SHapley Additive exPlanations Values Research Articles

Data-Driven Insights Into Post-Earthquake Reconnaissance Findings: 2023 Türkiye Earthquake Sequence

Post-earthquake reconnaissance stands as a vital endeavor, encompassing a systematic survey and data collection to gain insights into the aftermath of seismic events. Grappling with a multitude of multidimensional data poses inherent challenges in the proper interpretation and extraction of hidden information. This paper unfolds in two parts. Initially, we delineate the reconnaissance mission following the February 06 Türkiye earthquake sequence, elucidating the challenges, complexities, and nature of the amassed data. Subsequently, a paradigm shift toward automated machine learning (AutoML) is embraced to ascertain the optimal model, categorize observed damages, and unveil underlying patterns within the collected data. The most accurate machine learning model is then coupled with Shapley Additive Explanations (SHAP) to explicate observations, steering away from a black-box model. SHAP values discern and prioritize factors significantly contributing to various damage levels. The results reveal a test set accuracy of 0.75 and 0.95 for multi- and binary-class problems, respectively, employing both raw and composite features. Building-related features exert more control over light damage, while earthquake-related factors dominate severe damage. In critical damage scenarios, duration- and velocity-dependent intensity measures assume significance. Furthermore, findings indicate that if column and wall indices are below 0.07 and 0.08, respectively, they positively contribute to the likelihood of critical structural damage.

Explaining deep learning models for age-related gait classification based on acceleration time series

BackgroundGait analysis holds significant importance in monitoring daily health, particularly among older adults. Advancements in sensor technology enable the capture of movement in real-life environments and generate big data. Machine learning, notably deep learning (DL), shows promise to use these big data in gait analysis. However, the inherent black-box nature of these models poses challenges for their clinical application. This study aims to enhance transparency in DL-based gait classification for aged-related gait patterns using Explainable Artificial Intelligence, such as SHapley Additive exPlanations (SHAP). MethodsIn this cross-sectional study, a total of 244 participants, comprising 129 adults and 115 older adults (age>65), were included. They performed a 3-min walking task while accelerometers were affixed to the lumbar segment L3. DL models, convolutional neural network (CNN) and gated recurrent unit (GRU), were trained using 1-stride and 8-stride accelerations, respectively, to classify adult and older adult groups. SHAP was employed to explain the models' predictions. ResultsCNN achieved a satisfactory performance with an accuracy of 81.4 % and an AUC of 0.89, and GRU demonstrated promising results with an accuracy of 84.5 % and an AUC of 0.94. SHAP analysis revealed that both CNN and GRU assigned higher SHAP values to the data from vertical and walking directions, particularly emphasizing data around heel contact, spanning from the terminal swing to loading response phases. Furthermore, SHAP values indicated that GRU did not treat every stride equally. ConclusionCNN accurately distinguished between adults and older adults based on the characteristics of a single stride's data. GRU achieved accurate classification by considering the relationships and subtle differences between strides. In both models, data around heel contact emerged as most critical, suggesting differences in acceleration and deceleration patterns during walking between different age groups.

Can neotectonic faults influence soil air radon levels in the Upper Rhine Graben? An exploratory machine learning assessment

Radon (Rn) is a naturally occurring radioactive gas that poses a significant lung cancer risk. Subsurface fault zones can act as pathways for fluid and gas migration, potentially amplifying Rn accumulation. This study investigates the impact of fault zones on Rn concentrations within a 25 km2 area in the Northern Upper Rhine Graben, Germany — a region with available detailed geophysical exploration data and active neotectonic faulting. We conducted 597 Rn soil air measurements along precisely located fault zones, integrating a comprehensive range of environmental parameters. Utilizing the advanced machine learning model eXtreme Gradient Boosting (XGBoost) in conjunction with SHapley Additive exPlanations (SHAP) values, we dissected the influence of soil types, environmental factors, and proximity to fault zones on soil air Rn concentrations at a 1-meter depth. Our results reveal that clay-rich soils and cumulative 30-day precipitation are the primary drivers of elevated Rn levels. Proximity to fault zones also significantly influences Rn concentrations, though its impact is less pronounced than the factors mentioned above. Additionally, environmental factors such as wind speed, air pressure, and temperature exhibited lesser effects on Rn levels. The negligible influence of measuring devices and operating personnel increases confidence in data integrity in extensive environmental studies. This study demonstrates the effectiveness of integrating XGBoost with SHAP values to identify and quantify key factors influencing Rn concentrations. By providing a robust framework for enhancing Rn prediction models through machine learning, our findings contribute to improved risk assessments and mitigation strategies, thereby advancing public health and environmental management.

Explainable and Reduced-Feature Machine learning models for shape and drag prediction of a freely moving drop in the sub-critical Weber number regime

Accurately predicting the shape and drag of a moving drop is crucial in many spray applications. However, due to the complex interaction between the drag force and drop shape deformation, accurate prediction of drop shape and drag from Computational Fluid Dynamics (CFD) simulation requires large computational resources and time. A novel data-driven approach using NARXNN (Non-linear Auto Regressive eXogenous input Neural Network) is proposed in this study, which recurrently predicts the drop shape and drag for a given Weber number (We) and Reynolds number (Re), in the sub-critical We regime where there is no drop breakup. The average error in radius and drag coefficient for the test cases, as low as 0.36% and 0.49%, respectively, is achieved. Most importantly, SHAP (SHapley Additive exPlanations) analysis is performed to identify important features for the predictions, hence enabling the explainability of the physics involved in the predictions of machine learning (ML) models for drop deformation and the interaction between the drop and gas flows. Global interpretation of SHAP values identifies We and Re as the most important features to predict the drop shape and drag. Based on the SHAP analysis, several reduced-feature models are developed. All the reduced-feature models can predict drop shape and drag coefficient within 4% average error in radius and drag coefficient for any We and Re combinations within the prescribed parameter space. The ML models developed in this study demonstrate a tremendous computational advantage over the traditional CFD simulations. The drop shape and drag coefficient evolution can be predicted within seconds for a single case, whereas the CFD simulation takes several days to a week on a single processor.

A Predictive Model for Perinatal Brain Injury Using Machine Learning Based on Early Birth Data

Background/Objective: It is difficult to predict perinatal brain injury, and performing brain magnetic resonance imaging (MRI) based on suspected injury remains a clinical challenge. Therefore, we aimed to develop a reliable method for predicting perinatal brain injury using a machine learning model with early birth data. Methods: Neonates admitted to our institution from January 2017 to June 2024 with a gestational age of ≥36 weeks, a birth weight of ≥1800 g, admission within 6 h of birth, and who underwent brain MRI to confirm perinatal brain injury were included. Various machine learning models, including gradient boosting, were trained using early birth data to predict perinatal brain injury. Synthetic minority over-sampling and adaptive synthetic sampling (ADASYN) were applied to address class imbalance. Model performance was evaluated using accuracy, F1 score, and ROC curves. Feature importance scores and Shapley additive explanations (SHAP) values were also calculated. Results: Among 179 neonates, 39 had perinatal brain injury. There were significant differences between the injury and non-injury groups in mode of delivery, Apgar scores, capillary pH, lactate dehydrogenase (LDH) levels, and whether therapeutic hypothermia was performed. The gradient boosting model with the ADASYN method achieved the best performance. In terms of feature importance scores, the 1 min Apgar score was the most influential predictor. Additionally, SHAP analysis showed that LDH levels had the highest SHAP values. Conclusion: the gradient boosting model with ADASYN oversampling effectively predicts perinatal brain injury, potentially improving early detection for predicting long-term outcomes, reducing unnecessary MRI scans, and lowering healthcare costs.

Prediction of outcome in malignant hypertension patients using machine learning: a report from the West Birmingham malignant hypertension registry

Abstract Objectives Malignant hypertension (MHT) is a rare, yet severe condition characterized by high morbidity and mortality rates. This study aimed to assess the potential of machine learning (ML) algorithms in forecasting prognostic outcomes in MHT patients. Methods The dataset from the West Birmingham MHT Registry was used. We evaluated the efficacy of 9 ML algorithms, CatBoost, Decision Tree (DT), Light-Gradient Boosting Machine (LightGBM), K-Nearest Neighbors (KNN), Logistic Regression (LR), Multi-Layer Perceptron (MLP), Random Forest (RF), Support Vector Machine (SVM) and XGBoost, in predicting a composite outcome of all-cause mortality or dialysis. Evaluation metrics included the area under the receiver operating characteristic curve (AUC) and the F1 score. The SHapley Additive exPlanations values were employed to quantify the importance of each feature in the models. Results The study cohort comprised 385 individuals diagnosed with MHT (mean age of 48±13 years, 66% male). During a median follow-up period of 11 years (interquartile range: 3-18 years), 282 patients (73%) experienced the composite outcome. From 24 demographic and clinical variables, 16 were selected into the ML models. Among 9 models, the SVM and LR models exhibited robust predictive performance, achieving AUCs of 0.86 (95% CI: 0.76-0.93) and 0.81 (95% CI: 0.72-0.93), respectively. Furthermore, these models demonstrated high F1 scores (SVM: 0.89, LR: 0.85) and sensitivity (SVM: 0.89, LR: 0.83) (Table). Age, smoking history, systolic blood pressure at follow-up and baseline creatinine levels were commonly identified as primary prognostic features in both SVM and LR models (Figure). Conclusions The application of ML algorithms facilitates effective prediction of prognostic outcomes in patients with MHT, illustrating their potential utility in clinical decision-making through more targeted risk stratification and individualized patient care in these high-risk patients.

Predicting atrial fibrillation within 7 years follow-up using machine learning models

Abstract Background The prevalence of atrial fibrillation (AF), the most common type of cardiac arrhythmia, has increased globally. AF is also the leading cardiac factor for several cardiovascular and cerebrovascular events. Early prediction and treatment of AF prevents major cardiovascular and cerebrovascular events. Purpose Our objective is to predict AF risk within 7-years follow-up using only twelve basic factors and applying machine learning (ML) models. Methods Data from the German epidemiological trial on ankle brachial index (getABI) were utilized in this work. GetABI includes 6,457 patients with mean age of 72.53 years. 3,985 individuals were assessed for AF of which 393 patients suffered from AF at 7-years follow-up. The feature selection was initially done manually, ending up in 45 features. The 10 best features were extracted from this dataset according to the SHAP (SHapley Additive exPlanations) values. The final features were body mass index, sensitive CRP, cholesterol, LDL-cholesterol, troponin I, NT-pro-BNP, glomerular filtration rate, ratio of triglycerides and HDL-cholesterol, gamma-GT, heart rate, including age and sex as major factors. The handling of missing values was achieved by SimpleImputer, estimating the mean values. XGBoost and Random Forest predicted AF. Logistic Regression was also used as a baseline model. The data were split into training and test set and ML models were applied for 100 iterative runs, extracting results with mean values. Results XGBoost was the most accurate model, predicting the AF with 71.19 % mean accuracy, 66.37 % mean sensitivity and 75.61 % mean specificity. Figure 1 depicts the performance of all models. Additionally, Receiver Operating Characteristic Curve (ROC) and Area Under Curve (AUC) for each model are presented in Figure 2. XGBoost also outperformed all the others models, achieving the highest mean AUC value, equal to 70.99 %. Contrariwise, logistic regression had the lowest overall performance. Conclusion ML models, specifically XGBoost, achieved accurate AF prediction in 3,985 individuals utilizing only twelve clinical-routine data.Evaluation results of each used model.ROC curve and AUC value of each model.

ASCVD risk scores versus a novel cancer-specific Machine Learning-based calculator among patients with breast, colorectal, lung or prostate cancer

Abstract Background Patients with cancer have up to a 3-fold higher risk for cardiovascular disease (CVD) than the general population. Consequently, traditional risk scores developed to predict CVD in the general population (i.e., Pooled Cohort Equations (PCE), Predicting Risk of cardiovascular disease EVENTs [PREVENT]) may provide less accurate risk prediction for this population. Purpose Compare cancer-specific Machine Learning (ML)-based score to PCE and PREVENT to predict 10-year CVD risk in patients with breast cancer (BC), colorectal cancer (CRC), lung cancer (LC), or prostate cancer (PC). Methods Patients aged ≥18 years, diagnosed with BC, CRC, LC, or PC between 2005-2012 at a large hybrid academic-community practice in Northeast Ohio, United States, were included. An ML Extreme Gradient Boosting (XGBoost) algorithm, adapted for survival modeling, was developed using a training subset of the cohort (55% train + 20% test) and ranked 40 to 50 covariates (including social determinants of health, cancer treatment) for predicting 10-year CVD risk for each cancer type using SHAP (SHapley Additive exPlanations) values. The top 10 ML predictors were utilized to create a predictive equation using logistic regression models. This equation was then tested in the cohort validation subset (25%) and subsequently compared to the PCE and PREVENT (both simple and enhanced versions) using the area under the curve (AUC) of the time-dependent receiver operating characteristic curve. The CVD events were identified using ICD codes corresponding to Acute Coronary Syndrome, Ischemic Stroke, and Heart Failure. CVD risk was assessed through Kaplan-Meier. Results We included 10,240 patients (Table 1). The actual 10-year CVD events were: BC 21% (95% CI 19-23), CRC 10% (95% CI 8-12), LC 27% (95% CI 24-30), and PC 20% (95% CI 17-23). The ACC/AHA PCE predicted mean 10-year CVD risks of 16% (95% CI 15-18) for BC, 19% (95% CI 17-21) for CRC, 28% (95% CI 25-32) for LC, and 19% (95% CI 19-20) for PC, with corresponding AUCs of 0.75, 0.65, 0.76, and 0.61, respectively. The PREVENT simple version predicted mean 10-year CVD risks of 13% (95% CI 13-14) for BC, 13% (95% CI 12-13) for CRC, 15% (95% CI 14-15) for LC, and 16% (95% CI 12-19) for PC, with AUCs of 0.55, 0.55, 0.55, and 0.53, respectively. The PREVENT enhanced equation predicted mean 10-year CVD risks of 17% (95% CI 17-18) for BC, 15% (95% CI 14-15) for CRC, 15% (95% CI 15-16) for LC, and 17% (95% 16-19) for PC, with AUCs of 0.75, 0.62, 0.79, and 0.63, respectively. The predictive equation derived from the top 10 ML predictors (Table 2) achieved AUCs of 0.84 for BC, 0.76 for CRC, 0.83 for LC, and 0.71 for PC. Conclusion(s) Conventional CVD equations inadequately assess the risk in patients with BC, CRC, LC, or PC, often resulting in underestimation or overestimation. Cancer-specific ML-derived equations show good performance and emphasize the importance of integrating cancer-related covariates for precise prediction.Table 1Table 2

Tailoring Hematite Photoanodes for Improved PEC Performance: The Role of Alcohol Species Revealed by SHAP Analysis.

We explore the synergistic effects of TiO2 underlayers and varied alcohol species in the precursor solutions on the photoelectrochemical (PEC) performance of hematite photoanodes. Utilizing a robust machine learning (ML) framework combined with comprehensive analytical data sets, we systematically investigate how these modifications influence key physical and chemical properties, directly impacting the efficiency of water splitting processes. Our approach employs an ML model that integrates SHapley Additive exPlanations (SHAP) to quantitatively assess the impact of each dominant descriptor selected in the analytical data on the PEC performance, and they were combined with the SHAP values' dependence on the experimental operations. Specifically, we focus on the type of alcohol (methanol, ethanol, butanol, and 2-ethyl-1-butanol) used in the precursor solutions as the experimental operation, examining their effects on the dominant descriptors selected in the analytical data. The results from the SHAP analysis reveal that different alcohol species significantly alter the physicochemical properties at the hematite/TiO2 interface and in bulk hematite. These changes are primarily manifested in the modulation of the density of states and resistance to promote the charge carrier transport. For example, ethanol and butanol were found to enhance the electron density of states at the interface, which correlates with higher photocurrent outputs and improved PEC activity. In contrast, methanol showed a less pronounced effect, suggesting a nuanced interaction between the alcohol molecular structure and hematite surface chemistry. These findings not only underscore the importance of tailored precursor solution chemistry for enhancing PEC performance but also highlight the power of ML tools in uncovering the underlying physical and chemical mechanisms that govern the behavior of complex material systems. This study sets a foundational approach where ML can bridge the gap between empirical observations and theoretical understanding, leading to the rational design of energy materials.

Disease progression associated cytokines in COVID-19 patients with deteriorating and recovering health conditions.

Understanding the immune response to COVID-19 is challenging due to its high variability among individuals. To identify differentially expressed cytokines between the deteriorating and recovering phases, we analyzed the Electronic Health Records (EHR) and cytokine profile data in a COVID-19 cohort of 444 infected patients and 145 non-infected healthy individuals. We categorized each patient's progression into Deterioration Phase (DP) and Recovery Phase (RP) using longitudinal neutrophil, lymphocyte and lactate dehydrogenase levels. A random forest model was built using healthy and severe patients to compute the contribution of each cytokine toward disease progression using Shapley Additive Explanations (SHAP). SHAP values were used for supervised clustering to identify DP and RP-related samples and their associated cytokines. The identified clusters effectively discriminated DP and RP samples, suggesting that the cytokine profiles differed between deteriorating and recovering health conditions. Especially, CXCL10, GDF15, PTX3, and TNFSF10 were differentially expressed between the DP and RP samples, which are involved in the JAK-STAT, NF- B, and MAPK signaling pathways contributing to the inflammatory response. Collectively, we characterized the immune response in terms of disease progression of COVID-19 with deteriorating and recovering health conditions.

Predicting Unplanned Trauma ICU Admissions for Initial Nonoperative, Non-ICU Patients.

Unplanned intensive care unit (ICU) admissions are associated with increased morbidity and mortality. This study uses interpretable machine learning to predict unplanned ICU admissions for initial nonoperative trauma patients admitted to non-ICU locations. TQIP (2020-2021) was queried for initial nonoperative adult patients admitted to non-ICU locations. Univariable analysis compared patients who required an unplanned ICU admission to those who did not. Using variables that could be known at hospital admission, gradient boosting machines (CatBoost, LightGBM, XGBoost) were trained on 2021 data and tested on 2020 data. SHapley Additive exPlanations (SHAP) were used for interpretation. The cohort had 1,107,822 patients; 1.6% had an unplanned ICU admission. Unplanned ICU admissions were older (71 [58-80] vs. 61 [39-76] years, p < 0.01), had a higher Injury Severity Score (ISS) (9 [8-13] vs. 9 [4-10], p < 0.01), longer length of stay (11 [7-17] vs. 4 [3-6] days, p < 0.01), higher rates of all complications and most comorbidities and injuries (p < 0.05). All models had an AUC of 0.78 and an F1 score of 0.12, indicating poor performance in predicting the minority class. Mean absolute SHAP values revealed ISS (0.46), age (0.29), and absence of comorbidities (0.16) as most influential in predictions. Dependency plots showed greater SHAP values for greater ISS, age, and presence of comorbidities. Machine learning may outperform prior attempts at predicting the risk of unplanned ICU admissions in trauma patients while identifying unique predictors. Despite this progress, further research is needed to improve predictive performance by addressing class imbalance limitations.

Relationship between microbiota awareness, nutrition literacy, and health literacy among adolescents.

Microbiota awareness, nutritional literacy, and health literacy levels in adolescents have a significant impact on their health and well-being. This research was conducted to examine the relationship between microbiota awareness, nutrition literacy, and health literacy in adolescents. This research was structured with a descriptive-correlational design. The study population comprised adolescents aged 10-19 years, living in Türkiye (n = 739), between June 2022 and February 2024. Data were analyzed using SPSS 22.0, G*Power 3.1, and R programming language 4.1.3. The total effect of the health literacy variable on nutritional literacy was 0.2311, and this was statistically significant at a 95% confidence interval (CI) (p < 0.05). In terms of the health literacy variable, the direct effect of the nutrition literacy variable on the microbiota awareness variable was 0.2888, and this was statistically significant at the 95% CI (p < 0.05). In terms of the nutritional literacy variable, the direct effect of the health literacy variable on the microbiota awareness variable was 0.1707, and this was statistically significant at the 95% CI (p < 0.05). Nutrition literacy had a partial mediating role in the effect of health literacy on microbiota awareness (lower limit CI: 0.045; upper limit CI: 0.0894). The most accurate prediction of machine learning approaches to predict microbiota awareness was made with random forest with shapley additive explanations values, and the most important variable that should be in the model to predict the microbiota awareness variable was the nutrition literacy variable. Microbiota awareness increased as health literacy and nutrition literacy increased. In the machine learning approach prediction, the most important variables affecting microbiota awareness were health literacy and nutritional literacy. Longitudinal studies on microbiota awareness are recommended.

Association between sensitivity to thyroid hormones and non-high-density lipoprotein cholesterol levels in patients with type 2 diabetes mellitus

BACKGROUND Dyslipidemia and type 2 diabetes mellitus (T2DM) are chronic conditions with substantial public health implications. Effective management of lipid metabolism in patients with T2DM is critical. However, there has been insufficient attention given to the relationship between thyroid hormone sensitivity and dyslipidemia in the T2DM population, particularly concerning non-high-density lipoprotein cholesterol (non-HDL-C). AIM To clarify the association between thyroid hormone sensitivity and dyslipidemia in patients with T2DM. METHODS In this cross-sectional study, thyroid hormone sensitivity indices, the thyroid feedback quantile-based index (TFQI), the thyroid-stimulating hormone index (TSHI), the thyrotrophic T4 resistance index (TT4RI), and the free triiodothyronine (FT3)/free thyroxine (FT4) ratio were calculated. Logistic regression analysis was performed to determine the associations between those composite indices and non-HDL-C levels. Random forest variable importance and Shapley Additive Explanations (SHAP) summary plots were used to identify the strength and direction of the association between hyper-non-HDL-C and its major predictor. RESULTS Among the 994 participants, 389 (39.13%) had high non-HDL-C levels. Logistic regression analysis revealed that the risk of hyper-non-HDL-C was positively correlated with the TFQI (OR: 1.584; 95%CI: 1.088-2.304; P = 0.016), TSHI (OR: 1.238; 95%CI: 1.034-1.482; P = 0.02), and TT4RI (OR: 1.075; 95%CI: 1.006-1.149; P = 0.032) but was not significantly correlated with the FT3/FT4 ratio. The relationships between composite indices of the thyroid system and non-HDL-C levels differed according to sex. An increased risk of hyper-non-HDL-C was associated with elevated TSHI levels in men (OR: 1.331; 95%CI: 1.003-1.766; P = 0.048) but elevated TFQI levels in women (OR: 2.337; 95%CI: 1.4-3.901; P = 0.001). Among the analyzed variables, the average SHAP values were highest for TSHI, followed by TT4RI. CONCLUSION Impaired sensitivity to thyroid hormones was associated with high non-HDL-C levels in patients with T2DM.

Iron Deficiency: Global Trends and Projections from 1990 to 2050.

Iron deficiency (ID) remains the leading cause of anemia, affects a vast number of persons globally, and continues to be a significant global health burden. Comprehending the patterns of ID burden is essential for developing targeted public health policies. Using data from the Global Burden of Disease (GBD) 2021 study for the years 1990-2021, the XGBoost model was constructed to predict prevalence and disability-adjusted life years (DALYs) for the period 2022-2050, based on key demographic variables. Shapley Additive exPlanations (SHAP) values were applied to interpret the contributions of each variable to the model's predictions. Additionally, the Age-Period-Cohort (APC) model was used to evaluate the effects of age, period, and birth cohort on both prevalence and DALYs. The relationship between the Socio-Demographic Index (SDI) and ID's age-standardized prevalence rate (ASPR) as well as the age-standardized DALYs rate (ASDR) was also analyzed to assess the influence of socioeconomic development on disease burden. The global prevalent cases of ID grew from 984.61 million in 1990 to 1270.64 million in 2021 and are projected to reach 1439.99 million by 2050. Similarly, global DALYs from ID increased from 28.41 million in 1990 to 32.32 million in 2021, with a projected rise to 36.13 million by 2050. The ASPR declined from 18,204/100,000 in 1990 to 16,433/100,000 in 2021, with an estimated annual percentage change (EAPC) of -0.36% over this period. It is expected to decrease further to 15,922 by 2050, with an EAPC of -0.09% between 2021 and 2050. The ASDR was 518/100,000 in 1990 and 424/100,000 in 2021, with an EAPC of -0.68% from 1990 to 2021. It is expected to remain relatively stable at 419/100,000 by 2050, with an EAPC of -0.02% between 2021 and 2050. In 2021, the highest ASPRs were recorded in Senegal (34,421/100,000), Mali (34,233/100,000), and Pakistan (33,942/100,000). By 2050, Mali (35,070/100,000), Senegal (34,132/100,000), and Zambia (33,149/100,000) are projected to lead. For ASDR, Yemen (1405/100,000), Mozambique (1149/100,000), and Mali (1093/100,000) had the highest rates in 2021. By 2050, Yemen (1388/100,000), Mali (1181/100,000), and Mozambique (1177/100,000) are expected to remain the highest. SHAP values demonstrated that gender was the leading predictor of ID, with age and year showing negative contributions. Females aged 10 to 60 consistently showed higher prevalence and DALYs rates compared to males, with the under-5 age group having the highest rates for both. Additionally, men aged 80 and above exhibited a rapid increase in prevalence. Furthermore, the ASPR and ASDR were significantly higher in regions with a lower SDI, highlighting the greater burden of ID in less developed regions. ID remains a significant global health concern, with its burden projected to persist through 2050, particularly in lower-SDI regions. Despite declines in ASPR and ASDR, total cases and DALYs are expected to rise. SHAP analysis revealed that gender had the greatest influence on the model's predictions, while both age and year showed overall negative contributions to ID risk. Children under 5, women under 60, and elderly men aged 80+ were the most vulnerable groups. These findings underscore the need for targeted interventions, such as improved nutrition, early screening, and addressing socioeconomic drivers through iron supplementation programs in low-SDI regions.

Machine learning and Bayesian network analyses identifies associations with insomnia in a national sample of 31,285 treatment-seeking college students

BackgroundA better understanding of the relationships between insomnia and anxiety, mood, eating, and alcohol-use disorders is needed given its prevalence among young adults. Supervised machine learning provides the ability to evaluate which mental disorder is most associated with heightened insomnia among U.S. college students. Combined with Bayesian network analysis, probable directional relationships between insomnia and interacting symptoms may be illuminated.MethodsThe current exploratory analyses utilized a national sample of college students across 26 U.S. colleges and universities collected during population-level screening before entering a randomized controlled trial. We used a 4-step statistical approach: (1) at the disorder level, an elastic net regularization model examined the relative importance of the association between insomnia and 7 mental disorders (major depressive disorder, generalized anxiety disorder, social anxiety disorder, panic disorder, post-traumatic stress disorder, anorexia nervosa, and alcohol use disorder); (2) This model was evaluated within a hold-out sample. (3) at the symptom level, a completed partially directed acyclic graph (CPDAG) was computed via a Bayesian hill-climbing algorithm to estimate potential directionality among insomnia and its most associated disorder [based on SHAP (SHapley Additive exPlanations) values)]; (4) the CPDAG was then tested for generalizability by assessing (in)equality within a hold-out sample using structural hamming distance (SHD).ResultsOf 31,285 participants, 20,597 were women (65.8%); mean (standard deviation) age was 22.96 (4.52) years. The elastic net model demonstrated clinical significance in predicting insomnia severity in the training sample [R2 = .44 (.01); RMSE = 5.00 (0.08)], with comparable performance in the hold-out sample (R2 = .33; RMSE = 5.47). SHAP values indicated that the presence of any mental disorder was associated with higher insomnia scores, with major depressive disorder as the most important disorder associated with heightened insomnia (mean |SHAP|= 3.18). The training CPDAG and hold-out CPDAG (SHD = 7) suggested depression symptoms presupposed insomnia with depressed mood, fatigue, and self-esteem as key parent nodes.ConclusionThese findings provide insights into the associations between insomnia and mental disorders among college students and warrant further investigation into the potential direction of causality between insomnia and depression.Trial registrationTrial was registered on the National Institute of Health RePORTER website (R01MH115128 || 23/08/2018).

B-123 Prediction of Opioid Use Duration in Post-Surgical Orthopedic Patients Using CYP2D6 Inhibitor Index and Routine Laboratory Markers

Abstract Background Hydrocodone, the most commonly prescribed opioid in the U.S., is typically used for pain management following surgery and is a known potential drug of abuse with non-negligible risk of addiction. Significant variability in patients’ response to hydrocodone therapy have been reported, owing to both iatrogenic and endogenous sources. The objective of this study is to evaluate the feasibility of predicting post-discharge hydrocodone use in patients who have undergone orthopedic surgery using patient demographics, genotyping, concurrently prescribed medications, and routinely collected laboratory test results. Methods Targeted genotyping was performed on each patient with genes known to affect hydrocodone pharmacokinetics or pharmacodynamics, including CYP2D6, CYP3A4, CYP3A5, CYP2B6, COMT, OPRM1, and ABCB1. An inhibition index for CYP2D6 was calculated by ranking concurrently prescribed medications as weak, moderate, or strong inhibitors of enzyme activity. Pain index, average hydrocodone dose per day during hospital stay, and CYP2D6 estimated activity scores based on allele functionality were included as predictors. In addition, a retrospective chart review of electronic hospital records was performed where basic metabolic panel and complete blood count results were screened within 10 days of each patient's surgery date. Two types of classification and decision tree algorithms were tuned and validated as prediction models using 5-fold cross validation, including a ‘Fast and Frugal Decision Tree’ (FFTree) and an extreme gradient boosting machine (xgBoost). Clinical predictive features were ranked according to the Shapley Additive Explanations (SHAP) values using the xgBoost model. Results Out of 79 total patients with hydrocodone use duration information, 25 were categorized as “Short” and 54 categorized as “Long” using a 1-week threshold. FFTree model selected the top three features of CYP2D6 inhibition index, blood sodium level, and blood creatinine level and obtained a sensitivity and specificity of 0.80 and 0.76, respectively. The xgBoost model, using the same three features, achieved a sensitivity and specificity of 0.872 and 0.628, respectively, an AUROC of 0.814, PRAUC of 0.912, net benefit of 0.359, and was well-calibrated with a Brier score of 0.161. Consistent with previously published results, the predictive capacity of all genotypes analyzed were unremarkable and consistently ranked as the least important features by the xgBoost model SHAP values, due to their low predictive value. Conclusions Both the FFTree and xgBoost models were able to adequately classify the patients according to a 1-week stratification criteria with good to excellent classification performance. Patients with larger CYP2D6 inhibition index were more likely to be classified as “Long” duration. Sodium and creatinine level, which may reflect hydration status and renal function, and therefore hydromorphone elimination rate, were key predictors of post-discharge hydrocodone use duration when combined with the CYP2D6 inhibition index. Further refinement of the proposed models and validation of model robustness and stability using external cohorts is necessary. These results may have implications for providers when prescribing hydrocodone for pain management, where the ability to risk-stratify patients according to their predicted probability of developing opioid use disorder is of clinical importance.

Machine learning aided prediction of martensite transformation temperature of NiTi-based shape memory alloy

The martensitic transformation temperature (Ms) of shape memory alloys plays a crucial role in their performance. To achieve rapid prediction of Ms in shape memory alloys, this study employs machine learning techniques, using NiTi-based SMAs, the most widely used type, as an example. The results show that the gradient boosting machine algorithm provides the best prediction accuracy, with an R² of 0.92 and a mean absolute error (MAE) of 23.42 °C on the test set. Through correlation analysis, recursive feature elimination, and exhaustive search methods, six key features were identified. Subsequently, SHapley Additive exPlanations (SHAP) values were introduced to analyze the model interpretability, revealing the feature importance ranking and the dependence relationship between features and SHAP values. To address the issue of low Ms in NiTi-based shape memory alloys, the concept of high-entropy alloys was introduced. The study designed a TiZrHfNiCu high-entropy shape memory alloy and efficiently screened alloy compositions using the predictive model. This approach successfully increased the Ms, with the Ti19Zr19Hf19Ni37Cu6 alloy exhibiting an Ms exceeding 400 °C, significantly enhancing high-temperature application performance.

Applicability of ensemble learning in total organic carbon and porosity evaluation of shales

Accurate evaluation of total organic carbon (TOC) content and porosity is of paramount significance for assessment and target interval selection for shale reservoirs. This study takes shales from the western Chongqing area as an exemplary case to delve into the applicability and reliability of ensemble learning in evaluating TOC content and porosity. The results indicate that although both Light Gradient Boosting Machine (LightGBM) and Random Forest (RF) algorithms are suitable for evaluating TOC content and porosity in shales, LightGBM algorithm is preferred due to its comprehensive advantages, including higher accuracy, stronger generalization capability, and faster operating speed. For TOC content evaluation, the four most important logging parameters identified by LightGBM and RF are consistent, but exhibit different orders: DEN (compensated density) &amp;gt; GR (gamma ray) &amp;gt; U (uranium) &amp;gt; CNL (compensated neutron) and DEN &amp;gt; U &amp;gt; GR &amp;gt; CNL, respectively. For porosity evaluation, LightGBM and RF identify the same type and order of the three most important logging parameters: AC (acoustic transit time) &amp;gt; DEN &amp;gt; U. This similarity may be attributed to the fact that both algorithms utilize Classification and Regression Tree (CART) as base learners. The dependence plots between SHAP (SHapley Additive exPlanations) values and logging parameters reveal that the role of each logging parameter in the evaluation model is segmented, rather than exhibiting a continuous linear contribution. In conclusion, given the exceptional performance of ensemble learning algorithms, they, especially LightGBM algorithm, are highly recommended for shale evaluation.

Polynomial-SHAP analysis of liver disease markers for capturing of complex feature interactions in machine learning models

Liver disease diagnosis is pivotal for effective patient management, and machine learning techniques have shown promise in this domain. In this study, we investigate the impact of Polynomial-SHapley Additive exPlanations analysis on enhancing the performance and interpretability of machine learning models for liver disease classification. Our results demonstrate significant improvements in accuracy, precision, recall, F1_score, and Matthews correlation coefficient across various algorithms when polynomial- SHapley Additive exPlanations analysis is applied. Specifically, the Light Gradient Boosting Machine model achieves exceptional performance with 100 % accuracy in both scenarios. Furthermore, by comparing the results obtained with and without the approach, we observe substantial differences in the performance, highlighting the importance of incorporating Polynomial-SHapley Additive exPlanations analysis for improved model performance. The Polynomial features and SHapley Additive exPlanations values also enhance the interpretability of machine learning models by capturing complex feature interactions, enabling users to gain deeper insights into the underlying mechanisms driving the diagnosis. Moreover, data rebalancing using Synthetic Minority Over-sampling Technique and parameter tuning were employed to optimize the performance of the models. These findings underscore the significance of employing this analytical approach in machine-learning-based diagnostic systems for liver diseases, offering superior performance and enhanced interpretability for informed decision-making in clinical practice.

Personalized Prediction of Long-Term Renal Function Prognosis Following Nephrectomy Using Interpretable Machine Learning Algorithms: Case-Control Study.

Acute kidney injury (AKI) is a common adverse outcome following nephrectomy. The progression from AKI to acute kidney disease (AKD) and subsequently to chronic kidney disease (CKD) remains a concern; yet, the predictive mechanisms for these transitions are not fully understood. Interpretable machine learning (ML) models offer insights into how clinical features influence long-term renal function outcomes after nephrectomy, providing a more precise framework for identifying patients at risk and supporting improved clinical decision-making processes. This study aimed to (1) evaluate postnephrectomy rates of AKI, AKD, and CKD, analyzing long-term renal outcomes along different trajectories; (2) interpret AKD and CKD models using Shapley Additive Explanations values and Local Interpretable Model-Agnostic Explanations algorithm; and (3) develop a web-based tool for estimating AKD or CKD risk after nephrectomy. We conducted a retrospective cohort study involving patients who underwent nephrectomy between July 2012 and June 2019. Patient data were randomly split into training, validation, and test sets, maintaining a ratio of 76.5:8.5:15. Eight ML algorithms were used to construct predictive models for postoperative AKD and CKD. The performance of the best-performing models was assessed using various metrics. We used various Shapley Additive Explanations plots and Local Interpretable Model-Agnostic Explanations bar plots to interpret the model and generated directed acyclic graphs to explore the potential causal relationships between features. Additionally, we developed a web-based prediction tool using the top 10 features for AKD prediction and the top 5 features for CKD prediction. The study cohort comprised 1559 patients. Incidence rates for AKI, AKD, and CKD were 21.7% (n=330), 15.3% (n=238), and 10.6% (n=165), respectively. Among the evaluated ML models, the Light Gradient-Boosting Machine (LightGBM) model demonstrated superior performance, with an area under the receiver operating characteristic curve of 0.97 for AKD prediction and 0.96 for CKD prediction. Performance metrics and plots highlighted the model's competence in discrimination, calibration, and clinical applicability. Operative duration, hemoglobin, blood loss, urine protein, and hematocrit were identified as the top 5 features associated with predicted AKD. Baseline estimated glomerular filtration rate, pathology, trajectories of renal function, age, and total bilirubin were the top 5 features associated with predicted CKD. Additionally, we developed a web application using the LightGBM model to estimate AKD and CKD risks. An interpretable ML model effectively elucidated its decision-making process in identifying patients at risk of AKD and CKD following nephrectomy by enumerating critical features. The web-based calculator, found on the LightGBM model, can assist in formulating more personalized and evidence-based clinical strategies.