Factors affecting subjective cognitive decline: an automated machine learning approach

To construct and validate the best predictive model for 28-day death risk in patients with septic shock based on different supervised machine learning algorithms. The patients with septic shock meeting the Sepsis-3 criteria were selected from Medical Information Mart for Intensive Care-IV v2.0 (MIMIC-IV v2.0). According to the principle of random allocation, 70% of these patients were used as the training set, and 30% as the validation set. Relevant predictive variables were extracted from three aspects: demographic characteristics and basic vital signs, serum indicators within 24 hours of intensive care unit (ICU) admission and complications possibly affecting indicators, functional scoring and advanced life support. The predictive efficacy of models constructed using five mainstream machine learning algorithms including decision tree classification and regression tree (CART), random forest (RF), support vector machine (SVM), linear regression (LR), and super learner [SL; combined CART, RF and extreme gradient boosting (XGBoost)] for 28-day death in patients with septic shock was compared, and the best algorithm model was selected. The optimal predictive variables were determined by intersecting the results from LASSO regression, RF, and XGBoost algorithms, and a predictive model was constructed. The predictive efficacy of the model was validated by drawing receiver operator characteristic curve (ROC curve), the accuracy of the model was assessed using calibration curves, and the practicality of the model was verified through decision curve analysis (DCA). A total of 3 295 patients with septic shock were included, with 2 164 surviving and 1 131 dying within 28 days, resulting in a mortality of 34.32%. Of these, 2 307 were in the training set (with 792 deaths within 28 days, a mortality of 34.33%), and 988 in the validation set (with 339 deaths within 28 days, a mortality of 34.31%). Five machine learning models were established based on the training set data. After including variables at three aspects, the area under the ROC curve (AUC) of RF, SVM, and LR machine learning algorithm models for predicting 28-day death in septic shock patients in the validation set was 0.823 [95% confidence interval (95%CI) was 0.795-0.849], 0.823 (95%CI was 0.796-0.849), and 0.810 (95%CI was 0.782-0.838), respectively, which were higher than that of the CART algorithm model (AUC = 0.750, 95%CI was 0.717-0.782) and SL algorithm model (AUC = 0.756, 95%CI was 0.724-0.789). Thus above three algorithm models were determined to be the best algorithm models. After integrating variables from three aspects, 16 optimal predictive variables were identified through intersection by LASSO regression, RF, and XGBoost algorithms, including the highest pH value, the highest albumin (Alb), the highest body temperature, the lowest lactic acid (Lac), the highest Lac, the highest serum creatinine (SCr), the highest Ca2+, the lowest hemoglobin (Hb), the lowest white blood cell count (WBC), age, simplified acute physiology score III (SAPS III), the highest WBC, acute physiology score III (APS III), the lowest Na+, body mass index (BMI), and the shortest activated partial thromboplastin time (APTT) within 24 hours of ICU admission. ROC curve analysis showed that the Logistic regression model constructed with above 16 optimal predictive variables was the best predictive model, with an AUC of 0.806 (95%CI was 0.778-0.835) in the validation set. The calibration curve and DCA curve showed that this model had high accuracy and the highest net benefit could reach 0.3, which was significantly outperforming traditional models based on single functional score [APS III score, SAPS III score, and sequential organ failure assessment (SOFA) score] with AUC (95%CI) of 0.746 (0.715-0.778), 0.765 (0.734-0.796), and 0.625 (0.589-0.661), respectively. The Logistic regression model, constructed using 16 optimal predictive variables including pH value, Alb, body temperature, Lac, SCr, Ca2+, Hb, WBC, SAPS III score, APS III score, Na+, BMI, and APTT, is identified as the best predictive model for the 28-day death risk in patients with septic shock. Its performance is stable, with high discriminative ability and accuracy.

To construct machine learning prediction model for sepsis-associated encephalopathy (SAE), and analyze the application value of the model on early identification of SAE risk in elderly septic patients. Patients aged over 60 years with a primary diagnosis of sepsis admitted to intensive care unit (ICU) from 2008 to 2023 were selected from Medical Information Mart for Intensive Care-IV 2.2 (MIMIC-IV 2.2). Demographic variables, disease severity scores, comorbidities, interventions, laboratory indicators, and hospitalization details were collected. Key factors associated with SAE were identified using univariate Logistic regression analysis. The data were randomly divided into training and validation sets in a 7 : 3 ratio. Multivariable Logistic regression analysis was conducted in the training set and visualized using a nomogram model for prediction of SAE. The discrimination of the model was evaluated in the validation set using the receiver operator characteristic curve (ROC curve), and its calibration was assessed using calibration curve. Furthermore, multiple machine learning algorithms, including multi-layer perceptron (MLP), support vector machine (SVM), naive bayes (NB), gradient boosting machine (GBM), random forest (RF), and extreme gradient boosting (XGB), were constructed in the training set. Their predictive performance was subsequently evaluated on the validation set. Taking the XGB model as an example, the interpretability of the model through the SHapley Additive exPlanations (SHAP) algorithm was enhanced to identify the key predictive factors and their contributions. A total of 2 204 septic patients were finally enrolled, of whom 840 developed SAE (38.1%). A total of 21 variables associated with SAE were screened through univariate Logistic regression analysis. Multivariable Logistic regression analysis showed that endotracheal intubation [odds ratio (OR) = 0.40, 95% confidence interval (95%CI) was 0.19-0.88, P < 0.001], oxygen therapy (OR = 0.76, 95%CI was 0.53-0.95, P = 0.023), tracheotomy (OR = 0.20, 95%CI was 0.07-0.53, P < 0.001), continuous renal replacement therapy (CRRT; OR = 0.32, 95%CI was 0.15-0.70, P < 0.001), cerebrovascular disease (OR = 0.31, 95%CI was 0.16-0.60, P < 0.001), rheumatic disease (OR = 0.44, 95%CI was 0.19-0.99, P < 0.001), male (OR = 0.68, 95%CI was 0.54-0.86, P = 0.001), and maximum anion gap (AG; OR = 0.95, 95%CI was 0.93-0.97, P < 0.001) were associated with an decreased probability of SAE, and age (OR = 1.05, 95%CI was 1.03-1.06, P < 0.001), acute physiology score III (APSIII; OR = 1.02, 95%CI was 1.01-1.02, P < 0.001), Oxford acute severity of illness score (OASIS; OR = 1.04, 95%CI was 1.03-1.06, P < 0.001), and length of hospital stay (OR = 1.01, 95%CI was 1.01-1.02, P < 0.001) were associated with an increased probability of SAE. A nomogram model was constructed based on these variables. In the validation set, ROC curve analysis showed that the model achieved an area under the ROC curve (AUC) of 0.723, and the calibration curve showed good consistency between the predicted probability of the model and the observed probability. Among the machine learning algorithms, including MLP, SVM, NB, GBM, RF, and XGB, the SVM model and RF model demonstrated relatively good predictive performance, with AUC of 0.748 and 0.739, respectively, and the sensitivity was both exceeding 85%. The predictive performance of the XGB model was explained through SHAP analysis, and the results indicated that APSIII score (SHAP value was 0.871), age (SHAP value was 0.521), and OASIS score (SHAP value was 0.443) were important factors affecting the predictive performance of the model. The machine learning-based SAE prediction model exhibits good predictive capability and holds significant application value for the early identification of SAE risk in elderly septic patients.

To evaluate the value of a machine learning model using enhanced CT radiomics features in the prediction of lymphovascular invasion (LVI) of esophageal squamous cell carcinoma (ESCC) before treatment. We reviewed and analyzed the enhanced CT images of 258 ESCC patients from June 2017 to December 2019. We randomly assigned the patients in a ratio of 7:3 to a training set (182 cases) and a validation (76 cases) set. Clinical risk factors and CT image characteristics were recorded, and multifactor logistic regression was used to screen independent risk factors of LVI of ESCC patients. We extracted the CT radiomics features using the FAE software and screened radiomics features using maximum relevance and minimum redundancy (MRMR) and least absolute shrinkage and selection operator (LASSO) algorithms, and finally, the radiomics labels of each patient were established. Five machine learning algorithms, namely, support vector machine (SVM), K-nearest neighbor (KNN), logistic regression (LR), Gauss naive Bayes (GNB), and multilayer perceptron (MLP), were used to construct the model of radiomics labels, and its clinical features were screened. The predictive efficacy of the machine learning model for LVI of ESCC was evaluated using the receiver operating characteristic (ROC) curve. Tumor thickness [OR = 1.189, 95% confidence interval (CI) 1.060-1.351, P = 0.005], tumor-to-normal wall enhancement ratio (TNR) (OR = 2.966, 95% CI 1.174-7.894, P = 0.024), and clinical N stage (OR = 5.828, 95% CI 1.752-20.811, P= 0.005) were determined as independent risk factors of LVI. We extracted 1,316 features from preoperative enhanced CT images and selected 14 radiomics features using MRMR and LASSO to construct the radiomics labels. In the test set, SVM, KNN, LR, and GNB showed high predictive performance, while the MLP model had poor performance. In the training set, the area under the curve (AUC) values were 0.945 and 0.905 in the KNN and SVM models, but these decreased to 0.866 and 0.867 in the validation set, indicating significant overfitting. The GNB and LR models had AUC values of 0.905 and 0.911 in the training set and 0.900 and 0.893 in the validation set, with stable performance and good fitting and predictive ability. The MLP model had AUC values of 0.658 and 0.674 in the training and validation sets, indicating poor performance. A multiscale combined model constructed using multivariate logistic regression has an AUC of 0.911 (0.870-0.951) and 0.893 (0.840-0.962), accuracy of 84.4% and 79.7%, sensitivity of 90.8% and 87.1%, and specificity of 80.5% and 79.0% in the training and validation sets, respectively. Machine learning models can preoperatively predict the condition of LVI effectively in patients with ESCC based on enhanced CT radiomics features. The GNB and LR models exhibit good stability and may bring a new way for the non-invasive prediction of LVI condition in ESCC patients before treatment.

To develop an integrated model that combines multimodal fundus image features for accurately predicting the individualized risk of progression from early to late stages of age-related macular degeneration (AMD). A retrospective analysis was conducted on the data of 324 patients with AMD. The patients were randomly divided into a training set (n = 227) and a validation set (n = 97) at a ratio of 7:3. The follow-up period was 3 years, and patients with disease progression were defined as the progression group. In the training set, indicators related to prognosis were screened through univariate analysis. After variable compression by LASSO regression, independent influencing factors for poor prognosis were determined using multivariate logistic regression. Random Forest, Support Vector Machine, XG BOOST, and K-Nearest Neighbor algorithm prediction models were constructed using Python software. The performance of the models was evaluated by the area under the receiver operating characteristic curve (AUC), and the optimal model was selected. There were no significant differences in the baseline characteristics between the training set and the validation set patients (P > 0.05), indicating comparability. In the training set, multivariate logistic regression analysis showed that pigmentary abnormalities, the total area of drusen in the macular area, and the ellipsoid zone were independent risk factors for disease progression (P < 0.05), while subfoveal choroidal thickness and choroidal capillary blood flow density were independent protective factors (P < 0.05). The AUC values of the Random Forest model (0.779 in the training set and 0.700 in the validation set) were significantly higher than those of the K-Nearest Neighbor algorithm (0.717 in the training set and 0.596 in the validation set), the Support Vector Machine model (0.768 in the training set and 0.646 in the validation set), and XG BOOST (0.702 in the training set and 0.762 in the validation set), making it the optimal prediction model. In this study, an AMD progression prediction model based on multimodal fundus images was successfully developed, which can effectively identify patients at high risk of progression and provide a new paradigm for clinical individualized precision medicine.

Background: Coronavirus disease 2019 (COVID-19) is an emerging infectious disease and has spread worldwide. Clinical risk factors associated with the severity in COVID-19 patients have not yet been well delineated. The aim of this study was to explore the risk factors related with the progression of severe COVID-19 and establish a prediction model for severity in COVID-19 patients. Methods: We retrospectively recruited patients with confirmed COVID-19 admitted in Enze Hospital, Taizhou Enze Medical Center (Group) and Nanjing Drum Tower Hospital between January 24 and March 12, 2020. Take the Taizhou cohort as the training set and the Nanjing cohort as the validation set. Severe case was defined based on the World Health Organization Interim Guidance Report criteria for severe pneumonia. The patients were divided into severe and non-severe groups. Epidemiological, laboratory, clinical, and imaging data were recorded with data collection forms from the electronic medical record. The predictive model of severe COVID-19 was constructed, and the efficacy of the predictive model in predicting the risk of severe COVID-19 was analyzed by the receiver operating characteristic curve (ROC). Results: A total of 402 COVID-19 patients were included in the study, including 98 patients in the training set (Nanjing cohort) and 304 patients in the validation set (Nanjing cohort). There were 54 cases (13.43%) in severe group and 348 cases (86.57%) in non-severe group. Logistic regression analysis showed that body mass index (BMI) and lymphocyte count were independent risk factors for severe COVID-19 (all P &lt; 0.05). Logistic regression equation based on risk factors was established as follows: Logit (BL)=–5.552–5.473 × L + 0.418 × BMI. The area under the ROC curve (AUC) of the training set and the validation set were 0.928 and 0.848, respectively (all P &lt; 0.001). The model was simplified to get a new model (BMI and lymphocyte count ratio, BLR) for predicting severe COVID-19 patients, and the AUC in the training set and validation set were 0.926 and 0.828, respectively (all P &lt; 0.001). Conclusions: Higher BMI and lower lymphocyte count are critical factors associated with severity of COVID-19 patients. The simplified BLR model has a good predictive value for the severe COVID-19 patients. Metabolic factors involved in the development of COVID-19 need to be further investigated.

BackgroundGenetic neurodevelopmental disorders (g-NDDs), a complex group of idiopathic syndromes with a heterogeneous genetic etiology, are defined as global developmental delay/intellectual disability (GDD/ID) with other common neurodevelopmental comorbidity (NDC), such as autism spectrum disorder (ASD). Although significant progress has been made in trio-based whole-exome sequencing (trio-WES) that enable the detection of exon-level variants, the diagnostic efficacy of using trio-WES for g-NDDs is still not satisfactory. Therefore, exploring key phenotypic variables for forecasting the diagnostic probability of trio-WES is extremely necessary to implement personalized diagnosis for children with g-NDDs.MethodsA total of 265 g-NDDs children who received trio-WES at Sun Yat-sen Memorial Hospital between Sep 2016 and Dec 2022 were enrolled and clustered temporally into training and internal validation sets [163 cases (Oct 2016 ~ Dec 2022) and 102 cases (Sep 2016 ~ Sep 2018), respectively]. A total of 97 g-NDDs children who underwent trio-WES at Weierkang Children’s Rehabilitation Center between Jan 2023 and Dec 2023 were enrolled and served as an independent external validation set. Univariate and multivariate logistic regression were conducted in the training set to screen out independent diagnosis-related phenotypic signifiers and establish an alignment diagram model. The model was further validated in internal and external validation sets.ResultsThrough univariate and multivariate analyses, independent diagnosis-related predictive signifiers, including GDD/ID severity, NDC complexity, ASD, and head circumference abnormality, were identified in the training cohort and used to construct a model. The model showed good discrimination power with an area under the ROC curves (AUC) in the training set of 0.821 (95% CI: 0.756–0.886), yielding a F1 score of 0.76. The model also showed powerful prediction in both the internal (AUC: 0.905 with 95% CI: 0.842–0.968 and F1 score: 0.77) and external validation sets (AUC: 0.919 with 95% CI: 0.858–0.979 and F1 score: 0.79).ConclusionWe found the potential linear relationship between trio-WES-diagnostic rates and the phenotypic enrichments in g-NDDs patients for the first time, indicating the possibility of applying a logistic regression model based on phenotypic features to predict the personalized diagnostic rates of using trio-WES in children with g-NDDs.

Gallbladder cholesterol polyp (GCP) and gallbladder adenoma (GA) are easily confused in clinical diagnosis. This study aims to establish a nomogram prediction model for preoperative prediction of the risk of GA patients. We retrospectively collected clinical data of GCP or GA patients who underwent laparoscopic cholecystectomy (LC) between January 2020 and April 2023. We compared and analyzed the differences between the GCP group and the GA group. The data were divided into a training set and a validation set in a 7:3 ratio. Independent risk factors were determined using LASSO and Logistic regression analysis, and a nomogram model was established. The model was comprehensively validated and evaluated using the area under the ROC curve (AUC), Hosmer-Lemeshow test and clinical decision curve analysis (DCA). This study ultimately included 497 patients. The independent predictors of the nomogram model include blood type (O-type blood, OR 2.00, 95% CI 1.02-3.94; P = 0.046), number of lesions (solitary, OR 2.11; 95% CI 1.08-4.12; P = 0.033), sessile polyp (OR 2.04; 95% CI 1.06-3.92; P = 0.033), age (OR 1.10; 95% CI 1.07-1.20; P < 0.001), diameter (OR 1.30; 95% CI 1.17-1.45; P < 0.001). For the training and validation set, the area under the ROC curve (AUC) was 0.843 and 0.837, respectively, and the P-value for the Hosmer-Lemeshow test was 0.056 and 0.300, respectively. In addition, the calibration curve and DCA curve indicate that the model has accurate predictive ability and reliable clinical practicality. The blood type, number of lesions, sessile polyp, age and diameter are significant risk factors for GA. This nomogram model can use simple and readily available clinical data to predict the risk of having GA and can assist in guiding surgical decisions.

IntroductionThe objective of this study is to develop a model based on indicators in the routine examination of neonates to effectively predict neonatal apnea.MethodsWe retrospectively analysed 8024 newborns from the MIMIC IV database, building logistic regression models and decision tree models. The performance of the model is examined by decision curves, calibration curves and ROC curves. Variables were screened by stepwise logistic regression analysis and LASSO regression.ResultsA total of 7 indicators were ultimately included in the model: gestational age, birth weight, ethnicity, gender, monocytes, lymphocytes and acetaminophen. The mean AUC (the area under the ROC curve) of the 5-fold cross-validation of the logistic regression model in the training set and the AUC in the validation set are 0.879 and 0.865, respectively. The mean AUC (the area under the ROC curve) of the 5-fold cross-validation of the decision tree model in the training set and the AUC in the validation set are 0.861 and 0.850, respectively. The calibration and decision curves in the two cohorts also demonstrated satisfactory predictive performance of the model. However, the logistic regression model performs relatively well.DiscussionOur results proved that blood indicators were valuable and effective predictors of neonatal apnea, which could provide effective predictive information for medical staff.

High-resolution NMR metabolomics of patients with subjective cognitive decline plus: Perturbations in the metabolism of glucose and branched-chain amino acids

To establish a machine learning model based on extreme gradient boosting (XGBoost) algorithm for early prediction of severe acute pancreatitis (SAP), and explore its predictive efficiency. A retrospective cohort study was conducted. The patients with acute pancreatitis (AP) who admitted to the First Affiliated Hospital of Soochow University, the Second Affiliated Hospital of Soochow University and Changshu Hospital Affiliated to Soochow University from January 1, 2020 to December 31, 2021 were enrolled. Demography information, etiology, past history, and clinical indicators and imaging data within 48 hours of admission were collected according to the medical record system and image system, and the modified CT severity index (MCTSI), Ranson score, bedside index for severity in acute pancreatitis (BISAP) and acute pancreatitis risk score (SABP) were calculated. The data sets of the First Affiliated Hospital of Soochow University and Changshu Hospital Affiliated to Soochow University were randomly divided into training set and validation set according to 8 : 2. Based on XGBoost algorithm, the SAP prediction model was constructed on the basis of hyperparameter adjustment by 5-fold cross validation and loss function. The data set of the Second Affiliated Hospital of Soochow University was served as independent test set. The predictive efficacy of the XGBoost model was evaluated by drawing the receiver operator characteristic curve (ROC curve), and compared it with the traditional AP related severity score; variable importance ranking diagram and Shapley additive explanation (SHAP) diagram were drawn to visually explain the model. A total of 1 183 AP patients were enrolled finally, of which 129 (10.9%) developed SAP. Among the patients from the First Affiliated Hospital of Soochow University and Changshu Hospital Affiliated to Soochow University, there were 786 patients in the training set and 197 in the validation set; 200 patients from the Second Affiliated Hospital of Soochow University were used as the test set. Analysis of all three datasets showed that patients who advanced to SAP exhibited pathological manifestation such as abnormal respiratory function, coagulation function, liver and kidney function, and lipid metabolism. Based on the XGBoost algorithm, an SAP prediction model was constructed, and ROC curve analysis showed that the accuracy for prediction of SAP reached 0.830, the area under the ROC curve (AUC) was 0.927, which was significantly improved compared with the traditional scoring systems including MCTSI, Ranson, BISAP and SABP, the accuracy was 0.610, 0.690, 0.763, 0.625, and the AUC was 0.689, 0.631, 0.875, and 0.770, respectively. The feature importance analysis based on the XGBoost model showed that the top ten items ranked by the importance of model features were admission pleural effusion (0.119), albumin (Alb, 0.049), triglycerides (TG, 0.036), Ca2+ (0.034), prothrombin time (PT, 0.031), systemic inflammatory response syndrome (SIRS, 0.031), C-reactive protein (CRP, 0.031), platelet count (PLT, 0.030), lactate dehydrogenase (LDH, 0.029), and alkaline phosphatase (ALP, 0.028). The above indicators were of great significance for the XGBoost model to predict SAP. The SHAP contribution analysis based on the XGBoost model showed that the risk of SAP increased significantly when patients had pleural effusion and decreased Alb. A SAP prediction scoring system was established based on the machine automatic learning XGBoost algorithm, which can predict the SAP risk of patients within 48 hours of admission with good accuracy.

Occupational contact dermatitis (OCD) is a prevalent work-related skin condition among nurses and remains a significant occupational health issue due to its impact on well-being, productivity, and workforce sustainability. However, reliable tools for early risk stratification in this population are lacking. This study aimed to develop and validate a nomogram-based prediction model to estimate the individual risk of OCD among nurses. A multicenter cross-sectional survey was conducted among 2,852 nurses from 40 hospitals across China. Participants were randomly assigned to a training cohort (n = 2,000) and a validation cohort (n = 852). Independent predictors were identified using univariate and multivariable logistic regression analyses. A nomogram was constructed based on the final multivariable model. Model performance was assessed using the area under the ROC curve (AUC), bootstrapped calibration plots, and decision curve analysis (DCA). Nine predictors were independently associated with OCD: age, dermatitis history, glove type, glove-wearing hours, handwashing frequency during work, hospital level, hand-cream habit, baseline skin condition, and sleep duration. The model showed excellent discrimination (AUC = 0.925 in the training set; 0.931 in the validation set). Calibration curves demonstrated close agreement between predicted and observed risks. DCA indicated consistently higher net benefit compared with the "treat-all" and "treat-none" strategies across wide threshold probability ranges (0.01-0.98 in the training set; 0.02-0.96 in the validation set). The resulting nomogram provides an intuitive, point-based tool for individualized OCD risk prediction. A robust, well-validated prediction model and nomogram were developed to estimate OCD risk among nurses. This tool may support occupational health screening, early risk identification, and targeted preventive strategies in healthcare institutions.

While a binary classifier aims to distinguish positives from negatives, a ranker orders instances from high to low expectation that the instance is positive. Most classification models in machine learning output some score of ‘positiveness’, and hence can be used as rankers. Conversely, any ranker can be turned into a classifier if we have some instance-independent means of splitting the ranking into positive and negative segments. This could be a fixed score threshold; a point obtained from fixing the slope on the ROC curve; the break-even point between true positive and true negative rates; to mention just a few possibilities.These connections between ranking and classification notwithstanding, there are considerable differences as well. Classification performance on n examples is measured by accuracy, an O(n) operation; ranking performance, on the other hand, is measured by the area under the ROC curve (AUC), an O(n logn) operation. The model with the highest AUC does not necessarily dominate all other models, and thus it is possible that another model would achieve a higher accuracy for certain operating conditions, even if its AUC is lower.However, within certain model classes good ranking performance and good classification performance are more closely related than suggested by the previous remarks. For instance, there is evidence that certain classification models, while designed to optimise accuracy, in effect optimise an AUC-based loss function [1]. It has also been known for some time that decision tree yield convex training set ROC curves by construction [2], and thus optimising training set accuracy is likely to lead to good training set AUC. In this talk I will investigate the relation between ranking and classification more closely.I will also consider the connection between ranking and probability estimation. The quality of probability estimates can be measured by, e.g., mean squared error in the probability estimates (the Brier score). However, like accuracy, this is an O(n) operation that doesn’t fully take ranking performance into account. I will show how a novel decomposition of the Brier score into calibration loss and refinement loss [3] sheds light on both ranking and probability estimation performance. While previous decompositions are approximate [4], our decomposition is an exact one based on the ROC convex hull. (The connection between the ROC convex hull and calibration was independently noted by [5]). In the case of decision trees, the analysis explains the empirical evidence that probability estimation trees produce well-calibrated probabilities [6].KeywordsProbability EstimationRanking PerformanceBrier ScoreTrue Negative RateLearn Decision TreeThese keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.

Rationale and ObjectivesThe human epidermal growth factor receptor 2 (HER2) gene status is crucial for determining treatment efficacy. This study assessed preoperative HER2 classification in breast cancer using machine learning based on clinicopathological and MRI characteristics.Materials and MethodsThis retrospective study involved 1,015 patients (1,030 lesions) across two centers. Patients were divided into training, internal validation, and external validation sets. Nomograms were developed using clinicopathological and MRI features. Predictive models were constructed using decision trees (DT), support vector machines (SVM), k-nearest neighbors (k-NN), artificial neural networks (ANN), and multivariable logistic regression (LR). Model performance was evaluated using receiver operating characteristic curves, decision curve analysis, and calibration curves. Model interpretability was achieved by developing nomograms and employing SHAP (SHapley Additive exPlanations) analysis.ResultsKey variables for distinguishing HER2-positive from HER2-negative cases included regional N category, estrogen receptor, PR (progesterone receptor) status, Ki-67 status, lesion number, distribution quadrant, and accompanying signs. The SVM model achieved the highest AUC of 0.86 (95% confidence interval (CI): 0.81–0.90) in the training set, while the ANN model had an AUC of 0.77 (95% CI: 0.67–0.86) in the internal validation set. In the external validation set, the LR model achieved the highest AUC of 0.66 (95% CI: 0.56–0.76), although the overall performance was modest. For HER2-low versus HER2-zero differentiation, Ki-67 status, lesion number, distribution quadrant, mass shape, early enhancement rate, and ADC (apparent diffusion coefficient) were significant. The SVM model attained the highest AUC of 0.87 (95% CI: 0.83–0.91) in the training set, while the LR model demonstrated superior generalizability, yielding the highest AUCs in both the internal and external validation sets (internal: 0.67, 95% CI: 0.58–0.76; external: 0.74, 95% CI: 0.65–0.83). Radiologists benefited from the nomogram for improved diagnostic accuracy, especially junior radiologists. SHAP analysis revealed that PR status was paramount for HER2-positive classification, whereas mass shape and ADC values were dominant for identifying HER2-low status.ConclusionIntegrating machine learning with clinicopathological and MRI characteristics improves the accuracy of HER2 status classification in breast cancer and enhances diagnostic capabilities for radiologists in clinical practice.

ObjectiveTo explore the predictive value of peak systolic velocity (PSV) ratio, PSV1 and PSV2 of ophthalmic artery Doppler in pregnant women for small for gestational age (SGA) infants and to construct a nomogram prediction model.MethodsA total of 201 pregnant women who visited our hospital from March 2022 to June 2024 were selected as the research subjects, and their clinical data and ophthalmic artery Doppler parameters were collected. The data were randomly divided into a training set (n = 295) and a verification set (n = 126) in a 7:3 ratio. The independent risk factors for SGA were screened by univariate and multivariate Logistic regression analysis, and the nomogram model was constructed. The model calibration degree, prediction efficiency and clinical value were evaluated.ResultsThe incidence of SGA in the training and validation sets was 25.42 and 25.31%, respectively. Multivariate Logistic regression analysis showed that PSV ratio of ophthalmic artery, PSV1, PSV2, Pulse index 1 (PI1), PI2, and gestational week were the independent risk factors for SGA (all p < 0.05). The C-index of the nomogram model in the training set and the verification set was 0.858 and 0.835, respectively, the area under the ROC curve (AUC) was 0.858 (95% CI: 0.804–0.912) and 0.835 (95% CI: 0.734–0.936), and the sensitivity and specificity were 0.887, 0.747 and 0.636 and 0.833, respectively. The calibration curve showed good agreement between the predicted and actual values, respectively, which was good by the Hosmer-Lemeshow test. The analysis of decision curve showed that the model had high clinical application value in the range of threshold probability 0.10–0.80.ConclusionThe nomogram model based on PSV ratio of ophthalmic artery Doppler, PSV1, and PSV2 has good prediction performance for SGA, and provides a new tool for early clinical recognition of high-risk pregnant women with SGA. However, the accuracy and applicability of the model still need to be further verified in the multi-center large sample study.

ObjectiveTo develop and validate a computed tomography (CT)-based radiomics model for predicting tumor deposits (TDs) preoperatively in patients with rectal cancer (RC).MethodsThis retrospective study enrolled 254 patients with pathologically confirmed RC between December 2017 and December 2019. Patients were divided into a training set (n = 203) and a validation set (n = 51). A large number of radiomics features were extracted from the portal venous phase images of CT. After selecting features with L1-based method, we established Rad-score by using the logistic regression analysis. Furthermore, a combined model incorporating Rad-score and clinical factors was developed and visualized as the nomogram. The models were evaluated by the receiver operating characteristic curve (ROC) analysis and area under the ROC curve (AUC).ResultsOne hundred and seventeen of 254 patients were eventually found to be TDs+. Rad-score and clinical factors including carbohydrate antigen (CA) 19-9, CT-reported T stage (cT), and CT-reported peritumoral nodules (+/-) were significantly different between the TDs+ and TDs- groups (all P < 0.001). These factors were all included in the combined model by the logistic regression analysis (odds ratio = 2.378 for Rad-score, 2.253 for CA19-9, 2.281 for cT, and 4.485 for peritumoral nodules). This model showed good performance to predict TDs in the training and validation cohorts (AUC = 0.830 and 0.832, respectively). Furthermore, the combined model outperformed the clinical model incorporating CA19-9, cT, and peritumoral nodules (+/-) in both training and validation cohorts for predicting TDs preoperatively (AUC = 0.773 and 0.718, P = 0.008 and 0.039).ConclusionsThe combined model incorporating Rad-score and clinical factors could provide a preoperative prediction of TDs and help clinicians guide individualized treatment for RC patients.