Evaluating polygenic risk scores for predicting cardiometabolic traits and disease risks in the Taiwan Biobank.

Type 2 diabetes (T2D) and hypertension are common comorbidities and, along with hyperlipidemia, serve as risk factors for cardiovascular diseases. This study aimed to evaluate the predictive value of polygenic risk scores (PRSs) on cardiometabolic traits related to T2D, hypertension, and hyperlipidemia and the incidence of these three diseases in Taiwan Biobank samples. Using publicly available, large-scale genome-wide association studies summary statistics, we constructed cross-ethnic PRSs for T2D, hypertension, body mass index, and nine quantitative traits typically used to define the three diseases. A composite PRS (cPRS) for each of the nine traits was constructed by aggregating the significant PRSs of its genetically correlated traits. The associations of each of the nine traits at baseline as well as the change of trait values during a 3- to 6-year follow-up period with its cPRS were evaluated. The predictive performances of cPRSs in predicting future incidences of T2D, hypertension, and hyperlipidemia were assessed. The cPRSs had significant associations with baseline and changes of trait values in 3-6 years and explained a higher proportion of variance for all traits than individual PRSs. Furthermore, models incorporating disease-related cPRSs, along with clinical features and relevant trait measurements achieved area under the curve values of 87.8%, 83.7%, and 75.9% for predicting future T2D, hypertension, and hyperlipidemia in 3-6 years, respectively.

Prevalence and predictive modeling of undiagnosed diabetes and impaired fasting glucose in Taiwan: a Taiwan Biobank study.

INTRODUCTION: We investigated the prevalence of undiagnosed diabetes and impaired fasting glucose (IFG) in individuals without known diabetes in Taiwan and developed a risk prediction model for identifying undiagnosed diabetes and IFG. RESEARCH DESIGN AND METHODS: Using data from a large population-based Taiwan Biobank study linked with the National Health Insurance Research Database, we estimated the standardized prevalence of undiagnosed diabetes and IFG between 2012 and 2020. We used the forward continuation ratio model with the Lasso penalty, modeling undiagnosed diabetes, IFG, and healthy reference group (individuals without diabetes or IFG) as three ordinal outcomes, to identify the risk factors and construct the prediction model. Two models were created: Model 1 predicts undiagnosed diabetes, IFG_110 (ie, fasting glucose between 110 mg/dL and 125 mg/dL), and the healthy reference group, while Model 2 predicts undiagnosed diabetes, IFG_100 (ie, fasting glucose between 100 mg/dL and 125 mg/dL), and the healthy reference group. RESULTS: The standardized prevalence of undiagnosed diabetes for 2012-2014, 2015-2016, 2017-2018, and 2019-2020 was 1.11%, 0.99%, 1.16%, and 0.99%, respectively. For these periods, the standardized prevalence of IFG_110 and IFG_100 was 4.49%, 3.73%, 4.30%, and 4.66% and 21.0%, 18.26%, 20.16%, and 21.08%, respectively. Significant risk prediction factors were age, body mass index, waist to hip ratio, education level, personal monthly income, betel nut chewing, self-reported hypertension, and family history of diabetes. The area under the curve (AUC) for predicting undiagnosed diabetes in Models 1 and 2 was 80.39% and 77.87%, respectively. The AUC for predicting undiagnosed diabetes or IFG in Models 1 and 2 was 78.25% and 74.39%, respectively. CONCLUSIONS: Our results showed the changes in the prevalence of undiagnosed diabetes and IFG. The identified risk factors and the prediction models could be helpful in identifying individuals with undiagnosed diabetes or individuals with a high risk of developing diabetes in Taiwan.

Predicting Postoperative Mortality With Deep Neural Networks and Natural Language Processing: Model Development and Validation.

BACKGROUND: Machine learning (ML) achieves better predictions of postoperative mortality than previous prediction tools. Free-text descriptions of the preoperative diagnosis and the planned procedure are available preoperatively. Because reading these descriptions helps anesthesiologists evaluate the risk of the surgery, we hypothesized that deep learning (DL) models with unstructured text could improve postoperative mortality prediction. However, it is challenging to extract meaningful concept embeddings from this unstructured clinical text. OBJECTIVE: This study aims to develop a fusion DL model containing structured and unstructured features to predict the in-hospital 30-day postoperative mortality before surgery. ML models for predicting postoperative mortality using preoperative data with or without free clinical text were assessed. METHODS: We retrospectively collected preoperative anesthesia assessments, surgical information, and discharge summaries of patients undergoing general and neuraxial anesthesia from electronic health records (EHRs) from 2016 to 2020. We first compared the deep neural network (DNN) with other models using the same input features to demonstrate effectiveness. Then, we combined the DNN model with bidirectional encoder representations from transformers (BERT) to extract information from clinical texts. The effects of adding text information on the model performance were compared using the area under the receiver operating characteristic curve (AUROC) and the area under the precision-recall curve (AUPRC). Statistical significance was evaluated using P<.05. RESULTS: The final cohort contained 121,313 patients who underwent surgeries. A total of 1562 (1.29%) patients died within 30 days of surgery. Our BERT-DNN model achieved the highest AUROC (0.964, 95% CI 0.961-0.967) and AUPRC (0.336, 95% CI 0.276-0.402). The AUROC of the BERT-DNN was significantly higher compared to logistic regression (AUROC=0.952, 95% CI 0.949-0.955) and the American Society of Anesthesiologist Physical Status (ASAPS AUROC=0.892, 95% CI 0.887-0.896) but not significantly higher compared to the DNN (AUROC=0.959, 95% CI 0.956-0.962) and the random forest (AUROC=0.961, 95% CI 0.958-0.964). The AUPRC of the BERT-DNN was significantly higher compared to the DNN (AUPRC=0.319, 95% CI 0.260-0.384), the random forest (AUPRC=0.296, 95% CI 0.239-0.360), logistic regression (AUPRC=0.276, 95% CI 0.220-0.339), and the ASAPS (AUPRC=0.149, 95% CI 0.107-0.203). CONCLUSIONS: Our BERT-DNN model has an AUPRC significantly higher compared to previously proposed models using no text and an AUROC significantly higher compared to logistic regression and the ASAPS. This technique helps identify patients with higher risk from the surgical description text in EHRs.

DeepDRG: Performance of Artificial Intelligence Model for Real-Time Prediction of Diagnosis-Related Groups.

Nowadays, the use of diagnosis-related groups (DRGs) has been increased to claim reimbursement for inpatient care. The overall benefits of using DRGs depend upon the accuracy of clinical coding to obtain reasonable reimbursement. However, the selection of appropriate codes is always challenging and requires professional expertise. The rate of incorrect DRGs is always high due to the heavy workload, poor quality of documentation, and lack of computer assistance. We therefore developed deep learning (DL) models to predict the primary diagnosis for appropriate reimbursement and improving hospital performance. A dataset consisting of 81,486 patients with 128,105 episodes was used for model training and testing. Patients' age, sex, drugs, diseases, laboratory tests, procedures, and operation history were used as inputs to our multiclass prediction model. Gated recurrent unit (GRU) and artificial neural network (ANN) models were developed to predict 200 primary diagnoses. The performance of the DL models was measured by the area under the receiver operating curve, precision, recall, and F1 score. Of the two DL models, the GRU method, had the best performance in predicting the primary diagnosis (AUC: 0.99, precision: 83.2%, and recall: 66.0%). However, the performance of ANN model for DRGs prediction achieved AUC of 0.99 with a precision of 0.82 and recall of 0.57. The findings of our study show that DL algorithms, especially GRU, can be used to develop DRGs prediction models for identifying primary diagnosis accurately. DeepDRGs would help to claim appropriate financial incentives, enable proper utilization of medical resources, and improve hospital performance.

Acute Exacerbation of a Chronic Obstructive Pulmonary Disease Prediction System Using Wearable Device Data, Machine Learning, and Deep Learning: Development and Cohort Study.

BACKGROUND: The World Health Organization has projected that by 2030, chronic obstructive pulmonary disease (COPD) will be the third-leading cause of mortality and the seventh-leading cause of morbidity worldwide. Acute exacerbations of chronic obstructive pulmonary disease (AECOPD) are associated with an accelerated decline in lung function, diminished quality of life, and higher mortality. Accurate early detection of acute exacerbations will enable early management and reduce mortality. OBJECTIVE: The aim of this study was to develop a prediction system using lifestyle data, environmental factors, and patient symptoms for the early detection of AECOPD in the upcoming 7 days. METHODS: This prospective study was performed at National Taiwan University Hospital. Patients with COPD that did not have a pacemaker and were not pregnant were invited for enrollment. Data on lifestyle, temperature, humidity, and fine particulate matter were collected using wearable devices (Fitbit Versa), a home air quality-sensing device (EDIMAX Airbox), and a smartphone app. AECOPD episodes were evaluated via standardized questionnaires. With these input features, we evaluated the prediction performance of machine learning models, including random forest, decision trees, k-nearest neighbor, linear discriminant analysis, and adaptive boosting, and a deep neural network model. RESULTS: The continuous real-time monitoring of lifestyle and indoor environment factors was implemented by integrating home air quality-sensing devices, a smartphone app, and wearable devices. All data from 67 COPD patients were collected prospectively during a mean 4-month follow-up period, resulting in the detection of 25 AECOPD episodes. For 7-day AECOPD prediction, the proposed AECOPD predictive model achieved an accuracy of 92.1%, sensitivity of 94%, and specificity of 90.4%. Receiver operating characteristic curve analysis showed that the area under the curve of the model in predicting AECOPD was greater than 0.9. The most important variables in the model were daily steps walked, stairs climbed, and daily distance moved. CONCLUSIONS: Using wearable devices, home air quality-sensing devices, a smartphone app, and supervised prediction algorithms, we achieved excellent power to predict whether a patient would experience AECOPD within the upcoming 7 days. The AECOPD prediction system provided an effective way to collect lifestyle and environmental data, and yielded reliable predictions of future AECOPD events. Compared with previous studies, we have comprehensively improved the performance of the AECOPD prediction model by adding objective lifestyle and environmental data. This model could yield more accurate prediction results for COPD patients than using only questionnaire data.