Predicting post-liver transplant outcomes in patients with acute-on-chronic liver failure using Expert-Augmented Machine Learning.

Liver transplantation (LT) is a treatment for acute-on-chronic liver failure (ACLF), but high post-LT mortality has been reported. Existing post-LT models in ACLF have been limited. We developed an Expert-Augmented Machine Learning (EAML) model to predict post-LT outcomes. We identified ACLF patients who underwent LT in the University of California Health Data Warehouse. We applied the RuleFit machine learning (ML) algorithm to extract rules from decision trees and create intermediate models. We asked human experts to rate the rules generated by RuleFit and incorporated these ratings to generate final EAML models. We identified 1384 ACLF patients. For death at 1 year, areas under the receiver-operating characteristic curve were 0.707 (confidence interval [CI] 0.625-0.793) for EAML and 0.719 (CI 0.640-0.800) for RuleFit. For death at 90 days, areas under the receiver-operating characteristic curve were 0.678 (CI 0.581-0.776) for EAML and 0.707 (CI 0.615-0.800) for RuleFit. In pairwise comparisons, both EAML and RuleFit models outperformed cross-sectional models. Significant discrepancies between experts and ML occurred in rankings of biomarkers used in clinical practice. EAML may serve as a method for ML-guided hypothesis generation in further ACLF research.

Expert-augmented machine learning.

Machine learning is proving invaluable across disciplines. However, its success is often limited by the quality and quantity of available data, while its adoption is limited by the level of trust afforded by given models. Human vs. machine performance is commonly compared empirically to decide whether a certain task should be performed by a computer or an expert. In reality, the optimal learning strategy may involve combining the complementary strengths of humans and machines. Here, we present expert-augmented machine learning (EAML), an automated method that guides the extraction of expert knowledge and its integration into machine-learned models. We used a large dataset of intensive-care patient data to derive 126 decision rules that predict hospital mortality. Using an online platform, we asked 15 clinicians to assess the relative risk of the subpopulation defined by each rule compared to the total sample. We compared the clinician-assessed risk to the empirical risk and found that, while clinicians agreed with the data in most cases, there were notable exceptions where they overestimated or underestimated the true risk. Studying the rules with greatest disagreement, we identified problems with the training data, including one miscoded variable and one hidden confounder. Filtering the rules based on the extent of disagreement between clinician-assessed risk and empirical risk, we improved performance on out-of-sample data and were able to train with less data. EAML provides a platform for automated creation of problem-specific priors, which help build robust and dependable machine-learning models in critical applications.

Structural and Functional Brain Parameters Related to Cognitive Performance Across Development: Replication and Extension of the Parieto-Frontal Integration Theory in a Single Sample.

The parieto-frontal integration theory (PFIT) identified a fronto-parietal network of regions where individual differences in brain parameters most strongly relate to cognitive performance. PFIT was supported and extended in adult samples, but not in youths or within single-scanner well-powered multimodal studies. We performed multimodal neuroimaging in 1601 youths age 8-22 on the same 3-Tesla scanner with contemporaneous neurocognitive assessment, measuring volume, gray matter density (GMD), mean diffusivity (MD), cerebral blood flow (CBF), resting-state functional magnetic resonance imaging measures of the amplitude of low frequency fluctuations (ALFFs) and regional homogeneity (ReHo), and activation to a working memory and a social cognition task. Across age and sex groups, better performance was associated with higher volumes, greater GMD, lower MD, lower CBF, higher ALFF and ReHo, and greater activation for the working memory task in PFIT regions. However, additional cortical, striatal, limbic, and cerebellar regions showed comparable effects, hence PFIT needs expansion into an extended PFIT (ExtPFIT) network incorporating nodes that support motivation and affect. Associations of brain parameters became stronger with advancing age group from childhood to adolescence to young adulthood, effects occurring earlier in females. This ExtPFIT network is developmentally fine-tuned, optimizing abundance and integrity of neural tissue while maintaining a low resting energy state.

Building more accurate decision trees with the additive tree.

The expansion of machine learning to high-stakes application domains such as medicine, finance, and criminal justice, where making informed decisions requires clear understanding of the model, has increased the interest in interpretable machine learning. The widely used Classification and Regression Trees (CART) have played a major role in health sciences, due to their simple and intuitive explanation of predictions. Ensemble methods like gradient boosting can improve the accuracy of decision trees, but at the expense of the interpretability of the generated model. Additive models, such as those produced by gradient boosting, and full interaction models, such as CART, have been investigated largely in isolation. We show that these models exist along a spectrum, revealing previously unseen connections between these approaches. This paper introduces a rigorous formalization for the additive tree, an empirically validated learning technique for creating a single decision tree, and shows that this method can produce models equivalent to CART or gradient boosted stumps at the extremes by varying a single parameter. Although the additive tree is designed primarily to provide both the model interpretability and predictive performance needed for high-stakes applications like medicine, it also can produce decision trees represented by hybrid models between CART and boosted stumps that can outperform either of these approaches.

Dominant hemisphere lateralization of cortical parasympathetic control as revealed by frontotemporal dementia.

The brain continuously influences and perceives the physiological condition of the body. Related cortical representations have been proposed to shape emotional experience and guide behavior. Although previous studies have identified brain regions recruited during autonomic processing, neurological lesion studies have yet to delineate the regions critical for maintaining autonomic outflow. Even greater controversy surrounds hemispheric lateralization along the parasympathetic-sympathetic axis. The behavioral variant of frontotemporal dementia (bvFTD), featuring progressive and often asymmetric degeneration that includes the frontoinsular and cingulate cortices, provides a unique lesion model for elucidating brain structures that control autonomic tone. Here, we show that bvFTD is associated with reduced baseline cardiac vagal tone and that this reduction correlates with left-lateralized functional and structural frontoinsular and cingulate cortex deficits and with reduced agreeableness. Our results suggest that networked brain regions in the dominant hemisphere are critical for maintaining an adaptive level of baseline parasympathetic outflow.

Age-Related Effects and Sex Differences in Gray Matter Density, Volume, Mass, and Cortical Thickness from Childhood to Young Adulthood.

Developmental structural neuroimaging studies in humans have long described decreases in gray matter volume (GMV) and cortical thickness (CT) during adolescence. Gray matter density (GMD), a measure often assumed to be highly related to volume, has not been systematically investigated in development. We used T1 imaging data collected on the Philadelphia Neurodevelopmental Cohort to study age-related effects and sex differences in four regional gray matter measures in 1189 youths ranging in age from 8 to 23 years. Custom T1 segmentation and a novel high-resolution gray matter parcellation were used to extract GMD, GMV, gray matter mass (GMM; defined as GMD × GMV), and CT from 1625 brain regions. Nonlinear models revealed that each modality exhibits unique age-related effects and sex differences. While GMV and CT generally decrease with age, GMD increases and shows the strongest age-related effects, while GMM shows a slight decline overall. Females have lower GMV but higher GMD than males throughout the brain. Our findings suggest that GMD is a prime phenotype for the assessment of brain development and likely cognition and that periadolescent gray matter loss may be less pronounced than previously thought. This work highlights the need for combined quantitative histological MRI studies.SIGNIFICANCE STATEMENT This study demonstrates that different MRI-derived gray matter measures show distinct age and sex effects and should not be considered equivalent but complementary. It is shown for the first time that gray matter density increases from childhood to young adulthood, in contrast with gray matter volume and cortical thickness, and that females, who are known to have lower gray matter volume than males, have higher density throughout the brain. A custom preprocessing pipeline and a novel high-resolution parcellation were created to analyze brain scans of 1189 youths collected as part of the Philadelphia Neurodevelopmental Cohort. A clear understanding of normal structural brain development is essential for the examination of brain-behavior relationships, the study of brain disease, and, ultimately, clinical applications of neuroimaging.

Quantitative assessment of structural image quality.

Data quality is increasingly recognized as one of the most important confounding factors in brain imaging research. It is particularly important for studies of brain development, where age is systematically related to in-scanner motion and data quality. Prior work has demonstrated that in-scanner head motion biases estimates of structural neuroimaging measures. However, objective measures of data quality are not available for most structural brain images. Here we sought to identify quantitative measures of data quality for T1-weighted volumes, describe how these measures relate to cortical thickness, and delineate how this in turn may bias inference regarding associations with age in youth. Three highly-trained raters provided manual ratings of 1840 raw T1-weighted volumes. These images included a training set of 1065 images from Philadelphia Neurodevelopmental Cohort (PNC), a test set of 533 images from the PNC, as well as an external test set of 242 adults acquired on a different scanner. Manual ratings were compared to automated quality measures provided by the Preprocessed Connectomes Project's Quality Assurance Protocol (QAP), as well as FreeSurfer's Euler number, which summarizes the topological complexity of the reconstructed cortical surface. Results revealed that the Euler number was consistently correlated with manual ratings across samples. Furthermore, the Euler number could be used to identify images scored "unusable" by human raters with a high degree of accuracy (AUC: 0.98-0.99), and out-performed proxy measures from functional timeseries acquired in the same scanning session. The Euler number also was significantly related to cortical thickness in a regionally heterogeneous pattern that was consistent across datasets and replicated prior results. Finally, data quality both inflated and obscured associations with age during adolescence. Taken together, these results indicate that reliable measures of data quality can be automatically derived from T1-weighted volumes, and that failing to control for data quality can systematically bias the results of studies of brain maturation.

Impact of puberty on the evolution of cerebral perfusion during adolescence.

Puberty is the defining biological process of adolescent development, yet its effects on fundamental properties of brain physiology such as cerebral blood flow (CBF) have never been investigated. Capitalizing on a sample of 922 youths ages 8-22 y imaged using arterial spin labeled MRI as part of the Philadelphia Neurodevelopmental Cohort, we studied normative developmental differences in cerebral perfusion in males and females, as well as specific associations between puberty and CBF. Males and females had conspicuously divergent nonlinear trajectories in CBF evolution with development as modeled by penalized splines. Seventeen brain regions, including hubs of the executive and default mode networks, showed a robust nonlinear age-by-sex interaction that surpassed Bonferroni correction. Notably, within these regions the decline in CBF was similar between males and females in early puberty and only diverged in midpuberty, with CBF actually increasing in females. Taken together, these results delineate sex-specific growth curves for CBF during youth and for the first time to our knowledge link such differential patterns of development to the effects of puberty.

The impact of quality assurance assessment on diffusion tensor imaging outcomes in a large-scale population-based cohort.

BACKGROUND: Diffusion tensor imaging (DTI) is applied in investigation of brain biomarkers for neurodevelopmental and neurodegenerative disorders. However, the quality of DTI measurements, like other neuroimaging techniques, is susceptible to several confounding factors (e.g., motion, eddy currents), which have only recently come under scrutiny. These confounds are especially relevant in adolescent samples where data quality may be compromised in ways that confound interpretation of maturation parameters. The current study aims to leverage DTI data from the Philadelphia Neurodevelopmental Cohort (PNC), a sample of 1601 youths with ages of 8-21 who underwent neuroimaging, to: 1) establish quality assurance (QA) metrics for the automatic identification of poor DTI image quality; 2) examine the performance of these QA measures in an external validation sample; 3) document the influence of data quality on developmental patterns of typical DTI metrics. METHODS: All diffusion-weighted images were acquired on the same scanner. Visual QA was performed on all subjects completing DTI; images were manually categorized as Poor, Good, or Excellent. Four image quality metrics were automatically computed and used to predict manual QA status: Mean voxel intensity outlier count (MEANVOX), Maximum voxel intensity outlier count (MAXVOX), mean relative motion (MOTION) and temporal signal-to-noise ratio (TSNR). Classification accuracy for each metric was calculated as the area under the receiver-operating characteristic curve (AUC). A threshold was generated for each measure that best differentiated visual QA status and applied in a validation sample. The effects of data quality on sensitivity to expected age effects in this developmental sample were then investigated using the traditional MRI diffusion metrics: fractional anisotropy (FA) and mean diffusivity (MD). Finally, our method of QA is compared with DTIPrep. RESULTS: TSNR (AUC=0.94) best differentiated Poor data from Good and Excellent data. MAXVOX (AUC=0.88) best differentiated Good from Excellent DTI data. At the optimal threshold, 88% of Poor data and 91% Good/Excellent data were correctly identified. Use of these thresholds on a validation dataset (n=374) indicated high accuracy. In the validation sample 83% of Poor data and 94% of Excellent data was identified using thresholds derived from the training sample. Both FA and MD were affected by the inclusion of poor data in an analysis of an age, sex and race matched comparison sample. In addition, we show that the inclusion of poor data results in significant attenuation of the correlation between diffusion metrics (FA and MD) and age during a critical neurodevelopmental period. We find higher correspondence between our QA method and DTIPrep for Poor data, but we find our method to be more robust for apparently high-quality images. CONCLUSION: Automated QA of DTI can facilitate large-scale, high-throughput quality assurance by reliably identifying both scanner and subject induced imaging artifacts. The results present a practical example of the confounding effects of artifacts on DTI analysis in a large population-based sample, and suggest that estimates of data quality should not only be reported but also accounted for in data analysis, especially in studies of development.

Linked Sex Differences in Cognition and Functional Connectivity in Youth.

Sex differences in human cognition are marked, but little is known regarding their neural origins. Here, in a sample of 674 human participants ages 9-22, we demonstrate that sex differences in cognitive profiles are related to multivariate patterns of resting-state functional connectivity MRI (rsfc-MRI). Males outperformed females on motor and spatial cognitive tasks; females were faster in tasks of emotion identification and nonverbal reasoning. Sex differences were also prominent in the rsfc-MRI data at multiple scales of analysis, with males displaying more between-module connectivity, while females demonstrated more within-module connectivity. Multivariate pattern analysis using support vector machines classified subject sex on the basis of their cognitive profile with 63% accuracy (P < 0.001), but was more accurate using functional connectivity data (71% accuracy; P < 0.001). Moreover, the degree to which a given participant's cognitive profile was "male" or "female" was significantly related to the masculinity or femininity of their pattern of brain connectivity (P = 2.3 × 10(-7)). This relationship was present even when considering males and female separately. Taken together, these results demonstrate for the first time that sex differences in patterns of cognition are in part represented on a neural level through divergent patterns of brain connectivity.

Functional maturation of the executive system during adolescence.

Adolescence is characterized by rapid development of executive function. Working memory (WM) is a key element of executive function, but it is not known what brain changes during adolescence allow improved WM performance. Using a fractal n-back fMRI paradigm, we investigated brain responses to WM load in 951 human youths aged 8-22 years. Compared with more limited associations with age, WM performance was robustly associated with both executive network activation and deactivation of the default mode network. Multivariate patterns of brain activation predicted task performance with a high degree of accuracy, and also mediated the observed age-related improvements in WM performance. These results delineate a process of functional maturation of the executive system, and suggest that this process allows for the improvement of cognitive capability seen during adolescence.

Intrinsic connectivity network disruption in progressive supranuclear palsy.

OBJECTIVE: Progressive supranuclear palsy (PSP) has been conceptualized as a large-scale network disruption, but the specific network targeted has not been fully characterized. We sought to delineate the affected network in patients with clinical PSP. METHODS: Using task-free functional magnetic resonance imaging, we mapped intrinsic connectivity to the dorsal midbrain tegmentum (dMT), a region that shows focal atrophy in PSP. Two healthy control groups (1 young, 1 older) were used to define and replicate the normal connectivity pattern, and patients with PSP were compared to an independent matched healthy control group on measures of network connectivity. RESULTS: Healthy young and older subjects showed a convergent pattern of connectivity to the dMT, including brainstem, cerebellar, diencephalic, basal ganglia, and cortical regions involved in skeletomotor, oculomotor, and executive control. Patients with PSP showed significant connectivity disruptions within this network, particularly within corticosubcortical and cortico-brainstem interactions. Patients with more severe functional impairment showed lower mean dMT network connectivity scores. INTERPRETATION: This study defines a PSP-related intrinsic connectivity network in the healthy brain and demonstrates the sensitivity of network-based imaging methods to PSP-related physiological and clinical changes.

Predicting the Effect of Proton Beam Therapy Technology on Pulmonary Toxicities for Patients With Locally Advanced Lung Cancer Enrolled in the Proton Collaborative Group Prospective Clinical Trial.

PURPOSE: This study aimed to predict the probability of grade ≥2 pneumonitis or dyspnea within 12 months of receiving conventionally fractionated or mildly hypofractionated proton beam therapy for locally advanced lung cancer using machine learning. METHODS AND MATERIALS: Demographic and treatment characteristics were analyzed for 965 consecutive patients treated for lung cancer with conventionally fractionated or mildly hypofractionated (2.2-3 Gy/fraction) proton beam therapy across 12 institutions. Three machine learning models (gradient boosting, additive tree, and logistic regression with lasso regularization) were implemented to predict Common Terminology Criteria for Adverse Events version 4 grade ≥2 pulmonary toxicities using double 10-fold cross-validation for parameter hyper-tuning without leak of information. Balanced accuracy and area under the curve were calculated, and 95% confidence intervals were obtained using bootstrap sampling. RESULTS: The median age of the patients was 70 years (range, 20-97), and they had predominantly stage IIIA or IIIB disease. They received a median dose of 60 Gy in 2 Gy/fraction, and 46.4% received concurrent chemotherapy. In total, 250 (25.9%) had grade ≥2 pulmonary toxicity. The probability of pulmonary toxicity was 0.08 for patients treated with pencil beam scanning and 0.34 for those treated with other techniques (P = 8.97e-13). Use of abdominal compression and breath hold were highly significant predictors of less toxicity (P = 2.88e-08). Higher total radiation delivered dose (P = .0182) and higher average dose to the ipsilateral lung (P = .0035) increased the likelihood of pulmonary toxicities. The gradient boosting model performed the best of the models tested, and when demographic and dosimetric features were combined, the area under the curve and balanced accuracy were 0.75 ± 0.02 and 0.67 ± 0.02, respectively. After analyzing performance versus the number of data points used for training, we observed that accuracy was limited by the number of observations. CONCLUSIONS: In the largest analysis of prospectively enrolled patients with lung cancer assessing pulmonary toxicities from proton therapy to date, advanced machine learning methods revealed that pencil beam scanning, abdominal compression, and lower normal lung doses can lead to significantly lower probability of developing grade ≥2 pneumonitis or dyspnea.

Heterogeneous impact of motion on fundamental patterns of developmental changes in functional connectivity during youth.

Several independent studies have demonstrated that small amounts of in-scanner motion systematically bias estimates of resting-state functional connectivity. This confound is of particular importance for studies of neurodevelopment in youth because motion is strongly related to subject age during this period. Critically, the effects of motion on connectivity mimic major findings in neurodevelopmental research, specifically an age-related strengthening of distant connections and weakening of short-range connections. Here, in a sample of 780 subjects ages 8-22, we re-evaluate patterns of change in functional connectivity during adolescent development after rigorously controlling for the confounding influences of motion at both the subject and group levels. We find that motion artifact inflates both overall estimates of age-related change as well as specific distance-related changes in connectivity. When motion is more fully accounted for, the prevalence of age-related change as well as the strength of distance-related effects is substantially reduced. However, age-related changes remain highly significant. In contrast, motion artifact tends to obscure age-related changes in connectivity associated with segregation of functional brain modules; improved preprocessing techniques allow greater sensitivity to detect increased within-module connectivity occurring with development. Finally, we show that subject's age can still be accurately estimated from the multivariate pattern of functional connectivity even while controlling for motion. Taken together, these results indicate that while motion artifact has a marked and heterogeneous impact on estimates of connectivity change during adolescence, functional connectivity remains a valuable phenotype for the study of neurodevelopment.

Network-level structural covariance in the developing brain.

Intrinsic or resting state functional connectivity MRI and structural covariance MRI have begun to reveal the adult human brain's multiple network architectures. How and when these networks emerge during development remains unclear, but understanding ontogeny could shed light on network function and dysfunction. In this study, we applied structural covariance MRI techniques to 300 children in four age categories (early childhood, 5-8 y; late childhood, 8.5-11 y; early adolescence, 12-14 y; late adolescence, 16-18 y) to characterize gray matter structural relationships between cortical nodes that make up large-scale functional networks. Network nodes identified from eight widely replicated functional intrinsic connectivity networks served as seed regions to map whole-brain structural covariance patterns in each age group. In general, structural covariance in the youngest age group was limited to seed and contralateral homologous regions. Networks derived using primary sensory and motor cortex seeds were already well-developed in early childhood but expanded in early adolescence before pruning to a more restricted topology resembling adult intrinsic connectivity network patterns. In contrast, language, social-emotional, and other cognitive networks were relatively undeveloped in younger age groups and showed increasingly distributed topology in older children. The so-called default-mode network provided a notable exception, following a developmental trajectory more similar to the primary sensorimotor systems. Relationships between functional maturation and structural covariance networks topology warrant future exploration.

Underinsurance And Multiple Surgical Treatments for Kidney Stones.

OBJECTIVE: To further elucidate the relationship between low socioeconomic status (SES) and larger, more complex stones requiring staged surgical interventions. Specifically, we aimed to determine if underinsurance (Medicaid, Medicare, and self-pay insurance types) is associated with multiple surgeries within 1 year. METHODS: We performed a retrospective longitudinal analysis of prospectively collected data from the California statewide Department of Health Care Access and Information (HCAI) dataset. We included adult patients who had their first recorded kidney stone encounter between 2009 and 2018 and underwent at least 1 urologic stone procedure. We followed these patients within the dataset for one year after their initial surgery to assess for factors predicting multiple surgical treatments for stones. RESULTS: A total of 156,319 adults were included in the study. The proportions of individuals in private insurance, Medicaid, Medicare and self-pay/indigent groups differed by the presence or absence of additional surgeries (64.0%, 13.5%, 19.4%, and 0.1%, vs 70.3%, 10.1%, 16.6%, and 0.1%, respectively). Compared to private insurance, Medicaid (1.46 [1.40-1.53] P < .001) and Medicare (1.15 [1.10-1.20] P < .001) insurance types were associated with significantly greater odds of multiple surgeries, whereas no significant association was seen in the self-pay/indigent insurance type (1.35 [0.83-2.19], P = 1.0). CONCLUSION: In a statewide, California database from 2009 to 2018, underinsured adults had higher odds of undergoing a second procedure for kidney stones within 1 year of initial surgical treatment. This study adds to the expanding body of literature linking suboptimal healthcare access and disparate outcomes for kidney stone patients.

Association of SARS-CoV-2 nucleocapsid viral antigen and the receptor for advanced glycation end products with development of severe disease in patients presenting to the emergency department with COVID-19.

Introduction: There remains a need to better identify patients at highest risk for developing severe Coronavirus Disease 2019 (COVID-19) as additional waves of the pandemic continue to impact hospital systems. We sought to characterize the association of receptor for advanced glycation end products (RAGE), SARS-CoV-2 nucleocapsid viral antigen, and a panel of thromboinflammatory biomarkers with development of severe disease in patients presenting to the emergency department with symptomatic COVID-19. Methods: Blood samples were collected on arrival from 77 patients with symptomatic COVID-19, and plasma levels of thromboinflammatory biomarkers were measured. Results: Differences in biomarkers between those who did and did not develop severe disease or death 7 days after presentation were analyzed. After adjustment for multiple comparisons, RAGE, SARS-CoV-2 nucleocapsid viral antigen, interleukin (IL)-6, IL-10 and tumor necrosis factor receptor (TNFR)-1 were significantly elevated in the group who developed severe disease (all p<0.05). In a multivariable regression model, RAGE and SARS-CoV-2 nucleocapsid viral antigen remained significant risk factors for development of severe disease (both p<0.05), and each had sensitivity and specificity >80% on cut-point analysis. Discussion: Elevated RAGE and SARS-CoV-2 nucleocapsid viral antigen on emergency department presentation are strongly associated with development of severe disease at 7 days. These findings are of clinical relevance for patient prognostication and triage as hospital systems continue to be overwhelmed. Further studies are warranted to determine the feasibility and utility of point-of care measurements of these biomarkers in the emergency department setting to improve patient prognostication and triage.

The Conditional Super Learner.

Using cross validation to select the best model from a library is standard practice in machine learning. Similarly, meta learning is a widely used technique where models previously developed are combined (mainly linearly) with the expectation of improving performance with respect to individual models. In this article we consider the Conditional Super Learner (CSL), an algorithm that selects the best model candidate from a library of models conditional on the covariates. The CSL expands the idea of using cross validation to select the best model and merges it with meta learning. We propose an optimization algorithm that finds a local minimum to the problem posed and proves that it converges at a rate faster than Op(n-1/4). We offer empirical evidence that: (1) CSL is an excellent candidate to substitute stacking and (2) CLS is suitable for the analysis of Hierarchical problems. Additionally, implications for global interpretability are emphasized.