Objective:To assess the fall risk of hospitalized patients with cardiovascular diseases, analyze the related influencing factors, and construct a prediction model based on machine learning technology, so as to provide a basis for the fall management of hospitalized patients with cardiovascular diseases.Methods:This study was a retrospective cohort study. A total of 450 patients admitted to the Department of Cardiology, the First Affiliated Hospital of Nanjing Medical University from June 2017 to June 2024 were selected as the research objects by convenience sampling method. By reviewing electronic medical records, trained nurses extracted the patients' general information and Activities of Daily Living Scale (ADL) scores during hospitalization. Lasso regression was used to screen risk factors, and machine learning libraries were used to construct support vector machine (SVM), decision tree, XGBoost, and neural network models. Bootstrap resampling method and area under the curve (AUC) were used to verify the model performance.Results:Among the 450 patients, there were 261 males and 189 females, with a mean age of (66.0 ± 8.4) years. Among them, 90 patients fell during hospitalization and 360 patients did not fall. The results of Lasso regression showed that ADL score ≤60 points, use of hypnotics, hypokalemia, nighttime toilet visits≥2 times, use of antihypertensive drugs, no caregiver, and history of atrial fibrillation were all risk factors for falls in hospitalized patients with cardiovascular diseases (regression coefficients ranging from 0.61 to 1.20, all P<0.01). Among the machine learning models, XGBoost had the best comprehensive performance (AUC=0.98), which was better than decision tree (AUC=0.66), SVM (AUC=0.95), and neural network (AUC=0.87). Conclusions:The fall risk of hospitalized patients with cardiovascular diseases is jointly affected by physiological, medication and behavioral factors, and the XGBoost model can effectively identify high-risk groups. In actual clinical work, nursing strategies can be optimized in combination with risk factors, and the application of intelligent fall prediction and assessment tools can be promoted.

Objective:To assess the fall risk of hospitalized patients with cardiovascular diseases, analyze the related influencing factors, and construct a prediction model based on machine learning technology, so as to provide a basis for the fall management of hospitalized patients with cardiovascular diseases.Methods:This study was a retrospective cohort study. A total of 450 patients admitted to the Department of Cardiology, the First Affiliated Hospital of Nanjing Medical University from June 2017 to June 2024 were selected as the research objects by convenience sampling method. By reviewing electronic medical records, trained nurses extracted the patients' general information and Activities of Daily Living Scale (ADL) scores during hospitalization. Lasso regression was used to screen risk factors, and machine learning libraries were used to construct support vector machine (SVM), decision tree, XGBoost, and neural network models. Bootstrap resampling method and area under the curve (AUC) were used to verify the model performance.Results:Among the 450 patients, there were 261 males and 189 females, with a mean age of (66.0 ± 8.4) years. Among them, 90 patients fell during hospitalization and 360 patients did not fall. The results of Lasso regression showed that ADL score ≤60 points, use of hypnotics, hypokalemia, nighttime toilet visits≥2 times, use of antihypertensive drugs, no caregiver, and history of atrial fibrillation were all risk factors for falls in hospitalized patients with cardiovascular diseases (regression coefficients ranging from 0.61 to 1.20, all P<0.01). Among the machine learning models, XGBoost had the best comprehensive performance (AUC=0.98), which was better than decision tree (AUC=0.66), SVM (AUC=0.95), and neural network (AUC=0.87). Conclusions:The fall risk of hospitalized patients with cardiovascular diseases is jointly affected by physiological, medication and behavioral factors, and the XGBoost model can effectively identify high-risk groups. In actual clinical work, nursing strategies can be optimized in combination with risk factors, and the application of intelligent fall prediction and assessment tools can be promoted.

Objective:To evaluate the measurement agreement of Roche 25(OH)D immunoassay(evaluation method) with LC-MS/MS (reference method).Methods:A total of 909 residual serum samples from routine health check participants were collected from May to June in 2019. 25(OH)D concentrations were measured by evaluation method and LC-MS/MS, respectively. Passing-bablok regression, intraclass correlation coefficient (ICC), Bland Altman plots and Kappa test were used to analyze the consistency and bias on the results derived from the two measurement methods.Results:The 25(OH)D concentration derived from evaluation method was significantly different from those from LC-MS/MS method ( P<0.001). Slope of regression for evaluation method and LC-MS/MS was 0.962(95% CI 0.919-1.007), while intercept was -0.185 (95% CI -1.191-0.745). The ICC was 0.765 (95% CI 0.735-0.792). Altman plot showed that the average deviation between evaluation method and LC-MS/MS was -0.902 ng/ml (0.300%). The coincidence rate of evaluation method′s judgment of vitamin D sufficiency, insufficiency and deficiency with LC-MS/MS was 83.39%, and the weighted Kappa values was 0.790. Conclusion:Roche automatic 25(OH)D immunoassay shows acceptable correlation and agreement with LC-MS/MS, however, it is to note that the deviation between immunoassay and LC-MS/MS may lead to wrong judgment of vitamin D nutritional status. It is recommended that each laboratory should establish own corresponding reference values for 25(OH)D concentrations derived from these two methods.

After the stratification of the normal glucose tolerance, the changes of insulin resistance and βcell function in the development of type 2 diabetes mellitus were investigated. A retrospective analysis on data of 275 cases with oral glucose insulin releasing tests. The area under the insulin curve (AUCINS ) 108. 43 mU/ L was taken as the critical value of diagnosis. Normal glucose tolerance subjects were divided into the NGT-a group(AUCINS<108. 43 mU/ L) and the NGT-b group(AUCINS≥108. 43 mU/ L). The plasma glucose, insulin, insulin sensitivity, and β cell function were compared among the 4 groups: NGT-a group (n=96), NGT-b group (n=49), prediabetes group (n=71), and type 2 diabetes mellitus group ( n = 59). Among the fasting insulin, 2 h insulin, AUCINS , early-phase insulin secretion index(△I30 / △G30), the ratio of total insulin area under curve, and total glucose area under curve, disposition index, homeostasis model assessment for insulin resistance, and Matsuda insulin sensitivity index, the relationship as follows: NGT-b group>prediabetes group>NGT-a group>type 2 diabetes mellitus group. The NGT-b group was always the highest, prediabetes group was lower, NGT-a group and type 2 diabetes mellitus group were the lowest, there were significant differences (all P<0. 05). Making the NGT-a group as the basic state, in the NGT-b group, β cell function has begun to appear compensation and insulin resistance, and β cell function compensation reached the peak, the β cell function in the prediabetes group was beginning to compensate for the deficiency, the function of β cell in type 2 diabetes mellitus group decreased further. These findings suggest that the development process of type 2 diabetes mellitus could be the following four stages according to the function of β cell: β cell function normal, β cell functional compensation, β cell function loss of compensation, and finally β cell function failure.

[Summary] The high-risk subjects of type 2 diabetes mellitus ( T2DM) in normal glucose tolerance ( NGT) were screened. The subjects with NGT at baseline were divided into high-risk and low-risk groups according to the diagnostic threshold of insulin area under the curve ( AUCINS ) 108. 43 mU/L. The incidence of prediabetes and/or T2DM was significantly increased in high risk group in comparison with low risk group ( 29. 41 vs 2. 21%, P<0. 01). The result suggests that the diagnosis threshold for AUCINS≥108. 43 mU/L can be used to screen the high-risk subjects of T2DM in NGT.