Multivariate Classification Tree Research Articles

Incidence and risk factors of early onset VOD/SOS differ in younger vs older adults after stem cell transplantation

AbstractVeno-occlusive disease (VOD) is a rare but potentially life-threatening complication after allogeneic hematopoietic stem cell transplantation (allo-SCT). Although increasing awareness and modern transplant techniques have mitigated risk, the interaction of historic risk factors in the current era with posttransplant cyclophosphamide (PTCy) is unknown. We performed a retrospective single-center analysis of adult patients aged ≥18 years undergoing allo-SCT (N = 1561) using predominately PTCy as graft-versus-host disease (GVHD) prophylaxis (72%). We found a higher rate of VOD at 16.8% (20 of 119) in those aged ≤25 years compared with 3.8% (55 of 1442) in those aged >25 years, with unique predictors of VOD within each cohort. Multivariate classification and regression tree (CART) analysis confirmed age as the primary independent determinant of the rate of VOD. Among patients aged 18 to 25 years, disease risk index (DRI; 31% with high/very high DRI vs 12% low/intermediate DRI; P = .03) and prior lines of chemotherapy (24% with >1 vs 6% with ≤1; P = .03) were the strongest predictors of VOD. Incidence of VOD in patients aged >25 years of age consistently ranged between 3% and 5% across most risk factors evaluated, with only hepatic factors (baseline elevation of bilirubin, aspartate transferase, alanine aminotransferase) or gemtuzumab exposure associated with increased rates of VOD. There was no significant difference in rates of VOD in those receiving PTCy compared with those receiving alternate GVHD prophylaxis. Our data highlight the differences in incidence and predictors of VOD between younger (≤25) and older (>25) adults undergoing allo-SCT.

Incidence and Risk Factors of Veno-Occlusive Disease Are Different in Younger Versus Older Adults Undergoing Allogeneic Hematopoietic Stem Cell Transplantation

Background: Veno-occlusive disease (VOD) is a rare but potentially life-threatening complication following allogeneic stem cell transplantation (allo-SCT). Improved transplant techniques and increased awareness of modifiable risk factors have reduced VOD incidence. However, the interaction of historic risk factors in the current era, particularly with the increasing use of calicheamicin-based therapies and post-transplant cyclophosphamide as GVHD prophylaxis, remains unclear. Methods: We conducted a retrospective, single center, chart-based study of consecutive adult patients (age ≥ 18 years) undergoing allo-SCT at MD Anderson Cancer Center between January 1 st, 2017 and December 31 st, 2021. VOD cases were defined using the classic VOD EBMT criteria or if patients received defibrotide treatment for VOD based on clinical judgment of the treating physician. Risk factors for VOD were evaluated in univariate analysis using Fine and Grey regression analysis. Multivariate analysis was performed using Classification and Regression Tree (CART) analysis to validate the findings of univariate analysis and evaluate independent effects accounting for potential interaction effects. Results: A total of 1561 patients were included in our analysis. Median age was 56 years and 60% were male. Patients primarily had acute myeloid leukemia/myelodysplastic syndrome (AML/MDS; 60%), acute lymphoblastic leukemia (ALL; 13%), or myeloproliferative neoplasms (13%). Most patients had matched related (28%) or unrelated donors (48%), and 21% underwent haploidentical transplantation. Forty-nine (3%) patients received inotuzumab ozogamicin (InO) containing regimens prior to allo-SCT, and 30 (2%) were exposed to gemtuzumab ozogamicin (GO). Post-transplant cyclophosphamide (PTCy) was used as GVHD prophylaxis in 72% of patients. Incidence of VOD in the entire study population was 4.8%. There was a significant difference in the median age of patients diagnosed with VOD compared to those without VOD (46 years vs 56 years, respectively, p=0.001). Further examination of the incidence of VOD within different age groups identified a VOD rate of 16.8% (20/119) in those aged ≤ 25 years compared to 3.8% (55/1442) in those > 25 years (Table 1). Multivariate classification and regression tree (CART) analysis confirmed age as the primary independent determinant of the rate of VOD (Figure 1). Given these findings, risk factor analysis was stratified according to age. In patients ≤25 years of age, disease risk index (DRI) (31% with high/very high DRI vs 12% low/intermediate DRI; p=0.03) and prior lines of chemotherapy (24% with >1 vs 6% with ≤1, p=0.03) were the strongest predictors of VOD. Within the younger cohort of patients with AML/MDS, 2 of 3 patients (67%) who received gemtuzumab ozogamicin (GO) developed VOD compared to 10% who did not receive GO (p=0.05). In patients aged >25 years, elevated baseline bilirubin, AST, ALT, and GO exposure were associated with increased VOD rates. VOD incidence based on baseline bilirubin levels (WNL, >ULN to 1.5x ULN, and >1.5x ULN) was 3%, 14%, and 16%, respectively (p<0.001). For baseline ALT levels (WNL, >ULN to 2.5x ULN, and >2.5x ULN), VOD rates were 4%, 3%, and 8%, respectively (p=0.02). Similarly, VOD rates with baseline AST levels (WNL, >ULN to 2.5x ULN, and >2.5x ULN) were 3%, 4%, and 27%, respectively (p≤0.001). In patients aged >25 years with AML/MDS, those receiving GO had a higher incidence of VOD compared to AML/MDS patients not exposed to GO (15% vs. 3%; p=0.01). There was no significant difference in VOD rates between those receiving PTCy and those receiving alternate GVHD prophylaxis in either age cohort. Conclusion: In summary, our data highlight that young adults (age 18 - 25 years) represent a distinct adult population with increased rates of VOD compared to their older counterparts, which are exacerbated by disease and treatment related factors (DRI and number lines of therapy). In contrast, patients > 25 years of age had low rates of VOD even in the presence of historical predictors, with only hepatic risk factors identified as increasing baseline VOD risk. There was no observed increased incidence of VOD among those receiving PTCy as GVHD prophylaxis.

Association Between Depression and Urinary Heavy Metal Levels

Introduction: Growing concerns about heavy metal pollution due to urbanization and industrialization have highlighted potential links between heavy metal levels and neurological disorders, including depression. This project aims to investigate the relationship between urinary heavy metal levels and depression status. Methods: The US National Health and Nutrition Examination Survey (NHANES) 2011-2018 data were used. Depression was assessed using a nine-item version of the Patient Health Questionnaire (PHQ-9), with a cut-off point of 10 for depression cases. 13 urinary heavy metals were included. Both univariate analyses, the weighted Wilcoxon test and weighted logistic regression with heavy metal variables transformed into quintiles, and multivariate analyses, Classification and Regression Tree (CART) and random forest, were conducted to investigate the association. Results: The weighted Wilcoxon test found higher levels of cadmium (Cd), antimony (Sb), tin (Sn) and tungsten (Tu) and lower levels of mercury (Hg) and arsenic (As) in the depression group. Weighted logistic regression revealed higher depression risks in the fifth quintile of Cd, the third, fourth and fifth quintiles of Sb, and the third and fifth quintiles of Tu levels. Lower risk was detected in the fifth quintile of As levels. Multivariate analysis identified Sn, Cd, As, cesium (Cs), and thallium (Tl) as crucial metals for classifying depression. Conclusion: In conclusion, this project reveals the complex relationship between urinary heavy metals and depression. Depression was associated with different sets of metals depending on the testing method used, and additional investigation is required to explore the potential interactions.

Predicting body weight through biometric measurements in growing hair sheep using data mining and machine learning algorithms.

Determination of live weight, which is one of the most important features that determine meat production, is a very important issue for herd management and sustainable livestock. In this context, the necessity of finding alternative methods has emerged, especially in rural conditions, due to the difficulties to be experienced in finding the weighing tool. Especially for conditions with no weighing tool, it has been tried to establish relations between the information obtained from body measurements and live weight. Since these studies will differ from species to species and breed to breed, the need for new studies is extremely high. For this aim, it is to evaluate the body measurement information obtained with the present study using several statistical approaches. To implement this aim, several data mining and machine learning algorithms such as multivariate adaptive regression splines (MARS), classification and regression tree (CART), and support vector machine regression (SVR) algorithms were used for training (70%) and test (30%) sets. To predict final body weight, 280 hair sheep breeds (162 female and 118 male) ranging from 2 months to 3 years were used with different data mining and machine learning approaches. Various goodness-of-fit criteria were used to evaluate the performances of the aforementioned algorithms. Although the MARS and SVR algorithms gave the same and highest results in terms of R2 and r values for both the train and the test sets, the SVR algorithm is one of the methods to be recommended as a result of this study, especially when other goodness-of-fit criteria are evaluated. In conclusion, the usage of SVR algorithms may be a useful tool of machine learning approaches for detecting the hair sheep breed standards and may contribute to increasing the sheep meat quality in Mexico.

Mpox Outbreak 2022: A Comparative Analysis of the Characteristics of Individuals Receiving MVA-BN Vaccination and People Diagnosed with Mpox Infection in Milan, Italy.

Mpox caused a worldwide outbreak in 2022, disproportionately affecting MSM reporting high-risk sexual behaviors. The aim of this study was to compare the characteristics of people receiving MVA-BN vaccination with those of individuals diagnosed with mpox to guide future vaccination policies. This was a retrospective study on people with mpox infection or vaccination at San Raffaele Scientific Institute, Milan, Italy, from May to November 2022. Characteristics were compared using Mann-Whitney or chi-square/Fisher's exact tests; multivariable logistic regression and classification tree analysis were applied. Overall, 473 vaccinated individuals and 135 with mpox were included; 472/473 and 134/135 were MSM. People with mpox were more frequently living with HIV (48.9% vs. 22.4%, p < 0.001), had ≥1 previous STI (75.6% vs. 35.7%, p < 0.001), were chemsex users (37.8% vs. 6.34%, p < 0.001), were with a higher number of partners (23.0% vs. 1.69%, p < 0.001), and had engaged in group sex (55.6% vs. 24.1%, p < 0.001). At multivariable analysis, PLWH (aOR = 2.86, 95%CI = 1.59-5.19, p < 0.001), chemsex users (aOR = 2.96, 95%CI = 1.52-5.79, p = 0.001), those with previous syphilis (aOR = 4.11, 95%CI = 2.22-7.72, p < 0.001), and those with >10 partners (aOR = 11.56, 95%CI = 6.60-21.09, p < 0.001) had a higher risk of infection. This study underscores the importance of prioritizing MSM with prior STIs and multiple partners as well as chemsex users in vaccination policies to curb mpox spread. A destigmatized assessment of sexual history is vital for comprehensive sexual health strategies.

Viral Hepatitis and Human Papillomavirus Vaccination During HIV Pre-Exposure Prophylaxis: Factors Associated With Missed Vaccination.

We evaluated factors associated with lack of triple vaccination (hepatitis A virus [HAV], hepatitis B virus [HBV], and human papillomavirus [HPV]) among men who have sex with men using pre-exposure prophylaxis (PrEP). PrEP users at the San Raffaele Scientific Institute, Italy, with ≥1 follow-up visit (May 2017-2022). Participants were considered protected if (1) before PrEP access: positive serology (IgG-HAV+, hepatitis B surface antigen >10 mUI/mL) or vaccination history was recorded and (2) after starting PrEP: ≥1 dose of each vaccination was administered. Individuals were considered fully protected if they received the following before/during PrEP access: HAV vaccination/infection, HBV vaccination/infection, and HPV vaccination. χ 2 and Kruskal-Wallis tests were used to compare characteristics of those fully, partially, and not protected. Factors associated with the lack of triple vaccination were assessed by using multivariable logistic regression and classification tree analysis. Overall, 473 men who have sex with men were considered: 146 (31%) were fully protected, 231 (48%) partially, and 96 (20%) were not. Daily-based PrEP users (fully: 93, 63.7%; partially: 107, 46.3%; and not protected: 40, 41.7%; P = 0.001) and those with a sexually transmitted infection at the first visit (43, 29.5%; 55, 23.8%; 15, 15.6%; P = 0.048) were more frequently fully protected. At multivariable analysis, the odds of lack of triple vaccination was lower among daily-based users (adjusted odds ratio = 0.47, 95% confidence interval = 0.31-0.70, P < 0.001). Classification tree analysis showed that among daily-based users, with sexually transmitted infection prior and at the first PrEP visit, there was a lower chance of lack of triple vaccination ( P = 44%). Strategies targeting PrEP users at risk of missing HAV, HBV, and HPV vaccinations need to be implemented, focusing mostly on event-based users.

The Use and Misuse of Tracking Error

Equity portfolio tracking error to a benchmark is a most ubiquitous restriction for active portfolios as prescribed by fiduciaries. The restriction is typically a tight range with minimum and maximum ex-ante extremes. The typical capitalization-weighted benchmark (i.e., the S&P 500 Index) has significant changes in diversification character overtime. This brings into question the sensibility of holding tracking error (TE) constant for a skilled active manager. The authors demonstrate that as the concentration of names in the S&P 500 increases, its diversification fades. When this happens, constant tracking error is the enemy of the skilled diversified manager, other things equal. Using multivariate classification and regression trees (CART), they show that high tracking dominates low tracking when index concentration is low, trending lower, and return dispersion is high. Low tracking dominates when the opposite index conditions exist. The authors conclude that the power of a <italic>flexible</italic> TE process in wealth creation is dominant over <italic>inflexibility</italic> in the benchmark.

On multivariate randomized classification trees: [formula omitted]-based sparsity, VC dimension and decomposition methods

Decision trees are widely-used classification and regression models because of their interpretability and good accuracy. Classical methods such as CART are based on greedy approaches but a growing attention has recently been devoted to optimal decision trees. We investigate the nonlinear continuous optimization formulation proposed in Blanquero et al. (2020) for training sparse optimal randomized classification trees. Sparsity is important not only for feature selection but also to improve interpretability. We first consider alternative methods to sparsify such trees based on concave approximations of the l0 “norm”. Promising results are obtained on 24 datasets in comparison with the original l1 and l∞ regularizations. Then, we derive bounds on the VC dimension of multivariate randomized classification trees. Finally, since training is computationally challenging for large datasets, we propose a general node-based decomposition scheme and a practical version of it. Experiments on larger datasets show that the proposed decomposition method is able to significantly reduce the training times without compromising the testing accuracy.

A data-driven approach finds RNA polymerase III antibody and tendon friction rubs as enrichment tools for early diffuse scleroderma trials.

Clinical trials in early diffuse SSc have consistently shown a placebo group response with a declining modified Rodnan skin score (mRSS), with negative outcomes. Our objective was to identify strategies using clinical characteristics or laboratory values to improve trial design. We identified early diffuse SSc patients first seen at the University of Pittsburgh from 1980-2015. Eligible patients had ≥3 visits, with at least two mRSS scores within the first year of follow-up. We performed Kaplan-Meier analyses, group-based trajectory analysis of mRSS scores, followed by multivariable regression analysis and classification tree analysis. We applied the results to the abatacept in early diffuse systemic sclerosis (ASSET) trial outcome data. We identified 403 patients with <18 months, and 514 with <36 months disease duration. The median number of mRSS follow-up scores was 14 (interquartile range 8, 25). All methodologic approaches identified skin thickness progression rate, RNA polymerase III (RNAP3) antibody positivity and presence of tendon friction rubs (TFR) as predictors of mRSS trajectory over 5 years of follow-up, and thereby as potential enrichment variables. When applied to the ASSET data, adjustment for both RNAP3 and TFR demonstrated reduction of the placebo mRSS response, particularly at 6 months. A significant difference in the ACR Composite Response Index in Systemic Sclerosis (CRISS) score was found with adjustment by RNAP3 at 6 months, and TFR or RNAP3 at 12 months. Adjustment for both RNAP3 and TFR predicts mRSS trajectory and diminished the mRSS decline in ASSET placebo group, and identified significant differences in CRISS. RNAP3, particularly, is a stratification or enrichment approach to improve early diffuse SSc trial design.

A classification and regression tree analysis identifies subgroups of childhood type 1 diabetes.

SummaryBackgroundDiabetes in childhood and adolescence includes autoimmune and non-autoimmune forms with heterogeneity in clinical and biochemical presentations. An unresolved question is whether there are subtypes, endotypes, or theratypes within these forms of diabetes.MethodsThe multivariable classification and regression tree (CART) analysis method was used to identify subgroups of diabetes with differing residual C-peptide levels in patients with newly diagnosed diabetes before 20 years of age (n=1192). The robustness of the model was assessed in a confirmation and prognosis cohort (n=2722).FindingsThe analysis selected age, haemoglobin A1c (HbA1c), and body mass index (BMI) as split parameters that classified patients into seven islet autoantibody-positive and three autoantibody-negative groups. There were substantial differences in genetics, inflammatory markers, diabetes family history, lipids, 25-OH-Vitamin D3, insulin treatment, insulin sensitivity and insulin autoimmunity among the groups, and the method stratified patients with potentially different pathogeneses and prognoses. Interferon-ɣ and/or tumour necrosis factor inflammatory signatures were enriched in the youngest islet autoantibody-positive groups and in patients with the lowest C-peptide values, while higher BMI and type 2 diabetes characteristics were found in older patients. The prognostic relevance was demonstrated by persistent differences in HbA1c at 7 years median follow-up.InterpretationThis multivariable analysis revealed subgroups of young patients with diabetes that have potential pathogenetic and therapeutic relevance.FundingThe work was supported by funds from the German Federal Ministry of Education and Research (01KX1818; FKZ 01GI0805; DZD e.V.), the Innovative Medicine Initiative 2 Joint Undertaking INNODIA (grant agreement No. 115797), the German Robert Koch Institute, and the German Diabetes Association.

Mapping Gully Erosion Variability and Susceptibility Using Remote Sensing, Multivariate Statistical Analysis, and Machine Learning in South Mato Grosso, Brazil

In Brazil, the development of gullies constitutes widespread land degradation, especially in the state of South Mato Grosso, where fighting against this degradation has become a priority for policy makers. However, the environmental and anthropogenic factors that promote gully development are multiple, interact, and present a complexity that can vary by locality, making their prediction difficult. In this framework, a database was constructed for the Rio Ivinhema basin in the southern part of the state, including 400 georeferenced gullies and 13 geo-environmental descriptors. Multivariate statistical analysis was performed using principal component analysis (PCA) to identify the processes controlling the variability in gully development. Susceptibility maps were created through four machine learning models: multivariate discriminant analysis (MDA), logistic regression (LR), classification and regression tree (CART), and random forest (RF). The predictive performance of the models was analyzed by five evaluation indices: accuracy (ACC), sensitivity (SST), specificity (SPF), precision (PRC), and Receiver Operating Characteristic curve (ROC curve). The results show the existence of two major processes controlling gully erosion. The first is the surface runoff process, which is related to conditions of slightly higher relief and higher rainfall. The second also reflects high surface runoff conditions, but rather related to high drainage density and downslope, close to the river network. Human activity represented by peri-urban areas, construction of small earthen dams, and extensive rotational farming contribute significantly to gully formation. The four machine learning models yielded fairly similar results and validated susceptibility maps (ROC curve &gt; 0.8). However, we noted a better performance of the random forest (RF) model (86% and 89.8% for training and test, respectively, with an ROC curve value of 0.931). The evaluation of the contribution of the parameters shows that susceptibility to gully erosion is not governed primarily by a single factor, but rather by the interconnection between different factors, mainly elevation, geology, precipitation, and land use.

An Inductive Approach to Developing Ecological Site Concepts with Existing Monitoring Data

Ecological sites comprise a land classification system that represents potential vegetation states and their management needs for different soils and climates. In the Rio Grande del Norte National Monument (RGdNNM) in northern New Mexico, uncertainty about the patterns and drivers of vegetation states impedes sustainable land management. Similar challenges are ubiquitous across terrestrial ecosystems and in particular landscapes with high spatial variability in soils and climate. Lack of suitable data has been a barrier to large-scale ecological site development based on quantitative observations. We used data from existing federal monitoring programs alongside spatial, environmental, and land use data to test for the role of climate, geomorphology, soils, and land use history on vegetation communities in RGdNNM. The monitoring dataset was collected with standardized monitoring methods implemented by the Bureau of Land Management's Assessment, Inventory, and Monitoring program and the Natural Resources Conservation Service's Landscape Monitoring Framework program. Eleven ecological site concepts and paired vegetation communities were identified using multivariate fuzzy clustering and classification tree analysis to determine the influence of abiotic variables on vegetation communities. The ecological site and vegetation community concepts developed for RGdNNM demonstrate how existing monitoring data can be used to interpret the structural and functional characteristics of landscapes. A workflow for applying monitoring data to landscape classification is presented to support the broader framework for ecological site development.

Hippocampal Metastasis Rate Based on Non-Small Lung Cancer TNM Stage and Molecular Markers.

Hippocampal-avoidance whole-brain radiation therapy (HA-WBRT) is justified because of low hippocampal brain metastases (BM) rate and its prevention of cognitive decline. However, we hypothesize that the risk of developing BM in the hippocampal-avoidance region (HAR) may differ depending on the lung-cancer stage and molecular status. We retrospectively reviewed 123 patients with non-small cell lung cancer (NSCLC) at the initial diagnosis of BM. The number of BMs within the HAR (5 mm expansion) was counted. The cohort was divided into patients with and without BMs in the HAR, and their clinical variables, TNM stage, and epidermal growth factor receptor (EGFR) status were compared. The most influential variable predicting BMs in the HAR was determined using multi-variable logistic regression, classification and regression tree (CART) analyses, and gradient boosting method (GBM). The feasibility of HAR expansion was tested using generalized estimating equation marginal model. Patients with BMs in the HAR were more frequently non-smokers, and more likely to have extra-cranial metastases and EGFR mutations (p<0.05). Multi-variable analysis revealed that extra-cranial metastases were independently associated with the presence of BM in the HAR (odds ratio=8.75, p=0.04). CART analysis and GBM revealed that the existence of extra-cranial metastasis was the most influential variable predicting BM occurrence in the HAR (variable importance: 23% and relative influence: 37.38). The estmated BM incidence of patients without extra-cranial metastases in th extended HAR (7.5-mm and 10-mm expansion) did not differ significantly from that in the conventional HAR. In conclusion, NSCLC patients with extra-cranial metastases were more likely to have BMs in the HAR than those without extra-cranial metastases.

Use of Multivariate Adaptive Regression Splines, Classification Tree and Roc Curve in Diagnosis of Subclinical Mastitis in Dairy Cattle

Subclinical mastitis is one of the most significant diseases that cause economic losses in dairy cattle farming. This investigation was conducted on 112 head Holstein Friesian cows in order to reveal relationship between subclinical mastitis and electrical conductivity milk composition and milk quality. In the study, CMT (California Mastitis Test) and CSCC (Classified Somatic Cell Count) used in diagnosis of subclinical mastitis were used as a binary response variable i.e. healthy and unhealthy. Potential predictors included here were lactation number, days in milk (DIM), L, a, b, H, C, milk fat, milk protein, lactose, milk freezing point, SNF, density, solids, pH and electrical conductivity. CART, CHAID, Exhaustive CHAID, QUEST and multivariate adaptive regression splines (MARS) were used as data mining algorithms that help to make an accurate decision about detecting influential factors increasing risk of subclinical mastitis.In conclusion, better classification performances of CART and MARS data mining algorithms were determined compared with those of remaining algorithms in order to correctly discriminate healthy and unhealthy cows.

Prediction of rainfall time series using soft computing techniques.

Past studies indicate that increasing temperatures would accelerate the Earth's water cycle and in turn would increase the evaporation rate. Increased evaporation will result in more frequent and intense storms; hence, most researchers focus on climate change and its effect on Earth, particularly the precipitation. In the last two decades, the Udaipur district, India, faces water scarcity and flooding situations twice. The present study focuses on the prediction of rainfall using the most advanced soft computing techniques (SCT) such as multivariate adaptive regression splines (MARS), classification and regression trees (CART), and gene expression programming (GEP) in India's Udaipur district. The performance of these SCT was evaluated to test the capability to predict the rainfall. Results showed that the MARS model for rainfall prediction showed better performance than the GEP model.

Prediction of Geosmin at Different Depths of Lake Using Machine Learning Techniques.

Geosmin is a major concern in the management of water sources worldwide. Thus, we predicted concentration categories of geosmin at three different depths of lakes (i.e., surface, middle, and bottom), and analyzed relationships between geosmin concentration and factors such as phytoplankton abundance and environmental variables. Data were collected monthly from three major lakes (Uiam, Cheongpyeong, and Paldang lakes) in Korea from May 2014 to December 2015. Before predicting geosmin concentration, we categorized it into four groups based on the boxplot method, and multivariate adaptive regression splines, classification and regression trees, and random forest (RF) were applied to identify the most appropriate modelling to predict geosmin concentration. Overall, using environmental variables was more accurate than using phytoplankton abundance to predict the four categories of geosmin concentration based on AUC and accuracy in all three models as well as each layer. The RF model had the highest predictive power among the three SDMs. When predicting geosmin in the three water layers, the relative importance of environmental variables and phytoplankton abundance in the sensitivity analysis was different for each layer. Water temperature and abundance of Cyanophyceae were the most important factors for predicting geosmin concentration categories in the surface layer, whereas total abundance of phytoplankton exhibited relatively higher importance in the bottom layer.

Machine learning-based risk prediction of malignant arrhythmia in hospitalized patients with heart failure.

AimsPredicting the risk of malignant arrhythmias (MA) in hospitalized patients with heart failure (HF) is challenging. Machine learning (ML) can handle a large volume of complex data more effectively than traditional statistical methods. This study explored the feasibility of ML methods for predicting the risk of MA in hospitalized HF patients.Methods and resultsWe evaluated the baseline data and MA events of 2794 hospitalized HF patients in the HF cohort in Anhui Province and randomly divided the study population into training and validation sets in a 7:3 ratio. The Lasso‐logistic regression, multivariate adaptive regression splines (MARS), classification and regression tree (CART), random forest (RF), and eXtreme gradient boosting (XGBoost) algorithms were used to construct risk prediction models in the training set, and model performance was verified in the validation set. The area under the receiver operating characteristic curve (AUC) and Brier score were employed to evaluate the discrimination and calibration of the model, respectively. Clinical utility of the Lasso‐logistic regression model was analysed using decision curve analysis (DCA). The median (Q1, Q3) age of the study population was 70 (61, 77) years, and 39.5% were female. MA events occurred in 117 patients (4.2%) during hospitalization. In the training set (n = 1964), the AUC of the XGBoost model was 0.998 [95% confidence interval (CI) 0.997–1.000], which was higher than the other models (all P < 0.001). In the validation set (n = 830), there was no significant difference in AUC of Lasso‐logistic model 1 [AUC: 0.867 (95% CI 0.819–0.915)], Lasso‐logistic model 2 [AUC: 0.828 (95% CI 0.764–0.892)], MARS model [AUC: 0.852 (95% CI 0.793–0.910)], RF model [AUC: 0.804 (95% CI 0.726–0.881)], and XGBoost model [AUC: 0.864 (95% CI 0.810–0.918); all P > 0.05], which were higher than that of CART model [AUC: 0.743 (95% CI 0.661–0.824); all P < 0.05]. Brier scores for all prediction models were less than 0.05. DCA results showed that the Lasso‐logistic model had a net clinical benefit. Oral antiarrhythmic drug, left bundle branch block, serum magnesium, d‐dimer, and random blood glucose were significant predictors in half or more of the models.ConclusionsThe current study findings suggest that ML models based on the Lasso‐logistic regression, MARS, RF, and XGBoost algorithms can effectively predict the risk of MA in hospitalized HF patients. The Lasso‐logistic model had better clinical interpretability and ease of use than the other models.

Investigating correlation of physico-mechanical parameters and P-wave velocity of rocks: a comparative intelligent study

The mechanical characteristics of rocks and rock masses are considered as the determining factors in making plans in the mining and civil engineering projects. Two factors that determine how rocks responds in varying stress conditions are P-wave velocity (PWV) and its isotropic properties. Therefore, achieving a high-accurate method to estimate PWV is a very important task. This work investigates the use of different intelligent models such as multivariate adaptive regression splines (MARS), classification and regression tree (CART), group method of data handling (GMDH), and gene expression programming (GEP) for the prediction of PWV. The proposed models are then evaluated using several error statistics, i.e. squared correlation coefficient (R2) and root mean squared error (RMSE). The values of R2 obtained from the CART, MARS, GMDH, and GEP models are 0.983, 0.999, 0.995, and 0.998, respectively. Furthermore, the CART, MARS, GMDH, and GEP models predict PWV with the RMSE values of 0.037, 0.007, 0.023, and 0.020, respectively. According to the aforementioned amounts, the models presented in this work predict PWV with a good performance. Nevertheless, the results obtained reveal that the MARS model yields a better prediction in comparison to the GEP, GMDH, and CART models. Accordingly, MARS can be offered as an accurate model for predicting the aims in other rock mechanics and geotechnical fields.

Graft Failure Post Allogeneic Stem Cell Transplant: Conditioning Regimen Improves Engraftment and Survival

▪Introduction- Untreated Graft Failure (GF) post allogeneic stem cell transplant (allo-SCT) is potentially lethal. Salvage stem cells with or without additional conditioning regimen can be curative but concerns about toxicity and the need for adequate immunosuppression needs to be balanced in the therapeutic decisions. Due to the rarity of GF and paucity of data no consensus exists on the most preferred combinations to opt at the time of intervention. To address this unmet scenario we performed a large single institution retrospective study.Methods- From January 1996- February 2016, 97 patients received treatment for GF following an allo-SCT. We excluded from the analysis patients (n-15) who received autologous cells as their sole rescue and those who developed GF after progression of the underlying malignancy. Both primary GF (48%) and secondary GF (52%) post 1st allo-SCT were included. Median time from 1st allo-SCT to GF rescue intervention was 56 (range 21-738) days. Conditioning regimen prior to allogeneic rescue cell infusions was categorized as : 1) no preparative regimen (21%) 2) single agent antithymocyte globulin (ATG) or Campath (6%) 3) Reduced intensity conditioning (RIC) chemotherapy or low dose total body irradiation (TBI) (200cGY) (28%) (4) Fludarabine (Flu) + ATG (41%) and 5) miscellaneous other combinations (4%). Engraftment (absolute neutrophil count ≥ 500 for 3 consecutive days) and overall survival (OS) from the time of intervention were the main outcomes of interest. The following factors were considered for their association with primary outcomes following GF rescue intervention: 1) donor/ patient demographics, 2) disease status, 3) graft- source (related, unrelated donors, peripheral blood, bone marrow and umbilical cord blood), cell dose, graft modification (t cell deplete vs replete), level of human leucocyte antigen (HLA)/ABO matching between donor and recipient and 4) treatment related metrics (conditioning regimens and graft versus host disease prophylaxis).Results- Among patients (n=82) who received GF rescue intervention, 56 (68%) engrafted, 10 (12%) failed to engraft and 16 (20%) had early death (< 28 days). With a median follow up of 61 (range 11-242) months, 5-y OS was 18%. Only conditioning regimen correlated with engraftment with a significantly higher proportion of patients engrafting after Flu/Mel (92%) or TBI (91%) based preparative regimens (Table 1). In comparison, 38%, 40%, 67%, and 33% engrafted following no preparative regimen, and conditioning with ATG or Campath, Flu/ATG, and miscellaneous other group, respectively. Change of donor, graft source and graft modification were not significant predictors for engraftment. On multivariate Classification and Regression Tree (CART) analysis, use of 1 HLA antigen mismatched related or unrelated donor was associated with poor overall survival (HR 3.7, 95 % CI 1.5-8.9 P value 0.004). In the absence of 1 antigen related or unrelated donor the use of 2nd allo-SCT was associated with better survival compared with boost only (i.e. no preparative regimen prior to cell infusion) (HR- 0.5, 95 % CI 0.3-0.9, p value 0.03)(Figure-1). Graft rejection/failure (39.7%) and infections (23.5%) were the commonest primary causes of death.Conclusions- GF post allo-SCT is a rare entity with sparse long term survivors despite interventions. For patients eligible to receive conditioning, use of RIC portends better engraftment and survival prior to cell rescue. [Display omitted] DisclosuresOran:Celgene: Research Funding; AROG: Research Funding; Astex: Research Funding. Khouri:Novartis: Research Funding.

Practical clinical and radiological models to diagnose COVID-19 based on a multicentric teleradiological emergency chest CT cohort

Our aim was to develop practical models built with simple clinical and radiological features to help diagnosing Coronavirus disease 2019 [COVID-19] in a real-life emergency cohort. To do so, 513 consecutive adult patients suspected of having COVID-19 from 15 emergency departments from 2020-03-13 to 2020-04-14 were included as long as chest CT-scans and real-time polymerase chain reaction (RT-PCR) results were available (244 [47.6%] with a positive RT-PCR). Immediately after their acquisition, the chest CTs were prospectively interpreted by on-call teleradiologists (OCTRs) and systematically reviewed within one week by another senior teleradiologist. Each OCTR reading was concluded using a 5-point scale: normal, non-infectious, infectious non-COVID-19, indeterminate and highly suspicious of COVID-19. The senior reading reported the lesions’ semiology, distribution, extent and differential diagnoses. After pre-filtering clinical and radiological features through univariate Chi-2, Fisher or Student t-tests (as appropriate), multivariate stepwise logistic regression (Step-LR) and classification tree (CART) models to predict a positive RT-PCR were trained on 412 patients, validated on an independent cohort of 101 patients and compared with the OCTR performances (295 and 71 with available clinical data, respectively) through area under the receiver operating characteristics curves (AUC). Regarding models elaborated on radiological variables alone, best performances were reached with the CART model (i.e., AUC = 0.92 [versus 0.88 for OCTR], sensitivity = 0.77, specificity = 0.94) while step-LR provided the highest AUC with clinical-radiological variables (AUC = 0.93 [versus 0.86 for OCTR], sensitivity = 0.82, specificity = 0.91). Hence, these two simple models, depending on the availability of clinical data, provided high performances to diagnose positive RT-PCR and could be used by any radiologist to support, modulate and communicate their conclusion in case of COVID-19 suspicion. Practically, using clinical and radiological variables (GGO, fever, presence of fibrotic bands, presence of diffuse lesions, predominant peripheral distribution) can accurately predict RT-PCR status.