BMC Medical Informatics and Decision Making最新文献

Background: Predicting the length of stay in advance will not only benefit the hospitals both clinically and financially but enable healthcare providers to better decision-making for improved quality of care. More importantly, understanding the length of stay of severe patients who require general anesthesia is key to enhancing health outcomes.

Objective: Here, we aim to discover how machine learning can support resource allocation management and decision-making resulting from the length of stay prediction.

Methods: A retrospective cohort study was conducted from January 2018 to October 2020. A total cohort of 240,000 patients' medical records was collected. The data were collected exclusively for preoperative variables to accurately analyze the predictive factors impacting the postoperative length of stay. The main outcome of this study is an analysis of the length of stay (in days) after surgery until discharge. The prediction was performed with ridge regression, random forest, XGBoost, and multi-layer perceptron neural network models.

Results: The XGBoost resulted in the best performance with an average error within 3 days. Moreover, we explain each feature's contribution over the XGBoost model and further display distinct predictors affecting the overall prediction outcome at the patient level. The risk factors that most importantly contributed to the stay after surgery were as follows: a direct bilirubin laboratory test, department change, calcium chloride medication, gender, and diagnosis with the removal of other organs. Our results suggest that healthcare providers take into account the risk factors such as the laboratory blood test, distributing patients, and the medication prescribed prior to the surgery.

Conclusion: We successfully predicted the length of stay after surgery and provide explainable models with supporting analyses. In summary, we demonstrate the interpretation with the XGBoost model presenting insights on preoperative features and defining higher risk predictors to the length of stay outcome. Our development in explainable models supports the current in-depth knowledge for the future length of stay prediction on electronic medical records that aids the decision-making and facilitation of the operation department.

Background: Multiscale sample entropy (MSE) is a prevalent complexity metric to characterize a time series and has been extensively applied to the physiological signal analysis. However, for a short-term time series, the likelihood of identifying comparable subsequences decreases, leading to higher variability in the Sample Entropy (SampEn) calculation. Additionally, as the scale factor increases in the MSE calculation, the coarse-graining process further shortens the time series. Consequently, each newly generated time series at a larger scale consists of fewer data points, potentially resulting in unreliable or undefined entropy values, particularly at higher scales. To overcome the shortcoming, a modified multiscale Renyi distribution entropy (MMRDis) was proposed in our present work.

Methods: The MMRDis method uses a moving-averaging procedure to acquire a family of time series, each of which quantify the dynamic behaviors of the short-term time series over the multiple temporal scales. Then, MMRDis is constructed for the original and the coarse-grained time series.

Results: The MMRDis method demonstrated superior computational stability on simulated Gaussian white and 1/f noise time series, effectively avoiding undefined measurements in short-term time series. Analysis of short-term heart rate variability (HRV) signals from healthy elderly individuals, healthy young people, and subjects with congestive heart failure and atrial fibrillation revealed that MMRDis complexity measurement values decreased with aging and disease. Additionally, MMRDis exhibited better distinction capability for short-term HRV physiological/pathological signals compared to several recently proposed complexity metrics.

Conclusions: MMRDis was a promising measurement for screening cardiovascular condition within a short time.

Background: Anomaly detection is crucial in healthcare data due to challenges associated with the integration of smart technologies and healthcare. Anomaly in electronic health record can be associated with an insider trying to access and manipulate the data. This article focuses around the anomalies under different contexts.

Methodology: This research has proposed methodology to secure Electronic Health Records (EHRs) within a complex environment. We have employed a systematic approach encompassing data preprocessing, labeling, modeling, and evaluation. Anomalies are not labelled thus a mechanism is required that predicts them with greater accuracy and less false positive results. This research utilized unsupervised machine learning algorithms that includes Isolation Forest and Local Outlier Factor clustering algorithms. By calculating anomaly scores and validating clustering through metrics like the Silhouette Score and Dunn Score, we enhanced the capacity to secure sensitive healthcare data evolving digital threats. Three variations of Isolation Forest (IForest)models (SVM, Decision Tree, and Random Forest) and three variations of Local Outlier Factor (LOF) models (SVM, Decision Tree, and Random Forest) are evaluated based on accuracy, sensitivity, specificity, and F1 Score.

Results: Isolation Forest SVM achieves the highest accuracy of 99.21%, high sensitivity (99.75%) and specificity (99.32%), and a commendable F1 Score of 98.72%. The Isolation Forest Decision Tree also performs well with an accuracy of 98.92% and an F1 Score of 99.35%. However, the Isolation Forest Random Forest exhibits lower specificity (72.84%) than the other models.

Conclusion: The experimental results reveal that Isolation Forest SVM emerges as the top performer showcasing the effectiveness of these models in anomaly detection tasks. The proposed methodology utilizing isolation forest and SVM produced better results by detecting anomalies with less false positives in this specific EHR of a hospital in North England. Furthermore the proposal is also able to identify new contextual anomalies that were not identified in the baseline methodology.

Background: Timely and accurate prediction of disease progress is crucial for facilitating early intervention and treatment for various chronic diseases. However, due to the complicated and longitudinal nature of disease progression, the capacity and completeness of clinical data required for training deep learning models remains a significant challenge. This study aims to explore a new method that reduces data dependency and achieves predictive performance comparable to existing research.

Methods: This study proposed DAPNet, a deep learning-based disease progression prediction model that solely utilizes the comorbidity duration (without relying on multi-modal data or comprehensive medical records) and disease associations from biomedical knowledge graphs to deliver high-performance prediction. DAPNet is the first to apply multi-view graph contrastive learning to disease progression prediction tasks. Compared with other studies on comorbidities, DAPNet innovatively integrates molecular-level disease association information, combines disease co-occurrence and ICD10, and fully explores the associations between diseases; RESULTS: This study validated DAPNet using a de-identified clinical dataset derived from medical claims, which includes 2,714 patients and 10,856 visits. Meanwhile, a kidney dataset (606 patients) based on MIMIC-IV has also been constructed to fully validate its performance. The results showed that DAPNet achieved state-of-the-art performance on the severe pneumonia dataset (F1=0.84, with an improvement of 8.7%), and outperformed the six baseline models on the kidney disease dataset (F1=0.80, with an improvement of 21.3%). Through case analysis, we elucidated the clinical and molecular associations identified by the DAPNet model, which facilitated a better understanding and explanation of potential disease association, thereby providing interpretability for the model.

Conclusions: The proposed DAPNet, for the first time, utilizes comorbidity duration and disease associations network, enabling more accurate disease progression prediction based on a multi-view graph contrastive learning, which provides valuable insights for early diagnosis and treatment of patients. Based on disease association networks, our research has enhanced the interpretability of disease progression predictions.

Background: This study aimed to identify the risk factors of acute ischemic stroke (AIS) occurring during hospitalization in patients following off-pump coronary artery bypass grafting (OPCABG) and utilize Bayesian network (BN) methods to establish predictive models for this disease.

Methods: Data were collected from the electronic health records of adult patients who underwent OPCABG at Beijing Anzhen Hospital from January 2018 to December 2022. Patients were allocated to the training and test sets in an 8:2 ratio according to the principle of randomness. Subsequently, a BN model was established using the training dataset and validated against the testing dataset. The BN model was developed using a tabu search algorithm. Finally, receiver operating characteristic (ROC) and calibration curves were plotted to assess the extent of disparity in predictive performance between the BN and logistic models.

Results: A total of 10,184 patients (mean (SD) age, 62.45 (8.7) years; 2524 (24.7%) females) were enrolled, including 151 (1.5%) with AIS and 10,033 (98.5%) without AIS. Female sex, history of ischemic stroke, severe carotid artery stenosis, high glycated albumin (GA) levels, high D-dimer levels, high erythrocyte distribution width (RDW), and high blood urea nitrogen (BUN) levels were strongly associated with AIS. Type 2 diabetes mellitus (T2DM) was indirectly linked to AIS through GA and BUN. The BN models exhibited superior performance to logistic regression in both the training and testing sets, achieving accuracies of 72.64% and 71.48%, area under the curve (AUC) of 0.899 (95% confidence interval (CI), 0.876-0.921) and 0.852 (95% CI, 0.769-0.935), sensitivities of 91.87% and 89.29%, and specificities of 72.35% and 71.24% (using the optimal cut-off), respectively.

Conclusion: Female gender, IS history, carotid stenosis (> 70%), RDW-CV, GA, D-dimer, BUN, and T2DM are potential predictors of IS in our Chinese cohort. The BN model demonstrated greater efficiency than the logistic regression model. Hence, employing BN models could be conducive to the early diagnosis and prevention of AIS after OPCABG.

Background: Mental health presentations account for a considerable proportion of paramedic workload; however, the decision-making involved in managing these cases is poorly understood. This study aimed to explore how paramedics perceive their clinical decision-making when managing mental health presentations.

Methods: A qualitative descriptive study design was employed. Overall, 73 paramedics participated in semi structured interviews, and data were analyzed from transcribed interviews in NVivo.

Results: Four themes emerged that reflected participants' perceptions: the assessment process, experience, the use of documents and standard procedures, and consultation with other healthcare providers. There were conflicting perceptions about the clinical decision-making process, with perception of role having a potential impact. The dual process theory of clinical decision-making, which includes both analytical and intuitive approaches, was evident in the decision-making process.

Conclusion: Incorporating dual process theory into education and training, which highlights the strengths and weaknesses of analytical and intuitive decision-making, may reduce clinical errors made by cognitive bias. To further support clinical decision-making, additional education and training are warranted to promote critical thinking and clarify the scope of practice and roles when attending to mental health-related presentations.

Background: Lung cancer is characterized by high morbidity and mortality due to the lack of practical early diagnostic and prognostic tools. The present study uses machine learning algorithms to construct a clinical predictive model for non-small cell lung cancer (NSCLC) patients.

Methods: Laboratory indices of the NSCLC patients at their initial visit were collected for quality control and exploratory analysis. By comparing the levels of the above indices between the survival and death groups, the statistically significant indices were selected for subsequent machine learning modeling. Ten machine learning algorithms were then employed to develop the predictive models with survival and recurrence as outcomes, respectively. Moreover, regression models were constructed using the random survival forest algorithm by incorporating the survival time dimension. Finally, critical variables in the optimal model were screened based on the interpretable algorithms to build a decision tree to facilitate clinical application.

Results: 682 patients were enrolled according to the inclusion and exclusion criteria. The preliminary comparison results revealed that except for fast blood glucose, CD₃⁺T cell proportion, NK cell proportion, and CA72-4, there were significant statistical differences in other tumor markers, inflammation, metabolism, and immune-related indices between the survival and death groups (p < 0.01). Subsequently, indices with statistical differences were incorporated into machine learning modeling and evaluation. The results showed that among the ten prognostic models constructed using survival status as the outcome, the neural network model obtained the best predictive performance, with accuracy, sensitivity, specificity, AUC, and precision values of 0.993, 0.987, 1.000, 0.994, and 1.000, respectively. The corresponding SHAP16 algorithm revealed that the top five variables in terms of importance were interleukin6 (IL-6), soluble interleukin2 receptor (sIL-2R), cholesterol, CEA, and Cy211, respectively. The random survival forest model also confirmed the critical role of CEA, sIL-2R, and IL-6 in predicting the prognosis of NSCLC patients. A decision tree model with seven cut-off points based on the above three indices was eventually built for clinical application.

Conclusion: The neural network model exhibited ideal predictive performance in the survival status of NSCLC patients, and the decision tree model constructed based on selected important variables was conducive to rapid bedside prognosis assessment and decision-making.

Background: MRI is critical for diagnosing lumbar spine disorders but its complexity challenges diagnostic accuracy. This study proposes a BERT-based large language model (LLM) to enhance precision in classifying lumbar spine disorders through the integration of MRI data, textual reports, and numerical measurements.

Methods: The segmentation quality of MRI data is evaluated using dice coefficients (cut-off: 0.92) and intersection over union (IoU) metrics (cut-off: 0.88) to ensure precise anatomical feature extraction. The CNN extracts key lumbar spine features, such as lumbar lordotic angle (LLA) and disc heights, which are tokenized as direct scalar values representing positional relationships. A data source of 28,065 patients with various disorders, including degenerative disc disease, spinal stenosis, and spondylolisthesis, is used to establish diagnostic standards. These standards are refined through post-CNN processing of MRI texture features. The BERT-based spinal LLM model integrates these CNN-extracted MRI features and numerical values through early fusion layers.

Results: Segmentation analysis illustrate various lumbar spine disorders and their anatomical changes. The model achieved high performance, with all key metrics nearing 0.9, demonstrating its effectiveness in classifying conditions like spondylolisthesis, herniated disc, and spinal stenosis. External validation further confirmed the model's generalizability across different populations. External validation on 514 expert-validated MRI cases further confirms the model's clinical relevance and generalizability. The BERT-based model classifies 61 combinations of lumbar spine disorders.

Conclusions: The BERT-based spinal LLM significantly improves the precision of lumbar spine disorder classification, supporting accurate diagnosis and treatment planning.

Objective: To construct a highly accurate and interpretable feeding intolerance (FI) risk prediction model for preterm newborns based on machine learning (ML) to assist medical staff in clinical diagnosis.

Methods: In this study, a sample of 350 hospitalized preterm newborns were retrospectively analysed. First, dual feature selection was conducted to identify important feature variables for model construction. Second, ML models were constructed based on the logistic regression (LR), decision tree (DT), support vector machine (SVM) and eXtreme Gradient Boosting (XGBoost) algorithms, after which random sampling and tenfold cross-validation were separately used to evaluate and compare these models and identify the optimal model. Finally, we apply the SHapley Additive exPlanation (SHAP) interpretable framework to analyse the decision-making principles of the optimal model and expound upon the important factors affecting FI in preterm newborns and their modes of action.

Results: The accuracy of XGBoost was 87.62%, and the area under the curve (AUC) was 92.2%. After the application of tenfold cross-validation, the accuracy was 83.43%, and the AUC was 89.45%, which was significantly better than those of the other models. Analysis of the XGBoost model with the SHAP interpretable framework showed that a history of resuscitation, use of probiotics, milk opening time, interval between two stools and gestational age were the main factors affecting the occurrence of FI in preterm newborns, yielding importance scores of 0.632, 0.407, 0.313, 0.313, and 0.258, respectively. A history of resuscitation, first milk opening time ≥ 24 h and interval between stools ≥ 3 days were risk factors for FI, while the use of probiotics and gestational age ≥ 34 weeks were protective factors against FI in preterm newborns.

Conclusions: In practice, we should improve perinatal care and obstetrics with the aim of reducing the occurrence of hypoxia and preterm delivery. When feeding, early milk opening, the use of probiotics, the stimulation of defecation and other measures should be implemented with the aim of reducing the occurrence of FI.

Introduction: Health information systems play a crucial role in the delivery of efficient and effective healthcare. Poor usability is one of the reasons for their lack of acceptance and low usage by users. The aim of this study was to identify the usability problems of a national comprehensive health information system using the concurrent think-aloud method in the recording of childcare data.

Methods: A descriptive cross-sectional study was conducted in the health centers of Kashan University of Medical Sciences, Iran, in 2020. Ten healthcare providers as system's users were purposively selected to evaluate the system. To identify problems, a concurrent think-aloud evaluation was conducted. Two administrators of the system designed scenarios for ten childcare data recording tasks. By analysing the recorded files, usability problems were identified. The severity of the problems was then determined with the help of the users and problems were assigned to usability attributes based on their impact on the user.

Results: A total of 68 unique problems were identified in the system, of which 47.1% were rated as catastrophic problems. The participants assigned 47 problems (69%) to the user satisfaction attribute and 45 problems (66%) to the efficiency attribute; they also did not assign any problems to the effectiveness attribute.

Conclusion: The problems identified in the national comprehensive health information system using the think-aloud method were rated as major and catastrophic, which indicates poor usability of this system. Therefore, resolving the system problems will help increase user satisfaction and system efficiency, allowing more time to be spent on patient care and parent's education as well as improving overall quality of care.