Number Of False Negatives Research Articles

Exploring false negatives in self-reported trauma exposure status.

Trauma exposure is often assessed using checklists such as the Life Events Checklist for DSM-5 (LEC-5; Weathers et al., 2013b). When participants endorse multiple events, respondents are asked to identify a single, worst event (i.e., index event). Recent work indicates that the "worst event" method leads to a concerning number of false negatives. The purpose of the current study was to replicate previous findings of false negatives and extend them by examining characteristics associated with false negatives, such as trauma type, means of exposure, recency of trauma, and posttraumatic stress disorder (PTSD) symptom severity. Adults (n = 476) provided data on trauma history assessed using a revised version of the LEC-5 that asked participants to provide follow-up information for each traumatic event endorsed. Participants also provided demographic data and completed the PTSD Checklist for DSM-5. Results: Two hundred thirty-four participants (49.16%) reported a worst event that met the DSM-5 definition of Criterion A trauma ("primary Criterion A" group). However, of the 242 participants who did not, 138 participants (57.02%, or 28.99% of the total sample) reported a secondary event that did meet Criterion A ("secondary Criterion A" group). The secondary Criterion A group most commonly reported serious life-threatening illnesses/injuries and "other" stressful life experiences as their index trauma that did not fulfill Criterion A. Participants in the primary and secondary Criterion A groups reported similar levels of PTSD symptoms. No differences were observed in means of exposure and recency of index trauma between the Criterion A groups. Findings raise questions regarding the efficiency and accuracy of the worst event method to determine trauma exposure status via self-report. Researchers should consider alternative methods for assessing trauma exposure rather than relying on the worst event scoring method. (PsycInfo Database Record (c) 2024 APA, all rights reserved).

Refined selection of individuals for preventive cardiovascular disease treatment with a transformer-based risk model

Abstract Background Although statistical models have been commonly used to identify patients at risk of cardiovascular disease (CVD) for preventive therapy, these models tend to over-recommend therapy [1]. Moreover, in populations with pre-existing diseases, the current approach is to indiscriminately treat all individuals, as modelling in this context is currently inadequate. Aims To develop and validate a novel deep learning model, the Transformer-based Risk assessment survival model (TRisk) model, for prediction of 10-year risk of CVD in both the general population and individuals with diabetes. Methods An open cohort of 3 million adults aged 25 to 84 years free of CVD and statin use was identified using linked primary and secondary electronic health records from 291 and 98 general practices in England, which were used for TRisk model development and external validation, respectively. Comparison against two commonly recommended statistical models (QRISK3 [2] and SCORE2 [3]) and a deep learning derivation of QRISK3 was conducted. Additional analyses compared discriminatory performance in other age groups, by sex, and across categories of socioeconomic status. Similar analyses were conducted in a subgroup of 59 thousand patients with diabetes. Results TRisk demonstrated superior discrimination (C index in the general population: 0.910; 95% confidence interval [CI]: 0.906 to 0.913). TRisk’s performance was found to be less sensitive to population age range than the benchmark models and outperformed other models also in all stratified analyses. All models were overall well-calibrated. In decision curve analyses, TRisk demonstrated greater net benefit than benchmark models across the range of relevant thresholds. In impact analyses, at the widely recommended 10% risk threshold and the higher 15% threshold for the general population, TRisk reduced both the total number of patients classified at high risk (by 27% and 39% respectively) and the number of false negatives as compared with currently recommended strategies. TRisk similarly outperformed other models in patients with diabetes. Compared with the widely recommended treat-all policy approach for patients with diabetes, TRisk at a 10% risk threshold would lead to deselection of 24% of individuals with a small fraction of false negatives (0.2% of cohort). Conclusion TRisk enabled a more targeted selection of individuals at risk of CVD compared to benchmark statistical and deep learning models, in both the general population and patients with diabetes. Incorporation of TRisk into routine clinical care would allow a reduction in the number of treatment-eligible patients in general population and subgroup with diabetes by approximately one-third and one-quarter respectively while preventing at least as many events as with currently adopted approaches.Models' discrimination on both cohortsImpact analyses for both cohorts

Echocaridographic assessment of pulmonary hypertension: improving the probability

Abstract Introduction Pulmonary hypertension (PH) is a cardio-pulmonary disorder characterised by raised pulmonary artery pressure (mPAP >20mmHg) leading to right ventricular failure, and death if left untreated. Transthoracic echocardiography (TTE) assessment of PH uses a probability-based approach, combining peak tricuspid regurgitant velocity with other supporting signs. We evaluated the latest ESC TTE PH guidelines in a real-world UK dataset and identified markers that may improve the sensitivity of the existing algorithm. Method Retrospective analysis of 447 patients referred for initial PH assessment. This was a select population of patients who underwent both TTE and right heart catheter (RHC) (mean time to RHC = 40±30 days). In addition to current TTE PH probability markers we analysed right ventricular free wall strain (RVFWS). Measurements were made by a single operator using Tomtec Arena 2020. Crosstabulation, ROC curve analysis and T-tests were used to compare variables between PH and no PH. Results Mean sample age was 64±15 yrs (61% female). 6.9% (n=4) of the no PH group had pulmonary mid-systolic notching, which may be due to pseudo-notching attributable to incomplete Doppler envelope. Sensitivity and specificity of current ESC TTE probability guidelines were 83%, and 65% respectively (95% C.I.). However, within the low probability group (i.e. those felt to be less likely to have PH) 62% (n=63) had confirmed PH at RHC (mPAP = 28±9). The addition of RVFWS as an additional marker to the existing supporting signs of PH (Table 1) resulted in an improvement of the sensitivity of the algorithm to 92% with a mild reduction in specificity (63%). The number of false negatives were reduced by 51%. Of these, the majority were in the low TTE probability group (n = 32). Conclusions Using current PH TTE probability algorithms classifies a significant proportion of those with PH as having a low TTE PH probability. The addition of RVFWS improves the sensitivity of the existing algorithm to 92% and reduces false negatives by 51%.

Evaluation of a Rapid Drug Test Device for Urine Fentanyl Compared to Mass Spectrometry and 2 Urine Fentanyl Assays.

A new Rapid Drug Test Device (RDTD) is available for analysis of urine fentanyl. With the rise in fentanyl abuse in the United States, we evaluated the analytical performance of the RDTD test strip compared to mass spectrometry and 2 urine fentanyl immunoassays. Leftover, deidentified urine samples collected from inpatients and outpatients from our psychiatric hospital and other clinics were frozen at <-70°C, thawed at room temperature, and centrifuged. Aliquots were tested with the RDTD (CLIA Waived, Inc.) test strips and 2 urine fentanyl immunoassays: the ARK Fentanyl II assay (ARK Diagnostics Inc.) and the Immunalysis SEFRIA Fentanyl assay (Immunalysis Corporation). Both assays were conducted on the Abbott Alinity c chemistry analyzer (Abbott Laboratories). Mass spectrometry analysis was performed at ARUP Laboratories. All assays had a 1 ng/mL positive cutoff. A total of 142 urine samples included 70 positive and 72 negative samples. The RDTD strips had lower sensitivity (84.3%) and efficiency (85.9%) and showed a specificity of 87.5% compared to the other assays. The ARK Fentanyl II assay showed identical sensitivity (95.7%) to the Immunalysis SEFRIA Fentanyl assay but had higher specificity (94.4% vs 81.9%) and overall efficiency (95.1% vs 88.7%). Differences were noted in the number of false negatives and positives among the assays. The RDTD demonstrated acceptable performance in detecting urine fentanyl in our patient population and would provide faster test results at point-of-care testing sites in our healthcare enterprise.

Research of data mining methods for classification of imbalanced data sets

With the rapid development of information technology, which is widely used in all spheres of human life and activity, extremely large amounts of data have been accumulated today. By applying machine learning methods to this data, new practically useful knowledge can be obtained. The main goal of this paper is to study different machine learning methods for solving the classification problem and compare their efficiency and accuracy. A separate task is data pre-processing aimed at solving the problem of sample imbalance, as well as identifying the principal components that will be used to solve the classification problem. For this purpose, an information system for classifying the bankruptcy of a company with specified economic and financial characteristics was researched and developed. The study uses a dataset on the basis of which the efficiency and quality of application of several existing classification algorithms are evaluated. These classifiers are: conventional and linear Support Vector Machine, Extra Trees, Random Forest, Decision Tree, Logistic Regression, Multilayer perceptron Classifier, Gradient Boosting, Naive Bayes Classifier. For data pre-processing, we scaled the data, used the SMOTE method to get rid of the imbalance of the training sample, and performed principal component analysis and L1 regularisation. Principal component analysis allowed us to identify 15 principal components that have the greatest impact on classification accuracy and, accordingly, use them in the classification process. Analysing the results, we found that the best classifier was Random Forest with 95.9 % accuracy, and the worst was Naive Bayes with 85.1 %. To evaluate the quality of classification and select the best classifier, the Confusion matrix is used, which takes into account the number of true positive (TP) and true negative (TN) values, as well as the number of false negative (FN) and false positive (FP) classification results, and the values of such metrics as accuracy, precision, sensitivity, F1, and ROC. Accuracy is the percentage of correct answers given by the algorithm, while Recall is the number of TPs divided by the number of TPs plus the number of FNs. F1 indicates the balance between accuracy and sensitivity. Precision is the number of true positive predictions divided by the number of false positive and true negative predictions. ROC AUC is a tool for measuring performance for classification tasks at different thresholds. It shows how well a model can distinguish between classes. The conclusions present the main results of the study and indicate the main future direction of the work, namely, the study of classification results for other datasets and more efficient processing and analysis.

What Interval of Daily Pain Assessment Is Required to Reliably Diagnose Chronic Pain in Sickle Cell Disease? the Pisces Study

Background: In Sickle Cell Disease (SCD), chronic pain, i.e., pain on most days over a 6 month time period, is a major complication associated with substantial morbidity and poor outcomes. In the landmark Pain in Sickle Cell Epidemiology Study (PiSCES), the largest study of pain in SCD, one-third of individuals age 16 years or older had daily or near-daily pain, and &amp;gt;50% had chronic pain. However, PiSCES required patients to have 5-6 months of completed diaries in order to qualify for chronic pain analysis. Practically, the International Association for the Study of Pain (IASP) has used three months as the interval upon which to base a definition of chronic pain. Shorter daily assessment intervals upon which to reliably base a definition of chronic pain would reduce respondent burden, and would improve the frequency and accuracy of pain phenotyping for clinical and research purposes. Method: We chose as the gold-standard for chronic pain definition subjects enrolled in PiSCES that had at least 5 months of daily diaries (n=127). We excluded 10 patients who had at least 182 diaries (essentially 6 months), but had at least 50% of diaries missing in any of months 1-4. We simulated daily diary assessment over less burdensome, shorter intervals, always beginning on day one of assessment, to mimic what would be done clinically. We chose the time frames of 2 weeks (2w), 4 weeks (4w), 1 month (1m), 2 months (2m), 3 months (3m), and 4 months (4m). To operationalize the definition of chronic pain, we used a definition of pain severity rated as &amp;gt;0 on &amp;gt; 50% of diary days, regardless of interval. To determine if using diaries collected over a shorter period of time would still classify patients correctly, we calculated the sensitivity and the number of false negatives, as well as the specificity and number of false positives of each simulated diary interval. Results: Within the gold standard sample, 51.3% of patients had chronic pain based on all diaries submitted. The Table shows the percent of patients diagnosed with chronic pain, the sensitivity as well as the number of false negatives, and the specificity as well as the number of false positives, for each simulated diary interval. Simulated intervals of two months or more of daily diaries yielded chronic pain percentage estimates with both high sensitivity and high specificity, compared to the gold standard estimate. Simulated intervals of two week and four weeks yielded increasingly lower specificity, i.e., increasingly more false-positive indications of chronic pain, as the interval shortened. However, each interval still yielded high sensitivity-few patients with gold-standard chronic pain were missed, even at two weeks. Results were very similar even if data from the entire 127 participants was used. Conclusion : For determining the presence of chronic SCD pain, defined as pain on more than 50% of days enrolled in PiSCES, simulated intervals of two months or more of daily diary collection yield both very good sensitivity and very good specificity, compared to intervals of 5-6 months. Clinicians and researchers may reliably choose to adopt two months of dairy collection, as a matter of practicality, to reduce respondent burden when phenotyping SCD pain as acute versus chronic.

The Effectiveness of Software Designed to Detect AI-Generated Writing: A Comparison of 16 AI Text Detectors

Abstract This study evaluates the accuracy of 16 publicly available AI text detectors in discriminating between AI-generated and human-generated writing. The evaluated documents include 42 undergraduate essays generated by ChatGPT-3.5, 42 generated by ChatGPT-4, and 42 written by students in a first-year composition course without the use of AI. Each detector’s performance was assessed with regard to its overall accuracy, its accuracy with each type of document, its decisiveness (the relative number of uncertain responses), the number of false positives (human-generated papers designated as AI by the detector), and the number of false negatives (AI-generated papers designated as human). Three detectors – Copyleaks, TurnItIn, and Originality.ai – have high accuracy with all three sets of documents. Although most of the other 13 detectors can distinguish between GPT-3.5 papers and human-generated papers with reasonably high accuracy, they are generally ineffective at distinguishing between GPT-4 papers and those written by undergraduate students. Overall, the detectors that require registration and payment are only slightly more accurate than the others.

Inherent lack of CMFDA/FDA staining in certain algae and its implication for ballast water testing

The approved method for testing the efficacy of ballast water management systems with respect to killing 10–50 μm organisms uses movements of the organisms or the vital stains CMFDA/FDA. The present study demonstrates that certain freshwater coccoid chlorophytes, known or suspected to contain a highly resistant cell wall component (algaenan), stain poorly with CMFDA/FDA, resulting in false negatives. The staining rates for the most dominant species were determined and were approx. 3–70 %. The use of Crystal Violet as an indicator for the presence of algaenan gave inconclusive results. The number of the 10–50 μm organisms in a small pond was found to be 10,183 organisms/mL (Lugol's fixed sample) vs. 2335 organisms/mL (CMFDA/FDA-stained sample). Using the staining rates obtained, it was estimated that the number of false negatives could make 40–50 %. The implications for biological performance evaluation of ballast water management systems are discussed.

A new multimodal paradigm for biomarkers longitudinal monitoring: a clinical application to women steroid profiles in urine and blood

BackgroundMost current state-of-the-art strategies to generate individual adaptive reference ranges are designed to monitor one clinical parameter at a time. An innovative methodology is proposed for the simultaneous longitudinal monitoring of multiple biomarkers. The estimation of individual thresholds is performed by applying a Bayesian modeling strategy to a multivariate score integrating several biomarkers (compound concentration and/or ratio). This multimodal monitoring was applied to data from a clinical study involving 14 female volunteers with normal menstrual cycles receiving testosterone via transdermal route, as to test its ability to detect testosterone administration. The study samples consisted of urine and blood collected during 4 weeks of a control phase and 4 weeks with a daily testosterone gel application. ResultsIntegrating multiple biomarkers improved the detection of testosterone gel administration with substantially higher sensitivity compared with the distinct follow-up of each biomarker, when applied to selected urine and serum steroid biomarkers, as well as the combination of both. Among the 175 known positive samples, 38% were identified by the multimodal approach using urine biomarkers, 79% using serum biomarkers and 83% by combining biomarkers from both biological matrices, whereas 10%, 67% and 64% were respectively detected using standard unimodal monitoring. Significance and noveltyThe detection of abnormal patterns can be improved using multimodal approaches. The combination of urine and serum biomarkers reduced the overall number of false-negatives, thus evidencing promising complementarity between urine and blood sampling for doping control, as highlighted in the case of the use of transdermal testosterone preparations. The generation in a multimodal setting of adaptive and personalized reference ranges opens up new opportunities in clinical and anti-doping profiling. The integration of multiple parameters in a longitudinal monitoring is expected to provide a more complete evaluation of individual profiles generating actionable intelligence to further guide sample collection, analysis protocols and decision-making in clinics and anti-doping.

Artificial intelligence for localization of the acute ischemic stroke by non-contrast computed tomography.

A non-contrast cranial computer tomography (ncCT) is often employed for the diagnosis of the early stage of the ischemic stroke. However, the number of false negatives is high. More accurate results are obtained by an MRI. However, the MRI is not available in every hospital. Moreover, even if it is available in the clinic for the routine tests, emergency often does not have it. Therefore, this paper proposes an end-to-end framework for detection and segmentation of the brain infarct on the ncCT. The computer tomography perfusion (CTp) is used as the ground truth. The proposed ensemble model employs three deep convolution neural networks (CNNs) to process three end-to-end feature maps and a hand-craft features characterized by specific contra-lateral features. To improve the accuracy of the detected infarct area, the spatial dependencies between neighboring slices are employed at the postprocessing step. The numerical experiments have been performed on 18 ncCT-CTp paired stroke cases (804 image-pairs). The leave-one-out approach is applied for evaluating the proposed method. The model achieves 91.16% accuracy, 65.15% precision, 77.44% recall, 69.97% F1 score, and 0.4536 IoU.

Health outcomes modelling of RNA-based versus DNA-based detection of driver gene rearrangements in non-small cell lung cancer.

e18837 Background: There are several FDA-approved targeted therapies for NSCLC with driver gene rearrangements (DGRs) involving ALK, ROS1, RET, NTRK1-3, and MET exon 14. DNA-based sequencing panels miss 15-20% of DGRs compared to RNA-based methods (Benayed et al., 2019; Solomon et al., 2020). We modelled the years of life gained by using RNA- instead of DNA-based methods for DGR detection. Methods: We used the IVI-NSCLC open-source model to analyze the relationship between stable disease, progression, and death as previously described in a multi-state model. Previously reported PFS hazard ratios for targeted therapies (versus chemotherapy) treating DGRs range from 0.17 to 0.52 (Hoang et al., 2020). Therefore, we assumed that the HRs corresponding to the parameter estimates for the targeted therapy option built into the model would approximate treatment-specific relative effects of targeted therapies for DGRs (PFS = 0.3; OS = 0.8). We simulated life-years and quality-adjusted life years (QALYs) for a cohort of 1,000 patients with DGRs matched to appropriate targeted therapies over a 5-year time horizon. We used population-level prevalence data to estimate the number of patients with metastatic NSCLC and DGRs. We then estimated the number of false negatives generated by DNA-only sequencing and calculated the additional life-years gained if RNA-based methods were used instead. Results: In a simulated cohort of 1,000 patients with DGRs detected through RNA sequencing and receiving targeted therapy, median PFS and OS were 1.8 and 3.7 years, respectively. The use of targeted therapy (instead of chemotherapy) in these patients increased life-years by 3.46 years (95% CI, 3.04-3.90) and QALYs by 2.27 years (95% CI 1.91-2.71) on average per patient. There were approximately 235,768 new lung cancers in the US in 2021, 56% of which were metastatic at diagnosis (SEER). Approximately 50% of lung cancers are adenocarcinomas, and approximately 12% of adenocarcinomas have DGRs: ALK (5%) , MET (4%) , ROS1 (2%) , RET (1%) , and NTRK1-3 (&lt; 1%). Thus, there were approximately 7,922 new metastatic NSCLC cases with DGRs in 2021. Prior studies have shown that DNA-only methods have a sensitivity of 80-85% for DGRs compared to RNA-based methods. At the level of the US population, this implies 1,188-1,584 false negative cases in 2021. The clinical impact of eliminating these false negatives by using RNA- instead of DNA-based methods to identify DGRs amounts to 4,110–5,480 life-years or 2,696–3,595 QALYs gained by appropriately matching patients with DGRs to targeted therapy over chemotherapy. Conclusions: DNA-only sequencing methods have limited sensitivity for DGRs in lung cancer, leading to false negatives and missed opportunities for targeted therapy. This modeling study suggests that the elimination of such false negatives using RNA-based detection methods would lead to meaningful gains in patient outcomes.

Doppler trans-thoracic echocardiography for detection of pulmonary hypertension in adults.

Doppler trans-thoracic echocardiography for detection of pulmonary hypertension in adults.

Cardiac Disease Prediction using Supervised Machine Learning Techniques.

Diagnosis of cardiac disease requires being more accurate, precise, and reliable. The number of death cases due to cardiac attacks is increasing exponentially day by day. Thus, practical approaches for earlier diagnosis of cardiac or heart disease are done to achieve prompt management of the disease. Various supervised machine learning techniques like K-Nearest Neighbour, Decision Tree, Logistic Regression, Naïve Bayes, and Support Vector Machine (SVM) model are used for predicting cardiac disease using a dataset that was collected from the repository of the University of California, Irvine (UCI). The results depict that Logistic Regression was better than all other supervised classifiers in terms of the performance metrics. The model is also less risky since the number of false negatives is low as compared to other models as per the confusion matrix of all the models. In addition, ensemble techniques can be approached for the accuracy improvement of the classifier. Jupyter notebook is the best tool, for the implementation of Python Programming having many types of libraries, header files, for accurate and precise work.

Sense the moment: A highly sensitive antimicrobial activity predictor based on hydrophobic moment

BackgroundComputer-aided identification and design tools are indispensable for developing antimicrobial agents for controlling antibiotic-resistant bacteria. Antimicrobial peptides (AMPs) have aroused intense interest, since they have a broad spectrum of activity, and therefore, several systems for predicting antimicrobial peptides have been developed, using scalar physicochemical properties; however, regardless of the machine learning algorithm, these systems often fail in discriminating AMPs from their shuffled versions, leading to the need for new training methods to overcome this bias. Aiming to solve this bias, here we present “Sense the Moment”, a prediction system capable of discriminating AMPs and shuffled versions. MethodsThe system was trained using 776 entries: 388 from known AMPs and another 388 based on shuffled versions of known AMPs. Each entry contained the geometric average of three hydrophobic moments measured with different scales. ResultsThe model showed good accuracy (>80%) and excellent sensitivity (>90%) for AMP prediction, exceeding deep-learning-based methods. ConclusionOur results demonstrate the system's applicability, aiding in identifying and discarding non-AMPs, since the number of false negatives is lower than false positives. General significanceThe application of this model in virtual screening protocols for identifying and/or creating antimicrobial agents could aid in the identification of potential drugs to control pathogenic microorganisms and in solving the antibiotic resistance crisis. AvailabilityThe system was implemented as a web application, available at <http://portoreports.com/stm/>.

Computed Tomography Image Based on Intelligent Segmentation Algorithm in the Diagnosis of Ovarian Tumor

This study was to explore the application of computed tomography (CT) images based on intelligent segmentation algorithms in the analysis of ovarian tumors, so as to provide a theoretical basis for clinical diagnosis of ovarian tumors. In this study, 100 patients with ovarian tumors were selected as the research objects and performed CT imaging examinations; a convolutional neural networks (CNN) algorithm model was constructed and applied to CT diagnostic image segmentation of patients with ovarian tumors, so as to analyze the effectiveness of the proposed algorithm for CT image segmentation. As a result, the image was segmented three times under the CNN algorithm, and the numbers of true positives (TP) were 50, 49, and 50, respectively; the numbers of false positives (FP) were 1, 2, and 1, respectively; the numbers of false negatives (FN) were 2, 3, and 2, respectively; and the numbers of true negatives (TN) were 47, 46, and 47, respectively. Thus, there was no great difference in the three measured values P ≥ 0.05 . The accuracy of the CNN algorithm was 0.97, 0.95, and 0.97, respectively, for the three times of segmentation; the precision was 0.98, 0.96, and 0.98, respectively; the recall was 0.96, 0.94, and 0.96, respectively. Thus, the accuracy, precision, and recall of the three measurements were not greatly different P ≥ 0.05 . In addition, the F1 values of three measurements were 0.97, 0.94, and 0.97, respectively, which all were close to 1, showing no statistically great difference P ≥ 0.05 . The segmentation accuracy, precision, and recall of the algorithm in this study were greatly greater than the SE-Res Block U-shaped CNN algorithm, and the density peak clustering algorithm, and the differences were statistically significant P &lt; 0.05 . In short, the CNN algorithm showed high accuracy, precision, recall, and comprehensive evaluation values for CT image segmentation, which made the diagnosis of malignant or benign ovarian tumors more effective and provided reliable theoretical guidance for clinical analysis of ovarian tumors.

Can AI Help Pediatricians? Diagnosing Kawasaki Disease Using DRSA.

The DRSA method (dominance-based rough set approach) was used to create decision-making rules based on the results of physical examination and additional laboratory tests in the differential diagnosis of Kawasaki disease (KD), infectious mononucleosis and S. pyogenes pharyngitis in children. The study was conducted retrospectively. The search was based on the ICD-10 (International Classification of Diseases) codes of final diagnosis. Demographic and laboratory data from one Polish hospital (Poznan) were collected. Traditional statistical methods and the DRSA method were applied in data analysis. The algorithm formed 45 decision rules recognizing KD. The rules with the highest sensitivity (number of false negatives equals zero) were based on the presence of conjunctivitis and CRP (C-reactive Protein) ≥ 40.1 mg/L, thrombocytosis and ESR (Erythrocyte Sedimentation Rate) ≥ 77 mm/h; fair general condition and fever ≥ 5 days and rash; fair general condition and fever ≥ 5 days and conjunctivitis; fever ≥ 5 days and rash and CRP ≥ 7.05 mg/L. The DRSA analysis may be helpful in diagnosing KD at an early stage of the disease. It can be used even with a small amount of clinical or laboratory data.

Support vector machine and deep-learning object detection for localisation of hard exudates

Hard exudates are one of the main clinical findings in the retinal images of patients with diabetic retinopathy. Detecting them early significantly impacts the treatment of underlying diseases; therefore, there is a need for automated systems with high reliability. We propose a novel method for identifying and localising hard exudates in retinal images. To achieve fast image pre-scanning, a support vector machine (SVM) classifier was combined with a faster region-based convolutional neural network (faster R-CNN) object detector for the localisation of exudates. Rapid pre-scanning filtered out exudate-free samples using a feature vector extracted from the pre-trained ResNet-50 network. Subsequently, the remaining samples were processed using a faster R-CNN detector for detailed analysis. When evaluating all the exudates as individual objects, the SVM classifier reduced the false positive rate by 29.7% and marginally increased the false negative rate by 16.2%. When evaluating all the images, we recorded a 50% reduction in the false positive rate, without any decrease in the number of false negatives. The interim results suggested that pre-scanning the samples using the SVM prior to implementing the deep-network object detector could simultaneously improve and speed up the current hard exudates detection method, especially when there is paucity of training data.

Implementation of a maternal early warning system during early postpartum. A prospective observational study.

ObjectiveTo evaluate the implementation of a maternal early warning system (MEWS) for monitoring patients during the first two hours after delivery in a tertiary level hospital.MethodsImplementation of the criteria between 15 March and 15 September 2018 was evaluated in 1166 patients. The parameters collected were systolic and diastolic blood pressure, heart rate, oxygen saturation, urine output, uterine involution, and bleeding. Out-of-range values of any of these parameters triggered a warning, and an obstetrician was called to examine the patient. The obstetrician then decided whether to call the anesthesiologist. We carried out a sensitivity-specificity study of triggers and a multivariate analysis of the factors involved in developing potentially fatal disorders (PFD), reintervention, critical care admission, and stay.ResultsThe MEWS was triggered in 75 patients (6.43%). Leading trigger was altered systolic blood pressure in 32 patients (42.7%), and 11 patients had a PFD. Twenty-eight triggers were false-negatives. Sensitivity and specificity of the system was 0.28 (0.15, 0.45) and 0.94 (0.93, 0.96), respectively. The multivariate analysis showed a correlation between trigger activation and PFD.ConclusionOur MEWS presented low sensitivity and high specificity, with a significant number of false-negatives.

AEC Classifier: A Tree-Based Classifier with Error Control for Medical Disease Diagnosis and Other Applications

In medical disease diagnosis, the cost of a false negative could greatly outweigh the cost of a false positive. This is because the former could cost a life, whereas the latter may only cause medical costs and stress to the patient. The unique nature of this problem highlights the need of asymmetric error control for binary classification applications. In this domain, traditional machine learning classifiers may not be ideal as they do not provide a way to control the number of false negatives below a certain threshold. This paper proposes a novel tree-based binary classification algorithm that can control the number of false negatives with a mathematical guarantee, based on Neyman–Pearson (NP) Lemma. This classifier is evaluated on the data obtained from different heart studies and it predicts the risk of cardiac disease, not only with comparable accuracy and AUC-ROC score but also with full control over the number of false negatives. The methodology used to construct this classifier can be expanded to many more use cases, not only in medical disease diagnosis but also beyond as shown from analysis on different diverse datasets.

Prediction of student attrition risk using machine learning

PurposeThe prediction of student attrition is critical to facilitate retention mechanisms. This study aims to focus on implementing a method to predict student attrition in the upper years of a physiotherapy program.Design/methodology/approachMachine learning is a computer tool that can recognize patterns and generate predictive models. Using a quantitative research methodology, a database of 336 university students in their upper-year courses was accessed. The participant's data were collected from the Financial Academic Management and Administration System and a platform of Universidad Autónoma de Chile. Five quantitative and 11 qualitative variables were chosen, associated with university student attrition. With this database, 23 classifiers were tested based on supervised machine learning.FindingsAbout 23.58% of males and 17.39% of females were among the attrition student group. The mean accuracy of the classifiers increased based on the number of variables used for the training. The best accuracy level was obtained using the “Subspace KNN” algorithm (86.3%). The classifier “RUSboosted trees” yielded the lowest number of false negatives and the higher sensitivity of the algorithms used (78%) as well as a specificity of 86%.Practical implicationsThis predictive method identifies attrition students in the university program and could be used to improve student retention in higher grades.Originality/valueThe study has developed a novel predictive model of student attrition from upper-year courses, useful for unbalanced databases with a lower number of attrition students.