Test AUC Research Articles

249-OR: A Randomized, Double-Blind Phase 2 Trial of the CXCR1/2 Inhibitor Ladarixin in Adult Patients with New-Onset Type 1 Diabetes

In preclinical studies, the investigational CXCR1/2 inhibitor ladarixin was shown to prevent and reverse diabetes in NOD mice. On this basis, a multicenter clinical trial in adult patients with new onset type 1 diabetes (T1D) was conducted (NCT02814838). Patients (Table 1) were randomized 2:1 to ladarixin (LDX) 400 mg BID or placebo (PLA) and randomized within 100 days from first insulin administration. Patients received 3 treatment cycles, each consisting of 14 days on and 14 days off drug, and were then followed up for 12 months. The difference between LDX and PLA in mixed meal tolerance test (MMTT) AUC of C peptide was statistically significant neither at week 13 (primary endpoint, p=0.330) nor at week 26 (p=0.404). Conversely, at week 26 patients having HbA1c<7% and daily insulin requirements <0.50 IU/kg were 76.6% and 45.8% in LDX and PLA, respectively (p=0.0095). In patients with baseline fasting C-peptide below the median value of trial population (prespecified subgroup analysis), the difference between LDX and PLA in MMTT AUC of C peptide trended at week 13 (p=0.111) and reached statistical significance at week 26 (p=0.041). Adverse Reactions occurring in ≥10% of patients included dyspepsia (16% LDX vs. 0% PLA) and headache (16% LDX vs. 15.4% PLA). Although primary endpoint was not met, secondary and subgroup analyses suggest further investigation of LDX in new onset T1D may be warranted. Disclosure E. Bosi: Consultant; Self; Abbott, Eli Lilly and Company, Medtronic, Novartis AG, Roche Diabetes Care, Sanofi. P. Pozzilli: Advisory Panel; Self; Abbott, AstraZeneca, Eli Lilly and Company. Research Support; Self; Medtronic, Sanofi. F. Giorgino: Advisory Panel; Self; AstraZeneca, Eli Lilly and Company, Novo Nordisk Inc., Roche Diabetes Care, Sanofi. Consultant; Self; AstraZeneca, Boehringer Ingelheim Pharmaceuticals, Inc., LifeScan, Inc., MedImmune, Merck Sharp & Dohme Corp., Roche Diabetes Care, Sanofi. Research Support; Self; Eli Lilly and Company, LifeScan, Inc., Takeda Development Centre Europe Ltd. E. Cossu: None. T. Linn: Advisory Panel; Self; Dompe Pharmaceuticals. L. Rose: None. B. Keymeulen: None. P. Gillard: None. L. Daffonchio: Employee; Self; Domp farmaceutici S.p.A. P.A. Ruffini: Employee; Self; Domp farmaceutici S.p.A., Domp farmaceutici S.p.A. M. Allegretti: Employee; Self; Domp Farmaceutici spa. L. Piemonti: Advisory Panel; Self; AstraZeneca, Roche Diabetes Care, Semma Therapeutics, Inc.

MP50-06 UTILIZATION OF RADIOMIC MARKERS IN LOCALIZED RENAL CELL CARCINOMA TO PREDICT STAGE

INTRODUCTION AND OBJECTIVE: Incidental pT3a upstaging following partial nephrectomy (PN) for cT1 renal masses is associated with inferior oncologic outcomes. The ability to identify tumor characteristics associated with pathologic upstaging using current imaging protocols is limited. The inclusion of quantitative radiomic imaging characteristics in pre-operative staging may improve the detection of adverse pathologic features. We sought to evaluate whether any radiomic features were associated with pathologic upstaging following PN for cT1 renal masses. METHODS: Pre-operative CT imaging from a total of 130 patients with cT1 renal masses and final pathology demonstrating either pT1 (n=92) or pT3a (n=45) renal cell carcinoma were retrospectively reviewed. T1 patients were matched by age and tumor size with pT3a patients in a 2:1 ratio. An Auto-Initialized Cascaded Level Set (AI-CALS) system was used to segment 139 renal lesions. 49 radiomic descriptors including morphological features (volume, shape) and gray level features were extracted from segmented lesions. A predictive model was developed using automatic feature selection and linear discriminant analysis classifier. A leave-one-lesion out cross validation method was used for model training and performance testing. The area under the receiver operating characteristic curve (AUC) was calculated for the predictor model to evaluate its performance in predicting cancer stage (pT1 or pT3a). RESULTS: We identified one gray level and 4 shape features of significance in our predictive model. Features related to surface irregularity of the renal mass appeared to be highly predictive in distinguishing between pT1a and pT3a cancers. The training AUC was 0.72 and the test AUC was 0.70. The results from the test AUC were graphed. The AUC for our training set was 0.72 dropping only slightly to 0.70 in our validation set (figure). CONCLUSIONS: Machine learning techniques show promise in decision support for renal cell carcinoma staging. The ability to predict pathologic stage prior to intervention could help guide treatment selection or identify patients for future neoadjuvant trials. Future work will focus on tool and feature optimization and incorporating clinical and biopsy data into the predictive model.Source of Funding: None

MP75-17 ETHNIC VARIATION IN PROSTATE CANCER DETECTION

INTRODUCTION AND OBJECTIVE: The Stockholm3 test has not been validated in an American cohort where clinical pathways to biopsy and race differ. We aimed to match clinical data stratified by race and location in contemporary cohorts from both Chicago and Stockholm, to compare the detection of clinically significant prostate cancer (csPCa) (ISUP Grade Group 2 or higher). Secondarily, we aimed to observe the Stockholm3 test performance in a subgroup of Swedish men matched to the American cohort. METHODS: All men at UIC age 49-71 years who underwent prostate biopsies from June 2016 to July 2019 were evaluated. Propensity scores (PS) were estimated for each person using logistic regression for age, PSA, prostate volume, family history of PCa, use of 5-alpha reductase inhibitor, and a prior negative biopsy with stratification for location and race. Matching was performed on the logit of the estimated PS using caliper width 0.20 times the standard deviation of the logit of the PS and using a 3:1 nearest neighbor algorithm with replacement. Standardized mean difference was used to check for covariate balance. Area under the receiver operator curve (AUC) of Stockholm3 for detection of csPCa on biopsy was assessed in all sub-cohorts. Logistic regression models were used to compare detection rates of csPCa. RESULTS: A total of 504 from Chicago and 6,980 men from STHLM3 were included. The Chicago cohort was 59.5% African American (AA), 21.3% Hispanic, and 12.6% Caucasian. In AA men, 51.2% of men had csPCa detected, while in the matched STHLM3 men, 33.6% of men had csPCa detected (p <0.001). Whereas there was no difference in csPCa detected when matching Chicago Caucasian or Hispanic men (p = 0.57, 0.95). The AUC for the Stockholm3 test of the entire STHLM3 cohort was 0.78, whereas the matched STHLM3 cohort for AA men was 0.85. When matching covariates of AA men in Chicago to Caucasian men in Chicago the detection of csPCa was 40.6% vs. 21.7% (p = 0.01). CONCLUSIONS: STHLM3 differs in that it is a screening by invitation cohort, explaining the low prevalence of csPCa. However, there was a higher detection of csPCa in AA men when matched to Caucasian men from both Stockholm and Chicago. In all matched sub-cohorts of Swedish men to the Chicago, the Stockholm3 test AUC was increased. Studies are currently underway validating the Stockholm3 test in an American population.Source of Funding: None

Beta‐cell‐specific Loss of the Inhibitory G protein, Gαz, has Sex‐dependent Effects on Development and Pathophysiology of Type 1 Diabetes

The mechanisms that underlie the beta‐cell pathophysiology of Type 1 Diabetes (T1D) are not fully understood. Our lab has discovered a role for the unique inhibitory G‐protein, Gαz, in beta‐cell function, survival, and replication in both normal conditions and those of pathological stress. Gαz, when acting as a canonical inhibitory G‐protein, reduces production of cAMP through inhibition of adenylate cyclase. Cyclic AMP acts as essential player in the amplification of glucose‐stimulated insulin secretion, while also mediating pathways that regulate beta‐cell death and replication. Previous work in our lab has shown that non‐obese diabetic (NOD) mice with global loss of Gαz are protected from developing T1D‐like hyperglycemia through increased beta‐cell function, increased beta‐cell replication, and decreased beta‐cell death. In order to determine if Gαz in the beta‐cell was solely responsible for the protected phenotype, we generated a beta‐cell‐specific Gαzknockout mouse by breeding GzFlox/Flox mice with Rat Insulin Promoter (RIP) driven CreHerrmice, on the NOD background for more than 10 generations. Experimental mice were sNP genotyped to ensure preservation of NOD loci. Weekly random‐fed blood glucose measurements were taken beginning at weaning. Loss of beta‐cell Gαz(GαzβKO), only provided a modest protection against diabetes development; 69% of Gαz βKO females remained euglycemic over the first 30 weeks of life compared to 59% in Cre+ NOD controls. Consistent with this phenotype, neither pancreases from Gαzβ‐cell‐KO males nor females had increased beta‐cell fractional area as compared to Cre+ controls. Overall, both euglycemic Gαzβ‐cell‐KO males and females had improved glucose tolerance at 12 weeks of age, with this effect persisting at 28 weeks of age only in females. Yet, the mean insulitis score at 28 weeks‐of‐age was dramatically reduced in GαzβKO males only, and this score was strongly correlated with 28‐week glucose tolerance test AUC. The effects of Gαzβ‐cell‐KO on plasma insulin and ex vivo islet insulin secretion are also differentially regulated by sex: effects that appear to correlate with baseline expression of the constitutively‐active splice variant of the Gαz coupled EP3 receptor. Finally, changes in intra‐islet hormone production and secretion in full‐body vs. beta‐cell‐specific Gαz‐KO mice provide a potential mechanism for the incomplete protection from hyperglycemia in Gαzβ‐cell‐KO mice. Overall, our results show a divergent, sex‐dependent role for beta‐cell Gαz in T1D pathophysiology and implicate Gαz as a hub of intra‐islet cell‐cell signaling.

Development and validation of a pancreatic cancer prediction model from electronic health records using machine learning.

679 Background: Pancreatic Adenocarcinoma (PDAC) is often diagnosed at an advanced stage. We sought to develop a model for early PDAC prediction in the general population, using electronic health records (EHRs) and machine learning. Methods: We used three EHR datasets from Beth-Israel Deaconess Medical Center (BIDMC) and Partners Healthcare (PHC): 1. “BIDMC-Development-Data” (BIDMC-DD) for model development, using a feed-forward neural network (NN) and L2-regularized logistic regression,randomly split (80:20) into training and test groups. We tuned hyperparameters using cross-validation in training, and report performance on the test split. 2. “BIDMC-Large-Data” (BIDMC-LD) to re-fit and calibrate models. 3. “PHC-Data” for external validation. We evaluate using Area Under the Receiver Operating Characteristic Curve (AUC) and compute 95% CI using empirical bootstrap over test data. PDAC patients were selected using ICD9/-10 codes and validated with tumor registries. In contrast to prior work, we did not predefine feature sets based on known clinical correlates and instead employed data-driven feature selection, specifically importance-based feature pruning, regularization, and manual validation, to identify diagnostic-based features. Results: BIDMC-DD included demographics, diagnoses, labs and medications for 1018 patients (cases = 509; age-sex paired controls). BIDMC-LD included diagnoses for 547,917 patients (cases = 509), and PHC included diagnoses for 160,593 patients (cases = 408). We compared our approach to adapted and re-fitted published baselines. With a 365-day lead-time, NN obtained a BIDMC-DD test AUC of 0.84 (CI 0.79 - 0.90) versus the previous best baseline AUC of 0.70 (CI 0.62 - 0.78). We also validated using BIDMC-DD’s test cancer patients and BIDMC LD controls. The AUC was 0.71 (CI 0.67 - 0.76) at the 365-day cutoff. NN’s external validation AUC on PHC-Data was 0.71 (CI 0.63 - 0.79), outperforming an existing model’s AUC of 0.61 (CI 0.52 - 0.70) (Baecker et al, 2019). Conclusions: Models based on data-driven feature selection outperform models that use predefined sets of known clinical correlates and can help in early prediction of PDAC development.

Predicting potential current distribution of Lycorma delicatula (Hemiptera: Fulgoridae) using MaxEnt model in South Korea

Abstract Lycorma delicatula (Hemiptera: Fulgoridae) is an invasive insect in Korea which causes plant damages by sucking and sooty molds. Lycorma delicatula was first detected in South Korea in 2004, where its introduction and spreading possibly were affected by human activity-related factors. Here, we used MaxEnt to describe current distribution of L. delicatula in Korea and tried to find out the impact of human influences on distribution. We used 143 sites of occurrence data, 19 bioclimatic variables, duration of temperature below −11 °C, average daily minimum temperature in January, cumulative thermal unit variable, the distribution of grape orchard variable and human footprint to create models. These models were estimated by two sets of 24 candidates with feature combinations and regularization multipliers. In addition, these two sets were created as models with and without footprint to assess human influence on distribution. Model selection for optimal model was performed by selecting a model with a lowest sum of each rank in small sample-size corrected Akaike’s information criterion and difference between training and test AUC. Model of LQ10 parameter combinations was selected as optimal models for both model sets. Consequently, both of distribution maps from these models showed similar patterns of presence probability for L. delicatula. Both models expected that low altitude regions were relatively more suitable for L. delicatula than mountain areas in Korea. Footprint might be limited for the distribution and L. delicatula might already occupy most of available habitats. Human-related factors might contribute to spread of L. delicatula to uninfected areas.

Ensemble of deep learning models to predict platinum resistance in high grade serous ovarian cancer

Deep learning has benefits to find complicated interactions between biological entities and could make useful estimation. In clinical practice, the prediction of platinum resistance in ovarian cancer is an important problem because it alters treatment options for patients and subsequently their quality of life. In this paper, Deep Neural Network (DNN) models are designed and evaluated with several feature selection approaches. Among the feature selection approaches, genes selected by group difference in gene expression showed the best performance of 0.838 in test accuracy and 0.889 in test AUC. Hybrid ensemble approaches displayed a performance of 0.903 in test accuracy and 0.869 in test AUC. An alternative hybrid ensemble model with removed partially sensitive samples displayed the performance of 0.903 in test accuracy and 0.914 in test AUC. These results suggest that a hybrid ensemble approach could help prediction of platinum resistance in ovarian cancer and subsequent treatment practice in clinics.

2019 Novel Coronavirus-Infected Pneumonia on CT: A Feasibility Study of Few-Shot Learning for Computerized Diagnosis of Emergency Diseases.

COVID-19 is an emerging disease with transmissibility and severity. So far, there are no effective therapeutic drugs or vaccines for COVID-19. The most serious complication of COVID-19 is a type of pneumonia called 2019 novel coronavirus-infected pneumonia (NCIP) with about 4.3% mortality rate. Comparing to chest Digital Radiography (DR), it is recently reported that chest Computed Tomography (CT) is more useful to serve as the early screening and diagnosis tool for NCIP. In this study, aimed to help physicians make the diagnostic decision, we develop a machine learning (ML) approach for automated diagnosis of NCIP on chest CT. Different from most ML approaches which often require training on thousands or millions of samples, we design a few-shot learning approach, in which we combine few-shot learning with weakly supervised model training, for computerized NCIP diagnosis. A total of 824 patients are retrospectively collected from two Hospitals with IRB approval. We first use 9 patients with clinically confirmed NCIP and 20 patients without known lung diseases for training a location detector which is a multitask deep convolutional neural network (DCNN) designed to output a probability of NCIP and the segmentation of targeted lesion area. An experienced radiologist manually localizes the potential locations of NCIPs on chest CTs of 9 COVID-19 patients and interactively segments the area of the NCIP lesions as the reference standard. Then, the multitask DCNN is furtherly fine-tuned by a weakly supervised learning scheme with 291 case-level labeled samples without lesion labels. A test set of 293 patients is independently collected for evaluation. With our NCIP-Net, the test AUC is 0.91. Our system has potential to serve as the NCIP screening and diagnosis tools for the fight of COVID-19’s endemic and pandemic.

Using aerial surveys and citizen science to create species distribution models for an imperiled grouse

Estimating species distributions requires species presence data of sufficient quantity from reputable sources that are geographically representative of the species’ space use. Collecting presence data that meets these standards can be costly and is often complicated by limited land access. Citizen science projects are an appealing alternative source of presence data as these data are freely available and collected globally. Websites such as eBird have become increasingly large repositories of citizen science data. The vulnerable lesser prairie-chicken (LPC; Tympanuchus pallidicinctus) is a species well-represented in the eBird database, with presence observations from 140 unique locations from 2012 to 2014. During that same period, a distribution-wide, standardized aerial survey with state and federal support recorded 106 LPC detections. Our objective was to compare species distribution models (SDMs) made with eBird data to models made with aerial survey data to determine the potential for citizen science data to contribute to conservation planning. We used maximum entropy modeling to create SDMs based on eBird data, aerial survey data, and a combination of both data sets using variables of biological significance to LPCs. We obtained comparable model performance using aerial survey data only [standardized test omission rate (STO): 23.4%, test AUC: 0.76] and with eBird data only (STO: 23.8%, AUC: 0.76). The I statistic confirmed a very high degree of similarity between the outputs of the two model sets (I = 0.929). However, a road bias existed within both data sets (positive and negative biases), potentially confounding some environmental correlates. Despite this bias, our combined model predicted an increase of 1,732,500 ha of unique area suitable for the LPC. Our results indicated that eBird data could be used as a low-cost source for species occurrence data to create species distribution models, though biases in these datasets should be assessed to guide interpretability of the predicted outputs.

Insolvency Prediction Analysis of Italian Small Firms by Deep Learning

To improve credit risk management, there is a lot of interest in bankruptcy predictive models. Academic research has mainly used traditional statistical techniques, but interest in the capability of machine learning methods is growing. This Italian case study pursues the goal of developing a commercial firms insolvency prediction model. In compliance with the Basel II Accords, the major objective of the model is an estimation of the probability of default over a given time horizon, typically one year. The collected dataset consists of absolute values as well as financial ratios collected from the balance sheets of 14.966 Italian micro-small firms, 13,846 ongoing and 1,120 bankrupted, with 82 observed variables. The volume of data processed places the research on a scale like that used by Moody’s in the development of its rating model for public and private companies, RiskcalcTM. The study has been conducted using Gradient Boosting, Random Forests, Logistic Regression and some deep learning techniques: Convolutional Neural Networks and Recurrent Neural Networks. The results were compared with respect to the predictive performance on a test set, considering accuracy, sensitivity and AUC. The results obtained show that the choice of the variables was very effective, since all the models show good performances, better than those obtained in previous works. Gradient Boosting was the preferred model, although an increase in observation times would probably favour Recurrent Neural Networks.

Predicting rehospitalization within 2 years of initial patient admission for a major depressive episode: a multimodal machine learning approach

Machine learning methods show promise to translate univariate biomarker findings into clinically useful multivariate decision support systems. At current, works in major depressive disorder have predominantly focused on neuroimaging and clinical predictor modalities, with genetic, blood-biomarker, and cardiovascular modalities lacking. In addition, the prediction of rehospitalization after an initial inpatient major depressive episode is yet to be explored, despite its clinical importance. To address this gap in the literature, we have used baseline clinical, structural imaging, blood-biomarker, genetic (polygenic risk scores), bioelectrical impedance and electrocardiography predictors to predict rehospitalization within 2 years of an initial inpatient episode of major depression. Three hundred and eighty patients from the ongoing 12-year Bidirect study were included in the analysis (rehospitalized: yes = 102, no = 278). Inclusion criteria was age ≥35 and <66 years, a current or recent hospitalisation for a major depressive episode and complete structural imaging and genetic data. Optimal performance was achieved with a multimodal panel containing structural imaging, blood-biomarker, clinical, medication type, and sleep quality predictors, attaining a test AUC of 67.74 (p = 9.99−05). This multimodal solution outperformed models based on clinical variables alone, combined biomarkers, and individual data modality prognostication for rehospitalization prediction. This finding points to the potential of predictive models that combine multimodal clinical and biomarker data in the development of clinical decision support systems.

Application of a Maximum Entropy Model for Mineral Prospectivity Maps

The effective integration of geochemical data with multisource geoscience data is a necessary condition for mapping mineral prospects. In the present study, based on the maximum entropy principle, a maximum entropy model (MaxEnt model) was established to predict the potential distribution of copper deposits by integrating 43 ore-controlling factors from geological, geochemical and geophysical data. The MaxEnt model was used to screen the ore-controlling factors, and eight ore-controlling factors (i.e., stratigraphic combination entropy, structural iso-density, Cu, Hg, Li, La, U, Na2O) were selected to establish the MaxEnt model to determine the highest potential zone of copper deposits. The spatial correlation between each ore-controlling factor and the occurrence of a copper mine was studied using a response curve, and the relative importance of each ore-controlling factor was determined by jackknife analysis in the MaxEnt model. The results show that the occurrence of copper ore is positively correlated with the content of Cu, Hg, La, structural iso-density and stratigraphic combination entropy, and negatively correlated with the content of Na2O, Li and U. The model’s performance was evaluated by the area under the receiver operating characteristic curve (AUC), Cohen’s maximized Kappa and true skill statistic (TSS) (training AUC = 0.84, test AUC = 0.8, maximum Kappa = 0.5 and maximum TSS = 0.6). The results indicate that the model can effectively integrate multi-source geospatial data to map mineral prospectivity.

Endometrial Cancer Risk Prediction and Stratification Using Personal Health Data

While the 5-year survival rate of endometrial cancer is relatively high (82%), the incidence and death rates are rising. Currently screening methods (e.g. endometrial biopsy, pelvic ultrasound) are primarily used for women with a genetic predisposition (e.g. Lynch syndrome). So far, there is no effective way to conduct population- based screening. It is therefore crucial to have a risk stratification tool, based on comprehensive personal health data, which can identify high risk individuals who would benefit from screening. Such a stratification scheme could also help direct preventative interventions for individuals. In this study we developed a machine learning model capable of accurately predicting endometrial cancer risk from personal health data. We used the following inputs: age, BMI and weight chance, race, smoking habits, diabetes, emphysema, stroke, hypertension, heart disease, arthritis, another cancer, family history of cancer, gynecological surgeries, menarche age, parity, use of birth control, and age at menopause extracted from the Prostate, Lung, Colorectal, Ovarian Cancer Screening Trail data. Participants in the trial were followed for 13 years or until cancer diagnosis with 78,215 women participants, 952 of whom developed endometrial cancer within 5 years of enrolling. Splitting the data into training (70%) and testing (30%) sets we tested logistic regression, decision tree, random forest, linear discriminant analysis, support vector machine, naive Bayes, and neural networks (NN). The top model was then used to stratify endometrial cancer risk into low, medium, and high risk categories. The NN outperformed the other methods with a test AUC of 0.88. Using the NN, we classified 57.2% of those who developed cancer within 5 years as high risk, 41.8% as medium risk, and 1.1% as low risk. For those who did not develop cancer within 5 years we classified 0.9%, 71.0%, and 28.2% as high, medium, and low risk, respectively.Abstract 2304; Table 1# Low Risk% Low Risk# Medium Risk% Medium Risk# High Risk% High RiskCancer31.1%11941.8%16357.2%Non-Cancer6,52728.2%16,45471.0%1970.9%Risk stratification results on the test set by our NN. Open table in a new tab Risk stratification results on the test set by our NN. Our results indicate that the use of a NN based on personal health information can accurately discriminate between those at high risk of developing endometrial cancer and those who are not, offering a cost-effective and non-invasive way to stratify endometrial cancer risk for targeted screening and prevention.

Applying machine learning to predict future adherence to physical activity programs

BackgroundIdentifying individuals who are unlikely to adhere to a physical exercise regime has potential to improve physical activity interventions. The aim of this paper is to develop and test adherence prediction models using objectively measured physical activity data in the Mobile Phone-Based Physical Activity Education program (mPED) trial. To the best of our knowledge, this is the first to apply Machine Learning methods to predict exercise relapse using accelerometer-recorded physical activity data.MethodsWe use logistic regression and support vector machine methods to design two versions of a Discontinuation Prediction Score (DiPS), which uses objectively measured past data (e.g., steps and goal achievement) to provide a numerical quantity indicating the likelihood of exercise relapse in the upcoming week. The respective prediction accuracy of these two versions of DiPS are compared, and then numerical simulation is performed to explore the potential of using DiPS to selectively allocate financial incentives to participants to encourage them to increase physical activity.Resultswe had access to a physical activity trial data that were continuously collected every 60 sec every day for 9 months in 210 participants. By using the first 15 weeks of data as training and test on weeks 16–30, we show that both versions of DiPS have a test AUC of 0.9 with high sensitivity and specificity in predicting the probability of exercise adherence. Simulation results assuming different intervention regimes suggest the potential benefit of using DiPS as a score to allocate resources in physical activity intervention programs in reducing costs over other allocation schemes.ConclusionsDiPS is capable of making accurate and robust predictions for future weeks. The most predictive features are steps and physical activity intensity. Furthermore, the use of DiPS scores can be a promising approach to determine when or if to provide just-in-time messages and step goal adjustments to improve compliance. Further studies on the use of DiPS in the design of physical activity promotion programs are warranted.Trial registrationClinicalTrials.gov NCT01280812 Registered on January 21, 2011.

Environmental niche modeling for some species of the genus Anthrax Scopoli (Diptera: Bombyliidae) in Egypt, with special notes on St. Catherine protected area as a suitable habitat

The genus Anthrax Scopoli is a well-known genus of bee flies, with 16 species recorded from Egypt, out of about 250 species known worldwide. Six of these species seem to be endangered or threatened (at least in a part of their distribution range) as they disappeared from most of their Egyptian localities which faced severe changes in the floral and faunal composition during the last 50 years. These changes may be attributed to urbanization or modern agricultural activities that caused severe degradation of natural habitats of these species. Maximum entropy (Maxent) modeling was used to predict the potential climatic niches of these six Anthrax species, namely: Anthrax aethiops, Anthrax chionanthrax, Anthrax dentate, Anthrax melanista, Anthrax lucidus, and Anthrax trifasciatus. The Maxent model performed better than random for the six species, with an average test AUC value of 0.95 for A. aethiops bezzii, 0.83 for A. chionanthrax, and 0.98 for A. dentate, 0.86 for A. melanista, 0.96 for A. lucidus and 0.99 for A. trifasciatus. A set of four variables were selected for each species from 19 bioclimatic and 3 topographic variables, based on correlation analyses. The study showed that the current distributions of A. dentata, A. lucidus and A. trifasciatus were largely explained by the altitude, but A. chionanthrax and A. melanista were so far dominated by temperature, whereas on A. aethiops bezzii was strongly related to precipitation variables. St. Catherine protected area showed high habitat suitability for all species except A. lucidus. The suitable habitat for A. trifasciatus was restricted to this protected area, however, future scenarios of climatic changes through the area predict a great degradation in habitat suitability for this species. These results can be used for making powerful conservation management decisions in St. Catherine Protectorate to incorporate these Anthrax spp. in any present or future conservation programs.

Detection of Acute Respiratory Distress Syndrome by Incorporation of Label Uncertainty and Partially Available Privileged Information.

Acute respiratory distress syndrome (ARDS) is a fulminant inflammatory lung injury that develops in patients with critical illnesses including sepsis, pneumonia, and trauma. However, many patients with ARDS are not recognized when they develop this syndrome nor given outcome-improving treatments. Because ARDS is a clinical syndrome, physicians may not be certain about a patient's diagnosis (label uncertainty). In addition, the diagnosis requires a chest x-ray, which may not be always be available in a clinical setting (privileged information). For this paper, we implemented the Learning Using Label Uncertainty and Partially Available Privileged Information (LULUPAPI) paradigm, built on classical SVM, to detect ARDS using Electronic Health Record (EHR) data and chest radiography. In comparison to SVM, this resulted in a 3.55 percent improvement of test AUC.

The status and potential distribution of Hydrocotyle umbellata L. and Salvinia auriculata Aubl. under climate change scenarios

Aquatic ecosystems are susceptible to human-induced disturbance, including climate changes and biological invasions. The aim of this study was to assess the current and future potential distribution of two introduced aquatic species that have become invasive in some places where they were introduced. Hydrocotyle umbellata L. and Salvinia auriculata Aubl. are free-floating macrophytes native to North, Central, and South America. Both can quickly colonize aquatic environments because of their high growth rate and reproductive capacity similar to water hyacinth. Both species were introduced to Egypt for ornamental purposes. We have applied species distribution models using the Maxent approach and bioclimatic variables. Occurrence records from the entire range of the two species were obtained from the Global Biodiversity Information Facility and used for modelling their habitat suitability and assessing the potentiality of their spread in other new habitats. To project future changes in the two macrophytes’ distributions with respect to climate change, we used four representative concentration pathway scenarios (RCP 2.6, 4.5, 6.0 and 8.5) of the IPCC 5th assessment, based on different assumptions of greenhouse gas emissions for the future period of 2050s. The results showed that Maxent approach has successfully predicted the distribution of the two species with test AUC > 0.92. Bioclimatic variables that contributed the most to the prediction of the two species distribution included isothermality, temperature seasonality, mean temperature of the coldest quarter, and minimum temperature of the coldest month. Results showed that the range of S. auriculata is predicted to increase by 2050 under all climatic scenarios. A decline in the current climatically suitable habitats of H. umbellata is projected to occur in its native range, especially in South America, while it is predicted to gain more suitable habitats out of its native range in Europe and Africa. Both species are predicted to gain habitats outside their native range, while their ranges are expected to face a decline in their native region. The study can help in the identification of areas with high potential vulnerability to future invasions by the two studied aquatic macrophytes and thus can assist in prioritization of monitoring actions and management plans. This can reduce any ecological and socio-economic consequences due to invasion by these two aquatic plants.

Predicting the invasion of a southern African savannah by the black wattle (Acacia mearnsii)

Understanding the drivers of biological invasions in landscapes is a major goal in invasion ecology. The control of biological invasions has increasingly become critical in the past few decades because invasive species are thought to be a major threat to endemism. In this study, by examining the key variables that influence Acacia mearnsii, we sought to understand its potential invasion in eastern Zimbabwe. We used the maximum entropy (MaxEnt) method against a set of environmental variables to predict the potential invasion front of A. mearnsii. Our study showed that the predictor variables, i.e., aspect, elevation, distance from streams, soil type and distance from the nearest A. mearnsii plantation adequately explained (training AUC = 0.96 and test AUC = 0.93) variability in the spatial distribution of invading A. mearnsii. The front of invasion by A. mearnsii seemed also to occur next to existing A. mearnsii plantations. Results from our study could be useful in identifying priority areas that could be targeted for controlling the spread of A. mearnsii in Zimbabwe and other areas under threat from A. mearnsii invasion. We recommend that the plantation owners pay for the control of A. mearnsii invasion about their plantations.

The influence of data source and species distribution modelling method on spatial conservation priorities

AbstractAimSpecies distribution models are an important conservation tool; however, performance can vary with factors including data inputs and modelling method. Model outputs are often under‐evaluated for explanatory and predictive capacity. Our aim was to evaluate the capacity of existing data for seven small mammal species to provide useful inferences for management planning.LocationBathurst and Melville (collectively the Tiwi) Islands, Northern Territory, Australia.MethodsWe developed species distribution models (SDMs) with generalized linear models (GLMs) and boosted regression trees (BRTs) using survey data (351 sites) of small mammals, with two sets of environmental predictors: (a) field‐study measurements and (b) available remotely sensed rasters. Predictive capacity of models was evaluated using percentage of deviance explained (%DE) and area under the receiver operating characteristic curve (AUC). We used Marxan to evaluate the influence of different model and data types as input for identifying spatial priorities.ResultsField‐informed SDMs performed well across both modelling methods, with relatively high test AUC values (mean = 0.82, range = 0.64–0.97) and test %DE (mean = 22.5%, range = 3.5%–65.8%). Remotely sensed models performed relatively poorly, with lower test AUC values (mean = 0.7, range = 0.56–0.86) and lower test %DE (mean = 8.9%, range = 0.03%–24.9%). A notable exception was remotely sensed models forMelomys burtoni(AUC = 0.85 &amp; 0.86, %DE = 23.3% &amp; 24.9%, Bathurst and Melville respectively). Marxan site irreplaceability rankings demonstrated low to marginal agreement using field‐informed and remotely sensed inputs (Pearson correlation coefficient = 0.3), and similarly, using GLM and BRT model inputs (0.29).Main conclusionsThe occurrence of small mammals on the Tiwi Islands can be reasonably explained with field‐informed variables, but not with remotely sensed alternatives. Different models lead to different conservation priorities. Our work emphasizes the importance of thoroughly testing SDMs prior to decision‐making.

Automatic recognition and analysis of metal streak artifacts in head and neck computed tomography for radiomics modeling.

Background and purposeComputed tomography (CT) radiomics of head and neck cancer (HNC) images is susceptible to dental implant artifacts. This work devised and validated an automated algorithm to detect CT metal artifacts and investigate their impact on subsequent radiomics analyses. A new method based on features from total variation, gradient directional distribution, and Hough transform was developed and evaluated.Materials and methodsTwo HNC datasets were analyzed: a training set of 131 patients for developing the detection algorithm and a testing set of 220 patients. Seven designated features were extracted from ROIs (regions of interest) and machine learning with random forests was used for building the artifact detection algorithm. Performance was assessed using the area under the receiver operating characteristics curve (AUC).ResultsThe testing results of artifacts detection yielded a cross-validated AUC of 0.91 (95% CI: 0.89–0.94), and a test AUC of 0.89. External testing validation yielded an accuracy of 0.82. For radiomics model prediction, training with artifacts yielded an AUC of 0.64 (95% CI: 0.63–0.65), while training on images without artifacts improved the AUC to 0.75 (95% CI: 0.74–0.76). This was compared to visual inspection of artifacts (AUC = 0.71 [95% CI: 0.69–0.73]).ConclusionWe developed a new method for automated and efficient detection of streak artifacts. We also showed that such streak artifacts in HNC CT images can worsen the performance of radiomics modeling.