Important Predictor Variables Research Articles

Gdaphen, R pipeline to identify the most important qualitative and quantitative predictor variables from phenotypic data

BackgroundIn individuals or animals suffering from genetic or acquired diseases, it is important to identify which clinical or phenotypic variables can be used to discriminate between disease and non-disease states, the response to treatments or sexual dimorphism. However, the data often suffers from low number of samples, high number of variables or unbalanced experimental designs. Moreover, several parameters can be recorded in the same test. Thus, correlations should be assessed, and a more complex statistical framework is necessary for the analysis. Packages already exist that provide analysis tools, but they are not found together, rendering the decision method and implementation difficult for non-statisticians.ResultWe present Gdaphen, a fast joint-pipeline allowing the identification of most important qualitative and quantitative predictor variables to discriminate between genotypes, treatments, or sex. Gdaphen takes as input behavioral/clinical data and uses a Multiple Factor Analysis (MFA) to deal with groups of variables recorded from the same individuals or anonymize genotype-based recordings. Gdaphen uses as optimized input the non-correlated variables with 30% correlation or higher on the MFA-Principal Component Analysis (PCA), increasing the discriminative power and the classifier’s predictive model efficiency. Gdaphen can determine the strongest variables that predict gene dosage effects thanks to the General Linear Model (GLM)-based classifiers or determine the most discriminative not linear distributed variables thanks to Random Forest (RF) implementation. Moreover, Gdaphen provides the efficacy of each classifier and several visualization options to fully understand and support the results as easily readable plots ready to be included in publications. We demonstrate Gdaphen capabilities on several datasets and provide easily followable vignettes.ConclusionsGdaphen makes the analysis of phenotypic data much easier for medical or preclinical behavioral researchers, providing an integrated framework to perform: (1) pre-processing steps as data imputation or anonymization; (2) a full statistical assessment to identify which variables are the most important discriminators; and (3) state of the art visualizations ready for publication to support the conclusions of the analyses. Gdaphen is open-source and freely available at https://github.com/munizmom/gdaphen, together with vignettes, documentation for the functions and examples to guide you in each own implementation.

Comparative performance of Sentinel-2 MSI and Landsat-8 OLI data in canopy cover prediction using Random Forest model: Comparing model performance and tuning parameters

Quantifying canopy cover using Random Forest (RF) model’s appropriate tuning parameters value and sensor based predictor variables is always challenging, especially in fragmented dry deciduous forests. Therefore, this study was designed to compare the performances of Sentinel-2 and Landsat-8 based models using the RF model for predicting canopy cover with assessing variables’ relative importance and correlation. Sentinel-2 and Landsat-8 based bands and spectral indices were used as predictor variables. We compared different mtry, ntree and bag fraction values of the RF model. R-square (R2) and root mean square error (RMSE) were used for comparing the model performance. The results showed that the lowest RMSE value was associated with the default value (predictors/3) or more than the default value of mtry, with bag fraction 0.3–0.7 for Sentinel-2 and 0.3–0.4 for Landsat-8. Model accuracy has increased and stabilized with increase of ntree, and received the lowest RMSE to ntree of more than 1000. Except for SWIR indices based model of Landsat-8, all other Landsat-8 based model’s accuracy was lesser compared to Sentinel-2 based models. Model accuracy of Sentinel-2 based full model (except red edge indices) was marginally better (R2 = 0.899, RMSE = 6.883 %) than Landsat-8 based full model (R2 = 0.886, RMSE = 7.089 %). But with the incorporation of red edge indices, full model RMSE had decreased further from 6.883 % to 6.747 %, and R2 had increased from 0.899 to 0.918. The full model of Sentinel-2 tended to spread variable importance among more variables, but the full model of Landsat-8 slightly tends to concentrate variable importance with fewer variables. However, SWIR bands and indices were the most important predictor variables and highly correlated with canopy cover. These findings can solve the parameter value choice of RF model, and the use of the Sentinel-2 based model will be superior to Landsat-8 based model.

On the value of expert knowledge in estimation and forecasting of solar photovoltaic power generation

Reliable estimates and forecasts of Photovoltaic (PV) power output form a fundamental basis to support its large-scale integration. This is recognized in literature, where a growing amount of studies deal with the development of PV power estimation and forecasting models. In particular, machine learning techniques received significant attention in the past decade. Yet, the importance of predictor variables are consistently ignored in such developments and as a result those models fail to acknowledge the value of including physics-based models. In this study we quantify the value of predictor variables for PV power estimation and forecasting, assess deficiencies in estimation and forecasting models, and introduce a number of pre-processing steps to improve the overall estimation or forecasting performance. To this end, we use common physical models to create so-called expert variables and test their impact on the performance of single-point and probabilistic models. In addition, we investigate the optimal selection of predictor variables for PV power estimation and forecasting. By means of a sensitivity analysis, the paper shows how the value of expert variables is affected by the tilt angle of the PV system. To allow for a deeper insight into the importance of predictor variables, two case studies in different climate regions are considered in the numerical evaluation.

New insights into the benthic macrofauna composition and structure in a southern-west Mediterranean coastal lagoon after restoration actions: Spatial and Seasonal patterns

The Marchica lagoon (Morocco, Alboran Sea), one of the largest lagoons in the Mediterranean, was subject to various restoration actions in 2011 aiming at enhancing its biodiversity and ecosystem services. Here, benthic macrofauna was investigated (28 stations sampled seasonally in 2015) to assess the functioning on the lagoon after these restorations actions. The ultimate aim was to investigate the spatial and seasonal variation of the benthic macrofauna, the environmental drivers, and the potential effects of the recent bioinvasions. Overall, 131 taxa were identified among which two non-indigenous species (NIS) were recorded: Caprella scaura and Paracerceis sculpta. Multivariate analysis of abundances (cluster analysis and non-metric multi-dimensional scaling) allowed the identification of three main groups of stations totalizing seven macrofaunal assemblages. The species which are dominant and/or discriminate each assemblage are Cymodoce truncata, Elasmopus rapax, Ericthonius brasiliensis, Gammarus insensibilis, Heteromastus sp., Aonides oxycephala, Nereis sp., Loripes orbiculatus, P. sculpta and C. scaura. The macrofaunal assemblage composition are subject to seasonal changes, supported by significant differences in several univariate measures, including total number of individuals, number of taxa, and Shannon–Wiener’s diversity index. Furthermore, the Marchica lagoon showed high values of macrofaunal species richness further supporting the idea that the restoration measures had positive consequences. The distribution of the macrobenthic fauna in the study area was affected by a complex of various factors, with depth, temperature, salinity, sediment characteristics and vegetation biomass being the most important predictor variables. Macrofaunal assemblages significantly correlated with environmental variables; and exhibited a “High” to “Good” status in terms of the impact of alien species according to the Alien Biotic Index. However, the high abundance of NIS detected suggests caution about their impact and further spread.

Estimating the mean cutting force of conical picks using random forest with salp swarm algorithm

Conical picks are widely used as cutting tools in shearers and roadheaders, and the mean cutting force (MCF) is one of the important parameters affecting conical pick performance. As MCF depends on a number of parameters and due to that the existing empirical and theoretical formulas and numerical modelling are not sufficient enough and reliable to predict MCF in a proficient manner. So, in this research, a novel intelligent model based on a random forest algorithm (RF) and a heuristic algorithm called the salp swarm algorithm (SSA) have been applied to determine the optimal hyper-parameters in RF, and root mean square error is used as a fitness function. A total of 188 data samples including 50 rock types and seven parameters (tensile strength of the rock σt, compressive strength of the rock σc, cone angle θ, cutting depth d, attack angle γ, rake angle α and back-clearance angle β) were collected to develop an SSA-RF model for mean cutting force prediction. The prediction results of the SSA-RF model were compared with seven influential formulas and four classical models, such as random forest, extreme learning machine, support vector machine and radial basis function neural network. The mean absolute error (MAE), root mean square error (RMSE), mean absolute percentage error (MAPE) and Pearson correlation coefficient (R2) were employed as evaluation indexes to compare the capability of different predicting models. The MAE (0.509 and 0.996), RMSE (0.882 and 1.165), MAPE (0.146 and 0.402) and R2 (0.975 and 0.910) values between measured and predicted MCF for training and testing phases of the SSA-RF model clearly demonstrate the superiority in prediction compared to the other tools. A sensitivity analysis has also been performed to understand the influence of each input parameter on MCF, which indicates that σc, d and σt are the most important variables for MCF prediction.

Predicting the occurrence of an endangered salamander in a highly urbanized landscape

Effective protection of threatened species living in highly urbanized landscapes requires detailed information on their population distribution. For species that are difficult to detect, species distribution models (SDMs) can be valuable tools for predicting their occurrence. We created an SDM to predict breeding pond locations of the endangered Jefferson salamander Ambystoma jeffersonianum in southern Ontario, the most highly developed and populated region in Canada. Using a maximum entropy modelling algorithm (Maxent), we combined known breeding pond occurrences with climate, land type, soil, and topography data to capture the ecological niche of the Jefferson salamander. Our SDM showed excellent performance (AUC = 0.919), with land type being the most important predictor variable. We produced a continuous habitat suitability map that predicted most hotspots of suitable habitat to occur along the Niagara Escarpment, with small patches in hedge rows and forest fragments in surrounding agricultural areas. Our refined presence-absence map predicted a high suitability area of 305 km2 with high specificity and moderate overall accuracy. Over half of this area was within the Ontario Greenbelt, demonstrating the importance of protecting this land from future development. Our work demonstrates how SDMs can be used to inform decisions on endangered species and direct conservation efforts towards critical habitats.

A data-driven on-site injury severity assessment model for car-to-electric-bicycle collisions based on positional relationship and random forest

<abstract> <p>Vulnerable road users (VRUs) are usually more susceptible to fatal injuries. Accurate and rapid assessment of VRU injury severity at the accident scene can provide timely support for decision-making in emergency response. However, evaluating VRU injury severity at the accident scene usually requires medical knowledge and medical devices. Few studies have explored the possibility of using on-site positional relationship to assess injury severity, which could provide a new perspective for on-site transportation professionals to assess accident severity. This study proposes a data-driven on-site injury severity assessment model for car-to-electric-bicycle accidents based on the relationship between the final resting positions of the car, electric bicycle and cyclist at the accident scene. Random forest is employed to learn the accident features from the at-scene positional relationship among accident participants, by which injury severity of the cyclist is assessed. Conditional permutation importance, which can account for correlation among predictor variables, is adopted to reflect the importance of predictor variables more accurately. The proposed model is demonstrated using simulated car-to-electric-bicycle collision data. The results show that the proposed model has good performance in terms of overall accuracy and is balanced in recognizing both fatal and non-fatal accidents. Model performance under partial information confirms that the position information of the electric bicycle is more important than the position information of the cyclist in assessing injury severity.</p> </abstract>

The Quality of Life of Seniors with Eye Diseases during COVID-19.

To assess health-related quality of life (HRQoL), vision-related quality of life (VRQoL), depression and anxiety symptoms, and social support and community integration of seniors with eye diseases and to identify important predictor variables of the outcomes. A cross-sectional survey was performed in seniors with eye diseases (n = 90). Demographic characteristics and questionnaire scores were summarized. Linear regression analysis with backward stepwise selection was used to predict the value of the outcomes of eye disease. Preference-based HRQoL of the study patients with eye diseases during the pandemic was likely good with a mean utility value of 0.88. VRQoL and sleep quality appeared to be good as well. Depression and anxiety symptoms appeared to be low, while community integration and social support were moderate. Furthermore, the presence of retinal disease, number of nonocular comorbidities, and education appeared to have significant negative effects on social support and community integration. The presence of retinal disease and the number of nonocular comorbidities both appeared to negatively impact VRQoL. The use of a mobility aid appeared to negatively affect depressive symptoms and sleep quality. Overall quality of life and wellness among seniors with eye diseases appeared to be good during the COVID-19 pandemic. However, the presence of retinal disease and the number of nonocular comorbidities both appeared to negatively impact VRQoL and social support and community integration. Education appeared to impact social support and community integration negatively. The use of a mobility aid appeared to negatively affect depressive symptoms and sleep quality.

Spatial modeling of two mosquito vectors of West Nile virus using integrated nested Laplace approximations

AbstractThe abundance of Culex restuans and Culex pipiens in relation to ecological predictors is poorly understood in regions of the United States where their ranges overlap. It is suspected that these species play different roles in spreading West Nile virus (WNV) in these regions, but few studies have modeled these species separately or accounted for spatial correlation using Bayesian methods. We used mosquito surveillance data collected by the Pennsylvania Department of Environmental Protection from 2002 to 2016 and integrated nested Laplace approximations with the stochastic partial differential equation approach to predict C. restuans and C. pipiens abundance in relation to several ecological predictors. We then made a predictive risk surface of abundance for each species at locations that were not sampled. Explanatory variables in the models included ecological variables previously described to be important predictors of the abundance of these mosquito species. Developed habitat, temperature, and precipitation were important predictor variables for the abundance of C. restuans, whereas developed habitat, snow water equivalent, and normalized difference water index were important predictor variables for the abundance of C. pipiens. The abundance of C. restuans had a negative relationship with developed habitat in contrast to C. pipiens abundance, which had a positive relationship with developed habitat. Julian date was modeled as a temporal trend for both species and showed C. restuans to be more abundant from late April through late June and C. pipiens to be more abundant from July through September. A seasonal crossover was observed between these two species on Julian day 185, 4 July. We observed different spatial patterns of abundance in the predictive risk maps of each of the species. Our results indicate that modeling the abundance of these species spatially and separately in regions where these two mosquito vectors coexist can help gain further insight into understanding the epidemiology of WNV in human and susceptible animal populations.

Extreme Rainfall and Flood Risk Prediction over the East Coast of South Africa

Extreme rainfall associated with mid-tropospheric cut-off low (COL) pressure systems affected the entire east coast of South Africa during April 2022, leading to flooding and destruction of homes, electricity power lines, and road infrastructure, and leaving 448 people confirmed dead. Therefore, this study investigated the evolution of the two COLs and their impacts, including the occurrence of extreme rainfall and cold weather over the southeast coast of the country. We analysed observed and reanalysis meteorological data and mapped areas at risk to impacts of flood hazards on the east coast of South Africa. Extreme rainfall (>500 mm) accumulated over 16 days was observed along the east coast, with the amount of rainfall progressively decreasing inland. We found that the rainfall associated with the first COL was significantly enhanced by the interactions between a strong low-level onshore airflow across the Agulhas Current and the coastal escarpment, resulting in deep convection and lifting. An unusual surface cyclone with tropical characteristics developed over the subtropical southwest Indian Ocean, driving onshore southeasterly winds which enhanced low-level convergence. Moreover, the flood risk results revealed that, amongst others, land cover/use (52.8%), elevation (16.8%) and lithology (15.5%) were the most important flood predictor variables in this study. Much of the study area was found to have very low (28.33%), low (31.82%), and moderate (21.66%) flood risk, whilst the high- and very-high-risk areas accounted for only 17.5% of the total land area. Nonetheless, the derived flood risk map achieved an acceptable level of accuracy of about 89.9% (Area Under Curve = 0.899). The findings of this study contribute to understanding extreme rainfall events and the vulnerability of settlements on South Africa’s east coast to flood risk, which can be used towards natural disaster risk reduction.

Analysis of the Correlation Between Healthy Lifestyle Patterns and Work Stress in Psychiatric Nurses

Introduction: This study explores the status of mental healthcare workers in the eastern region with respect to work pressure, health-promoting lifestyle patterns, and their related factors, which can provide an important reference to aid the formulation and development of health promotion strategies for such individuals. Methods: A total of 115 questionnaires were distributed and 112 valid questionnaires were recovered, with a recovery rate of 97.4%. The number of samples required for regression analysis was 99, as calculated using G-power software. The effect of work stress on health-promoting lifestyle patterns (β=-0.342, p<0.001) was negative, indicating that a health-promoting lifestyle was better and less stressful. Results: This accounted for 11.70% of the total variation in the health-promoting lifestyle scores (F=14.164, p<0.001, R=0.342). The important predictive variables of health-promoting lifestyle patterns had the greatest influence on work stress (β=-0.302, p=0.001). Discussion: The results of the comprehensive analysis showed that the standardized score of the health-promoting lifestyle of psychiatric nurses was 60.86 points. The overall health-promoting lifestyle of this study was lower than that of clinical nurses in two regional hospitals in Pingtung (the standardized score index is 64.84). This shows that the health-promoting lifestyle of psychiatric nursing staff is only better than that of public health-related staff in the same nursing work, and its overall health-promoting lifestyle is slightly worse than that of general nursing staff and school nurses. Conclusions: This influence was negative, indicating that the greater the work pressure, the worse the implementation of health-promoting lifestyle patterns. Creating a friendly workplace and implementing workplace health-promoting programs, organizing and encouraging psychiatric caregivers to participate in networking activities, and form families can help reduce mental health care workers’ work pressure, improve health, and promote lifestyle participation thus maintaining physical and mental health status.

115: MEASURING ASSOCIATIONS BETWEEN ELECTRONIC HEALTH RECORD USE AND INPATIENT PEDIATRIC PROVIDER BURNOUT

Introduction: EHRs are associated with improved patient outcomes, but the inefficiency of time spent on EHR may contribute to healthcare provider stress, exhaustion and burnout. EHR use associates with higher prevalence of burnout in Neonatal Intensive Care Units (NICUs), but the relationship between EHR use and burnout in other inpatient settings remains unclear. This study sought to identify subjective and objective markers of EHR use associated with provider burnout in the pediatric inpatient setting. Methods: Attendings, fellows, hospitalists, and nurse practitioners from a quaternary children’s hospital who work primarily in the inpatient setting were included in our study. We collected five months (June-October 2020) of EHR-use and schedule data for these providers. This data was then linked to 2020 annual wellness survey (September-October) data which included a 4-item EHR experience score and the Stanford Professional Fulfillment Index. We evaluated associations using Pearson correlation, lasso regression for variable selection, and mixed effect linear regression. Results: Of 246 eligible providers, 179 (73%) responded to the survey. Average EHR experience score was 2.2±0.8, range 0-4, and average burnout score was 1.1±0.7, range 0-4, with burnout prevalence of 32%. Average number of notes/orders placed was independently associated with the EHR experience scores (coefficient -0.013, p=.04). EHR experience scores were negatively correlated with provider burnout scores (r -0.18, p=.03), but not independently associated after adjustment for work setting and role. EHR task count per patient, division, and role were selected as important predictor variables for burnout via lasso regression. Among these, the Pediatric Intensive Care Unit (PICU) division was independently associated with higher burnout scores (coefficient 0.36, p=.016). None of the EHR use measures were independently associated with burnout. Conclusions: Burnout is prevalent among inpatient providers, and being a PICU provider was independently associated with burnout. Although routinely collected EHR use measures may associate with negative EHR experience, we did not identify independent associations with burnout among this sample of inpatient pediatric providers. Larger studies are needed to investigate this relationship further.

Predictive accuracy of post-fire conifer death declines over time in models based on crown and bole injury.

A key uncertainty of empirical models of post-fire tree mortality is understanding the drivers of elevated post-fire mortality several years following fire, known as delayed mortality. Delayed mortality can represent a substantial fraction of mortality, particularly for large trees that are a conservation focus in western US coniferous forests. Current post-fire tree mortality models have undergone limited evaluation of how injury level and time since fire interact to influence model accuracy and predictor variable importance. Less severe injuries potentially serve as an indicator for vulnerability to additional stressors such as bark beetle attack or moisture stress. We used a collection of 164,293 individual tree records to examine post-fire tree mortality in eight western USA conifers: Abies concolor, Abies grandis, Calocedrus decurrens, Larix occidentalis, Pinus contorta, Pinus lambertiana, Pinus ponderosa, and Pseudotsuga menziesii. We evaluated the importance of fire injury predictors on discriminating between surviving trees versus immediate and delayed post-fire mortality. We fit balanced random forest models for each species using cumulative tree mortality from 1 to 5-years post-fire. We compared these results to multi-class random forest models using first-year mortality, 2-5-year mortality, and survival 5-years post-fire as a response variable. Crown volume scorched, diameter at breast height, and relative bark char height, were used as predictor variables. The cumulative mortality models all predicted trees that died within 1-year of fire with high accuracy but failed to predict 2-5-year mortality. The multi-class models were an improvement but had lower accuracy for predicting 2-5-year mortality. Multi-class model accuracies ranged from 85% to 95% across all species for predicting 1-year post-fire mortality, 42%-71% for predicting 2-5-year mortality, and 64%-85% for predicting trees that lived past 5-years. Our study highlights the differences in tree species tolerance to fire injury and suggests that including second-order predictors such as beetle attack or climatic water stress before and after fire will be critical to improve accuracy and better understand the mechanisms and patterns of fire-caused tree death. Random forest models have potential for management applications such as post-fire harvesting and simulating future stand dynamics.

Species distribution modelling of the endangered Mahogany Glider (Petaurus gracilis) reveals key areas for targeted survey and conservation

AbstractThe Mahogany Glider (Petaurus gracilis) is one of the most endangered marsupials in Australia. Its known distribution is an approximately 120 km strip of fragmented coastal woodland in north‐east Queensland, from north of Townsville to the Tully area. Records are clustered in a number of well‐surveyed areas, with significant areas of lowland habitat unsurveyed. Around 30% of historic records fall in areas that were subsequently cleared for farmland, and ongoing clearing and fragmentation of lowland sclerophyll forest continues within the potential distribution. Resolving the distribution is an urgent requirement to guide conservation but Mahogany Gliders are difficult to detect in the field. Species distribution modelling offers a technique for estimating the fine‐scale distribution and for targeting further field survey and conservation efforts. We used known occurrence records (N = 481) to predict the distribution of Mahogany Gliders across the Wet Tropics bioregion. We used climatic, topographic and other environmental predictors to generate distribution models using Maxent and Random Forest algorithms, each with two bias correction methods. The predictions revealed that many unknown populations may exist within the currently defined distribution and in important areas beyond this (e.g. Hinchinbrook Island). There was reasonable congruence between models, and we include syntheses of the models to present the most likely current distribution. The most important predictor variables across the models were precipitation seasonality (high seasonality), elevation (generally <100 m), soil type (hydrosols) and vegetation type (including Eucalyptus and Melaleuca woodlands). Our results identify core habitat and reveal key areas that require targeted field surveys. Importantly, the predicted suitable habitat is highly fragmented and ongoing conservation efforts need to improve habitat connectivity and limit further fragmentation.

A machine learning method for improving the accuracy of radiation biodosimetry by combining data from the dicentric chromosomes and micronucleus assays

A large-scale malicious or accidental radiological event can expose vast numbers of people to ionizing radiation. The dicentric chromosome (DCA) and cytokinesis-block micronucleus (CBMN) assays are well-established biodosimetry methods for estimating individual absorbed doses after radiation exposure. Here we used machine learning (ML) to test the hypothesis that combining automated DCA and CBMN assays will improve dose reconstruction accuracy, compared with using either cytogenetic assay alone. We analyzed 1349 blood sample aliquots from 155 donors of different ages (3–69 years) and sexes (49.1% males), ex vivo irradiated with 0–8 Gy at dose rates from 0.08 Gy/day to ≥ 600 Gy/s. We compared the performances of several state-of-the-art ensemble ML methods and found that random forest generated the best results, with R2 for actual vs. reconstructed doses on a testing data subset = 0.845, and mean absolute error = 0.628 Gy. The most important predictor variables were CBMN and DCA frequencies, and age. Removing CBMN or DCA data from the model significantly increased squared errors on testing data (p-values 3.4 × 10–8 and 1.1 × 10–6, respectively). These findings demonstrate the promising potential of combining CBMN and DCA assay data to reconstruct radiation doses in realistic scenarios of heterogeneous populations exposed to a mass-casualty radiological event.

A Random Forest Classification Model for Spasticity after Stroke

<h3>Research Objectives</h3> The objective of this analysis was to developed a Random Forest Classification Model to determine which sociodemographic and clinical health characteristics predict the development of post-stroke spasticity (PSS). <h3>Design</h3> Random Forest Classification Modelling with data collected from a retrospective review of patients' medical charts and the National Rehabilitation Reporting System. <h3>Setting</h3> Specialized stroke rehabilitation inpatient and outpatient units from a large, urban centre in Ontario, Canada. <h3>Participants</h3> A total of 922 individuals (102 spastic) attended stroke rehabilitation between July 1, 2015 and June 30, 2020. A synthetic minority oversampling technique was used to generate a dataset of 1632 individuals (408 spastic). <h3>Interventions</h3> None. <h3>Main Outcome Measures</h3> Clinical health data (i.e., number, type and location of stroke, age at stroke), length of stay, LOS); presence of spasticity, presence of aphasia, endovascular or tissue plasminogen activator treatment, Chedoke-McMaster impairment inventory scores, Functional Independence Measure motor and cognitive sub-scores. <h3>Results</h3> Based on the results of a feature hyper-parameterization procedure, a model with 1501 trees was selected with 8 variables (mtry=8). A random forest model was then trained using these parameters on 80% of the dataset. The relative importance of predictor variables was assessed by examining the decrease in model accuracy when a particular variable is permutated. Chedoke McMaster impairment scores, age at stroke onset, Functional Independence Measure motor sub-scores and length of stay appear to be the most important predictor variables when compared to the remaining 7. When testing the model performance on the untrained 20% of the dataset, the model performed with a balanced accuracy of 82.5%, a Cohen's kappa value of 0.71 and a spasticity classification error of 32.1%. <h3>Conclusions</h3> Overall, our model performed relatively well, but could be tuned more for sensitivity to prioritize the correct identification of spasticity. In the future, more data will be required to validate the generalizability of the model and robustness of fit, as well as to eliminate the need for synthetically created data. <h3>Author(s) Disclosures</h3> None.

Confirmation of the southern African distribution of the marine sponge Hymeniacidon perlevis (Montagu, 1814) in the context of its global dispersal.

Intertidal rocky shore surveys along the South African coastline (∼3,000 km) have demonstrated the presence and abundance of the encrusting orange sponge Hymeniacidon perlevis (Montagu, 1814), a well-known globally distributed species. After analysing the southern African populations, we gained a better understanding of the genetic structure of this now-accepted global species. Apart from confirming the presence of a single population of H. perlevis, we also determined its distribution in the southern African intertidal rocky shore ecosystem, compared its genetic diversity to congeners, predict its global distribution via environmental niche modelling, and discussed possible underlying mechanisms controlling the species' global distribution. We surveyed the South African coastline and sampled sponges at 53 rocky shore sites spanning over 3,000 km, from Grosse Bucht south of Lüderitz (Namibia) to Kosi Bay on the east coast of South Africa. DNA sequences of the nuclear rDNA internal transcribed spacer (ITS1) and the COI mitochondrial gene were obtained from 61 samples and compared them to a world-wide sample of other H. perlevis sequences. Using environmental predictor variables from the global dataset BIO-ORACLE, we predicted the probability of global occurrence of the species using an ensemble of eight distribution models. South African specimens were found to be 99-100% identical to other populations of H. perlevis (=H. sinapium) from other world-wide regions. The presence of a single population of H. perlevis in southern Africa is supported by genetic data, extending its distribution to a relatively wide geographical range spanning more than 4,000 km along the temperate southern African coast. The predicted global occurrence by ensemble model matched well with the observed distribution. Surface temperature mean and range were the most important predictor variables. While H. perlevis appears to have been introduced in many parts of the world, its origins in Europe and southern Africa are unclear.

Predictive mapping of wetland soil types in the Canadian Prairie Pothole Region using high-resolution digital elevation model terrain derivatives

Wetland soil types, which can be distinguished based on calcium carbonate content, vary in their effect on ecosystem functions like phosphorus retention, salinity contributions, and greenhouse gas forcing. These soil types may be predictively mapped with machine learning models that use terrain derivatives calculated from high-resolution digital elevation models. Soil profiles from three Saskatchewan study sites were classified into three functional categories—upland, calcareous wetland, or noncalcareous wetland—and used to train random forest models for predictive soil mapping. Multiple terrain derivatives were included as predictor variables to capture local- and landscape-scale morphometry and hydrology influences, including five derivatives developed for this study. Models were developed at three spatial resolutions: 2, 5, and 10 m, and tested via internal cross-validation and independent validation with datasets from previous studies. Predictive accuracies were highest when mapping at 2 m resolution (independent validation accuracy range = 64%–100%) but also successful when mapping at 5 and 10 m resolutions (independent validation accuracy range = 63%–100%); however, visual inspection determined that the maps generated at 10 m resolution were less detailed and occasionally featured questionable discontinuous soil distributions. Three of the five terrain derivatives developed for this study were among the most important predictor variables (first, second, and 10th most important). Models trained using only data from a specific site had slightly better performance than models trained using data from all sites, except in regions where training data were lacking.

The role of extracranial carotid duplex in predicting functional outcome in first time lacunae strokes

ABSTRACT Background Stroke is the third largest cause of death both worldwide and in Taiwan. Among the various stroke subtypes, lacunae strokes account for 20 to 30% of the total stroke population. Through vigorous risk control factors, the effective prevention and the long-term functional outcome remains are yet to be investigated. Carotid duplex is a non-invasive neuroimaging modality that is routinely applied to stroke patients. In the current research, we correlate baseline carotid duplex examination parameters with functional outcome assessment in an attempt to ascertain important variables for early outcome prediction. Method We retrospectively recruited lacunae ischemic stroke patients from the Department of Neurology, Changhua Christian Hospital. Patient baseline demographics, biochemistry, neuroimaging, and outcome assessments were thoroughly evaluated via a modified Rankin Scale (mRS), NIHSS, and Barthel index scores, both prior and subsequent to patient discharge for a period of up to one year. Patients were divided into two groups based on their functional recovery status. This current study utilizes the Kaplan–Meier method to draw the survival curve and adopts the Cox proportional hazards model to estimate the hazard ratio between the two groups. The risk modeling method is also applied to formulate the best fit in seeking and identifying the variables, which has the most impact on determining theoutcome predictor. All statistical analyses were performed using R for Windows (version 3.6.3), with p < 0.05 meaning statistical significance and 0.05 < p < 0.1 denoting marginal statistical significance. Results One hundred forty-one ischemic lacunae stroke patients are included in the study, of which the follow-ups of 69 patients showed no physical functionality improvement, whereas 72 patients demonstrated favorable functional outcomes. In terms of underlying diseases, more than 80% of the patients suffered from hyperlipidemia upon hospitalization and more than 40% of the patients suffered from diabetes mellitus and heart disease. There was a significant difference in pulmonary disease between improvement and non-improvement in disability (1.4% vs 10.1%, p = 0.034). The average age of the improved group was lower than that of the non-improved group (65.5 vs. 72, p = 0.014), and the WBC was higher (8 vs 6.95, p = 0.005); furthermore, higher NIHSS scores (6.5 vs 3, p = 0.001) and lower Barthel index scores (47.5 vs 80, p < 0.001) were also noted. The group with improved mRS required a longer hospital stay (with a median of 10 days), while their un-improved counterpart required 8 days (p < 0.05). The majority of patients were discharged within 30 days, and there was a slight difference between the two survival curves (marginal statistical significance, 0.05 < p < 0.1). The hazard ratio of the improved group was 0.7188. There was a difference in the end diastolic velocity (EDV) of common carotid artery (CCA) between the lesion and non-lesion sides of the patients (p < 0.05). Risk-stratifying models indicate that diabetes mellitus, peak systolic velocity (PSV) of the internal carotid artery (ICA) on the lesion side, the diameter of the ICA on the non-lesion side, and the pulsatility index (PI) of the ICA and external carotid artery (ECA) were selected as the most important factors affecting the end diastolic velocity (EDV) of the common carotid artery (CCA) in all four models. Conclusion In the current study, the most suitable length of stay in hospital for lacunae stroke patients to fully recuperate is ten days. Additionally, CCA EDV might be the key determinant, with baseline diabetes mellitus acting in a medium role, capable of altering long-term outcome functionality and recovery.

Comparison of machine learning models for bluetongue risk prediction: a seroprevalence study on small ruminants

BackgroundBluetongue (BT) is a disease of concern to animal breeders, so the question on their minds is whether they can predict the risk of the disease before it occurs. The main objective of this study is to enhance the accuracy of BT risk prediction by relying on machine learning (ML) approaches to help in fulfilling this inquiry. Several risk factors of BT that affect the occurrence and magnitude of animal infection with the virus have been reported globally. Additionally, risk factors, such as sex, age, species, and season, unevenly affect animal health and welfare. Therefore, the seroprevalence study data of 233 apparently healthy animals (125 sheep and 108 goats) from five different provinces in Egypt were used to analyze and compare the performance of the algorithms in predicting BT risk.ResultsLogistic regression (LR), decision tree (DT), random forest (RF), and a feedforward artificial neural network (ANN) were used to develop predictive BT risk models and compare their performance to the base model (LR). Model performance was assessed by the area under the receiver operating characteristics curve (AUC), accuracy, true positive rate (TPR), false positive rate (FPR), false negative rate (FNR), precision, and F1 score. The results indicated that RF performed better than other models, with an AUC score of 81%, ANN of 79.6%, and DT of 72.85%. In terms of performance and prediction, LR showed a much lower value (AUC = 69%). Upon further observation of the results, it was discovered that age and season were the most important predictor variables reported in classification and prediction.ConclusionThe findings of this study can be utilized to predict and control BT risk factors in sheep and goats, with better diagnostic discrimination in terms of accuracy, TPR, FNR, FPR, and precision of ML models over traditional and commonly used LR models. Our findings advocate that the implementation of ML algorithms, mainly RF, in farm decision making and prediction is a promising technique for analyzing cross-section studies, providing adequate predictive power and significant competence in identifying and ranking predictors representing potential risk factors for BT.