Area Under The Receiver Operating Characteristic Research Articles

Establishment of Scoring to Predict Severe Complication After Pyogenic Spondylodiscitis Surgery.

Retrospective cohort study using prospective database. This study aimed to establish a risk-scoring system for predicting severe complications after pyogenic spondylodiscitis surgery. Pyogenic spondylodiscitis surgery can cause severe complications. Grades III, IV, and V complications in the CD classification were defined as severe complications. A predictive scoring system for severe complications was developed using 7 risk factors identified from a cohort of 143 PS surgery patients from January 2013 to December 2017 described in a previous study. External validation used a separate cohort of 70 patients from 9 institutions identified from January 2018 to December 2021. This first study proposed a risk predictive scoring system for severe complications [updated Charlson comorbidity index (≥3), 2; chronic pulmonary disease, 3; diabetes, 1; Gram-negative bacteria, 3; pyogenic osteoarthritis, 3; preoperative white blood count (≥1.0×104/μL), 2; preoperative platelet count (≤2.4×105/μL), 2]. The scoring system could well predict severe complications [area under the receiver operating curve (AUROC) value of 0.851]. There was classification into 3 risk groups: low-moderate risk (≤4), high risk (5 or 6), and very high risk (≥7), further simplified by the scoring system. Complication rates were 5.2% (low-moderate), 40.1% (high), and 84.2% (very high). In addition, external validation showed a very good AUROC value of 0.820. We developed a simple, externally validated scoring system for predicting severe complications after pyogenic spondylodiscitis surgery that will be helpful for clinicians involved in informed consent and intensive care unit management of high-risk patients.

Insights into prescribing patterns for antidepressants: an evidence-based analysis

BackgroundAntidepressants are a primary treatment for depression, yet prescribing them poses significant challenges due to the absence of clear guidelines for selecting the most suitable option for individual patients. This study aimed to analyze prescribing patterns for antidepressants across healthcare providers, including physicians, physician assistants, nurse practitioners, and pharmacists, to better understand the complex factors influencing these patterns in the management of depression.MethodsLeast Absolute Shrinkage and Selection Operator (LASSO) regression was employed to identify variables that explained the variation in the prescribed antidepressants, utilizing a large number of claims. Models were created to identify the prescription patterns of the 14 most common antidepressants, including amitriptyline, bupropion, citalopram, desvenlafaxine, doxepin, duloxetine, escitalopram, fluoxetine, mirtazapine, nortriptyline, paroxetine, sertraline, trazodone, and venlafaxine. The accuracy of predictions was measured through the Area under the Receiver Operating Curve (AROC).ResultsOur analysis revealed several key factors influencing prescribing patterns, including patients’ comorbidities, previous medications, age, and gender. A history of high antidepressant use (four or more prior medications) was the most common factor across antidepressants. Age influenced prescribing patterns, with mirtazapine and trazodone more frequent among older patients, while fluoxetine and sertraline were more common in younger individuals. Condition-specific factors included trazodone for insomnia, and amitriptyline or nortriptyline for headaches. Paroxetine, venlafaxine, and sertraline more often prescribed to females, while bupropion and doxepin were commonly prescribed for patients with tobacco use disorder and opioid dependence. Predictive factors per medicine ranged from 51 (doxepin) to 168 (citalopram), with cross-validated AROC scores averaging 76.3%.ConclusionsOur findings provide valuable insights into the nuanced factors that shape evidence-based antidepressant prescribing practices, offering a foundation for more personalized, effective depression treatment. Further research is needed to validate these models in other extant databases. These findings contribute to a more comprehensive understanding of antidepressant prescribing practices and have the potential to improve patient outcomes in the management of depression.

Predicting the risk of heart failure after acute myocardial infarction using an interpretable machine learning model

Background Early prediction of heart failure (HF) after acute myocardial infarction (AMI) is essential for personalized treatment. We aimed to use interpretable machine learning (ML) methods to develop a risk prediction model for HF in AMI patients.MethodsWe retrospectively included patients initially with AMI who received percutaneous coronary intervention (PCI) in our hospital from November 2016 to February 2020. The primary endpoint was the occurrence of HF within 3 years after operation. For developing a predictive model for HF risk in AMI patients, the least absolute shrinkage and selection operator (LASSO) Regression was used to feature selection, and four ML algorithms including Random Forest (RF), Extreme Gradient Boost (XGBoost), Support Vector Machine (SVM), and Logistic Regression (LR) were employed to develop the model on the training set. The performance evaluation of the prediction model was carried out on the training set and the testing set, utilizing metrics including AUC (Area under the receiver operating characteristic curve), calibration plot, and decision curve analysis (DCA). In addition, we used the Shapley Additive Explanations (SHAP) value to determine the importance of the selected features and interpret the optimal model.ResultsA total of 1220 AMI patients were included and 244 (20%) patients developed HF during follow-up. Among the four evaluated ML models, the XGBoost model exhibited exceptional accuracy, with an AUC value of 0.922. The SHAP method showed that left ventricular ejection fraction (LVEF), left ventricular end-systolic diameter (LVDs) and lactate dehydrogenase (LDH) were identified as the three most important characteristics to predict HF risk in AMI patients. Individual risk assessment was performed using SHAP plots and waterfall plot analysis.ConclusionsOur research demonstrates the potential of ML methods in the early prediction of HF risk in AMI patients. Furthermore, it enhances the interpretability of the XGBoost model through SHAP analysis to guide clinical decision-making.

Comparing clinical only and combined clinical laboratory models for ECPR outcomes in refractory cardiac arrest

Extracorporeal cardiopulmonary resuscitation (ECPR) improves survival for prolonged cardiac arrest (CA) but carries significant risks and costs due to ECMO. Previous predictive models have been complex, incorporating both clinical data and parameters obtained after CPR or ECMO initiation. This study aims to compare a simpler clinical-only model with a model that includes both clinical and pre-ECMO laboratory parameters, to refine patient selection and improve ECPR outcomes. Medical records between January 2012 and January 2019 in our institution were retrospectively reviewed. Patients who met the following criteria were enrolled in the ECPR program: age 18–75 years, CCPR started with CA in < 5 min, CA was assumed to be of heart origin, and refractory CA. Survivors had similar underlying diseases and younger age without statistical significance (57.0 vs. 61.0 years, p = 0.117). Survivors had significantly higher rates of initial shockable rhythm, pulseless ventricular tachycardia and ventricular fibrillation, shorter low-flow time (CPR-to-ECMO time), lower lactate levels, and higher initial pH. Survival to discharge was higher for emergency department CA than for out-of-hospital and in-hospital CA (63.3% vs. 35.3%, p = 0.007). Two models were used for evaluating survival to discharge and good neurological outcomes. Model 1, short version based on clinical factors, (S1, survival score 1; F1, function score 1) included the patient’s characteristics before ECPR, whereas Model 2, full version included clinical factors and laboratory data including lactate and pH levels (S2, survival score 2; F2, function score 2). Both Model 1(S1) and Model 2(S2) showed good predictive ability for survival to discharge with areas under the receiver operating characteristic (AUROCs) of 0.79 and 0.83, respectively. Model 1(F1) and Model 2(F2) revealed prediction power for good neurological outcomes, with AUROCs of 0.80 and 0.79, respectively. The AUROCs of survival score Model 1(S1) and 2(S2) and function score Model 1(F1) and 2(F2) were not significantly different. This study demonstrates that clinical factors alone can effectively predict survival to discharge and favorable neurological outcomes at 6 months. This emphasizes the importance of early prognostic evaluation and supports the use of clinical data as a practical tool for clinicians in decision-making for this difficult situation.

A Systematic Review: The Role of Artificial Intelligence in Lung Cancer Screening in Detecting Lung Nodules on Chest X-Rays

Background: Lung cancer remains one of the leading causes of cancer-related deaths worldwide. Artificial intelligence (AI) holds significant potential roles in enhancing the detection of lung nodules through chest X-ray (CXR), enabling earlier diagnosis and improved outcomes. Methods: Papers were identified through a comprehensive search of the Web of Science (WOS), Scopus, and Ovid Medline databases for publications dated between 2020 and 2024. Following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines, 34 studies that met the inclusion criteria were selected for quality assessment and data extraction. Results: AI demonstrated sensitivity rates of 56.4–95.7% and specificities of 71.9–97.5%, with the area under the receiver operating characteristic (AUROC) values between 0.89 and 0.99, compared to radiologists’ mean area under the curve (AUC) of 0.81. AI performed better with larger nodules (&gt;2 cm) and solid nodules, showing higher AUC values for calcified (0.71) compared to non-calcified nodules (0.55). Performance was lower in hilar areas (30%) and lower lung fields (43.8%). A combined AI-radiologist approach improved overall detection rates, particularly benefiting less experienced readers; however, AI showed limitations in detecting ground-glass opacities (GGOs). Conclusions: AI shows promise as a supplementary tool for radiologists in lung nodule detection. However, the variability in AI results across studies highlights the need for standardized assessment methods and diverse datasets for model training. Future studies should focus on developing more precise and applicable algorithms while evaluating the effectiveness and cost-efficiency of AI in lung cancer screening interventions.

P0687 Development and validation of a novel multimodal model based on histology of colonoscopic biopsy to predict primary non-response to infliximab in patients with Crohn’s disease

Abstract Background Crohn’s disease (CD) is a chronic inflammatory condition of the gastrointestinal tract, and while biologics like infliximab (IFX) offer significant benefits, 10-30% of patients remain non-responders initially. This study aims to refine the prediction of treatment responses using artificial intelligence to integrate multimodal information, potentially enhancing treatment precision. Methods We conducted a retrospective analysis of clinical data from Crohn’s disease patients treated at The First Affiliated Hospital of Sun Yat-sen University from October 2016 to October 2021. The endpoint event was defined as primary non-response within 14 weeks of treatment. This study included clinical, laboratory, endoscopic, imaging, and histological data from colonoscopic biopsies. The data were split into a test set and an internal validation set in a 7:3 ratio. A deep learning (DL) model was developed using a multi-instance learning framework to incorporate this multimodal information for predicting primary non-response to IFX. Finally, the multimodal model was validated in an independent external validation cohort.[Figure1] Results The study included 170 patients, with 93 in the internal training set, 38 in the test set, and 39 in the external validation set. The DL model utilizing pathological features achieved an area under the receiver operating characteristic (AUROC) of 0.79 in 5-fold cross-validation. When combining pathological and radiological features, the AUC was 0.76, and integrating clinical features raised the AUC to 0.82. External validation upheld the efficacy of model with the highest AUC at 0.73 for the multimodal model. Conclusion The study developed several DL models to predict primary non-response to IFX with high accuracy. The most effective model integrated pathology, radiomics, and clinical data, supporting targeted treatment strategies and advancing personalized medicine for CD.

DOP083 «Sonorescue»-treatment response in acute severe Ulcerative Colitis measured with ultrasound

Abstract Background First line medical treatment in acute severe ulcerative colitis (ASUC) is intravenous corticosteroids. Patients with insufficient response receives rescue treatment. The decision is made by the treating physician and is based on disease severity and improvement in stool frequency, rectal bleeding, biochemical parameters, and general well-being. Gastrointestinal ultrasound (GIUS) can be used to find the degree of inflammation in ulcerative colitis, but can it be used to detect the lack of treatment response in ASUC? Methods Patients with ASUC were included from June 2020 to Nov 2024 at two hospitals in Bergen, Norway. They were examined with GIUS prior to treatment with iv corticosteroids and after 2-3 days of treatment. Patients with a normal GIUS, endoscopic MAYO&amp;lt;2 or isolated proctitis were excluded. Bowel wall thickness, bowel wall vessel density with colour Doppler, loss of haustration and fatty tissue reaction was registered for every segment of the colon as well as an activity score based on these parameters1. The two main outcomes of the study were the clinical decision to give rescue with infliximab and Lindgren’s criteria for the risk of colectomy2. Results In total 33/64 patients received rescue treatment. In the sigmoid colon the wall was 4.3±1,3 mm in patients receiving rescue treatment and 2.5±1mm in the others (p&amp;lt;0.001). There was also a significant difference in vessel density, haustrations, and fatty tissue reaction. Furthermore, there were significant differences for wall thickness in all the other colon segments including the sum of these measurements. The full score based on all parameters was 15.2±5,2 vs 8.3±5,3 (p&amp;lt;0.001). The areas under the receiver operating curve (AUROC) for wall thickness in the sigmoid colon, total bowel wall thickness and the activity score were 0.87 (0.79-0.96), 0.84 (0.75-0.94) and 0.84 (0.74-0.94), respectively. According to Lindgren’s Criteria 23/64 had a high risk for colectomy. AUROCs for the same parameters were 0.86(0.75-0.96), 0.88(0.80-0.97) and 0.85 (0.76-0.94), respectively. If a low risk of colectomy is required a total bowel wall thickness ≥ 9mm at day 2-3 predicts the risk of colectomy with 100% sensitivity and 50% specificity. For the single measurement in the sigmoid colon and a corresponding sensitivity of 96% (cut off 2.4 mm) the specificity will wall fall to 32.5%. Conclusion GIUS can be used to detect insufficient treatment response in ASUC already after 2-3 days of intravenous corticosteroids which confirms previous findings3. A single measurement in the sigmoid colon could be sufficient, but the combined sum of bowel wall thickness in all colon segments could be better for selecting patients that should get rescue treatment.

Automated extraction of post-stroke functional outcomes from unstructured electronic health records.

Population level tracking of post-stroke functional outcomes is critical to guide interventions that reduce the burden of stroke-related disability. However, functional outcomes are often missing or documented in unstructured notes. We developed a natural language processing (NLP) model that reads electronic health records (EHR) notes to automatically determine the modified Rankin Scale (mRS). We included consecutive patients (⩾18 years) with acute stroke admitted to our center (2015-2024). mRS scores were obtained from the Get With the Guidelines registry and clinical notes (if documented), and used as the gold standard to compare against NLP-generated scores. We used text-based features from notes, along with age, sex, discharge status, and outpatient follow-up to train a logistic regression for prediction of good (0-2) versus poor (3-6) mRS, and a linear regression for the full range of mRS scores. The models were trained for prediction of mRS at hospital discharge and post-discharge. The models were externally validated in a dataset of patients with brain injuries from a different healthcare center. We included 5307 patients, 5006 in train and test and 301 in validation; average age was 69 (SD 15) and 65 (SD 17) years, respectively; 47% female. The logistic regression achieved an area under the receiver operating curve (AUROC) of 0.94 [CI 0.93-0.95] (test) and 0.94 [0.91-0.96] (validation), and the linear model a root mean squared error (RMSE) of 0.91 [0.87-0.94] (test) and 1.17 [1.06-1.28] (validation). The NLP-based model is suitable for use in large-scale phenotyping of stroke functional outcomes and population health research.

Predicting 90-day return to the emergency department in orthopaedic trauma patients in the Southeastern USA: a machine-learning approach

IntroductionReturn-to-acute-care metrics, such as early emergency department (ED) visits, are key indicators of healthcare quality, with ED returns following surgery often considered avoidable and costly events. Proactively identifying patients at...

Benchmarking protein language models for protein crystallization

The problem of protein structure determination is usually solved by X-ray crystallography. Several in silico deep learning methods have been developed to overcome the high attrition rate, cost of experiments and extensive trial-and-error settings, for predicting the crystallization propensities of proteins based on their sequences. In this work, we benchmark the power of open protein language models (PLMs) through the TRILL platform, a be-spoke framework democratizing the usage of PLMs for the task of predicting crystallization propensities of proteins. By comparing LightGBM / XGBoost classifiers built on the average embedding representations of proteins learned by different PLMs, such as ESM2, Ankh, ProtT5-XL, ProstT5, xTrimoPGLM, SaProt with the performance of state-of-the-art sequence-based methods like DeepCrystal, ATTCrys and CLPred, we identify the most effective methods for predicting crystallization outcomes. The LightGBM classifiers utilizing embeddings from ESM2 model with 30 and 36 transformer layers and 150 and 3000 million parameters respectively have performance gains by 3- than all compared models for various evaluation metrics, including AUPR (Area Under Precision-Recall Curve), AUC (Area Under the Receiver Operating Characteristic Curve), and F1 on independent test sets. Furthermore, we fine-tune the ProtGPT2 model available via TRILL to generate crystallizable proteins. Starting with 3000 generated proteins and through a step of filtration processes including consensus of all open PLM-based classifiers, sequence identity through CD-HIT, secondary structure compatibility, aggregation screening, homology search and foldability evaluation, we identified a set of 5 novel proteins as potentially crystallizable.

Prognostic nutritional index as a predictor of mortality in acute ischemic stroke.

Nutritional and immune status have been recognized as important factors influencing the prognosis of acute ischemic stroke. This study aims to evaluate the prognostic value of the Prognostic Nutritional Index (PNI) in predicting 30-day mortality in patients with acute ischemic stroke. A retrospective cohort study was conducted in a tertiary care hospital emergency department. Patients diagnosed with acute ischemic stroke between January 1, 2019, and January 1, 2024, were included. PNI was calculated using serum albumin levels and total lymphocyte count. Data on demographics, clinical parameters, laboratory results, and 30-day mortality were collected from electronic medical records and patient follow-up via phone calls. Appropriate statistical tests were applied based on the data distribution, and p-values < 0.05 were considered statistically significant. A total of 169 patients were included in the study, with a mean age of 74.2 ± 10.5 years, and 55 % were male. Of these patients, 25 (14.8 %) did not survive within the 30-day follow-up period. PNI was significantly lower in deceased patients compared to survivors (p < 0.001). The area under the receiver operating characteristic (AUROC) curve was 0.831, indicating good predictive accuracy for 30-day mortality. The optimal PNI cut-off of 37.21 demonstrated a sensitivity of 72.0 % and a specificity of 81.2 %. The PNI is a reliable predictor of 30-day mortality in acute ischemic stroke patients. Lower PNI values were associated with higher mortality, highlighting the importance of nutritional and immune status in stroke outcomes. PNI could be used in clinical practice to help identify high-risk patients and inform more appropriate treatment decisions.

Effect of Characteristic Inflammatory and Structural Pelvic Magnetic Resonance Imaging Lesions on Expert Assessment of Axial Juvenile Spondyloarthritis.

To evaluate the influence of pelvic magnetic resonance imaging (MRI) findings on axial disease assessment in juvenile spondyloarthritis (JSpA). This was a cross-sectional study of patients with JSpA with suspected axial disease. Three experts reviewed each case and rated their confidence (-3 to +3) in the presence of axial disease, first with clinical data and second with clinical and MRI data. Agreement was defined as ≥ 2/3 clinical experts with a rating of ≤ -1 or ≥ 1, and high confidence agreement as ≤ -2 or ≥ 2. The association of clinical features and both global assessments was tested with modified Poisson regression models. Two hundred seventy-two of 303 cases (89.8%) achieved agreement with clinical data alone. Adding imaging data affected agreement in 38.9% (118/303) and directionality of agreement in 23.4% (71/303). Agreement was facilitated in 26/31 cases and lost in 21/272 cases. Of those 71 cases that changed directionality, 33 changed from axial disease being absent to present and 38 from present to absent. The final model had an area under the receiver-operating characteristic (AUROC) curve of 0.93 and 3 factors were independently associated with expert agreement (HLA-B27: relative risk [RR] 1.41, 95% CI 1.14-1.74; pain improvement with activity: RR 1.27, 95% CI 1.05-1.54; and bone marrow edema on MRI: RR 4.08, 95% CI 2.91-5.73). The addition of imaging data affected directionality and improved high confidence agreement of expert assessment of axial disease. These results underscore the integral role of MRI in the determination of axial disease in JSpA.

Diagnostic Accuracy of a Point-Of-Care aMMP-8 Test for Discriminating Periodontal Health Status in Adults: Validation Trials and Updated Meta-Analysis.

To evaluate the diagnostic accuracy of an active matrix metalloproteinase-8 (aMMP-8) point-of-care oral rinse test (POC-ORT) for predicting periodontitis in treatment-naïve subjects in two independent studies and update a recent meta-analysis. The aMMP-8 POC-ORT index test was performed in a representative population in Hong Kong, China, and a consecutive convenience sample in Shanghai, China. The reference standard was the 2017 World Workshop classification of periodontal diseases. Sensitivity, specificity, and the area under the receiver operating characteristic (AUROC) curve were assessed. The original data were used to update a recent Bayesian meta-analysis following the current Cochrane guideline for diagnostic trials. The GRADE framework was used to interpret the strength and certainty of the evidence. Three-hundred and eighty-four and 390 subjects were enrolled in the Hong Kong and Shanghai studies, respectively; 74.5% and 67.2% had periodontitis. An aMMP-8-positive test predicted periodontitis with an AUROC of 0.661 and 0.669 in the two studies. The updated systematic review and meta-analysis included eight studies and 2048 subjects. After considering the risk of bias, indirectness, inconsistency, imprecision and publication bias, it showed moderate certainty of a sensitivity of 0.59 (95% CrI: 0.42-0.75), a specificity of 0.82 (95% CrI: 0.68-0.93) and a hierarchical summary AUROC of 0.77 (95% CrI: 0.74-0.81). There is moderate certainty that the aMMP-8 POC-ORT test predicts periodontitis with low to moderate sensitivity, moderate to high specificity, and moderate accuracy. Its high false-negative rate does not allow the replacement of clinical examinationswhen available. The moderate to high positive predictive value shows the potential utility of a positive test for self-detection or co-management of periodontitis in a medical setting and its incorporation in multi-test diagnostics. Further investigations are highly warranted.

Factors Associated with Abusive Head Trauma in Young Children Presenting to Emergency Medical Services Using a Large Language Model

Objectives Abusive head trauma (AHT) is a leading cause of death in young children. Analyses of patient characteristics presenting to Emergency Medical Services (EMS) are often limited to structured data fields. Artificial Intelligence (AI) and Large Language Models (LLM) may identify rare presentations like AHT through factors not found in structured data. Our goal was to apply AI and LLM to EMS narrative documentation of young children to detect AHT. Methods This is a retrospective cohort study of EMS transports of children <36 months of age with a diagnosis of head injury from the 2018–2019 ESO Research Data Collaborative. Non-abusive closed head injury (NA-CHI) was distinguished from AHT and child maltreatment (AHT-CAN) through 2 expert reviewers; kappa statistic (k) assessed inter-rater reliability. A Natural Language Processing (NLP) framework using an LLM augmented with expert derived n-grams was developed to identify AHT-CAN. We compared test characteristics (sensitivity, specificity, negative predictive value (NPV)) between this NLP framework to a Generative Pretrained Transformer (GPT) or n-grams only models to detect AHT-CAN. Association of specific word tokens with AHT-CAN was analyzed using Pearson’s chi-square. Area Under the Receiver Operator Curve (AUROC) and Area Under the Precision-Recall Curve (AUPRC) are also reported. Results There were 1082 encounters in our cohort; 1030 (95.2%) NA-CHI and 52 (4.8%) AHT-CAN. Inter-rater agreement was substantial (k = 0.71). The augmented NLP framework had a specificity and sensitivity of 72.4% and 92.3%, respectively with a NPV of 99.5%. In comparison, the GPT model had a sensitivity of 69.2%, specificity of 97.1% and NPV of 98.4% and n-grams alone had a sensitivity of 53.8%, specificity of 62.0%, NPV of 96.4%. AUROC was 0.91 and AUPRC was 0.52. A total of 44 n-grams and bi-grams were positively associated with AHT-CAN including “domestic,” “various,” “bruise,” “cheek,” “multiple,” “doa,” “not respond,” “see EMS.” Conclusions AI and LLMs have high sensitivity and specificity to detect AHT-CAN in EMS free-text narratives. Words associated with physical signs of trauma are strongly associated with AHT-CAN. LLMs augmented with a list of n-grams may help EMS identify signs of trauma that aid in the detection of AHT in young children.

Risk Estimation of Severe Primary Graft Dysfunction in Heart Transplant Recipients Using a Smartphone.

Currently, there are no standardized guidelines for graft allocation in heart transplants (HTxs), particularly when considering organs from marginal donors and donors after cardiocirculatory arrest. This complexity highlights the need for an effective risk analysis tool for primary graft dysfunction (PGD), a severe complication in HTx. Existing score systems for predicting PGD lack superior predictive capability and are often too complex for routine clinical use. This study sought to develop a user-friendly score integrating variables from these systems to enhance the efficacy of the organ allocation process. Severe PGD was defined as the need for mechanical circulatory support and/or death from an unknown etiology within the first 24 hours following HTx. We used a meta-analytical approach to create a derivation cohort to identify risk factors. We then applied a logistic regression analysis to generate an equation predicting severe PGD risk. We used our previous experience in HTx to create a validation cohort. Subsequently, we implemented the formula in a smartphone application. The meta-analysis comprising six studies revealed a 10.5% ( 95% confidence interval (CI): 5.3-12.4) incidence rate of severe PGD and related 30-day mortality of 38.6%. Eleven risk factors were identified: female donors, female donor to male recipient, undersized donor, donor age, recipient on ventricular assist device support, recipient on amiodarone treatment, recipient with diabetes and renal dysfunction, re-sternotomy, graft ischemic time, and bypass time. An equation to predict the risk, including the 11 parameters (GREF-11), was created using logistic regression models and validated based on our experience involving 116 patients. In our series, 29 recipients (25%) required extracorporeal membrane oxygenation support within 24 hours post-HTx. The overall 30-day mortality was 4.3%, 3.4%, and 6.8% in the non-PGD and severe PGD groups, respectively. The area under the receiver operating characteristic (AU-ROC) curve of the model in the validation cohort was 0.804. The GREF-11 application should offer HTx teams several benefits, including standardized risk assessment and bedside clinical decision support, thereby helping minimize the risk of severe PGD post-HTx.

Development of novel nomograms for predicting prostate cancer in biopsy-naive patients with PSA < 10 ng/ml and PI-RADS ≤ 3 lesions.

To develop novel nomograms for predicting prostate cancer (PCa) and clinically significant prostate cancer (csPCa) in patients with prostate-specific antigen (PSA) < 10 ng/ml and PI-RADS v2.1 score ≤ 3. We retrospectively collected data from 327 men with PSA < 10 ng/ml and PI-RADS score ≤ 3 from June 2020 to June 2024 in our hospital. Clinical data were compared among the PI-RADS scores 1-3 population, PI-RADS scores 1-2 population, and PI-RADS score 3 population. Logistic regression analyses were conducted to identify independent risk factors for PCa or csPCa, and nomograms were subsequently developed. The nomograms were evaluated via receiver operating curves (ROC), calibration curves, and decision curve analysis (DCA). Internal validation was conducted using bootstrap methods. Among the 327 patients, 224 (68.50%) were diagnosed with benign, 65 (19.87%) with csPCa, and 38 (11.62%) with clinically insignificant prostate cancer (cisPCa). Prostate-specific antigen density (PSAD), lesion volume (LV), lesion location, and apparent diffusion coefficient (ADC) were found to be independent risk factors for PCa and csPCa in PI-RADS scores 1-3 population. PSAD and lesion location were independent risk factors for PCa in the PI-RADS scores 1-2 population, while PSAD, lesion location and ADC were independent risk factors for PCa in the PI-RADS score 3 population. Four nomograms were established based on these variables. For the population with PI-RADS scores 1-3, the area under the ROC (AUC) for predicting PCa and csPCa was 0.78 and 0.79, respectively. For patients with PI-RADS scores 1-2, the AUC for predicting PCa was 0.75. For patients with PI-RADS score 3, the AUC for predicting PCa was 0.78. The calibration curves revealed good concordance between the predicted probability and the actual probability. DCA demonstrated the net benefit of nomograms. Internal validation revealed strong discrimination of the nomograms. We developed novel nomograms with acceptable discriminability for predicting PCa and csPCa in patients with PSA < 10 ng/ml and PI-RADS score ≤ 3. These models can assist urologists in determining the necessity of prostate biopsy.

Predicting the risk of type 2 diabetes mellitus (T2DM) emergence in 5 years using mammography images: a comparison study between radiomics and deep learning algorithm.

The prevalence of type 2 diabetes mellitus (T2DM) has been steadily increasing over the years. We aim to predict the occurrence of T2DM using mammography images within 5 years using two different methods and compare their performance. We examined 312 samples, including 110 positive cases (developed T2DM after 5 years) and 202 negative cases (did not develop T2DM) using two different methods. In the first method, a radiomics-based approach, we utilized radiomics features and machine learning (ML) algorithms. The entire breast region was chosen as the region of interest for extracting radiomics features. Then, a binary breast image was created from which we extracted 668 features and analyzed them using various ML algorithms. In the second method, a complex convolutional neural network (CNN) with a modified ResNet architecture and various kernel sizes was applied to raw mammography images for the prediction task. A nested, stratified five-fold cross-validation was done for both parts A and B to compute accuracy, sensitivity, specificity, and area under the receiver operating curve (AUROC). Hyperparameter tuning was also done to enhance the model's performance and reliability. The radiomics approach's light gradient boosting model gave 68.9% accuracy, 30.7% sensitivity, 89.5% specificity, and 0.63 AUROC. The CNN method achieved an AUROC of 0.58 over 20 epochs. Radiomics outperformed CNN by 0.05 in terms of AUROC. This may be due to the more straightforward interpretability and clinical relevance of predefined radiomics features compared with the complex, abstract features learned by CNNs.

DOME: Directional medical embedding vectors from Electronic Health Records.

The increasing availability of Electronic Health Record (EHR) systems has created enormous potential for translational research. Recent developments in representation learning techniques have led to effective large-scale representations of EHR concepts along with knowledge graphs that empower downstream EHR studies. However, most existing methods require training with patient-level data, limiting their abilities to expand the training with multi-institutional EHR data. On the other hand, scalable approaches that only require summary-level data do not incorporate temporal dependencies between concepts. We introduce a DirectiOnal Medical Embedding (DOME) algorithm to encode temporally directional relationships between medical concepts, using summary-level EHR data. Specifically, DOME first aggregates patient-level EHR data into an asymmetric co-occurrence matrix. Then it computes two Positive Pointwise Mutual Information (PPMI) matrices to correspondingly encode the pairwise prior and posterior dependencies between medical concepts. Following that, a joint matrix factorization is performed on the two PPMI matrices, which results in three vectors for each concept: a semantic embedding and two directional context embeddings. They collectively provide a comprehensive depiction of the temporal relationship between EHR concepts. We highlight the advantages and translational potential of DOME through three sets of validation studies. First, DOME consistently improves existing direction-agnostic embedding vectors for disease risk prediction in several diseases, for example achieving a relative gain of 5.5% in the area under the receiver operating characteristic (AUROC) for lung cancer. Second, DOME excels in directional drug-disease relationship inference by successfully differentiating between drug side effects and indications, correspondingly achieving relative AUROC gain over the state-of-the-art methods by 10.8% and 6.6%. Finally, DOME effectively constructs directional knowledge graphs, which distinguish disease risk factors from comorbidities, thereby revealing disease progression trajectories. The source codes are provided at https://github.com/celehs/Directional-EHR-embedding.

Global climate change and its impact on the distribution and efficacy of Bacillus thuringiensis as a biopesticide.

This study is the first modeling exercise to assess the impacts of climate change on the current and future global distribution of Bacillus thuringiensis (Bt). Bt is a common Gram-positive, rod-shaped bacterium widely distributed in various environments, including soil and water. It is widely recognized as a source of effective and safe agricultural biopesticides for pest management in various climatic regions globally. In the present work, ensemble species distribution models were developed for Bt based on the generalized linear model (GLM), generalized boosting model (GBM), random forest (RF), and maximum entropy (MaxEnt) under two distinct scenarios, SSP2-4.5 (optimistic) and SSP5-8.5 (pessimistic) for the year of 2050, 2070, and 2090. The performance of our models was evaluated based on true skill statistics (TSS) and the area under the receiver-operator curve (AUC) indices. Both AUC and TSS values were observed in an acceptable range, with AUC at 0.84 and TSS at 0.512, respectively. Results indicate that most of the areas currently suitable for Bt will likely remain stable in the future, particularly Central America, Central and South Africa, South Asia, and parts of Oceania. Norway, Peru, and the UK will have notable habitat gains by 2090 based on SSP2-4.5 and SSP5-8.5 scenarios. On the contrary, Serbia, Guinea, Poland, Croatia, Spain and Romania showed notable losses under both scenarios. Our results underscore Bt potential to improve pest control, crop yields, and environmental sustainability, especially in regions where agriculture is predominant. Our research highlights the need to understand ecological dynamics for future conservation and agricultural planning in the face of climate change.

AI‐assisted diagnosis of multiple sclerosis via optical coherence tomography

Aims/Purpose: This study aims to evaluate the ability of inter‐eye asymmetry and retinal structure thicknesses, obtained via the Posterior Pole protocol in Optical Coherence Tomography (OCT), to distinguish between healthy controls and patients with recently diagnosed multiple sclerosis (MS). Additionally, it aims to develop a convolutional neural network (CNN) for assisted diagnosis using these measurements.Methods: Two cohorts, one with relapsing‐remitting MS (RRMS) patients (n = 79) and another with healthy controls (n = 69), were recruited. Structural retinal measurements were obtained using the Spectralis OCT device. The analysis focused on thickness differences in nine retinal layers and their inter‐eye asymmetry. Statistical methods included the Shapiro‐Wilk test, Student's t‐test, χ2 test, and area under the receiver operating characteristic (AUROC) curve. A CNN was trained to classify patients and controls based on the most discriminant retinal measurements.Results: The study found significant thinning in the ganglion cell layer (GCL) and inner plexiform layer (IPL) of RRMS patients compared to controls, with AUROC values of 0.82 and 0.78, respectively. Inter‐eye asymmetry was also notable in these layers, particularly the GCL (AUROC = 0.75) and IRL (AUROC = 0.74). The CNN, utilizing GCL thickness and IPL inter‐eye difference as inputs, achieved an accuracy of 0.87, sensitivity of 0.82, and specificity of 0.92.Conclusions: The study demonstrates that neuroretinal thinning and inter‐eye differences in specific retinal layers, measurable by the Posterior Pole protocol in OCT, can effectively discriminate between MS patients and healthy controls. The CNN developed shows high accuracy in MS diagnosis, supporting the potential of OCT and AI integration in clinical settings. Further research is needed to validate these findings across larger and more diverse populations.