Important Predictor Variables Research Articles

Association between opioid utilization and patient-reported outcome measures following lumbar spine surgery

BACKGROUND CONTEXTThe patient-reported outcomes measurement information system (PROMIS), created by the National institute of Health, is a reliable and valid survey for patients with lumbar spine pathology. Preoperative opioid use has been shown to be an important predictor variable of self-reported health status in legacy patient-reported outcome measures. PURPOSETo investigate the impact of chronic preoperative opiate use on PROMIS survey scores. STUDY DESIGNRetrospective database analysis. PATIENT SAMPLEBetween March 2019 and November 2021, 227 patients underwent lumbar decompression ± ≤ 2 level fusion. Fifty-seven patients (25.11%) had chronic preoperative opioid use. OUTCOME MEASURESOswestry disability index (ODI) and PROMIS survey scores. METHODSA retrospective analysis of a prospectively maintained single center patient-reported outcome database was performed with a minimum of 2 year follow-up. PROMIS Anxiety, Depression, Fatigue, Pain Interference (PI), Physical Function (PF), Sleep disturbance (SD), and Social Roles (SR) surveys were recorded at preoperative intake with subsequent follow-up at 6, 12, and 24 months postoperatively. Patients were grouped into chronic opioid users as defined by >6-month duration of use. Differences in mean survey scores were evaluated using Welch t-tests. RESULTSTwo hundred and twenty-seven patients met our inclusion criteria of completed PROMIS surveys at the designated timepoints. A total of 57 (25.11%) were chronic opioid users (COU) prior to surgery. Analysis of patient-reported health outcomes shows that long term opioid use correlated with worse ODI and PROMIS scores at baseline compared to nonchronic users (NOU). At 1 and 2 year follow-up, the COU cohort continued to have significantly worse ODI, PROMIS Fatigue, PF, PI, SD, and SR scores. There is a statistical difference in the magnitude of change in health status between the 2 cohorts at 1 year follow-up in PROMIS Depression (-5.04±7.88 vs -2.49±8.73, p=.042), PF (6.25±7.11 vs 9.03±9.04, p=.019), and PI (-7.40±7.37 vs -10.58±9.87, p=.011) and 2 year follow-up in PROMIS PF (5.58±6.84 vs 7.99±9.64, p=.041) and PI (-6.71±8.32 vs -9.62±10.06, p=.032). Mean improvement in PROMIS scores for the COU cohort at 2 year follow-up exceeded minimal clinically important difference (MCID) in all domains except PROMIS Depression, SR and SD. CONCLUSIONPatients with chronic opioid use status have worse baseline PROMIS scores compared with patients who had nonchronic use. However, patients in the COU cohort displayed clinically significant postoperative improvement in multiple PROMIS domains. These results show that patients with chronic opioid use can benefit greatly from surgical intervention and will allow physicians to better set expectations with their patients.

Mapping stony rise landforms using a novel remote sensing, geophysical, and machine learning approach

A basalt lava plain covers much of central and western Victoria in Australia. Many of the younger flows (<1 Ma) have hummocky surfaces with abundant tumuli and lava rises, known locally as stony rises due to the protruding basalt boulders. Stony rises are noteworthy geologically but have ecological importance and are culturally significant to Aboriginal Australians. Ecologically, stony rises host many native plant species. Culturally, stony rises provide evidence of Aboriginal land use and occupation through the presence of artefact deposits. Currently, mapping of stony rises is carried out by conducting field surveys on a study-by-study basis. This could be performed more rapidly using remote sensing datasets such as LiDAR, satellite and aerial imagery, and geophysical data. This research mapped stony rises in an area that is experiencing rapid urban development using a combination of remotely sensed datasets that characterise the morphology of stony rises through a Machine Learning approach that utilises Object Based Image Analysis (OBIA) segmentation and Random Forests (RF) classification techniques. The research successfully identified stony rises in the landscape to an accuracy of 89.97 %, with 314 potential new stony rise sites identified. The research identified the most important predictor variables were slope, local elevation, the concentrations of thorium and uranium (obtained from airborne radiometric data), aeromagnetic data (total magnetic intensity), the Normalised Difference Water Index (NDWI), and the Clay Mineral Ratio. The results from this study lay a foundation for standardising the definition of a stony rise within Australia. Furthermore, by utilising various predictor variables, this study highlights the success of a multi-faceted approach into detecting a specific landform at a landscape level.

Application of the Boruta algorithm to assess the multidimensional determinants of malnutrition among children under five years living in southern Punjab, Pakistan

BackgroundMalnutrition causes nutrient deficiencies that have both physical and clinical consequences in severe acute malnutrition children. Globally, there were 47 million wasted children under the age of five in 2019. One in four were located in sub-Saharan Africa, with half being in South Asia. This study aims to apply the Boruta algorithm to identify the determinants of undernutrition among children under five living in Dera Ghazi Khan, one of the marginalized districts of densely populated Punjab Province in Pakistan.MethodsA multicenter cross-sectional study design was used to collect data from 185 children with severe acute malnutrition aged under five years visiting the OTPs centers located in Dera Ghazi Khan, Punjab, Pakistan. A purposive sampling technique was used to collect data using a pretested structured questionnaire from parents/caregivers regarding family sociodemographic characteristics, child nutrition, and biological and healthcare characteristics. Anthropometric measurements, including height, weight, and mid-upper arm circumference, were collected. The Boruta models were used to incorporate the children’s anthropometric, nutritional, and household factors to determine the important predictive variables for undernutrition using the Boruta package in R studio.ResultsThis study included 185 children, with a mean age of 15.36 ± 10.23 months and an MUAC of 10.19 ± 0.96 cm. The Boruta analysis identifies age, mid-upper arm circumference, weaning practices, and immunization status as important predictors of undernutrition. Income per month, exclusive breastfeeding, and immunization status were found to be key factors of undernutrition in children under the age of five.ConclusionThis study highlights age, mid-upper arm circumference, weaning practices, and immunization status as key determinants of weight-for-height and weight-for-age in children under five years. It also suggests that economic context may influence undernutrition. The findings can guide targeted strategies for combating undernutrition.

Effects of climate on the distribution and conservation of commonly observed European earthworms.

Belowground biodiversity distribution does not necessarily reflect aboveground biodiversity patterns, but maps of soil biodiversity remain scarce because of limited data availability. Earthworms belong to the most thoroughly studied soil organisms and-in their role as ecosystem engineers-have a significant impact on ecosystem functioning. We used species distribution modeling (SDMs) and available data sets to map the spatial distribution of commonly observed (i.e., frequently recorded) earthworm species (Annelida, Oligochaeta) across Europe under current and future climate conditions. First, we predicted potential species distributions with commonly used models (i.e., MaxEnt and Biomod) and estimated total species richness (i.e., number of species in a 5 × 5km grid cell). Second, we determined how much the different types of protected areas covered predicted earthworm richness and species ranges (i.e., distributions) by estimating the respective proportion of the range area. Earthworm species richness was high in central western Europe and low in northeastern Europe. This pattern was mainly associated with annual mean temperature and precipitation seasonality, but the importance of predictor variables to species occurrences varied among species. The geographical ranges of the majority of the earthworm species were predicted to shift to eastern Europe and partly decrease under future climate scenarios. Predicted current and future ranges were only poorly covered by protected areas, such as national parks. More than 80% of future earthworm ranges were on average not protected at all (mean [SD] = 82.6% [0.04]). Overall, our results emphasize the urgency of considering especially vulnerable earthworm species, as well as other soil organisms, in the design of nature conservation measures.

Using machine learning to identify indicators of rare earth element enrichment in sedimentary strata with applications for metal prospectivity

Rare earth elements (REE), classified as critical minerals which are crucial for clean energy technologies, face soaring demand. While economic deposits are found in limited geologic environments including carbonatites and ion-adsorption clays, unconventional, secondary sources such as those from sedimentary basins could hold potential to meet this increased demand. Coal and its associated combustion by-products, phosphorites, oil sands tailings, and formation waters have all garnered interest for REE recovery, yet they remain significantly underexplored. Accordingly, new tools for data analysis and optimization such as machine learning can assist in mineral prospectivity, with these tools being subject to rapid proliferation in the Earth sciences.This work leverages compositional data analysis principles and machine learning to probe geochemical relationships and predict REE abundances in sedimentary lithologies using unsupervised (correlation, principal component, and cluster analysis) and supervised (regression, support vector machine, random forest, and boosting) machine learning models. These three unsupervised models display similar results, with REE typically being associated with incompatible elements (e.g., Th, Nb, and Hf). Gradient boosting, Adaboost, and Random Forest had the highest performance for predicting REE concentrations, with Th and P commonly being the most important predictor variables. Identifying geochemical indicators of REE enrichment that may be used to assist in discovering potentially exploitable REE resources based on existing data, as well as increasing the understanding of metal behaviour in sedimentary systems, is a step forward in understanding novel secondary and unconventional REE sources. Although REE concentrations from these sources are generally lower than primary ore deposits, the amount of available feedstock, potentially simpler, cheaper, and less environmentally taxing extraction processes, and the added benefit of remediating waste streams and contributing to the circular economy make these sources alluring.

Estimating Lithium Concentrations in Groundwater Used as Drinking Water for the Conterminous United States.

Lithium (Li) concentrations in drinking-water supplies are not regulated in the United States; however, Li is included in the 2022 U.S. Environmental Protection Agency list of unregulated contaminants for monitoring by public water systems. Li is used pharmaceutically to treat bipolar disorder, and studies have linked its occurrence in drinking water to human-health outcomes. An extreme gradient boosting model was developed to estimate geogenic Li in drinking-water supply wells throughout the conterminous United States. The model was trained using Li measurements from ∼13,500 wells and predictor variables related to its natural occurrence in groundwater. The model predicts the probability of Li in four concentration classifications, ≤4 μg/L, >4 to ≤10 μg/L, >10 to ≤30 μg/L, and >30 μg/L. Model predictions were evaluated using wells held out from model training and with new data and have an accuracy of 47-65%. Important predictor variables include average annual precipitation, well depth, and soil geochemistry. Model predictions were mapped at a spatial resolution of 1 km2 and represent well depths associated with public- and private-supply wells. This model was developed by hydrologists and public-health researchers to estimate Li exposure from drinking water and compare to national-scale human-health data for a better understanding of dose-response to low (<30 μg/L) concentrations of Li.

Artificial intelligence-based prediction models for acute myeloid leukemia using real-life data: A DATAML registry study

We designed artificial intelligence-based prediction models (AIPM) using 52 diagnostic variables from 3687 patients included in the DATAML registry treated with intensive chemotherapy (IC, N = 3030) or azacitidine (AZA, N = 657) for an acute myeloid leukemia (AML). A neural network called multilayer perceptron (MLP) achieved a prediction accuracy for overall survival (OS) of 68.5% and 62.1% in the IC and AZA cohorts, respectively. The Boruta algorithm could select the most important variables for prediction without decreasing accuracy. Thirteen features were retained with this algorithm in the IC cohort: age, cytogenetic risk, white blood cells count, LDH, platelet count, albumin, MPO expression, mean corpuscular volume, CD117 expression, NPM1 mutation, AML status (de novo or secondary), multilineage dysplasia and ASXL1 mutation; and 7 variables in the AZA cohort: blood blasts, serum ferritin, CD56, LDH, hemoglobin, CD13 and disseminated intravascular coagulation (DIC). We believe that AIPM could help hematologists to deal with the huge amount of data available at diagnosis, enabling them to have an OS estimation and guide their treatment choice. Our registry-based AIPM could offer a large real-life dataset with original and exhaustive features and select a low number of diagnostic features with an equivalent accuracy of prediction, more appropriate to routine practice.

Optimising carbon fixation through agroforestry: Estimation of aboveground biomass using multi-sensor data synergy and machine learning

As agricultural land expansion is the primary driver of deforestation, agroforestry could be an optimal land use strategy for climate change mitigation and reducing pressure on forests. Agroforestry is a promising method for carbon sequestration. With recent advancements in geospatial and data science technology, the ability to predict aboveground biomass (AGB) and assess ecosystem services in agroforestry is rapidly expanding. This study was conducted in the Belpada Block of Balangir, Odisha, a forest-dominated region of eastern India. We recorded species occurrence and measured plant parameters, including Circumference at Breast Height (CBH), height, and geolocation, in 196 plots (0.09 ha) in agroforestry intervention sites and noted the tree species. This study used Sentinel-1 and Sentinel-2 multi sensor data to achieve data synergy in AGB estimation. Three machine learning models were used: Random Forest (RF), Support Vector Machine (SVM), and Artificial Neural Network (ANN). The RF model exhibited the highest level of prediction accuracy (R2 = 0.69 and RMSE = 17.07 Mg/ha), followed by the ANN model (R2 = 0.63 and RMSE = 19.35 Mg/ha), SVM model (R2 = 0.54, RMSE = 21.97 Mg/ha. The spectral vegetation indices that are (Normalized Difference Vegetation Index (NDVI), Soil-Adjusted Vegetation Index (SAVI), Enhanced Vegetation Index (EVI), Modified Simple Ratio (MSR), Modified Soil-Adjusted Vegetation Index (MSAVI), Difference Vegetation Index (DVI), and SAR backscatter values, were found important variables for AGB prediction. The findings revealed that agroforestry interventions and plantations resulted in an average carbon stock increase of 15 Mg/ha over five years in the study area. The Plant Value Index (PVI), which indicates the importance of species in the local economy and biomass carbon storage, showed that Tectona grandis was the dominant species with the highest PVI value (88.35), followed by Eucalyptus globulus (56.87), Mangifera indica (53.75), and Azadirachta indica (15.45). This approach enables the expansion of monitoring efforts to assess carbon stock in agroforestry systems, thereby promoting effective management strategies.

Building Machine Learning Models to Improve Hypertension Diagnosis

Hypertension (HT) is a major risk factor for heart disease, stroke, and kidney disease. A diagnosis of HT is based on two or more blood pressure (BP) readings taken on two or more visits, and such diagnosis may vary depending on the threshold BP values utilized by specific guidelines. Our hypertension identification approach has been recently proposed and validated for HT diagnosis and could be used as ground truth in modeling of latent HT diagnosis. Our approach to modeling of latent HT diagnosis with high precision will leverage analytics to “Big Data”, such as electronic health records (EHRs). In this work, we will review the time complexity underlying the classical ML methods XGBoost (XGB) and Artificial Neural Networks (ANN). In particular, we compare the XGB and ANN to leverage their strengths. The performance of all algorithms for diagnosing HT will be characterized using the area under the curve (AUC) approach on a big EHR longitudinal dataset. Predictor variable importance and model interpretability will be assessed using the Shapley values approach.

Predicting traumatic brain injury outcomes using a posterior dominant rhythm.

Predicting severe traumatic brain injury (sTBI) outcomes is challenging, and existing models have limited applicability to individual patients. This study aimed to identify metrics that could predict recovery following sTBI. The researchers strived to demonstrate that a posterior dominant rhythm on electroencephalography is strongly associated with positive outcomes and to develop a novel machine learning-based model that accurately forecasts the return of consciousness. In this retrospective study, the authors assessed all intubated adults admitted with sTBI (Glasgow Coma Scale [GCS] score ≤ 8) from 2010 to 2021, who underwent EEG recording < 30 days from sTBI (n = 195). Seventy-three clinical, radiographic, and EEG variables were collected. Based on the presence of a PDR within 30 days of injury, two cohorts were created-those with a PDR (PDR[+] cohort, n = 51) and those without (PDR[-] cohort, n = 144)-to assess differences in presentation and four outcomes: in-hospital survival, recovery of command following, Glasgow Outcome Scale-Extended (GOS-E) score at discharge, and GOS-E score at 6 months post discharge. AutoScore, a machine learning-based clinical score generator that selects and assigns weights to important predictive variables, was used to create a prognostic model that predicts in-hospital survival and recovery of command following. Lastly, the MRC-CRASH and IMPACT traumatic brain injury predictive models were used to compare expected patient outcomes with true outcomes. At presentation, the PDR(-) cohort had a lower mean GCS motor subscore (1.97 vs 2.45, p = 0.048). Despite no difference in predicted outcomes (via MRC-CRASH and IMPACT), the PDR(+) cohort had superior rates of in-hospital survival (84.3% vs 63.9%, p = 0.007), recovery of command following (76.5% vs 53.5%, p = 0.004), and mean discharge GOS-E score (3.00 vs 2.39, p = 0.006). There was no difference in the 6-month GOS-E score. AutoScore was then used to identify the 7 following variables that were highly predictive of in-hospital survival and recovery of command: age, body mass index, systolic blood pressure, pupil reactivity, blood glucose, and hemoglobin (all at presentation), and a PDR on EEG. This model had excellent discrimination for predicting in-hospital survival (area under the curve [AUC] 0.815) and recovery of command following (AUC 0.700). A PDR on EEG in sTBI patients predicts favorable outcomes. The authors' prognostic model has strong accuracy in predicting these outcomes, and performed better than previously reported models. The authors' model can be valuable in clinical decision-making as well as counseling families following these types of injuries.

Personalized Prognosis with Machine Learning Models for Predicting In-Hospital Outcomes Following Intracranial Meningioma Resections

Meningiomas display diverse biological traits and clinical behaviors, complicating patient outcome prediction. This heterogeneity, along with varying prognoses, underscores the need for a precise, personalized evaluation of postoperative outcomes. Data from the American College of Surgeons National Surgical Quality Improvement Program database identified patients who underwent intracranial meningioma resections from 2014 to 2020. We focused on 5 outcomes: prolonged LOS, nonhome discharges, 30-day readmissions, unplanned reoperations, and major complications. Six machine learning algorithms, including TabPFN, TabNet, XGBoost, LightGBM, Random Forest, and Logistic Regression, coupled with the Optuna optimization library for hyperparameter tuning, were tested. Models with the highest area under the receiver operating characteristic (AUROC) values were included in the web application. SHapley Additive exPlanations were used to evaluate the importance of predictor variables. Our analysis included 7000 patients. Of these patients, 1658 (23.7%) had prolonged LOS, 1266 (18.1%) had nonhome discharges, 573 (8.2%) had 30-day readmission, 253 (3.6%) had unplanned reoperation, and 888 (12.7%) had major complications. Performance evaluation indicated that the top-performing models for each outcome were the models built with LightGBM and Random Forest algorithms. The LightGBM models yielded AUROCs of 0.842 and 0.846 in predicting prolonged LOS and nonhome discharges, respectively. The Random Forest models yielded AUROCs of 0.717, 0.76, and 0.805 in predicting 30-day readmissions, unplanned reoperations, and major complications, respectively. The study successfully demonstrated the potential of machine learning models in predicting short-term adverse postoperative outcomes after meningioma resections. This approach represents a significant step forward in personalizing the information provided to meningioma patients.

Estimating the Surface Fuel Load of the Plant Physiognomy of the Cerrado Grassland Using Landsat 8 OLI Products

Techniques and tools meant to aid fire management activities in the Cerrado, such as accurately determining the fuel load and composition spatially and temporally, are pretty scarce. The need to obtain fuel information for more efficient management in a considerably heterogeneous, biodiverse, and fire-dependent environment requires a constant search for improved remote sensing techniques for determining fuel characteristics. This study presents the following objectives: (1) to assess the use of data from Landsat 8 OLI images to estimate the fine surface fuel load of the Cerrado during the dry season by adjusting multiple linear regression equations, (2) to estimate the fuel load through random forest and k-nearest neighbor (k-NN) algorithms in comparison to regression analyses, and (3) to evaluate the importance of predictor variables from satellite images. Therefore, 64 sampling units were collected, and the pixel values associated with the field plots were extracted in a 3 × 3-pixel window surrounding the reference pixel. For multiple linear regression analyses, the R2 values ranged from 0.63 to 0.78, while the R2 values of the models fitted using the random forest algorithm ranged from 0.52 to 0.83 and the R2 values of those fitted using the k-NN algorithm ranged from 0.30 to 0.68. The estimates made through multiple linear regression analyses showed better results for the equations adjusted for the beginning of the dry season (May and June). Adopting the random forest algorithm resulted in improvements in the statistical metrics of evaluation of the fuel load estimates for the Cerrado grassland relative to multiple linear regression analyses. The variable fraction-soil (FS) exerted the most significant effect on surface fuel load estimates, followed by the vegetation indices NDII, GVMI, DER56, NBR, and MSI, all of which use near-infrared and short-wave infrared channels in their calculations.

Predicting 30-Day Non-Seizure Outcomes Following Temporal Lobectomy with Personalized Machine Learning Models

Temporal lobe epilepsy is the most common reason behind drug-resistant seizures and temporal lobectomy (TL) is performed after all other efforts have been taken for a Temporal lobe epilepsy. Our study aims to develop multiple machine learning (ML) models capable of predicting postoperative outcomes following TL surgery. Data from the American College of Surgeons National Surgical Quality Improvement Program database identified patients who underwent TL surgery. We focused on 3 outcomes: prolonged length of stay (LOS), nonhome discharges, and 30-day readmissions. Six ML algorithms, TabPFN, XGBoost, LightGBM, Support Vector Machine, Random Forest, and Logistic Regression, coupled with the Optuna optimization library for hyperparameter tuning, were tested. Models with the highest area under the receiver operating characteristic (AUROC) values were included in the web application. SHapley Additive exPlanations was used to evaluate importance of predictor variables. Our analysis included 423 patients. Of these patients, 111 (26.2%) experienced prolonged LOS, 33 (7.8%) had nonhome discharges, and 29 (6.9%) encountered 30-day readmissions. The top-performing models for each outcome were those built with the Random Forest algorithm. The Random Forest models yielded AUROCs of 0.868, 0.804, and 0.742 in predicting prolonged LOS, nonhome discharges, and 30-day readmissions, respectively. Our study uses ML to forecast adverse postoperative outcomes following TL. We developed accessible predictive models that enhance prognosis prediction for TL surgery. Making ML models available for this purpose represents a significant advancement in shifting toward a more patient-centric, data-driven paradigm.

Changes of Volatile Organic Compounds of Different Flesh Texture Pears during Shelf Life Based on Headspace Solid-Phase Microextraction with Gas Chromatography-Mass Spectrometry.

Aroma is an important sensory factor in evaluating the quality of pear fruits. This study used headspace solid-phase microextraction (HS-SPME) combined with gas chromatography-mass spectrometry (GC-MS) to analyze the volatile organic compounds (VOCs) of three crispy pears and five soft pears during shelf life, and the changes in soluble solids content (SSC) were analyzed. The results showed that the SSC of the soft pears such as Nanguoli, Jingbaili and Louis was always higher than that of the crispy pears throughout shelf life. A total of 160 VOCs were detected in the eight pear varieties. Orthogonal partial least squares discriminant analysis (OPLS-DA) and hierarchical cluster analysis (HCA) combined with predictor variable importance projection (VIP) showed that the eight pear varieties could be obviously classified into six groups according to the differences in their VOCs, and 31 differential VOCs were screened out, which could be used to differentiate between pears with different flesh textures. The results of clustering heat map analysis showed that, with the extension of shelf life, the content of each different VOC did not change much in crispy pears, whereas the difference in soft pears was larger. This study confirmed the potential of determining the optimal shelf life of different pear varieties about aroma evaluation and studying the mechanism of differences in VOCs in the future.

Ammonia emissions from a dairy housing and wastewater treatment plant quantified with an inverse dispersion method accounting for deposition loss

ABSTRACT Ammonia (NH3) emissions negatively impact air, soil, and water quality, hence human health and biodiversity. Significant emissions, including the largest sources, originate from single or multiple structures, such as livestock facilities and wastewater treatment plants (WWTPs). The inverse dispersion method (IDM) is effective in measuring total emissions from such sources, although depositional loss between the source and point of measurement is often not accounted for. We applied IDM with a deposition correction to determine total emissions from a representative dairy housing and WWTP during several months in autumn and winter in Switzerland. Total emissions were 1.19 ± 0.48 and 2.27 ± 1.53 kg NH3 d−1 for the dairy housing and WWTP, respectively, which compared well with literature values, despite the paucity of WWTP data. A concurrent comparison with an inhouse tracer ratio method at the dairy housing indicated an offset of the IDM emissions by < 20%. Diurnal emission patterns were evident at both sites mostly driven by changes in air temperature with potential lag effects such as following sludge agitation. Modeled deposition corrections to adjust the concentration loss detected at the measurement point with the associated footprint were 22–28% of the total emissions and the cumulative fraction of deposition to emission modeled with distance from the source was between 7% and 12% for the measurement distances (60–150 m). Although estimates of depositional loss were plausible, the approach is still connected with substantial uncertainty, which calls for future validation measurements. Longer measurement periods encompassing more management activities and environmental conditions are required to assess predictor variable importance on emission dynamics. Combined, IDM with deposition correction will allow the determination of emission factors at reduced efforts and costs, thereby supporting the development and assessment of emission reducing methods and expand the data availability for emission inventories. Implications: Ammonia emissions must be measured to determine emission factors and reporting national inventories. Measurements from structures like farms and industrial plants are complex due to the many different emitting surfaces and the building configuration leading to a poor data availability. Micrometeorological methods provide high resolution emission data from the entire structure, but suffer from uncertainties, as the instruments must be placed at a distance from the structure resulting in a greater loss of the emitted ammonia via dry deposition before it reaches the measurement. This study constrains such emission measurements from a dairy housing and wastewater treatment plant by applying a simple correction to account for the deposition loss and compares the results to other methods.

A Hybrid Decision Tree and Deep Learning Approach Combining Medical Imaging and Electronic Medical Records to Predict Intubation Among Hospitalized Patients With COVID-19: Algorithm Development and Validation.

Early prediction of the need for invasive mechanical ventilation (IMV) in patients hospitalized with COVID-19 symptoms can help in the allocation of resources appropriately and improve patient outcomes by appropriately monitoring and treating patients at the greatest risk of respiratory failure. To help with the complexity of deciding whether a patient needs IMV, machine learning algorithms may help bring more prognostic value in a timely and systematic manner. Chest radiographs (CXRs) and electronic medical records (EMRs), typically obtained early in patients admitted with COVID-19, are the keys to deciding whether they need IMV. We aimed to evaluate the use of a machine learning model to predict the need for intubation within 24 hours by using a combination of CXR and EMR data in an end-to-end automated pipeline. We included historical data from 2481 hospitalizations at The Mount Sinai Hospital in New York City. CXRs were first resized, rescaled, and normalized. Then lungs were segmented from the CXRs by using a U-Net algorithm. After splitting them into a training and a test set, the training set images were augmented. The augmented images were used to train an image classifier to predict the probability of intubation with a prediction window of 24 hours by retraining a pretrained DenseNet model by using transfer learning, 10-fold cross-validation, and grid search. Then, in the final fusion model, we trained a random forest algorithm via 10-fold cross-validation by combining the probability score from the image classifier with 41 longitudinal variables in the EMR. Variables in the EMR included clinical and laboratory data routinely collected in the inpatient setting. The final fusion model gave a prediction likelihood for the need of intubation within 24 hours as well. At a prediction probability threshold of 0.5, the fusion model provided 78.9% (95% CI 59%-96%) sensitivity, 83% (95% CI 76%-89%) specificity, 0.509 (95% CI 0.34-0.67) F1-score, 0.874 (95% CI 0.80-0.94) area under the receiver operating characteristic curve (AUROC), and 0.497 (95% CI 0.32-0.65) area under the precision recall curve (AUPRC) on the holdout set. Compared to the image classifier alone, which had an AUROC of 0.577 (95% CI 0.44-0.73) and an AUPRC of 0.206 (95% CI 0.08-0.38), the fusion model showed significant improvement (P<.001). The most important predictor variables were respiratory rate, C-reactive protein, oxygen saturation, and lactate dehydrogenase. The imaging probability score ranked 15th in overall feature importance. We show that, when linked with EMR data, an automated deep learning image classifier improved performance in identifying hospitalized patients with severe COVID-19 at risk for intubation. With additional prospective and external validation, such a model may assist risk assessment and optimize clinical decision-making in choosing the best care plan during the critical stages of COVID-19.

Predictive geochemical mapping using machine learning in western Kenya

Digital soil mapping techniques represent a cost-effective method for obtaining detailed information regarding the spatial distribution of chemical elements in soils. Machine learning (ML) algorithms using random forest (RF) models have been developed for classification, pattern recognition and regression tasks, they are capable of modelling non-linear relationships using a range of datasets, identifying hierarchical relationships, and determining the importance of predictor variables. In this study, we describe a framework for spatial prediction based on RF modelling where inverse distance weighted (IDW) predictors are used in conjunction with ancillary environmental covariates. The model was applied to predict the total concentration (mg kg−1) and assess the prediction uncertainty of 56 elements, soil pH and organic matter content using 466 soil samples in western Kenya; the results of iodine (I), selenium (Se), zinc (Zn) and soil pH are highlighted in this work. These elements were selected due to contrasting biogeochemical cycles and widespread dietary deficiencies in sub-Saharan Africa, whilst soil pH is an important parameter controlling soil chemical reactions. Algorithm performance was evaluated determining the relative importance of each predictor variable and the model's response using partial dependence profiles. The accuracy and precision of each RF model were assessed by evaluating out-of-bag predicted values. The models R2 values range from 0.31 to 0.64 whilst CCC values range from 0.51 to 0.77. The IDW predictor variables had the greatest impact on assessing the distribution of soil properties in the study area, however, the inclusion of ancillary environmental data improved model performance for all soil properties. The results presented in this paper highlight the benefits of ML algorithms which can incorporate multiple layers of data for spatial prediction, uncertainty assessment and attributing variable importance. Additional research is now required to ensure health practitioners and the agri-community utilise the geochemical maps presented here for assessing the relationship between environmental geochemistry, endemic diseases and preventable micronutrient deficiency.

IDENTIFYING VARIABLES THAT PREDICT STUDENTS’ GEOGRAPHICAL INQUIRY SKILLS DURING THE COVID-19 PANDEMIC

The purpose of this study was to observe the predictive power of the practices carried out in distance geography courses conducted during the Covid-19 pandemic in students’ self-efficacy in geographical inquiry skills. Fourteen variables were determined for this purpose. In this context, questions covering the individual characteristics of the students, systems followed by the students (synchronous-asynchronous), students’ interests in the course and their follow-ups, and their learning experiences were included. “The Self-Assessment Scale for Geographic Inquiry Process Skills” was used to determine the students’ self-efficacy levels. The data were collected from 493 students attending 11th and 12th grades in eighteen high schools in the spring semester of the 2021-2022 academic year. The screening model was used in the research and the data were analyzed using multiple hierarchical regression methods. The results of the study showed that nine variables statistically predicted 89% of the total variance. In order of relative importance, grade, school type, and gender are the first and most important predictor variables. Students’ asking questions, doing homework, and using supplementary materials come next. Based on this, it is recommended that teachers take on the responsibility of raising their students as individuals who are independent and learned to learn.

Estimation of historical daily airborne pollen concentrations across Switzerland using a spatio temporal random forest model

High concentrations of airborne pollen trigger seasonal allergies and possibly more severe adverse respiratory and cardiovascular health events. Predicting pollen concentration accurately is valuable for epidemiological studies, in order to study the effects of pollen exposure. We aimed to develop a spatiotemporal machine learning model predicting daily pollen concentrations at a spatial resolution of 1 × 1 km across Switzerland between 2000 and 2019. Daily pollen concentrations for five common, highly allergenic pollen types (hazel, alder, birch ash, and grasses) were available from fourteen measurement sites across Switzerland. We considered several predictors such as elevation, species distribution, wind speed, wind direction, temperature, precipitation, relative humidity, satellite-observed Normalized Difference Vegetation Index, and land-use (CORINE, Landsat satellite) to explain variation in pollen concentration. We employed feature engineering techniques to encode categorical variables and fill in missing values. We applied a random forest machine learning model with 5-fold cross-validation. The 5th–99th percentiles for concentrations of hazel, alder, birch, ash, and grass pollen at the pollen monitoring stations were 0–298, 0–306, 0–1153, 0–800, and 0–290 pollen grains/m3, respectively. The results of a predictive model for these concentrations yielded overall R2 values of 0.87, 0.84, 0.89, 0.88, and 0.91, and temporal root mean squared errors (RMSEs) of 16.07, 16.72, 69.04, 41.50, and 22.45 pollen grains/m3. An analysis of predictor variable importance indicates that the average national daily pollen concentration is the most important predictor of pollen concentrations for all pollen types. Furthermore, meteorological variables including temperature, total precipitation, humidity, boundary layer height, wind speed, and wind direction, as well as date and satellite features, are important factors in pollen concentration prediction. These spatiotemporal pollen models will serve to estimate individual residential pollen exposure for epidemiological studies. Resulting estimates will enable us to study respiratory and cardiovascular mortality and hospital admissions in Switzerland.

Potential of circulating receptor-interacting protein kinase 3 levels as a marker of acute liver injury

The pathogenesis of acute liver failure (ALF) involves cell death. Necroptosis is a newly suggested programmed cell death, and receptor-interacting protein kinase 3 (RIPK3) has been reported as a marker for necroptosis. However, there are few reports on necroptosis in ALF. Therefore, we evaluated the role of cell death markers such as cytokeratin (CK) 18, cleaved CK (cCK) 18, and RIPK3 in ALF, as well as cytokines and hepatocyte growth factor (HGF). Seventy-one hospitalized patients with acute liver injury (38 nonsevere hepatitis [non-SH]/22 severe hepatitis [SH]/11 ALF) were studied. No significant difference was found for cytokines, but a substantial increase in HGF levels was found following the severity of hepatitis. The non-SH group had lower levels of CK18 and cCK18 than the SH/ALF group. RIPK3 was significantly lower in the non-SH/SH group than in the ALF group. HGF, RIPK3, and albumin levels were found to be important predictive variables. The present study suggests that cCK18, CK18, and RIPK3 are associated with the severity of hepatitis. RIPK3 and other markers related cell death may be useful for understanding the pathogenesis of ALF and as a prognostic marker of acute liver injury.