A dosing strategy model of deep deterministic policy gradient algorithm for sepsis patients.

BACKGROUND: A growing body of research suggests that the use of computerized decision support systems can better guide disease treatment and reduce the use of social and medical resources. Artificial intelligence (AI) technology is increasingly being used in medical decision-making systems to obtain optimal dosing combinations and improve the survival rate of sepsis patients. To meet the real-world requirements of medical applications and make the training model more robust, we replaced the core algorithm applied in an AI-based medical decision support system developed by research teams at the Massachusetts Institute of Technology (MIT) and IMPERIAL College London (ICL) with the deep deterministic policy gradient (DDPG) algorithm. The main objective of this study was to develop an AI-based medical decision-making system that makes decisions closer to those of professional human clinicians and effectively reduces the mortality rate of sepsis patients. METHODS: We used the same public intensive care unit (ICU) dataset applied by the research teams at MIT and ICL, i.e., the Multiparameter Intelligent Monitoring in Intensive Care III (MIMIC-III) dataset, which contains information on the hospitalizations of 38,600 adult sepsis patients over the age of 15. We applied the DDPG algorithm as a strategy-based reinforcement learning approach to construct an AI-based medical decision-making system and analyzed the model results within a two-dimensional space to obtain the optimal dosing combination decision for sepsis patients. RESULTS: The results show that when the clinician administered the exact same dose as that recommended by the AI model, the mortality of the patients reached the lowest rate at 11.59%. At the same time, according to the database, the baseline mortality rate of the patients was calculated as 15.7%. This indicates that the patient mortality rate when difference between the doses administered by clinicians and those determined by the AI model was zero was approximately 4.2% lower than the baseline patient mortality rate found in the dataset. The results also illustrate that when a clinician administered a different dose than that recommended by the AI model, the patient mortality rate increased, and the greater the difference in dose, the higher the patient mortality rate. Furthermore, compared with the medical decision-making system based on the Deep-Q Learning Network (DQN) algorithm developed by the research teams at MIT and ICL, the optimal dosing combination recommended by our model is closer to that given by professional clinicians. Specifically, the number of patient samples administered by clinicians with the exact same dose recommended by our AI model increased by 142.3% compared with the model based on the DQN algorithm, with a reduction in the patient mortality rate of 2.58%. CONCLUSIONS: The treatment plan generated by our medical decision-making system based on the DDPG algorithm is closer to that of a professional human clinician with a lower mortality rate in hospitalized sepsis patients, which can better help human clinicians deal with complex conditional changes in sepsis patients in an ICU. Our proposed AI-based medical decision-making system has the potential to provide the best reference dosing combinations for additional drugs.

Can the combination of time-lapse parameters and clinical features predict embryonic ploidy status or implantation?

RESEARCH QUESTION: Can models based on artificial intelligence predict embryonic ploidy status or implantation potential of euploid transferred embryos? Can the addition of clinical features into time-lapse monitoring (TLM) parameters as input data improve their predictive performance? DESIGN: A single academic fertility centre, retrospective cohort study. A total of 773 high-grade euploid and aneuploid blastocysts from 212 patients undergoing preimplantation genetic testing (PGT) between July 2016 and July 2021 were studied for ploidy prediction. Among them, 170 euploid embryos were single-transferred and included for implantation analysis. Five machine learning models and two types of deep learning networks were used to develop the predictive algorithms. The predictive performance was measured using the area under the receiver operating characteristic curve (AUC), in addition to accuracy, precision, recall and F1 score. RESULTS: The most predictive model for ploidy prediction had an AUC, accuracy, precision, recall and F1 score of 0.70, 0.64, 0.64, 0.50 and 0.56, respectively. The DNN-LSTM model showed the best predictive performance with an AUC of 0.78, accuracy of 0.77, precision of 0.79, recall of 0.86 and F1 score of 0.83. The predictive power was improved after the addition of clinical features for the algorithms in ploidy prediction and implantation prediction. CONCLUSION: Our findings emphasize that clinical features can largely improve embryo prediction performance, and their combination with TLM parameters is robust to predict high-grade euploid blastocysts. The models for ploidy prediction, however, were not highly predictive, suggesting they cannot replace preimplantation genetic testing currently.

Construction of machine learning tools to predict threatened miscarriage in the first trimester based on AEA, progesterone and ß-hCG in China: a multicentre, observational, case-control study.

BACKGROUND: Endocannabinoid anandamide (AEA), progesterone (P4) and ß-human chorionic gonadotrophin (ß-hCG) are associated with the threatened miscarriage in the early stage. However, no study has investigated whether combing these three hormones could predict threatened miscarriage. Thus, we aim to establish machine learning models utilizing these three hormones to predict threatened miscarriage risk. METHODS: This is a multicentre, observational, case-control study involving 215 pregnant women. We recruited 119 normal pregnant women and 96 threatened miscarriage pregnant women including 58 women with ongoing pregnancy and 38 women with inevitable miscarriage. P4 and ß-hCG levels were detected by chemiluminescence immunoassay assay. The level of AEA was tested by ultra-high-performance liquid chromatography-tandem mass spectrometry. Six predictive machine learning models were established and evaluated by the confusion matrix, area under the receiver operating characteristic (ROC) curve (AUC), accuracy and precision. RESULTS: The median concentration of AEA was significantly lower in the healthy pregnant women group than that in the threatened miscarriage group, while the median concentration of P4 was significantly higher in the normal pregnancy group than that in the threatened miscarriage group. Only the median level of P4 was significantly lower in the inevitable miscarriage group than that in the ongoing pregnancy group. Moreover, AEA is strongly positively correlated with threatened miscarriage, while P4 is negatively correlated with both threatened miscarriage and inevitable miscarriage. Interestingly, AEA and P4 are negatively correlated with each other. Among six models, logistic regression (LR), support vector machine (SVM) and multilayer perceptron (MLP) models obtained the AUC values of 0.75, 0.70 and 0.70, respectively; and their accuracy and precision were all above 0.60. Among these three models, the LR model showed the highest accuracy (0.65) and precision (0.70) to predict threatened miscarriage. CONCLUSIONS: The LR model showed the highest overall predictive power, thus machine learning combined with the level of AEA, P4 and ß-hCG might be a new approach to predict the threatened miscarriage risk in the near feature.

Relationship between small dense low-density lipoprotein cholesterol with carotid plaque in Chinese individuals with abnormal carotid artery intima-media thickness.

AIM: To investigate the relationship of small dense low-density lipoprotein cholesterol (sdLDL-C) to carotid artery intima-media thickness (CA-IMT) and carotid plaque (CAP) in Chinese general population, and to evaluate whether sdLDL-C could be an independent risk factor for individuals with subclinical atherosclerosis. METHODS: A total of 729 subjects were randomly collected from consecutive individuals from April 2019 to April 2020 for an annual health checkup. CA-IMT > 1.0 mm was defined as abnormal IMT. Plaque stability was measured by ultrasound examination based on the property of the echo. And sdLDL-C levels were detected by LipoPrint system. Multivariate logistic regression analysis was performed to identify factors associated with CA-IMT and carotid plaque. RESULTS: The abnormal IMT group had significantly higher sdLDL-C levels than control group (p < 0.0001). And sdLDL-C levels were significantly positively correlated with IMT value (r = 0.1396, p = 0.0021) and presence of carotid plaque (r = 0.14, p = 0.002) in the subjects with abnormal IMT. In addition, subjects with higher levels of sdLDL-C (r = 0.11, p = 0.035) tended to have unstable CAP. After adjustment for age, gender and blood glucose, sdLDL-C level was an independent risk factor of the presence of CAP (OR = 1.59, 95% CI: 1.02-1.83, p = 0.034) in subjects with abnormal IMT. CONCLUSION: SdLDL-C is an independent risk factor of the occurrence of CAP in the Chinese subjects with abnormal IMT. Our findings provide supporting evidence that sdLDL-C might be an alternative way to predict CVD in early stage.

Development of Machine Learning Tools for Predicting Coronary Artery Disease in the Chinese Population.

Purpose: Coronary artery disease (CAD) is one of the major cardiovascular diseases and the leading cause of death globally. Blood lipid profile is associated with CAD early risk. Therefore, we aim to establish machine learning models utilizing blood lipid profile to predict CAD risk. Methods: In this study, 193 non-CAD controls and 2001 newly-diagnosed CAD patients (1647 CAD patients who received lipid-lowering therapy and 354 who did not) were recruited. Clinical data and the result of routine blood lipids tests were collected. Moreover, low-density lipoprotein cholesterol (LDL-C) subfractions (LDLC-1 to LDLC-7) were classified and quantified using the Lipoprint system. Six predictive models (k-nearest neighbor classifier (KNN), logistic regression (LR), support vector machine (SVM), decision tree (DT), multilayer perceptron (MLP), and extreme gradient boosting (XGBoost)) were established and evaluated by the confusion matrix, area under the receiver operating characteristic (ROC) curve (AUC), recall (sensitivity), accuracy, precision, and F1 score. The selected features were analyzed and ranked. Results: While predicting the CAD development risk of the CAD patients without lipid-lowering therapy in the test set, all models obtained AUC values above 0.94, and the accuracy, precision, recall, and F1 score were above 0.84, 0.85, 0.92, and 0.88, respectively. While predicting the CAD development risk of all CAD patients in the test set, all models obtained AUC values above 0.91, and the accuracy, precision, recall, and F1 score were above 0.87, 0.94, 0.87, and 0.92, respectively. Importantly, small dense LDL-C (sdLDL-C) and LDLC-4 play pivotal roles in predicting CAD risk. Conclusions: In the present study, machine learning tools combining both clinical data and blood lipid profile showed excellent overall predictive power. It suggests that machine learning tools are suitable for predicting the risk of CAD development in the near future.

Prediction models constructed for Hashimoto's thyroiditis risk based on clinical and laboratory factors.

Background: Hashimoto's thyroiditis (HT) frequently occurs among autoimmune diseases and may simultaneously appear with thyroid cancer. However, it is difficult to diagnose HT at an early stage just by clinical symptoms. Thus, it is urgent to integrate multiple clinical and laboratory factors for the early diagnosis and risk prediction of HT. Methods: We recruited 1,303 participants, including 866 non-HT controls and 437 diagnosed HT patients. 44 HT patients also had thyroid cancer. Firstly, we compared the difference in thyroid goiter degrees between controls and patients. Secondly, we collected 15 factors and analyzed their significant differences between controls and HT patients, including age, body mass index, gender, history of diabetes, degrees of thyroid goiter, UIC, 25-(OH)D, FT3, FT4, TSH, TAG, TC, FPG, low-density lipoprotein cholesterol, and high-density lipoprotein cholesterol. Thirdly, logistic regression analysis demonstrated the risk factors for HT. For machine learning modeling of HT and thyroid cancer, we conducted the establishment and evaluation of six models in training and test sets. Results: The degrees of thyroid goiter were significantly different among controls, HT patients without cancer (HT-C), and HT patients with thyroid cancer (HT+C). Most factors had significant differences between controls and patients. Logistic regression analysis confirmed diabetes, UIC, FT3, and TSH as important risk factors for HT. The AUC scores of XGBoost, LR, SVM, and MLP models indicated appropriate predictive power for HT. The features were arranged by their importance, among which, 25-(OH)D, FT4, and TSH were the top three high-ranking factors. Conclusions: We firstly analyzed comprehensive factors of HT patients. The proposed machine learning modeling, combined with multiple factors, are efficient for thyroid diagnosis. These discoveries will extensively promote precise diagnosis, personalized therapies, and reduce unnecessary cost for thyroid diseases.

Application of machine learning tools: Potential and useful approach for the prediction of type 2 diabetes mellitus based on the gut microbiome profile.

The gut microbiota plays an important role in the regulation of the immune system and the metabolism of the host. The aim of the present study was to characterize the gut microbiota of patients with type 2 diabetes mellitus (T2DM). A total of 118 participants with newly diagnosed T2DM and 89 control subjects were recruited in the present study; six clinical parameters were collected and the quantity of 10 different types of bacteria was assessed in the fecal samples using quantitative PCR. Taking into consideration the six clinical variables and the quantity of the 10 different bacteria, 3 predictive models were established in the training set and test set, and evaluated using a confusion matrix, area under the receiver operating characteristic curve (AUC) values, sensitivity (recall), specificity, accuracy, positive predictive value and negative predictive value (npv). The abundance of Bacteroides, Eubacterium rectale and Roseburia inulinivorans was significantly lower in the T2DM group compared with the control group. However, the abundance of Enterococcus was significantly higher in the T2DM group compared with the control group. In addition, Faecalibacterium prausnitzii, Enterococcus and Roseburia inulinivorans were significantly associated with sex status while Bacteroides, Bifidobacterium, Enterococcus and Roseburia inulinivorans were significantly associated with older age. In the training set, among the three models, support vector machine (SVM) and XGboost models obtained AUC values of 0.72 and 0.70, respectively. In the test set, only SVM obtained an AUC value of 0.77, and the precision and specificity were both above 0.77, whereas the accuracy, recall and npv were above 0.60. Furthermore, Bifidobacterium, age and Roseburia inulinivorans played pivotal roles in the model. In conclusion, the SVM model exhibited the highest overall predictive power, thus the combined use of machine learning tools with gut microbiome profiling may be a promising approach for improving early prediction of T2DM in the near feature.