Bayesian Additive Regression Trees Research Articles

Influential Observations in Bayesian Regression Tree Models

Bayesian Classification and Regression Trees (BCART) and Bayesian Additive Regression Trees (BART) are popular Bayesian regression models widely applicable in modern regression problems. Their popularity is intimately tied to the ability to flexibly model complex responses depending on high-dimensional inputs while simultaneously being able to quantify uncertainties. This ability to quantify uncertainties is key, as it allows researchers to perform appropriate inferential analyses in settings that have generally been too difficult to handle using the Bayesian approach. However, surprisingly little work has been done to evaluate the sensitivity of these modern regression models to violations of modeling assumptions. In particular, we will consider influential observations, which one reasonably would imagine to be common—or at least a concern—in the big-data setting. In this article, we consider both the problem of detecting influential observations and adjusting predictions to not be unduly affected by such potentially problematic data. We consider three detection diagnostics for Bayesian tree models, one an analogue of Cook’s distance and the others taking the form of a divergence measure and a conditional predictive density metric, and then propose an importance sampling algorithm to re-weight previously sampled posterior draws so as to remove the effects of influential data in a computationally efficient manner. Finally, our methods are demonstrated on real-world data where blind application of the models can lead to poor predictions and inference. Supplementary materials for this article are available online.

The Variation of Withdrawal of Life Sustaining Therapy in Older Adults With Traumatic Brain Injury

IntroductionThe decision to withdraw life sustaining treatment (WDLST) in older adults with traumatic brain injury is subject to wide variability leading to nonbeneficial interventions and unnecessary use of hospital resources. We hypothesized that patient and hospital factors are associated with WDLST and WDLST timing. MethodsAll traumatic brain injury patients ≥65 with Glasgow coma scores (GCS) of 4-11 from 2018 to 2019 at level I and II centers were selected from the National Trauma Data Bank. Patients with head abbreviated injury scores 5-6 or death within 24 h were excluded. Bayesian additive regression tree analysis was performed to identify the cumulative incidence function (CIF) and the relative risks (RR) over time for withdrawal of care, discharge to hospice (DH), and death. Death alone (no WDLST or DH) served as the comparator group for all analyses. A subanalysis of the composite outcome WDLST/DH (defined as end-of-life-care), with death (no WDLST or DH) as a comparator cohort was performed. ResultsWe included 2126 patients, of whom 1957 (57%) underwent WDLST, 402 (19%) died, and 469 (22%) were DH. 60% of patients were male, and the mean age was 80 y. The majority of patients were injured by fall (76%, n = 1644). Patients who were DH were more often female (51% DH versus 39% WDLST), had a past medical history of dementia (45% DH versus 18% WDLST), and had lower admission injury severity score (14 DH versus 18.6 WDLST) (P < 0.001). Compared to those who DH, those who underwent WDLST had a lower GCS (9.8 versus 8.4, P < 0.001). CIF of WDSLT and DH increased with age, stabilizing by day 3. At day 3, patients ≥90 y had an increased RR of DH compared to WDLST (RR 2.5 versus 1.4). As GCS increased, CIF and RR of WDLST decreased, while CIF and RR of DH increased (RR on day 3 for GCS 12: WDLST 0.42 versus DH 1.31).Patients at nonprofit institutions were more likely to undergo WDLST (RR 1.15) compared to DH (0.68). Compared to patients of White race, patients of Black race had a lower RR of WDLST at all timepoints. ConclusionsPatient and hospital factors influence the practice of end-of-life-care (WDLST, DH, and death), highlighting the need to better understand variability to target palliative care interventions and standardize care across populations and trauma centers.

Predicting pedestrian crash occurrence and injury severity in Texas using tree-based machine learning models

ABSTRACT This study investigates the frequency and injury severity of pedestrian crashes across Texas using tree-based machine learning models. Ten years of police records are used along with roadway inventory and other sources to map 78,000 + pedestrian crashes over 700,000 road segments. Methods like random forests (RF), gradient boosting (LightGBM and XGBoost), and Bayesian additive regression trees (XBART) are applied and compared. The crash frequency models indicate that highway design variables have significant impacts on crash frequencies. Severity models show how higher speed limits significantly increase the likelihood of pedestrian fatalities and severe injuries, and how intoxication (of drivers or pedestrians) lead to more severe injuries. The 4 specifications perform similarly in predicting crash counts, with LightGBM having much faster computing times. Across the crash-severity models, XBART achieved greater precision values but with significantly higher computating times.

Bayesian risk prediction model for colorectal cancer mortality through integration of clinicopathologic and genomic data

Routine tumor-node-metastasis (TNM) staging of colorectal cancer is imperfect in predicting survival due to tumor pathobiological heterogeneity and imprecise assessment of tumor spread. We leveraged Bayesian additive regression trees (BART), a statistical learning technique, to comprehensively analyze patient-specific tumor characteristics for the improvement of prognostic prediction. Of 75 clinicopathologic, immune, microbial, and genomic variables in 815 stage II–III patients within two U.S.-wide prospective cohort studies, the BART risk model identified seven stable survival predictors. Risk stratifications (low risk, intermediate risk, and high risk) based on model-predicted survival were statistically significant (hazard ratios 0.19–0.45, vs. higher risk; P < 0.0001) and could be externally validated using The Cancer Genome Atlas (TCGA) data (P = 0.0004). BART demonstrated model flexibility, interpretability, and comparable or superior performance to other machine-learning models. Integrated bioinformatic analyses using BART with tumor-specific factors can robustly stratify colorectal cancer patients into prognostic groups and be readily applied to clinical oncology practice.

Bayesian Additive Regression Trees for Classification of Unbalanced Class of Credit Collectability Data

Aims: This study aims at determining the classification results using the Bayesian Additive Regression Trees (BART) method on bank credit collectability data, where there is a class imbalance in the data.&#x0D; Study Design: Quantitative Design.&#x0D; Place and Duration of Study: The used data are secondary data in the form of bank debtor’s credit collectability data with nine predictor variables and one response variable in the form of credit collectability. They are collected from Banks in East Java, Indonesia, from the date of 01 May 1986 to 31 May 2018.&#x0D; Methodology: The Bayesian approach is one of the estimation methods in statistics that is currently being popularly used, this is because the rapid development of technology makes computational challenges no longer a problem. The Bayesian estimation continues to develop and can be used in various statistical methods, for instance both for regression and classification. The Classification and Regression Trees (CART) method is one of the most popular classification methods used. Debtors, in a bank, who have delinquent credit have a small proportion compared to debtors who have current credit. Standard classifier methods such as CART are not suitable for handling this case, as CART is sensitive to classes that have a high degree. Hence, additional methods such as ensemble BART (Bayesian Additive Regression Trees), are needed in order to increase the accuracy of classification in cases of class imbalance.&#x0D; Results: The results of the cross-validation on the BART show a high consistency of classification accuracy, 83.49%. This indicates that the BART method can work consistently even though there is a class imbalance. The results of this study indicate that the classification accuracy of the training data is 84.53%, while the accuracy in the testing data is 85.48%. These results also show that the BART method has ability to overcome overfitting in the classification method, where overfitting often occurs in most of the classification methods that have very good classification abilities.&#x0D; Conclusion: The testing data show that the accuracy is relatively similar to the one of the training data, this indicates that the BART method has been able to capture patterns in the data.

Does Anyone Suffer From Teenage Motherhood? Mental Health Effects of Teen Motherhood in Great Britain Are Small and Homogeneous

Teen mothers experience disadvantage across a wide range of outcomes. However, previous research is equivocal with respect to possible long-term mental health consequences of teen motherhood and has not adequately considered the possibility that effects on mental health may be heterogeneous. Drawing on data from the 1970 British Birth Cohort Study, this article applies a novel statistical machine-learning approach—Bayesian Additive Regression Trees—to estimate the effects of teen motherhood on mental health outcomes at ages 30, 34, and 42. We extend previous work by estimating not only sample-average effects but also individual-specific estimates. Our results show that sample-average mental health effects of teen motherhood are substantively small at all time points, apart from age 30 comparisons to women who first became mothers at age 25‒30. Moreover, we find that these effects are largely homogeneous for all women in the sample—indicating that there are no subgroups in the data who experience important detrimental mental health consequences. We conclude that there are likely no mental health benefits to policy and interventions that aim to prevent teen motherhood.

Causal modelling demonstrates metabolic power is largely affected by gait kinematics and motor control in children with cerebral palsy.

Metabolic power (net energy consumed while walking per unit time) is, on average, two-to-three times greater in children with cerebral palsy (CP) than their typically developing peers, contributing to greater physical fatigue, lower levels of physical activity and greater risk of cardiovascular disease. The goal of this study was to identify the causal effects of clinical factors that may contribute to high metabolic power demand in children with CP. We included children who 1) visited Gillette Children's Specialty Healthcare for a quantitative gait assessment after the year 2000, 2) were formally diagnosed with CP, 3) were classified as level I-III under the Gross Motor Function Classification System and 4) were 18 years old or younger. We created a structural causal model that specified the assumed relationships of a child's gait pattern (i.e., gait deviation index, GDI) and common impairments (i.e., dynamic and selective motor control, strength, and spasticity) with metabolic power. We estimated causal effects using Bayesian additive regression trees, adjusting for factors identified by the causal model. There were 2157 children who met our criteria. We found that a child's gait pattern, as summarized by the GDI, affected metabolic power approximately twice as much as the next largest contributor. Selective motor control, dynamic motor control, and spasticity had the next largest effects. Among the factors we considered, strength had the smallest effect on metabolic power. Our results suggest that children with CP may benefit more from treatments that improve their gait pattern and motor control than treatments that improve spasticity or strength.

Assessing Longitudinal Impacts of Mentor Role on Youth Outcomes

Youth who report prosocial relationships with natural mentors show increased social and academic success and life satisfaction. However, little research investigates how mentors impact aspects of youth identity such as athleticism or religiosity. The present study applies a Bayesian Additive Regression Trees (BART) analysis model to data from a national longitudinal study ( N = 15,701) to predict specific outcomes in educational, athletic, and religious domains. This analytic approach is uniquely well-suited for accurately drawing conclusions within highly collinear, high dimensional datasets. Analyses included demographic variables and childhood base rates of academic success, fitness, and religious beliefs as covariates. Findings indicate that the presence of an academic mentor during adolescence longitudinally predicts educational attainment, while athletic or religious mentors did not have significant impacts in terms of increasing athleticism or religiosity. These results suggest that academic mentors may have more longitudinal impacts on student success than other types of mentors.

Aquatic species shows asymmetric distribution range shifts in native and non-native areas

IntroductionClimate change is reshaping the geographical distribution of species across the globe. In marine ecosystems, climate change creates novel challenges to an environment impacted by numerous anthropogenic stressors. Forecasting shifts in species distribution, including the expansion of non-indigenous species under climate change scenarios, is a management challenge for today’s world.MethodsWe applied Bayesian Additive Regression Tree (BART) models to investigate the environmental factors modulating the occurrence and habitat preferences of the Atlantic blue crab Callinectes sapidus Rathbun, 1896 in native and non-native areas. With BART models, we also aimed to predict its current and future distributions under two climate change scenarios (RCPs 4.5 and 8.5). BART models were performed using global occurrences of the Atlantic blue crab – recorded from 1830 to 2022 – and several environmental covariates (i.e., water temperature, salinity, current velocity, and rugosity). Model accuracy was assessed with the Area Under the Curve (AUC) and True Skill Statistics (TSS) criteria. Cross-validation experiments were made to balance accuracy prediction and uncertainty model prediction intervals.Results and discussionAUC and TSS values indicated that data validation was successful for the BART model. Water temperature was the most critical variable affecting the presence probability of the Atlantic blue crab. The BART model predicts asymmetric distribution range shifts on both sides of the Atlantic Ocean. The non-native populations will experience broader shifts in their distribution range than in the native range, and the RCP 8.5 scenario model outputs predict a wider distribution range by the end of the century. Overall, we anticipate significant ecological changes in native and non-native areas. The range expansion in native areas is often equivalent to the ecological shifts induced by invasive species, so lessons learned by ecologists and managers in non-native areas will provide actionable insights to managers in native areas.

Effective communications on invasive alien species: Identifying communication needs of Swedish domestic garden owners

Invasive alien species threaten biodiversity with domestic gardens acting as a major pathway for the introduction of alien species. Even though the Nordic region is not currently a hotspot for biological invasions, the number of invasions in the Nordic area has been predicted to increase due to climate change. Given a time lag between introduction and invasion, many non-invasive horticultural alien species already introduced into gardens may become invasive in the future. This study aimed to identify the communication needs of Swedish garden owners regarding their management of invasive alien species. A survey among domestic garden owners, informed by topic specialists and local area experts, and interviews with garden owners were conducted in three different bio-climatic areas in a latitudinal gradient across Sweden. The questions targeted invasive alien species and their relations to biodiversity loss and climate change, as well as measures taken to control these species. Analysing the survey data collected in relation to measures taken to control invasive species, Bayesian Additive Regression Tree (BART) modelling was used to identify geographically varying communication needs of the domestic garden owners. In all study areas, the garden owners' measures taken to control invasive alien species were correlated with their strength of beliefs in having experienced local biodiversity loss. A majority of the garden owners were, moreover, uncertain about the impact of climate change on the invasiveness of alien species. In addition, the garden owners' capacity for identifying invasive alien species was often in need of improvement, in particular with respect to the species Impatiens glandulifera, Reynoutria japonica and Rosa rugosa. The results suggest that the evidence-based guidelines for effective communications we developed, have the potential to help communicators meet the local communication needs of garden owners across Sweden, in relation to the management of invasive alien garden species.

Causal Bayesian machine learning to assess treatment effect heterogeneity by dexamethasone dose for patients with COVID-19 and severe hypoxemia

The currently recommended dose of dexamethasone for patients with severe or critical COVID-19 is 6 mg per day (mg/d) regardless of patient features and variation. However, patients with severe or critical COVID-19 are heterogenous in many ways (e.g., age, weight, comorbidities, disease severity, and immune features). Thus, it is conceivable that a standardized dosing protocol may not be optimal. We assessed treatment effect heterogeneity in the COVID STEROID 2 trial, which compared 6 mg/d to 12 mg/d, using a causal inference framework with Bayesian Additive Regression Trees, a flexible modeling method that detects interactive effects and nonlinear relationships among multiple patient characteristics simultaneously. We found that 12 mg/d of dexamethasone, relative to 6 mg/d, was probably associated with better long-term outcomes (days alive without life support and mortality after 90 days) among the entire trial population (i.e., no signals of harm), and probably more beneficial among those without diabetes mellitus, that were older, were not using IL-6 inhibitors at baseline, weighed less, or had higher level respiratory support at baseline. This adds more evidence supporting the use of 12 mg/d in practice for most patients not receiving other immunosuppressants and that additional study of dosing could potentially optimize clinical outcomes.

Epidemic intelligence data of Crimean-Congo haemorrhagic fever, European Region, 2012 to 2022: a new opportunity for risk mapping of neglected diseases.

BackgroundThe Epidemic Intelligence from Open Sources (EIOS) system, jointly developed by the World Health Organisation (WHO), the Joint Research Centre (JRC) of the European Commission and various partners, is a web-based platform that facilitate the monitoring of information on public health threats in near real-time from thousands of online sources.AimsTo assess the capacity of the EIOS system to strengthen data collection for neglected diseases of public health importance, and to evaluate the use of EIOS data for improving the understanding of the geographic extents of diseases and their level of risk.MethodsA Bayesian additive regression trees (BART) model was implemented to map the risk of Crimean-Congo haemorrhagic fever (CCHF) occurrence in 52 countries and territories within the European Region between January 2012 and March 2022 using data on CCHF occurrence retrieved from the EIOS system.ResultsThe model found a positive association between all temperature-related variables and the probability of CCHF occurrence, with an increased risk in warmer and drier areas. The highest risk of CCHF was found in the Mediterranean basin and in areas bordering the Black Sea. There was a general decreasing risk trend from south to north across the entire European Region.ConclusionThe study highlights that the information gathered by public health intelligence can be used to build a disease risk map. Internet-based sources could aid in the assessment of new or changing risks and planning effective actions in target areas.

Cardiovascular disease incidence prediction by machine learning and statistical techniques: a 16-year cohort study from eastern Mediterranean region

BackgroundCardiovascular diseases (CVD) are the predominant cause of early death worldwide. Identification of people with a high risk of being affected by CVD is consequential in CVD prevention. This study adopts Machine Learning (ML) and statistical techniques to develop classification models for predicting the future occurrence of CVD events in a large sample of Iranians.MethodsWe used multiple prediction models and ML techniques with different abilities to analyze the large dataset of 5432 healthy people at the beginning of entrance into the Isfahan Cohort Study (ICS) (1990–2017). Bayesian additive regression trees enhanced with “missingness incorporated in attributes” (BARTm) was run on the dataset with 515 variables (336 variables without and the remaining with up to 90% missing values). In the other used classification algorithms, variables with more than 10% missing values were excluded, and MissForest imputes the missing values of the remaining 49 variables. We used Recursive Feature Elimination (RFE) to select the most contributing variables. Random oversampling technique, recommended cut-point by precision-recall curve, and relevant evaluation metrics were used for handling unbalancing in the binary response variable.ResultsThis study revealed that age, systolic blood pressure, fasting blood sugar, two-hour postprandial glucose, diabetes mellitus, history of heart disease, history of high blood pressure, and history of diabetes are the most contributing factors for predicting CVD incidence in the future. The main differences between the results of classification algorithms are due to the trade-off between sensitivity and specificity. Quadratic Discriminant Analysis (QDA) algorithm presents the highest accuracy (75.50 ± 0.08) but the minimum sensitivity (49.84 ± 0.25); In contrast, decision trees provide the lowest accuracy (51.95 ± 0.69) but the top sensitivity (82.52 ± 1.22). BARTm.90% resulted in 69.48 ± 0.28 accuracy and 54.00 ± 1.66 sensitivity without any preprocessing step.ConclusionsThis study confirmed that building a prediction model for CVD in each region is valuable for screening and primary prevention strategies in that specific region. Also, results showed that using conventional statistical models alongside ML algorithms makes it possible to take advantage of both techniques. Generally, QDA can accurately predict the future occurrence of CVD events with a fast (inference speed) and stable (confidence values) procedure. The combined ML and statistical algorithm of BARTm provide a flexible approach without any need for technical knowledge about assumptions and preprocessing steps of the prediction procedure.

BAYESIAN ADDITIVE REGRESSION TREE APPLICATION FOR PREDICTING MATERNITY RECOVERY RATE OF GROUP LONG-TERM DISABILITY INSURANCE

Bayesian Additive Regression Tree (BART) is a sum-of-trees model used to approximate classification or regression cases. The main idea of this method is to use a prior distribution to keep the tree size small and a likelihood from data to get the posterior. By fixing the tree size as small as possible, the approximation of each tree would have a little effect on the posterior, which is the sum of all output from all the trees used. Bayesian additive regression tree method will be used for predicting the maternity recovery rate of group long-term disability insurance data from the Society of Actuaries (SOA). The decision tree-based models such as Gradient Boosting Machine, Random Forest, Decision Tree, and Bayesian Additive Regression Tree model are compared to find the best model by comparing mean squared error and program runtime. After comparing some models, the Bayesian Additive Regression Tree model gives the best prediction based on smaller root mean squared error values and relatively short runtime.

Using a Multi-Site RCT to Predict Impacts for a Single Site: Do Better Data and Methods Yield More Accurate Predictions?

Multi-site randomized controlled trials (RCTs) provide unbiased estimates of the average impact in the study sample. However, their ability to accurately predict the impact for individual sites outside the study sample, to inform local policy decisions, is largely unknown. To extend prior research on this question, we analyzed six multi-site RCTs and tested modern prediction methods—lasso regression and Bayesian Additive Regression Trees (BART)—using a wide range of moderator variables. The main study findings are that: (1) all of the methods yielded accurate impact predictions when the variation in impacts across sites was close to zero (as expected); (2) none of the methods yielded accurate impact predictions when the variation in impacts across sites was substantial; and (3) BART typically produced “less inaccurate” predictions than lasso regression or than the Sample Average Treatment Effect. These results raise concerns that when the impact of an intervention varies considerably across sites, statistical modeling using the data commonly collected by multi-site RCTs will be insufficient to explain the variation in impacts across sites and accurately predict impacts for individual sites.

Age-dependent failure probabilities of corroding concrete sewer pipes under traffic loads

Concrete pipes are frequently used in the wastewater infrastructure. These buried structures are prone to ageing and deterioration, as well as damage under excessive stresses from traffic. This study presents the first set of temporal fragility models for underground concrete sewer pipes under the compounding effects of corrosion and truck loads. To address the high computational costs of fragility modeling, Bayesian Additive Regression Trees (BART) – a machine-learning based meta-model – are trained using a high-fidelity computational model of buried concrete pipes. The BART models are subsequently applied to develop age-dependent fragility models for serviceability and collapse of pipes conditional on truck loads. Considered uncertainties include statistically significant concrete and soil properties, truck load, and environmental factors that affect the corrosion of concrete sewer pipes. Results indicate that the direction of traffic stream, the degree of loss of wall thickness due to corrosion, and the maximum allowed truck load are critical factors that must be considered when planning inspection, rehabilitation, and retrofit of this underground infrastructure. According to the results, concrete sewer pipes buried in perpendicular to the direction of traffic flow are more vulnerable than those that are parallel to the direction of traffic flow. The results highlight that the probability of collapse is insignificant when it comes to vehicles weighing less than the design truck weight; however, the probability of collapse significantly increases for the moderate and severe corroded sewer pipes when it comes to the overweight vehicles. The age-dependent fragility models indicate that for sewer pipes buried perpendicular to the direction of traffic stream with service life of less than 25 years, the probability of the first major structural defect is less than 5%. This is also true for sewer pipes that have been in service for less than 45 years if the traffic stream is parallel to the buried pipe. The developed fragility models also constitute a key component of quantitative risk assessment of underground wastewater collection systems.

Motor control complexity can be dynamically simplified during gait pattern exploration using motor control-based biofeedback.

Understanding how the central nervous system coordinates diverse motor outputs has been a topic of extensive investigation. Although it is generally accepted that a small set of synergies underlies many common activities, such as walking, whether synergies are equally robust across a broader array of gait patterns or can be flexibly modified remains unclear. Here, we evaluated the extent to which synergies changed as nondisabled adults (n = 14) explored gait patterns using custom biofeedback. Secondarily, we used Bayesian additive regression trees to identify factors that were associated with synergy modulation. Participants explored 41.1 ± 8.0 gait patterns using biofeedback, during which synergy recruitment changed depending on the type and magnitude of gait pattern modification. Specifically, a consistent set of synergies was recruited to accommodate small deviations from baseline, but additional synergies emerged for larger gait changes. Synergy complexity was similarly modulated; complexity decreased for 82.6% of the attempted gait patterns, but distal gait mechanics were strongly associated with these changes. In particular, greater ankle dorsiflexion moments and knee flexion through stance, as well as greater knee extension moments at initial contact, corresponded to a reduction in synergy complexity. Taken together, these results suggest that the central nervous system preferentially adopts a low-dimensional, largely invariant control strategy but can modify that strategy to produce diverse gait patterns. Beyond improving understanding of how synergies are recruited during gait, study outcomes may also help identify parameters that can be targeted with interventions to alter synergies and improve motor control after neurological injury.NEW & NOTEWORTHY We used a motor control-based biofeedback system and machine learning to characterize the extent to which nondisabled adults can modulate synergies during gait pattern exploration. Results revealed that a small library of synergies underlies an array of gait patterns but that recruitment from this library changes as a function of the imposed biomechanical constraints. Our findings enhance understanding of the neural control of gait and may inform biofeedback strategies to improve synergy recruitment after neurological injury.

Association between proteomic biomarkers and myocardial fibrosis measured by MRI: the multi-ethnic study of atherosclerosis.

Cardiac magnetic resonance imaging (CMR) determines the extent of interstitial fibrosis, measured by increased extracellular volume (ECV), and replacement fibrosis with late gadolinium myocardial enhancement (LGE). Despite advances in detection, the pathophysiology of subclinical myocardial fibrosis is incompletely understood. Targeted proteomic discovery technologies enable quantification of low abundance circulating proteins to elucidate cardiac fibrosis mechanisms. Using a cross-sectional design, we selected 92 LGE+cases and 92 LGE- demographically matched controls from the Multi-Ethnic Study of Atherosclerosis. Similarly, we selected 156 cases from the highest ECV quartile and matched with 156 cases from the lowest quartile. The plasma serum proteome was analyzed using proximity extension assays to determine differential regulation of 92 proteins previously implicated with cardiovascular disease. Results were analyzed using volcano plots of statistical significance vs. magnitude of change and Bayesian additive regression tree (BART) models to determine importance. After adjusting for false discovery, higher ECV was significantly associated with 17 proteins. Using BART, Plasminogen activator inhibitor 1, Insulin-like growth factor-binding protein 1, and N-terminal pro-B-type natriuretic peptide were associated with higher ECV after accounting for other proteins and traditional cardiovascular risk factors. In contrast, no circulating proteins were associated with replacement fibrosis. Our results suggest unique circulating proteomic signatures associated with interstitial fibrosis emphasizing its systemic influences. With future validation, protein panels may identify patients who may develop interstitial fibrosis with progression to heart failure. This research was supported by contracts and grants from NHLBI, NCATS and the Inova Heart and Vascular Institute.

Nonparametric failure time: Time-to-event machine learning with heteroskedastic Bayesian additive regression trees and low information omnibus Dirichlet process mixtures.

Many popular survival models rely on restrictive parametric, or semiparametric, assumptions that could provide erroneous predictions when the effects of covariates are complex. Modern advances in computational hardware have led to an increasing interest in flexible Bayesian nonparametric methods for time-to-event data such as Bayesian additive regression trees (BART). We propose a novel approach that we call nonparametric failure time (NFT) BART in order to increase the flexibility beyond accelerated failure time (AFT) and proportional hazard models. NFT BART has three key features: (1)a BART prior for the mean function of the event time logarithm; (2)a heteroskedastic BART prior to deduce a covariate-dependent variance function; and (3)a flexible nonparametric error distribution using Dirichlet process mixtures (DPM). Our proposed approach widens the scope of hazard shapes including nonproportional hazards, can be scaled up to large sample sizes, naturally provides estimates of uncertainty via the posterior and can be seamlessly employed for variable selection. We provide convenient, user-friendly, computer software that is freely available as a reference implementation. Simulations demonstrate that NFT BART maintains excellent performance for survival prediction especially when AFT assumptions are violated by heteroskedasticity. We illustrate the proposed approach on a study examining predictors for mortality risk in patients undergoing hematopoietic stem cell transplant (HSCT) for blood-borne cancer, where heteroskedasticity and nonproportional hazards are likelypresent.

Can we explain machine learning-based prediction for rupture status assessments of intracranial aneurysms?

Although applying machine learning (ML) algorithms to rupture status assessment of intracranial aneurysms (IA) has yielded promising results, the opaqueness of some ML methods has limited their clinical translation. We presented the first explainability comparison of six commonly used ML algorithms: multivariate logistic regression (LR), support vector machine (SVM), random forest (RF), extreme gradient boosting (XGBoost), multi-layer perceptron neural network (MLPNN), and Bayesian additive regression trees (BART). A total of 112 IAs with known rupture status were selected for this study. The ML-based classification used two anatomical features, nine hemodynamic parameters, and thirteen morphologic variables. We utilized permutation feature importance, local interpretable model-agnostic explanations (LIME), and SHapley Additive exPlanations (SHAP) algorithms to explain and analyze 6 Ml algorithms. All models performed comparably: LR area under the curve (AUC) was 0.71; SVM AUC was 0.76; RF AUC was 0.73; XGBoost AUC was 0.78; MLPNN AUC was 0.73; BART AUC was 0.73. Our interpretability analysis demonstrated consistent results across all the methods; i.e., the utility of the top 12 features was broadly consistent. Furthermore, contributions of 9 important features (aneurysm area, aneurysm location, aneurysm type, wall shear stress maximum during systole, ostium area, the size ratio between aneurysm width, (parent) vessel diameter, one standard deviation among time-averaged low shear area, and one standard deviation of temporally averaged low shear area less than 0.4 Pa) were nearly the same. This research suggested that ML classifiers can provide explainable predictions consistent with general domain knowledge concerning IA rupture. With the improved understanding of ML algorithms, clinicians’ trust in ML algorithms will be enhanced, accelerating their clinical translation.