Aprendizaje automático para la toma de decisiones en cardiología: una revisión narrativa para ayudar a recorrer el nuevo panorama / Machine learning for decision-making in cardiology: a narrative review to aid navigating the new landscape

El aprendizaje automático (machine learning) en cardiología es cada vez más frecuente en la literatura médica, pero los modelos de aprendizaje automático aún no han producido un cambio generalizado de la práctica clínica. En parte esto se debe a que el lenguaje utilizado para describir el aprendizaje automático procede de la informática y resulta menos familiar a los lectores de revistas clínicas. En esta revisión narrativa se proporcionan, en primer lugar, algunas orientaciones sobre cómo leer las revistas de aprendizaje automático y, a continuación, orientaciones adicionales para quienes se plantean iniciar un estudio utilizando el aprendizaje automático. Por último, se ilustra el estado actual de la técnica con breves resúmenes de 5 artículos que van desde un modelo de aprendizaje automático muy sencillo hasta otros muy sofisticados.(AU)

Machine learning in cardiology is becoming more commonplace in the medical literature; however, machine learning models have yet to result in a widespread change in practice. This is partly due to the language used to describe machine, which is derived from computer science and may be unfamiliar to readers of clinical journals. In this narrative review, we provide some guidance on how to read machine learning journals and additional guidance for investigators considering instigating a study using machine learning. Finally, we illustrate the current state of the art with brief summaries of 5 articles describing models that range from the very simple to the highly sophisticated.(AU)

Shape Prediction of Nasal Bones by Digital 2D-Photogrammetry of the Nose Based on Convolution and Back-Propagation Neural Network.

In rhinoplasty, it is necessary to consider the correlation between the anthropometric indicators of the nasal bone, so that it prevents surgical complications and enhances the patient's satisfaction. The penetrating form of high-energy electromagnetic radiation is highly impacted on human health, which has often raised concerns of alternative method for facial analysis. The critical stage to assess nasal morphology is the nasal analysis on its anthropology that is highly reliant on the understanding of the structural features of the nasal radix. For example, the shape and size of nasal bone features, skin thickness, and also body factors aggregated from different facial anthropology values. In medical diagnosis, however, the morphology of the nasal bone is determined manually and significantly relies on the clinician's expertise. Furthermore, the evaluation anthropological keypoint of the nasal bone is nonrepeatable and laborious, also finding widely differ and intralaboratory variability in the results because of facial soft tissue and equipment defects. In order to overcome these problems, we propose specialized convolutional neural network (CNN) architecture to accurately predict nasal measurement based on digital 2D photogrammetry. To boost performance and efficacy, it is deliberately constructed with many layers and different filter sizes, with less filters and optimizing parameters. Through its result, the back-propagation neural network (BPNN) indicated the correlation between differences in human body factors mentioned are height, weight known as body mass index (BMI), age, gender, and the nasal bone dimension of the participant. With full of parameters could the nasal morphology be diagnostic continuously. The model's performance is evaluated on various newest architecture models such as DenseNet, ConvNet, Inception, VGG, and MobileNet. Experiments were directly conducted on different facials. The results show the proposed architecture worked well in terms of nasal properties achieved which utilize four statistical criteria named mean average precision (mAP), mean absolute error (MAE), R-square (R 2), and T-test analyzed. Data has also shown that the nasal shape of Southeast Asians, especially Vietnamese, could be divided into different types in two perspective views. From cadavers for bony datasets, nasal bones can be classified into 2 morphological types in the lateral view which "V" shape was presented by 78.8% and the remains were "S" shape evaluated based on Lazovic (2015). With 2 angular dimension averages are 136.41 ± 7.99 and 104.25 ± 5.95 represented by the nasofrontal angle (g-n-prn) and the nasomental angle (n-prn-sn), respectively. For frontal view, classified by Hwang, Tae-Sun, et al. (2005), nasal morphology of Vietnamese participants could be divided into three types: type A was present in 57.6% and type B was present in 30.3% of the noses. In particular, types C, D, and E were not a common form of Vietnamese which includes the remaining number of participants. In conclusion, the proposed model performed the potential hybrid of CNN and BPNN with its application to give expected accuracy in terms of keypoint localization and nasal morphology regression. Nasal analysis can replace MRI imaging diagnostics that are reflected by the risk to human body.

Building quantum mechanics quality force fields of proteins with the generalized energy-based fragmentation approach and machine learning.

We combined our generalized energy-based fragmentation (GEBF) approach and machine learning (ML) technique to construct quantum mechanics (QM) quality force fields for proteins. In our scheme, the training sets for a protein are only constructed from its small subsystems, which capture all short-range interactions in the target system. The energy of a given protein is expressed as the summation of atomic contributions from QM calculations of various subsystems, corrected by long-range Coulomb and van der Waals interactions. With the Gaussian approximation potential (GAP) method, our protocol can automatically generate training sets with high efficiency. To facilitate the construction of training sets for proteins, we store all trained subsystem data in a library. If subsystems in the library are detected in a new protein, corresponding datasets can be directly reused as a part of the training set on this new protein. With two polypeptides, 4ZNN and 1XQ8 segment, as examples, the energies and forces predicted by GEBF-GAP are in good agreement with those from conventional QM calculations, and dihedral angle distributions from GEBF-GAP molecular dynamics (MD) simulations can also well reproduce those from ab initio MD simulations. In addition, with the training set generated from GEBF-GAP, we also demonstrate that GEBF-ML force fields constructed by neural network (NN) methods can also show QM quality. Therefore, the present work provides an efficient and systematic way to build QM quality force fields for biological systems.

Comparison of Machine Learning Methods for Predicting Outcomes After In-Hospital Cardiac Arrest.

OBJECTIVES: Prognostication of neurologic status among survivors of in-hospital cardiac arrests remains a challenging task for physicians. Although models such as the Cardiac Arrest Survival Post-Resuscitation In-hospital score are useful for predicting neurologic outcomes, they were developed using traditional statistical techniques. In this study, we derive and compare the performance of several machine learning models with each other and with the Cardiac Arrest Survival Post-Resuscitation In-hospital score for predicting the likelihood of favorable neurologic outcomes among survivors of resuscitation. DESIGN: Analysis of the Get With The Guidelines-Resuscitation registry. SETTING: Seven-hundred fifty-five hospitals participating in Get With The Guidelines-Resuscitation from January 1, 2001, to January 28, 2017. PATIENTS: Adult in-hospital cardiac arrest survivors. INTERVENTIONS: None. MEASUREMENTS AND MAIN RESULTS: Of 117,674 patients in our cohort, 28,409 (24%) had a favorable neurologic outcome, as defined as survival with a Cerebral Performance Category score of less than or equal to 2 at discharge. Using patient characteristics, pre-existing conditions, prearrest interventions, and periarrest variables, we constructed logistic regression, support vector machines, random forests, gradient boosted machines, and neural network machine learning models to predict favorable neurologic outcome. Events prior to October 20, 2009, were used for model derivation, and all subsequent events were used for validation. The gradient boosted machine predicted favorable neurologic status at discharge significantly better than the Cardiac Arrest Survival Post-Resuscitation In-hospital score (C-statistic: 0.81 vs 0.73; p < 0.001) and outperformed all other machine learning models in terms of discrimination, calibration, and accuracy measures. Variables that were consistently most important for prediction across all models were duration of arrest, initial cardiac arrest rhythm, admission Cerebral Performance Category score, and age. CONCLUSIONS: The gradient boosted machine algorithm was the most accurate for predicting favorable neurologic outcomes in in-hospital cardiac arrest survivors. Our results highlight the utility of machine learning for predicting neurologic outcomes in resuscitated patients.

Improving Stroke Risk Prediction in the General Population: A Comparative Assessment of Common Clinical Rules, a New Multimorbid Index, and Machine-Learning-Based Algorithms.

BACKGROUND: There are few large studies examining and predicting the diversified cardiovascular/noncardiovascular comorbidity relationships with stroke. We investigated stroke risks in a very large prospective cohort of patients with multimorbidity, using two common clinical rules, a clinical multimorbid index and a machine-learning (ML) approach, accounting for the complex relationships among variables, including the dynamic nature of changing risk factors. METHODS: We studied a prospective U.S. cohort of 3,435,224 patients from medical databases in a 2-year investigation. Stroke outcomes were examined in relationship to diverse multimorbid conditions, demographic variables, and other inputs, with ML accounting for the dynamic nature of changing multimorbidity risk factors, two clinical risk scores, and a clinical multimorbid index. RESULTS: Common clinical risk scores had moderate and comparable c indices with stroke outcomes in the training and external validation samples (validation-CHADS2: c index 0.812, 95% confidence interval [CI] 0.808-0.815; CHA2DS2-VASc: c index 0.809, 95% CI 0.805-0.812). A clinical multimorbid index had higher discriminant validity values for both the training/external validation samples (validation: c index 0.850, 95% CI 0.847-0.853). The ML-based algorithms yielded the highest discriminant validity values for the gradient boosting/neural network logistic regression formulations with no significant differences among the ML approaches (validation for logistic regression: c index 0.866, 95% CI 0.856-0.876). Calibration of the ML-based formulation was satisfactory across a wide range of predicted probabilities. Decision curve analysis demonstrated that clinical utility for the ML-based formulation was better than that for the two current clinical rules and the newly developed multimorbid tool. Also, ML models and clinical stroke risk scores were more clinically useful than the "treat all" strategy. CONCLUSION: Complex relationships of various comorbidities uncovered using a ML approach for diverse (and dynamic) multimorbidity changes have major consequences for stroke risk prediction. This approach may facilitate automated approaches for dynamic risk stratification in the significant presence of multimorbidity, helping in the decision-making process for risk assessment and integrated/holistic management.

Stool microbiota are superior to saliva in distinguishing cirrhosis and hepatic encephalopathy using machine learning.

BACKGROUND & AIMS: Saliva and stool microbiota are altered in cirrhosis. Since stool is logistically difficult to collect compared to saliva, it is important to determine their relative diagnostic and prognostic capabilities. We aimed to determine the ability of stool vs. saliva microbiota to differentiate between groups based on disease severity using machine learning (ML). METHODS: Controls and outpatients with cirrhosis underwent saliva and stool microbiome analysis. Controls vs. cirrhosis and within cirrhosis (based on hepatic encephalopathy [HE], proton pump inhibitor [PPI] and rifaximin use) were classified using 4 ML techniques (random forest [RF], support vector machine, logistic regression, and gradient boosting) with AUC comparisons for stool, saliva or both sample types. Individual microbial contributions were computed using feature importance of RF and Shapley additive explanations. Finally, thresholds for including microbiota were varied between 2.5% and 10%, and core microbiome (DESeq2) analysis was performed. RESULTS: Two hundred and sixty-nine participants, including 87 controls and 182 patients with cirrhosis, of whom 57 had HE, 78 were on PPIs and 29 on rifaximin were included. Regardless of the ML model, stool microbiota had a significantly higher AUC in differentiating groups vs. saliva. Regarding individual microbiota: autochthonous taxa drove the difference between controls vs. patients with cirrhosis, oral-origin microbiota the difference between PPI users/non-users, and pathobionts and autochthonous taxa the difference between rifaximin users/non-users and patients with/without HE. These were consistent with the core microbiome analysis results. CONCLUSIONS: On ML analysis, stool microbiota composition is significantly more informative in differentiating between controls and patients with cirrhosis, and those with varying cirrhosis severity, compared to saliva. Despite logistic challenges, stool should be preferred over saliva for microbiome analysis. LAY SUMMARY: Since it is harder to collect stool than saliva, we wanted to test whether microbes from saliva were better than stool in differentiating between healthy people and those with cirrhosis and, among those with cirrhosis, those with more severe disease. Using machine learning, we found that microbes in stool were more accurate than saliva alone or in combination, therefore, stool should be preferred for analysis and collection wherever possible.

Analyzing the Check-In Behavior of Visitors through Machine Learning Model by Mining Social Network's Big Data.

The current article paper is aimed at assessing and comparing the seasonal check-in behavior of individuals in Shanghai, China, using location-based social network (LBSN) data and a variety of spatiotemporal analytic techniques. The article demonstrates the uses of location-based social network's data by analyzing the trends in check-ins throughout a three-year term for health purpose. We obtained the geolocation data from Sina Weibo, one of the biggest renowned Chinese microblogs (Weibo). The composed data is converted to geographic information system (GIS) type and assessed using temporal statistical analysis and spatial statistical analysis using kernel density estimation (KDE) assessment. We have applied various algorithms and trained machine learning models and finally satisfied with sequential model results because the accuracy we got was leading amongst others. The location cataloguing is accomplished via the use of facts about the characteristics of physical places. The findings demonstrate that visitors' spatial operations are more intense than residents' spatial operations, notably in downtown. However, locals also visited outlying regions, and tourists' temporal behaviors vary significantly while citizens' movements exhibit a more steady stable behavior. These findings may be used in destination management, metro planning, and the creation of digital cities.

Performance of Machine Learning Algorithms for Predicting Progression to Dementia in Memory Clinic Patients.

Importance: Machine learning algorithms could be used as the basis for clinical decision-making aids to enhance clinical practice. Objective: To assess the ability of machine learning algorithms to predict dementia incidence within 2 years compared with existing models and determine the optimal analytic approach and number of variables required. Design, Setting, and Participants: This prognostic study used data from a prospective cohort of 15 307 participants without dementia at baseline to perform a secondary analysis of factors that could be used to predict dementia incidence. Participants attended National Alzheimer Coordinating Center memory clinics across the United States between 2005 and 2015. Analyses were conducted from March to May 2021. Exposures: 258 variables spanning domains of dementia-related clinical measures and risk factors. Main Outcomes and Measures: The main outcome was incident all-cause dementia diagnosed within 2 years of baseline assessment. Results: In a sample of 15 307 participants (mean [SD] age, 72.3 [9.8] years; 9129 [60%] women and 6178 [40%] men) without dementia at baseline, 1568 (10%) received a diagnosis of dementia within 2 years of their initial assessment. Compared with 2 existing models for dementia risk prediction (ie, Cardiovascular Risk Factors, Aging, and Incidence of Dementia Risk Score, and the Brief Dementia Screening Indicator), machine learning algorithms were superior in predicting incident all-cause dementia within 2 years. The gradient-boosted trees algorithm had a mean (SD) overall accuracy of 92% (1%), sensitivity of 0.45 (0.05), specificity of 0.97 (0.01), and area under the curve of 0.92 (0.01) using all 258 variables. Analysis of variable importance showed that only 6 variables were required for machine learning algorithms to achieve an accuracy of 91% and area under the curve of at least 0.89. Machine learning algorithms also identified up to 84% of participants who received an initial dementia diagnosis that was subsequently reversed to mild cognitive impairment or cognitively unimpaired, suggesting possible misdiagnosis. Conclusions and Relevance: These findings suggest that machine learning algorithms could accurately predict incident dementia within 2 years in patients receiving care at memory clinics using only 6 variables. These findings could be used to inform the development and validation of decision-making aids in memory clinics.

Characterizing Opioid Overdoses Using Emergency Medical Services Data : A Case Definition Algorithm Enhanced by Machine Learning.

OBJECTIVES: Tracking nonfatal overdoses in the escalating opioid overdose epidemic is important but challenging. The objective of this study was to create an innovative case definition of opioid overdose in North Carolina emergency medical services (EMS) data, with flexible methodology for application to other states' data. METHODS: This study used de-identified North Carolina EMS encounter data from 2010-2015 for patients aged >12 years to develop a case definition of opioid overdose using an expert knowledge, rule-based algorithm reflecting whether key variables identified drug use/poisoning or overdose or whether the patient received naloxone. We text mined EMS narratives and applied a machine-learning classification tree model to the text to predict cases of opioid overdose. We trained models on the basis of whether the chief concern identified opioid overdose. RESULTS: Using a random sample from the data, we found the positive predictive value of this case definition to be 90.0%, as compared with 82.7% using a previously published case definition. Using our case definition, the number of unresponsive opioid overdoses increased from 3412 in 2010 to 7194 in 2015. The corresponding monthly rate increased by a factor of 1.7 from January 2010 (3.0 per 1000 encounters; n = 261 encounters) to December 2015 (5.1 per 1000 encounters; n = 622 encounters). Among EMS responses for unresponsive opioid overdose, the prevalence of naloxone use was 83%. CONCLUSIONS: This study demonstrates the potential for using machine learning in combination with a more traditional substantive knowledge algorithm-based approach to create a case definition for opioid overdose in EMS data.

Automated Assessment of Brain CT After Cardiac Arrest-An Observational Derivation/Validation Cohort Study.

OBJECTIVES: Prognostication of outcome is an essential step in defining therapeutic goals after cardiac arrest. Gray-white-matter ratio obtained from brain CT can predict poor outcome. However, manual placement of regions of interest is a potential source of error and interrater variability. Our objective was to assess the performance of poor outcome prediction by automated quantification of changes in brain CTs after cardiac arrest. DESIGN: Observational, derivation/validation cohort study design. Outcome was determined using the Cerebral Performance Category upon hospital discharge. Poor outcome was defined as death or unresponsive wakefulness syndrome/coma. CTs were automatically decomposed using coregistration with a brain atlas. SETTING: ICUs at a large, academic hospital with circulatory arrest center. PATIENTS: We identified 433 cardiac arrest patients from a large previously established database with brain CTs within 10 days after cardiac arrest. INTERVENTIONS: None. MEASUREMENTS AND MAIN RESULTS: Five hundred sixteen brain CTs were evaluated (derivation cohort n = 309, validation cohort n = 207). Patients with poor outcome had significantly lower radiodensities in gray matter regions. Automated GWR_si (putamen/posterior limb of internal capsule) was performed with an area under the curve of 0.86 (95%-CI: 0.80-0.93) for CTs taken later than 24 hours after cardiac arrest (similar performance in the validation cohort). Poor outcome (Cerebral Performance Category 4-5) was predicted with a specificity of 100% (95% CI, 87-100%, derivation; 88-100%, validation) at a threshold of less than 1.10 and a sensitivity of 49% (95% CI, 36-58%, derivation) and 38% (95% CI, 27-50%, validation) for CTs later than 24 hours after cardiac arrest. Sensitivity and area under the curve were lower for CTs performed within 24 hours after cardiac arrest. CONCLUSIONS: Automated gray-white-matter ratio from brain CT is a promising tool for prediction of poor neurologic outcome after cardiac arrest with high specificity and low-to-moderate sensitivity. Prediction by gray-white-matter ratio at the basal ganglia level performed best. Sensitivity increased considerably for CTs performed later than 24 hours after cardiac arrest.

Investigating Bacterial Volatilome for the Classification and Identification of Mycobacterial Species by HS-SPME-GC-MS and Machine Learning.

Species of Mycobacteriaceae cause disease in animals and humans, including tuberculosis and leprosy. Individuals infected with organisms in the Mycobacterium tuberculosis complex (MTBC) or non-tuberculous mycobacteria (NTM) may present identical symptoms, however the treatment for each can be different. Although the NTM infection is considered less vital due to the chronicity of the disease and the infrequency of occurrence in healthy populations, diagnosis and differentiation among Mycobacterium species currently require culture isolation, which can take several weeks. The use of volatile organic compounds (VOCs) is a promising approach for species identification and in recent years has shown promise for use in the rapid analysis of both in vitro cultures as well as ex vivo diagnosis using breath or sputum. The aim of this contribution is to analyze VOCs in the culture headspace of seven different species of mycobacteria and to define the volatilome profiles that are discriminant for each species. For the pre-concentration of VOCs, solid-phase micro-extraction (SPME) was employed and samples were subsequently analyzed using gas chromatography-quadrupole mass spectrometry (GC-qMS). A machine learning approach was applied for the selection of the 13 discriminatory features, which might represent clinically translatable bacterial biomarkers.

A transfer learning approach to facilitate ComBat-based harmonization of multicentre radiomic features in new datasets.

PURPOSE: To facilitate the demonstration of the prognostic value of radiomics, multicenter radiomics studies are needed. Pooling radiomic features of such data in a statistical analysis is however challenging, as they are sensitive to the variability in scanner models, acquisition protocols and reconstruction settings, which is often unavoidable in a multicentre retrospective analysis. A statistical harmonization strategy called ComBat was utilized in radiomics studies to deal with the "center-effect". The goal of the present work was to integrate a transfer learning (TL) technique within ComBat-and recently developed alternate versions of ComBat with improved flexibility (M-ComBat) and robustness (B-ComBat)-to allow the use of a previously determined harmonization transform to the radiomic feature values of new patients from an already known center. MATERIAL AND METHODS: The proposed TL approach were incorporated in the four versions of ComBat (standard, B, M, and B-M ComBat). The proposed approach was evaluated using a dataset of 189 locally advanced cervical cancer patients from 3 centers, with magnetic resonance imaging (MRI) and positron emission tomography (PET) images, with the clinical endpoint of predicting local failure. The impact performance of the TL approach was evaluated by comparing the harmonization achieved using only parts of the data to the reference (harmonization achieved using all the available data). It was performed through three different machine learning pipelines. RESULTS: The proposed TL technique was successful in harmonizing features of new patients from a known center in all versions of ComBat, leading to predictive models reaching similar performance as the ones developed using the features harmonized with all the data available. CONCLUSION: The proposed TL approach enables applying a previously determined ComBat transform to new, previously unseen data.

A hybrid CNN-LSTM model for pre-miRNA classification.

miRNAs (or microRNAs) are small, endogenous, and noncoding RNAs construct of about 22 nucleotides. Cumulative evidence from biological experiments shows that miRNAs play a fundamental and important role in various biological processes. Therefore, the classification of miRNA is a critical problem in computational biology. Due to the short length of mature miRNAs, many researchers are working on precursor miRNAs (pre-miRNAs) with longer sequences and more structural features. Pre-miRNAs can be divided into two groups as mirtrons and canonical miRNAs in terms of biogenesis differences. Compared to mirtrons, canonical miRNAs are more conserved and easier to be identified. Many existing pre-miRNA classification methods rely on manual feature extraction. Moreover, these methods focus on either sequential structure or spatial structure of pre-miRNAs. To overcome the limitations of previous models, we propose a nucleotide-level hybrid deep learning method based on a CNN and LSTM network together. The prediction resulted in 0.943 (%95 CI ± 0.014) accuracy, 0.935 (%95 CI ± 0.016) sensitivity, 0.948 (%95 CI ± 0.029) specificity, 0.925 (%95 CI ± 0.016) F1 Score and 0.880 (%95 CI ± 0.028) Matthews Correlation Coefficient. When compared to the closest results, our proposed method revealed the best results for Acc., F1 Score, MCC. These were 2.51%, 1.00%, and 2.43% higher than the closest ones, respectively. The mean of sensitivity ranked first like Linear Discriminant Analysis. The results indicate that the hybrid CNN and LSTM networks can be employed to achieve better performance for pre-miRNA classification. In future work, we study on investigation of new classification models that deliver better performance in terms of all the evaluation criteria.

Predictive Risk Models for Wound Infection-Related Hospitalization or ED Visits in Home Health Care Using Machine-Learning Algorithms.

OBJECTIVE: Wound infection is prevalent in home healthcare (HHC) and often leads to hospitalizations. However, none of the previous studies of wounds in HHC have used data from clinical notes. Therefore, the authors created a more accurate description of a patient's condition by extracting risk factors from clinical notes to build predictive models to identify a patient's risk of wound infection in HHC. METHODS: The structured data (eg, standardized assessments) and unstructured information (eg, narrative-free text charting) were retrospectively reviewed for HHC patients with wounds who were served by a large HHC agency in 2014. Wound infection risk factors were identified through bivariate analysis and stepwise variable selection. Risk predictive performance of three machine learning models (logistic regression, random forest, and artificial neural network) was compared. RESULTS: A total of 754 of 54,316 patients (1.39%) had a hospitalization or ED visit related to wound infection. In the bivariate logistic regression, language describing wound type in the patient's clinical notes was strongly associated with risk (odds ratio, 9.94; P < .05). The areas under the curve were 0.82 in logistic regression, 0.75 in random forest, and 0.78 in artificial neural network. Risk prediction performance of the models improved (by up to 13.2%) after adding risk factors extracted from clinical notes. CONCLUSIONS: Logistic regression showed the best risk prediction performance in prediction of wound infection-related hospitalization or ED visits in HHC. The use of data extracted from clinical notes can improve the performance of risk prediction models.

Performance of a Machine Learning Algorithm Using Electronic Health Record Data to Identify and Estimate Survival in a Longitudinal Cohort of Patients With Lung Cancer.

Importance: Electronic health records (EHRs) provide a low-cost means of accessing detailed longitudinal clinical data for large populations. A lung cancer cohort assembled from EHR data would be a powerful platform for clinical outcome studies. Objective: To investigate whether a clinical cohort assembled from EHRs could be used in a lung cancer prognosis study. Design, Setting, and Participants: In this cohort study, patients with lung cancer were identified among 76 643 patients with at least 1 lung cancer diagnostic code deposited in an EHR in Mass General Brigham health care system from July 1988 to October 2018. Patients were identified via a semisupervised machine learning algorithm, for which clinical information was extracted from structured and unstructured data via natural language processing tools. Data completeness and accuracy were assessed by comparing with the Boston Lung Cancer Study and against criterion standard EHR review results. A prognostic model for non-small cell lung cancer (NSCLC) overall survival was further developed for clinical application. Data were analyzed from March 2019 through July 2020. Exposures: Clinical data deposited in EHRs for cohort construction and variables of interest for the prognostic model were collected. Main Outcomes and Measures: The primary outcomes were the performance of the lung cancer classification model and the quality of the extracted variables; the secondary outcome was the performance of the prognostic model. Results: Among 76 643 patients with at least 1 lung cancer diagnostic code, 42 069 patients were identified as having lung cancer, with a positive predictive value of 94.4%. The study cohort consisted of 35 375 patients (16 613 men [47.0%] and 18 756 women [53.0%]; 30 140 White individuals [85.2%], 1040 Black individuals [2.9%], and 857 Asian individuals [2.4%]) after excluding patients with lung cancer history and less than 14 days of follow-up after initial diagnosis. The median (interquartile range) age at diagnosis was 66.7 (58.4-74.1) years. The area under the receiver operating characteristic curves of the prognostic model for overall survival with NSCLC were 0.828 (95% CI, 0.815-0.842) for 1-year prediction, 0.825 (95% CI, 0.812-0.836) for 2-year prediction, 0.814 (95% CI, 0.800-0.826) for 3-year prediction, 0.814 (95% CI, 0.799-0.828) for 4-year prediction, and 0.812 (95% CI, 0.798-0.825) for 5-year prediction. Conclusions and Relevance: These findings suggest the feasibility of assembling a large-scale EHR-based lung cancer cohort with detailed longitudinal clinical measurements and that EHR data may be applied in cancer progression with a set of generalizable approaches.

Data-driven and interpretable machine-learning modeling to explore the fine-scale environmental determinants of malaria vectors biting rates in rural Burkina Faso.

BACKGROUND: Improving the knowledge and understanding of the environmental determinants of malaria vector abundance at fine spatiotemporal scales is essential to design locally tailored vector control intervention. This work is aimed at exploring the environmental tenets of human-biting activity in the main malaria vectors (Anopheles gambiae s.s., Anopheles coluzzii and Anopheles funestus) in the health district of Diébougou, rural Burkina Faso. METHODS: Anopheles human-biting activity was monitored in 27 villages during 15 months (in 2017-2018), and environmental variables (meteorological and landscape) were extracted from high-resolution satellite imagery. A two-step data-driven modeling study was then carried out. Correlation coefficients between the biting rates of each vector species and the environmental variables taken at various temporal lags and spatial distances from the biting events were first calculated. Then, multivariate machine-learning models were generated and interpreted to (i) pinpoint primary and secondary environmental drivers of variation in the biting rates of each species and (ii) identify complex associations between the environmental conditions and the biting rates. RESULTS: Meteorological and landscape variables were often significantly correlated with the vectors' biting rates. Many nonlinear associations and thresholds were unveiled by the multivariate models, for both meteorological and landscape variables. From these results, several aspects of the bio-ecology of the main malaria vectors were identified or hypothesized for the Diébougou area, including breeding site typologies, development and survival rates in relation to weather, flight ranges from breeding sites and dispersal related to landscape openness. CONCLUSIONS: Using high-resolution data in an interpretable machine-learning modeling framework proved to be an efficient way to enhance the knowledge of the complex links between the environment and the malaria vectors at a local scale. More broadly, the emerging field of interpretable machine learning has significant potential to help improve our understanding of the complex processes leading to malaria transmission, and to aid in developing operational tools to support the fight against the disease (e.g. vector control intervention plans, seasonal maps of predicted biting rates, early warning systems).

Tissue outcome prediction in hyperacute ischemic stroke: Comparison of machine learning models.

Machine Learning (ML) has been proposed for tissue fate prediction after acute ischemic stroke (AIS), with the aim to help treatment decision and patient management. We compared three different ML models to the clinical method based on diffusion-perfusion thresholding for the voxel-based prediction of final infarct, using a large MRI dataset obtained in a cohort of AIS patients prior to recanalization treatment. Baseline MRI (MRI0), including diffusion-weighted sequence (DWI) and Tmax maps from perfusion-weighted sequence, and 24-hr follow-up MRI (MRI24h) were retrospectively collected in consecutive 394 patients AIS patients (median age = 70 years; final infarct volume = 28mL). Manually segmented DWI24h lesion was considered the final infarct. Gradient Boosting, Random Forests and U-Net were trained using DWI, apparent diffusion coefficient (ADC) and Tmax maps on MRI0 as inputs to predict final infarct. Tissue outcome predictions were compared to final infarct using Dice score. Gradient Boosting had significantly better predictive performance (median [IQR] Dice Score as for median age, maybe you can replace the comma with an equal sign for consistency 0.53 [0.29-0.68]) than U-Net (0.48 [0.18-0.68]), Random Forests (0.51 [0.27-0.66]), and clinical thresholding method (0.45 [0.25-0.62]) (P < 0.001). In this benchmark of ML models for tissue outcome prediction in AIS, Gradient Boosting outperformed other ML models and clinical thresholding method and is thus promising for future decision-making.