Explainable El Niño predictability from climate mode interactions.

The El Niño-Southern Oscillation (ENSO) provides most of the global seasonal climate forecast skill1-3, yet, quantifying the sources of skilful predictions is a long-standing challenge4-7. Different sources of predictability affect ENSO evolution, leading to distinct global effects. Artificial intelligence forecasts offer promising advancements but linking their skill to specific physical processes is not yet possible8-10, limiting our understanding of the dynamics underpinning the advancements. Here we show that an extended nonlinear recharge oscillator (XRO) model shows skilful ENSO forecasts at lead times up to 16-18 months, better than global climate models and comparable to the most skilful artificial intelligence forecasts. The XRO parsimoniously incorporates the core ENSO dynamics and ENSO's seasonally modulated interactions with other modes of variability in the global oceans. The intrinsic enhancement of ENSO's long-range forecast skill is traceable to the initial conditions of other climate modes by means of their memory and interactions with ENSO and is quantifiable in terms of these modes' contributions to ENSO amplitude. Reforecasts using the XRO trained on climate model output show that reduced biases in both model ENSO dynamics and in climate mode interactions can lead to more skilful ENSO forecasts. The XRO framework's holistic treatment of ENSO's global multi-timescale interactions highlights promising targets for improving ENSO simulations and forecasts.

Northwestern Pacific Oceanic circulation shaped by ENSO.

The intricate currents of the Northwest Pacific Ocean, with strong manifestations along the westside rim, connect tropical and subtropical gyres and significantly influence East Asian and global climates. The El Niño/Southern Oscillation (ENSO) originates in the tropical Pacific Ocean and disrupts this ocean circulation system. However, the spatiotemporal dependence of the impact of ENSO events has yet to be elucidated because of the complexities of both ENSO events and circulation systems, as well as the increased availability of observational data. We thus combined altimeter and drifter observations to demonstrate the distinct tropical and subtropical influences of the circulation system on ENSO diversity. During El Niño years, the North Equatorial Current, North Equatorial Countercurrent, Mindanao Current, Indonesian Throughflow, and the subtropical Kuroshio Current and its Extension region exhibit strengthening, while the tropical Kuroshio Current weakens. The tropical impact is characterized by sea level changes in the warm pool, whereas the subtropical influence is driven by variations in the wind stress curl. The tropical and subtropical influences are amplified during the Centra Pacific El Niño years compared to the Eastern Pacific El Niño years. As the globe warms, these impacts are anticipated to intensify. Thus, strengthening observation systems and refining climate models are essential for understanding and projecting the enhancing influences of ENSO on the Northwest Pacific Oceanic circulation.

Ocean internal tides suppress tropical cyclones in the South China Sea.

Tropical Cyclones (TCs) are devastating natural disasters. Analyzing four decades of global TC data, here we find that among all global TC-active basins, the South China Sea (SCS) stands out as particularly difficult ocean for TCs to intensify, despite favorable atmosphere and ocean conditions. Over the SCS, TC intensification rate and its probability for a rapid intensification (intensification by ≥ 15.4 m s-1 day-1) are only 1/2 and 1/3, respectively, of those for the rest of the world ocean. Originating from complex interplays between astronomic tides and the SCS topography, gigantic ocean internal tides interact with TC-generated oceanic near-inertial waves and induce a strong ocean cooling effect, suppressing the TC intensification. Inclusion of this interaction between internal tides and TC in operational weather prediction systems is expected to improve forecast of TC intensity in the SCS and in other regions where strong internal tides are present.

[Expression of MicroRNA-3162-3p in Different Clinical Stages of Children with Primary Immune Thrombocytopenia and Its Significance].

OBJECTIVE: To explore the expression of microRNA-3162-3p in different clinical stages of childhood primary immune thrombocytopenia (ITP) and its significance. METHODS: Ninety-six children with ITP were enrolled and divided into new diagnosis group (n=40), persistent group (n=30) and chronic group (n=26) according to the course of disease. 80 healthy children were selected as the control group. Peripheral blood mononuclear cells (PBMNC) of ITP children and healthy children were isolated and cultured, and the expression of microRNA-3162-3p in PBMNC of subjects was detected by real-time fluorescence quantitative PCR. The contents of IL-17, IL-23, IL-10 and TGF-ß in PBMNC of subjects were determined by ELISA. The correlation between microRNA-3162-3p and platelet count, IL-17, IL-23, IL-10 and TGF-ß was analyzed. RESULTS: Compared with the control group, the expression of microRNA-3162-3p and IL-10 in PBMNC and platelet count of ITP children were significantly decreased（P < 0.05）, while IL-17, IL-23 and TGF-ß were significantly increased (P < 0.05). With the prolongation of the disease course, the expressions of microRNA-3162-3p and IL-10 in PBMNC and platelet count were significantly decreased（P < 0.05）, while the expressions of IL-17, IL-23 and TGF-ß were significantly increased (P < 0.05). The expression of microRNA-3162-3p in PBMNC was positively correlated with platelet count and IL-10 (r =0.716, 0.667), and negatively correlated with IL-17, IL-23, and TGF-ß (r =-0.540, -0.641, -0.560). CONCLUSION: MicroRNA-3162-3p expression is significantly reduced in PBMNC of children with ITP, and is involved in the regulation of Th17/Treg imbalance, which can be used as a potential therapeutic target of ITP.

Hysteresis of the El Niño-Southern Oscillation to CO2 forcing.

El Niño-Southern Oscillation (ENSO) is the strongest interannual climate variability with far-reaching socioeconomic consequences. Many studies have investigated ENSO-projected changes under future greenhouse warming, but its responses to plausible mitigation behaviors remain unknown. We show that ENSO sea surface temperature (SST) variability and associated global teleconnection patterns exhibit strong hysteretic responses to carbon dioxide (CO2) reduction based on the 28-member ensemble simulations of the CESM1.2 model under an idealized CO2 ramp-up and ramp-down scenario. There is a substantial increase in the ensemble-averaged eastern Pacific SST anomaly variance during the ramp-down period compared to the ramp-up period. Such ENSO hysteresis is mainly attributed to the hysteretic response of the tropical Pacific Intertropical Convergence Zone meridional position to CO2 removal and is further supported by several selected single-member Coupled Model Intercomparison Project Phase 6 (CMIP6) model simulations. The presence of ENSO hysteresis leads to its amplified and prolonged impact in a warming climate, depending on the details of future mitigation pathways.

Challenges and Solutions in Implementing eSource Technology for Real-World Studies in China: Qualitative Study Among Different Stakeholders.

BACKGROUND: eSources consist of data that were initially documented in an electronic structure. Typically, an eSource encompasses the direct acquisition, compilation, and retention of electronic information (such as electronic health records [EHRs] or wearable devices), which serves to streamline clinical research. eSources have the potential to enhance the accuracy of data, promote patient safety, and minimize expenses associated with clinical trials. An opinion study published in September 2020 by TransCelerate outlined a road map for the future application of eSource technology and identified 5 key areas of challenges. The background of this study concerns the use of eSource technology in clinical research. OBJECTIVE: The aim of this study was to present challenges and possible solutions for the implementation of eSource technology in real-world studies by summarizing team experiences and lessons learned from an eSource record (ESR) project. METHODS: After initially developing a simple prototype of the ESR software that can be demonstrated systematically, the researchers conducted in-depth interviews and interacted with different stakeholders to obtain guidance and suggestions. The researchers selected 5 different roles for interviewees: regulatory authorities, pharmaceutical company representatives, hospital information department employees, medical system providers, and clinicians. RESULTS: After screening all consultants, the researchers concluded that there were 25 representative consultants. The hospital information department needs to implement many demands from various stakeholders, which makes the existing EHR system unable to meet all the demands of eSources. The emergence of an ESR is intended to divert the burden of the hospital information department from the enormous functional requirements of the outdated EHR system. The entire research process emphasizes multidisciplinary and multibackground expert opinions and considers the complexity of the knowledge backgrounds of personnel involved in the chain of clinical source data collection, processing, quality control, and application in real-world scenarios. To increase the readability of the results, the researchers classified the main results in accordance with the paragraph titles in "Use of Electronic Health Record Data in Clinical Investigations," a guide released by the US Food and Drug Administration. CONCLUSIONS: This study introduces the requirement dependencies of different stakeholders and the challenges and recommendations for designing ESR software when implementing eSource technology in China. Experiences based on ESR projects will provide new insights into the disciplines that advance the eSource research field. Future studies should engage patients directly to understand their experiences, concerns, and preferences regarding the implementation of eSource technology. Moreover, involving additional stakeholders, including community health care providers and social workers, will provide valuable insights into the challenges and potential solutions across various health care settings.

Influence of El Niño on the variability of global shoreline position.

Coastal zones are fragile and complex dynamical systems that are increasingly under threat from the combined effects of anthropogenic pressure and climate change. Using global satellite derived shoreline positions from 1993 to 2019 and a variety of reanalysis products, here we show that shorelines are under the influence of three main drivers: sea-level, ocean waves and river discharge. While sea level directly affects coastal mobility, waves affect both erosion/accretion and total water levels, and rivers affect coastal sediment budgets and salinity-induced water levels. By deriving a conceptual global model that accounts for the influence of dominant modes of climate variability on these drivers, we show that interannual shoreline changes are largely driven by different ENSO regimes and their complex inter-basin teleconnections. Our results provide a new framework for understanding and predicting climate-induced coastal hazards.

Epidemiology and the economic burden of traumatic fractures in China: A population-based study.

Objectives: National data on the admission rate, distribution, in-hospital mortality, and economic burden of traumatic fractures in China is unclear. We aimed to conduct a cross-sectional population-based study to determine such above data at the national level in China. Methods: A national administrative database was used to review all traumatic fracture hospitalizations in China during 2020, from which a total of 2,025,169 inpatients with traumatic fractures was retrieved. Admission rates and in-hospital mortality rates stratified by age, sex, and region were calculated. The causes of traumatic fracture and economic burden were described. Results: The admission rate of traumatic fractures of all China population in 2020 was 1.437‰. The admission rate increased with age and varied with genders and causes of injuries. Falls are the leading cause of traumatic fracture hospitalization, followed by road traffic injuries. The most common diagnoses were femoral neck fractures, with a number of 138,377. The in-hospital mortality was 1.209‰. Road traffic injuries led to the highest in-hospital mortality. The median length of stay was 10 days, with the median hospitalization cost of ¥20,900 (about $3,056). Conclusion: Traumatic fractures are concerning conditions with a high admission rate and in-hospital mortality in China, which are mainly caused by falls and road traffic injuries. The government should implement more public health policies to enhance the health of the elderly and improve transportation safety to prevent traumatic fractures.

Electronic Source Data Transcription for Electronic Case Report Forms in China: Validation of the Electronic Source Record Tool in a Real-world Ophthalmology Study.

BACKGROUND: As researchers are increasingly interested in real-world studies (RWSs), improving data collection efficiency and data quality has become an important challenge. An electronic source (eSource) generally includes direct capture, collection, and storage of electronic data to simplify clinical research. It can improve data quality and patient safety and reduce clinical trial costs. Although there are already large projects on eSource technology, there is a lack of experience in using eSource technology to implement RWSs. Our team designed and developed an eSource record (ESR) system in China. In a preliminary prospective study, we selected a cosmetic medical device project to evaluate ESR software's effect on data collection and transcription. As the previous case verification was simple, we plan to choose more complicated ophthalmology projects to further evaluate the ESR. OBJECTIVE: We aimed to evaluate the data transcription efficiency and quality of ESR software in retrospective studies to verify the feasibility of using eSource as an alternative to traditional manual transcription of data in RWS projects. METHODS: The approved ophthalmic femtosecond laser project was used for ESR case validation. This study compared the efficiency and quality of data transcription between the eSource method using ESR software and the traditional clinical research model of manually transcribing the data. Usability refers to the quality of a user's experience when interacting with products or systems including websites, software, devices, or applications. To evaluate the system availability of ESR, we used the System Usability Scale (SUS). The questionnaire consisted of the following 2 parts: participant information and SUS evaluation of the electronic medical record (EMR), electronic data capture (EDC), and ESR systems. By accessing log data from the EDC system previously used by the research project, all the time spent from the beginning to the end of the study could be counted. RESULTS: In terms of transcription time cost per field, the eSource method can reduce the time cost by 81.8% (11.2/13.7). Compared with traditional manual data transcription, the eSource method has higher data transcription quality (correct entry rate of 2356/2400, 98.17% vs 47,991/51,424, 93.32%). A total of 15 questionnaires were received with a response rate of 100%. In terms of usability, the average overall SUS scores of the EMR, EDC, and ESR systems were 50.3 (SD 21.9), 51.5 (SD 14.2), and 63.0 (SD 11.3; contract research organization experts: 69.5, SD 11.5; clinicians: 59.8, SD 10.2), respectively. The Cronbach α for the SUS items of the EMR, EDC, and ESR systems were 0.591 (95% CI -0.012 to 0.903), 0.588 (95% CI -0.288 to 0.951), and 0.785 (95% CI 0.576-0.916), respectively. CONCLUSIONS: In real-world ophthalmology studies, the eSource approach based on the ESR system can replace the traditional clinical research model that relies on the manual transcription of data.

Clinical Source Data Production and Quality Control in Real-world Studies: Proposal for Development of the eSource Record System.

BACKGROUND: An eSource generally includes the direct capture, collection, and storage of electronic data to simplify clinical research. It can improve data quality and patient safety and reduce clinical trial costs. There has been some eSource-related research progress in relatively large projects. However, most of these studies focused on technical explorations to improve interoperability among systems to reuse retrospective data for research. Few studies have explored source data collection and quality control during prospective data collection from a methodological perspective. OBJECTIVE: This study aimed to design a clinical source data collection method that is suitable for real-world studies and meets the data quality standards for clinical research and to improve efficiency when writing electronic medical records (EMRs). METHODS: On the basis of our group's previous research experience, TransCelerate BioPharm Inc eSource logical architecture, and relevant regulations and guidelines, we designed a source data collection method and invited relevant stakeholders to optimize it. On the basis of this method, we proposed the eSource record (ESR) system as a solution and invited experts with different roles in the contract research organization company to discuss and design a flowchart for data connection between the ESR and electronic data capture (EDC). RESULTS: The ESR method included 5 steps: research project preparation, initial survey collection, in-hospital medical record writing, out-of-hospital follow-up, and electronic case report form (eCRF) traceability. The data connection between the ESR and EDC covered the clinical research process from creating the eCRF to collecting data for the analysis. The intelligent data acquisition function of the ESR will automatically complete the empty eCRF to create an eCRF with values. When the clinical research associate and data manager conduct data verification, they can query the certified copy database through interface traceability and send data queries. The data queries are transmitted to the ESR through the EDC interface. The EDC and EMR systems interoperate through the ESR. The EMR and EDC systems transmit data to the ESR system through the data standards of the Health Level Seven Clinical Document Architecture and the Clinical Data Interchange Standards Consortium operational data model, respectively. When the implemented data standards for a given system are not consistent, the ESR will approach the problem by first automating mappings between standards and then handling extensions or corrections to a given data format through human evaluation. CONCLUSIONS: The source data collection method proposed in this study will help to realize eSource's new strategy. The ESR solution is standardized and sustainable. It aims to ensure that research data meet the attributable, legible, contemporaneous, original, accurate, complete, consistent, enduring, and available standards for clinical research data quality and to provide a new model for prospective data collection in real-world studies.

Existing barriers and recommendations of real-world data standardisation for clinical research in China: a qualitative study.

OBJECTIVE: To investigate the existing barriers and recommendations of real-world data (RWD) standardisation for clinical research through a qualitative study on different stakeholders. DESIGN: This qualitative study involved five types of stakeholders based on five interview outlines. The data analysis was performed using the constructivist grounded theory analysis process. SETTING: Eight hospitals, four hospital system vendors, three big data companies, six medical products companies and four regulatory institutions were included. PARTICIPANTS: In total, 62 participants from 25 institutions were interviewed through purposive sampling. RESULTS: The findings showed that the lack of clinical applicability in existing terminology standards, lack of generalisability in existing research databases, and lack of transparency in existing data standardisation process were the barriers of data standardisation of RWD for clinical research. Enhancing terminology standards by incorporating locally used clinical terminology, reducing burden in the usage of terminology standards, improving generalisability of RWD for research by using clinical data models, and improving traceability to source data for transparency might be feasible suggestions for solving the current problems. CONCLUSIONS: Efficient and reliable data standardisation of RWD for clinical research can help generate better evidence used to support regulatory evaluation of medical products. This research suggested enhancing terminology standards by incorporating locally used clinical terminology, reducing burden in the usage of terminology standards, improving generalisability of RWD for research by using clinical data models, and improving traceability to source data for transparency to guide efforts in data standardisation in the future.

Global decline in ocean memory over the 21st century.

Ocean memory, the persistence of ocean conditions, is a major source of predictability in the climate system beyond weather time scales. We show that ocean memory, as measured by the year-to-year persistence of sea surface temperature anomalies, is projected to steadily decline in the coming decades over much of the globe. This global decline in ocean memory is predominantly driven by shoaling of the upper-ocean mixed layer depth in response to global surface warming, while thermodynamic and dynamic feedbacks can contribute substantially regionally. As the mixed layer depth shoals, stochastic forcing becomes more effective in driving sea surface temperature anomalies, increasing high-frequency noise at the expense of persistent signals. Reduced ocean memory results in shorter lead times of skillful persistence-based predictions of sea surface thermal conditions, which may present previously unknown challenges for predicting climate extremes and managing marine biological resources under climate change.

Evaluation of the clinical application effect of eSource record tools for clinical research.

BACKGROUND: Electronic sources (eSources) can improve data quality and reduce clinical trial costs. Our team has developed an innovative eSource record (ESR) system in China. This study aims to evaluate the efficiency, quality, and system performance of the ESR system in data collection and data transcription. METHODS: The study used time efficiency and data transcription accuracy indicators to compare the eSource and non-eSource data collection workflows in a real-world study (RWS). The two processes are traditional data collection and manual transcription (the non-eSource method) and the ESR-based source data collection and electronic transmission (the eSource method). Through the system usability scale (SUS) and other characteristic evaluation scales (system security, system compatibility, record quality), the participants' experience of using ESR was evaluated. RESULTS: In terms of the source data collection (the total time required for writing electronic medical records (EMRs)), the ESR system can reduce the time required by 39% on average compared to the EMR system. In terms of data transcription (electronic case report form (eCRF) filling and verification), the ESR can reduce the time required by 80% compared to the non-eSource method (difference: 223 ± 21 s). The ESR accuracy in filling the eCRF field is 96.92%. The SUS score of ESR is 66.9 ± 16.7, which is at the D level and thus very close to the acceptable margin, indicating that optimization work is needed. CONCLUSIONS: This preliminary evaluation shows that in the clinical medical environment, the ESR-based eSource method can improve the efficiency of source data collection and reduce the workload required to complete data transcription.

Exploratory COVID-19 death risk score based on basic laboratory tests and physiological clinical measurements.

BACKGROUND: In the event of a sudden shortage of medical resources, a rapid, simple, and accurate prediction model is essential for the 30-day mortality rate of patients with COVID-19. METHODS: This retrospective study compared the characteristics of the survivals and non-survivals of 278 patients with COVID-19. Logistic regression analysis was performed to obtain the "COVID-19 death risk score" (CDRS) model. Using the area under the receiver operating characteristic (AUROC) curve and Hosmer-Lemeshow goodness-of-fit test, discrimination and calibration were assessed. Internal validation was conducted using a regular bootstrap method. RESULTS: A total of 63 (22.66%) of 278 included patients died. The logistic regression analysis revealed that high-sensitivity C-reactive protein (hsCRP; odds ratio [OR]=1.018), D-dimer (OR=1.101), and respiratory rate (RR; OR=1.185) were independently associated with 30-day mortality. CDRS was calculated as follows: CDRS=-10.245+(0.022×hsCRP)+(0.172×D-dimer)+(0.203×RR). CDRS had the same predictive effect as the sequential organ failure assessment (SOFA) and "confusion, uremia, respiratory rate, blood pressure, and age over 65 years" (CURB-65) scores, with AUROCs of 0.984 for CDRS, 0.975 for SOFA, and 0.971 for CURB-65, respectively. And CDRS showed good calibration. The AUROC through internal validations was 0.980 (95% confidence interval [CI]: 0.965-0.995). Regarding the clinical value, the decision curve analysis of CDRS showed a net value similar to that of CURB-65 in this cohort. CONCLUSION: CDRS is a novel, efficient and accurate prediction model for the early identification of COVID-19 patients with poor outcomes. Although it is not as advanced as the other models, CDRS had a similar performance to that of SOFA and CURB-65.

On the influence of ENSO complexity on Pan-Pacific coastal wave extremes.

Wind-generated waves are dominant drivers of coastal dynamics and vulnerability, which have considerable impacts on littoral ecosystems and socioeconomic activities. It is therefore paramount to improve coastal hazards predictions through the better understanding of connections between wave activity and climate variability. In the Pacific, the dominant climate mode is El Niño Southern Oscillation (ENSO), which has known a renaissance of scientific interest leading to great theoretical advances in the past decade. Yet studies on ENSO's coastal impacts still rely on the oversimplified picture of the canonical dipole across the Pacific. Here, we consider the full ENSO variety to delineate its essential teleconnection pathways to tropical and extratropical storminess. These robust seasonally modulated relationships allow us to develop a mathematical model of coastal wave modulation essentially driven by ENSO's complex temporal and spatial behavior. Accounting for this nonlinear climate control on Pan-Pacific wave activity leads to a much better characterization of waves' seasonal to interannual variability (+25% in explained variance) and intensity of extremes (+60% for strong ENSO events), therefore paving the way for significantly more accurate forecasts than formerly possible with the previous baseline understanding of ENSO's influence on coastal hazards.

Health Status of Left-Behind Children and Parenting Behaviors of Caregivers in Poor Rural Areas - 6 Provinces, China, 2018.

SUMMARY: What is already known about this topic? China has a significant population of left-behind children, and their health and living environments remain a major public health challenge. Children under 6 years old are especially vulnerable to poor health knowledge and behaviors of their caregivers. What is added by this report? The prevalence of stunting, being underweight, and often sick were 13%, 3.4%, and 5%, respectively. Only 53.9% of left-behind children could eat meat often, and 59.5% could control their intake of sugary drinks. The proportions of children who had a safe home environment, a safe play environment, and no family violence were 22.5%, 75.3%, and 45.9%, respectively. The percentages of caregivers who ensured that they rarely left their children alone and were always in their sight are 76.1% and 92.4%, respectively. What are the implications for public health practice? The implementation of early home visits is necessary to improve the physical health and safety of the living environment of left-behind children. Primary health workers should pay attention to improving the health knowledge and behaviors of caregivers.

Spurious North Tropical Atlantic precursors to El Niño.

The El Niño-Southern Oscillation (ENSO), the primary driver of year-to-year global climate variability, is known to influence the North Tropical Atlantic (NTA) sea surface temperature (SST), especially during boreal spring season. Focusing on statistical lead-lag relationships, previous studies have proposed that interannual NTA SST variability can also feed back on ENSO in a predictable manner. However, these studies did not properly account for ENSO's autocorrelation and the fact that the SST in the Atlantic and Pacific, as well as their interaction are seasonally modulated. This can lead to misinterpretations of causality and the spurious identification of Atlantic precursors for ENSO. Revisiting this issue under consideration of seasonality, time-varying ENSO frequency, and greenhouse warming, we demonstrate that the cross-correlation characteristics between NTA SST and ENSO, are consistent with a one-way Pacific to Atlantic forcing, even though the interpretation of lead-lag relationships may suggest otherwise.

Consensus-Based Linguistic Distribution Large-Scale Group Decision Making Using Statistical Inference and Regret Theory.

Large-scale group decision-making (LSGDM) deals with complex decision- making problems which involve a large number of decision makers (DMs). Such a complex scenario leads to uncertain contexts in which DMs elicit their knowledge using linguistic information that can be modelled using different representations. However, current processes for solving LSGDM problems commonly neglect a key concept in many real-world decision-making problems, such as DMs' regret aversion psychological behavior. Therefore, this paper introduces a novel consensus based linguistic distribution LSGDM (CLDLSGDM) approach based on a statistical inference principle that considers DMs' regret aversion psychological characteristics using regret theory and which aims at obtaining agreed solutions. Specifically, the CLDLSGDM approach applies the statistical inference principle to the consensual information obtained in the consensus process, in order to derive the weights of DMs and attributes using the consensus matrix and adjusted decision-making matrices to solve the decision-making problem. Afterwards, by using regret theory, the comprehensive perceived utility values of alternatives are derived and their ranking determined. Finally, a performance evaluation of public hospitals in China is given as an example in order to illustrate the implementation of the designed method. The stability and advantages of the designed method are analyzed by a sensitivity and a comparative analysis.

Development of a Malignancy Potential Binary Prediction Model Based on Deep Learning for the Mitotic Count of Local Primary Gastrointestinal Stromal Tumors.

OBJECTIVE: The mitotic count of gastrointestinal stromal tumors (GIST) is closely associated with the risk of planting and metastasis. The purpose of this study was to develop a predictive model for the mitotic index of local primary GIST, based on deep learning algorithm. MATERIALS AND METHODS: Abdominal contrast-enhanced CT images of 148 pathologically confirmed GIST cases were retrospectively collected for the development of a deep learning classification algorithm. The areas of GIST masses on the CT images were retrospectively labelled by an experienced radiologist. The postoperative pathological mitotic count was considered as the gold standard (high mitotic count, > 5/50 high-power fields [HPFs]; low mitotic count, ≤ 5/50 HPFs). A binary classification model was trained on the basis of the VGG16 convolutional neural network, using the CT images with the training set (n = 108), validation set (n = 20), and the test set (n = 20). The sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) were calculated at both, the image level and the patient level. The receiver operating characteristic curves were generated on the basis of the model prediction results and the area under curves (AUCs) were calculated. The risk categories of the tumors were predicted according to the Armed Forces Institute of Pathology criteria. RESULTS: At the image level, the classification prediction results of the mitotic counts in the test cohort were as follows: sensitivity 85.7% (95% confidence interval [CI]: 0.834-0.877), specificity 67.5% (95% CI: 0.636-0.712), PPV 82.1% (95% CI: 0.797-0.843), NPV 73.0% (95% CI: 0.691-0.766), and AUC 0.771 (95% CI: 0.750-0.791). At the patient level, the classification prediction results in the test cohort were as follows: sensitivity 90.0% (95% CI: 0.541-0.995), specificity 70.0% (95% CI: 0.354-0.919), PPV 75.0% (95% CI: 0.428-0.933), NPV 87.5% (95% CI: 0.467-0.993), and AUC 0.800 (95% CI: 0.563-0.943). CONCLUSION: We developed and preliminarily verified the GIST mitotic count binary prediction model, based on the VGG convolutional neural network. The model displayed a good predictive performance.