[Construction and evaluation of an artificial intelligence-based risk prediction model for death in patients with nasopharyngeal cancer]

OBJECTIVE

To screen the risk factors for death in patients with nasopharyngeal carcinoma (NPC) using artificial intelligence (AI) technology and establish a risk prediction model.

METHODS

The clinical data of NPC patients obtained from SEER database (1973-2015). The patients were randomly divided into model building and verification group at a 7∶3 ratio. Based on the data in the model building group, R software was used to identify the risk factors for death in NPC patients using 4 AI algorithms, namely eXtreme Gradient Boosting (XGBoost), Decision Tree (DT), Least absolute shrinkage and selection operator (LASSO) and random forest (RF), and a risk prediction model was constructed based on the risk factor identified. The C-Index, decision curve analysis (DCA), receiver operating characteristic (ROC) curve and calibration curve (CC) were used for internal validation of the model; the data in the validation group and clinical data of 96 NPC patients (collected from First Affiliated Hospital of Bengbu Medical College) were used for internal and external validation of the model.

RESULTS

The clinical data of a total of 2116 NPC patients were included (1484 in model building group and 632 in verification group). Risk factor screening showed that age, race, gender, stage M, stage T, and stage N were all risk factors of death in NPC patients. The risk prediction model for NPC-related death constructed based on these factors had a C-index of 0.76 for internal evaluation, an AUC of 0.74 and a net benefit rate of DCA of 9%-93%. The C-index of the model in internal verification was 0.740 with an AUC of 0.749 and a net benefit rate of DCA of 3%-89%, suggesting a high consistency of the two calibration curves. In external verification, the C-index of this model was 0.943 with a net benefit rate of DCA of 3%-97% and an AUC of 0.851, and the predicted value was consistent with the actual value.

CONCLUSIONS

Gender, age, race and TNM stage are risk factors of death of NPC patients, and the risk prediction model based on these factors can accurately predict the risks of death in NPC patients.

A reactive molecular dynamics study of thermal pyrolysis behavior and mechanisms of lignin during the hydrothermal process: The function of the water molecules

The lignin hydrothermal processing is an important option but a full understanding of the role played by the water molecules in the depolymerization of lignin is still lacking. In order to clarify the role of the water molecules in the depolymerization of lignin, the evolution of chemical bonds, microstructural changes, and possible mechanisms of product generation were compared during the pyrolysis process under vacuum and water conditions using Reactive Molecular Dynamics Simulation. Compared with vacuum conditions, the role of water changes with temperature, identifying three stages: promotion (1200-1800 K)-inhibition (2100-2400 K)-promotion (2700-3000 K). Also compared with vacuum conditions, hydrothermal processing can promote the cleavage of the ether bonds while inhibiting the destruction of carbocycles. Water molecules promote the depolymerization of lignin into more C₄-molecules, thereby generating more combustible gas resources. Based on the research results, the pyrolysis conditions of lignin can be flexibly controlled to obtain solids, liquids or gases.

Density Functional Studies on the Atomistic Structure and Properties of Iron Oxides: A Parametric Study

With the aim to find the best simulation routine to accurately predict the ground-state structures and properties of iron oxides (hematite, magnetite, and wustite) using density functional theory (DFT) with Hubbard-U correction, a significant amount of DFT calculations were conducted to investigate the influence of various simulation parameters (energy cutoff, K-point, U value, magnetization setting, smearing value, etc.) and pseudopotentials on the structures and properties of iron oxides. With optimized simulation parameters, the obtained equation of state, lattice constant, bulk moduli, and band gap is much closer to the experimental values compared with previous studies. Due to the strong coupling between the 2p orbital of O and the 3d orbital of Fe, it was found that Hubbard-U correction obviously improved the results for all three kinds of iron oxides including magnetite which has not yet been tested with U correction before, but the U value should be different for different oxides (3 ev, 4 ev, 4 ev for hematite, magnetite, and wustite, respectively). Two kinds of spin magnetism settings for FeO are considered, which should be chosen according to different calculation purposes. The detailed relationship between the parameter settings and the atomic structures and properties were analyzed, and the general principles for future DFT calculation of iron oxides were provided.

[Correlation between serum uric acid and creatinine ratio and metabolic syndrome based on physical examination population in Nantong area]

To investigate the relationship between serum uric acid to creatinine ratio (SUA/Cr) and metabolic syndrome (MS) and other indexes on physical examination population in Nantong area. Using the method of cross-sectional study, 8 148 physical examiners in the physical examination center of the Affiliated Hospital of Nantong University from January 2017 to April 2020 were used as the research objects, and the clinical data and serum biochemical indicators such as smoking and alcohol addiction, physical examination and so on were collected. According to the standard diagnosis of MS of Diabetes Society of Chinese Medical Association, the patients were grouped according to the quartile of SUA/Cr and the clinical data of each group were compared. Pearson correlation analysis and logistic regression analysis were used to explore the correlation between SUA/Cr and clinical indicators and the relationship between SUA/Cr and the risk of MS. The results showed that UA and SUA/Cr were the lowest in normal metabolism group, followed by abnormal metabolism group and the highest in MS group, The difference between the two groups was statistically significant (H=919.21 and 629.34, P<0.001). According to the SUA/Cr quartile, the population was divided into four groups. After adjusting for gender, age, smoking history and drinking history, SUA/Cr in group Q₁ was positively correlated with BMI and TG (r=0.061 and 0.080, P<0.05), but negatively correlated with HDL-C (r=-0.057, P<0.05). Multivariate logistic regression results showed that after adjusting for age, sex, smoking history and drinking history, the risk of MS for BMI, SBP, DBP, FBG, TG, HDL-C and SUA/Cr [OR (95%CI)] were: 1.44 (1.41-1.47), 1.07 (1.06-1.07), 1.10 (1.10-1.11), 1.83 (1.73-1.92), 1.89 (1.79-1.99), 0.08 (0.06-0.10) and 1.54 (1.47-1.62). Compared with SUA/Cr group Q₁, the risk of MS in group Q₂, Q₃ and Q₄ increased by 75%, 162% and 346%, respectively. In conclusion, there was an independent positive correlation between SUA/Cr and MS risk in Nantong area.

[A comparison of CAS risk model and CHA2DS2-VASc risk model in guiding anticoagulation treatment in Chinese patients with non-valvular atrial fibrillation]

Objective: To compare the differences between CAS risk model and CHA₂DS₂-VASc risk score in predicting all cause death, thromboembolic events, major bleeding events and composite endpoint in patients with nonvalvular atrial fibrillation. Methods: This is a retrospective cohort study. From the China Atrial Fibrillation Registry cohort study, the patients with atrial fibrillation who were>18 years old were randomly divided into CAS risk score group and CHA₂DS₂-VASc risk score group respectively. According to the anticoagulant status at baseline and follow-up, patients in the 2 groups who complied with the scoring specifications for anticoagulation were selected for inclusion in this study. Baseline information such as age and gender in the two groups were collected and compared. Follow-up was performed periodically to collect information on anticoagulant therapy and endpoints. The endpoints were all-cause death, thromboembolism events and major bleeding, the composite endpoint events were all-cause death and thromboembolism events. The incidence of endpoints in CAS group and CHA₂DS₂-VASc group was analyzed, and multivariate Cox proportional risk model was used to analyze whether the incidence of the endpoints was statistically different between the two groups. Results: A total of 5 206 patients with AF were enrolled, average aged (63.6±12.2) years, and 2092 (40.2%) women. There were 2 447 cases (47.0%) in CAS risk score group and 2 759 cases (53.0%) in CHA₂DS₂-VASc risk score group. In the clinical baseline data of the two groups, the proportion of left ventricular ejection fraction<55%, non-paroxysmal atrial fibrillation, oral warfarin and HAS BLED score in the CAS group were lower than those in the CHA₂DS₂-VASc group, while the proportion of previous diabetes history and history of antiplatelet drugs in the CAS group was higher than that in the CHA₂DS₂-VASc group, and there was no statistical difference in other baseline data. Patients were followed up for (82.8±40.8) months. In CAS risk score group, 225(9.2%) had all-cause death, 186 (7.6%) had thromboembolic events, 81(3.3%) had major bleeding, and 368 (15.0%) had composite endpoint. In CHA₂DS₂-VASc risk score group, 261(9.5%) had all-cause death 209(7.6%) had thromboembolic events, 112(4.1%) had major bleeding, and 424 (15.4%) had composite endpoint. There were no significant differences in the occurrence of all-cause death, thromboembolic events, major bleeding and composite endpoint between anticoagulation in CAS risk score group and anticoagulation in CHA₂DS₂-VASc risk score group (log-rank P =0.643, 0.904, 0.126, 0.599, respectively). Compared with CAS risk score, multivariable Cox proportional hazards regression models showed no significant differences for all-cause death, thromboembolic events, major bleeding and composite endpoint between the two groups with HR(95%CI) 0.95(0.80-1.14), 1.00(0.82-1.22), 0.83(0.62-1.10), 0.96(0.84-1.11), respectively. All P>0.05. Conclusions: There were no significant differences between CAS risk model and CHA₂DS₂-VASc risk score in predicting all-cause death, thromboembolic events, and major bleeding events in Chinese patients with non-valvular atrial fibrillation.

New Insights into the Traditional Charge Compensation Theory: Amphoteric Behavior of TiO2 under the Guidance of Supply-Demand Relationship

Polymerization degree theory and traditional charge compensation theory are the most fundamental principles to understand the structure and properties of oxide melts. It can well explain the behavior characteristics of acidic oxides and basic oxides in a melt. However, the amphoteric behavior of oxides cannot be explained well by these two theories. Herein, the octahedral connection mode and the behavior of the amphoteric transition of TiO₂ are analyzed by molecular dynamics simulation, and then, a calculation model which can quantitatively calculate the amphoteric transition of the oxide is established by analyzing a large number of data. On the basis of the model, a novel theory of supply and demand is put forward, which can explain the amphoteric transition behavior of oxides very well. To a great extent, the supply and demand theory makes up for the deficiency of the atomic structure theory of oxide melts and provides mechanism explanation and model prediction for the oxide amphoteric transformation behavior.

[The relationship between fasting blood glucose level and thromboembolism events in patients with non-valvular atrial fibrillation]

Objective: To explore the relationship between fasting blood glucose level and thromboembolism events in patients with non-valvular atrial fibrillation (NVAF). Methods: This was an observational study based on data from a multicenter, prospective Chinese atrial fibrillation registry cohort, which included 18 703 consecutive patients with atrial fibrillation (AF) in 31 hospitals in Beijing from August 2011 to December 2018. Patients were divided into 5 groups according to status of comorbid diabetes and fasting glucose levels at admission: normal blood glucose (normal glucose group), pre-diabetes group, strict glycemic control group, average glycemic control group and poor glycemic control group. Patients were followed up by telephone or outpatient service every 6 months. The primary follow-up endpoint was thromboembolic events, including ischemic stroke and systemic embolism. The secondary endpoint was the composite endpoint of cardiovascular death and thromboembolic events. Kaplan-Meier survival analysis and multifactorial Cox regression were used to analyze the correlation between fasting glucose levels and endpoint events. Results: The age of 18 703 patients with NVAF was (63.8±12.0) years, and there were 11 503 (61.5%) male patients. There were 11 877 patients (63.5%) in normal blood glucose group, 2 023 patients (10.8%)in pre-diabetes group, 1 131 patients (6.0%) in strict glycemic control group, 811 patients in average glycemic control group and 2 861 patients(4.3%) in poor glycemic control group. Of the 4 803 diabetic patients, 1 131 patients (23.5%) achieved strict glycemic control, of whom 328 (29.0%) were hypoglycemic (fasting blood glucose level<4.4 mmol/L at admission). During a mean follow-up of (51±23) months (up to 82 months), thromboembolic events were reported in 984 patients (5.3%). The survival curve analysis of Kaplan Meier showed that the incidence rates of thromboembolic events in normal glucose group, pre-diabetes group, strict glycemic control group, average glycemic control group and poor glycemic control group were 1.10/100, 1.41/100, 2.09/100, 1.46/100 and 1.71/100 person-years, respectively (χ²=53.0, log-rank P<0.001). The incidence rates of composite endpoint events were 1.86/100, 2.17/100, 4.08/100, 2.58/100, 3.16/100 person-years (χ²=72.3, log-rank P<0.001). The incidence of thromboembolic events and composite endpoint events in the other four groups were higher than that in the normal blood glucose group (P<0.001). Multivariate Cox regression analysis showed that compared with normal glucose group, the risk of thromboembolism increased in pre-diabetes group(HR=1.23, 95%CI 1.00-1.51, P=0.049), strict glycemic control group(HR=1.32, 95%CI 1.06-1.65, P=0.013) and poor glycemic control group(HR=1.26, 95%CI 1.01-1.58, P=0.044). Conclusion: Both high or low fasting glucose may be an independent risk factor for thromboembolic events in patients with NVAF.

The Inverted U-Shaped Association of Caffeine Intake with Serum Uric Acid in U.S. Adults

BACKGROUND AND AIMS

Caffeine is a worldwide popularly consumed constituent in foods that can exert physiological effects. However, previous researches about the relationship between caffeine intake and serum uric acid (SUA) were limited and controversial. Therefore, we sought to investigate that relationship in U.S. adults.

METHODS

In this cross-sectional study, the total sample of 7888 selected participants (3838 males and 4050 females) were identified from the National Health and Nutritional Examination Surveys (NHANES) 2015-2018. All subjects were tested for serum uric acid levels (μmmol/L), and their daily caffeine intakes (mg/d) were obtained by an average of two 24-hour dietary recalls. Multivariate linear regression models were used to evaluate the association between two variables in total subjects and subgroup analyses. Generalized additive models with smooth curve fittings were also performed.

RESULTS

Multivariate regression analyses showed caffeine intake was negatively correlated with SUA after adjustment of other confounders. The subgroup analyses stratified by gender showed the negative correlation of caffeine intake with SUA was statistically significant in males but not in females. Furthermore, we observed a nonlinear inverse association of caffeine intake with SUA (P nonlinear <0.001) in the generalized additive model, followed by an inverted U-shaped curve (inflection point: 60.5mg/d) for all participants. This inverted U-shaped relationship between them could also be found in both genders, individuals aged below 60 years old, overweight (BMI of 25 to 30), and Non-Hispanic White individuals.

CONCLUSIONS

This study indicated that caffeine intake exhibited an inverse correlation with SUA, especially in males. In addition, this inverse relationship was nonlinear, which followed an inverted U-shaped curve.

J-shaped relationship between admission diastolic blood pressure and 2-year cardiovascular mortality in elderly patients with acute coronary syndrome

Objective

To investigate the relationship between admission diastolic blood pressure (DBP) and subsequent cardiovascular and all-cause mortality in elderly patients with acute coronary syndrome (ACS).

Methods

This is a retrospective observational study. Consecutive patients ≥65 years of age admitted for ACS at a 2,300-bed tertiary hospital from December 2012 to July 2019 were included. The association between admission DBP and cardiovascular and all-cause mortality during hospitalization and over the follow-up period among this population were analyzed using multivariate COX regression model. Results were presented according to DBP quartiles: Q1, less than 67 mm Hg; Q2, from 67 to 72 mm Hg; Q3, from 73 to 80 mm Hg; and Q4, above 80 mm Hg.

Results

A total of 6 785 patients were included in this cohort study. Mean (SD) patient age was 74.0 (6.5) years, and 47.6% were women. Mean (SD) follow-up time was 2.54 (1.82) years. A non-linear relation was observed between DBP at admission and cardiovascular and all-cause mortality during hospitalization and over the follow-up period using restricted cubic splines. After adjustment for potential confounders, patients in Q3 or Q2 had lower risk for 2-year cardiovascular death by Cox proportional hazard model compared with patients in Q4 (hazard ratio [HR] 0.66; 95% confidence interval [CI], 0.48–0.90, P=0.010, for Q3 vs Q4; and HR 0.72; 95% CI, 0.53–0.99, P=0.041, for Q2vs Q4), while patients in Q1 had similar risk for cardiovascular death with that of patients in Q4. Meanwhile, when compared with patients in Q1, patients in Q3 had lower risk for 2-year cardiovascular death (HR, 0.72; 95% CI, 0.53–0.97, P=0.033). However, lower or higher admission DBP was not an independent predictor of 2-year all-cause mortality in this population.

Conclusion

Among patients aged ≥65 years admitted for ACS, there is a J-curve relationship between supine admission DBP and risk for 2-year cardiovascular death, with a nadir at 73–80 mm Hg.

Funding Acknowledgement

Type of funding sources: Other. Main funding source(s): the Beijing Municipal Administration of Hospitals Clinical Medicine Development of Special Funding Support Study population and selectionAdjusted multivariate COX regression

[Safety and effectiveness of left atrial appendage occlusion in patients with atrial fibrillation and end-stage renal disease undergoing long-term hemodialysis]

Objective: To evaluate the safety and effectiveness of left atrial appendage occlusion (LAAO) in patients with atrial fibrillation and end-stage renal disease undergoing long-term hemodialysis. Methods: Six patients with AF and end-stage renal disease(ESRD)on long term hemodialysis who underwent LAAO from March 2017 to March 2021 in Beijing Anzhen Hospital were enrolled. Baseline characteristics such as age, sex, types of arrhythmia, stroke and bleeding score, and continuous dialysis time were collected. Four patients underwent LAAO, two patients underwent the combined procedure of catheter ablation and LAAO. Perioperative treatment and serious complications were recorded. Transesophageal echocardiography was repeated at 45 days and 60 days after the procedure. Telephone follow-up was conducted at 3, 6 and 12 months after the procedure, and every 6 months thereafter. Thromboembolism and major bleeding events and survival were evaluated. Results: The average age was (66.7±17.0) years old, and 5 were male (5/6). There were 4 patients with paroxysmal AF (4/6), and 2 patients with persistent AF (2/6). The mean CHA₂DS₂-VASc score was (4.8±1.5), and the HAS-BLED score was (3.5±1.4). The duration of hemodialysis was 2.6 (1.1, 8.3) years. Successfully Watchman implantation was achieved in all patients. There were no severe perioperative complications, and no device related thrombosis or leaks were observed by transesophageal echocardiography. During a mean of 22.0 (12.0, 32.0) months follow-up, there was no thromboembolism or major bleeding events. A total of 2 patients died, one from sudden cardiac death, and another one from heart failure. Conclusions: LAAO may be a safe and effective therapeutic option for prevention of thromboembolism in patients with atrial fibrillation and end-stage renal disease undergoing long-term hemodialysis, further studies with larger patient cohort are needed to confirm our results.

Probiotic and prebiotic interventions for non-alcoholic fatty liver disease: a systematic review and network meta-analysis

Many studies have associated altered intestinal bacterial communities and non-alcoholic fatty liver disease, but the putative effects are inconclusive. The purpose of this network meta-analysis (NMA) was to evaluate the effects of probiotics, prebiotics, and synbiotics on non-alcoholic fatty liver disease through randomised intervention trials. Literature searches were performed until March 2020. For each outcome, a random NMA was performed, the surface under the cumulative ranking curve (SUCRA) was determined. A total of 22 randomised trials comparing prebiotic, probiotic, and synbiotic treatments included 1301 participants. Considering all seven results (aspartate aminotransferase, alanine aminotransferase, body mass index, weight, total cholesterol, high-density lipoprotein cholesterol, and low-density lipoprotein cholesterol) together, the highest SUCRA values are probiotics (94%), synbiotics (61%) and prebiotics (56%), respectively. NMA results provide evidence that probiotics, prebiotics, and synbiotics can alleviate non-alcoholic fatty liver disease. However, due to the lack of high-quality randomised trials, this research also has some limitations.

[Clinical analysis of left atrial appendage occlusion for stroke prevention in elderly patients with atrial fibrillation]

To evaluate the efficacy and safety of left atrial appendage occlusion (LAAO) in patients with atrial fibrillation (AF) over 75 years. A total of 82 patients with AF who underwent LAAO successfully in Beijing Anzhen Hospital from March 2014 to March 2019 were divided into two groups according to age: the elderly group (aged>75 years) and the young group (aged ≤75 years). Risk of perioperative complications and incidence of ischemic stroke and major bleeding during follow-up were retrospectively analyzed. The results showed that there were no significant differences in procedure-related ischemic stroke(0 vs.1.6%,P=0.768) and major bleeding (0 vs.1.6%,P=0.768) during perioperative period between the two groups. No complications as death or pericardial tamponade occurred in the two group. During a (25.9±15.9) months period of followed up, ischemic stroke event rate was 3.6/100 person-years in the elderly group and 4.9/100 person-years in the young group, respectively. Major bleeding event rate was 2.5/100 person-years in the elderly group and 0/100 person-years in the young group, respectively. Compared with the expected ones, the relative risk reduction (RRR) of stroke in the elderly group was more profound than that in the young group (32.0% vs. 25.0%), while the risk of major bleeding in the young group was significantly lower than that in the elderly group (RRR 100% vs. 56.9%). Therefore, LAAO might be suitable for stroke prevention in the elderly AF patients.

MiRNA-15a-3p inhibits the metastasis of hepatocellular carcinoma by interacting with HMOX1

OBJECTIVE

To uncover the relationship between microRNA-15a-3p (miRNA-15a-3p) level and clinical features of hepatocellular carcinoma (HCC), and to explore the influence of miRNA-15a-3p on metastasis of HCC cells.

PATIENTS AND METHODS

HCC and paracancerous tissues were surgically resected from 44 HCC patients. Their clinical data and follow-up files were recorded. Differential expressions of miRNA-15a-3p in HCC samples were determined. The relationship between miRNA-15a-3p level and clinical features of HCC patients was analyzed. Changes in proliferative, migratory and invasive potentials in Huh7 and HepG2 cells overexpressing miRNA-15a-3p were examined. The downstream gene of miRNA-15a-3p and its involvement in HCC development were finally explored.

RESULTS

MiRNA-15a-3p was downregulated in HCC tissues. High metastasis rate and poor prognosis were observed in HCC patients expressing a low level of miRNA-15a-3p. Overexpression of miRNA-15a-3p attenuated proliferative, migratory and invasive potentials in HCC. Protein levels of HMOX1, CD31, c-Myc, MMP-2 and MMP-9 were downregulated in HCC cells after overexpression of miRNA-15a-3p. HMOX1 was the downstream gene of miRNA-15a-3p, which was upregulated in HCC samples. Highly expressed HMOX1 was unfavorable to the prognosis in HCC. Overexpression of HMOX1 abolished the regulatory effects of miRNA-15a-3p on HCC cell phenotypes.

CONCLUSIONS

MiRNA-15a-3p is closely linked to lymphatic metastasis, distant metastasis and poor prognosis in HCC. It inhibits the malignant development of HCC by interacting with HMOX1.

[Causes of death and influencing factors of atrial fibrillation patients undergoing anticoagulation therapy]

Objective: To investigate the causes of death and predictors in patients with nonvalvular atrial fibrillation (AF) undergoing anticoagulation therapy. Methods: Consecutive anticoagulated nonvalvular AF patients were recruited from the China Atrial Fibrillation Registry (China-AF) Study from August 2011 to December 2018. After exclusion of patients with hypertrophic cardiomyopathy, dilated cardiomyopathy, or loss of follow-up within 1 year, 2 248 patients were included in this analysis. Enrolled patients were followed up were followed up for 3 and 6 months, and then every 6 months. The primary endpoint was death, including cardiovascular death, non-cardiovascular death and undetermined death. The patients were divided into survival group and death group according to the survival status after follow-up. Clinical information such as age and sex was collected. Cox proportional hazards regression was performed to identify associated risk factors for all-cause mortality, and Fine-Gray competing risk model was used to identify associated risk factors for cardiovascular mortality. Results: A total of 2 248 patients with atrial fibrillation receiving anticoagulant therapy died over a mean follow-up of (42±24) months, mean age was (67±10) years old and 41.1% (923/2 248) patients were female. The mortality rate was 2.8 deaths per 100 patient-years. The most common cause of death was cardiovascular deaths, accounted for 55.0% (120/218). Worsening heart failure was the most common cause of cardiovascular deaths (18.3% (40/218)), followed by bleeding events (12.9% (28/218)) and ischemic stroke (8.7% (19/218)). Multivariate Cox regression analysis showed that age (HR = 1.05, 95%CI 1.04-1.07, P<0.001), anemia (HR = 1.81, 95%CI 1.02-3.18, P = 0.041), heart failure (HR=2.40, 95%CI 1.75-3.30, P<0.001), ischemic stroke/transient ischemic attack (TIA)(HR = 1.59, 95%CI 1.21-2.13, P = 0.001) and myocardial infarction (HR = 2.93, 95%CI 1.79-4.81, P<0.001) were independently associated with all-cause death. Fine-Gray competing risk model showed that age (HR=1.05, 95%CI 1.02-1.08, P<0.001), heart failure (HR=2.81, 95%CI 1.79-4.39, P<0.001), ischemic stroke/TIA (HR=1.50, 95%CI 1.02-2.22, P=0.041) and myocardial infarction (HR=3.31, 95%CI 1.72-6.37, P<0.001) were independently associated with cardiovascular death. Conclusions: In anticoagulated nonvalvular AF patients, ischemic stroke represents only a small subset of deaths, whereas worsening heart failure is the most common cause of cardiovascular deaths. Heart failure, ischemic stroke/TIA, and myocardial infarction are associated with increased mortality.

Physics-informed machine learning: case studies for weather and climate modelling

Machine learning (ML) provides novel and powerful ways of accurately and efficiently recognizing complex patterns, emulating nonlinear dynamics, and predicting the spatio-temporal evolution of weather and climate processes. Off-the-shelf ML models, however, do not necessarily obey the fundamental governing laws of physical systems, nor do they generalize well to scenarios on which they have not been trained. We survey systematic approaches to incorporating physics and domain knowledge into ML models and distill these approaches into broad categories. Through 10 case studies, we show how these approaches have been used successfully for emulating, downscaling, and forecasting weather and climate processes. The accomplishments of these studies include greater physical consistency, reduced training time, improved data efficiency, and better generalization. Finally, we synthesize the lessons learned and identify scientific, diagnostic, computational, and resource challenges for developing truly robust and reliable physics-informed ML models for weather and climate processes. This article is part of the theme issue 'Machine learning for weather and climate modelling'.