Machine Learning Prediction Models for Mechanically Ventilated Patients: Analyses of the MIMIC-III Database

Explainable machine learning model for predicting paratracheal lymph node metastasis in cN0 papillary thyroid cancer

Prophylactic dissection of paratracheal lymph nodes in clinically lymph node-negative (cN0) papillary thyroid carcinoma (PTC) remains controversial. This study aims to integrate preoperative and intraoperative variables to compare traditional nomograms and machine learning (ML) models, developing and validating an interpretable predictive model for paratracheal lymph node metastasis (PLNM) in cN0 PTC patients. We retrospectively selected 3213 PTC patients treated at the First Affiliated Hospital of Chongqing Medical University from 2016 to 2020. They were randomly divided into the training and test datasets with a 7:3 ratio. The 533 PTC patients treated at the Guangyuan Central Hospital from 2019 to 2022 were used as an external test sets. We developed and validated nine ML models using 10-fold cross-validation and grid search for hyperparameter tuning. The predictive performance was evaluated using ROC curves, decision curve analysis (DCA), calibration curves, and precision-recall curves. The best model was compared to a traditional logistic regression-based nomogram. The XGBoost model achieved AUC values of 0.935, 0.857, and 0.775 in the training, validation, and test sets, respectively, significantly outperforming the traditional nomogram model with AUCs of 0.85, 0.844, and 0.769, respectively. SHapley Additive exPlanations (SHAP)-based visualization identified the top 10 predictive features of the XGBoost model, and a web-based calculator was created based on these features. ML is a reliable tool for predicting PLNM in cN0 PTC patients. The SHAP method provides valuable insights into the XGBoost model, and the resultant web-based calculator is a clinically useful tool to assist in the surgical planning for paratracheal lymph node dissection.

A machine learning-based prediction of hospital mortality in mechanically ventilated ICU patients

BACKGROUND

Mechanical ventilation (MV) is vital for critically ill ICU patients but carries significant mortality risks. This study aims to develop a predictive model to estimate hospital mortality among MV patients, utilizing comprehensive health data to assist ICU physicians with early-stage alerts.

METHODS

We developed a Machine Learning (ML) framework to predict hospital mortality in ICU patients receiving MV. Using the MIMIC-III database, we identified 25,202 eligible patients through ICD-9 codes. We employed backward elimination and the Lasso method, selecting 32 features based on clinical insights and literature. Data preprocessing included eliminating columns with over 90% missing data and using mean imputation for the remaining missing values. To address class imbalance, we used the Synthetic Minority Over-sampling Technique (SMOTE). We evaluated several ML models, including CatBoost, XGBoost, Decision Tree, Random Forest, Support Vector Machine (SVM), K-Nearest Neighbors (KNN), and Logistic Regression, using a 70/30 train-test split. The CatBoost model was chosen for its superior performance in terms of accuracy, precision, recall, F1-score, AUROC metrics, and calibration plots.

RESULTS

The study involved a cohort of 25,202 patients on MV. The CatBoost model attained an AUROC of 0.862, an increase from an initial AUROC of 0.821, which was the best reported in the literature. It also demonstrated an accuracy of 0.789, an F1-score of 0.747, and better calibration, outperforming other models. These improvements are due to systematic feature selection and the robust gradient boosting architecture of CatBoost.

CONCLUSION

The preprocessing methodology significantly reduced the number of relevant features, simplifying computational processes, and identified critical features previously overlooked. Integrating these features and tuning the parameters, our model demonstrated strong generalization to unseen data. This highlights the potential of ML as a crucial tool in ICUs, enhancing resource allocation and providing more personalized interventions for MV patients.

Revolutionary Point-of-Care Wearable Diagnostics for Early Disease Detection and Biomarker Discovery through Intelligent Technologies

Early-stage disease detection, particularly in Point-Of-Care (POC) wearable formats, assumes pivotal role in advancing healthcare services and precision-medicine. Public benefits of early detection extend beyond cost-effectively promoting healthcare outcomes, to also include reducing the risk of comorbid diseases. Technological advancements enabling POC biomarker recognition empower discovery of new markers for various health conditions. Integration of POC wearables for biomarker detection with intelligent frameworks represents ground-breaking innovations enabling automation of operations, conducting advanced large-scale data analysis, generating predictive models, and facilitating remote and guided clinical decision-making. These advancements substantially alleviate socioeconomic burdens, creating a paradigm shift in diagnostics, and revolutionizing medical assessments and technology development. This review explores critical topics and recent progress in development of 1) POC systems and wearable solutions for early disease detection and physiological monitoring, as well as 2) discussing current trends in adoption of smart technologies within clinical settings and in developing biological assays, and ultimately 3) exploring utilities of POC systems and smart platforms for biomarker discovery. Additionally, the review explores technology translation from research labs to broader applications. It also addresses associated risks, biases, and challenges of widespread Artificial Intelligence (AI) integration in diagnostics systems, while systematically outlining potential prospects, current challenges, and opportunities.

Predicting dyslipidemia incidence: unleashing machine learning algorithms on Lifestyle Promotion Project data

BACKGROUND

Dyslipidemia, characterized by variations in plasma lipid profiles, poses a global health threat linked to millions of deaths annually.

OBJECTIVES

This study focuses on predicting dyslipidemia incidence using machine learning methods, addressing the crucial need for early identification and intervention.

METHODS

The dataset, derived from the Lifestyle Promotion Project (LPP) in East Azerbaijan Province, Iran, undergoes a comprehensive preprocessing, merging, and null handling process. Target selection involves five distinct dyslipidemia-related variables. Normalization techniques and three feature selection algorithms are applied to enhance predictive modeling.

RESULT

The study results underscore the potential of different machine learning algorithms, specifically multi-layer perceptron neural network (MLP), in reaching higher performance metrics such as accuracy, F1 score, sensitivity and specificity, among other machine learning methods. Among other algorithms, Random Forest also showed remarkable accuracies and outperformed K-Nearest Neighbors (KNN) in metrics like precision, recall, and F1 score. The study's emphasis on feature selection detected meaningful patterns among five target variables related to dyslipidemia, indicating fundamental shared unities among dyslipidemia-related factors. Features such as waist circumference, serum vitamin D, blood pressure, sex, age, diabetes, and physical activity related to dyslipidemia.

CONCLUSION

These results cooperatively highlight the complex nature of dyslipidemia and its connections with numerous factors, strengthening the importance of applying machine learning methods to understand and predict its incidence precisely.

Machine learning for the prediction of in-hospital mortality in patients with spontaneous intracerebral hemorrhage in intensive care unit

This study aimed to develop a machine learning (ML)-based tool for early and accurate prediction of in-hospital mortality risk in patients with spontaneous intracerebral hemorrhage (sICH) in the intensive care unit (ICU). We did a retrospective study in our study and identified cases of sICH from the MIMIC IV (n = 1486) and Zhejiang Hospital databases (n = 110). The model was constructed using features selected through LASSO regression. Among five well-known models, the selection of the best model was based on the area under the curve (AUC) in the validation cohort. We further analyzed calibration and decision curves to assess prediction results and visualized the impact of each variable on the model through SHapley Additive exPlanations. To facilitate accessibility, we also created a visual online calculation page for the model. The XGBoost exhibited high accuracy in both internal validation (AUC = 0.907) and external validation (AUC = 0.787) sets. Calibration curve and decision curve analyses showed that the model had no significant bias as well as being useful for supporting clinical decisions. XGBoost is an effective algorithm for predicting in-hospital mortality in patients with sICH, indicating its potential significance in the development of early warning systems.

Compact machine learning model for the accurate prediction of first 24-hour survival of mechanically ventilated patients

Background

The field of machine learning has been evolving and applied in medical applications. We utilised a public dataset, MIMIC-III, to develop compact models that can accurately predict the outcome of mechanically ventilated patients in the first 24 h of first-time hospital admission.

Methods

67 predictive features, grouped into 6 categories, were selected for the classification and prediction task. 4 tree-based algorithms (Decision Tree, Bagging, eXtreme Gradient Boosting and Random Forest), and 5 non-tree-based algorithms (Logistic Regression, K-Nearest Neighbour, Linear Discriminant Analysis, Support Vector Machine and Naïve Bayes), were employed to predict the outcome of 18,883 mechanically ventilated patients. 5 scenarios were crafted to mirror the target population as per existing literature. S1.1 reflected an imbalanced situation, with significantly fewer mortality cases than survival ones, and both the training and test sets played similar target class distributions. S1.2 and S2.2 featured balanced classes; however, instances from the majority class were removed from the test set and/or the training set. S1.3 and S 2.3 generated additional instances of the minority class via the Synthetic Minority Over-sampling Technique. Standard evaluation metrics were used to determine the best-performing models for each scenario. With the best performers, Autofeat, an automated feature engineering library, was used to eliminate less important features per scenario.

Results

Tree-based models generally outperformed the non-tree-based ones. Moreover, XGB consistently yielded the highest AUC score (between 0.91 and 0.97), while exhibiting relatively high Sensitivity (between 0.58 and 0.88) on 4 scenarios (1.2, 2.2, 1.3, and 2.3). After reducing a significant number of predictors, the selected calibrated ML models were still able to achieve similar AUC and MCC scores across those scenarios. The calibration curves of the XGB and BG models, both prior to and post dimension reduction in Scenario 2.2, showed better alignment to the perfect calibration line than curves produced from other algorithms.

Conclusion

This study demonstrated that dimension-reduced models can perform well and are able to retain the important features for the classification tasks. Deploying a compact machine learning model into production helps reduce costs in terms of computational resources and monitoring changes in input data over time.

Machine learning improves early prediction of organ failure in hyperlipidemia acute pancreatitis using clinical and abdominal CT features

BACKGROUND

This study aimed to investigate and validate machine-learning predictive models combining computed tomography and clinical data to early predict organ failure (OF) in Hyperlipidemic acute pancreatitis (HLAP).

METHODS

Demographics, laboratory parameters and computed tomography imaging data of 314 patients with HLAP from the First Affiliated Hospital of Wenzhou Medical University between 2017 and 2021, were retrospectively analyzed. Sixty-five percent of patients (n = 204) were assigned to the training group and categorized as patients with and without OF. Parameters were compared by univariate analysis. Machine-learning methods including random forest (RF) were used to establish model to predict OF of HLAP. Areas under the curves (AUCs) of receiver operating characteristic were calculated. The remaining 35% patients (n = 110) were assigned to the validation group to evaluate the performance of models to predict OF.

RESULTS

Ninety-three (45.59%) and fifty (45.45%) patients from the training and the validation cohort, respectively, developed OF. The RF model showed the best performance to predict OF, with the highest AUC value of 0.915. The sensitivity (0.828) and accuracy (0.814) of RF model were both the highest among the five models in the study cohort. In the validation cohort, RF model continued to show the highest AUC (0.820), accuracy (0.773) and sensitivity (0.800) to predict OF in HLAP, while the positive and negative likelihood ratios and post-test probability were 3.22, 0.267 and 72.85%, respectively.

CONCLUSIONS

Machine-learning models can be used to predict OF occurrence in HLAP in our pilot study. RF model showed the best predictive performance, which may be a promising candidate for further clinical validation.

A Continuous Non-Invasive Blood Pressure Prediction Method Based on Deep Sparse Residual U-Net Combined with Improved Squeeze and Excitation Skip Connections

Arterial blood pressure (ABP) serves as a pivotal clinical metric in cardiovascular health assessments, with the precise forecasting of continuous blood pressure assuming a critical role in both preventing and treating cardiovascular diseases. This study proposes a novel continuous non-invasive blood pressure prediction model, DSRUnet, based on deep sparse residual U-net combined with improved SE skip connections, which aim to enhance the accuracy of using photoplethysmography (PPG) signals for continuous blood pressure prediction. The model first introduces a sparse residual connection approach for path contraction and expansion, facilitating richer information fusion and feature expansion to better capture subtle variations in the original PPG signals, thereby enhancing the network's representational capacity and predictive performance and mitigating potential degradation in the network performance. Furthermore, an enhanced SE-GRU module was embedded in the skip connections to model and weight global information using an attention mechanism, capturing the temporal features of the PPG pulse signals through GRU layers to improve the quality of the transferred feature information and reduce redundant feature learning. Finally, a deep supervision mechanism was incorporated into the decoder module to guide the lower-level network to learn effective feature representations, alleviating the problem of gradient vanishing and facilitating effective training of the network. The proposed DSRUnet model was trained and tested on the publicly available UCI-BP dataset, with the average absolute errors for predicting systolic blood pressure (SBP), diastolic blood pressure (DBP), and mean blood pressure (MBP) being 3.36 ± 6.61 mmHg, 2.35 ± 4.54 mmHg, and 2.21 ± 4.36 mmHg, respectively, meeting the standards set by the Association for the Advancement of Medical Instrumentation (AAMI), and achieving Grade A according to the British Hypertension Society (BHS) Standard for SBP and DBP predictions. Through ablation experiments and comparisons with other state-of-the-art methods, the effectiveness of DSRUnet in blood pressure prediction tasks, particularly for SBP, which generally yields poor prediction results, was significantly higher. The experimental results demonstrate that the DSRUnet model can accurately utilize PPG signals for real-time continuous blood pressure prediction and obtain high-quality and high-precision blood pressure prediction waveforms. Due to its non-invasiveness, continuity, and clinical relevance, the model may have significant implications for clinical applications in hospitals and research on wearable devices in daily life.

Artificial intelligence differentiates abdominal Henoch-Schönlein purpura from acute appendicitis in children

OBJECTIVE

This study aims to construct an artificial intelligence (AI) model capable of effectively discriminating between abdominal Henoch-Schönlein purpura (AHSP) and acute appendicitis (AA) in pediatric patients.

METHODS

A total of 6965 participants, comprising 2201 individuals with AHSP and 4764 patients with AA, were enrolled in the study. Additionally, 53 laboratory indicators were taken into consideration. Five distinct artificial intelligence (AI) models were developed employing machine learning algorithms, namely XGBoost, AdaBoost, Gaussian Naïve Bayes (GNB), MLPClassifier (MLP), and support vector machine (SVM). The performance of these prediction models was assessed through receiver operating characteristic (ROC) curve analysis, calibration curve assessment, and decision curve analysis (DCA).

RESULTS

We identified 32 discriminative indicators (p < .05) between AHSP and AA. Five indicators, namely the lymphocyte ratio (LYMPH ratio), eosinophil ratio (EO ratio), eosinophil count (EO count), neutrophil ratio (NEUT ratio), and C-reactive protein (CRP), exhibited strong performance in distinguishing AHSP from AA (AUC ≥ 0.80). Among the various prediction models, the XGBoost model displayed superior performance evidenced by the highest AUC (XGBoost = 0.895, other models < 0.89), accuracy (XGBoost = 0.824, other models < 0.81), and Kappa value (XGBoost = 0.621, other models < 0.60) in the validation set. After optimization, the XGBoost model demonstrated remarkable diagnostic performance for AHSP and AA (AUC > 0.95). Both the calibration curve and decision curve analysis suggested the promising clinical utility and net benefits of the XGBoost model.

CONCLUSION

The AI-based machine learning model exhibits high prediction accuracy and can differentiate AHSP and AA from a data-driven perspective.

Machine learning-based prediction of low-value care for hospitalized patients

Objective

Low-value care (i.e., costly health care treatments that provide little or no benefit) is an ongoing problem in United States hospitals. Traditional strategies for reducing low-value care are only moderately successful. Informed by behavioral science principles, we sought to use machine learning to inform a targeted prompting system that suggests preferred alternative treatments at the point of care but before clinicians have made a decision.

Methods

We used intravenous administration of albumin for fluid resuscitation in intensive care unit (ICU) patients as an exemplar of low-value care practice, identified using the electronic health record of a multi-hospital health system. We divided all ICU episodes into 4-h periods and defined a set of relevant clinical features at the period level. We then developed two machine learning models: a single-stage model that directly predicts if a patient will receive albumin in the next period; and a two-stage model that first predicts if any resuscitation fluid will be administered and then predicts albumin only among the patients with a high probability of fluid use.

Results

We examined 87,489 ICU episodes divided into approximately 1.5 million 4-h periods. The area under the receiver operating characteristic curve was 0.86 for both prediction models. The positive predictive value was 0.21 (95% confidence interval: 0.20, 0.23) for the single-stage model and 0.22 (0.20, 0.23) for the two-stage model. Applying either model in a targeted prompting system could prevent 10% of albumin administrations, with an attending physician receiving one prompt every 4.2 days of ICU service.

Conclusion

Prediction of low-value care is feasible and could enable a point-of-care, targeted prompting system that offers suggestions ahead of the moment of need before clinicians have already decided. A two-stage approach does not improve performance but does interject new levers for the calibration of such a system.

A Retrospective Cohort Study: Predicting 90-Day Mortality for ICU Trauma Patients with a Machine Learning Algorithm Using XGBoost Using MIMIC-III Database

Objective

The aim of this study was to develop and validate a machine learning-based predictive model that predicts 90-day mortality in ICU trauma patients.

Methods

Data of patients with severe trauma were extracted from the Medical Information Mart for Intensive Care III (MIMIC-III) database. The performances of mortality prediction models generated using nine machine learning extreme gradient boosting (XGBoost), logistic regression, random forest, AdaBoost, multilayer perceptron (MLP) neural networks, support vector machine (SVM), light gradient boosting machine (GBM), k nearest neighbors (KNN) and gaussian naive bayes (GNB). The performance of the model was evaluated in terms of discrimination, calibration and clinical application.

Results

We found that the accuracy, sensitivity, specificity, PPV, NPV and F1 score of our proposed XGBoost model were 82.8%, 79.7%, 77.6%, 51.2%, 91.5% and 0.624, respectively. Among the nine models, the XGBoost model performed best. Compared with traditional logistic regression, the calibration curves of the XGBoost model and decision curve analysis (DCA) performed well.

Conclusion

Our study shows that the XGBoost model outperforms other machine learning models in predicting 90-day mortality in trauma patients. It can be used to assist clinicians in the early identification of mortality risk factors and early intervention to reduce mortality.

Predictive value of the Oxford Acute Severity of Illness Score in acute stroke patients with stroke-associated pneumonia

Background

Stroke-associated pneumonia (SAP) is associated with a poor prognosis and a high mortality rate in stroke patients. However, the accuracy of early prediction of SAP is insufficient, and there is a lack of effective prognostic evaluation methods. Therefore, in this study, we investigated the predictive value of the Oxford Acute Severity of Illness Score (OASIS) in SAP to provide a potential reference index for the incidence and prognosis of SAP.

Methods

We recruited a total of 280 patients with acute ischemic stroke who had been diagnosed and treated in the Zhumadian Central Hospital between January 2021 and January 2023. These patients were divided into an SAP group (86 cases) and a non-SAP group (194 cases) according to SAP diagnostic criteria by expert consensus on the diagnosis and treatment of SAP. We collated general and clinical data from all patients, including the survival of SAP patients during the follow-up period. Multivariate logistic regression was used to analyze the risk factors for SAP. Kaplan-Meier and multivariate COX regression analyses were used to investigate the relationship between OASIS and the prognosis of SAP, and a receiver operating characteristic (ROC) curve was drawn to analyze the predictive value of OASIS for SAP.

Results

Our analyses identified body temperature, C-reactive protein, procalcitonin, OASIS, and a prolonged length of intensive care unit (ICU) stay as the main risk factors for SAP (all Ps < 0.05). Advanced age and an elevated OASIS were identified as the main risk factors for death in SAP patients (all Ps < 0.05). The risk of death in patients with OASIS of 31-42 points was significantly higher than that in patients with OASIS of 12-20 points (HR = 5.588, 95% CI = 1.531-20.401, P = 0.009). ROC curve analysis further showed that OASIS had a high predictive value for morbidity and the incidence of death in SAP patients.

Conclusion

OASIS can effectively predict the onset and death of SAP patients and provides a potential reference index for early diagnosis and the prediction of prognosis in patients with SAP. Our findings should be considered in clinical practice.

XGBoost-aided prediction of lip prominence based on hard-tissue measurements and demographic characteristics in an Asian population

INTRODUCTION

Prediction of lip prominence based on hard-tissue measurements could be helpful in orthodontic treatment planning and has been challenging and formidable thus far.

METHODS

A machine learning-based cross-sectional study was conducted on 1549 patients. Hard-tissue measurements and demographic information were used as the input features. Seven popular machine learning algorithms were applied to the datasets to predict upper and lower lip prominence. The algorithm that performed the best was selected for the construction of the prediction model. Evaluation of feature importance was conducted using 3 classical methods.

RESULTS

Among the 7 algorithms, the XGBoost model performed the best in the prediction of the distances between labrale superius or labrale inferius to the esthetics plane (UL-EP and LL-EP distances), with root mean square error values of 1.25, 1.49 and r2 values of 0.755 and 0.683, respectively. Among the 14 input features, the L1-NB distance contributed the most to the prominences of the upper and lower lips. A lip prominence predictor was developed to facilitate clinical application by deploying the prediction model into a downloadable tool kit.

CONCLUSIONS

The XGBoost model performed well with high accuracy and practicability in predicting upper and lower lip prominence. The artificial intelligence-aided predictor could serve as a reference for orthodontic treatment planning.

A dynamic machine learning model for prediction of NAFLD in a health checkup population: A longitudinal study

Background

Non-alcoholic fatty liver disease (NAFLD) is one of the most common liver diseases worldwide. Currently, most NAFLD prediction models are diagnostic models based on cross-sectional data, which failed to provide early identification or clarify causal relationships. We aimed to use time-series deep learning models with longitudinal health checkup records to predict the onset of NAFLD in the future, and update the model stepwise by incorporating new checkup records to achieve dynamic prediction.

Methods

10,493 participants with over 6 health checkup records from Beijing MJ Health Screening Center were included to conduct a retrospective cohort study, in which the constantly updated initial 5 checkup data were incorporated stepwise to predict the risk of NAFLD at and after their sixth health checkups. A total of 33 variables were considered, consisting of demographic characteristics, medical history, lifestyle, physical examinations, and laboratory tests. L1-penalized logistic regression (LR) was used for feature selection. The long short-term memory (LSTM) algorithm was introduced for model development, and five-fold cross-validation was conducted to tune and choose optimal hyperparameters. Both internal validation and external validation were conducted, using the 20% randomly divided holdout test dataset and previously unseen data from Shanghai MJ Health Screening Center, respectively, to evaluate model performance. The evaluation metrics included area under the receiver operating characteristic curve (AUROC), sensitivity, specificity, Brier score, and decision curve. Bootstrap sampling was implemented to generate 95% confidence intervals of all the metrics. Finally, the Shapley additive explanations (SHAP) algorithm was applied in the holdout test dataset for model interpretability to obtain time-specific and sample-specific contributions of each feature.

Results

Among the 10,493 participants, 1662 (15.84%) were diagnosed with NAFLD at and after their sixth health checkups. The predictive performance of the deep learning model in the internal validation dataset improved over the incorporation of the checkups, with AUROC increasing from 0.729 (95% CI: 0.698,0.760) at baseline to 0.818 (95% CI: 0.798,0.844) when consecutive 5 checkups were included. The external validation dataset, containing 1728 participants, was used to verify the results, in which AUROC increased from 0.700 (95% CI: 0.657,0.740) with only the first checkups to 0.792 (95% CI: 0.758,0.825) with all five. The results of feature significance showed that body fat percentage, alanine transaminase (ALT), and uric acid owned the greatest impact on the outcome, time-specific, individual-specific and dynamic feature contributions were also produced for model interpretability.

Conclusion

A dynamic prediction model was successfully established in our study, and the prediction capability kept improving with the renewal of the latest checkup records. In addition, we identified key features associated with the onset of NAFLD, making it possible to optimize the prevention and control strategies of the disease in the general population.

A long short-temory relation network for real-time prediction of patient-specific ventilator parameters

Accurate prediction of patient-specific ventilator parameters is crucial for optimizing patient-ventilator interaction. Current approaches encounter difficulties in concurrently observing long-term, time-series dependencies and capturing complex, significant features that influence the ventilator treatment process, thereby hindering the achievement of accurate prediction of ventilator parameters. To address these challenges, we propose a novel approach called the long short-term memory relation network (LSTMRnet). Our approach uses a long, short-term memory bank to store rich information and an important feature selection step to extract relevant features related to respiratory parameters. This information is obtained from the prior knowledge of the follow up model. We also concatenate the embeddings of both information types to maintain the joint learning of spatio-temporal features. Our LSTMRnet effectively preserves both time-series and complex spatial-critical feature information, enabling an accurate prediction of ventilator parameters. We extensively validate our approach using the publicly available medical information mart for intensive care (MIMIC-III) dataset and achieve superior results, which can be potentially utilized for ventilator treatment (i.e., sleep apnea-hypopnea syndrome ventilator treatment and intensive care units ventilator treatment.

Construction and validation of prognostic models in critically Ill patients with sepsis-associated acute kidney injury: interpretable machine learning approach

BACKGROUND

Acute kidney injury (AKI) is a common complication in critically ill patients with sepsis and is often associated with a poor prognosis. We aimed to construct and validate an interpretable prognostic prediction model for patients with sepsis-associated AKI (S-AKI) using machine learning (ML) methods.

METHODS

Data on the training cohort were collected from the Medical Information Mart for Intensive Care IV database version 2.2 to build the model, and data of patients were extracted from Hangzhou First People's Hospital Affiliated to Zhejiang University School of Medicine for external validation of model. Predictors of mortality were identified using Recursive Feature Elimination (RFE). Then, random forest, extreme gradient boosting (XGBoost), multilayer perceptron classifier, support vector classifier, and logistic regression were used to establish a prognosis prediction model for 7, 14, and 28 days after intensive care unit (ICU) admission, respectively. Prediction performance was assessed using the receiver operating characteristic (ROC) curve and decision curve analysis (DCA). SHapley Additive exPlanations (SHAP) were used to interpret the ML models.

RESULTS

In total, 2599 patients with S-AKI were included in the analysis. Forty variables were selected for the model development. According to the areas under the ROC curve (AUC) and DCA results for the training cohort, XGBoost model exhibited excellent performance with F1 Score of 0.847, 0.715, 0.765 and AUC (95% CI) of 0.91 (0.90, 0.92), 0.78 (0.76, 0.80), and 0.83 (0.81, 0.85) in 7 days, 14 days and 28 days group, respectively. It also demonstrated excellent discrimination in the external validation cohort. Its AUC (95% CI) was 0.81 (0.79, 0.83), 0.75 (0.73, 0.77), 0.79 (0.77, 0.81) in 7 days, 14 days and 28 days group, respectively. SHAP-based summary plot and force plot were used to interpret the XGBoost model globally and locally.

CONCLUSIONS

ML is a reliable tool for predicting the prognosis of patients with S-AKI. SHAP methods were used to explain intrinsic information of the XGBoost model, which may prove clinically useful and help clinicians tailor precise management.

Machine learning algorithm to predict mortality in critically ill patients with sepsis-associated acute kidney injury

This study aimed to establish and validate a machine learning (ML) model for predicting in-hospital mortality in patients with sepsis-associated acute kidney injury (SA-AKI). This study collected data on SA-AKI patients from 2008 to 2019 using the Medical Information Mart for Intensive Care IV. After employing Lasso regression for feature selection, six ML approaches were used to build the model. The optimal model was chosen based on precision and area under curve (AUC). In addition, the best model was interpreted using SHapley Additive exPlanations (SHAP) values and Local Interpretable Model-Agnostic Explanations (LIME) algorithms. There were 8129 sepsis patients eligible for participation; the median age was 68.7 (interquartile range: 57.2-79.6) years, and 57.9% (4708/8129) were male. After selection, 24 of the 44 clinical characteristics gathered after intensive care unit admission remained linked with prognosis and were utilized developing ML models. Among the six models developed, the eXtreme Gradient Boosting (XGBoost) model had the highest AUC, at 0.794. According to the SHAP values, the sequential organ failure assessment score, respiration, simplified acute physiology score II, and age were the four most influential variables in the XGBoost model. Individualized forecasts were clarified using the LIME algorithm. We built and verified ML models that excel in early mortality risk prediction in SA-AKI and the XGBoost model performed best.

Identification of Distinct Clinical Phenotypes of Heterogeneous Mechanically Ventilated ICU Patients Using Cluster Analysis

This retrospective study aimed to derive the clinical phenotypes of ventilated ICU patients to predict the outcomes on the first day of ventilation. Clinical phenotypes were derived from the eICU Collaborative Research Database (eICU) cohort via cluster analysis and were validated in the Medical Information Mart for Intensive Care (MIMIC-IV) cohort. Four clinical phenotypes were identified and compared in the eICU cohort (n = 15,256). Phenotype A (n = 3112) was associated with respiratory disease, had the lowest 28-day mortality (16%), and had a high extubation success rate (~80%). Phenotype B (n = 3335) was correlated with cardiovascular disease, had the second-highest 28-day mortality (28%), and had the lowest extubation success rate (69%). Phenotype C (n = 3868) was correlated with renal dysfunction, had the highest 28-day mortality (28%), and had the second-lowest extubation success rate (74%). Phenotype D (n = 4941) was associated with neurological and traumatic diseases, had the second-lowest 28-day mortality (22%), and had the highest extubation success rate (>80%). These findings were validated in the validation cohort (n = 10,813). Additionally, these phenotypes responded differently to ventilation strategies in terms of duration of treatment, but had no difference in mortality. The four clinical phenotypes unveiled the heterogeneity of ICU patients and helped to predict the 28-day mortality and the extubation success rate.

Preliminary prediction of semen quality based on modifiable lifestyle factors by using the XGBoost algorithm

Introduction

Semen quality has decreased gradually in recent years, and lifestyle changes are among the primary causes for this issue. Thus far, the specific lifestyle factors affecting semen quality remain to be elucidated.

Materials and methods

In this study, data on the following factors were collected from 5,109 men examined at our reproductive medicine center: 10 lifestyle factors that potentially affect semen quality (smoking status, alcohol consumption, staying up late, sleeplessness, consumption of pungent food, intensity of sports activity, sedentary lifestyle, working in hot conditions, sauna use in the last 3 months, and exposure to radioactivity); general factors including age, abstinence period, and season of semen examination; and comprehensive semen parameters [semen volume, sperm concentration, progressive and total sperm motility, sperm morphology, and DNA fragmentation index (DFI)]. Then, machine learning with the XGBoost algorithm was applied to establish a primary prediction model by using the collected data. Furthermore, the accuracy of the model was verified via multiple logistic regression following k-fold cross-validation analyses.

Results

The results indicated that for semen volume, sperm concentration, progressive and total sperm motility, and DFI, the area under the curve (AUC) values ranged from 0.648 to 0.697, while the AUC for sperm morphology was only 0.506. Among the 13 factors, smoking status was the major factor affecting semen volume, sperm concentration, and progressive and total sperm motility. Age was the most important factor affecting DFI. Logistic combined with cross-validation analysis revealed similar results. Furthermore, it showed that heavy smoking (>20 cigarettes/day) had an overall negative effect on semen volume and sperm concentration and progressive and total sperm motility (OR = 4.69, 6.97, 11.16, and 10.35, respectively), while age of >35 years was associated with increased DFI (OR = 5.47).

Conclusion

The preliminary lifestyle-based model developed for semen quality prediction by using the XGBoost algorithm showed potential for clinical application and further optimization with larger training datasets.

Machine learning for the prediction of acute kidney injury in patients with sepsis

BACKGROUND

Acute kidney injury (AKI) is the most common and serious complication of sepsis, accompanied by high mortality and disease burden. The early prediction of AKI is critical for timely intervention and ultimately improves prognosis. This study aims to establish and validate predictive models based on novel machine learning (ML) algorithms for AKI in critically ill patients with sepsis.

METHODS

Data of patients with sepsis were extracted from the Medical Information Mart for Intensive Care III (MIMIC- III) database. Feature selection was performed using a Boruta algorithm. ML algorithms such as logistic regression (LR), k-nearest neighbors (KNN), support vector machine (SVM), decision tree, random forest, Extreme Gradient Boosting (XGBoost), and artificial neural network (ANN) were applied for model construction by utilizing tenfold cross-validation. The performances of these models were assessed in terms of discrimination, calibration, and clinical application. Moreover, the discrimination of ML-based models was compared with those of Sequential Organ Failure Assessment (SOFA) and the customized Simplified Acute Physiology Score (SAPS) II model.

RESULTS

A total of 3176 critically ill patients with sepsis were included for analysis, of which 2397 cases (75.5%) developed AKI during hospitalization. A total of 36 variables were selected for model construction. The models of LR, KNN, SVM, decision tree, random forest, ANN, XGBoost, SOFA and SAPS II score were established and obtained area under the receiver operating characteristic curves of 0.7365, 0.6637, 0.7353, 0.7492, 0.7787, 0.7547, 0.821, 0.6457 and 0.7015, respectively. The XGBoost model had the best predictive performance in terms of discrimination, calibration, and clinical application among all models.

CONCLUSION

The ML models can be reliable tools for predicting AKI in septic patients. The XGBoost model has the best predictive performance, which can be used to assist clinicians in identifying high-risk patients and implementing early interventions to reduce mortality.

Establishment of ICU Mortality Risk Prediction Models with Machine Learning Algorithm Using MIMIC-IV Database

Objective: The mortality rate of critically ill patients in ICUs is relatively high. In order to evaluate patients’ mortality risk, different scoring systems are used to help clinicians assess prognosis in ICUs, such as the Acute Physiology and Chronic Health Evaluation III (APACHE III) and the Logistic Organ Dysfunction Score (LODS). In this research, we aimed to establish and compare multiple machine learning models with physiology subscores of APACHE III—namely, the Acute Physiology Score III (APS III)—and LODS scoring systems in order to obtain better performance for ICU mortality prediction. Methods: A total number of 67,748 patients from the Medical Information Database for Intensive Care (MIMIC-IV) were enrolled, including 7055 deceased patients, and the same number of surviving patients were selected by the random downsampling technique, for a total of 14,110 patients included in the study. The enrolled patients were randomly divided into a training dataset (n = 9877) and a validation dataset (n = 4233). Fivefold cross-validation and grid search procedures were used to find and evaluate the best hyperparameters in different machine learning models. Taking the subscores of LODS and the physiology subscores that are part of the APACHE III scoring systems as input variables, four machine learning methods of XGBoost, logistic regression, support vector machine, and decision tree were used to establish ICU mortality prediction models, with AUCs as metrics. AUCs, specificity, sensitivity, positive predictive value, negative predictive value, and calibration curves were used to find the best model. Results: For the prediction of mortality risk in ICU patients, the AUC of the XGBoost model was 0.918 (95%CI, 0.915–0.922), and the AUCs of logistic regression, SVM, and decision tree were 0.872 (95%CI, 0.867–0.877), 0.872 (95%CI, 0.867–0.877), and 0.852 (95%CI, 0.847–0.857), respectively. The calibration curves of logistic regression and support vector machine performed better than the other two models in the ranges 0–40% and 70%–100%, respectively, while XGBoost performed better in the range of 40–70%. Conclusions: The mortality risk of ICU patients can be better predicted by the characteristics of the Acute Physiology Score III and the Logistic Organ Dysfunction Score with XGBoost in terms of ROC curve, sensitivity, and specificity. The XGBoost model could assist clinicians in judging in-hospital outcome of critically ill patients, especially in patients with a more uncertain survival outcome.

Machine Learning and Antibiotic Management

Machine learning and cluster analysis applied to the clinical setting of an intensive care unit can be a valuable aid for clinical management, especially with the increasing complexity of clinical monitoring. Providing a method to measure clinical experience, a proxy for that automatic gestalt evaluation that an experienced clinician sometimes effortlessly, but often only after long, hard consideration and consultation with colleagues, relies upon for decision making, is what we wanted to achieve with the application of machine learning to antibiotic therapy and clinical monitoring in the present work. This is a single-center retrospective analysis proposing methods for evaluation of vitals and antimicrobial therapy in intensive care patients. For each patient included in the present study, duration of antibiotic therapy, consecutive days of treatment and type and combination of antimicrobial agents have been assessed and considered as single unique daily record for analysis. Each parameter, composing a record was normalized using a fuzzy logic approach and assigned to five descriptive categories (fuzzy domain sub-sets ranging from “very low” to “very high”). Clustering of these normalized therapy records was performed, and each patient/day was considered to be a pertaining cluster. The same methodology was used for hourly bed-side monitoring. Changes in patient conditions (monitoring) can lead to a shift of clusters. This can provide an additional tool for assessing progress of complex patients. We used Fuzzy logic normalization to descriptive categories of parameters as a form nearer to human language than raw numbers.

Unstructured clinical notes within the 24 hours since admission predict short, mid & long-term mortality in adult ICU patients

Mortality prediction for intensive care unit (ICU) patients is crucial for improving outcomes and efficient utilization of resources. Accessibility of electronic health records (EHR) has enabled data-driven predictive modeling using machine learning. However, very few studies rely solely on unstructured clinical notes from the EHR for mortality prediction. In this work, we propose a framework to predict short, mid, and long-term mortality in adult ICU patients using unstructured clinical notes from the MIMIC III database, natural language processing (NLP), and machine learning (ML) models. Depending on the statistical description of the patients' length of stay, we define the short-term as 48-hour and 4-day period, the mid-term as 7-day and 10-day period, and the long-term as 15-day and 30-day period after admission. We found that by only using clinical notes within the 24 hours of admission, our framework can achieve a high area under the receiver operating characteristics (AU-ROC) score for short, mid and long-term mortality prediction tasks. The test AU-ROC scores are 0.87, 0.83, 0.83, 0.82, 0.82, and 0.82 for 48-hour, 4-day, 7-day, 10-day, 15-day, and 30-day period mortality prediction, respectively. We also provide a comparative study among three types of feature extraction techniques from NLP: frequency-based technique, fixed embedding-based technique, and dynamic embedding-based technique. Lastly, we provide an interpretation of the NLP-based predictive models using feature-importance scores.

Unifying Diagnosis Identification and Prediction Method Embedding the Disease Ontology Structure From Electronic Medical Records

OBJECTIVE

The reasonable classification of a large number of distinct diagnosis codes can clarify patient diagnostic information and help clinicians to improve their ability to assign and target treatment for primary diseases. Our objective is to identify and predict a unifying diagnosis (UD) from electronic medical records (EMRs).

METHODS

We screened 4,418 sepsis patients from a public MIMIC-III database and extracted their diagnostic information for UD identification, their demographic information, laboratory examination information, chief complaint, and history of present illness information for UD prediction. We proposed a data-driven UD identification and prediction method (UDIPM) embedding the disease ontology structure. First, we designed a set similarity measure method embedding the disease ontology structure to generate a patient similarity matrix. Second, we applied affinity propagation clustering to divide patients into different clusters, and extracted a typical diagnosis code co-occurrence pattern from each cluster. Furthermore, we identified a UD by fusing visual analysis and a conditional co-occurrence matrix. Finally, we trained five classifiers in combination with feature fusion and feature selection method to unify the diagnosis prediction.

RESULTS

The experimental results on a public electronic medical record dataset showed that the UDIPM could extracted a typical diagnosis code co-occurrence pattern effectively, identified and predicted a UD based on patients' diagnostic and admission information, and outperformed other fusion methods overall.

CONCLUSIONS

The accurate identification and prediction of the UD from a large number of distinct diagnosis codes and multi-source heterogeneous patient admission information in EMRs can provide a data-driven approach to assist better coding integration of diagnosis.