Machine Learning as a Tool for Early Detection: A Focus on Late-Stage Colorectal Cancer across Socioeconomic Spectrums.

PURPOSE: To assess the efficacy of various machine learning (ML) algorithms in predicting late-stage colorectal cancer (CRC) diagnoses against the backdrop of socio-economic and regional healthcare disparities. METHODS: An innovative theoretical framework was developed to integrate individual- and census tract-level social determinants of health (SDOH) with sociodemographic factors. A comparative analysis of the ML models was conducted using key performance metrics such as AUC-ROC to evaluate their predictive accuracy. Spatio-temporal analysis was used to identify disparities in late-stage CRC diagnosis probabilities. RESULTS: Gradient boosting emerged as the superior model, with the top predictors for late-stage CRC diagnosis being anatomic site, year of diagnosis, age, proximity to superfund sites, and primary payer. Spatio-temporal clusters highlighted geographic areas with a statistically significant high probability of late-stage diagnoses, emphasizing the need for targeted healthcare interventions. CONCLUSIONS: This research underlines the potential of ML in enhancing the prognostic predictions in oncology, particularly in CRC. The gradient boosting model, with its robust performance, holds promise for deployment in healthcare systems to aid early detection and formulate localized cancer prevention strategies. The study's methodology demonstrates a significant step toward utilizing AI in public health to mitigate disparities and improve cancer care outcomes.

Development and Validation of a Clinical Frailty Index for the World Trade Center General Responder Cohort.

Objectives: To develop and validate a clinical frailty index to characterize aging among responders to the 9/11 World Trade Center (WTC) attacks. Methods: This study was conducted on health monitoring data on a sample of 6197 responders. A clinical frailty index, WTC FI-Clinical, was developed according to the cumulative deficit model of frailty. The validity of the resulting index was assessed using all-cause mortality as an endpoint. Its association with various cohort characteristics was evaluated. Results: The sample's median age was 51 years. Thirty items were selected for inclusion in the index. It showed a strong correlation with age, as well as significant adjusted associations with mortality, 9/11 exposure severity, sex, race, pre-9/11 occupation, education, and smoking status. Discussion: The WTC FI-Clinical highlights effects of certain risk factors on aging within the 9/11 responder cohort. It will serve as a useful instrument for monitoring and tracking frailty within this cohort.

Early life metal dysregulation in amyotrophic lateral sclerosis.

OBJECTIVE: Deficiencies and excess of essential elements and toxic metals are implicated in amyotrophic lateral sclerosis (ALS), but the age when metal dysregulation appears remains unknown. This study aims to determine whether metal uptake is dysregulated during childhood in individuals eventually diagnosed with ALS. METHODS: Laser ablation-inductively coupled plasma-mass spectrometry was used to obtain time series data of metal uptake using biomarkers in teeth from autopsies or dental extractions of ALS (n = 36) and control (n = 31) participants. Covariate data included sex, smoking, occupational exposures, and ALS family history. Case-control differences were identified in temporal profiles of metal uptake for individual metals using distributed lag models. Weighted quantile sum (WQS) regression was used for metals mixture analyses. Similar analyses were performed on an ALS mouse model to further verify the relevance of dysregulation of metals in ALS. RESULTS: Metal levels were higher in cases than in controls: 1.49 times for chromium (1.11-1.82; at 15 years), 1.82 times for manganese (1.34-2.46; at birth), 1.65 times for nickel (1.22-2.01; at 8 years), 2.46 times for tin (1.65-3.30; at 2 years), and 2.46 times for zinc (1.49-3.67; at 6 years). Co-exposure to 11 elements indicated that childhood metal dysregulation was associated with ALS. The mixture contribution of metals to disease outcome was likewise apparent in tooth biomarkers of an ALS mouse model, and differences in metal distribution were evident in ALS mouse brains compared to brains from littermate controls. INTERPRETATION: Overall, our study reveals direct evidence that altered metal uptake during specific early life time windows is associated with adult-onset ALS.

Lagged WQS regression for mixtures with many components.

The developmental timing of exposures to toxic chemicals or combinations of chemicals may be as important as the dosage itself. This concept is called "critical windows of exposure." The time boundaries of such windows can be detected if exposure data are collected repeatedly in short time intervals. The development of tooth-matrix biomarkers which provide prenatal and postnatal exposure measures in repeated intervals can provide such data. Using teeth, we use reverse distributed lagged models (DLMs) to incorporate weekly prenatal and postnatal measures of exposures to estimate time-varying associations with developmental effects. The analysis of such data using lagged weighted quantile sum (WQS) regression as an extension to reverse DLMs for complex mixtures was first proposed by Bello et al. This prior algorithm was not operationally generalizable to large numbers of components (say, more than five or six). We propose a revised algorithm that may be useful for larger mixtures by combining time-specific WQS(t) indices in a reverse DLM. We demonstrate the new algorithm using tooth data in association with a neurodevelopmental score and in simulated data from 3 cases wherein different components of a mixture have time varying associations and in the case where none have associations. The new algorithm correctly detects the simulated associations when the number of samples within the time-specific analyses is moderate to large.

Cell-bound complement activation products associate with lupus severity in SLE.

OBJECTIVES: To evaluate the association between lupus severity and cell-bound complement activation products (CB-CAPs) or low complement proteins C3 and C4. METHODS: All subjects (n=495) fulfilled the American College of Rheumatology (ACR) classification criteria for SLE. Abnormal CB-CAPs (erythrocyte-bound C4d or B-lymphocyte-bound C4d levels >99th percentile of healthy) and complement proteins C3 and C4 were determined using flow cytometry and turbidimetry, respectively. Lupus severity was estimated using the Lupus Severity Index (LSI). Statistical analysis consisted of multivariable linear regression and groups comparisons. RESULTS: Abnormal CB-CAPs were more prevalent than low complement values irrespective of LSI levels (62% vs 38%, respectively, p<0.0001). LSI was low (median 5.44, IQR: 4.77-6.93) in patients with no complement abnormality, intermediate in patients with abnormal CB-CAPs (median 6.09, IQR: 5.31-8.20) and high in the group presenting with both abnormal CB-CAPs and low C3 and/or C4 (median 7.85, IQR: 5.51-8.37). Odds of immunosuppressant use was higher in subjects with LSI ≥5.95 compared with subjects with LSI <5.95 (1.60 vs 0.53, p<0.0001 for both). Multivariable regression analysis revealed that higher LSI scores associated with abnormal CB-CAPs-but not low C3/C4-after adjusting for younger age, race and longer disease duration (p=0.0001), which were also independent predictors of disease severity (global R2=0.145). CONCLUSION: Abnormalities in complement activation as measured by CB-CAPs are associated with increased LSI.

Blood manganese levels during pregnancy and postpartum depression: A cohort study among women in Mexico.

BACKGROUND: Occupational studies have shown an association between elevated Mn exposure and depressive symptoms. Blood Mn (BMn) naturally rises during pregnancy due to mobilization from tissues, suggesting it could contribute to pregnancy and postpartum depressive symptoms. OBJECTIVES: To assess the association between BMn levels during pregnancy and postpartum depression (PPD), creating opportunities for possible future interventions. METHODS: We studied 561 women from the reproductive longitudinal Programming Research in Obesity, Growth, Environment, and Social Stressors (PROGRESS) cohort in Mexico City. BMn was measured at the 2nd and 3rd trimesters, as well as delivery. The Edinburgh Postnatal Depression Scale (EPDS) was used to assess PPD symptoms at 12-months postpartum. We used a generalized linear model assuming a Poisson distribution to assess the association between BMn levels and PPD, with adjustments for age, stress and depressive symptoms during pregnancy, education, socioeconomic status, and contemporaneous blood lead levels. RESULTS: The mean ±â€¯standard deviation (SD) EPDS score at 12-months postpartum was 6.51 ±â€¯5.65, and 17.11% of women met the criteria for possible PPD (score ≥ 13). In adjusted models, BMn during the 3rd trimester (ß: 0.13, 95% CI: 0.04-0.21) and BMn levels averaged at the 2nd and 3rd trimester (ß: 0.14, 95% CI: 0.02-0.26) had a positive association with EPDS scores at 12 months postpartum. BMn at the 2nd trimester (ß: 0.07, 95% CI: -0.09-0.22) and delivery (ß: 0.03, 95% CI: -0.04-0.10) had a non-significant positive association with EPDS scores at 12-months postpartum. Stress and depressive symptoms during pregnancy was associated with higher EPDS scores at 12-months postpartum in all of the adjusted models but were only significant when either BMn during 3rd trimester or BMn averaged across 2nd and 3rd trimester was assessed as the exposure. DISCUSSION: Our results demonstrate that elevated BMn levels during pregnancy predict PPD symptoms and could be a potential pathway for intervention and prevention of PPD.

Deep learning cardiac motion analysis for human survival prediction.

Motion analysis is used in computer vision to understand the behaviour of moving objects in sequences of images. Optimising the interpretation of dynamic biological systems requires accurate and precise motion tracking as well as efficient representations of high-dimensional motion trajectories so that these can be used for prediction tasks. Here we use image sequences of the heart, acquired using cardiac magnetic resonance imaging, to create time-resolved three-dimensional segmentations using a fully convolutional network trained on anatomical shape priors. This dense motion model formed the input to a supervised denoising autoencoder (4Dsurvival), which is a hybrid network consisting of an autoencoder that learns a task-specific latent code representation trained on observed outcome data, yielding a latent representation optimised for survival prediction. To handle right-censored survival outcomes, our network used a Cox partial likelihood loss function. In a study of 302 patients the predictive accuracy (quantified by Harrell's C-index) was significantly higher (p = .0012) for our model C=0.75 (95% CI: 0.70 - 0.79) than the human benchmark of C=0.59 (95% CI: 0.53 - 0.65). This work demonstrates how a complex computer vision task using high-dimensional medical image data can efficiently predict human survival.

Automatic 3D Bi-Ventricular Segmentation of Cardiac Images by a Shape-Refined Multi- Task Deep Learning Approach.

Deep learning approaches have achieved state-of-the-art performance in cardiac magnetic resonance (CMR) image segmentation. However, most approaches have focused on learning image intensity features for segmentation, whereas the incorporation of anatomical shape priors has received less attention. In this paper, we combine a multi-task deep learning approach with atlas propagation to develop a shape-refined bi-ventricular segmentation pipeline for short-axis CMR volumetric images. The pipeline first employs a fully convolutional network (FCN) that learns segmentation and landmark localization tasks simultaneously. The architecture of the proposed FCN uses a 2.5D representation, thus combining the computational advantage of 2D FCNs networks and the capability of addressing 3D spatial consistency without compromising segmentation accuracy. Moreover, a refinement step is designed to explicitly impose shape prior knowledge and improve segmentation quality. This step is effective for overcoming image artifacts (e.g., due to different breath-hold positions and large slice thickness), which preclude the creation of anatomically meaningful 3D cardiac shapes. The pipeline is fully automated, due to network's ability to infer landmarks, which are then used downstream in the pipeline to initialize atlas propagation. We validate the pipeline on 1831 healthy subjects and 649 subjects with pulmonary hypertension. Extensive numerical experiments on the two datasets demonstrate that our proposed method is robust and capable of producing accurate, high-resolution, and anatomically smooth bi-ventricular 3D models, despite the presence of artifacts in input CMR volumes.

Development of a Physiological Frailty Index for the World Trade Center General Responder Cohort.

Responders to the 9/11/2001 WTC attacks were exposed to multiple toxic pollutants. Since 2002, the health of the responder cohort has been continuously tracked by the WTC Health Monitoring Program. However, no assessments have been made of frailty, an important health metric given the current average age of the WTC responder cohort (55 years). In this study, we use laboratory test results and other physiological parameters to construct a physiological frailty index (FI-Lab) for this cohort. The study sample comprised responders aged 40 years or older who completed a health monitoring visit at Mount Sinai Center within the past 5 years. For each subject, FI-Lab was computed as the proportion of 20 physiological parameters (lab tests, pulmonary function, and blood pressure) on which the subject had abnormal values. Using negative binomial regression models, we tested FI-Lab's association with the SF-12 wellbeing score and various demographic characteristics. FI-Lab showed strong associations with the physical and mental components of the SF-12 as well as age, race, and smoking status. Using a cutoff of 0.25 to define presence of physiological/preclinical frailty, we found frailty prevalence in the study sample to be approximately 12%. This study demonstrates the feasibility of assessing preclinical frailty in the WTC responder cohort.

The adoption of generic immunosuppressant medications in kidney, liver, and heart transplantation among recipients in Colorado or nationally with Medicare part D.

The transplant community is divided regarding whether substitution with generic immunosuppressants is appropriate for organ transplant recipients. We estimated the rate of uptake over time of generic immunosuppressants using US Medicare Part D Prescription Drug Event (PDE) and Colorado pharmacy claims (including both Part D and non-Part D) data from 2008 to 2013. Data from 26 070 kidney, 15 548 liver, and 6685 heart recipients from Part D, and 1138 kidney and 389 liver recipients from Colorado were analyzed. The proportions of patients with PDEs or claims for generic and brand-name tacrolimus or mycophenolate mofetil were calculated over time by transplanted organ and drug. Among Part D kidney, liver, and heart beneficiaries, the proportion dispensed generic tacrolimus reached 50%-56% at 1 year after first generic approval and 78%-81% by December 2013. The proportion dispensed generic mycophenolate mofetil reached 70%-73% at 1 year after generic market entry and 88%-90% by December 2013. There was wide interstate variability in generic uptake, with faster uptake in Colorado compared with most other states. Overall, generic substitution for tacrolimus and mycophenolate mofetil for organ transplant recipients increased rapidly following first availability, and utilization of generic immunosuppressants exceeded that of brand-name products within a year of market entry.

Assessment of cumulative health risk in the World Trade Center general responder cohort.

BACKGROUND: Multiple comorbidities have been reported among rescue/recovery workers responding to the 9/11/2001 WTC disaster. In this study, we developed an index that quantifies the cumulative physiological burden of comorbidities and predicts life expectancy in this cohort. METHODS: A machine learning approach (gradient boosting) was used to model the relationship between mortality and several clinical parameters (laboratory test results, blood pressure, pulmonary function measures). This model was used to construct a risk index, which was validated by assessing its association with a number of health outcomes within the WTC general responder cohort. RESULTS: The risk index showed significant associations with mortality, self-assessed physical health, and onset of multiple chronic conditions, particularly COPD, hypertension, asthma, and sleep apnea. CONCLUSION: As an aggregate of several clinical parameters, this index serves as a cumulative measure of physiological dysregulation and could be utilized as a prognostic indicator of life expectancy and morbidity risk.

Association of 2,4-dichlorophenol urinary concentrations and olfactory dysfunction in a national sample of middle-aged and older U.S. adults.

Olfaction is a key sensory mechanism in humans. Deficits in this chemosensory function have wide-ranging impacts on overall health and quality of life. This study examines the role of environmental phenols as risk factors for olfactory dysfunction among a random sample of 839 middle-aged and older U.S. adults. Olfactory function assessment was carried out using a short 8-item test, scores on which were used to classify subjects into normal or impaired olfactory function groups. Logistic regression models were used to test for associations between olfactory impairment and creatinine-adjusted urinary levels of 8 common environmental phenols, adjusting for potentially confounding covariates. A statistically significant association between 2,4-dichlorophenol levels and olfactory impairment (OR = 1.02 [95 % CI: (1.003, 1.04)]; p = 0.02) was found. 2,4-dichlorophenol is a hazardous pollutant with widespread exposure via industrial and indoor air pollution, diet, and the use of pesticides and herbicides. This study is the first to reveal its role in olfactory impairment.

Optimal and lead-in adaptive allocation for binary outcomes: a comparison of Bayesian methodologies.

We compare posterior and predictive estimators and probabilities in response-adaptive randomization designs for two- and three-group clinical trials with binary outcomes. Adaptation based upon posterior estimates are discussed, as are two predictive probability algorithms: one using the traditional definition, the other using a skeptical distribution. Optimal and natural lead-in designs are covered. Simulation studies show: efficacy comparisons lead to more adaptation than center comparisons, though at some power loss; skeptically predictive efficacy comparisons and natural lead-in approaches lead to less adaptation but offer reduced allocation variability. Though nuanced, these results help clarify the power-adaptation trade-off in adaptive randomization.

New Research Strategy for Measuring Pre- and Postnatal Metal Dysregulation in Psychotic Disorders.

While previous studies have found evidence for detrimental effects of metals on neurodevelopment, the long-term effects on mental health remain unclear. The objective was to explore the effect of early metal exposure on risk of psychotic disorder and on symptom severity following illness onset. Through the use of validated tooth-biomarkers, we estimated pre- and postnatal exposure levels of essential elements (copper, magnesium, manganese, and zinc) and elements associated with neurotoxicity (lead, arsenic, lithium, and tin). We found consistently higher levels of lithium in patients compared to controls. Higher levels of magnesium and lower levels of zinc were associated with more severe psychopathology over 20 years after metal exposure. The results show promise for the use of teeth biomarkers in examining early environmental risk for psychosis and underscore the relevance of studying metal exposure during critical neurodevelopmental periods.

Current Developments in Machine Learning Techniques in Biological Data Mining.

This supplement is intended to focus on the use of machine learning techniques to generate meaningful information on biological data. This supplement under Bioinformatics and Biology Insights aims to provide scientists and researchers working in this rapid and evolving field with online, open-access articles authored by leading international experts in this field. Advances in the field of biology have generated massive opportunities to allow the implementation of modern computational and statistical techniques. Machine learning methods in particular, a subfield of computer science, have evolved as an indispensable tool applied to a wide spectrum of bioinformatics applications. Thus, it is broadly used to investigate the underlying mechanisms leading to a specific disease, as well as the biomarker discovery process. With a growth in this specific area of science comes the need to access up-to-date, high-quality scholarly articles that will leverage the knowledge of scientists and researchers in the various applications of machine learning techniques in mining biological data.

Extending the Distributed Lag Model framework to handle chemical mixtures.

Distributed Lag Models (DLMs) are used in environmental health studies to analyze the time-delayed effect of an exposure on an outcome of interest. Given the increasing need for analytical tools for evaluation of the effects of exposure to multi-pollutant mixtures, this study attempts to extend the classical DLM framework to accommodate and evaluate multiple longitudinally observed exposures. We introduce 2 techniques for quantifying the time-varying mixture effect of multiple exposures on an outcome of interest. Lagged WQS, the first technique, is based on Weighted Quantile Sum (WQS) regression, a penalized regression method that estimates mixture effects using a weighted index. We also introduce Tree-based DLMs, a nonparametric alternative for assessment of lagged mixture effects. This technique is based on the Random Forest (RF) algorithm, a nonparametric, tree-based estimation technique that has shown excellent performance in a wide variety of domains. In a simulation study, we tested the feasibility of these techniques and evaluated their performance in comparison to standard methodology. Both methods exhibited relatively robust performance, accurately capturing pre-defined non-linear functional relationships in different simulation settings. Further, we applied these techniques to data on perinatal exposure to environmental metal toxicants, with the goal of evaluating the effects of exposure on neurodevelopment. Our methods identified critical neurodevelopmental windows showing significant sensitivity to metal mixtures.

A Random Forest-Induced Distance-Based Measure of Physiologic al Dysregulation.

BACKGROUND: Aging involves gradual, multisystemic Physiological Dysregulation (PD) which increases risk of age-related comorbidities. Ability to quantify age-related PD could provide insights into biological mechanisms underlying the aging process. One approach to measuring PD exploits the fact that increasing PD manifests as a gradual deviation of physiological parameters away from normal levels. A recent geometric approach for quantifying PD uses Mahalanobis distance to measure the extent to which an individual's physiological parameters (measured via biomarkers from clinical blood biochemistry panels) deviate from normal levels. While useful, this approach has shortcomings that may impact its accuracy, primarily the incorrect assumption of multivariate normality among biomarkers, and identical weighting of biomarkers. Herein, we develop a more robust multivariate distance-based measure of PD. METHOD: Proximity matrices induced by survival tree ensembles (Random Survival Forests) were used to compute a robust distance metric for quantifying how abnormal an individual's biomarker profile is. This approach requires no distributional assumptions and allows differential weighting of biomarkers based on association with mortality. Using receiver operating characteristic analysis and model fit statistics we compared performance of our measure to the standard approach based on Mahalanobis distance. RESULTS & CONCLUSION: Our new metric showed statistically significant improvements in predicting mortality, health status and biological age, compared to the standard approach. Additional advantages offered by our method are the ability to handle missing values in biomarkers and to accommodate categorical risk factors. These results suggest our approach could provide greater precision in the evaluation of PD, which could enable better characterization of the extent and impact of degenerative processes resulting from aging.

Development and validation of a simple lupus severity index using ACR criteria for classification of SLE.

OBJECTIVE: To develop a simple systemic lupus erythematosus (SLE) severity index that requires knowledge of only American College of Rheumatology (ACR) criteria and subcriteria. METHODS: This study used demographic, mortality and medical records data of 1915 patients with lupus from the Lupus Family Registry and Repository. The data were randomly split (2:1 ratio) into independent training and validation sets. A logistic regression with ridge penalty was used to model the probability of being prescribed major immunosuppressive drugs-a surrogate indicator of lupus severity. ACR criteria and subcriteria were used as predictor variables in this model, and the resulting regression coefficient estimates obtained from the training data were used as item weightings to construct the severity index. RESULTS: The resulting index was tested on the independent validation dataset and was found to have high predictive accuracy for immunosuppressive use and early mortality. The index was also found to be strongly correlated with a previously existing severity score for lupus. In addition, demographic factors known to influence lupus severity (eg, age of onset, gender and ethnicity) all showed robust associations with our severity index that were consistent with observed clinical trends. CONCLUSIONS: This new index can be easily computed using ACR criteria, which may be among the most readily available data elements from patient medical records. This tool may be useful in lupus research, especially large dataset analyses to stratify patients by disease severity, an important prognostic indicator in SLE.

Development and Application of a Genetic Algorithm for Variable Optimization and Predictive Modeling of Five-Year Mortality Using Questionnaire Data.

The problem of selecting important variables for predictive modeling of a specific outcome of interest using questionnaire data has rarely been addressed in clinical settings. In this study, we implemented a genetic algorithm (GA) technique to select optimal variables from questionnaire data for predicting a five-year mortality. We examined 123 questions (variables) answered by 5,444 individuals in the National Health and Nutrition Examination Survey. The GA iterations selected the top 24 variables, including questions related to stroke, emphysema, and general health problems requiring the use of special equipment, for use in predictive modeling by various parametric and nonparametric machine learning techniques. Using these top 24 variables, gradient boosting yielded the nominally highest performance (area under curve [AUC] = 0.7654), although there were other techniques with lower but not significantly different AUC. This study shows how GA in conjunction with various machine learning techniques could be used to examine questionnaire data to predict a binary outcome.