Special Issue:Digital Intelligence for Healthcare
The primary healthcare system is key to achieving a health equity. In China, great obstacles are challenged by imbalanced medical resources, shortage of primary healthcare providers, and the prevention and treatment of chronic diseases. Artificial intelligence large language models have demonstrated strong advantages in the medical system. This article deeply explored the application of large language models in the primary healthcare system and the challenges. The large language models are expected to assist the diagnosis and treatment of common diseases in grassroot medical institutions, promote intelligent health education and chronic disease management, underpin primary health services in the undeveloped and remote areas, stimulate the leapfrog development of general medicine, and accelerate the industrialization of large language models in general diagnosis and treatment and primary health services, thus providing important support for the construction of healthy China.
The creation of health records for Chinese residents is a key task for deepening the reform of the pharmaceutical and healthcare system, and an important action for promoting the equity of essential public health services. However, domestic studies on resident health records are mainly using the data from a city or community, and those using the national data from a demand-side perspective are rather scarce.
To understand the creation and utilization of health records in Chinese residents.
From November to December 2019, multistage sampling was used to select three provinces/municipality (Zhejiang, Shanxi and Chongqing) from eastern, central and western China (one was extracted from each geographical region) , then from each of them, one urban district and one county were extracted. Randomly selected 2 community health centers/stations, township health centers/village clinics in the corresponding districts (counties) . Finally, 20 community health service centers/township health centers were selected, the visitors of these institutions were invited to attend a questionnaire survey for understanding their information about the creation of health records, and the access to the health records, as well as satisfaction with the services. For ease of analysis, the visitors were classified into six categories (0-6-year-olds, pregnant women, over 65-year-olds, hypertensioners, diabetics, and general population) in accordance with the population groups defined in the Essential Public Health Service Programs.
Altogether, 10 067 residents were included for final analysis. Among them, 9 119 (90.58%) self-reported that they had received health records creation services. The rates of creation of health records in 0-6-year-olds, pregnant women, over 65-year-olds without hypertension/diabetes, over 65-year-olds with hypertension, under 65-year-olds with hypertension, over 65-year-olds with diabetes, and under 65-year-olds with diabetes, as well as general population were 94.09% (2 787/2 962) , 95.60% (956/1 000) , 87.87% (616/701) , 88.87% (1 414/1 591) , 92.91% (747/804) , 89.41% (895/1 001) , 92.72% (471/508) , and 82.20% (1 233/1 500) , respectively. Among those with health records created, 67.02% (5 990 / 8 938) could access to their health records at any time, and the health records accessed by most of them were printed〔75.76% (4 538/5 990) 〕. However, 12.40% (1 108/8 938) of residents reported that they had no access to their health records, and other 20.59% (1 840/8 938) indicated that they had never tried to gain access to their health records. The rate of satisfaction with health records services in residents was 83.31% (4 352/5 224) . The rate of health records creation and rate of accessing the health records differed significantly by province, district or country, household monthly income per person, education level, and category of population (P<0.05) . The rates of satisfaction with the creation of and access to health records differed significantly by province, type of visited health institution, district or country, household monthly income per person, education level, and category of population (P<0.05) .
Generally, the rate of creation of health records in Chinese residents has significantly increased. The rate of utilization of the records has also enhanced, but needs further improvement. Moreover, residentssatisfaction with health records services may be at a moderate level.
The number of chronic obstructive pulmonary disease (COPD) patients in China is huge, and respiratory rehabilitation training, as an important part of the management of COPD patients in the stabilization period, can effectively improve their lung function and quality of life, as well as reduce the burden on their families and society. Current data from Europe and the United States have shown that the implementation of respiratory rehabilitation under telemedicine management can improve the lung function and QOL of patients, however, there is a lack of relevant practice in China, especially in the west.
To assess the impact of respiratory rehabilitation training via telemedicine management in combination with conventional therapy on improving ventilatory capacity and lung function in elderly patients with moderate-to-severe COPD.
This study was a prospective randomized controlled study, enrolling consecutive COPD patients who attended the Fourth People's Hospital of Sichuan Province and five joint community clinics from June 2021 to June 2022. The included patients were randomly divided into the experimental group and control group by simple randomized grouping method using random number table. The control group received traditional long-term regular inhalation bronchodilator and oral medication, and the experimental group was guided by telemedicine on the basis of the treatment plan of the control group. A six-month study was conducted on two groups of patients, lung function, Borg score, 6MWT, and quality of life score (QOL score) were recorded at 1 month before and 1, 3, 6 months after intervention.
The study subjects were divided into 72 cases in the control group and 73 cases in the experimental group, and there was no significant difference in gender, age and lung function at baseline [the forced expiratory volume in one second/predicted value ratio (FEV1%pred) , and the ratio of the forced expiratory volume in one second to the forced vital capacity (FEV1/FVC) ] between the two groups (P>0.05) . There was an interaction between time and group for dyspnea and mood in FEV1%pred, FEV1/FVC, 6MWT level and QOL score (P<0.05) . After 1, 3, and 6 months of intervention, FEV1%pred, FEV1/FVC, 6MWT, Borg score, and QOL score of the experimental group were better than those of the control group (P<0.05) ; FEV1%pred, FEV1/FVC, Borg score, 6MWT, and QOL scores at 3 and 6 months post-intervention were better than those at 1 month post-intervention in the experimental group (P<0.05) .
The use of telemedicine technology for respiratory rehabilitation of elderly moderate-to-severe COPD patients in the stable stage can effectively improve the pulmonary function, quality of life and the quality of survival of this group of patients after 3, 6-months intervention.
China has the considerable disease burden of cervical cancer, with the mortality and morbidity of cervical cancer showing an increasing and younger trend. Facing to the critical situation of cervical cancer control, it is urgent to explore the new methods that suitable for different resource areas for the early detection and treatment of cervical cancer. Recently, great progress has been made in the field of AI image classification, and scientists have developed many algorithms to identify cervical lesions and conducted corresponding studies on their accuracy. Here, by reviewing the papers published at home and aboard, which studied the applications value of AI in cervical cytology screening, colposcopy examination, diagnosis and treatment of cervical cancer, we summarized and discussed the current progress and challenges for AI's application in the area of cervical cancer control, in order to provide solid evidence for the future use of AI in improving human health.
With the development of information technology, artificial intelligence shows great potentials for clinical diagnosis and treatment. Nevertheless, bias in algorithms derived by artificial intelligence can lead to problems such as unequal distribution of healthcare resources, which significantly affect patients' health equity. Algorithmic bias is a technical manifestation of human bias, whose formation strongly correlates with the entire development process of artificial intelligence, starting from data collection, model training and optimization to output application. Healthcare providers, as the key direct participants in ensuring patients' health, should take corresponding measures to prevent algorithmic bias to avoid its related health equity issues. It is important for healthcare providers to ensure the authenticity and unbiasedness of health data, optimize the fairness of artificial intelligence, and enhance the transparency of its output application. In addition, healthcare providers need to consider how to tackle bias-related health inequity, so as to comprehensively ensure patients' health equity. In this study, we reviewed the causes and coping strategies related to algorithmic bias in healthcare, with the aim of improving healthcare providers' awareness and ability to identify and address algorithmic bias, and laying a foundation for ensuring patients' health equity in the information age.
Twenty-six patients characteristics associated with thrombolytic effect were included for establishing models. The dimensionalities were reduced to two principal components by PCA, explaining 93.1% of the total variance. Comparison analysis revealed that the Wide&Deep model had the best predictive performance with an accuracy of 0.815, and an F-index of 0.871. Furthermore, the values of the area under the receiver operating characteristic (AUC) curve of the Wide&Deep model in predicting the thrombolytic effect in patients in the training set and test set were 0.753 and 0.793, respectively. The number of hidden layers and neurons in each layer of the model was 7 and 15, respectively. Using sigmoid as the activation function showed that the model parameters were optimal. The feature-engineering analysis of factors influencing the improvement of neurological function showed that the importance of medication type, administration mode and dosage ranked high, and the importance ranking in a descending order was: cerebrovascular disease history, type of medication, mode of administration, single dose, atherosclerosis, therapeutic time window of thrombolytic therapy, prevalence of use of anticoagulant drugs and drugs for promoting blood circulation and removing blood stasis. After simplifying the independent variables of the model, the accuracy of the Wide&Deep model was 0.819, and its accuracy was 0.801 suggested by the external verification after model simplification, indicating good predictive performance and generalizability.Conclusion The Wide&Deep model has proven to have excellent evaluation indicators. The importance of influencing factors of thrombolytic effect in a descending order is: cerebrovascular disease history, type of medication, administration mode, single dose, atherosclerosis, therapeutic time window of thrombolytic therapy, prevalence of use of anticoagulants and blood-activating and stasis-removing drugs. It provides clinicians with timely and effective thrombolysis treatment support involving thrombolysis related factors and individualized administration using AI-based algorithms.
The prevalence of age-related cognitive impairment, including dementia, has significantly increased with population aging. It has been shown that cognitive function is associated with gait status. Previously, researchers used statistical analysis methods instead of machine learning methods to study the gait of amnestic mild cognitive impairment (aMCI) and Alzheimer's disease (AD) .
To develop a model to identify aMCI and AD based on gait status using machine learning methods, explore gait markers differentiating between aMCI and AD, and to assess their possible values as aided tools in diagnosing aMCI and AD.
We recruited 102 cases from the Rehabilitation Hospital Affiliated to National Research Center for Rehabilitation Technical Aids, the First People's Hospital of Foshan, and Affiliated Zhongshan Hospital of Dalian University from December 2018 to December 2020, and included 98 of them according to the screening criteria, including 55 patients with aMCI, 10 patients with AD, and 33 healthy controls (HC) . The gait parameters of the participants were collected during performing single-task (free walking) , dual-task (counting backwards in sevens) and another dual-task (counting backwards from 100) using a wearable device. Random forest (RF) algorithm and gradient boosting decision tree (GBDT) algorithm were separately used to establish a model to compare the effect of two algorithms in recognizing three groups, with 10 gait parameters as predictive variables and the physical status (healthy, aMCI, AD) as response variables. Then important features were chosen using a machine learning algorithm combined with recursive feature elimination (RFE) .
No statistically significant differences were found among the three groups in terms of sex ratio, average age, height, body weight or shoe size (P>0.05) , while the differences in terms of average MMSE score and MoCA score were statistically significant (P<0.05) . In the free walking test, aMCI group and AD group had shorter average stride length and smaller average heel-to-ground angle (HtA) than HC group (P<0.05) . AD group had slower average gait speed and smaller average toe-off angle (ToA) than both HC group and aMCI group (P<0.05) . In performing the dual-task of counting backwards in sevens, compared with HC group, aMCI group and AD group had slower average gait speed and smaller average ToA and HtA (P<0.05) . AD group had longer average stance phase than HC group (P<0.05) . AD group had average smaller ToA than aMCI group (P<0.05) .In performing the dual-task of counting backwards from 100, AD group had slower average gait speed and smaller average HtA and ToA than both HC group and aMCI group (P<0.05) . Moreover, AD group had shorter average stride length than HC group (P<0.05) . The average HtA in aMCI group was smaller than that in HC group (P<0.05) . Using the GBDT-RFE method, we found important gait features in distinguishing between aMCI and AD to be the stride length, ToA and HtA, and the model using the RF algorithm performed better in identifying aMCI and AD, with an accuracy as high as 87.69%.
Stride length, ToA and HtA are important gait markers to identify aMCI and AD. These findings could help clinicians diagnose aMCI and AD in the future.
Scoliosis is a common abnormal curvature of the spine. Patients with mild scoliosis are usually treated with outpatient physiotherapy, but satisfactory efficacy is associated with appropriate treatment time and frequency. The efficacy of offline physiotherapy may be affected by limited medical resources and patients' treatment time and geographical location. Remote rehabilitation may save patients' treatment time and increase the geographical accessibility of physiotherapy, making the therapy more simple and convenient.
To explore the efficacy of remote rehabilitation combined with outpatient treatment in mild adolescent idiopathic scoliosis (AIS) .
Fifty-eight eligible mild AIS patients were selected from Department of Rehabilitation Medicine, Tianjin Hospital from September 2020 to September 2021, and divided into three groups according to patients and their parents' selection of treatment: online group (n=18), combined group (n=20) and offline group (n=20). The online group received WeChat- and Tencent Video-based physiotherapeutic scoliosis specific exercise (PSSE), the combined group received both outpatient and WeChat- and Tencent Video-based PSSE treatment, and the offline group received outpatient PSSE treatment. Data of three groups were collected, including the main curve Cobb angle, coronal balance (CB), thoracic kyphosis (TK) angle, lumbar lordosis (LL) angle, sagittal vertical axis (SVA), angle of axial trunk rotation (ATR), parietal vertebra rotation (Raimondi), pelvic incidence (PI), pelvic tilt (PT), sacral slope (SS), muscle activation rate (MAR) on both sides of paraspinal vertebrae, root mean square ratio (RMSR) of paraspinal muscles on both sides of paraspinal vertebrae, and the score of SRS-22 before and after treatment.
The main curve Cobb angle, TK, SVA, ATR, Raimondi, SS, MAR on paraspinal vertebrae, RMSR on the concave side of the parietal vertebra and SRS-22 self-image and mental health domain scores were significantly different from those before treatment in all groups (P<0.05). Specifically, the combined group was superior to the other two groups in improved ATR and treatment satisfaction. The combined group had significantly improved main curve Cobb angle after treatment than the online group. The improvement of the concave MAR in either the combined group or offline group was significantly better than that in the online group (P<0.05) .
In mild AIS patients, remote rehabilitation combined with outpatient treatment could effectively slow down the progression of AIS curve, improve sagittal abnormality of spine, abnormal posture and vertebral rotation, increase the activation rate of paraspinal muscles on the concave side of paraspinal vertebra and improve the balance of paraspinal muscles on both sides of paraspinal vertebrae. Moreover, the combined therapy also improved the quality of life.
Lung cancer ranks first in terms of incidence and mortality rates among cancers, with a 5-year survival rate of less than 20%. Many ways have been used to screen for early lung cancer, among which artificial intelligence (AI) has greatly improved the detection rate. However, how to use AI technologies to effectively manage atypical lung nodules to timely find early lung cancer, and to identify associated factors of lung nodule growth, which is an issue significantly associated with the guidance of clinical management of lung nodules.
To investigate the influencing factors of pulmonary nodules growth identified by AI-based follow-up and relevant clinical value.
A total of 175 patients with pulmonary nodules admitted to the Third Affiliated Hospital of Jinzhou Medical University in April 2019 were selected for a retrospective study. General clinical data, and AI-based analysis of imaging information related to pulmonary nodules was collected. The growth of pulmonary nodules〔solid nodules (in 82 cases) and ground-glass nodules (in 93 cases) classified by AI-based analysis〕 were observed by regular follow-ups. The influencing factors of pulmonary nodules growth were explored by Cox regression analysis.
Patients with solid nodules had higher prevalence of solid components, and mean CT quantitative parameters of nodules than those with ground-glass nodules (P<0.001) . Multivariate Cox regression analysis showed that average diameter〔HR=2.185, 95%CI (1.079, 4.425) , P=0.030〕, volume〔HR=1.001, 95%CI (1.000, 1.001) , P=0.022〕, malignant probability〔HR=2.232, 95%CI (1.036, 4.806) , P=0.040〕and surface signs〔HR=2.125, 95%CI (1.006, 4.489) , P=0.048〕 of the nodule were associated with solid nodular growth. The average diameter〔HR=2.458, 95%CI (1.053, 5.739) , P=0.038〕, volume〔HR=1.001, 95%CI (1.000, 1.002) , P=0.010〕, prevalence of solid components〔HR=1.022, 95%CI (1.002, 1.041) , P=0.030〕, malignant probability〔HR=2.386, 95%CI (1.174, 4.850) , P=0.016〕, surface signs〔HR=3.026, 95%CI (1.492, 6.136) , P=0.002〕, mean CT quantitative parameters〔HR=1.002, 95%CI (1.000, 1.003) , P=0.045〕 of the nodule were associated with the growth of ground-glass nodules.
The growth of pulmonary nodules was affected by many factors, such as original nodule size, mean CT quantitative parameters, presence of surface signs and malignant probability. It is suggested that clinicians determine the effective follow-up time based on the inflencing factors of pulmonary nodules growth identified by AI technologies, so as to detect the growth of pulmonary nodules as soon as possible and deliver treatment measures timely.
In recent years, artificial intelligence (AI) has shown rapid development in the medical field, and its application in diabetic retinopathy (DR) has been expanding.
To summarize the application of AI in DR through bibliometric analysis and elucidate the current status, hot spots and emerging trends of AI-related research in DR, with a view to providing ideas for future research.
The research was performed on the Web of Science database for the researches related to AI applications in DR from inception to 2022-11-04 and used CiteSpace software to conduct bibliometric analysis of the number of articles, countries, institutions, authors, co-citation and keywords in the literature.
A total of 1 770 papers were obtained, with an overall increasing trend in the number of publications and a peak of 402 papers in 2021. China was the top country in terms of the number of publications (440), and the UK was the country with the highest intermediary centrality (0.26). A total of 436 institutions were included in the institutional collaboration network mapping, represented by Sun Yat-sen University and Capital Medical University. A total of 601 authors were included in the author collaboration network mapping, represented by JIA Y L and HWANG T. Three highly cited authors, GULSHAN V, ABRàMOFF M D and TING D W, have made important contributions to the field. Ophthalmology, Invest Ophth Vis Sci and Ieee T Med Imaging are the three most influential journals in the field of AI applied to DR. The research hot spots were mainly focused on lesion segmentation and DR diagnosis. The future research trends may be efficacy prediction of diabetic macular edema as a complication of DR, disease management and improvement of AI algorithm performance.
Researchers can refer to the research hot spots and trends shown by this bibliometric analysis, focusing on AI in DR diagnosis, disease management and improvement of AI algorithm performance.
Using Machine Learning to Build an Early Warning Model for the Risk of Severe Airflow Limitation in Patients with Chronic Obstructive Pulmonary Disease
The degree of airflow limitation is a key indicator of the progression degree in COPD patients. However, problems such as contraindications to testing and compliance make it difficult for some patients to undergo the relevant tests and evaluate the severity of the disease.
To develop and evaluate a machine learning algorithm-based early warning model for the risk of severe airflow limitation in COPD patients.
A cross-sectional design was used to investigate COPD inpatients in a tertiary hospital in Sichuan Province from 2019-01 to 2020-06. General clinical indexes and pulmonary function test data were collected. The data were randomly divided into training and test sets in the ratio of 8∶2, and 216 risk warning models were constructed in the training set using four missing value filling methods, three feature screening methods, 17 machine learning and one integrated learning algorithm. The area under the ROC curve (AUC) , accuracy, precision, recall and F1 score were used to evaluate the predictive performance of the model; and the ten-fold cross-validation method and Bootstrapping were used for internal and external validation, respectively. The test set data was used for model testing and selection, the posterior method was used for sample size verification.
A total of 418 patients were included, of which 212 (50.7%) patients were at risk of severe airflow limitation. After four missing value treatments and three feature filters, a total of 12 processed datasets and the importance ranking of 12 factors affecting airflow limitation were obtained, and the results showed that modified medical research council dyspnea scale grade (mMRC) , age, body mass index (BMI) , smoking history (yes, no) , chronic obstructive pulmonary disease assessment test (CAT) score, and dyspnea (yes, no) were at the forefront inthe ranking of variable features and were key indicators for constructing the model, which had an important role in predicting the outcome. Using unfilled, Lasso screening, mMRC grade, smoking history (yes, no) , and dyspnea (yes, no) were the top 3 predictors, with mMRC grade accounting for 54.15% of feature importance. In which, using unfilled, Boruta screening, CAT score, age, and mMRC class were the top 3 predictors, and CAT score accounted for 26.64% of feature importance. A total of 216 prediction models were obtained using 17 machine learning algorithms and 1 integrated learning for each of the 12 datasets. 17 machine learning algorithms with 10-fold cross-validation showed that the differences were statistically significant (P<0.05) when comparing the prediction performance of different algorithms, and the average AUC of the stochastic gradient descent algorithm was maximum (0.738±0.089) . The results of external validation of the test set using the Bootstrapping algorithm showed that the differences were statistically significant (P<0.05) when comparing the prediction performance of the models obtained by different algorithms, and the average AUC of the integrated learning algorithm was maximum (0.757±0.057) . Evaluation of the prediction performance of four missing value treatments and three feature filters using the Bootstrapping algorithm showed that the performance of the model was improved when no padding and Lasso filtering were applied, with a statistically significant difference (P<0.05) . Using the test set data for 216 machine learning models, the best model had an AUC of 0.790 9, accuracy of 75.90%, precision of 75.00%, recall of 78.57%, and F1 value of 0.767 4. The sample size validation results suggested that the study sample size can meet the modeling needs.
In this study, a risk warning model for severe airflow limitation in COPD patients was developed and evaluated. mMRC class, age, BMI, CAT score, presence of smoking history and dyspnea were the key indicators affecting airflow limitation. The model has good predictive effect and has potential clinical application.
Osteoarthritis (OA) is a degenerative disease frequently encountered clinically, which can lead to loss of joint function in the late stage and is associated with a high disability rate. There is still no available cure for OA. Therefore, early diagnosis and precise treatment are the key to improving the therapeutic effect. Being an interdisciplinary research focus, artificial intelligence (AI) has been increasingly used in the diagnosis and treatment of OA recently, as it improves the diagnostic accuracy as well as clinical treatment and prognosis of OA. We summarized and systematically reviewed the literature on the application of AI in the diagnosis and treatment of OA, in which the application potential in assisting imaging diagnosis, surgical treatment, progression prediction and postoperative rehabilitation of OA was indicated, yet some limitations including non-standardized data collection and unstable algorithmic system were also identified. In the future, it is expected to establish a standardized clinical sample database and continuously optimize the algorithmic model, so as to better incorporate AI technologies in the diagnosis and treatment process of OA.
The past nearly 20-year period has seen a sudden increase in the use of artificial intelligence (AI) in esophageal cancer research, and an emergence of many systematic reviews and meta-analyses of the research. However, most of the reviews and meta-analyses only address a single aspect in summary, making it difficult for researchers to gain a comprehensive understanding of the latest developments and research hotspots in the field.
To perform a bibliometric analysis of the use of AI in esophageal cancer research, and the development, hotspots and emerging trend in this field.
All literature in English regarding esophageal cancer research using AI included in the Science Citation Index Expanded database of the Web of Science Core Collection was searched from 2000-01-01 to 2022-04-06. Microsoft Excel 2019, CiteSpace (5.8R3-64bit) and VOSviewer (1.6.18) were used to analyze the literature for annual number of publications, country, author, institution, co-citation and keywords.
Nine hundred and eighteen studies were retrieved, with a total of 23 490 times of being cited. The number of studies published between 2000 and 2016 grew slowly (from 6 to 40), but increased rapidly between 2017 and 2022 (from 62 to 216). Sixty countries, 118 institutions and 5 979 authors were involved in the studies. China (306 articles), the United States (238 articles) and the United Kingdom (113 articles) ranked the top three in terms of number of studies published. The top three institutions in terms of intensity of cooperation were University of Amsterdam (TLS=72), Catherine Hospital (TLS=64) and Eindhoven University of Technology (TLS=53). The top three authors in terms of number of publications were Jacques J G H M Bergman from the Netherlands (n=16), Tomohiro Tada from Japan (n=12), and Fons Van Der Sommen from the Netherlands (n=12). There were 39 962 co-cited authors and 42 992 co-cited studies. Thirty-three burst keywords were identified: the major burst keywords were p53 and mutations in 2001-2008 (early stage), and were esophageal cancer classification, new examination techniques (tomography), differentiation, identification and comparison between esophageal cancer and other cancers in 2013-2018 (middle stage), and were deep learning, convolutional neural network, and machine learning in esophageal cancer examination and diagnosis applications in 2019-2022 (late stage). Among which deep learning had the highest burst intensity (burst intensity of 13.89) .
AI application in esophageal cancer research has entered a new phase, moving gradually from genes and mutations toward accurate examination, diagnosis, and treatment. The latest major burst keywords in recent years (2019-2022) are deep learning, convolutional neural network, and machine learning in esophageal cancer examination and diagnosis. The future challenges to the use of AI in esophageal cancer research may include individual data collection, data quality assurance, data processing specifications, AI code reproduction, and reliability assurance of AI-assisted diagnostic decision-making.
Early intervention of blood glucose control in patients with new-onset type 2 diabetes mellitus (T2DM) can help delay the progression of diabetes. As a new form of health management, the effect of human-computer interaction intelligent blood glucose monitoring management on the progression of new-onset T2DM patients has not been clarified.
To explore the effect of human-computer interaction intelligent management on blood glucose control and self management capability in new-onset T2DM patients, so as to provide the reference for optimizing the control strategy in new-onset T2DM patients.
From June 2016 to December 2016, 200 patients with new-onset T2DM admitted to the Tianjin Medical University, Chu Hsien-I Memorial Hospital were selected by convenient sampling and randomly divided into the control group (n=100) and the monitoring group (n=100). The interventions in the monitoring group were the same as those in the control group except for the human-computer interaction intelligent monitoring. Blood glucose indexes〔fasting blood glucose (FBG), 2 h postprandial glucose (2 hPG) and glycated hemoglobin (HbA1c) 〕and self-management capability indexes〔diabetes management self-efficacy scale (DMSES), summary of diabetes self-care activities (SDSCA), diabetes self-care scale (2-DSCS) 〕were recorded at the time of enrollment and after 3 months of follow-up in the two groups.
After 3 months of follow-up, the monitoring group included 95 cases, the control group included 97 cases. Compared with the pre-intervention period, FBG, 2 hPG and HbA1c levels decreased in both groups after the intervention (P<0.05), and the scores of DMSES scores increased in both groups (P<0.05). FBG, 2 hPG and HbA1c were significantly lower in the post-intervention period of glucose monitoring group compared with the control group (P<0.05). 67 patients (70.5%) in the monitoring group reached the target level of FBG, 31 patients (32.0%) in the control group as well; besides 49 patients (51.6%) in the monitoring group reached the target level of 2 hPG, 30 patients (30.9%) in the control group as well; moreover, 67 patients (70.5%) in the monitoring group reached the target level of HbA1c, 29 cases (29.9%) in the control group as well, all the above rates of reaching in the monitoring group was higher than those in the control group (P<0.05). The total DMSES score, 2-DSCS score and SDSCA score in the monitoring group were higher than those in the control group (P<0.05). The score of DMSES in new-onset T2DM patients was positively correlated with the scores of 2-DSCS and SDSCA (rs=0.909, 0.872, P<0.01). The 2-DSCS scale score was positively correlated with the SDSCA scale score (rs=0.917, P<0.01). Multiple regression analysis showed that diet control, regular exercise, taking medication as instructed, blood glucose monitoring, prevention and management of high and low blood glucose behaviors were favorable factors for HbA1c reduction (P<0.05). The general diet, special diet and taking medication as instructed were the favorable factors for FBG and 2 hPG levels reduction (P<0.05), and the blood glucose monitoring was positive for 2 hPG levels reduction.
Human-computer interaction intelligent management was able to improve blood glucose control of new-onset T2DM patients effectively, which can promote the reaching to target blood glucose level, the subjective initiative of health behavior mainly through improving compliance of blood glucose monitoring, healthy diet, exercise and taking medication as instructed, which provide advice on effective intervention methods for new-onset T2DM patient management.
Telerehabilitation (TR) is an emerging model of rehabilitation service delivery based on communication technology, remote sensing and control technology, virtual reality technology and computer technology to to achieve cross-regional rehabilitation medical services. However, the effectiveness of TR in functional rehabilitation after stroke is still unclear, the methodological quality of related studies is uneven, and few researchers have systematically evaluated it.
To re-evaluate the systematic reviews/meta-analyses on the effectiveness of TR for functional rehabilitation after stroke.
In August 2021, PubMed, Web of science, the Cochrane Library, VIP, WanFang Data, CNKI and CBM were retrieved by computer for systematic reviews/meta-analyses on the effectiveness of TR applied to functional rehabilitation after stroke from the establishment of the database to August 2021. After the literature screen and data extract by two researchers independently, the methodological quality of the included literature was evaluated by AMSTAR 2 scale, and the evidence quality of the outcome index was graded by GRADE system. Descriptive analysis was used to analyze the effectiveness of TR in functional rehabilitation after stroke.
A total of 10 systematic reviews/meta-analyses were included, and the results of the AMSTAR 2 review showed that 2 systematic reviews were of high quality, 3 were of low quality, and 5 were of very low quality. The main reasons for the low methodological quality were the failure to report the preliminary study protocol, the list and reasons for excluded studies, the publication bias of the original study and the funding sources. The GRADE evidence quality assessment resulted in 10 systematic reviews addressing seven outcome measures, 41 bodies of evidence, with eight grade graded as intermediate, 23 grade graded as low, and 10 grade graded as very low. TR promoted the improvement of activities of daily living, motor function, quality of life, depressive symptoms and speech function of stroke patients to a certain extent, and had the same curative effect as face-to-face rehabilitation therapy or routine treatment, and even some TR rehabilitation effects were better than traditional rehabilitation therapy.
TR can promote the functional rehabilitation of stroke patients, but considering that the methodological quality and reliability of outcome measures of current systematic reviews/meta-analyses on the effectiveness of TR applied to functional recovery after stroke are mostly low, strict, standardized and comprehensive high-quality randomized controlled trials are still needed to provide evidence support; The results of this study can provide reference for the topic selection, research design and results report of future TR research.
Diet plays a critical role in the development, progression and prognosis of inflammatory bowel disease (IBD) . Given that specific nutritional guidelines are limited, nutritional management for patients with IBD remains challenging and fraught with uncertainty. Although previous studies have demonstrated that artificial intelligence (AI) shows promising applications in the nutritional management of patients with chronic diseases, research specifically focused on its application in the nutritional management of patients with IBD remains limited.
To conduct a scoping review of studies on AI in nutrition management of patients with IBD.
Following the methodology of scoping reviews, the databases of PubMed, Web of Science, Embase, Cochrane Library, CINAHL, IEEE Xplore, Association for Computing Machinery Digital Library, SinoMed, CNKI, Wanfang Data, and VIP were systematically searched from inception to March 2024 for studies on the application of AI in the nutritional management of patients with IBD. According to the established inclusion and exclusion criteria, two investigators independently screened the literature, and the basic characteristics of the selected studies were extracted.
A total of 15 studies were included. The applications of AI in this field include exploring the relationship between diet and IBD, assisting in nutritional assessment, and aiding nutritional interventions. The majority of utilization AI technologies in the included studies are machine learning, with some also employing additional techniques such as natural language processing and deep neural networks.
AI is beneficial for exploring healthy dietary patterns for patients with IBD and providing personalized nutritional guidance. However, its application in the field of nutritional management in patients with IBD is still in its infancy. Future efforts should focus on strengthening multidisciplinary collaboration, emphasizing the integration of clinical guidelines, and assessing the effectiveness of AI applications in clinical settings to enhance the rigor and accuracy of the results.
Research Progress of Machine Learning in Clinical Drug Therapy
With the advancement and development of concepts such as real-world research and precision treatment, the demand of researchers for medical big data processing keeps increasing. Because machine learning technology has unique advantages in processing massive, high-dimensional data and conducting predictive research, it has been deeply applied in the medical field in recent years. In addition to the application in disease diagnosis, image recognition and risk prediction, more and more studies have proved that machine learning can be applied to the decision support related research of clinical drug treatment. This article reviews the research progress of machine learning in clinical drug therapy.
The rapid development of new technologies such as artificial intelligence and large language models has brought new transformations to clinical medical practice. Both domestically and internationally, research and practical exploration of intelligent general practitioners have begun, but a consensus has yet to be formed. Against this backdrop, experts and scholars from Tsinghua University Vanke School of Public Health, Peking University School of Public Health, Chinese Association of General Practitioners of Chinese Medical Doctor Association and several other domestic institutions collaboratively developed a consensus. The background of these experts spans multiple disciplines, including general medicine, public health, artificial intelligence, and evidence-based medicine. Based on extensive literature review both domestically and internationally and through multiple rounds of expert discussions, the Chinese Expert Consensus on Artificial Intelligent General Practitioner (AIGP) was finally formulated. It includes 17 core consensus concerning the definition, characteristics, applications, challenges and recommendations of AIGP. This consensus aims to provide scientific references to promote the empowerment of general practitioners with intelligent technology and enhance the smart service level of primary healthcare.
The number of researches on the application of artificial intelligence (AI) to diagnosis and treatment of gastric cancer has been increasing in recent years, but no researcher has systematically analyzed it using bibliometric analysis.
To analyze the researches on the application of AI to diagnosis and treatment of gastric cancer, explore the research hotspots and development trends from 2003 to 2022. Methods On November 06, 2022, Web of Science (WOS) core collection database was searched by computer to obtain studies on the application of AI to gastric cancer diagnosis and treatment, and VOSviewer 1.6.18 software was used to visualize and analyze inter-country (region), inter-institution, and inter-author collaborations, co-cited authors, keyword co-occurrences and overlays through bibliometric analysis. CiteSpace 5.7.R5 software was used to perform institutional betweenness centrality analysis, journal biplot overlay, cluster analysis of co-cited literature for the last 6 years, co-cited literature clustering timeline graph analysis and reference bursting analysis. Excel 2019 software was used to plot bar graphs of the volume of publications and descriptive analysis tables of countries (regions), institutions, journals, authors, cited references and keywords.
A total of 703 papers were included, and the annual publication volume of the application of AI to gastric cancer diagnosis and treatment showed an overall increasing trend from 2003-2022, with a rapid increase after 2017 and the most rapid growth from 2019-2021. The top publishing country, institution and author was China, Chinese Academy of Sciences and TADA TOMOHIRO, respectively. The top three co-cited authors of BRAY FREDDIE, HIRASAWA TOSHIAKI and JIANG YUMING had made significant contributions to the field. Frontiers in Oncology was the journal with the highest publication volume, and Gastrointestinal Endoscopy was the most influential journal among the top ten journals for researches related to the application of AI to the diagnosis and treatment of gastric cancer. The citing journals mainly focused on the two fields of "Medicine, Medical, Clinical" and "Molecular, Biology, Immunology". And the cited journals mainly focused on the two fields of "Molecular, Biology, Genetics" and "Health, Nursing, Medicine". The top-ranked literature in terms of total citations titled Global cancer statistics 2018: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. All keywords were classified into 4 categories based on keyword clustering results, including AI-assisted biological research of gastric cancer, AI-assisted endoscopic diagnosis of gastric cancer, AI-assisted pathological diagnosis of gastric cancer, and AI-assisted non-endoscopic treatment and prognosis prediction of gastric cancer. Deep learning, convolutional neural network, imaging histology, gastrointestinal endoscopy, pathology and immunotherapy were the current research hotspots.
AI has a broad application prospect in gastric cancer diagnosis and treatment, and more and more scholars are devoted to AI in gastric cancer diagnosis and treatment. Currently, AI has been widely studied in the biology, diagnosis, staging, efficacy assessment and prognosis prediction of gastric cancer. The results of this study can provide a reference for scholars engaged in research work related to AI and gastric cancer.
The increasing prevalence of chronic diseases globally poses major challenges to the health of societies and individuals. Managing chronic diseases requires long-term treatment and monitoring, placing demands on patients' lifestyles. With the aging of the population and changes in lifestyle, chronic disease prevention and control are becoming more and more important. In recent years, as scientific and technological innovation in the field of healthcare develops in depth, and the application of artificial intelligence in healthcare has gradually become one of the important strategic directions of the country, the traditional method of chronic disease management relies too much on the offline communication between the doctor and the patient, which leads to the doctor not being able to maintain long-term and effective communication and follow up with the patient, and the patient may not be able to be detected and monitored by the doctor in a timely manner when his or her condition changes. In addition, the traditional chronic disease management approach is usually a generalized approach that fails to adequately consider the individual differences of each patient. Given the limitations of traditional chronic disease management methods, this study aims to provide more convenient and efficient primary care services using intelligent robots. Through personalized health management plans, assisted medical diagnosis, and timed medication reminders, the intelligent robot is committed to improving patients' quality of life, reducing the pressure on healthcare resources, and promoting the development of intelligent healthcare management globally.
Currently, the number of research papers on the application of artificial intelligence to the field of Alzheimer's disease (AD) has increased significantly. It is important to clarify the latest research hotspots and future development trends in this field.
To summarize the relevant research on the application of artificial intelligence to AD through bibliometric analysis, and clarify the research hotspots and trends from 2004 to 2023.
Literature on the application of artificial intelligence to AD from January 2004 to June 2023 was searched for in the Web of Science core database, and Microsoft Office Excel, CiteSpace, and VOSviewer software were used to visually analyze the number of publications, countries, authors, institutions, keywords, and co-citation networks of the literature.
Ultimately 3 189 articles were included. The number of literature on the application of artificial intelligence to AD has steadily increased since 2004 and has grown rapidly since 2015, with a maximum of over 600 articles. A total of 94 countries, 3 930 institutions, 13 563 authors, and 52 019 cited authors participated in this study. Among them, the United States and China were in a leading position in this field; Republic of Korea universities ranked first in terms of the number of publications; In addition, ZHANG DAOQIANG, LIU MINGXIA, SUK HEUNG-IL, and CLIFFORD R. JACK Jr were not only prolific authors but also the authors with the most citations. The visualization analysis of keywords and literature citations revealed that regarding the application of artificial intelligence to AD, the diagnosis and disease course classification of AD, as well as the prediction of its risk factors, are current research hotspots and that task analysis are future research trends.
The application of artificial intelligence to AD has attracted widespread attention from researchers worldwide. The diagnosis and classification of AD, as well as the prediction of its risk factors, are current research hotspots. Developing adjunctive drugs in task analysis, personalized treatment and care, and improving the algorithm performance of artificial intelligence may be research trends in the future.
Application of Artificial Intelligence Technologies in a Cloud-based Platform for ECG Analysis to Support the Diagnosis of a Critical Electrocardiography in Primary Care
The cloud-based platform for electrocardiography (ECG) analysis plays a supporting role in the prevention and treatment of cardiovascular diseases. During the construction of a cloud-based platform for ECG analysis, problems that should be focused and addressed are exploring ways to better use artificial intelligence (AI) technologies supporting ECG analysis, and improving the process and effectiveness of AI-aided diagnosis of a critical ECG.
To explore the use of AI technologies in a cloud-based platform for ECG analysis to support the diagnosis of a critical ECG in primary care.
The 12-lead resting ECGs (n=20 808) uploaded to Nalong Cloud-based ECG Analysis Platform by primary healthcare institutions were selected from June 2019 to June 2021. After being interpreted by AI-based algorithms and physicians, respectively, ECG findings were classified into critical group (critical ECGs) , normal group (normal ECGs) , and positive group (abnormal but not critical ECGs) . The results interpreted by the AI-based algorithm were compared with those interpreted by physicians (defined as the gold standard) to assess the diagnostic agreement and coincidence rate between AI-based and physician-based interpretations, and to assess the diagnostic sensitivity, and positive predictive value of AI-based interpretation. And the mean time for making diagnoses of three groups of ECGs was calculated.
By the AI-based interpretation, 619, 15 634 and 45 55 ECGs were included in the critical, positive, and normal groups, respectively. And by the physician-based interpretation, 619, 15 759 and 4 430 ECGs were included in the critical, positive, and normal groups, respectively. There was high agreement between AI-based and physician-based interpretation results of ECGs〔Kappa=0.984, 95%CI (0.982, 0.987) , P<0.001〕, with a diagnostic coincidence rate of 99.4%. The diagnostic sensitivity and positive predictive value of AI-based interpretation for ECGs was 99.4%, and 100.0%, respectively. The mean time for making diagnoses of critical ECGs, abnormal but not critical ECGs, and normal ECGs was statistically different (P<0.001) , the mean time of critical critical ECGs was shorter than normal ECGs and abnormal but not critical ECGs (P<0.001) .
AI technologies used in a cloud-based platform for ECG analysis could provide physicians with support for interpreting ECGs, which may contribute to improving the interpretation accuracy, optimizing the diagnostic process, shortening the time for diagnosing a critical ECG, and the treating of critical patients in primary care.
Chromosomal abnormalities are one of the common causes of birth defects, and karyotype analysis is still an important method for prenatal diagnosis of chromosomal abnormalities as well as an effective way to prevent and control birth defects. However, karyotype analysis, especially chromosomal image segmentation and classification mainly depends on manual work at present, which is laborious and time-consuming. As an emerging approach to karyotype analysis, it is of great significance to investigate the application value of artificial intelligence (AI) in prenatal chromosomal karyotype diagnosis.
To investigate the application effect and clinical value of AI in prenatal karyotype diagnosis.
A total of 1 000 pregnant women who received interventional prenatal diagnosis and karyotype analysis of amniotic fluid cells in the department of medical genetics and prenatal diagnosis of Wuxi Maternity and Child Health Care Hospital between 2020 and 2022 were selected as the study subjects. The karyotype analysis of all cases was performed using two-line mode, the results of the AI reading were reviewed by one geneticist in the first line, and another geneticist analyzed the karyotypes by Ikaros karyotype analysis workstation in the second line, the diagnostic results and time were recorded respectively. The final diagnosis of the samples were based on the manual review of the first line and the manual reading of the second line.
Among the 1 000 amniotic fluid samples, 735 cases were diagnosed as normal karyotype, 233 cases as aneuploidy, 0 case as structural abnormality and 32 cases as mosaicism by AI. The numbers of normal karyotype, aneuploidy, structural abnormality and mosaicism assessed by AI-assisted geneticist were 689, 233, 45 and 33, which were completely consistent with those evaluated by geneticist using Ikaros system. Compared with AI-assisted geneticist, AI-based diagnosis had strong consistency (Kappa=0.895, 95%CI=0.866-0.924, P<0.01). The diagnostic accuracy, sensitivity and positive predictive value of AI-based diagnosis was 95.4%, 95.4% and 100.0%, respectively, among which the normal karyotype, aneuploidy, structural abnormality and mosaicism were detected with a sensitivity of 100.0%, 100.0%, 0 and 97.0%, and the positive predictive value of 100.0%, 100.0%, 0 and 100.0%. The average diagnostic time of AI was shorter than that of AI-assisted geneticist and Ikaros-assisted geneticist (P<0.001), and AI-assisted geneticist took less time on average to diagnose than the Ikaros-assisted geneticist (P<0.001) .
AI-assisted karyotype analysis of amniotic fluid cells has a high degree of automation, but its ability to recognize chromosomal structural abnormalities needs to be improved. It is suggested that AI be combined with the geneticist for karyotype analysis in clinical application to ensure the quality of prenatal diagnosis and improve efficiency.
The research related to mHealth technology in chronic disease management has developed rapidly in recent years, however, the research trends, hotspots and cutting-edge issues in this field remain unclear.
To systematically review the application and development of mHealth technology in chronic disease management and provide reference for future research.
Using Web of Science Core Collection and PubMed as the source of literature data, the relevant literature was searched from 1997 to 2022 by CiteSpace 6.1.R 3 software on October 18, 2022, restricting the language to English, and excluding conference papers, conference abstracts, online publications, editorials, letters, book chapters, news, and other non-compliant contents. National regions, disciplinary intersections and keywords were analyzed to grasp the current status and hotspots of related research internationally, and the cutting-edge issues and research trends of mHealth technology in chronic disease management were comprehensively analyzed using keywords clustering analysis, keywords bursting analysis and timeline views.
A total of 7 622 papers were finally included in the study, with a significant growth trend in the volume of publications starting from 2011, in which the United States contributed the most with a total of 2 645 (34.70%). The journals in which the papers were published were mainly in the fields of medicine, psychology and health; and the top five high-frequency keywords were chronic disease (711 times), nursing (695 times), management (544 times), intervention (502 times) and health (448 times). A total of 10 meaningful clusters were formed, which can be categorized into 4 dimensions of research tools, research theories and methods, research objects, and research factors; combining with keywords bursting and timeline view, the hot issues mainly focus on telemedicine, telecare, and digital health.
The international research fervor for the application of mHealth technology in chronic disease management has continued, and the field of research has shifted from medicine to health science, with the focus on intervention research on chronic diseases through mHealth technology and the use of digital technology to provide integrated telehealth services for chronic diseases. It is suggested that our scholars should pay attention to the application of mHealth and digital technologies in chronic disease management, find high-quality health services for patients with chronic diseases in China through intervention studies, and provide strategies and suggestions for the high-quality development of chronic disease services and management in China.
Application of Scatter Diagram in Prehospital Screening for Arrhythmia Using Single Lead，Wearable Remote ECG Monitoring System
Arrhythmia is a common cardiovascular disease, which has a range of transient or paroxysmal conditions. Arrhythmia easily occurs outside of the hospital, but signals of its onset often could not be captured by traditional ECG devices since they can not be worn at any time.
To assess the effect of applying scatter diagram in prehospital screening for arrhythmia via analyzing patients' data monitored by the single lead, wearable remote ECG monitoring system.
Participants (n=1 076) were primary care patients who were selected from Yinchuan from September 2018 to September 2019. All of them used single lead, wearable remote ECG monitoring system to monitor cardiac rhythms prehospitally when they had palpitation, dizziness, chest tightness, shortness of breath and other symptoms, and real-timely uploaded 24-hour ambulatory ECG data to be used for screen for arrhythmia by different approaches: approach A (diagnosis made using scatter diagram analysis by primary care physicians) , approach B (diagnosis made using scatter diagram analysis by physicians from Remote ECG Center, the First People's Hospital of Yinchuan) , and approach C (diagnosis made using scatter diagram analysis and ECG analysis by physicians from Remote ECG Center, the First People's Hospital of Yinchuan) . Prevalence and types of arrhythmia detected by these approaches and diagnostic coincidence rate of these approaches were analyzed. The sensitivity, specificity, positive and negative predictive values of approaches A and B were assessed with those of approach C as the gold standard.
(1) The frequencies of arrhythmias detected by approaches A, B and C were 1 301, 1 323, and 1 647, respectively. The types of arrhythmias detected by approaches A, B and C were 14, 14, and 15, respectively. And the prevalence of arrhythmias detected by approaches A, B and C were 80.9%, 81.2% and 87.5%, respectively. (2) The diagnoses made by approaches A and B were highly consistent〔Kappa=0.891, 95%CI (0. 711, 1.071) , P=0.617〕, and the diagnostic coincidence rate was 96.7%. The diagnoses made by approaches B and C were highly consistent〔Kappa=0.759, 95%CI (0.489, 1.029) , P<0.001〕, and the diagnostic coincidence rate was 93.6%. The diagnoses made by approach A were relatively consistent with those by approach C〔Kappa=0.692, 95%CI (0.392, 0.992) , P<0.001〕, and the diagnostic coincidence rate was 91.7%. (3) The sensitivity, specificity, positive and negative predictive values of approach A in diagnosing arrhythmia were 91.5%, 93.3%, 99.0% and 61.2%, and those of approach B were 92.8%, 99.3%, 99.9% and 66.3%.
Using scatter diagram in prehospital screening for arrhythmia through analyzing the monitoring results of single lead, wearable remote ECG monitoring system will contribute to the development of arrhythmia diagnosis and treatment in primary care, and the establishment of an arrhythmia prevention and treatment network with the participation of residents, primary care physicians and remote ECG center physicians.
In China, the overall prevalence and incidence of cardiovascular disease (CVD) continues to increase, and the mortality rate from CVD in rural areas has exceeded that in urban areas recently. Remote ECG-based screening for CVD risk is a beneficial supplement for CVD risk screening in primary hospitals, but there are many difficulties during its implementation, which mainly include the following aspects: how to improve the awareness and credibility of remote ECG-based screening for CVD risk and sense of gain in residents? How to incentivize primary physicians to actively participate in the screening? How to improve insufficient management ability and experience of primary physicians who can only provide single screening and communication services? How to build a collaborative mechanism between primary and higher level hospitals involved in delivering referral services, and to provide continuous services by establishing multiple teams consisting of screening team, diagnosis team, evaluation team, treatment team and follow-up management team? To address these issues, we invited a group of experts to attend discussions, in which the following recommended solutions were put forward: using various resources rationally and efficiently; strengthening the division of labor and cooperation between team members to improve hierarchical diagnosis and treatment; giving full play to the capacities of nursing and public health teams to develop different screening programmes; strengthening the technical support of experts from higher level medical institutions for primary doctors, and increasing the social benefits of primary hospitals; carrying out workplace training to improve the professional level of primary care workers; integrating Internet technologies into primary care to enable referrals; building a big data database of cases; constructing medical and health groups with clear defined division of labor and cooperation.
With the prosperous development of Internet medical care and increasing flow of electronic prescriptions, electronic prescription review has become an important guarantee for rational drug use in the online medication environment. However, in the Internet scenario, inefficient manual review with low quality increases the risk of adverse drug events. The application of the intelligent auxiliary prescription review system can significantly reduce the working pressure of pharmacists reviewing prescriptions and improve the efficiency of review, but there is a lack of setting standards and standardized management measures in the system architecture, system functions and prescription review rules setting of the Internet medical intelligent auxiliary prescription review system at present, which cannot meet the rapidly developing needs of Internet medical care. Using the construction experience of the prescription review system in medical institutions as a reference, this expert consensus makes recommendations on the construction and application of the Internet medical intelligent auxiliary prescription review system based on the functions and methods of formulating prescription review rules of the existing prescription review system, to further promote the standardization of the Internet medical prescription review work and ensure rational drug use.
The Severe Acute Respiratory Syndrome Coronavirus 2 Omicron variant (SARS-CoV-2, Omicron) has been widely spread around the world. Since February 2022, Shenzhen was continuously affected by it as a major hub connecting domestic and international transportation, resulting in rapidly increasing number of infected cases.
To construct a modified susceptible-exposed-infected-recovered (SEIR) model for providing policy references and suggestions with applied value for epidemic prevention and control in Shenzhen, China, so as to alleviate the pressure of prevention and control.
This study developed a modified SEIR model targeting the epidemiological characteristics of the Omicron variant such as rapid transmission, high concealment, and general susceptibility of the population, introducing groups with policy characteristics as close contacts, secondary contacts, quarantined individuals and carriers, based on traditional SEIR model of infectious disease dynamics. The relevant parameters of the modified model were determined by fitting the Shenzhen epidemic data of February 18 to 28, 2022.
The predicted data of the model were basically consistent with the actual data from March 01 to 04, 2022, providing a reliable basis for predicting the subsequent development of the epidemic. Subsequently, the Omicron variant outbreak in Shenzhen between 5 to 19 March 2022 was forecasted through this modified model to provide guidance for epidemic prevention and control measures in terms of the degree and time of manual intervention in epidemic prevention and control, and healthcare resource requirements such as patient beds and isolation rooms.
The modified SEIR model developed in this study has proved its practical value in forecasting epidemic development, formulating and adjusting epidemic control measures, and allocating health resources.
Obstructive sleep apnea (OSA) is a high prevalent chronic disease that may lead to many complications, and cause great potential harm to health. Epidemiological studies have showed that OSA is closely related to the development of various cardiovascular diseases. There are about 66 million patients with moderate to severe OSA in China, but 80% of potential OSA patients have not been diagnosed and treated in time. OSA is mainly diagnosed and treated in a hospital-based sleep center currently, as the process is time-consuming and laborious, which may be lead to a delay in diagnosis and treatment of many patients. Supported by the development of Internet of Things, Internet technologies and other emerging technologies, remote medicine has been increasingly used in the diagnosis and management of chronic diseases owing to its advantages of easy access, interactivity, high efficiency, resource sharing, service continuity and without space-time constraints. Our center has initially built a management system for remote diagnosis and treatment of OSA, but its clinical efficacy and economic value need to be further verified. We designed a randomized controlled trial protocol to assess whether the clinical benefits of the low-cost remote healthcare model are similar to those of the traditional healthcare model by comparing them in terms of clinical efficacy and health economic benefits, hoping to provide a reference for the efficient use of medical resources and further promotion of remote diagnosis and treatment of chronic diseases.
Artificial Intelligence (AI) is an emerging technology to improve healthcare services. With the joint promotion of government agencies and academic departments around the world, a large number of studies have demonstrated that AI can improve the diagnosis, treatment and prevention of cardiovascular disease. However, there are still some limitations in its development and application, and it has not yet been widely used in clinical practice. Based on this, the American Heart Association (AHA) published the Use of Artificial Intelligence in Improving Outcomes in Heart Disease: a Scientific Statement from the American Heart Association in Circulation on April 2, 2024. This statement reviews the research progress of AI in the diagnosis, classification and treatment of cardiovascular disease, puts forward the existing problems and potential solutions, and builds a framework for the future application of AI in the cardiovascular disease. This article aims to interpret the statement for providing advice and direction for the application and research of AI in cardiovascular disease in China.
With the aging population, the incidence of pressure injury (PI) is gradually increasing. This not only severely impacts the quality of life for patients but also increases healthcare expenditures. However, the early detection and accurate staging of PI heavily depend on specialized training.
To construct and validate an artificial intelligence model for the automatic detection and staging of PI aimed at enhancing the real-time nature, accuracy, and objectivity of PI diagnostics.
A total of 693 PI images from the electronic management system of pressure ulcers at Changshu No.1 People's Hospital were selected from January 2021 to February 2024, the images were randomly divided into a training set (551 images) and a test set (142 images), and categorized into six stages according to National Pressure Ulcer Advisory Panel (NPUAP) guidelines: StageⅠ (154 images), StageⅡ (188 images), StageⅢ (160 images), StageⅣ (82 images), deep tissue injury (57 images), and unstageable (52 images). A deep learning object detection model for PI was established using five different versions of the YOLOv8 [nano (n), small (s), medium (m), large (l) and extra large (x) ] neural network and transfer learning. The model evaluation metrics included accuracy, sensitivity, specificity, false positive rate, and detection speed. Finally, the model was deployed to a mobile application via the Ultralytics Hub platform, facilitating the application of the AI model in clinical practice.
During the evaluation of a test set containing 142 PI images, the YOLOv8l version demonstrated high accuracy (0.827) and fast inference speed (68.49 fps), achieving the best balance between precision and speed among the YOLO versions. Specifically, it achieved an overall accuracy of 93.18% across all categories, a sensitivity of 76.52%, a specificity of 96.29%, and a false positive rate of 3.72%. Among the six stages of PI, the model achieved the highest accuracy for StageⅠat 95.97%. The accuracies for StageⅡ, StageⅢ, StageⅣ, deep tissue injury, and unstageable were 91.28%, 91.28%, 91.95%, 95.30%, and 93.29%, respectively. In terms of processing speed, YOLOv8l took a total of 2.07 seconds to process 142 images, averaging 68.49 PI images per second.
The AI model based on the YOLOv8l network can quickly and accurately detect and stage PI. Deploying this model to a mobile app allows for portable use in clinical practice, demonstrating significant potential for clinical application.
With the wide application of artificial intelligence (AI) in the medical field, more and more AI-based clinical decision support systems have been applied in the clinical diagnosis, screening, and other fields. Early-stage clinical evaluation is important for evaluating the clinical performance, safety, and human factors of AI-based clinical decision support systems, and laying the foundation for large-scale trials. However, the transparency and integrity of the clinical reports need to be improved. The Developmental and Exploratory Clinical Investigations of DEcision Support Systems Driven by Artificial Intelligence (DECIDE-AI) was officially published online in May 2022. Based on this guideline and relative literature, this paper explores the transparent reporting of early-stage clinical evaluation of AI-based clinical decision support systems, in order to help developers and researchers better understand and apply the relevant guidelines, and improve the reporting transparency of early-stage clinical evaluation of AI-based clinical decision support systems.
The diagnosis of Parkinson disease (PD) is mainly based on clinical symptoms, and there is a lack of objective methods for correct diagnosis. At present, there have been studies on retinal structural changes as a biomark for early diagnosis of PD, but machine learning based on retinal structural changes for predicting early PD has not yet been studied.
To construct a machine learning model based on the characteristics of retinal structural changes, explore its value in early PD diagnosis, and the accuracy of different machine learning algorithms for early PD diagnosis.
From October 2021 to September 2022, 49 PD patients aged 40 to 70 years old (PD group) who attended outpatient clinics and were hospitalized in the department of neurology of Henan Provincial People's Hospital (PD group) and 39 healthy people with matching age and sex (healthy control group) who came to the hospital for physical examination were collected. All study subjects underwent swept-source optical coherence tomography and swept-source optical coherence tomography angiography, the thickness and vessel density of the macular retina were also quantitatively analyzed. The 88 subjects were randomly divided into the 62 training sets and 26 validation set according to the ratio of 7∶3. Variables with significant differences between the PD group and healthy control group were selected as the characteristic variables for inclusion in the machine learning model, and Logistic regression (LR) , K-nearest neighbor algorithm (KNN) , decision tree (DT) , random forest (RF) and extreme gradient boosting (XGboost) models were constructed in the training set. The area under the curve (AUC) , accuracy, sensitivity and specificity of the receiver operating characteristic (ROC) curve were used to evaluate the predictive value of the machine learning model based on retinal structural changes for early PD.
Compared with the healthy control group, the density of the upper outer ring (A6) , the outer temporal outer ring (A7) , the lower outer ring (A8) and the outer nasal ring (A9) of the superficial capillaries in the PD group were reduced, the thickness of the upper inner ring (A2) , the inner temporal inner ring (A3) , the inferior inner ring (A4) , the inner ring of the nasal side (A5) of the retinal layer, A6, A7, A8 and A9, the thickness of A6 of the ganglion cell complex layer, the thickness of A7 of the nerve fiber layer, A2 and A4, A5, A6, A7, A8, A9 became thinner (P<0.05) . The reductions in A2 thickness of the retinal layer (OR=0.781, 95%CI=0.659-0.926) , A3 thickness of the retinal layer (OR=1.190, 95%CI=1.019-1.390) , A2 thickness of the outer retina (OR=0.748, 95%CI=0.603-0.929) , A6 thickness of the outer retina (OR=2.264, 95%CI=1.469-3.490) , A8 thickness of the outer retina (OR=0.723, 95%CI=0.576-0.906) , and A7 thickness of the nerve fiber layer (OR=0.592, 95%CI=0.454-0.773) , and the decrease in A7 density of the superficial capillaries (OR=1.966, 95%CI=1.399-2.765) were independent risk factors for the occurrence of early PD (P<0.05) . The above variables were involved to construct the machine learning model, the results showed that among the five models constructed, the LR model had the highest overall performance, with an AUC of 0.841, while the DT model has the highest accuracy at 0.846.
Machine learning model based on retinal features can accurately predict early PD, among which the DT model has high accuracy for early PD diagnosis.
