Clinical Practice For Prediction Research Articles

Nomogram for Predicting Survival in Locally Advanced Cervical Cancer with Concurrent Chemoradiotherapy plus or Not Adjuvant Chemotherapy: A Retrospective Analysis Based on 2018 FIGO Staging.

Background: The comprehensive treatment mode of combining concurrent chemoradiotherapy (CCRT) with adjuvant chemotherapy (AC) is a commonly used mainstream model in the clinical practice of locally advanced cervical cancer (LACC). However, the necessity for AC after CCRT lacks sufficient evidence-based medical support. This study constructs a predictive model for the survival time dependence of CCRT ± AC for LACC based on the 2018 International Federation of Gynecology and Obstetrics (FIGO) staging with internal validation, the prognosis was assessed with intensity-modulated radiotherapy (IMRT) and concurrent cisplatin, and provides guidance for future stratified treatment. Materials and Methods: The retrospective analysis included 482 patients with LACC who CCRT from January 2016 to January 2023. Patients who used the 2009 FIGO staging were all standardized for the 2018 FIGO staging. The 482 patients with LACC were divided into a training set (n = 290) and a validation set (n = 192) at a ratio of 6:4. COX multivariate regression model and LASSO regression were used to screen for independent prognostic factors affecting progression-free survival (PFS) and overall survival (OS), and a nomogram clinical prediction model was constructed based on these factors. Evaluate the effectiveness of the model through the receiver operating characteristic curve, calibration curve, decision curve, risk heat map, and survival curves for risk stratification. Results: The PFS and OS independent prognostic risk factors affecting the 2018 FIGO staging of LACC during CCRT were validated to be similar to the 2009 FIGO staging prediction model reported in previous literature. In the training cohort, area under the curve (AUC) values at 1, 3, and 5 years were 0.941, 0.882, and 0.885 for PFS, and 0.946, 0.946, and 0.969 for OS, respectively. When applied to a test cohort, the model also showed accurate prediction result (AUC at 1, 3, and 5 years were 0.869, 0.891, and 0.899 for PFS, and 0.891, 0.941 and 0.878 for OS, respectively). Subgroup analysis suggests that patients with LACC, adenocarcinoma, stage IVA, pelvic lymph node metastasis, pretreatment hemoglobin ≤100 g/l and residual tumor diameter >2 cm, who received CCRT in the 2018 FIGO stage, may benefit more from adjuvant chemtherapy. Conclusions: Based on the 2018 FIGO staging, a nomogram prediction model for PFS and OS in patients with LACC undergoing CCRT was developed. The model, established by combining weighted clinical and pathological factors, can provide more personalized treatment predictions in clinical practice. For patients with high-risk factors such as residual tumor diameter > 2 cm after CCRT for LACC, AC may bring benefits.

P066 Predictive Modeling of Infection Risk in Lupus Nephritis using Multiple Machine Learning Algorithms

Abstract Background/Aims The incidence of infections is considerably higher among patients with lupus nephritis (LN), and these infections significantly contribute to mortality rates. This study analysed the characteristics of peripheral blood lymphocyte subsets in LN patients, aiming at using machine learning (ML) methods to explore the risk factors of infection and establish a most effective ML algorithm to predict the occurrence of co-infection in LN. Methods This retrospective study encompassed LN patients, consisting of 111 non-infected individuals and 72 infected individuals. Additionally, 206 healthy controls (HCs) matched age and sex were recruited. The patient’s basic information, infection site, pathogen type, clinical manifestation, patient medication, and auxiliary laboratory indexes were recorded. Eight ML methods were compared to establish a corresponding model through a training group, and then verify the results in a test group. We trained the ML models, including Logistic Regression (LR), Decision Tree (DT), K-Nearest Neighbors (KNN), Support Vector Machine (SVM), Multi-Layer Perceptron (MLP), Random Forest (RF), Ada boost (Ada), Extreme Gradient Boosting (XGB), after selecting variables for LN infection prediction in the training set, and further evaluated potential predictors of infection. Results The levels of T, B, helper T, suppressor T, Natural killer cells in the infected patients were significantly decreased when compared with both non-infected LN patients and HCs. The number of regulatory T cells in LN was significantly lower than HCs, and the infected patients had the fewest regulatory T cells among all these groups. Regarding localization of infection, the respiratory tract was most frequently involved (69.4%), followed by gastrointestinal (20.8%) and urinary tract infection (12.5%). Concerning pathogens, bacterial infection was more common. Viral included Epstein Barr virus, cytomegalovirus, and respiratory syncytial virus. In training group, the AUC values, accuracy and precision of DT, RF, Ada, and XGB were 1, 100%, and 100%, respectively. As for the modeled effects of predicting LN co-infection in test groups by the eight algorithms, the three algorithms with the highest AUC values were SVM, RF and XGB, with values of 0.85, 0.81 and 0.76, respectively. Among the algorithms, XGB algorithm has the highest accuracy (80%) and precision (84.62%), thus it is recommended to be used for prediction in clinical practice. In addition, the top four weighting factors for infection identified of LN by the XGB in ML were T, red blood cell, hospital length of stay, and lymphocyte%. Conclusion Both innate and adaptive immune systems in LN patients are disturbed, and monitoring the lymphocyte subsets may provide a reference for the prevention and treatment of infection. Our research found that the XGB algorithm performs better than other models with the most accuracy in prediction, which may be one of the preferred algorithms in LN patients with co-infection for future research. Disclosure J. Zhang: None. P. Cai: None. J. Liu: None. Q. Xie: None.

Incidence and risk factors of systemic lupus erythematosus in patients with primary immune thrombocytopenia: a systematic review and meta-analysis.

Immune disorders and autoantibodies has been noted in both primary immune thrombocytopenia (ITP) and systemic lupus erythematosus (SLE). Whether the two disorders are correlated is unclear. The lack of evidence on the incidence of and risk factors for SLE in primary ITP patients poses a challenge for prediction in clinical practice. Therefore, we conducted this study. The protocol was registered with PROSPERO (CRD42023403665). Web of Science, Cochrane, PubMed, and EMBASE were searched for articles published from inception to 30 September 2023 on patients who were first diagnosed with primary ITP and subsequently developed into SLE. Furthermore, the risk factors were analyzed. Study quality was estimated using the Newcastle-Ottawa Scale. The statistical process was implemented using the R language. This systematic review included eight articles. The incidence of SLE during the follow-up after ITP diagnosis was 2.7% (95% CI [1.3-4.4%]), with an incidence of 4.6% (95% CI [1.6-8.6%]) in females and 0 (95% CI [0.00-0.4%]) in males. Older age (OR = 6.31; 95% CI [1.11-34.91]), positive antinuclear antibody (ANA) (OR = 6.64; 95% CI [1.40-31.50]), hypocomplementemia (OR = 8.33; 95% CI [1.62-42.91]), chronic ITP (OR = 24.67; 95% CI [3.14-100.00]), organ bleeding (OR = 13.67; 95% CI [2.44-76.69]), and female (OR = 20.50; 95% CI [4.94-84.90]) were risk factors for subsequent SLE in ITP patients. Patients with primary ITP are at higher risk of SLE. Specific follow-up and prevention strategies should be tailored especially for older females with positive ANA, hypocomplementemia, or chronic ITP. In subsequent studies, we need to further investigate the risk factors and try to construct corresponding risk prediction models to develop specific prediction strategies for SLE.

Prediction of conversion from mild cognitive impairment to Alzheimer’s disease and simultaneous feature selection and grouping using Medicaid claim data

BackgroundDue to the heterogeneity among patients with Mild Cognitive Impairment (MCI), it is critical to predict their risk of converting to Alzheimer’s disease (AD) early using routinely collected real-world data such as the electronic health record data or administrative claim data.MethodsThe study used MarketScan Multi-State Medicaid data to construct a cohort of MCI patients. Logistic regression with tree-guided lasso regularization (TGL) was proposed to select important features and predict the risk of converting to AD. A subsampling-based technique was used to extract robust groups of predictive features. Predictive models including logistic regression, generalized random forest, and artificial neural network were trained using the extracted features.ResultsThe proposed TGL workflow selected feature groups that were robust, highly interpretable, and consistent with existing literature. The predictive models using TGL selected features demonstrated higher prediction accuracy than the models using all features or features selected using other methods.ConclusionsThe identified feature groups provide insights into the progression from MCI to AD and can potentially improve risk prediction in clinical practice and trial recruitment.

Development and validation of a prehospital termination of resuscitation (TOR) rule for out - of hospital cardiac arrest (OHCA) cases using general purpose artificial intelligence (AI)

BackgroundPrehospital identification of futile resuscitation efforts (defined as a predicted probability of survival lower than 1%) for out-of-hospital cardiac arrest (OHCA) may reduce unnecessary transport. Reliable prediction variables for OHCA 'termination of resuscitation' (TOR) rules are needed to guide treatment decisions. The Universal TOR rule uses only three variables (Absence of Prehospital ROSC, Event not witnessed by EMS and no shock delivered on the scene) has been externally validated and is used by many EMS systems. Deep learning, an artificial intelligence (AI) platform is an attractive model to guide the development of TOR rule for OHCA. The purpose of this study was to assess the feasibility of developing an AI-TOR rule for neurologically favorable outcomes using general purpose AI and compare its performance to the Universal TOR rule. Methods: We identified OHCA cases of presumed cardiac etiology who were 18 years of age or older from 2016 to 2019 in the All-Japan Utstein Registry. We divided the dataset into 2 parts, the first half (2016–2017) was used as a training dataset for rule development and second half (2018–2019) for validation. The AI software (Prediction One®) created the model using the training dataset with internal cross-validation. It also evaluated the prediction accuracy and displayed the ranking of influencing variables. We performed validation using the second half cases and calculated the prediction model AUC. The top four of the 11 variables identified in the model were then selected as prognostic factors to be used in an AI-TOR rule, and sensitivity, specificity, positive predictive value, and negative predictive value were calculated from validation cohort. This was then compared to the performance of the Universal TOR rule using same dataset. ResultsThere were 504,561 OHCA cases, 18 years of age or older, 302,799 cases were presumed cardiac origin. Of these, 149,425 cases were used for the training dataset and 153,374 cases for the validation dataset. The model developed by AI using 11 variables had an AUC of 0.969, and its AUC for the validation dataset was 0.965. The top four influencing variables for neurologically favorable outcome were Prehospital ROSC, witnessed by EMS, Age (68 years old and younger) and nonasystole. The AUC calculated using the 4 variables for the AI-TOR rule was 0.953, and its AUC for the validation dataset was 0.952 (95%CI 0.949 –0.954). Of 80,198 patients in the validation cohort that satisfied all four criteria for the AI-TOR rule, 58 (0.07%) had a neurologically favorable one-month survival. The specificity of AI-TOR rule was 0.990, and the PPV was 0.999 for predicting lack of neurologically favorable survival, both the specificity and PPV were higher than that achieved with the universal TOR (0.959, 0.998). ConclusionsThe accuracy of prediction models using AI software to determine outcomes in OHCA was excellent and the AI-TOR rule’s variables from prediction model performed better than the Universal TOR rule. External validation of our findings as well as further research into the utility of using AI platforms for TOR prediction in clinical practice is needed.

Prognostic value of B-score for predicting joint replacement in the context of osteoarthritis phenotypes: Data from the osteoarthritis initiative

ObjectiveDeveloping new therapies for knee osteoarthritis (KOA) requires improved prediction of disease progression. This study evaluated the prognostic value of clinical clusters and machine-learning derived quantitative 3D bone shape B-score for predicting total and partial knee replacement (KR). DesignThis retrospective study used longitudinal data from the Osteoarthritis Initiative. A previous study used patients' clinical profiles to delineate phenotypic clusters. For these clusters, the distribution of B-scores was assessed (employing Tukey's method). The value of both cluster allocation and B-score for KR-prediction was then evaluated using multivariable Cox regression models and Kaplan-Meier curves for time-to-event analyses. The impact of using B-score vs. cluster was evaluated using a likelihood ratio test for the multivariable Cox model; global performances were assessed by concordance statistics (Harrell's C-index) and time dependent receiver operating characteristic (ROC) curves. ResultsB-score differed significantly for the individual clinical clusters (p ​< ​0.001). Overall, 9.4% of participants had a KR over 9 years, with a shorter time to event in clusters with high B-score at baseline. Those clusters were characterized clinically by a high rate of comorbidities and potential signs of inflammation. Both phenotype and B-score independently predicted KR, with better prediction if combined (P ​< ​0.001). B-score added predictive value in groups with less pain and radiographic severity but limited physical activity. ConclusionsB-scores correlated with phenotypes based on clinical patient profiles. B-score and phenotype independently predicted KR surgery, with higher predictive value if combined. This can be used for patient stratification in drug development and potentially risk prediction in clinical practice.

Development of a monoclonal antibody to ITPRIPL1 for immunohistochemical diagnosis of non-small cell lung cancers: accuracy and correlation with CD8+ T cell infiltration.

Introduction: Cancer biomarkers are substances or processes highly associated with the presence and progression of cancer, which are applicable for cancer screening, progression surveillance, and prognosis prediction in clinical practice. In our previous studies, we discovered that cancer cells upregulate inositol 1,4,5-triphosphate receptor-interacting protein-like 1 (ITPRIPL1), a natural CD3 ligand, to evade immune surveillance and promote tumor growth. We also developed a monoclonal ITPRIPL1 antibody with high sensitivity and specificity. Here, we explored the application of anti-ITPRIPL1 antibody for auxiliary diagnosis of non-small cell lung cancer (NSCLC). Methods: NSCLC patient tissue samples (n = 75) were collected and stained by anti-ITPRIPL1 or anti-CD8 antibodies. After excluding the flaked samples (n = 15), we evaluated the expression by intensity (0-3) and extent (0-100%) of staining to generate an h-score for each sample. The expression status was classified into negative (h-score < 20), low-positive (20-99), and high-positive (≥ 100). We compared the h-scores between the solid cancer tissue and stroma and analyzed the correlation between the h-scores of the ITPRIPL1 and CD8 expression in situ in adjacent tissue slices. Results: The data suggested ITPRIPL1 is widely overexpressed in NSCLC and positively correlates with tumor stages. We also found that ITPRIPL1 expression is negatively correlated with CD8 staining, which demonstrates that ITPRIPL1 overexpression is indicative of poorer immune infiltration and clinical prognosis. Therefore, we set 50 as the cutoff point of ITPRIPL1 expression H scores to differentiate normal and lung cancer tissues, which is of an excellent sensitivity and specificity score (100% within our sample collection). Discussion: These results highlight the potential of ITPRIPL1 as a proteomic immunohistochemical NSCLC biomarker with possible advantages over the existing NSCLC biomarkers, and the ITPRIPL1 antibody can be applied for accurate diagnosis and prognosis prediction.

De-accumulated error collaborative learning framework for predicting Alzheimer’s disease progression

Alzheimer’s disease (AD) is a chronic neurodegenerative disorder where precise prediction of progression is crucial for improving clinical diagnosis. However, missing data often complicates AD prediction in clinical practice. Existing studies primarily have employed recurrent neural networks to impute missing data, but this approach suffers from the “error accumulation” problem, leading to unsatisfactory prediction results. To address this issue, this study proposes the LSTM-TSGAIN collaborative learning framework, which consists of the following three main aspects: using a generative adversarial imputation method to reduce LSTM prediction errors; collaborative training of the adversarial imputation fusion module and the time series learning module to improve the performance of the model; the model inputs were adjusted to variable lengths to accommodate the differences in the number of visits of different subjects. The effectiveness of the model is demonstrated through experiments using longitudinal data from the ADNI dataset of 1256 subjects, which shows its superiority over state-of-the-art methods.

Abstract 15002: Cardiovascular Disease Risk Prediction by Random Survival Forest: The Korean National Health Insurance Service-National Health Screening Cohort

Introduction: Risk prediction is critical for cardiovascular disease (CVD) prevention and management. Existing CVD risk prediction models including the Framingham Risk Score (FRS) are based on the Cox proportional hazard ratio regression. Adaptation of a data mining approach could improve the accuracy of CVD risk prediction by being free from the linearity assumption. The aim of study was to construct Korean population-based CVD risk prediction models using random survival forest and to compare their predictive accuracy to the FRS in a 14-year Korean cohort. Methods: This was a secondary data analysis using a Korean population-based cohort. Out of 463,783 eligible participants, 70% were selected for training dataset using stratified random sampling, and the remaining 30% served as test dataset. Random survival forest was adopted to build risk prediction models using the training dataset. The test dataset was used to test the validity of the new models and the FRS. Results: Three random forest-based models were developed, and they were good in discrimination in both men and women with the c-statistics ranging from 0.698 to 0.747 while the c-statistics for the FRS were 0.692 and 0.733 in men and women, respectively. All random forest-based models showed good calibration, whereas the FRS tended to overestimate the CVD risk for men. Conclusions: The random survival forest-based models showed comparable performance with the FRS in a Korean population-based cohort. Continuous efforts are needed to improve the accuracy of CVD risk prediction for the Korean population and to utilize CVD risk prediction in clinical practice.

Development and validation of prognostic nomogram for T1-3N0M0 non-small cell lung cancer after curative resection

BackgroundRadical resection plus lymph node dissection is a common treatment for patients with T1-3N0M0 non-small cell lung cancer (NSCLC). Few models predicted the survival outcomes of these patients. This study aimed to developed a nomogram for predicting their overall survival (OS).Materials and methodsThis study involved 3002 patients with T1-3N0M0 NSCLC after curative resection between January 1999 and October 2013. 1525 Patients from Sun Yat-sen University Cancer Center were randomly allocated to training cohort and internal validation cohort in a ratio of 7:3. 1477 patients from ten institutions were recruited as external validation cohort. A nomogram was constructed based on the training cohort and validated by internal and external validation cohort to predict the OS of these patients. The accuracy and practicability were tested by Harrell's C-indexes, calibration plots and decision curve analyses (DCA).ResultsAge, sex, histological classification, pathological T stage, and HI standard were independent factors for OS and were included in our nomogram. The C-index of the nomogram for OS estimates were 0.671 (95% CI, 0.637–0.705),0.632 (95% CI, 0.581–0.683), and 0.645 (95% CI, 0.617–0.673) in the training cohorts, internal validation cohorts, and external validation cohort, respectively. The calibration plots and DCA for predictions of OS were in excellent agreement. An online version of the nomogram was built for convenient clinical practice.ConclusionsOur nomogram can predict the OS of patients with T1-3N0M0 NSCLC after curative resection. The online version of our nomogram offer opportunities for fast personalized risk stratification and prognosis prediction in clinical practice.

P046 A multinomial approach to prediction of methotrexate treatment response and discontinuation due to adverse events in rheumatoid arthritis

Abstract Background/Aims As recommended by the NICE guidelines, methotrexate (MTX) is typically prescribed as the first line therapy for patients with rheumatoid arthritis (RA). However, around 40% of patients do not respond to MTX at 6 months and around 80% experience adverse events (AEs). Our previous systematic review identified that clinical prediction models of MTX outcomes suffered from methodological limitations, including a lack of validation, suboptimal handling of missing data, and no consideration of competing risks, such as patients discontinuing due to adverse events (AEs). These shortcomings resulted in high risk of bias and should be addressed to aid the implementation of outcome prediction in clinical practice. Therefore, this study aimed to (i) develop a multinomial prediction model for estimating an individual’s risks of not achieving low disease activity (LDA) and discontinuing due to AEs at 6 months after starting MTX, using information observable at the starting point of treatment, (ii) update prognosis at 3 months, and (iii) internally validate these models to assess performance. Methods Data came from the national multi-centre (n = 38) Rheumatoid Arthritis Medication Study (RAMS), which recruited patients with a diagnosis of RA or undifferentiated polyarthritis starting MTX for the first time. We conducted a Patient and Public Involvement focus group with past and current MTX users, which informed our outcome choice of LDA (DAS28 ≤3.2). A multinomial logistic regression (MLR) was used to develop a prediction model that estimated the probabilities of three outcomes: 1) not in LDA, 2) achieved LDA, and 3) discontinued due to AEs. The updated model at 3 months was conditional on patients not having achieved LDA. Missing data was handled using multiple imputation and models were internally validated through bootstrapping. Calibration, discrimination, and variance explained was assessed to quantify model performance. Results A total of 1632 patients were included in the analysis. At 6 months, 756 (46%) patients achieved LDA, 730 (45%) were not in LDA, and 146 (9%) discontinued due to AEs. The updated model at 3 months included 1179 (72%) patients as they had not achieved LDA. For the LDA outcome pair, the c-statistic was 0.73 (0.70, 0.75), while it was 0.54 (0.49, 0.59) for the AEs outcome. The calibration slope was 0.99 (0.85, 1.12) and 0.74 (0.23, 1.23), respectively. Predictive performance at 3 months was similar to baseline. Conclusion Our MLR models handled outcomes of disease activity and AEs simultaneously. We were able to predict outcomes of LDA more accurately than AEs and performance was better in terms of calibration, which is crucial for the clinical utility of a prediction model. This modelling approach could provide more clinically useful and realistic predictions of MTX outcomes, as competing risks such as AEs have previously been ignored. Disclosure C.K. Gehringer: None. G.P. Martin: None. K.L. Hyrich: Honoraria; Abbvie. Grants/research support; BMS and Pfizer. S.M.M. Verstappen: None. J.C. Sergeant: None.

12-lead ECG signal processing and atrial fibrillation prediction in clinical practice.

Because clinically used 12-lead electrocardiography (ECG) devices have high falsepositive errors in automatic interpretations of atrial fibrillation (AF), they require substantial improvements before use. A clinical 12-lead ECG pre-processing method with a parallel convolutional neural network (CNN) model for 12-lead ECG automatic AF recognition is introduced. Raw AF diagnosis data from a 12-lead ECG device were collected and analyzed by two cardiologists to differentiate between true- and false-positives. Using a stationary wavelet transform (SWT) and independent component analysis (ICA) noise reduction was conducted and baseline wandering was corrected for the raw signals. AF patterns were learned and predicted using a parallel CNN deep learning (DL) model. (1) The proposed method alleviates the decreased ECG QRS amplitude enhances the signal-to-noise ratio and clearly shows atrial and ventricular activities. (2) After training, the CNNbased AF detector significantly reduced false-positive errors. The precision of AF diagnosis increased from 77.3% to 94.0 ± 1.5% as compared to ECG device interpretation. For AF screening, the model showed an average sensitivity of 96.8 ± 2.2%, specificity of 79.0 ± 5.8%, precision of 94.0 ± 1.5%, F1-measure of 95.2 ± 1.0%, and overall accuracy of 92.7 ± 1.5%. The method can bridge the gap between the research and clinical practice The ECG signal pre-processing and DL-based AF interpretation can be rapidly implemented clinically.

Implementation of five machine learning methods to predict the 52-week blood glucose level in patients with type 2 diabetes.

For the patients who are suffering from type 2 diabetes, blood glucose level could be affected by multiple factors. An accurate estimation of the trajectory of blood glucose is crucial in clinical decision making. Frequent glucose measurement serves as a good source of data to train machine learning models for prediction purposes. This study aimed at using machine learning methods to predict blood glucose for type 2 diabetic patients. We investigated various parameters influencing blood glucose, as well as determined the most effective machine learning algorithm in predicting blood glucose. 273 patients were recruited in this research. Several parameters such as age, diet, family history, BMI, alcohol intake, smoking status etal were analyzed. Patients who had glycosylated hemoglobin less than 6.5% after 52 weeks were considered as having achieved glycemic control and the rest as not achieving it. Five machine learning methods (KNN algorithm, logistic regression algorithm, random forest algorithm, support vector machine, and XGBoost algorithm) were compared to evaluate their performances in prediction accuracy. R 3.6.3 and Python 3.12 were used in data analysis. The statistical variables for which p< 0.05 was obtained were BMI, pulse, Na, Cl, AKP. Compared with the other four algorithms, XGBoost algorithm has the highest accuracy (Accuracy=99.54% in training set and 78.18% in testing set) and AUC values (1.0 in training set and 0.68 in testing set), thus it is recommended to be used for prediction in clinical practice. When it comes to future blood glucose level prediction using machine learning methods, XGBoost algorithm scores the highest in effectiveness. This algorithm could be applied to assist clinical decision making, as well as guide the lifestyle of diabetic patients, in pursuit of minimizing risks of hyperglycemic or hypoglycemic events.

A generalization performance study on the boosting radiotherapy dose calculation engine based on super-resolution

PurposeDuring the radiation treatment planning process, one of the time-consuming procedures is the final high-resolution dose calculation, which obstacles the wide application of the emerging online adaptive radiotherapy techniques (OLART). There is an urgent desire for highly accurate and efficient dose calculation methods. This study aims to develop a dose super resolution-based deep learning model for fast and accurate dose prediction in clinical practice. MethodA Multi-stage Dose Super-Resolution Network (MDSR Net) architecture with sparse masks module and multi-stage progressive dose distribution restoration method were developed to predict high-resolution dose distribution using low-resolution data. A total of 340 VMAT plans from different disease sites were used, among which 240 randomly selected nasopharyngeal, lung, and cervix cases were used for model training, and the remaining 60 cases from the same sites for model benchmark testing, and additional 40 cases from the unseen site (breast and rectum) was used for model generalizability evaluation. The clinical calculated dose with a grid size of 2 mm was used as baseline dose distribution. The input included the dose distribution with 4 mm grid size and CT images. The model performance was compared with HD U-Net and cubic interpolation methods using Dose-volume histograms (DVH) metrics and global gamma analysis with 1%/1 mm and 10% low dose threshold. The correlation between the prediction error and the dose, dose gradient, and CT values was also evaluated. ResultsThe prediction errors of MDSR were 0.06–0.84% of Dmean indices, and the gamma passing rate was 83.1–91.0% on the benchmark testing dataset, and 0.02–1.03% and 71.3–90.3% for the generalization dataset respectively. The model performance was significantly higher than the HD U-Net and interpolation methods (p < 0.05). The mean errors of the MDSR model decreased (monotonously by 0.03–0.004%) with dose and increased (by 0.01–0.73%) with the dose gradient. There was no correlation between prediction errors and the CT values. ConclusionThe proposed MDSR model achieved good agreement with the baseline high-resolution dose distribution, with small prediction errors for DVH indices and high gamma passing rate for both seen and unseen sites, indicating a robust and generalizable dose prediction model. The model can provide fast and accurate high-resolution dose distribution for clinical dose calculation, particularly for the routine practice of OLART.

The scientific progress and prospects of artificial intelligence in digestive endoscopy: A comprehensive bibliometric analysis.

Artificial intelligence (AI) has been used for diagnosis and outcome prediction in clinical practice. Furthermore, AI in digestive endoscopy has attracted much attention and shown promising and stimulating results. This study aimed to determine the development trends and research hotspots of AI in digestive endoscopy by visualizing articles. Publications on AI in digestive endoscopy research were retrieved from the Web of Science Core Collection on April 25, 2022. VOSviewer and CiteSpace were used to assess and plot the research outputs. This analytical research was based on original articles and reviews. A total of 524 records of AI research in digestive endoscopy, published between 2005 and 2022, were retrieved. The number of articles has increased 27-fold from 2017 to 2021. Fifty-one countries and 994 institutions contributed to all publications. Asian countries had the highest number of publications. China, the USA, and Japan were consistently the leading driving forces and mainly contributed (26%, 21%, and 14.31%, respectively). With a solid academic reputation in this area, Japan has the highest number of citations per article. Tada Tomohiro published the most articles and received the most citations.. Gastrointestinal endoscopy published the largest number of publications, and 4 of the top 10 cited papers were published in this journal. "The Classification," "ulcerative colitis," "capsule endoscopy," "polyp detection," and "early gastric cancer" were the leading research hotspots. Our study provides systematic elaboration for researchers to better understand the development of AI in gastrointestinal endoscopy.

Kinase Domain Mutation Testing and Clinical Outcome in a Nationwide Chronic Myeloid Leukemia Patient Population

Background: ABL1 kinase domain mutations (KDM) are associated with resistance to tyrosine kinase inhibitors (TKI) and inferior outcome in chronic myeloid leukemia (CML) patients. Current guidelines recommend KDM testing in case of failure or resistance during TKI treatment. In this analysis, we aimed to assess (1) the clinical practice of KDM testing in the Netherlands, (2) the observed responses to subsequent TKI in the presence of a KDM in relation to the in vitro half-maximal inhibitory concentration (IC50)-values, and (3) the clinical outcome of patients for whom KDM testing was done. Methods: Data were obtained from a nationwide population based CML registry (Pharos-HemoBase) and six CML-reference centers in the Netherlands. We included adult Ph+ CML patients who were treated with a TKI and who had at least one KDM analysis after December 2005, when second generation TKI became available in the Netherlands. Reasons for KDM testing were categorized in three main groups based on the pattern of TKI failure: (1) primary resistance (no response), (2) suboptimal response and (3) loss of response. A suboptimal response was defined as failure to achieve the ELN response milestones, but with the achievement of at least 1-log BCR::ABL1 reduction. Loss of response was defined as losing a previously achieved ELN response milestone. A cox regression model was used for the survival analysis. Results: A total of 391 KDM analyses performed in 275 patients between December 2015 and March 2022 were included. When focusing on CP CML, most KDM analyses were performed because of a suboptimal TKI response (n=155), however in only nine (6%) a KDM was detected. In patients with primary TKI resistance, a KDM was detected in 17 of 81 analyses (21%). The highest yield of KDM was in patients losing a previously achieved response with 43 of 89 (49%) of analyses showing a mutation. Twenty-five KDM analyses were performed outside the context of TKI failure (e.g. BCR::ABL1 fluctuations but no loss of MMR, molecular relapse after a TFR attempt) and in none of them a mutation was detected. In total, 84 KDM were detected (panel A). Sixty-nine (25%) patients harbored at least one KDM of whom 24 the T315I mutation and eight a compound mutation. Most KDM analyses were done with Sanger Sequencing (SS). Fifty-four TKI treatment lines were assessable for response for patients who harbored a KDM with known in vitro IC50-values for predicted effectiveness of TKI. Of the patients who were treated with a TKI with a good predicted response, 34 of 42 (81%) indeed obtained a durable major molecular response, while seven patients experienced failure (4 no response, 3 loss of response). In patients who were treated with a TKI for which intermediate sensitivity was predicted, 4 of 6 (67%) responded well. In patients who were treated with a TKI for which resistance was predicted, only 3 of 6 (50%) responded. Progression to advanced stage disease (AP/BC), hematopoietic stem cell transplantation (HSCT) and death occurred more often in patients with a KDM compared to patients without a KDM: AP/BC in 25 (36%) versus 29 (14%) patients, HSCT in 20 (29%) versus 43 (21%), death in 26 (38%) versus 29 (14%) and CML-related death in 12 (17%) versus 10 (5%), respectively. Ten-year overall survival (OS) after the first KDM analysis was 49% versus 82% in patients with versus without a KDM, respectively, with a median follow-up of 7.2 years (panel B). The HR of harboring a KDM for OS was 0.355 (95% CI, 0.209-0.603; p<0.001). After correction for age, sex, disease phase and Sokal score, the HR was 0.356 (95% CI, 0.182-0.696; p=0.003). Discussion: This is the first study describing the nationwide practice of KDM testing and assessing the clinical outcome of patients with a KDM in relation to IC50-values. More than 70% of the KDM were detected in patients with primary TKI resistance or in patients who lost a previously achieved response. Observed responses to TKI after the detection of a KDM were mainly in accordance with expected responses based on the IC-value, supporting the use of these in vitro derived predictions in clinical practice. In line with previous studies, the presence of a KDM translated in inferior outcomes with higher prevalence of progressive disease and (CML-related) death. Figure 1View largeDownload PPTFigure 1View largeDownload PPT Close modal

Automated model calibration with parallel MCMC: Applications for a cardiovascular system model.

Computational physiological models continue to increase in complexity, however, the task of efficiently calibrating the model to available clinical data remains a significant challenge. One part of this challenge is associated with long calibration times, which present a barrier for the routine application of model-based prediction in clinical practice. Another aspect of this challenge is the limited available data for the unique calibration of complex models. Therefore, to calibrate a patient-specific model, it may be beneficial to verify that task-specific model predictions have acceptable uncertainty, rather than requiring all parameters to be uniquely identified. We have developed a pipeline that reduces the set of fitting parameters to make them structurally identifiable and to improve the efficiency of a subsequent Markov Chain Monte Carlo (MCMC) analysis. MCMC was used to find the optimal parameter values and to determine the confidence interval of a task-specific prediction. This approach was demonstrated on numerical experiments where a lumped parameter model of the cardiovascular system was calibrated to brachial artery cuff pressure, echocardiogram volume measurements, and synthetic cerebral blood flow data that approximates what can be obtained from 4D-flow MRI data. This pipeline provides a cerebral arterial pressure prediction that may be useful for determining the risk of hemorrhagic stroke. For a set of three patients, this pipeline successfully reduced the parameter set of a cardiovascular system model from 12 parameters to 8-10 structurally identifiable parameters. This enabled a significant efficiency improvement in determining confidence intervals on predictions of pressure compared to performing a naive MCMC analysis with the full parameter set. This demonstrates the potential that the proposed pipeline has in helping address one of the key challenges preventing clinical application of such models. Additionally, for each patient, the MCMC approach yielded a 95% confidence interval on systolic blood pressure prediction in the middle cerebral artery smaller than ±10mmHg (±1.3kPa). The proposed pipeline exploits available high-performance computing parallelism to allow straightforward automation for general models and arbitrary data sets, enabling automated calibration of a parameter set that is specific to the available clinical data with minimal user interaction.

Prediction of prognosis in patients with panic disorder using pre-treatment brain white matter features

BackgroundThe early identification of patients with panic disorder (PD) with a poor prognosis is important for improving treatment outcomes; however, it is challenging due to a lack of objective biomarkers. We investigated the reliability of characterizing structural white matter (WM) connectivity and its ability to predict PD prognosis after pharmacotherapy. MethodsA total of 138 patients (59 men) with PD and 153 healthy controls (HCs; 73 men) participated in this study. PD symptom severity was measured using the Panic Disorder Severity Scale (PDSS) at baseline and follow-up periods of 8 weeks, 6 months, and 1 year. The least absolute shrinkage and selection operator (Lasso) was utilized to identify prognosis-related WM regions on diffusion imaging features. ResultsLasso identified seven prognosis-related WM regions: the bilateral posterior corona radiata, bilateral posterior limb of the internal capsule, the left retrolenticular part of the internal capsule, the left sagittal stratum, and the right fornix/stria terminalis. Some of these regions showed lower mean fractional anisotropy (FA) values in patients with PD than in HCs. The predicted PDSS scores using FA from these regions consistently correlated with the actual prognosis in all periods. LimitationsThis study had limited ability to evaluate individual longitudinal changes in detail owing to the data acquisition time and brain atlas resolution. ConclusionsOur findings suggest the possibility of using structural WM connectivity as a biomarker for the clinical characterization of PD. Our findings will expand our understanding of the neurobiology of PD and improve biomarker-based prognosis prediction in clinical practice.

Machine learning model to predict mental health crises from electronic health records

The timely identification of patients who are at risk of a mental health crisis can lead to improved outcomes and to the mitigation of burdens and costs. However, the high prevalence of mental health problems means that the manual review of complex patient records to make proactive care decisions is not feasible in practice. Therefore, we developed a machine learning model that uses electronic health records to continuously monitor patients for risk of a mental health crisis over a period of 28 days. The model achieves an area under the receiver operating characteristic curve of 0.797 and an area under the precision-recall curve of 0.159, predicting crises with a sensitivity of 58% at a specificity of 85%. A follow-up 6-month prospective study evaluated our algorithm’s use in clinical practice and observed predictions to be clinically valuable in terms of either managing caseloads or mitigating the risk of crisis in 64% of cases. To our knowledge, this study is the first to continuously predict the risk of a wide range of mental health crises and to explore the added value of such predictions in clinical practice.

ICU Cockpit: a platform for collecting multimodal waveform data, AI-based computational disease modeling and real-time decision support in the intensive care unit.

ICU Cockpit: a secure, fast, and scalable platform for collecting multimodal waveform data, online and historical data visualization, and online validation of algorithms in the intensive care unit. We present a network of software services that continuously stream waveforms from ICU beds to databases and a web-based user interface. Machine learning algorithms process the data streams and send outputs to the user interface. The architecture and capabilities of the platform are described. Since 2016, the platform has processed over 89 billion data points (N = 979 patients) from 200 signals (0.5-500Hz) and laboratory analyses (once a day). We present an infrastructure-based framework for deploying and validating algorithms for critical care. The ICU Cockpit is a Big Data platform for critical care medicine, especially for multimodal waveform data. Uniquely, it allows algorithms to seamlessly integrate into the live data stream to produce clinical decision support and predictions in clinical practice.