Machine learning for the prediction of volume responsiveness in patients with oliguric acute kidney injury in critical care

Interpretable machine learning models to predict decline in intrinsic capacity among older adults in China: a prospective cohort study

BACKGROUND

Monitoring intrinsic capacity and implementing appropriate interventions can support healthy aging. There are, though, few tools available for predicting decline in intrinsic capacity among older adults. This study aimed to develop and validate an interpretable machine learning model designed to identify populations at elevated risk of a decline in intrinsic capacity.

METHODS

Using data from the China Health and Retirement Longitudinal Study baseline (2011) and 4-year follow-up (2015), a total of 822 participants were randomly allocated to a training set and a testing set at a 7:3 ratio. Five machine learning methods were employed to train the model and assess its performance through various metrics. The SHapley Additive exPlanation method was subsequently used to interpret the optimal model.

RESULTS

The 4-year incidence of decline in intrinsic capacity among the older adults in the sample was 44.6 % (n = 367). Nine variables were screened for model construction, among which eXtreme gradient boosting demonstrated the best predictive performance, achieving an area under the receiver operating characteristic curve (AUC) of 0.715 (95 % CI 0.651-0.780) in the testing set. The SHapley Additive exPlanation method identified educational level, smoking, handgrip strength, self-rated health, and residence as the top five significant predictors.

CONCLUSIONS

The developed model can serve as a highly effective tool for primary care teams to identify older adults with early signs of decline in intrinsic capacity, facilitating the provision of subsequent screening and tailored interventions for intrinsic capacity.

An explainable non-invasive hybrid machine learning framework for accurate prediction of thyroid-stimulating hormone levels

Machine learning models, including thyroid biomarkers, are increasingly utilized in healthcare for biomarker prediction. These models offer the potential to enhance disease diagnosis through data-driven approaches relying on non-invasive techniques. However, no studies have explored the application of fully non-invasive methods for predicting thyroid-stimulating hormone (TSH) levels. Consequently, this study introduces a novel, fully non-invasive framework for predicting TSH levels by developing an innovative hybrid machine learning model that balances performance, complexity, and interpretability. Seven ML models were evaluated, and the best-performing models were integrated into a hybrid approach to balance performance, complexity, and interpretability. A dataset of 6190 instances from Jordan was used for model development. Four-dimensional non-invasive factors, including demographics, symptoms, family history, and newly engineered symptom scores, were incorporated into the model. The hybrid model achieved an R2 of 94.2 % and RMSE of 0.015, demonstrating superior predictive performance. Model interpretability was ensured using LIME and SHAP explainers, confirming aggregated symptom scores' critical role in enhancing prediction accuracy. A robust feature selection technique was implemented, reducing model complexity and enhancing performance. Among the top ten features for predicting TSH levels were hypothyroidism and hyperthyroidism symptom scores, family history, cold intolerance, itchy-dry skin, sweating, hand tremors, and palpitations. The model can be employed to develop cost-effective diagnostic tools for thyroid disorders. It also offers a robust framework that can be generalized to predict other biomarkers and applied in diverse contexts.

A predictive model for depression in elderly people with arthritis based on the TRIPOD guidelines

OBJECTIVE

Through machine learning algorithms, a prediction model for depression in arthritis patients was established to provide a basis for related interventions.

METHODS

Data from 4240 patients with arthritis were collected from the National Health and Nutrition Examination Survey database and divided into a training set (70 %) and a test set (30 %). LASSO Regression was employed for feature variable selection, and predictive models were constructed using five machine learning algorithms: Random Forest (RF), Logistic Regression (LR), Extreme Gradient Boosting (XGBoost), Gaussian Naive Bayes (GNB), and Gradient Boosting Decision Tree (GBDT). Model performance was evaluated through various metrics, including the Area Under the Receiver Operating Characteristic Curve (AUC), accuracy, sensitivity, cutoff, recall, Kappa, Positive Predictive Value(PPV) and Negative Predictive Value(NPV). Additionally, Shapley Additive Explanations (SHAP) analysis was conducted for personalized risk assessment.

RESULT

The predictive performance of the random forest model was the highest,with an area under curve (AUC) of 0.811 (SD:0.000) for the training set and 0.780 (SD:0.001) for the test set. The model identified eight significant variables associated with the occurrence of arthritis depression, including health status, difficulty standing for extended periods, difficulty getting in and out of bed, difficulty sitting for prolonged durations, education, gender, difficulty managing finances, and race. DCA demonstrated that the nomogram was clinically beneficial.

CONCLUSION

The predictive model developed for identifying depression in patients with arthritis exhibits high predictive ability, and good clinical applicability. This model can aid healthcare providers in the early detection of depression, thereby enabling timely interventions that can enhance patient prognosis and promote healthy aging. Future research should incorporate real-time biomarker monitoring to refine dynamic risk assessment.

Evaluation and comparison of machine learning algorithms for predicting discharge against medical advice in injured inpatients

BACKGROUND

Whether the application of machine learning algorithms offers an advantage over logistic regression in forecasting discharge against medical advice occurrences needs to be evaluated.

METHODS

This retrospective study included all inpatient records from January 1, 2018, to December 31, 2023. The foundational data set (2018-2021) was divided into a training set (80%) and a test set (20%) for model construction and internal validation. The temporal validation data set (2022-2023) was used to assess the model's prospective performance. Feature selection was performed using the BorutaShap method. Techniques including random oversampling, random undersampling, synthetic minority oversampling technique, and edited nearest neighbors were applied to address data imbalance. Model performance was evaluated using metrics including the area under the receiver operating characteristic curve, accuracy, specificity, sensitivity, F1 score, and geometric mean. The Shapley Additive Explanations analysis provided interpretation for the best machine learning model.

RESULTS

A total of 48,394 inpatient records for injured patients met the study criteria, of which 44,119 were discharged following medical advice and 4,275 chose discharge against medical advice, resulting in a ratio of 10.32:1. Among injury inpatients, 8.8% opted for discharge against medical advice. Based on the results of feature selection and multicollinearity analysis, 16 variables were ultimately selected for the construction and evaluation of the discharge against medical advice model. The light gradient boosting machine + edited nearest neighbors model showed the best generalization, with areas under the curves of 0.820 for internal validation and 0.837 for temporal validation. The Shapley Additive Explanations method was used to interpret the model, indicating that the grade of surgery is the most important variable.

CONCLUSIONS

The study is the first to use machine learning models to predict discharge against medical advice in injured inpatients, demonstrating its feasibility. In the future, health care institutions can learn from these models to optimize patient management and reduce discharge against medical advice incidents.

Transport of cyclic volatile methylsiloxanes in residence due to the use of personal care products

Emissions of cyclic volatile methylsiloxanes (cVMS) due to the use of personal care products pose adverse effect on indoor air quality. We make the first attempt to explore the emission characteristics of decamethylcyclopentasiloxane (D5, the most abundant cVMS) in an occupied residence via modelling and experimental approaches. A mass transfer model for characterizing the emissions of D5 from human skin lipids is developed, which incorporates the adsorption/desorption effect of D5 onto indoor surfaces that was ignored previously. A long-term field campaign in residence was conducted to examine the D5 emission behaviors, showing a number of concentration peaks after using personal care products. Special enhanced-ventilation and emission experiments were designed to determine the mass transfer parameters in the model via a hybrid optimization method, which were then used to predict D5 concentrations in some other test days. Good agreement between model predictions and observations demonstrates the effectiveness of the model and measured parameters. Feature importance analysis on indoor D5 transport is further performed, indicating that air exchange rate is the most important factor compared with temperature and relative humidity. This study comprehensively uncovers the transport dynamics of D5 in realistic indoor settings, which is helpful for human-related source characterization and control.

Using weak signals to predict spontaneous breathing trial success: a machine learning approach

BACKGROUND

Weaning from mechanical ventilation (MV) is a key phase in the management of intensive care unit (ICU) patient. According to the WEAN SAFE study, weaning from MV initiation is defined as the first attempt to separate a patient from the ventilator and the success is the absence of reintubation (or death) within 7 days of extubation. Mortality rates increase with the difficulty of weaning, reaching 38% for the most challenging cases. Predicting the success of weaning is difficult, due to the complexity of factors involved. The many biosignals that are measured in patients during ventilation may be considered "weak signals", a concept rarely used in medicine. The aim of this research is to investigate the performance of machine learning (ML) models based on biosignals to predict spontaneous breathing trial success (SBT) using biosignals and to identify the most important variables.

METHODS

This retrospective study used data from two centers (Nice University Hospital, Archet and Pasteur) collected from 232 intensive care patients who underwent MV (149 successfully and 83 unsuccessfully) between January, 2020 and April, 2023. The study focuses on the development of ML algorithms to predict the success of the spontaneous breathing trial based on a combination of discrete variables and biosignals (time series) recorded during the 24 h prior to the SBT.

RESULTS

For the models tested, the best results were obtained with Support Vector Classifier model: AUC-PR 0.963 (0.936-0.970, p = 0.001), AUROC 0.922 (0.871-0.940, p < 0.001).

CONCLUSIONS

We found that ML models are effective in predicting the success of SBT based on biosignals. Predicting weaning from mechanical ventilation thus appears to be a promising area for the application of AI, through the development of multidimensional models to analyze weak signals.

Leveraging Large Language Models for Predicting Postoperative Acute Kidney Injury in Elderly Patients

Objective: The objective of this work is to develop a framework based on large language models (LLMs) to predict postoperative acute kidney injury (AKI) outcomes in elderly patients. Impact Statement: Our study demonstrates that LLMs have the potential to address the issues of poor generalization and weak interpretability commonly encountered in disease prediction using traditional machine learning (ML) models. Introduction: AKI is a severe postoperative complication, especially in elderly patients with declining renal function. Current AKI prediction models rely on ML, but their lack of interpretability and generalizability limits clinical use. LLMs, with extensive pretraining and text generation capabilities, offer a new solution. Methods: We applied prompt engineering and knowledge distillation based on instruction fine-tuning to optimize LLMs for AKI prediction. The framework was tested on 2,649 samples from 2 private Chinese hospitals and one public South Korean dataset, which were divided into internal and external datasets. Results: The LLM framework showed robust external performance, with accuracy rates: commercial LLMs (internal: 63.73%, external: 68.73%), open-source LLMs (internal: 63.70%, external: 64.24%), and ML models (internal: 63.93%, external: 58.27%). LLMs also provided human-readable explanations for better clinical understanding. Conclusion: The proposed framework showcases the potential of LLMs to enhance generalization and interpretability in postoperative AKI prediction, paving the way for more robust and transparent predictive solutions in clinical settings.

Machine Learning-Based Pathomics Model Predicts Angiopoietin-2 Expression and Prognosis in Hepatocellular Carcinoma

Angiopoietin-2 (ANGPT2) shows promise as prognostic marker and therapeutic target in hepatocellular carcinoma (HCC). However, assessing ANGPT2 expression and prognostic potential using histopathology images viewed with naked eye is challenging. Herein, machine learning was employed to develop a pathomics model for analyzing histopathology images to predict ANGPT2 status. HCC cases obtained from The Cancer Genome Atlas (TCGA-HCC; n = 267) were randomly assigned to the training or testing set, and cases from a single center were employed as a validation set (n = 91). In the TCGA-HCC cohort, the group with high ANGPT2 expression had a significantly lower overall survival compared with the group with low ANGPT2. Histopathologic features in the training set were extracted, screened, and incorporated into a gradient-boosting machine model that generated a pathomics score, which successfully predicted ANGPT2 expression in the three data sets and showed remarkable risk stratification for overall survival in both the TCGA-HCC (P < 0.0001) and single-center cohorts (P = 0.001). Multivariate analysis suggested that the pathomics score could serve as a predictor of prognosis (P < 0.001). Bioinformatics analysis illustrated a distinction in tumor growth and development related gene-enriched pathways, vascular endothelial growth factor-related gene expression, and immune cell infiltration between high and low pathomics scores. This study indicates that the use of histopathology image features can enhance the prediction of molecular status and prognosis in HCC. The integration of image features with machine learning may improve prognosis prediction in HCC.

Research hotspots and frontiers of machine learning in renal medicine: a bibliometric and visual analysis from 2013 to 2024

BACKGROUND

The kidney, an essential organ of the human body, can suffer pathological damage that can potentially have serious adverse consequences on the human body and even affect life. Furthermore, the majority of kidney-induced illnesses are frequently not readily identifiable in their early stages. Once they have progressed to a more advanced stage, they impact the individual's quality of life and burden the family and broader society. In recent years, to solve this challenge well, the application of machine learning techniques in renal medicine has received much attention from researchers, and many results have been achieved in disease diagnosis and prediction. Nevertheless, studies that have conducted a comprehensive bibliometric analysis of the field have yet to be identified.

OBJECTIVES

This study employs bibliometric and visualization analyses to assess the progress of the application of machine learning in the renal field and to explore research trends and hotspots in the field.

METHODS

A search was conducted using the Web of Science Core Collection database, which yielded articles and review articles published from the database's inception to May 12, 2024. The data extracted from these articles and review articles were then analyzed. A bibliometric and visualization analysis was conducted using the VOSviewer, CiteSpace, and Bibliometric (R-Tool of R-Studio) software.

RESULTS

2,358 papers were retrieved and analyzed for this topic. From 2013 to 2024, the number of publications and the frequency of citations in the relevant research areas have exhibited a consistent and notable increase annually. The data set comprises 3734 institutions in 91 countries and territories, with 799 journals publishing the results. The total number of authors contributing to the data set is 14,396. China and the United States have the highest number of published papers, with 721 and 525 papers, respectively. Harvard University and the University of California System exert the most significant influence at the institutional level. Regarding authors, Cheungpasitporn, Wisit, and Thongprayoon Charat of the Mayo Clinic organization were the most prolific researchers, with 23 publications each. It is noteworthy that researcher Breiman I had the highest co-citation frequency. The journal with the most published papers was "Scientific Reports," while "PLoS One" had the highest co-citation frequency. In this field of machine learning applied to renal medicine, the article "A Clinically Applicable Approach to Continuous Prediction of Future Acute Kidney Injury" by Tomasev N et al., published in NATURE in 2019, emerged as the most influential article with the highest co-citation frequency. A keyword and reference co-occurrence analysis reveals that current research trends and frontiers in nephrology are the management of patients with renal disease, prediction and diagnosis of renal disease, imaging of renal disease, and development of personalized treatment plans for patients with renal disease. "Acute kidney injury," "chronic kidney disease," and "kidney tumors" are the most discussed diseases in medical research.

CONCLUSIONS

The field of renal medicine is witnessing a surge in the application of machine learning. On one hand, this study offers a novel perspective on applying machine learning techniques to kidney-related diseases based on bibliometric analysis. This analysis provides a comprehensive overview of the current status and emerging research areas in the field, as well as future trends and frontiers. Conversely, this study furnishes data on collaboration and exchange between countries, regions, institutions, journals, authors, keywords, and reference co-citations. This information can facilitate the advancement of future research endeavors, which aim to enhance interdisciplinary collaboration, optimize data sharing and quality, and further advance the application of machine learning in the renal field.

Risk factors and an interpretability tool of in-hospital mortality in critically ill patients with acute myocardial infarction

OBJECTIVE

We aim to develop and validate an interpretable machine-learning model that can provide critical information for the clinical treatment of critically ill patients with acute myocardial infarction (AMI).

METHODS

All data were extracted from the multi-centre database (training and internal validation cohorts: MIMIC-III/-IV, external validation cohort: eICU). After comparing different machine-learning models and several unbalanced data processing methods, the model with the best performance was selected. Lasso regression was used to build a compact model. Seven evaluation methods, PR and ROC curves were used to assess the model. The SHapley Additive exPlanations (SHAP) values were calculated to evaluate the feature's importance. The SHAP plots were adopted to explain and interpret the results. A web-based tool was developed to help application.

RESULTS

A total of 12,170 critically ill patients with AMI were included. The balance random forest (BRF) model had the best performance in predicting in-hospital mortality. The compact model did not differ from the full variable model in performance (AUC: 0.891 vs 0.885, P = 0.06). The external validation results also demonstrated the stability of the model (AUC: 0.784). All SHAP plots have shown the contribution ranking of all variables in the model, the relationship trend between variables and outcomes, and the interaction mode between variables. A web-based tool is constructed that can provide individualised risk stratification probabilities (https://github.com/huangtao36/BRF-web-tool).

CONCLUSION

We built the BRF model and the web-based tool by the model algorithm. The model effect has been verified externally. The tool can help clinical decision-making.

Unsupervised machine learning analysis to identify patterns of ICU medication use for fluid overload prediction

BACKGROUND

Fluid overload (FO) in the intensive care unit (ICU) is common, serious, and may be preventable. Intravenous medications (including administered volume) are a primary cause for FO but are challenging to evaluate as a FO predictor given the high frequency and time-dependency of their use and other factors affecting FO. We sought to employ unsupervised machine learning methods to uncover medication administration patterns correlating with FO.

METHODS

This retrospective cohort study included 927 adults admitted to an ICU for ≥72 h. FO was defined as a positive fluid balance ≥7% of admission body weight. After reviewing medication administration record data in 3-h periods, medication exposure was categorized into clusters using principal component analysis (PCA) and Restricted Boltzmann Machine (RBM). Medication regimens of patients with and without FO were compared within clusters to assess their temporal association with FO.

RESULTS

FO occurred in 127 (13.7%) of 927 included patients. Patients received a median (interquartile range) of 31(13-65) discrete intravenous medication administrations over the 72-h period. Across all 47,803 intravenous medication administrations, 10 unique medication clusters, containing 121 to 130 medications per cluster, were identified. The mean number of Cluster 7 medications administered was significantly greater in the FO cohort compared with patients without FO (25.6 vs.10.9, p < 0.0001). A total of 51 (40.2%) of 127 unique Cluster 7 medications were administered in more than five different 3-h periods during the 72-h study window. The most common Cluster 7 medications included continuous infusions, antibiotics, and sedatives/analgesics. Addition of Cluster 7 medications to an FO prediction model including the Acute Physiologic and Chronic Health Evaluation (APACHE) II score and receipt of diuretics improved model predictiveness from an Area Under the Receiver Operation Characteristic (AUROC) curve of 0.719 to 0.741 (p = 0.027).

CONCLUSIONS

Using machine learning approaches, a unique medication cluster was strongly associated with FO. Incorporation of this cluster improved the ability to predict FO compared to traditional prediction models. Integration of this approach into real-time clinical applications may improve early detection of FO to facilitate timely intervention.

The development and validation of a prediction model for post-AKI outcomes of pediatric inpatients

Background

Acute kidney injury (AKI) is common in hospitalized children. A post-AKI outcomes prediction model is important for the early detection of important clinical outcomes associated with AKI so that early management of pediatric AKI patients can be initiated.

Methods

Three retrospective cohorts were set up based on two pediatric hospitals in China, in which 8205 children suffered AKI during hospitalization. Two clinical outcomes were evaluated, i.e. hospital mortality and dialysis within 28 days after AKI occurrence. A Genetic Algorithm was used for feature selection, and a Random Forest model was built to predict clinical outcomes. Subsequently, a temporal validation set and an external validation set were used to evaluate the performance of the prediction model. Finally, the stratification ability of the prediction model for the risk of mortality was compared with a commonly used mortality risk score, the pediatric critical illness score (PCIS).

Results

The prediction model performed well for the prediction of hospital mortality with an area under the receiver operating curve (AUROC) of 0.854 [95% confidence interval (CI) 0.816-0.888], and the AUROC was >0.850 for both temporal and external validation. For the prediction of dialysis, the AUROC was 0.889 (95% CI 0.871-0.906). In addition, the AUROC of the prediction model for hospital mortality was superior to that of PCIS (P < .0001 in both temporal and external validation).

Conclusions

The new proposed post-AKI outcomes prediction model shows potential applicability in clinical settings.

A prognostic and predictive model based on deep learning to identify optimal candidates for intensity-modulated radiotherapy alone in patients with stage II nasopharyngeal carcinoma: A retrospective multicenter study

PURPOSE

To develop and validate a prognostic and predictive model integrating deep learning MRI features and clinical information in patients with stage II nasopharyngeal carcinoma (NPC) to identify patients with a low risk of progression for whom intensity-modulated radiotherapy (IMRT) alone is sufficient.

METHODS

This multicenter, retrospective study enrolled 999 patients with stage II NPC from two centers. 3DResNet was used to extract deep learning MRI features and eXtreme Gradient Boosting model was employed to integrate the pre-trained features and clinical information to obtain an overall score for each patient. Based on the optimal cutoff value of the overall score, patients were stratified into high- and low- risk groups. Model performance was evaluated using concordance indexes (C-indexes), the area under the curve (AUC) values and calibration tests. Survival curves were used to analyze the clinical benefits of additional chemotherapy in each risk group.

RESULTS

The combined model achieved a concordance index of 0.789 (95 % confidence interval [CI] 0.787-0.791), 0.768 (95 % CI 0.764-0.771), and 0.804 (95 % CI 0.801-0.807) for the training, internal validation, and external test cohorts, respectively, demonstrating a statistically significant improvement compared to the MRI model, T Stage, and N Stage. An overall score of < 0.405 in patients was significantly associated with a low risk of progression. In the low-risk group, patients treated with IMRT alone had comparable or even superior progression-free survival (PFS) compared to those who received additional chemotherapy.

CONCLUSION

The model demonstrated a satisfactory prognostic and predictive performance for PFS. Patients with stage II NPC were stratified into different risk groups to help identify optimal candidates who could benefit from IMRT alone.

[Predicting Intensive Care Unit Mortality in Patients With Heart Failure Combined With Acute Kidney Injury Using an Interpretable Machine Learning Model: A Retrospective Cohort Study]

Objective

Heart failure (HF) complicated by acute kidney injury (AKI) significantly impacts patient outcomes, and it is crucial to make early predictions of short-term mortality. This study is focused on developing an interpretable machine learning model to enhance early prediction accuracy in such clinical scenarios.

Methods

This retrospective cohort study utilized data from the Medical Information Mart for Intensive Care Ⅳ (MIMIC-Ⅳ, version 2.0) database. Data from the first 24 hours after admission to the ICU were extracted and divided into a training set (70%) and a validation set (30%). We utilized the SHapley Additive exPlanation (SHAP) method to interpret the workings of an extreme gradient boosting (XGBoost) model and identify key prognostic factors. The XGBoost model's predictive ability was evaluated against three other machine learning models using the area under the curve (AUC) metric, and its interpretation was enhanced using the SHAP method.

Results

The study included 8028 patients with HF complicated by AKI. The XGBoost model outperformed the other models, achieving an AUC of 0.93 (95% confidence interval [CI]: 0.78-0.94; accuracy = 0.89), while neural network model showed the worst performance (AUC = 0.79, 95% CI: 0.77-0.82; accuracy = 0.82). Decision curve analysis showed the superior net benefit of the XGBoost model within the 9% to 60% threshold probabilities. SHAP analysis was performed to identify the top 20 predictors, with age (mean SHAP value 1.29) and Glasgow Coma Scale score (mean SHAP value 1.24) emerging as significant factors.

Conclusions

Our interpretable model offers an enhanced ability to predict mortality risk in HF patients with AKI in ICUs. This model can be used to assist in formulating effective treatment plans and optimizing resource allocation.

Prediction of Distant Metastasis of Renal Cell Carcinoma Based on Interpretable Machine Learning: A Multicenter Retrospective Study

Background 

The traditional tool for predicting distant metastasis in renal cell carcinoma (RCC) is still insufficient. We aimed to establish an interpretable machine learning model for predicting distant metastasis in RCC patients.

Methods

We involved a population-based cohort of 121433 patients (mean age = 63 years; 63.58% men) diagnosed with RCC between 2004 and 2015 from the Surveillance, Epidemiology, and End Results (SEER) database. The lightGBM algorithm was used to develop prediction model and assessed by the area under the receiver-operating characteristic curve (AUC), accuracy, sensitivity, and specificity. The LightGBM model was then externally validated in 36395 RCC patients enrolled from the SEER database between 2016 and 2018. Shapley Additive exPlanations (SHAP) method was applied to provide insights into the model's outcome or prediction.

Results

Of 121433 patients involved in the study cohort, 10730 (8.84%) had distant metastasis. The LightGBM model showed good performance in the internal validation set (AUC: 0.955, 95% CI: 0.951-0.959) and temporal external validation sets (0.963, 95% CI: 0.959-0.967; 0.961, 95% CI: 0.954-0.966). Performance for the prediction model was also well performed in different sub-cohort stratified by age, gender, and ethnicity. The calibration curve indicated that the predicted values are highly consistent with the actual observed values. SHAP plots demonstrated that chemotherapy was the most vital variable for prediction of distant metastasis of RCC patients.

Conclusion

We developed an interpretable machine learning model that is capable of accurately predicting the risk of distant metastasis of RCC patients. The presented model could help identify high-risk patients who require additional treatment strategies and follow-up regimens.

Major Adverse Kidney Events in Hospitalized Older Patients With Acute Kidney Injury: Machine Learning-Based Model Development and Validation Study

BACKGROUND

Acute kidney injury (AKI) is a common complication in hospitalized older patients, associated with increased morbidity, mortality, and health care costs. Major adverse kidney events within 30 days (MAKE30), a composite of death, new renal replacement therapy, or persistent renal dysfunction, has been recommended as a patient-centered endpoint for clinical trials involving AKI.

OBJECTIVE

This study aimed to develop and validate a machine learning-based model to predict MAKE30 in hospitalized older patients with AKI.

METHODS

A total of 4266 older patients (aged ≥ 65 years) with AKI admitted to the Second Xiangya Hospital of Central South University from January 1, 2015, to December 31, 2020, were included and randomly divided into a training set and an internal test set in a ratio of 7:3. An additional cohort of 11,864 eligible patients from the Medical Information Mart for Intensive Care Ⅳ database served as an external test set. The Boruta algorithm was used to select the most important predictor variables from 53 candidate variables. The eXtreme Gradient Boosting algorithm was applied to establish a prediction model for MAKE30. Model discrimination was evaluated by the area under the receiver operating characteristic curve (AUROC). The SHapley Additive exPlanations method was used to interpret model predictions.

RESULTS

The overall incidence of MAKE30 in the 2 study cohorts was 28.3% (95% CI 26.9%-29.7%) and 26.7% (95% CI 25.9%-27.5%), respectively. The prediction model for MAKE30 exhibited adequate predictive performance, with an AUROC of 0.868 (95% CI 0.852-0.881) in the training set and 0.823 (95% CI 0.798-0.846) in the internal test set. Its simplified version achieved an AUROC of 0.744 (95% CI 0.735-0.754) in the external test set. The SHapley Additive exPlanations method showed that the use of vasopressors, mechanical ventilation, blood urea nitrogen level, red blood cell distribution width-coefficient of variation, and serum albumin level were closely associated with MAKE30.

CONCLUSIONS

An interpretable eXtreme Gradient Boosting model was developed and validated to predict MAKE30, which provides opportunities for risk stratification, clinical decision-making, and the conduct of clinical trials involving AKI.

TRIAL REGISTRATION

Chinese Clinical Trial Registry ChiCTR2200061610; https://tinyurl.com/3smf9nuw.

Prediction of sepsis among patients with major trauma using artificial intelligence: a multicenter validated cohort study

BACKGROUND

Sepsis remains a significant challenge in patients with major trauma in the ICU. Early detection and treatment are crucial for improving outcomes and reducing mortality rates. Nonetheless, clinical tools for predicting sepsis among patients with major trauma are limited. This study aimed to develop and validate an artificial intelligence (AI) platform for predicting the risk of sepsis among patients with major trauma.

PATIENTS AND METHODS

This study involved 961 patients, with a prospective analysis of data from 244 patients with major trauma at our hospital and a retrospective analysis of data from 717 patients extracted from a database in the United States. The patients from our hospital constituted the model development cohort, and the patients from the database constituted the external validation cohort. The patients in the model development cohort were randomly divided into a training cohort and an internal validation cohort at a ratio of 8:2. The machine-learning algorithms used to train models included logistic regression, decision tree, extreme gradient boosting machine (eXGBM), neural network (NN), random forest, and light gradient boosting machine (LightGBM).

RESULTS

The incidence of sepsis for the model development cohort was 43.44%. Twelve predictors, including gender, abdominal trauma, open trauma, red blood cell count, heart rate, respiratory rate, injury severity score, sequential organ failure assessment score, Glasgow coma scale, smoking, total protein concentrations, and hematocrit, were used as features in the final model. Internal validation showed that the NN model had the highest area under the curve (AUC) of 0.932 (95% CI: 0.917-0.948), followed by the LightGBM and eXGBM models with AUCs of 0.913 (95% CI: 0.883-0.930) and 0.912 (95% CI: 0.880-0.935), respectively. In the external validation cohort, the eXGBM model (AUC: 0.891, 95% CI: 0.866-0.914) had the highest AUC value, followed by the LightGBM model (AUC: 0.886, 95% CI: 0.860-0.906), and the AUC value of the NN model was only 0.787 (95% CI: 0.751-0.829). Considering the predictive performance for both the internal and external validation cohorts, the LightGBM model had the highest score of 82, followed by the eXGBM (81) and NN (76) models. Thus, the LightGBM has emerged as the optimal model, and it was deployed online as an AI application.

CONCLUSIONS

This study develops and validates an AI application to effectively assess the susceptibility of patients with major trauma to sepsis. The AI application equips healthcare professionals with a valuable tool to promptly identify individuals at high risk of developing sepsis. This will facilitate clinical decision-making and enable early intervention.

The relationship between low-carbohydrate diet score, dietary macronutrient intake, and rheumatoid arthritis: results from NHANES 2011-2016

BACKGROUND

This study sought to determine if dietary macronutrient consumption and the low-carbohydrate diet (LCD) score were linked to rheumatoid arthritis (RA).

METHODS

Participants ≥ 20 years were analyzed from the National Health and Nutrition Examination Survey (NHANES) 2011-2016. LCD score was calculated by summing the 11 quantiles values of the percentages of energy derived from carbohydrate, protein, and fat. Weighted logistic regression, eXtreme Gradient Boosting (XGBoost), and Light Gradient Boosting Machine (LightGBM) models were used to explore the relationship between LCD score, dietary macronutrient intake, and RA. Propensity score matching (PSM) were applied for sensitivity analysis.

RESULTS

Ultimately, 8118 participants (RA: 499, without RA: 7619) were analyzed. After fully adjusting for confounders, a negative association was found between the LCD score and the presence of RA [OR (95% CI), 0.97 (0.96, 0.99)]. A higher LCD score was also negatively associated with a lower likelihood of RA based on a categorical model. Among macronutrients, participants in the third and fourth quartiles had significantly increased odds of RA compared with the lowest carbohydrate intake. Regarding protein intake, individuals in the highest quartile of percentage of energy from protein had a 46% lower presence of RA compared with the lowest reference group. The relative importance of the LCD score on RA was determined based on XGBoost and LightGBM models. Moreover, the association between the LCD score, dietary macronutrient intake, and RA presence remained substantial after PSM.

CONCLUSIONS

LCD score was negatively associated with odds of RA in US adults. Moreover, a correlation was found between a lower likelihood of RA and high protein, and low carbohydrate consumption. Key Points • A significant negative association was found between LCD score and RA presence. • Machine learning models revealed the LCD score was a significant predictor of the presence of RA. • Low carbohydrate intake and high protein intake were correlated with a lower odds of RA.

An artificial intelligence application to predict prolonged dependence on mechanical ventilation among patients with critical orthopaedic trauma: an establishment and validation study

BACKGROUND

Prolonged dependence on mechanical ventilation is a common occurrence in clinical ICU patients and presents significant challenges for patient care and resource allocation. Predicting prolonged dependence on mechanical ventilation is crucial for improving patient outcomes, preventing ventilator-associated complications, and guiding targeted clinical interventions. However, specific tools for predicting prolonged mechanical ventilation among ICU patients, particularly those with critical orthopaedic trauma, are currently lacking. The purpose of the study was to establish and validate an artificial intelligence (AI) platform to assess the prolonged dependence on mechanical ventilation among patients with critical orthopaedic trauma.

METHODS

This study analyzed 1400 patients with critical orthopaedic trauma who received mechanical ventilation, and the prolonged dependence on mechanical ventilation was defined as not weaning from mechanical ventilation for ≧ 7 days. Patients were randomly classified into a training cohort and a validation cohort based on the ratio of 8:2. Patients in the training cohort were used to establish models using machine learning techniques, including logistic regression (LR), extreme gradient boosting machine (eXGBM), decision tree (DT), random forest (RF), support vector machine (SVM), and light gradient boosting machine (LightGBM), whereas patients in the validation cohort were used to validate these models. The prediction performance of these models was evaluated using discrimination and calibration. A scoring system was used to comprehensively assess and compare the prediction performance of the models, based on ten evaluation metrics. External validation of the model was performed in 122 patients with critical orthopaedic trauma from a university teaching hospital. Furthermore, the optimal model was deployed as an AI calculator, which was accessible online, to assess the risk of prolonged dependence on mechanical ventilation.

RESULTS

Among the developed models, the eXGBM model had the highest score of 50, followed by the LightGBM model (48) and the RF model (37). In detail, the eXGBM model outperformed other models in terms of recall (0.892), Brier score (0.088), log loss (0.291), and calibration slope (0.999), and the model was the second best in terms of area under the curve value (0.949, 95%: 0.933-0.961), accuracy (0.871), F1 score (0.873), and discrimination slope (0.647). The SHAP revealed that the most important five features were respiratory rate, lower limb fracture, glucose, PaO2, and PaCO2. External validation of the eXGBM model also demonstrated favorable prediction performance, with an AUC value of 0.893 (95%CI: 0.819-0.967). The eXGBM model was successfully deployed as an AI platform, which was at https://prolongedmechanicalventilation-lqsfm6ecky6dpd4ybkvohu.streamlit.app/ . By simply clicking the link and inputting features, users were able to obtain the risk of experiencing prolonged dependence on mechanical ventilation for individuals. Based on the risk of prolonged dependence on mechanical ventilation, patients were stratified into the high-risk or the low-risk groups, and corresponding therapeutic interventions were recommended, accordingly.

CONCLUSIONS

The AI model shows potential as a valuable tool for stratifying patients with a high risk of prolonged dependence on mechanical ventilation. The AI model may offer a promising approach for optimizing patient care and resource allocation in critical care settings.

CLINICAL TRIAL NUMBER

Not applicable.

XGBoost-SHAP-based interpretable diagnostic framework for knee osteoarthritis: a population-based retrospective cohort study

OBJECTIVE

To use routine demographic and clinical data to develop an interpretable individual-level machine learning (ML) model to diagnose knee osteoarthritis (KOA) and to identify highly ranked features.

METHODS

In this retrospective, population-based cohort study, anonymized questionnaire data was retrieved from the Wu Chuan KOA Study, Inner Mongolia, China. After feature selections, participants were divided in a 7:3 ratio into training and test sets. Class balancing was applied to the training set for data augmentation. Four ML classifiers were compared by cross-validation within the training set and their performance was further analyzed with an unseen test set. Classifications were evaluated using sensitivity, specificity, positive predictive value, negative predictive value, accuracy, area under the curve(AUC), G-means, and F1 scores. The best model was explained using Shapley values to extract highly ranked features.

RESULTS

A total of 1188 participants were investigated in this study, among whom 26.3% were diagnosed with KOA. Comparatively, XGBoost with Boruta exhibited the highest classification performance among the four models, with an AUC of 0.758, G-means of 0.800, and F1 scores of 0.703. The SHAP method reveals the top 17 features of KOA according to the importance ranking, and the average of the experience of joint pain was recognized as the most important features.

CONCLUSIONS

Our study highlights the usefulness of machine learning in unveiling important factors that influence the diagnosis of KOA to guide new prevention strategies. Further work is needed to validate this approach.

Relevance of combined influence of nutritional and inflammatory status on non-alcoholic fatty liver disease and advanced fibrosis: A mediation analysis of lipid biomarkers

BACKGROUND AND AIM

This study aimed to investigate the relationship between advanced lung cancer inflammation index (ALI) and non-alcoholic fatty liver disease (NAFLD) and advanced liver fibrosis (AF).

METHODS

A total of 5642 individuals from the National Health and Nutrition Examination Survey (NHANES) between 2017 and 2020 were examined. Limited cubic spline regression model, and weighted logistic regression were employed to determine if ALI levels were related to the prevalence of NAFLD and AF. Additionally, a mediating analysis was conducted to investigate the role of lipid biomarkers, such as total cholesterol (TC) and high-density lipoprotein cholesterol (HDL-C), in the effects of ALI on the prevalence of NAFLD and AF.

RESULTS

After adjusting for potential confounders, a significant positive association was found between ALI with NAFLD and AF prevalence. Compared with those in ALI Tertile 1, participants in Tertile 3 had higher odds of NAFLD prevalence (odds ratio [OR]: 3.16; 95% confidence interval [CI]: 2.52-3.97) and AF (OR: 3.17; 95% CI: 2.30-4.36). Participants in both Tertile 2 and Tertile 3 had lower odds of developing AF (P for trend = 0.005). Moreover, we discovered a nonlinear association between ALI and NAFLD. An inflection point of 74.25 for NAFLD was identified through a two-segment linear regression model. Moreover, TC and HDL-C levels mediated the association between ALI and NAFLD by 10.2% and 4.2%, respectively (both P < 0.001).

CONCLUSION

Our findings suggest that higher ALI levels are positively associated with an increased prevalence of NAFLD and AF, partly mediated by lipid biomarkers.

Effects of 24-hour urine-output trajectories on the risk of acute kidney injury in critically ill patients with cirrhosis: a retrospective cohort analysis

BACKGROUND

Acute kidney injury (AKI) is one of the most common complications for critically ill patients with cirrhosis, but it has remained unclear whether urine output fluctuations are associated with the risk of AKI in such patients. Thus, we explored the influence of 24-h urine-output trajectory on AKI in patients with cirrhosis through latent category trajectory modeling.

MATERIALS AND METHODS

This retrospective cohort study examined patients with cirrhosis using the MIMIC-IV database. Changes in the trajectories of urine output within 24 h after admission to the intensive care unit (ICU) were categorized using latent category trajectory modeling. The outcome examined was the occurrence of AKI during ICU hospitalization. The risk of AKI in patients with different trajectory classes was explored using the cumulative incidence function (CIF) and the Fine-Gray model with the sub-distribution hazard ratio (SHR) and the 95% confidence interval (CI) as size effects.

RESULTS

The study included 3,562 critically ill patients with cirrhosis, of which 2,467 (69.26%) developed AKI during ICU hospitalization. The 24-h urine-output trajectories were split into five classes (Classes 1-5). The CIF curves demonstrated that patients with continuously low urine output (Class 2), a rapid decline in urine output after initially high levels (Class 3), and urine output that decreased slowly and then stabilized at a lower level (Class 4) were at higher risk for AKI than those with consistently moderate urine output (Class 1). After fully adjusting for various confounders, Classes 2, 3, and 4 were associated with a higher risk of AKI compared with Class 1, and the respective SHRs (95% CIs) were 2.56 (1.87-3.51), 1.86 (1.34-2.59), and 1.83 1.29-2.59).

CONCLUSIONS

The 24-h urine-output trajectory is significantly associated with the risk of AKI in critically ill patients with cirrhosis. More attention should be paid to the dynamic nature of urine-output changes over time, which may help guide early intervention and improve patients' prognoses.

Is artificial intelligence prepared for the 24-h shifts in the ICU?

Integrating machine learning (ML) into intensive care units (ICUs) can significantly enhance patient care and operational efficiency. ML algorithms can analyze vast amounts of data from electronic health records, physiological monitoring systems, and other medical devices, providing real-time insights and predictive analytics to assist clinicians in decision-making. ML has shown promising results in predictive modeling for patient outcomes, early detection of sepsis, optimizing ventilator settings, and resource allocation. For instance, predictive algorithms have demonstrated high accuracy in forecasting patient deterioration, enabling timely interventions and reducing mortality rates. Despite these advancements, challenges such as data heterogeneity, integration with existing clinical workflows, and the need for transparency and interpretability of ML models persist. The deployment of ML in ICUs also raises ethical and legal considerations regarding patient privacy and the potential for algorithmic biases. For clinicians interested in the early embracing of AI-driven changes in clinical practice, in this review, we discuss the challenges of integrating AI and ML tools in the ICU environment in several steps and issues: (1) Main categories of ML algorithms; (2) From data enabling to ML development; (3) Decision-support systems that will allow patient stratification, accelerating the foresight of adequate individual care; (4) Improving patient outcomes and healthcare efficiency, with positive society and research implications; (5) Risks and barriers to AI-ML application to the healthcare system, including transparency, privacy, and ethical concerns.

Predictive models of sepsis-associated acute kidney injury based on machine learning: a scoping review

BACKGROUND

With the development of artificial intelligence, the application of machine learning to develop predictive models for sepsis-associated acute kidney injury has made potential breakthroughs in early identification, grading, diagnosis, and prognosis determination.

METHODS

Here, we conducted a systematic search of the PubMed, Cochrane Library, Embase (Ovid), Web of Science, and Scopus databases on April 28, 2023, and screened relevant literature. Then, we comprehensively extracted relevant data related to machine learning algorithms, predictors, and predicted objectives. We subsequently performed a critical evaluation of research quality, data aggregation, and analyses.

RESULTS

We screened 25 studies on predictive models for sepsis-associated acute kidney injury from a total of originally identified 2898 studies. The most commonly used machine learning algorithm is traditional logistic regression, followed by eXtreme gradient boosting. We categorized these predictive models into early identification models (60%), prognostic prediction models (32%), and subtype identification models (8%) according to their predictive purpose. The five most commonly used predictors were serum creatinine levels, lactate levels, age, blood urea nitrogen concentration, and diabetes mellitus. In addition, a single data source, insufficient assessment of clinical utility, lack of model bias assessment, and hyperparameter adjustment may be the main reasons for the low quality of the current research.

CONCLUSIONS

However, studies on the nondeath prognostic outcomes, the long-term clinical outcomes, and the subtype identification models are insufficient. Additionally, the poor quality of the research and the insufficient practicality of the model are problems that need to be addressed urgently.

Association of dementia with the 28-day mortality of sepsis: an observational and Mendelian randomization study

Background

Observational research suggests that individuals with dementia who have sepsis face a higher likelihood of death. However, whether there is a causal relationship between the two remains unknown.

Methods

We analyzed data from patients diagnosed with sepsis and dementia, extracted from the Medical Information Mart for Intensive Care IV (MIMIC-IV) database. To examine the correlation between dementia and 28-day mortality in sepsis, we utilized Cox proportional hazards models. Following this, we performed a Mendelian randomization (MR) study with two samples to investigate the potential link between dementia and mortality within 28 days in sepsis.

Results

This study included a total of 22,189 patients diagnosed with sepsis, among whom 1,346 cases (6.1%) had dementia. After adjusting for multiple confounding factors, dementia was associated with an increased risk of 28-day mortality in sepsis (HR = 1.25, 95% CI = 1.12-1.39, p < 0.001). In the MR analysis, there appeared to be a causal relationship between genetically predicted dementia with Lewy bodies (DLB) (OR = 1.093, 95% CI = 1.016-1.177, p = 0.017) and 28-day mortality in sepsis. However, there was no evidence of causality between any dementia (OR = 1.063, 95% CI = 0.91-1.243, p = 0.437), Alzheimer's disease (AD) (OR = 1.126, 95% CI = 0.976-1.299, p = 0.103), vascular dementia (VD) (OR = 1.008, 95% CI = 0.93-1.091, p = 0.844), and the risk of 28-day mortality in sepsis.

Conclusion

In the observational analysis, dementia was associated with an increased risk of 28-day mortality in septic patients. However, in the MR analysis, only DLB was associated with increased 28-day mortality in septic patients, with no observed correlation for other dementia subtypes.

Prediction of Early Mortality in Esophageal Cancer Patients with Liver Metastasis Using Machine Learning Approaches

Patients with esophageal cancer liver metastasis face a high risk of early mortality, making accurate prediction crucial for guiding clinical decisions. However, effective predictive tools are currently limited. In this study, we used clinicopathological data from 1897 patients diagnosed with esophageal cancer liver metastasis between 2010 and 2020, which were sourced from the SEER database. Prognostic factors were identified using univariate and multivariate logistic regression, and seven machine learning models, including extreme gradient boosting (XGBoost) and support vector machine (SVM), were developed to predict early mortality. The models were evaluated using Receiver Operating Characteristic (ROC) curves, calibration curves, decision curve analysis (DCA), and F1 scores. Results showed that 40% of patients experienced all-cause early mortality and 38% had cancer-specific early mortality. Key predictors of early mortality included age, location, chemotherapy, and lung metastasis. Among the models, XGBoost performed best in predicting all-cause early mortality, while SVM excelled in predicting cancer-specific early mortality. These findings demonstrate that machine learning models, particularly XGBoost and SVM, can serve as valuable tools for predicting early mortality in patients with esophageal cancer liver metastasis, aiding clinical decision making.

Artificial intelligence in healthcare: a mastery

There is a vast development of artificial intelligence (AI) in recent years. Computational technology, digitized data collection and enormous advancement in this field have allowed AI applications to penetrate the core human area of specialization. In this review article, we describe current progress achieved in the AI field highlighting constraints on smooth development in the field of medical AI sector, with discussion of its implementation in healthcare from a commercial, regulatory and sociological standpoint. Utilizing sizable multidimensional biological datasets that contain individual heterogeneity in genomes, functionality and milieu, precision medicine strives to create and optimize approaches for diagnosis, treatment methods and assessment. With the arise of complexity and expansion of data in the health-care industry, AI can be applied more frequently. The main application categories include indications for diagnosis and therapy, patient involvement and commitment and administrative tasks. There has recently been a sharp rise in interest in medical AI applications due to developments in AI software and technology, particularly in deep learning algorithms and in artificial neural network (ANN). In this overview, we enlisted the major categories of issues that AI systems are ideally equipped to resolve followed by clinical diagnostic tasks. It also includes a discussion of the future potential of AI, particularly for risk prediction in complex diseases, and the difficulties, constraints and biases that must be meticulously addressed for the effective delivery of AI in the health-care sector.

Interpretable machine learning model for new-onset atrial fibrillation prediction in critically ill patients: a multi-center study

BACKGROUND

New-onset atrial fibrillation (NOAF) is the most common arrhythmia in critically ill patients admitted to intensive care and is associated with poor prognosis and disease burden. Identifying high-risk individuals early is crucial. This study aims to create and validate a NOAF prediction model for critically ill patients using machine learning (ML).

METHODS

The data came from two non-overlapping datasets from the Medical Information Mart for Intensive Care (MIMIC), with MIMIC-IV used for training and subset of MIMIC-III used as external validation. LASSO regression was used for feature selection. Eight ML algorithms were employed to construct the prediction model. Model performance was evaluated based on identification, calibration, and clinical application. The SHapley Additive exPlanations (SHAP) method was used for visualizing model characteristics and individual case predictions.

RESULTS

Among 16,528 MIMIC-IV patients, 1520 (9.2%) developed AF post-ICU admission. A model with 23 variables was built, with XGBoost performing best, achieving an AUC of 0.891 (0.873-0.888) in validation and 0.769 (0.756-0.782) in external validation. Key predictors included age, mechanical ventilation, urine output, sepsis, blood urea nitrogen, percutaneous arterial oxygen saturation, continuous renal replacement therapy and weight. A risk probability greater than 0.6 was defined as high risk. A friendly user interface had been developed for clinician use.

CONCLUSION

We developed a ML model to predict the risk of NOAF in critically ill patients without cardiac surgery and validated its potential as a clinically reliable tool. SHAP improves the interpretability of the model, enables clinicians to better understand the causes of NOAF, helps clinicians to prevent it in advance and improves patient outcomes.

Quantitative texture analysis using machine learning for predicting interpretable pulmonary perfusion from non-contrast computed tomography in pulmonary embolism patients

BACKGROUND

Pulmonary embolism (PE) is life-threatening and requires timely and accurate diagnosis, yet current imaging methods, like computed tomography pulmonary angiography, present limitations, particularly for patients with contraindications to iodinated contrast agents. We aimed to develop a quantitative texture analysis pipeline using machine learning (ML) based on non-contrast thoracic computed tomography (CT) scans to discover intensity and textural features correlated with regional lung perfusion (Q) physiology and pathology and synthesize voxel-wise Q surrogates to assist in PE diagnosis.

METHODS

We retrospectively collected 99mTc-labeled macroaggregated albumin Q-SPECT/CT scans from patients suspected of PE, including an internal dataset of 76 patients (64 for training, 12 for testing) and an external testing dataset of 49 patients. Quantitative CT features were extracted from segmented lung subregions and underwent a two-stage feature selection pipeline. The prior-knowledge-driven preselection stage screened for robust and non-redundant perfusion-correlated features, while the data-driven selection stage further filtered features by fitting ML models for classification. The final classification model, trained with the highest-performing PE-associated feature combination, was evaluated in the testing cohorts based on the Area Under the Curve (AUC) for subregion-level predictability. The voxel-wise Q surrogate was then synthesized using the final selected feature maps (FMs) and model score maps (MSMs) to investigate spatial distributions. The Spearman correlation coefficient (SCC) and Dice similarity coefficient (DSC) were used to assess the spatial consistency between FMs or MSMs and Q-SPECT scans.

RESULTS

The optimal model performance achieved an AUC of 0.863 during internal testing and 0.828 on the external testing cohort. The model identified a combination containing 14 intensity and textural features that were non-redundant, robust, and capable of distinguishing between high- and low-functional lung regions. Spatial consistency assessment in the internal testing cohort showed moderate-to-high agreement between MSMs and reference Q-SPECT scans, with median SCC of 0.66, median DSCs of 0.86 and 0.64 for high- and low-functional regions, respectively.

CONCLUSIONS

This study validated the feasibility of using quantitative texture analysis and a data-driven ML pipeline to generate voxel-wise lung perfusion surrogates, providing a radiation-free, widely accessible alternative to functional lung imaging in managing pulmonary vascular diseases.

CLINICAL TRIAL NUMBER

Not applicable.

Machine learning models for mortality prediction in critically ill patients with acute pancreatitis-associated acute kidney injury

Background

The occurrence of acute kidney injury (AKI) was associated with an increased mortality rate among acute pancreatitis (AP) patients, indicating the importance of accurately predicting the mortality rate of critically ill patients with acute pancreatitis-associated acute kidney injury (AP-AKI) at an early stage. This study aimed to develop and validate machine learning-based predictive models for in-hospital mortality rate in critically ill patients with AP-AKI by comparing their performance with the traditional logistic regression (LR) model.

Methods

This study used data from three clinical databases. The predictors were identified by the Recursive Feature Elimination algorithm. The LR and two machine learning models-random forest (RF) and eXtreme Gradient Boosting (XGBoost)-were developed using 10-fold cross-validation to predict in-hospital mortality rate in AP-AKI patients.

Results

A total of 1089 patients from the Medical Information Mart for Intensive Care-IV (MIMIC-IV) and eICU Collaborative Research Database (eICU-CRD) were included in the training set and 176 patients from Xiangya Hospital were included in the external validation set. The in-hospital mortality rates of the training and external validation sets were 13.77% and 54.55%, respectively. Compared with the area under the curve (AUC) values of the LR model and the RF model, the AUC value of the XGBoost model {0.941 [95% confidence interval (CI) 0.931-0.952]} was significantly higher (both P < .001) and the XGBoost model had the smallest Brier score of 0.039 in the training set. In the external validation set, the performance of the XGBoost model was acceptable, with an AUC value of 0.724 (95% CI 0.648-0.800). However, it did not differ significantly from the LR and RF models.

Conclusions

The XGBoost model was superior to the LR and RF models in terms of both the discrimination and calibration in the training set. Whether the findings can be generalized needs to be further validated.

Machine learning for the prediction of delirium in elderly intensive care unit patients

PURPOSE

This study aims to develop and validate a prediction model for delirium in elderly ICU patients and help clinicians identify high-risk patients at the early stage.

METHODS

Patients admitted to ICU for at least 24 h and using the Confusion Assessment Method for the Intensive Care Unit (CAM-ICU) in the Medical Information Mart for Intensive Care-IV (MIMIC-IV) database (76,943 ICU stays from 2008 to 2019) were considered. Patients with a positive delirium test in the first 24 h and under 65 years of age were excluded. Two prediction models, machine learning extreme gradient boosting (XGBoost) and logistic regression (LR) model, were developed and validated to predict the onset of delirium.

RESULTS

Of the 18,760 patients included in the analysis, 3463(18.5%) were delirium positive. A total of 22 significant predictors were selected by LASSO regression. The XGBoost model demonstrated superior performance over the LR model, with the Area Under the Receiver Operating Characteristic (AUC) values of 0.853 (95% confidence interval [CI] 0.846-0.861) and 0.831 (95% CI 0.815-0.847) in the training and testing datasets, respectively. Moreover, the XGBoost model outperformed the LR model in both calibration and clinical utility. The top five predictors associated with the onset of delirium were sequential organ failure assessment (SOFA), infection, minimum platelets, maximum systolic blood pressure (SBP), and maximum temperature.

CONCLUSION

The XGBoost model demonstrated good predictive performance for delirium among elderly ICU patients, thus assisting clinicians in identifying high-risk patients at the early stage and implementing targeted interventions to improve outcome.

Association between Healthy Eating Index-2015 total and component food scores with reproductive lifespan among postmenopausal women: a population-based study from NHANES 2005-2016

BACKGROUND

Prior research has demonstrated that nutrition plays a crucial role in the establishment and maturation of the reproductive lifetime. Although the specific dietary components involved in preventing or postponing the reproductive lifespan are still unknown, a healthy diet can affect the reproductive lifespan. Here, the study aimed to explore the relationship between reproductive lifespan and diet quality by utilizing the Healthy Eating Index-2015 (HEI-2015).

METHODS

In this study, a total of 2761 postmenopausal women were selected from the National Health and Nutrition Examination Survey (NHANES) from 2005 to 2016. Diet quality was determined using HEI-2015 based on two 24-hour dietary recalls. Reproductive lifespan was defined as the number of years between self-reported age at menarche and menopause. Weighted linear regression and eXtreme Gradient Boosting (XGBoost) models were used to analyze the relationship between HEI-2015 and reproductive lifespan. Subsequently, the impact of various components of HEI-2015 on reproductive lifespan was assessed through weighted quantile sum (WQS) regression models.

RESULTS

Among 2761 postmenopausal women, the mean age was 63.7 years. 41.5% were obese, and 49.7% were non-Hispanic white. After adjusting for sociodemographic characteristics, lifestyle factors, and medical history, individuals in the highest tertile of HEI-2015 had a 4.81% (95% CI: 1.82-7.79%) longer reproductive time life. Higher HEI-2015 was also significantly associated with a higher likelihood of late menopause (p for trend < 0.05). Based on XGBoost models, the relative importance of HEI-2015 on reproductive lifespan was determined. Whole fruits, whole grains, total protein foods, and greens and beans significantly contributed to extending age at menopause and reproductive time life in the HEI-2015. The weights of the WQS index for age at menopause were 27.1%, 23.2%, 10.1%, and 7.5% respectively, while the weights of the WQS index for reproductive time life were 30.2%, 14.6%, 9.3%, and 14.0% respectively.

CONCLUSION

There is a positive association between the HEI-2015 and reproductive lifespan. This underscores the significance of enhancing adherence to healthy dietary patterns in preventing a shorter reproductive lifespan.

Artificial intelligence algorithms permits rapid acute kidney injury risk classification of patients with acute myocardial infarction

Objective

This study aimed to develop and validate several artificial intelligence (AI) models to identify acute myocardial infarction (AMI) patients at an increased risk of acute kidney injury (AKI) during hospitalization.

Methods

Included were patients diagnosed with AMI from the Medical Information Mart for Intensive Care (MIMIC) III and IV databases. Two cohorts of AMI patients from Changzhou Second People's Hospital and Xuzhou Center Hospital were used for external validation of the models. Patients' demographics, vital signs, clinical characteristics, laboratory results, and therapeutic measures were extracted. Totally, 12 AI models were developed. The area under the receiver operating characteristic curve (AUC) were calculated and compared.

Results

AKI occurred during hospitalization in 1098 (28.3 %) of the 3882 final enrolled patients, split into training (3105) and test (777) sets randomly. Among them, Random Forest (RF), C5.0 and Bagged CART models outperformed the other models in both the training and test sets. The AUCs for the test set were 0.754, 0.734 and 0.730, respectively. The incidence of AKI was 9.8 % and 9.5 % in 2202 patients in the Changzhou cohort and 807 patients in the Xuzhou cohort with AMI, respectively. The AUCs for patients in the Changzhou cohort were RF, 0.761; C5.0, 0.733; and bagged CART, 0.725, respectively, and Xuzhou cohort were RF, 0.799; C5.0, 0.808; and bagged CART, 0.784, respectively.

Conclusion

Several machines learning-based prediction models for AKI after AMI were developed and validated. The RF, C5.0 and Bagged CART model performed robustly in identifying high-risk patients earlier.

Clinical trial approval statement

This Trial was registered in the Chinese clinical trials registry: ChiCTR1800014583. Registered January 22, 2018 (http://www.chictr.org.cn/searchproj.aspx).

Combined influence of nutritional and inflammatory status and breast cancer: findings from the NHANES

BACKGROUND

Previous studies have hinted at the benefits of following an anti-inflammatory diet for potentially reducing breast cancer prevalence. However, the combined influence of diet and inflammation on breast cancer remains unclear.

METHODS

The advanced lung cancer inflammation index (ALI) was used to assess inflammation and nutritional status. Statistical methods, such as multivariable logistic regression, eXtreme Gradient Boosting (XGBoost) model, and subgroup analysis, were employed to analyze the impact of ALI on prevalence of BC. Additionally, a two-piece-wise logistic regression model with smoothing was used to determine the ALI threshold for BC prevalence. The study aimed to understand the mechanistic association between ALI levels and BC development.

RESULTS

The mean (SD) age of the study population was 50.0 (17.7) years, with 40.0% of individuals classified as obese. Comparing ALI tertiles to the lowest tertile, the odds ratios (95% CI) for breast cancer (BC) were 0.78 (0.62, 0.98) and 0.68 (0.52, 0.87) for T2-T3. The XGBoost machine learning model was employed to assess the importance of selected factors, revealing ALI as one of the top five variables influencing BC. Subgroup analysis identified a correlation between ALI, alcohol consumption, and menopausal status. Additionally, ALI levels were associated with decreased estradiol (E2) levels, increased total testosterone (TT)/E2 ratio, and TT/sex hormone-binding globulin (SHBG) ratio.

CONCLUSION

This study indicates a potential protective effect of ALI levels against breast cancer, possibly related to sex hormone disruption. The findings support the use of optimal therapeutic strategies for preventing breast cancer.

A two-tier feature selection method for predicting mortality risk in ICU patients with acute kidney injury

Acute kidney injury (AKI) is one of the most important lethal factors for patients admitted to intensive care units (ICUs), and timely high-risk prognostic assessment and intervention are essential to improving patient prognosis. In this study, a stacking model using the MIMIC-III dataset with a two-tier feature selection approach was developed to predict the risk of in-hospital mortality in ICU patients admitted for AKI. External validation was performed using separate MIMIC-IV and eICU-CRD. The area under the curve (AUC) was calculated using the stacking model, and features were selected using the Boruta and XGBoost feature selection methods. This study compares the performance of a stacking model using two-tier feature selection with a model using single-tier feature selection (XGBoost: 85; Boruta: 83; two-tier: 0.91). The predictive effectiveness of the stacking model was further validated by using different datasets (Validation 1: 0.83; Validation 2: 0.85) and comparing it with a simpler model and traditional clinical scores (SOFA: 0.65; APACH IV: 0.61). In addition, this study combined interpretable techniques and causal inference to analyze the causal relationship between features and predicted outcomes.

Interpretable machine learning for the prediction of death risk in patients with acute diquat poisoning

The aim of this study was to develop and validate predictive models for assessing the risk of death in patients with acute diquat (DQ) poisoning using innovative machine learning techniques. Additionally, predictive models were evaluated through the application of SHapley Additive ExPlanations (SHAP). A total of 201 consecutive patients from the emergency departments of the First Hospital and Shengjing Hospital of China Medical University admitted for deliberate oral intake of DQ from February 2018 to August 2023 were analysed. The initial clinical data of the patients with acute DQ poisoning were collected. Machine learning methods such as logistic regression, random forest, support vector machine (SVM), and gradient boosting were applied to build the prediction models. The whole sample was split into a training set and a test set at a ratio of 8:2. The performances of these models were assessed in terms of discrimination, calibration, and clinical decision curve analysis (DCA). We also used the SHAP interpretation tool to provide an intuitive explanation of the risk of death in patients with DQ poisoning. Logistic regression, random forest, SVM, and gradient boosting models were established, and the areas under the receiver operating characteristic curves (AUCs) were 0.91, 0.98, 0.96 and 0.94, respectively. The net benefits were similar across all four models. The four machine learning models can be reliable tools for predicting death risk in patients with acute DQ poisoning. Their combination with SHAP provides explanations for individualized risk prediction, increasing the model transparency.

Risk Factors and Prediction of 28-Day-All Cause Mortality Among Critically Ill Patients with Acute Pancreatitis Using Machine Learning Techniques: A Retrospective Analysis of Multi-Institutions

Objective

This study aimed to identify the risk factors and construct a reliable prediction model of 28-day all-cause mortality in critically ill patients with acute pancreatitis (AP) using machine learning techniques.

Methods

A total of 534 patients from three different institutions were included. Thirty-eight possible variables were collected from the Intensive care unit (ICU) admission for investigation. Patients were split into a training cohort (n = 400) and test cohort (n = 134) according to their source of hospital. The synthetic minority oversampling technique (SMOTE) was introduced to handle the inherent class imbalance. Six machine learning algorithms were applied in this study. The optimal machine learning model was chosen after patients in the test cohort were selected to validate the models. SHapley Additive exPlanation (SHAP) analysis was performed to rank the importance of variable. The predictive performance of the models was evaluated by the calibration curve, area under the receiver operating characteristics curves (AUROC), and decision clinical analysis.

Results

About 13.5% (72/534) of all patients eventually died of all-cause within 28 days of ICU admission. Eight important variables were screened out, including white blood cell count, platelets, body temperature, age, blood urea nitrogen, red blood cell distribution width, SpO2, and hemoglobin. The support vector machine (SVM) algorithm performed best in predicting 28-d all-cause death. Its AUROC reached 0.877 (95% CI: 0.809 to 0.927, p < 0.001), the Youden index was 0.634 (95% CI: 0.459 to 0.717). Based on the risk stratification system, the difference between the high-risk and low-risk groups was significantly different.

Conclusion

In conclusion, this study developed and validated SVM model, which better predicted 28-d all-cause mortality in critically ill patients with AP. In the future, we will continue to include patients from more institutions to conduct validation in different contexts and countries.

Using machine-learning models to predict extubation failure in neonates with bronchopulmonary dysplasia

AIM

To develop a decision-support tool for predicting extubation failure (EF) in neonates with bronchopulmonary dysplasia (BPD) using a set of machine-learning algorithms.

METHODS

A dataset of 284 BPD neonates on mechanical ventilation was used to develop predictive models via machine-learning algorithms, including extreme gradient boosting (XGBoost), random forest, support vector machine, naïve Bayes, logistic regression, and k-nearest neighbor. The top three models were assessed by the area under the receiver operating characteristic curve (AUC), and their performance was tested by decision curve analysis (DCA). Confusion matrix was used to show the high performance of the best model. The importance matrix plot and SHapley Additive exPlanations values were calculated to evaluate the feature importance and visualize the results. The nomogram and clinical impact curves were used to validate the final model.

RESULTS

According to the AUC values and DCA results, the XGboost model performed best (AUC = 0.873, sensitivity = 0.896, specificity = 0.838). The nomogram and clinical impact curve verified that the XGBoost model possessed a significant predictive value. The following were predictive factors for EF: pO2, hemoglobin, mechanical ventilation (MV) rate, pH, Apgar score at 5 min, FiO2, C-reactive protein, Apgar score at 1 min, red blood cell count, PIP, gestational age, highest FiO2 at the first 24 h, heart rate, birth weight, pCO2. Further, pO2, hemoglobin, and MV rate were the three most important factors for predicting EF.

CONCLUSIONS

The present study indicated that the XGBoost model was significant in predicting EF in BPD neonates with mechanical ventilation, which is helpful in determining the right extubation time among neonates with BPD to reduce the occurrence of complications.

The correlation between serum calcium levels and prognosis in patients with severe acute osteomyelitis

OBJECTIVE

To explore the relationship between serum calcium levels and the prognosis of severe acute osteomyelitis, and to assess the effectiveness of calcium levels in prognostic evaluation.

METHODS

Relevant patient records of individuals diagnosed with severe acute osteomyelitis were obtained for this retrospective study from the Medical Information Mart for Intensive Care (MIMIC-IV). The study aimed to assess the impact of different indicators on prognosis by utilizing COX regression analysis. To enhance prognostic prediction for critically ill patients, a nomogram was developed. The discriminatory capacity of the nomogram was evaluated using the Area Under the Curve (AUC) of the Receiver Operating Characteristic (ROC) curve, in addition to the calibration curve.

RESULT

The study analyzed a total of 1,133 cases of severe acute osteomyelitis, divided into the survivor group (1,025 cases) and the non-survivor group (108 cases). Significant differences were observed between the two groups in terms of age, hypertension, sepsis, renal injury, and various laboratory indicators, including WBC, PLT, Ca2+, CRP, hemoglobin, albumin, and creatinine (P<0.05). However, no significant differences were found in race, gender, marital status, detection of wound microbiota, blood sugar, lactate, and ALP levels. A multivariate COX proportional hazards model was constructed using age, hypertension, sepsis, Ca2+, creatinine, albumin, and hemoglobin as variables. The results revealed that hypertension and sepsis had a significant impact on survival time (HR=0.514, 95% CI 0.339-0.779, P=0.002; HR=1.696, 95% CI 1.056-2.723, P=0.029). Age, hemoglobin, Ca2+, albumin, and creatinine also showed significant effects on survival time (P<0.05). However, no statistically significant impact on survival time was observed for the other variables (P>0.05). To predict the survival time, a nomogram was developed using the aforementioned indicators and achieved an AUC of 0.841. The accuracy of the nomogram was further confirmed by the ROC curve and calibration curve.

CONCLUSION

According to the findings, this study establishes that a reduction in serum calcium levels serves as a distinct and standalone predictor of mortality among individuals diagnosed with severe acute osteomyelitis during their stay in the Intensive Care Unit (ICU) within a span of two years.

Early Prognostication of Critical Patients With Spinal Cord Injury: A Machine Learning Study With 1485 Cases

STUDY DESIGN

A retrospective case-series.

OBJECTIVE

The study aims to use machine learning to predict the discharge destination of spinal cord injury (SCI) patients in the intensive care unit.

SUMMARY OF BACKGROUND DATA

Prognostication following SCI is vital, especially for critical patients who need intensive care.

PATIENTS AND METHODS

Clinical data of patients diagnosed with SCI were extracted from a publicly available intensive care unit database. The first recorded data of the included patients were used to develop a total of 98 machine learning classifiers, seeking to predict discharge destination (eg, death, further medical care, home, etc.). The microaverage area under the curve (AUC) was the main indicator to assess discrimination. The best average-AUC classifier and the best death-sensitivity classifier were integrated into an ensemble classifier. The discrimination of the ensemble classifier was compared with top death-sensitivity classifiers and top average-AUC classifiers. In addition, prediction consistency and clinical utility were also assessed.

RESULTS

A total of 1485 SCI patients were included. The ensemble classifier had a microaverage AUC of 0.851, which was only slightly inferior to the best average-AUC classifier ( P =0.10). The best average-AUC classifier death sensitivity was much lower than that of the ensemble classifier. The ensemble classifier had a death sensitivity of 0.452, which was inferior to the top 8 death-sensitivity classifiers, whose microaverage AUC were inferior to the ensemble classifier ( P <0.05). In addition, the ensemble classifier demonstrated a comparable Brier score and superior net benefit in the DCA when compared with the performance of the origin classifiers.

CONCLUSIONS

The ensemble classifier shows an overall superior performance in predicting discharge destination, considering discrimination ability, prediction consistency, and clinical utility. This classifier system may aid in the clinical management of critical SCI patients in the early phase following injury.

LEVEL OF EVIDENCE

Level 3.

Development and validation of early prediction models for new-onset functional impairment in patients after being transferred from the ICU

A significant number of intensive care unit (ICU) survivors experience new-onset functional impairments that impede their activities of daily living (ADL). Currently, no effective assessment tools are available to identify these high-risk patients. This study aims to develop an interpretable machine learning (ML) model for predicting the onset of functional impairment in critically ill patients. Data for this study were sourced from a comprehensive hospital in China, focusing on adult patients admitted to the ICU from August 2022 to August 2023 without prior functional impairments. A least absolute shrinkage and selection operator (LASSO) model was utilized to select predictors for inclusion in the model. Four models, logistic regression, support vector machine (SVM), random forest (RF), and extreme gradient boosting (XGBoost), were constructed and validated. Model performance was assessed using the area under the curve (AUC), accuracy, sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV). Additionally, the DALEX package was employed to enhance the interpretability of the final models. The study ultimately included 1,380 patients, with 684 (49.6%) exhibiting new-onset functional impairment on the seventh day after leaving the ICU. Among the four models evaluated, the SVM model demonstrated the best performance, with an AUC of 0.909, accuracy of 0.838, sensitivity of 0.902, specificity of 0.772, PPV of 0.802, and NPV of 0.886. ML models are reliable tools for predicting new-onset functional impairments in critically ill patients. Notably, the SVM model emerged as the most effective, enabling early identification of patients at high risk and facilitating the implementation of timely interventions to improve ADL.

Interpretable machine learning model integrating clinical and elastosonographic features to detect renal fibrosis in Asian patients with chronic kidney disease

BACKGROUND

Non-invasive renal fibrosis assessment is critical for tailoring personalized decision-making and managing follow-up in patients with chronic kidney disease (CKD). We aimed to exploit machine learning algorithms using clinical and elastosonographic features to distinguish moderate-severe fibrosis from mild fibrosis among CKD patients.

METHODS

A total of 162 patients with CKD who underwent shear wave elastography examinations and renal biopsies at our institution were prospectively enrolled. Four classifiers using machine learning algorithms, including eXtreme Gradient Boosting (XGBoost), Support Vector Machine (SVM), Light Gradient Boosting Machine (LightGBM), and K-Nearest Neighbor (KNN), which integrated elastosonographic features and clinical characteristics, were established to differentiate moderate-severe renal fibrosis from mild forms. The area under the receiver operating characteristic curve (AUC) and average precision were employed to compare the performance of constructed models, and the SHapley Additive exPlanations (SHAP) strategy was used to visualize and interpret the model output.

RESULTS

The XGBoost model outperformed the other developed machine learning models, demonstrating optimal diagnostic performance in both the primary (AUC = 0.97, 95% confidence level (CI) 0.94-0.99; average precision = 0.97, 95% CI 0.97-0.98) and five-fold cross-validation (AUC = 0.85, 95% CI 0.73-0.98; average precision = 0.90, 95% CI 0.86-0.93) datasets. The SHAP approach provided visual interpretation for XGBoost, highlighting the features' impact on the diagnostic process, wherein the estimated glomerular filtration rate provided the largest contribution to the model output, followed by the elastic modulus, then renal length, renal resistive index, and hypertension.

CONCLUSION

This study proposed an XGBoost model for distinguishing moderate-severe renal fibrosis from mild forms in CKD patients, which could be used to assist clinicians in decision-making and follow-up strategies. Moreover, the SHAP algorithm makes it feasible to visualize and interpret the feature processing and diagnostic processes of the model output.

Community screening for dementia among older adults in China: a machine learning-based strategy

BACKGROUND

Dementia is a leading cause of disability in people older than 65 years worldwide. However, diagnosing dementia in its earliest symptomatic stages remains challenging. This study combined specific questions from the AD8 scale with comprehensive health-related characteristics, and used machine learning (ML) to construct diagnostic models of cognitive impairment (CI).

METHODS

The study was based on the Shenzhen Healthy Ageing Research (SHARE) project, and we recruited 823 participants aged 65 years and older, who completed a comprehensive health assessment and cognitive function assessments. Permutation importance was used to select features. Five ML models using BalanceCascade were applied to predict CI: a support vector machine (SVM), multilayer perceptron (MLP), AdaBoost, gradient boosting decision tree (GBDT), and logistic regression (LR). An AD8 score ≥ 2 was used to define CI as a baseline. SHapley Additive exPlanations (SHAP) values were used to interpret the results of ML models.

RESULTS

The first and sixth items of AD8, platelets, waist circumference, body mass index, carcinoembryonic antigens, age, serum uric acid, white blood cells, abnormal electrocardiogram, heart rate, and sex were selected as predictive features. Compared to the baseline (AUC = 0.65), the MLP showed the highest performance (AUC: 0.83 ± 0.04), followed by AdaBoost (AUC: 0.80 ± 0.04), SVM (AUC: 0.78 ± 0.04), GBDT (0.76 ± 0.04). Furthermore, the accuracy, sensitivity and specificity of four ML models were higher than the baseline. SHAP summary plots based on MLP showed the most influential feature on model decision for positive CI prediction was female sex, followed by older age and lower waist circumference.

CONCLUSIONS

The diagnostic models of CI applying ML, especially the MLP, were substantially more effective than the traditional AD8 scale with a score of ≥ 2 points. Our findings may provide new ideas for community dementia screening and to promote such screening while minimizing medical and health resources.

Comparing Open-Access Database and Traditional Intensive Care Studies Using Machine Learning: Bibliometric Analysis Study

BACKGROUND

Intensive care research has predominantly relied on conventional methods like randomized controlled trials. However, the increasing popularity of open-access, free databases in the past decade has opened new avenues for research, offering fresh insights. Leveraging machine learning (ML) techniques enables the analysis of trends in a vast number of studies.

OBJECTIVE

This study aims to conduct a comprehensive bibliometric analysis using ML to compare trends and research topics in traditional intensive care unit (ICU) studies and those done with open-access databases (OADs).

METHODS

We used ML for the analysis of publications in the Web of Science database in this study. Articles were categorized into "OAD" and "traditional intensive care" (TIC) studies. OAD studies were included in the Medical Information Mart for Intensive Care (MIMIC), eICU Collaborative Research Database (eICU-CRD), Amsterdam University Medical Centers Database (AmsterdamUMCdb), High Time Resolution ICU Dataset (HiRID), and Pediatric Intensive Care database. TIC studies included all other intensive care studies. Uniform manifold approximation and projection was used to visualize the corpus distribution. The BERTopic technique was used to generate 30 topic-unique identification numbers and to categorize topics into 22 topic families.

RESULTS

A total of 227,893 records were extracted. After exclusions, 145,426 articles were identified as TIC and 1301 articles as OAD studies. TIC studies experienced exponential growth over the last 2 decades, culminating in a peak of 16,378 articles in 2021, while OAD studies demonstrated a consistent upsurge since 2018. Sepsis, ventilation-related research, and pediatric intensive care were the most frequently discussed topics. TIC studies exhibited broader coverage than OAD studies, suggesting a more extensive research scope.

CONCLUSIONS

This study analyzed ICU research, providing valuable insights from a large number of publications. OAD studies complement TIC studies, focusing on predictive modeling, while TIC studies capture essential qualitative information. Integrating both approaches in a complementary manner is the future direction for ICU research. Additionally, natural language processing techniques offer a transformative alternative for literature review and bibliometric analysis.

An Explainable Artificial Intelligence Model to Predict Malignant Cerebral Edema after Acute Anterior Circulating Large-Hemisphere Infarction

INTRODUCTION

Malignant cerebral edema (MCE) is a serious complication and the main cause of poor prognosis in patients with large-hemisphere infarction (LHI). Therefore, the rapid and accurate identification of potential patients with MCE is essential for timely therapy. This study utilized an artificial intelligence-based machine learning approach to establish an interpretable model for predicting MCE in patients with LHI.

METHODS

This study included 314 patients with LHI not undergoing recanalization therapy. The patients were divided into MCE and non-MCE groups, and the eXtreme Gradient Boosting (XGBoost) model was developed. A confusion matrix was used to measure the prediction performance of the XGBoost model. We also utilized the SHapley Additive exPlanations (SHAP) method to explain the XGBoost model. Decision curve and receiver operating characteristic curve analyses were performed to evaluate the net benefits of the model.

RESULTS

MCE was observed in 121 (38.5%) of the 314 patients with LHI. The model showed excellent predictive performance, with an area under the curve of 0.916. The SHAP method revealed the top 10 predictive variables of the MCE such as ASPECTS score, NIHSS score, CS score, APACHE II score, HbA1c, AF, NLR, PLT, GCS, and age based on their importance ranking.

CONCLUSION

An interpretable predictive model can increase transparency and help doctors accurately predict the occurrence of MCE in LHI patients not undergoing recanalization therapy within 48 h of onset, providing patients with better treatment strategies and enabling optimal resource allocation.

A Real-Time Automated Machine Learning Algorithm for Predicting Mortality in Trauma Patients: Survey Says it's Ready for Prime-Time

BACKGROUND

Though artificial intelligence ("AI") has been increasingly applied to patient care, many of these predictive models are retrospective and not readily available for real-time decision-making. This survey-based study aims to evaluate implementation of a new, validated mortality risk calculator (Parkland Trauma Index of Mortality, "PTIM") embedded in our electronic healthrecord ("EHR") that calculates hourly predictions of mortality with high sensitivity and specificity.

METHODS

This is a prospective, survey-based study performed at a level 1 trauma center. An anonymous survey was sent to surgical providers and regarding PTIM implementation. The PTIM score evaluates 23 variables including Glasgow Coma Score (GCS), vital signs, and laboratory data.

RESULTS

Of the 40 completed surveys, 35 reported using PTIM in decision-making. Prior to reviewing PTIM, providers identified perceived top 3 predictors of mortality, including GCS (22/38, 58%), age (18/35, 47%), and maximum heart rate (17/35, 45%). Most providers reported the PTIM assisted their treatment decisions (27/35, 77%) and timing of operative intervention (23/35, 66%). Many providers agreed that PTIM integrated into rounds and patient assessment (22/36, 61%) and that it improved efficiency in assessing patients' potential mortality (21/36, 58%).

CONCLUSIONS

Artificial intelligence algorithms are mostly retrospective and lag in real-time prediction of mortality. To our knowledge, this is the first real-time, automated algorithm predicting mortality in trauma patients. In this small survey-based study, we found PTIM assists in decision-making, timing of intervention, and improves accuracy in assessing mortality. Next steps include evaluating the short- and long-term impact on patient outcomes.

Unsupervised machine learning analysis to identify patterns of ICU medication use for fluid overload prediction

INTRODUCTION

Intravenous (IV) medications are a fundamental cause of fluid overload (FO) in the intensive care unit (ICU); however, the association between IV medication use (including volume), administration timing, and FO occurrence remains unclear.

METHODS

This retrospective cohort study included consecutive adults admitted to an ICU ≥72 hours with available fluid balance data. FO was defined as a positive fluid balance ≥7% of admission body weight within 72 hours of ICU admission. After reviewing medication administration record (MAR) data in three-hour periods, IV medication exposure was categorized into clusters using principal component analysis (PCA) and Restricted Boltzmann Machine (RBM). Medication regimens of patients with and without FO were compared within clusters to assess for temporal clusters associated with FO using the Wilcoxon rank sum test. Exploratory analyses of the medication cluster most associated with FO for medications frequently appearing and used in the first 24 hours was conducted.

RESULTS

FO occurred in 127/927 (13.7%) of the patients enrolled. Patients received a median (IQR) of 31 (13-65) discrete IV medication administrations over the 72-hour period. Across all 47,803 IV medication administrations, ten unique IV medication clusters were identified with 121-130 medications in each cluster. Among the ten clusters, cluster 7 had the greatest association with FO; the mean number of cluster 7 medications received was significantly greater in patients in the FO cohort compared to patients who did not experience FO (25.6 vs.10.9. p<0.0001). 51 of the 127 medications in cluster 7 (40.2%) appeared in > 5 separate 3-hour periods during the 72-hour study window. The most common cluster 7 medications included continuous infusions, antibiotics, and sedatives/analgesics. Addition of cluster 7 medications to a prediction model with APACHE II score and receipt of diuretics improved the ability for the model to predict fluid overload (AUROC 5.65, p =0.0004).

CONCLUSIONS

Using ML approaches, a unique IV medication cluster was strongly associated with FO. Incorporation of this cluster improved the ability to predict development of fluid overload in ICU patients compared with traditional prediction models. This method may be further developed into real-time clinical applications to improve early detection of adverse outcomes.