Establish and validate the reliability of predictive models in bone mineral density by deep learning as examination tool for women.

While FRAX with BMD could be more precise in estimating the fracture risk, DL-based models were validated to slightly reduce the number of under- and over-treated patients when no BMD measurements were available. The validated models could be used to screen for patients at a high risk of fracture and osteoporosis. PURPOSE: Fracture risk assessment tool (FRAX) is useful in classifying the fracture risk level, and precise prediction can be achieved by estimating both clinical risk factors and bone mineral density (BMD) using dual X-ray absorptiometry (DXA). However, DXA is not frequently feasible because of its cost and accessibility. This study aimed to establish the reliability of deep learning (DL)-based alternative tools for screening patients at a high risk of fracture and osteoporosis. METHODS: Participants were enrolled from the National Bone Health Screening Project of Taiwan in this cross-sectional study. First, DL-based models were built to predict the lowest T-score value in either the lumbar spine, total hip, or femoral neck and their respective BMD values. The Bland-Altman analysis was used to compare the agreement between the models and DXA. Second, the predictive model to classify patients with a high fracture risk was built according to the estimated BMD from the first step and the FRAX score without BMD. The performance of the model was compared with the classification based on FRAX with BMD. RESULTS: Approximately 10,827 women (mean age, 65.4 ± 9.4 years) were enrolled. In the prediction of the lumbar spine BMD, total hip BMD, femoral neck BMD, and lowest T-score, the root-mean-square error (RMSE) was 0.099, 0.089, 0.076, and 0.68, respectively. The Bland-Altman analysis revealed a nonsignificant difference between the predictive models and DXA. The FRAX score with femoral neck BMD for major osteoporotic fracture risk was 9.7% ± 6.7%, whereas the risk for hip fracture was 3.3% ± 4.6%. Comparison between the classification of FRAX with and without BMD revealed the accuracy rate, positive predictive value (PPV), and negative predictive value (NPV) of 78.8%, 64.6%, and 89.9%, respectively. The area under the receiver operating characteristic curve (AUROC), accuracy rate, PPV, and NPV of the classification model were 0.913 (95% confidence interval: 0.904-0.922), 83.5%, 71.2%, and 92.2%, respectively. CONCLUSION: While FRAX with BMD could be more precise in estimating the fracture risk, DL-based models were validated to slightly reduce the number of under- and over-treated patients when no BMD measurements were available. The validated models could be used to screen for patients at a high risk of fracture and osteoporosis.

An ensemble model for predicting dispositions of emergency department patients.

OBJECTIVE: The healthcare challenge driven by an aging population and rising demand is one of the most pressing issues leading to emergency department (ED) overcrowding. An emerging solution lies in machine learning's potential to predict ED dispositions, thus leading to promising substantial benefits. This study's objective is to create a predictive model for ED patient dispositions by employing ensemble learning. It harnesses diverse data types, including structured and unstructured information gathered during ED visits to address the evolving needs of localized healthcare systems. METHODS: In this cross-sectional study, 80,073 ED patient records were amassed from a major southern Taiwan hospital in 2018-2019. An ensemble model incorporated structured (demographics, vital signs) and pre-processed unstructured data (chief complaints, preliminary diagnoses) using bag-of-words (BOW) and term frequency-inverse document frequency (TF-IDF). Two random forest base-learners for structured and unstructured data were employed and then complemented by a multi-layer perceptron meta-learner. RESULTS: The ensemble model demonstrates strong predictive performance for ED dispositions, achieving an area under the receiver operating characteristic curve of 0.94. The models based on unstructured data encoded with BOW and TF-IDF yield similar performance results. Among the structured features, the top five most crucial factors are age, pulse rate, systolic blood pressure, temperature, and acuity level. In contrast, the top five most important unstructured features are pneumonia, fracture, failure, suspect, and sepsis. CONCLUSIONS: Findings indicate that utilizing ensemble learning with a blend of structured and unstructured data proves to be a predictive method for determining ED dispositions.

Early short-term prediction of emergency department length of stay using natural language processing for low-acuity outpatients.

BACKGROUND: Low-acuity outpatients constitute the majority of emergency department (ED) patients, and these patients often experience an unpredictable length of stay (LOS). Effective LOS prediction might improve the quality of ED care and reduce ED crowding. OBJECTIVE: The objective of this study was to explore the potential of natural language processing (NLP) of the first ED physicians' clinical notes and to evaluate NLP-based short-term prediction models based on mixed-type clinical data. METHODS: A retrospective study was conducted at an ED of a tertiary teaching hospital in Taiwan from January 2017 to June 2017. In total, 12,962 low-acuity outpatients were enrolled. Using structured data (e.g., demographic variables and vital signs) and different sections of the first SOAP notes as predictors, we developed six NLP-based prediction models (i.e., term frequency-inverse document frequency (TF-IDF) and truncated singular value decomposition (SVD)) to predict LOS. The metric for model evaluation is the mean squared error (MSE). RESULTS: Of the six NLP-based models, the model using structured data and all the sections of the first SOAP notes processed by the TF-IDF and truncated SVD method performed the best, with an MSE of 3.00 [95% CI: 2.94-3.06]. In addition, ten important topics extracted by the TF-IDF and truncated SVD method had significant effects on the LOS (p < 0.001). CONCLUSION: NLP-based models can be used as an early short-term prediction of LOS and have the potential for mixed-type clinical data analysis. The proposed models would likely aid ED physicians' decision-making processes and improve ED quality of care.

Advanced Sampling Technique in Radiology Free-Text Data for Efficiently Building Text Mining Models by Deep Learning in Vertebral Fracture.

This study aims to establish advanced sampling methods in free-text data for efficiently building semantic text mining models using deep learning, such as identifying vertebral compression fracture (VCF) in radiology reports. We enrolled a total of 27,401 radiology free-text reports of X-ray examinations of the spine. The predictive effects were compared between text mining models built using supervised long short-term memory networks, independently derived by four sampling methods: vector sum minimization, vector sum maximization, stratified, and simple random sampling, using four fixed percentages. The drawn samples were applied to the training set, and the remaining samples were used to validate each group using different sampling methods and ratios. The predictive accuracy was measured using the area under the receiver operating characteristics (AUROC) to identify VCF. At the sampling ratios of 1/10, 1/20, 1/30, and 1/40, the highest AUROC was revealed in the sampling methods of vector sum minimization as confidence intervals of 0.981 (95%CIs: 0.980-0.983)/0.963 (95%CIs: 0.961-0.965)/0.907 (95%CIs: 0.904-0.911)/0.895 (95%CIs: 0.891-0.899), respectively. The lowest AUROC was demonstrated in the vector sum maximization. This study proposes an advanced sampling method, vector sum minimization, in free-text data that can be efficiently applied to build the text mining models by smartly drawing a small amount of critical representative samples.

Measurement of Body Surface Area for Psoriasis Using U-net Models.

During the evaluation of body surface area (BSA), precise measurement of psoriasis is crucial for assessing disease severity and modulating treatment strategies. Physicians usually evaluate patients subjectively through direct visual evaluation. However, judgment based on the naked eye is not reliable. This study is aimed at evaluating the use of machine learning methods, specifically U-net models, and developing an artificial neural network prediction model for automated psoriasis lesion segmentation and BSA measurement. The segmentation of psoriasis lesions using deep learning is adopted to measure the BSA of psoriasis so that the severity can be evaluated automatically in patients. An automated psoriasis lesion segmentation method based on the U-net architecture was used with a focus on high-resolution images and estimation of the BSA. The proposed method trained the model with the same patch size of 512 × 512 and predicted testing images with different patch sizes. We collected 255 high-resolution psoriasis images representing large anatomical sites, such as the trunk and extremities. The average residual of the ground truth image and the predicted image was approximately 0.033. The interclass correlation coefficient between the U-net and dermatologist's segmentations measured in the ratio of affected psoriasis over the body area in the test dataset was 0.966 (95% CI: 0.981-0.937), indicating strong agreement. Herein, the proposed U-net model achieved dermatologist-level performance in estimating the involved BSA for psoriasis.

Emergency department disposition prediction using a deep neural network with integrated clinical narratives and structured data.

BACKGROUND: Emergency department (ED) overcrowding has been a serious issue and demands effective clinical decision-making of patient disposition. In previous studies, emergency clinical narratives provide a rich context for clinical decisions. We aimed to develop the disposition prediction model using deep learning modeling strategy with the heterogeneous data, including the physicians' narratives. METHODS: We constructed a retrospective cohort of all 104,083 ED visits of non-trauma adults during 2017-18 from an academically affiliated ED in Taiwan. 18,308 visits were excluded based on the completeness of each record and the unpredictable dispositions, such as out-of-hospital cardiac arrest, against-advice discharge, and escapes. We integrated subjective section of the first physicians' clinical narratives and structured data (e.g., demographics, triage vital signs, etc.) as available predictors at the first physician-patient encounter. To predict final patient disposition (i.e., hospitalization or discharge), a deep neural network (DNN) model was developed with word embedding, a common natural language processing method. We compared the proposed model to a reference model using the Rapid Emergency Medicine Score, a logistic regression model with structured data, and a DNN model with paragraph vectors. F1 score was used to measure the predictive performance for each model. RESULTS: The F1 score (with 95 % CI) for the proposed model, the reference model, the logistic regression model with structured data, and the DNN model with paragraph vectors were 0.674 (0.669-0.679), 0.474 (0.469-0.479), 0.547 (0.543-0.551), and 0.602 (0.596-0.607), respectively. While analyzing the relationship between context length and predictive performance under the proposed model, the F1 score at 95th percentile of the word counts was higher than that at 25th percentile of the word counts in chief complaint [0.634 (0.629-0.640) vs. 0.624 (0.620-0.628)] and in present illness [0.671 (0.667-0.674) vs. 0.654 (0.651-0.658)], but not in past medical history [0.674 (0.669-0.679) vs. 0.673 (0.666-0.679)]. CONCLUSIONS: The proposed deep learning model with the usage of the first physicians' clinical narratives and structured data based on natural language processing outperformed the commonly used ones in terms of F1 score. It also evidenced the importance of the subjective section of clinical narratives, which serve as vital predictors for ED clinical decision-making.

Predicting nurses' intention to quit with a support vector machine: a new approach to set up an early warning mechanism in human resource management.

This project developed a Support Vector Machine for predicting nurses' intention to quit, using working motivation, job satisfaction, and stress levels as predictors. This study was conducted in three hospitals located in southern Taiwan. The target population was all nurses (389 valid cases). For cross-validation, we randomly split cases into four groups of approximately equal sizes, and performed four training runs. After the training, the average percentage of misclassification on the training data was 0.86, while that on the testing data was 10.8, resulting in predictions with 89.2% accuracy. This Support Vector Machine can predict nurses' intention to quit, without asking these nurses whether they have an intention to quit.