Search

Federated learning (FL) is a method used for training artificial intelligence models with data from multiple sources while maintaining data anonymity, thus removing many barriers to data sharing. Here we used data from 20 institutes across the globe to train a FL model, called EXAM (electronic medical record (EMR) chest X-ray AI model), that predicts the future oxygen requirements of symptomatic patients with COVID-19 using inputs of vital signs, laboratory data and chest X-rays. EXAM achieved an average area under the curve (AUC) >0.92 for predicting outcomes at 24 and 72 h from the time of initial presentation to the emergency room, and it provided 16% improvement in average AUC measured across all participating sites and an average increase in generalizability of 38% when compared with models trained at a single site using that site’s data. For prediction of mechanical ventilation treatment or death at 24 h at the largest independent test site, EXAM achieved a sensitivity of 0.950 and specificity of 0.882. In this study, FL facilitated rapid data science collaboration without data exchange and generated a model that generalized across heterogeneous, unharmonized datasets for prediction of clinical outcomes in patients with COVID-19, setting the stage for the broader use of FL in healthcare. Federated learning, a method for training artificial intelligence algorithms that protects data privacy, was used to predict future oxygen requirements of symptomatic patients with COVID-19 using data from 20 different institutes across the globe.

Federated learning (FL) is a method used for training artificial intelligence models with data from multiple sources while maintaining data anonymity, thus removing many barriers to data sharing. Here we used data from 20 institutes across the globe to train a FL model, called EXAM (electronic medical record (EMR) chest X-ray AI model), that predicts the future oxygen requirements of symptomatic patients with COVID-19 using inputs of vital signs, laboratory data and chest X-rays. EXAM achieved an average area under the curve (AUC)0.92 for predicting outcomes at 24 and 72 h from the time of initial presentation to the emergency room, and it provided 16% improvement in average AUC measured across all participating sites and an average increase in generalizability of 38% when compared with models trained at a single site using that site's data. For prediction of mechanical ventilation treatment or death at 24 h at the largest independent test site, EXAM achieved a sensitivity of 0.950 and specificity of 0.882. In this study, FL facilitated rapid data science collaboration without data exchange and generated a model that generalized across heterogeneous, unharmonized datasets for prediction of clinical outcomes in patients with COVID-19, setting the stage for the broader use of FL in healthcare.

Background Diabetes mellitus management is one of the most challenging health care issues in Taiwan. To improve guideline adherence and the quality of diabetes care, the diabetes mellitus pay-for-performance (DM-P4P) program was introduced in 2001. This study examined the effects of the DM-P4P program on guideline adherence among patients with diabetes. Methods This cross-sectional study used National Health Insurance (NHI) claim data to assess guideline adherence among three groups of patients with diabetes: patients enrolled in the DM-P4P program, patients not enrolled but treated by DM-P4P-participating physicians, and patients treated by non-P4P physicians. Guideline adherence measures included hemoglobin A1c, blood glucose, lipid, serum creatinine, alanine transaminase, urinalysis, and eye examinations. Multiple logistic regression with generalized estimated equation approach were used to assess the effect of the DM-P4P program while controlling for patient and physician characteristics. Results A total of 520,804 patients were included in the analysis. Patients enrolled in the DM-P4P program were more likely to receive all of the guideline-recommended tests/examinations than patients treated by non-P4P physicians. Patients who were not enrolled in the program but who were treated by DM-P4P-participating physicians were more likely to receive three of the seven recommended tests/examinations than were those treated by non-P4P physicians. Conclusion When physicians participated in the P4P program, this increased the likelihood that patients would receive guideline-recommended tests or examinations. Increasing the DM-P4P participation rates for physicians and patients are important and worthwhile objectives that can assist in the effort to more effectively manage diabetes in the general population.

Two-dimensional (2-D) device simulation is used to investigate the tunneling current of metal ultra-thin-oxide silicon tunneling diodes with different oxide roughness. With the conformal nature of ultrathin oxide, the tunneling current density is simulated in both direct tunneling and Fowler-Nordheim (FN) tunneling regimes with different oxide roughness. The results show that oxide roughness dramatically enhances the tunneling current density and the 2-D electrical effect is responsible for this increment of tunneling current density. Furthermore, a set of devices with controlled oxide roughness is fabricated to verify the simulation results and our model qualitatively agrees with the experiment results.

Background Platinum-based chemotherapy is the standard first-line therapy for patients with advanced lung squamous cell carcinoma (SCC). We compared the effectiveness of first-line chemotherapy regimens. Methods We searched the database of the Taiwan Cancer Registry for patients with newly diagnosed advanced lung SCC from 2004 to 2007. Medication prescription data were retrieved from the database of National Health Insurance, Taiwan. We identified patients who received standard first-line platinum-based chemotherapy, which was defined as chemotherapy with a platinum (P) compound (cisplatin or carboplatin) in addition to 1 of the 4 chemotherapy agents, including gemcitabine (G), docetaxel (D), paclitaxel (T), and vinorelbine (V). Deaths were identified by searching the National Death Registry. Overall survival (OS) was compared between patients who underwent different therapies. Results In total, 2790 patients were identified; 983 patients (35.2%) received standard first-line chemotherapy with P and G (58.1%), D (14.5%), T (11.6%), or V (15.8%). Older patients (age ≥ 70 years) were less likely to receive P + D than P + G, P + T, or P + V ( P = .018). Patients who received P + G, P + D, P + T, or P + V had similar OS (median, 8.9, 7.9, 9.5, and 8.2 months; P = .816). In multivariate analyses adjusting for age, sex, and stage, the first-line chemotherapy regimen was not a predictor for OS. With P + G as the reference group, the adjusted hazard ratios of P + D, P + T, and P + V were 1.03, 0.90, and 1.02, respectively ( P = .710). Conclusions In patients with advanced lung SCC, various regimens did not have a significant effect on survival outcomes.

Background Limited information about tumor status and the time at which antiviral therapy was initiated may have influenced effect estimation in previous research. The aim of this study was to investigate the effect of antiviral therapies on HBV-related HCC progression and deaths in patients receiving curative treatment based on clear clinical-pathological cancer status and the association of start time of adjuvant antiviral therapy initiation and outcomes. Methodology A nationwide inception cohort study of newly diagnosed HCC patients who suffered from viral hepatitis B and received curative HCC therapy as the first course of treatment were identified from the Taiwan Cancer Registry between January 1, 2004, and December 31, 2009. Matched Cox proportional hazards models based on propensity score matching and incorporated time-varying exposure were used to estimate adjusted hazard ratios and 95% confidence intervals (CIs). Findings Among 3,855 HCC patients with HBV, antiviral therapy was administered to 490 (12.7%) following curative treatment. Antiviral-treated patients had a higher percentage of young age, early stage, and smaller tumor size of HCC compared with untreated patients. After propensity score matching, treated patients demonstrated a higher risk of HCC progression (hazard ratio, 1.42; 95%CI, 1.20–1.69) and death from all causes (1.45; 1.15–1.82) than untreated patients. Similar results were also obtained in sub-cohort of patients who were alive with cancer-free status at least one year after receiving curative treatment and the sub-cohort of patients with liver resection. The interval length between initiation of antiviral therapy and first-line curative treatment did not show a significant association with all-cause mortality. Conclusions This study found that adjuvant antiviral therapy did not reduce the risk of HCC progression or mortality in HBV-related HCC patients after cancer status adjusting.

A metal-oxide-semiconductor (MOS) Ge quantum dot photodetector is demonstrated. The oxide is grown directly on Ge substrate by liquid phase deposition (LPD). The photodetector has the responsivity of 130, 0.16, and 0.08 mA/W under the wavelength of 820 nm, 1300 nm, and 1550 nm, respectively. The dark current is extremely low (0.06 mA/cm/sup 2/). The high performance of Ge quantum dot MOS photodetectors at 820 nm makes it feasible to integrate optoelectronic devices into the Si chip for short-reach optical communication.