An emerging outbreak of a novel coronavirus, COVID-19, has now been detected in at least 211 countries worldwide. Given this pandemic situation, robust risk communication is urgently needed, particularly in affected countries. Therefore, this study explored the potential use of Google Trends (GT) to monitor public restlessness toward COVID-19 infection in Taiwan. We retrieved GT data for the specific locations and subregions in Taiwan nationwide using defined search terms related to the coronavirus, handwashing, and face masks. Searches related to COVID-19 and face masks in Taiwan rapidly increased following the announcements of Taiwan's first imported case and reached a peak as locally acquired cases were reported. However, searches for handwashing gradually increased during the period of face-mask shortage. Moreover, high to moderate correlations between Google relative search volumes (RSVs) and COVID-19 cases were found in Taipei (lag-3), New Taipei (lag-2), Taoyuan (lag-2), Tainan (lag-1), Taichung (lag0), and Kaohsiung (lag0). In response to the ongoing outbreak, our results demonstrated that GT could potentially define the proper timing and location for practicing appropriate risk communication strategies for affected populations.

Background: Predictions in pregnancy care are complex because of interactions among multiple factors. Hence, pregnancy outcomes are not easily predicted by a single predictor using only one algorithm or modeling method.Objective: This study aims to review and compare the predictive performances between logistic regression (LR) and other machine learning algorithms for developing or validating a multivariable prognostic prediction model for pregnancy care to inform clinicians’ decision making.Methods: Research articles from MEDLINE, Scopus, Web of Science, and Google Scholar were reviewed following several guidelines for a prognostic prediction study, including a risk of bias (ROB) assessment. We report the results based on the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines. Studies were primarily framed as PICOTS (population, index, comparator, outcomes, timing, and setting): Population: men or women in procreative management, pregnant women, and fetuses or newborns; Index: multivariable prognostic prediction models using non-LR algorithms for risk classification to inform clinicians’ decision making; Comparator: the models applying an LR; Outcomes: pregnancy-related outcomes of procreation or pregnancy outcomes for pregnant women and fetuses or newborns; Timing: pre-, inter-, and peripregnancy periods (predictors), at the pregnancy, delivery, and either puerperal or neonatal period (outcome), and either short- or long-term prognoses (time interval); and Setting: primary care or hospital. The results were synthesized by reporting study characteristics and ROBs and by random effects modeling of the difference of the logit area under the receiver operating characteristic curve of each non-LR model compared with the LR model for the same pregnancy outcomes. We also reported between-study heterogeneity by using τ2 and I2.Results: Of the 2093 records, we included 142 studies for the systematic review and 62 studies for a meta-analysis. Most prediction models used LR (92/142, 64.8%) and artificial neural networks (20/142, 14.1%) among non-LR algorithms. Only 16.9% (24/142) of studies had a low ROB. A total of 2 non-LR algorithms from low ROB studies significantly outperformed LR. The first algorithm was a random forest for preterm delivery (logit AUROC 2.51, 95% CI 1.49-3.53; I2=86%; τ2=0.77) and pre-eclampsia (logit AUROC 1.2, 95% CI 0.72-1.67; I2=75%; τ2=0.09). The second algorithm was gradient boosting for cesarean section (logit AUROC 2.26, 95% CI 1.39-3.13; I2=75%; τ2=0.43) and gestational diabetes (logit AUROC 1.03, 95% CI 0.69-1.37; I2=83%; τ2=0.07).Conclusions: Prediction models with the best performances across studies were not necessarily those that used LR but also used random forest and gradient boosting that also performed well. We recommend a reanalysis of existing LR models for several pregnancy outcomes by comparing them with those algorithms that apply standard guidelines.Trial Registration: PROSPERO (International Prospective Register of Systematic Reviews) CRD42019136106; https://www.crd.york.ac.uk/prospero/display_record.php?RecordID=136106

The risk factors of diabetic retinopathy (DR) were investigated extensively in the past studies, but it remains unknown which risk factors were more associated with the DR than others. If we can detect the DR related risk factors more accurately, we can then exercise early prevention strategies for diabetic retinopathy in the most high-risk population. The purpose of this study is to build a prediction model for the DR in type 2 diabetes mellitus using data mining techniques including the support vector machines, decision trees, artificial neural networks, and logistic regressions. Experimental results demonstrated that prediction performance by support vector machines performed better than the other machine learning algorithms and achieved 79.5% and 0.839 in accuracy and area under the receiver operating characteristic curve using percentage split (i.e., data set divided into 80% as trainning and 20% as test), respectively. Evaluated by three-way data split scheme (i.e., data set divided into 60% as training, 20% as validation, and 20% as independent test), our method obtained slightly lower performance compared to percentage split, which suggested that three-way data split is a better way to evaluate the real performance and prevent overestimation. Moreover, we incorporated approaches proposed in previous studies to evaluate our data set and our prediction performance outperformed the other previous studies in most evaluation measures. This lends support to our assumption that appropriate machine learning algorithms combined with discriminative clinical features can effectively detect diabetic retinopathy. Our method identifies use of insulin and duration of diabetes as novel interpretable features to assist with clinical decisions in identifying the high-risk populations for diabetic retinopathy. If duration of DM increases by 1 year, the odds ratio to have DMR is increased by 9.3%. The odds ratio to have DR is increased by 3.561 times for patients who use insulin compared to patients who do not use insulin. Our results can be used to facilitate development of clinical decision support systems for clinical practice in the future.

We developed and validated an artificial intelligence (AI)-assisted prediction of preeclampsia applied to a nationwide health insurance dataset in Indonesia. The BPJS Kesehatan dataset have been preprocessed using a nested case-control design into preeclampsia/eclampsia (n = 3318) and normotensive pregnant women (n = 19,883) from all women with one pregnancy. The dataset provided 95 features consisting of demographic variables and medical histories started from 24 months to event and ended by delivery as the event. Six algorithms were compared by area under the receiver operating characteristics curve (AUROC) with a subgroup analysis by time to the event. We compared our model to similar prediction models from systematically reviewed studies. In addition, we conducted a text mining analysis based on natural language processing techniques to interpret our modeling results. The best model consisted of 17 predictors extracted by a random forest algorithm. Nine∼12 months to the event was the period that had the best AUROC in external validation by either geographical (0.88, 95% confidence interval (CI) 0.88-0.89) or temporal split (0.86, 95% CI 0.85-0.86). We compared this model to prediction models in seven studies from 869 records in PUBMED, EMBASE, and SCOPUS. This model outperformed the previous models in terms of the precision, sensitivity, and specificity in all validation sets. Our low-cost model improved preliminary prediction to decide pregnant women that will be predicted by the models with high specificity and advanced predictors. This work was supported by grant no. MOST108-2221-E-038-018 from the Ministry of Science and Technology of Taiwan.

Background: The prevalence of nonalcoholic fatty liver disease is increasing over time worldwide, with similar trends to those of diabetes and obesity. A liver biopsy, the gold standard of diagnosis, is not favored due to its invasiveness. Meanwhile, noninvasive evaluation methods of fatty liver are still either very expensive or demonstrate poor diagnostic performances, thus, limiting their applications. We developed neural network–based models to assess fatty liver and classify the severity using B-mode ultrasound (US) images. Methods: We followed standards for reporting of diagnostic accuracy guidelines to report this study. In this retrospective study, we utilized B-mode US images from a consecutive series of patients to develop four-class, two-class, and three-class diagnostic prediction models. The images were eligible if confirmed by at least two gastroenterologists. We compared pretrained convolutional neural network models, consisting of visual geometry group (VGG)19, ResNet-50 v2, MobileNet v2, Xception, and Inception v2. For validation, we utilized 20% of the dataset resulting in >100 images for each severity category. Results: There were 21,855 images from 2,070 patients classified as normal (N = 11,307), mild (N = 4,467), moderate (N = 3,155), or severe steatosis (N = 2,926). We used ResNet-50 v2 for the final model as the best ones. The areas under the receiver operating characteristic curves were 0.974 (mild steatosis vs others), 0.971 (moderate steatosis vs others), 0.981 (severe steatosis vs others), 0.985 (any severity vs normal), and 0.996 (moderate-to-severe steatosis/clinically abnormal vs normal-to-mild steatosis/clinically normal). Conclusion: Our deep learning models achieved comparable predictive performances to the most accurate, yet expensive, noninvasive diagnostic methods for fatty liver. Because of the discriminative ability, including for mild steatosis, significant impacts on clinical applications for fatty liver are expected. However, we need to overcome machine-dependent variation, motion artifacts, lacking of second confirmation from any other tools, and hospital-dependent regional bias.

Background: Preeclampsia and intrauterine growth restriction are placental dysfunction–related disorders (PDDs) that require a referral decision be made within a certain time period. An appropriate prediction model should be developed for these diseases. However, previous models did not demonstrate robust performances and/or they were developed from datasets with highly imbalanced classes.Objective: In this study, we developed a predictive model of PDDs by machine learning that uses features at 24-37 weeks’ gestation, including maternal characteristics, uterine artery (UtA) Doppler measures, soluble fms-like tyrosine kinase receptor-1 (sFlt-1), and placental growth factor (PlGF).Methods: A public dataset was taken from a prospective cohort study that included pregnant women with PDDs (66/95, 69%) and a control group (29/95, 31%). Preliminary selection of features was based on a statistical analysis using SAS 9.4 (SAS Institute). We used Weka (Waikato Environment for Knowledge Analysis) 3.8.3 (The University of Waikato, Hamilton, NZ) to automatically select the best model using its optimization algorithm. We also manually selected the best of 23 white-box models. Models, including those from recent studies, were also compared by interval estimation of evaluation metrics. We used the Matthew correlation coefficient (MCC) as the main metric. It is not overoptimistic to evaluate the performance of a prediction model developed from a dataset with a class imbalance. Repeated 10-fold cross-validation was applied.Results: The classification via regression model was chosen as the best model. Our model had a robust MCC (.93, 95% CI .87-1.00, vs .64, 95% CI .57-.71) and specificity (100%, 95% CI 100-100, vs 90%, 95% CI 90-90) compared to each metric of the best models from recent studies. The sensitivity of this model was not inferior (95%, 95% CI 91-100, vs 100%, 95% CI 92-100). The area under the receiver operating characteristic curve was also competitive (0.970, 95% CI 0.966-0.974, vs 0.987, 95% CI 0.980-0.994). Features in the best model were maternal weight, BMI, pulsatility index of the UtA, sFlt-1, and PlGF. The most important feature was the sFlt-1/PlGF ratio. This model used an M5P algorithm consisting of a decision tree and four linear models with different thresholds. Our study was also better than the best ones among recent studies in terms of the class balance and the size of the case class (66/95, 69%, vs 27/239, 11.3%).Conclusions: Our model had a robust predictive performance. It was also developed to deal with the problem of a class imbalance. In the context of clinical management, this model may improve maternal mortality and neonatal morbidity and reduce health care costs.