those with a > 20 to 50% or >50% rise from first to second measurement, though only at 30 days. The risk of death was very low in subjects with two normal hsTnT concentrations (0-30 days: 0.1%, 31-365 days: 0.9%) and did not depend on relative or absolute changes between measurements.

CONCLUSIONS

Individuals with suspected acute coronary syndrome and two consecutively elevated hsTnT concentrations consistently had the highest risk of death. Mortality was very low in subjects with two normal hsTnT concentrations, irrespective of changes between measurements.

Machine learning for early prediction of acute myocardial infarction or death in acute chest pain patients using electrocardiogram and blood tests at presentation

AIMS

In the present study, we aimed to evaluate the performance of machine learning (ML) models for identification of acute myocardial infarction (AMI) or death within 30 days among emergency department (ED) chest pain patients.

METHODS AND RESULTS

Using data from 9519 consecutive ED chest pain patients, we created ML models based on logistic regression or artificial neural networks. Model inputs included sex, age, ECG and the first blood tests at patient presentation: High sensitivity TnT (hs-cTnT), glucose, creatinine, and hemoglobin. For a safe rule-out, the models were adapted to achieve a sensitivity > 99% and a negative predictive value (NPV) > 99.5% for 30-day AMI/death. For rule-in, we set the models to achieve a specificity > 90% and a positive predictive value (PPV) of > 70%. The models were also compared with the 0 h arm of the European Society of Cardiology algorithm (ESC 0 h); An initial hs-cTnT < 5 ng/L for rule-out and ≥ 52 ng/L for rule-in. A convolutional neural network was the best model and identified 55% of the patients for rule-out and 5.3% for rule-in, while maintaining the required sensitivity, specificity, NPV and PPV levels. ESC 0 h failed to reach these performance levels.

DISCUSSION

An ML model based on age, sex, ECG and blood tests at ED arrival can identify six out of ten chest pain patients for safe early rule-out or rule-in with no need for serial blood tests. Future studies should attempt to improve these ML models further, e.g. by including additional input data.

Integrated Machine Learning Decision Tree Model for Risk Evaluation in Patients with Non-Valvular Atrial Fibrillation When Taking Different Doses of Dabigatran

The new generation of nonvitamin K antagonists are broadly applied for stroke prevention due to their notable efficacy and safety. Our study aimed to develop a suggestive utilization of dabigatran through an integrated machine learning (ML) decision-tree model. Participants taking different doses of dabigatran in the Randomized Evaluation of Long-Term Anticoagulant Therapy trial were included in our analysis and defined as the 110 mg and 150 mg groups. The proposed scheme integrated ML methods, namely naive Bayes, random forest (RF), classification and regression tree (CART), and extreme gradient boosting (XGBoost), which were used to identify the essential variables for predicting vascular events in the 110 mg group and bleeding in the 150 mg group. RF (0.764 for 110 mg; 0.747 for 150 mg) and XGBoost (0.708 for 110 mg; 0.761 for 150 mg) had better area under the receiver operating characteristic curve (AUC) values than logistic regression (benchmark model; 0.683 for 110 mg; 0.739 for 150 mg). We then selected the top ten important variables as internal nodes of the CART decision tree. The two best CART models with ten important variables output tree-shaped rules for predicting vascular events in the 110 mg group and bleeding in the 150 mg group. Our model can be used to provide more visualized and interpretable suggestive rules to clinicians managing NVAF patients who are taking dabigatran.

An online platform for COVID-19 diagnostic screening using a machine learning algorithm

OBJECTIVE

COVID-19 has brought emerging public health emergency and new challenges. It configures a complex panorama that has been requiring a set of coordinated actions and has innovation as one of its pillars. In particular, the use of digital tools plays an important role. In this context, this study presents a screening algorithm that uses a machine learning model to assess the probability of a diagnosis of COVID-19 based on clinical data.

METHODS

This algorithm was made available for free on an online platform. The project was developed in three phases. First, an machine learning risk model was developed. Second, a system was developed that would allow the user to enter patient data. Finally, this platform was used in teleconsultations carried out during the pandemic period.

RESULTS

The number of accesses during the period was 4,722. A total of 126 assistances were carried out from March 23, 2020, to June 16, 2020, and 107 satisfaction survey returns were received. The response rate to the questionnaires was 84.92%, and the ratings obtained regarding the satisfaction level were higher than 4.8 (on a 0-5 scale). The Net Promoter Score was 94.4.

CONCLUSION

To the best of our knowledge, this is the first online application of its kind that presents a probabilistic assessment of COVID-19 using machine learning models exclusively based on the symptoms and clinical characteristics of users. The level of satisfaction was high. The integration of machine learning tools in telemedicine practice has great potential.

Deep learning-based prediction of future myocardial infarction using invasive coronary angiography: a feasibility study

BACKGROUND

Angiographic parameters can facilitate the risk stratification of coronary lesions but remain insufficient in the prediction of future myocardial infarction (MI).

AIMS

We compared the ability of humans, angiographic parameters and deep learning (DL) to predict the lesion that would be responsible for a future MI in a population of patients with non-significant CAD at baseline.

METHODS

We retrospectively included patients who underwent invasive coronary angiography (ICA) for MI, in whom a previous angiogram had been performed within 5 years. The ability of human visual assessment, diameter stenosis, area stenosis, quantitative flow ratio (QFR) and DL to predict the future culprit lesion (FCL) was compared.

RESULTS

In total, 746 cropped ICA images of FCL and non-culprit lesions (NCL) were analysed. Predictive models for each modality were developed in a training set before validation in a test set. DL exhibited the best predictive performance with an area under the curve of 0.81, compared with diameter stenosis (0.62, p=0.04), area stenosis (0.58, p=0.05) and QFR (0.67, p=0.13). DL exhibited a significant net reclassification improvement (NRI) compared with area stenosis (0.75, p=0.03) and QFR (0.95, p=0.01), and a positive nonsignificant NRI when compared with diameter stenosis. Among all models, DL demonstrated the highest accuracy (0.78) followed by QFR (0.70) and area stenosis (0.68). Predictions based on human visual assessment and diameter stenosis had the lowest accuracy (0.58).

CONCLUSION

In this feasibility study, DL outperformed human visual assessment and established angiographic parameters in the prediction of FCLs. Larger studies are now required to confirm this finding.

Machine-learning Algorithms for Ischemic Heart Disease Prediction: A Systematic Review

PURPOSE

This review aims to summarize and evaluate the most accurate machinelearning algorithm used to predict ischemic heart disease.

METHODS

This systematic review was performed following PRISMA guidelines. A comprehensive search was carried out using multiple databases such as Science Direct, PubMed\ MEDLINE, CINAHL, and IEEE explore.

RESULTS

Thirteen articles published between 2017 to 2021 were eligible for inclusion. Three themes were extracted: the commonly used algorithm to predict ischemic heart disease, the accuracy of algorithms to predict ischemic heart disease, and the clinical outcomes to improve the quality of care. All methods have utilized supervised and unsupervised machine-learning.

CONCLUSION

Applying machine-learning is expected to assist clinicians in interpreting patients' data and implementing optimal algorithms for their datasets. Furthermore, machine-learning can build evidence-based that supports health care providers to manage individual situations who need invasive procedures such as catheterizations.

A Novel Inflammation-Based Risk Score Predicts Mortality in Acute Type A Aortic Dissection Surgery: The Additive Anti-inflammatory Action for Aortopathy and Arteriopathy Score

OBJECTIVE

To develop an inflammation-based risk stratification tool for operative mortality in patients with acute type A aortic dissection.

METHODS

Between January 1, 2016 and December 31, 2021, 3124 patients from Beijing Anzhen Hospital were included for derivation, 571 patients from the same hospital were included for internal validation, and 1319 patients from other 12 hospitals were included for external validation. The primary outcome was operative mortality according to the Society of Thoracic Surgeons criteria. Least absolute shrinkage and selection operator regression were used to identify clinical risk factors. A model was developed using different machine learning algorithms. The performance of the model was determined using the area under the receiver operating characteristic curve (AUC) for discrimination, calibration curves, and Brier score for calibration. The final model (5A score) was tested with respect to the existing clinical scores.

RESULTS

Extreme gradient boosting was selected for model training (5A score) using 12 variables for prediction-the ratio of platelet to leukocyte count, creatinine level, age, hemoglobin level, prior cardiac surgery, extent of dissection extension, cerebral perfusion, aortic regurgitation, sex, pericardial effusion, shock, and coronary perfusion-which yields the highest AUC (0.873 [95% confidence interval (CI) 0.845-0.901]). The AUC of 5A score was 0.875 (95% CI 0.814-0.936), 0.845 (95% CI 0.811-0.878), and 0.852 (95% CI 0.821-0.883) in the internal, external, and total cohort, respectively, which outperformed the best existing risk score (German Registry for Acute Type A Aortic Dissection score AUC 0.709 [95% CI 0.669-0.749]).

CONCLUSION

The 5A score is a novel, internally and externally validated inflammation-based tool for risk stratification of patients before surgical repair, potentially advancing individualized treatment.

TRIAL REGISTRATION

clinicaltrials.gov Identifier: NCT04918108.

Risk stratification of patients who present with chest pain and have normal troponins using a machine learning model

BACKGROUND

Risk stratification tools exist for patients presenting with chest pain to the emergency room and have achieved the recommended negative predictive value (NPV) of 99%. However, due to low positive predictive value (PPV), current stratification tools result in unwarranted investigations such as serial laboratory tests and cardiac stress tests (CSTs).

AIM

To create a machine learning model (MLM) for risk stratification of chest pain with a better PPV.

METHODS

This retrospective cohort study used de-identified hospital data from January 2016 until November 2021. Inclusion criteria were patients aged > 21 years who presented to the ER, had at least two serum troponins measured, were subsequently admitted to the hospital, and had a CST within 4 d of presentation. Exclusion criteria were elevated troponin value (> 0.05 ng/mL) and missing values for body mass index. The primary outcome was abnormal CST. Demographics, coronary artery disease (CAD) history, hypertension, hyperlipidemia, diabetes mellitus, chronic kidney disease, obesity, and smoking were evaluated as potential risk factors for abnormal CST. Patients were also categorized into a high-risk group (CAD history or more than two risk factors) and a low-risk group (all other patients) for comparison. Bivariate analysis was performed using a χ² test or Fisher’s exact test. Age was compared by t test. Binomial regression (BR), random forest, and XGBoost MLMs were used for prediction. Bootstrapping was used for the internal validation of prediction models. BR was also used for inference. Alpha criterion was set at 0.05 for all statistical tests. R software was used for statistical analysis.

RESULTS

The final cohort of the study included 2328 patients, of which 245 (10.52%) patients had abnormal CST. When adjusted for covariates in the BR model, male sex [risk ratio (RR) = 1.52, 95% confidence interval (CI): 1.2-1.94, P < 0.001)], CAD history (RR = 4.46, 95%CI: 3.08-6.72, P < 0.001), and hyperlipidemia (RR = 3.87, 95%CI: 2.12-8.12, P < 0.001) remained statistically significant. Incidence of abnormal CST was 12.2% in the high-risk group and 2.3% in the low-risk group (RR = 5.31, 95%CI: 2.75-10.24, P < 0.001). The XGBoost model had the best PPV of 24.33%, with an NPV of 91.34% for abnormal CST.

CONCLUSION

The XGBoost MLM achieved a PPV of 24.33% for an abnormal CST, which is better than current stratification tools (13.00%-17.50%). This highlights the beneficial potential of MLMs in clinical decision-making.

Risk Stratification of COVID-19 Using Routine Laboratory Tests: A Machine Learning Approach

The COVID-19 pandemic placed significant stress on an already overburdened health system. The diagnosis was based on detection of a positive RT-PCR test, which may be delayed when there is peak demand for testing. Rapid risk stratification of high-risk patients allows for the prioritization of resources for patient care. The study aims were to classify patients as severe or not severe based on outcomes using machine learning on routine laboratory tests. Data were extracted for all individuals who had at least one SARS-CoV-2 PCR test conducted via the NHLS between the periods of 1 March 2020 to 7 July 2020. Exclusion criteria: those 18 years, and those with indeterminate PCR tests. Results for 15437 patients (3301 positive and 12,136 negative) were used to fit six machine learning models, namely the logistic regression (LR) (the base model), decision trees (DT), random forest (RF), extreme gradient boosting (XGB), convolutional neural network (CNN) and self-normalising neural network (SNN). Model development was carried out by splitting the data into training and testing set of a ratio 70:30, together with a 10-fold cross-validation re-sampling technique. For risk stratification, admission to high care or ICU was the outcome for severe disease. Performance of the models varied: sensitivity was best for RF at 75% and accuracy of 75% for CNN. The area under the curve ranged from 57% for CNN to 75% for RF. RF and SNN were the best-performing models. Machine Learning (ML) can be incorporated into the laboratory information system and offers promise for early identification and risk stratification of COVID-19 patients, particularly in areas of resource-poor settings.

A Predictive Machine Learning Tool for Asthma Exacerbations: Results from a 12-Week, Open-Label Study Using an Electronic Multi-Dose Dry Powder Inhaler with Integrated Sensors

PURPOSE

Machine learning models informed by sensor data inputs have the potential to provide individualized predictions of asthma deterioration. This study aimed to determine if data from an integrated digital inhaler could be used to develop a machine learning model capable of predicting impending exacerbations.

PATIENTS AND METHODS

Adult patients with poorly controlled asthma were enrolled in a 12-week, open-label study using ProAir® Digihaler®, an electronic multi-dose dry powder inhaler (eMDPI) with integrated sensors, as reliever medication (albuterol, 90 µg/dose; 1-2 inhalations every 4 hours, as needed). Throughout the study, the eMDPI recorded inhaler use, peak inspiratory flow (PIF), inhalation volume, inhalation duration, and time to PIF. A model predictive of impending exacerbations was generated by applying machine learning techniques to data downloaded from the inhalers, together with clinical and demographic information. The generated model was evaluated by receiver operating characteristic area under curve (ROC AUC) analysis.

RESULTS

Of 360 patients included in the predictive analysis, 64 experienced a total of 78 exacerbations. Increased albuterol use preceded exacerbations; the mean number of inhalations in the 24-hours preceding an exacerbation was 7.3 (standard deviation 17.3). The machine learning model, using gradient-boosting trees with data from the eMDPI and baseline patient characteristics, predicted an impending exacerbation over the following 5 days with an ROC AUC of 0.83 (95% confidence interval: 0.77-0.90). The feature of the model with the highest weight was the mean number of daily inhalations during the 4 days prior to the day the prediction was made.

CONCLUSION

A machine learning model to predict impending asthma exacerbations using data from the eMDPI was successfully developed. This approach may support a shift from reactive care to proactive, preventative, and personalized management of chronic respiratory diseases.