Multiparameter Intelligent Monitoring in Intensive Care II: a public-access intensive care unit database

Explainable machine learning to identify risk factors for unplanned hospital readmissions in Nova Scotian hospitals

OBJECTIVE

A report from the Canadian Institute for Health Information found unplanned hospital readmissions (UHR) common, costly, and potentially avoidable, estimating a $1.8 billion cost to the Canadian healthcare system associated with inpatient readmissions within 30 days of discharge for the studied period (11 months). The first step towards addressing this costly problem is enabling early detection of patients at risk through detecting UHR risk factors.

METHODOLOGY

We utilized Machine Learning and explainability tools to examine risk factors for UHR within 30 days of discharge, utilizing data from Nova Scotian (Canada) healthcare institutions (2015-2022). To the best of our knowledge, our research constitutes the most comprehensive study on UHR risk factors for the province.

RESULTS

We found that predicting UHR solely from healthcare data has limitations, as discharge information often falls short of accurately predicting readmission occurrences. However, despite this inherent limitation, integrating explainability tools offers insights into the underlying factors contributing to readmission risk, empowering medical personnel with information to improve patient care and outcomes. As part of this work, we identify and report risk factors for UHR and build a guideline to support medical personnel's decision-making regarding targeted post-discharge follow-ups. We found that conditions such as heart failure and Chronic Obstructive Pulmonary Disease (COPD) are associated with a higher likelihood of readmission. Patients admitted for procedures related to childbirth have a lower probability of readmission. We studied the impact of the admission type, patient characteristics, and patient stay characteristics on UHR. For example, we found that new and elective admission patients are less likely to be readmitted, while patients who received a transfusion are more likely to be readmitted.

CONCLUSIONS

We validated the risk factors and the guidelines using real-world data. Our results suggested that our proposal correctly identifies risk factors and effectively produces valuable guidelines for medical personnel. The guideline evaluation suggests we can screen half the patients while capturing more than 72% of the readmission episodes. Our study contributes insights into the challenge of identifying risk factors for UHR while providing a practical guideline for healthcare professionals to identify factors influencing patient readmission, particularly within Nova Scotia.

Exploring the limitations of blood pressure estimation using the photoplethysmography signal

Objetive.Hypertension, a leading contributor to cardiovascular morbidity, underscores the need for accurate and continuous blood pressure (BP) monitoring. Photoplethysmography (PPG) emerges as a promising approach for continuous BP monitoring. However, the precision of BP estimates derived from PPG signals has been the subject of ongoing debate, requiring a comprehensive evaluation of their efficacy. This paper aims to provide the potentials and limitations regarding BP estimation from single-site PPG signals.Approach.We developed a calibration-based Siamese ResNet model for BP estimation. We compared the use of normalized PPG (N-PPG) against the normalized invasive arterial BP (N-IABP) signals as input. N-IABP signals, while not directly presenting systolic (SBP) and diastolic (DBP) BP values, are expected to offer more precise estimations than PPG since it is a direct pressure sensor inside the body. Thus, if N-IABP poses challenges in BP estimation, predicting BP from PPG signals might be even more challenging.Main results.Our evaluation, conducted using the AAMI and BHS standards on the VitalDB dataset, revealed that inference using N-IABP signals meet with AAMI standards for both SBP and DBP, with errors of1.29±6.33mmHg for systolic pressure and1.17±5.78for diastolic pressure. In contrast, N-PPG based inference exhibited inferior performance than N-IABP, presenting1.49±11.82mmHg and0.89±7.27mmHg for systolic and diastolic pressure respectively in their best setup.Significance.Our findings establish a critical benchmark for PPG performance, providing realistic expectations for its BP estimation capabilities. We concluded that while PPG signals contain BP-correlated information, they may not suffice for accurate prediction.

A Novel Experience-Driven and Federated Intelligent Threat-Defense Framework in IoMT

The Artificial Intelligence-enabled Internet of Medical Things (AI-IoMT) envisions the connectivity of medical devices encompassing advanced computing technologies to empower large-scale intelligent healthcare networks. The AI-IoMT continuously monitors patients' health and vital computations via IoMT sensors with enhanced resource utilization for providing progressive medical care services. However, the security concerns of these autonomous systems against potential threats are still underdeveloped. Since these IoMT sensor networks carry a bulk of sensitive data, they are susceptible to unobservable False Data Injection Attacks (FDIA), thus jeopardizing patients' health. This paper presents a novel threat-defense analysis framework that establishes an experience-driven approach based on a deep deterministic policy gradient to inject false measurements into IoMT sensors, computing vitals, causing patients' health instability. Subsequently, a privacy-preserved and optimized federated intelligent FDIA detector is deployed to detect malicious activity. The proposed method is parallelizable and computationally efficient to work collaboratively in a dynamic domain. Compared to existing techniques, the proposed threat-defense framework is able to thoroughly analyze severe systems' security holes and combats the risk with lower computing cost and high detection accuracy along with preserving the patients' data privacy.

ECG-Based Biometric Recognition: A Survey of Methods and Databases

This work presents a comprehensive and chronologically ordered survey of existing studies and data sources on Electrocardiogram (ECG) based biometric recognition systems. This survey is organized in terms of the two main goals pursued in it: first, a description of the main ECG features and recognition techniques used in the existing literature, including a comprehensive compilation of references; second, a survey of the ECG databases available and used by the referenced studies. The most relevant characteristics of the databases are identified, and a comprehensive compilation of databases is given. To date, no other work has presented such a complete overview of both studies and data sources for ECG-based biometric recognition. Readers interested in the subject can obtain an understanding of the state of the art, easily identifying specific key papers by using different criteria, and become aware of the databases where they can test their novel algorithms.

Machine learning analysis of integrated ABP and PPG signals towards early detection of coronary artery disease

Every year, Coronary Artery Disease (CAD) claims lives of over a million people. CAD occurs when the coronary arteries, responsible for supplying oxygenated blood to the heart, get occluded due to plaque deposits on their inner walls. The most critical fact about this disease is that it develops gradually over the years and by the time symptomatic changes such as angina or shortness of breath appear, the disease has already become severe. The overall aim of the proposed work is to detect CAD efficiently in its early stage while utilizing (radial) arterial blood pressure (ABP) along with photoplethysmogram (PPG) signals so that necessary clinical measures may be taken timely. To achieve this objective, firstly, ABP and PPG data of 73 CAD and 64 non-CAD (not suffering from any cardiac condition) subjects have been collected from MIMIC-II waveform database with matched subset. Secondly, the collected data is pre-processed using band pass filters having bandwidths of 2.5 to 16 Hz and 1.5 to 16 Hz for ABP and PPG respectively. Thirdly, nineteen features have been extracted from each of the two signals; some of the key features include mean of pulse duration, mean of rising slope and ratio of low frequency to high frequency. Finally, extensive analysis on CAD and non-CAD classification is carried out on the basis of extracted features while employing state-of-the-art classifiers such as support vector machines (SVM), K-nearest neighbors (KNN) and neural networks(NN). The numerical experiments have led to the interpretation that neural network outperforms other classifiers, claiming an accuracy of about 90%. Moreover, accuracy of the proposed approach is found to be better than the state-of-the-art works reported in literature where one of or combinations of cardiovascular signals, namely, electrocardiogram (ECG), phonocardiogram (PCG) and photoplethysmogram (PPG) have been utilized for the CAD detection.

Evaluating the influence MRSA Co-infection on 28-day mortality among sepsis patients: insights from the MIMIC-IV database

Background

Sepsis remains a leading cause of mortality in intensive care units (ICUs), with methicillin-resistant Staphylococcus aureus (MRSA) infections presenting significant treatment challenges. The impact of MRSA co-infection on sepsis outcomes necessitates further exploration.

Methods

We conducted a retrospective observational cohort study using the Medical Information Mart for Critical Care IV (MIMIC-IV-2.2) database. This cohort study included sepsis patients, scrutinizing baseline characteristics, MRSA co-infection, antimicrobial susceptibility, and their relations to mortality through Cox regression and Kaplan-Meier analyses.

Results

Among 453 sepsis patients analyzed, significant baseline characteristic differences were observed between survivors (N = 324) and non-survivors (N = 129). Notably, non-survivors were older (70.52 ± 14.95 vs. 64.42 ± 16.05, P < 0.001), had higher lactate levels (2.82 ± 1.76 vs. 2.04 ± 1.56 mmol/L, P < 0.001), and higher SOFA scores (8.36 ± 4.18 vs. 6.26 ± 3.65, P < 0.001). Cox regression highlighted SOFA score (HR = 1.122, P = 0.003), body temperature (HR = 0.825, P = 0.048), and age (HR = 1.030, P = 0.004) as significant predictors of 28-day mortality. MRSA co-infection was found in 98.7% of cases without a significant effect on 28-day mortality (P = 0.9). However, sensitivity to cephalosporins, meropenem, and piperacillin/tazobactam was associated with reduced mortality. The area under the ROC curve for the combined model of age, SOFA, and body temperature was 0.73, indicating a moderate predictive value for 28-day mortality.

Conclusion

While MRSA co-infection's direct impact on 28-day sepsis mortality is minimal, antimicrobial sensitivity, especially to cephalosporins, meropenem, and piperacillin/tazobactam, plays a critical role in improving outcomes, underscoring the importance of antimicrobial stewardship and personalized treatment strategies in sepsis care.

Predicting Mortality in Sepsis-Associated Acute Respiratory Distress Syndrome: A Machine Learning Approach Using the MIMIC-III Database

BackgroundTo develop and validate a mortality prediction model for patients with sepsis-associated Acute Respiratory Distress Syndrome (ARDS).MethodsThis retrospective cohort study included 2466 patients diagnosed with sepsis and ARDS within 24 h of ICU admission. Demographic, clinical, and laboratory parameters were extracted from Medical Information Mart for Intensive Care III (MIMIC-III) database. Feature selection was performed using the Boruta algorithm, followed by the construction of seven ML models: logistic regression, Naive Bayes, k-nearest neighbor, support vector machine, decision tree, Random Forest, and extreme gradient boosting. Model performance was evaluated using the area under the receiver operating characteristic curve, accuracy, sensitivity, specificity, positive predictive value, and negative predictive value.ResultsThe study identified 24 variables significantly associated with mortality. The optimal ML model, a Random Forest model, demonstrated an AUC of 0.8015 in the test set, with high accuracy and specificity. The model highlighted the importance of blood urea nitrogen, age, urine output, Simplified Acute Physiology Score II, and albumin levels in predicting mortality.ConclusionsThe model's superior predictive performance underscores the potential for integrating advanced analytics into clinical decision-making processes, potentially improving patient outcomes and resource allocation in critical care settings.

[Research and application implementation of the Internet of Things scheme for intensive care unit medical equipment]

The intensive care unit (ICU) is a highly equipment-intensive area with a wide variety of medical devices, and the accuracy and timeliness of medical equipment data collection are highly demanded. The integration of the Internet of Things (IoT) into ICU medical devices is of great significance for enhancing the quality of medical care and nursing, as well as for the advancement of digital and intelligent ICUs. This study focuses on the construction of the IOT for ICU medical devices and proposes innovative solutions, including the overall architecture design, devices connection, data collection, data standardization, platform construction and application implementation. The overall architecture was designed according to the perception layer, network layer, platform layer and application layer; three modes of device connection and data acquisition were proposed; data standardization based on Integrating the Healthcare Enterprise-Patient Care Device (IHE-PCD) was proposed. This study was practically verified in the Chinese People's Liberation Army General Hospital, a total of 122 devices in four ICU wards were connected to the IoT, storing 21.76 billion data items, with a data volume of 12.5 TB, which solved the problem of difficult systematic medical equipment data collection and data integration in ICUs. The remarkable results achieved proved the feasibility and reliability of this study. The research results of this paper provide a solution reference for the construction of hospital ICU IoT, offer more abundant data for medical big data analysis research, which can support the improvement of ICU medical services and promote the development of ICU to digitalization and intelligence.

A review of machine learning methods for non-invasive blood pressure estimation

Blood pressure is a very important clinical measurement, offering valuable insights into the hemodynamic status of patients. Regular monitoring is crucial for early detection, prevention, and treatment of conditions like hypotension and hypertension, both of which increasing morbidity for a wide variety of reasons. This monitoring can be done either invasively or non-invasively and intermittently vs. continuously. An invasive method is considered the gold standard and provides continuous measurement, but it carries higher risks of complications such as infection, bleeding, and thrombosis. Non-invasive techniques, in contrast, reduce these risks and can provide intermittent or continuous blood pressure readings. This review explores modern machine learning-based non-invasive methods for blood pressure estimation, discussing their advantages, limitations, and clinical relevance.

rU-Net, Multi-Scale Feature Fusion and Transfer Learning: Unlocking the Potential of Cuffless Blood Pressure Monitoring With PPG and ECG

This study introduces an innovative deep-learning model for cuffless blood pressure estimation using PPG and ECG signals, demonstrating state-of-the-art performance on the largest clean dataset, PulseDB. The rU-Net architecture, a fusion of U-Net and ResNet, enhances both generalization and feature extraction accuracy. Accurate multi-scale feature capture is facilitated by short-time Fourier transform (STFT) time-frequency distributions and multi-head attention mechanisms, allowing data-driven feature selection. The inclusion of demographic parameters as supervisory information further elevates performance. On the calibration-based dataset, our model excels, achieving outstanding accuracy (SBP MAE ± std: 4.49 ± 4.86 mmHg, DBP MAE ± std: 2.69 ± 3.10 mmHg), surpassing AAMI standards and earning a BHS Grade A rating. Addressing the challenge of calibration-free data, we propose a fine-tuning-based transfer learning approach. Remarkably, with only 10% data transfer, our model attains exceptional accuracy (SBP MAE ± std: 4.14 ± 5.01 mmHg, DBP MAE ± std: 2.48 ± 2.93 mmHg). This study sets the stage for the development of highly accurate and reliable wearable cuffless blood pressure monitoring devices.

The Complete Inpatient Record Using Comprehensive Electronic Data (CIRCE) project: A team-based approach to clinically validated, research-ready electronic health record data

Introduction

The rapid adoption of electronic health record (EHR) systems has resulted in extensive archives of data relevant to clinical research, hospital operations, and the development of learning health systems. However, EHR data are not frequently available, cleaned, standardized, validated, and ready for use by stakeholders. We describe an in-progress effort to overcome these challenges with cooperative, systematic data extraction and validation.

Methods

A multi-disciplinary team of investigators collaborated to create the Complete Inpatient Record Using Comprehensive Electronic Data (CIRCE) Project dataset, which captures EHR data from six hospitals within the University of Pennsylvania Health System. Analysts and clinical researchers jointly iteratively reviewed SQL queries and their output to validate desired data elements. Data from patients aged ≥18 years with at least one encounter at an acute care hospital or hospice occurring since 7/1/2017 were included. The CIRCE Project includes three layers: (1) raw data comprised of direct SQL query output, (2) cleaned data with errors removed, and (3) transformed data with standardized implementations of commonly used case definitions and clinical scores.

Results

Between July 1, 2017 and December 31, 2023, the dataset captured 1 629 920 encounters from 740 035 patients. Most encounters were emergency department only visits (n = 965 834, 59.3%), followed by inpatient admissions without an intensive care unit admission (n = 518 367, 23.7%). The median age was 46.9 years (25th-75th percentiles = 31.1-64.7) at the time of the first encounter. Most patients were female (n = 418 303, 56.5%), a significant proportion were of non-White race (n = 272 018, 36.8%), and 54 625 (7.4%) were of Hispanic/Latino ethnicity.

Conclusions

The CIRCE Project represents a novel cooperative research model to capture clinically validated EHR data from a large diverse academic health system in the greater Philadelphia region and is designed to facilitate collaboration and data sharing to support learning health system activities. Ultimately, these data will be de-identified and converted to a publicly available resource.

Robust modelling of arterial blood pressure reconstruction from photoplethysmography

Blood pressure is a crucial indicator of cardiovascular disease, and arterial blood pressure (ABP) waveforms contain information that reflects the cardiovascular status. We propose a novel deep-learning method that converts photoplethysmogram (PPG) signals into ABP waveforms. We used [Formula: see text]-Net as a feature extractor and designed a Bi-block to capture individualised time information in encoder feature extraction. We further enhanced the prediction accuracy of the ABP waveforms by applying a combined loss function to each layer of deep supervision. We also propose a total error index (TEI) to measure overall performance. Furthermore, we extended our method from the UCI dataset to the VitalDB dataset, achieving mean absolute error ± standard deviation (MAE ± STD) values of 2.48 ± 1.95, 1.42 ± 1.42, and 1.48 ± 1.36 mmHg for systolic blood pressure (SBP), diastolic blood pressure (DBP), and mean arterial pressure (MAP) in UCI dataset, and 2.16 ± 1.53, 1.12 ± 0.59, and 1.35 ± 0.84 mmHg in VitalDB dataset, respectively. The mean ± STD values of the TEI index are 0.29 ± 0.10 in UCI dataset and 0.29 ± 0.15 in VitalDB dataset. These results demonstrate the superiority of the proposed method over existing methods and its robustness to different sampling frequencies and devices.

Optimized robust learning framework based on big data for forecasting cardiovascular crises

Numerous Deep Learning (DL) scenarios have been developed for evolving new healthcare systems that leverage large datasets, distributed computing, and the Internet of Things (IoT). However, the data used in these scenarios tend to be noisy, necessitating the incorporation of robust pre-processing techniques, including data cleaning, preparation, normalization, and addressing imbalances. These steps are crucial for generating a robust dataset for training. Designing frameworks capable of handling such data without compromising efficiency is essential to ensuring robustness. This research aims to propose a novel healthcare framework that selects the best features and enhances performance. This robust deep learning framework, called (R-DLH2O), is designed for forecasting cardiovascular crises. Unlike existing methods, R-DLH2O integrates five distinct phases: robust pre-processing, feature selection, feed-forward neural network, prediction, and performance evaluation. This multi-phase approach ensures superior accuracy and efficiency in crisis prediction, offering a significant advancement in healthcare analytics. H2O is utilized in the R-DLH2O framework for processing big data. The main improvement of this paper lies in the unique form of the Whale Optimization Algorithm (WOA), specifically the Modified WOA (MWOA). The Gaussian distribution approach for random walks was employed with the diffusion strategy to choose the optimal MWOA solution during the growth phase. To validate the R-DLH2O framework, six performance tests were conducted. Surprisingly, the MWOA-2 outperformed other heuristic algorithms in speed, despite exhibiting lower accuracy and scalability. The suggested MWOA was further analyzed using benchmark functions from CEC2005, demonstrating its advantages in accuracy and robustness over WOA. These findings highlight that the framework's processing time is 436 s, mean per-class error is 0.150125, accuracy 95.93%, precision 92.57%, and recall 93.6% across all datasets. These findings highlight the framework's potential to produce significant and robust results, outperforming previous frameworks concerning time and accuracy.

MIMIC-BP: A curated dataset for blood pressure estimation

Blood pressure (BP) is one of the most prominent indicators of potential cardiovascular disorders. Traditionally, BP measurement relies on inflatable cuffs, which is inconvenient and limit the acquisition of such important health-related information in general population. Based on large amounts of well-collected and annotated data, deep-learning approaches present a generalization potential that arose as an alternative to enable more pervasive approaches. However, most existing work in this area currently uses datasets with limitations, such as lack of subject identification and severe data imbalance that can result in data leakage and algorithm bias. Thus, to offer a more properly curated source of information, we propose a derivative dataset composed of 380 hours of the most common biomedical signals, including arterial blood pressure, photoplethysmography, and electrocardiogram for 1,524 anonymized subjects, each having 30 segments of 30 seconds of those signals. We also validated the proposed dataset through experiments using state-of-the-art deep-learning methods, as we highlight the importance of standardized benchmarks for calibration-free blood pressure estimation scenarios.

Machine Learning Applied to Reference Signal-Less Detection of Motion Artifacts in Photoplethysmographic Signals: A Review

Machine learning algorithms have brought remarkable advancements in detecting motion artifacts (MAs) from the photoplethysmogram (PPG) with no measured or synthetic reference data. However, no study has provided a synthesis of these methods, let alone an in-depth discussion to aid in deciding which one is more suitable for a specific purpose. This narrative review examines the application of machine learning techniques for the reference signal-less detection of MAs in PPG signals. We did not consider articles introducing signal filtering or decomposition algorithms without previous identification of corrupted segments. Studies on MA-detecting approaches utilizing multiple channels and additional sensors such as accelerometers were also excluded. Despite its promising results, the literature on this topic shows several limitations and inconsistencies, particularly those regarding the model development and testing process and the measures used by authors to support the method's suitability for real-time applications. Moreover, there is a need for broader exploration and validation across different body parts and a standardized set of experiments specifically designed to test and validate MA detection approaches. It is essential to provide enough elements to enable researchers and developers to objectively assess the reliability and applicability of these methods and, therefore, obtain the most out of them.

Raw Photoplethysmography as an Enhancement for Research-Grade Wearable Activity Monitors

Wearable monitors continue to play a critical role in scientific assessments of physical activity. Recently, research-grade monitors have begun providing raw data from photoplethysmography (PPG) alongside standard raw data from inertial sensors (accelerometers and gyroscopes). Raw PPG enables granular and transparent estimation of cardiovascular parameters such as heart rate, thus presenting a valuable alternative to standard PPG methodologies (most of which rely on consumer-grade monitors that provide only coarse output from proprietary algorithms). The implications for physical activity assessment are tremendous, since it is now feasible to monitor granular and concurrent trends in both movement and cardiovascular physiology using a single noninvasive device. However, new users must also be aware of challenges and limitations that accompany the use of raw PPG data. This viewpoint paper therefore orients new users to the opportunities and challenges of raw PPG data by presenting its mechanics, pitfalls, and availability, as well as its parallels and synergies with inertial sensors. This includes discussion of specific applications to the prediction of energy expenditure, activity type, and 24-hour movement behaviors, with an emphasis on areas in which raw PPG data may help resolve known issues with inertial sensing (eg, measurement during cycling activities). We also discuss how the impact of raw PPG data can be maximized through the use of open-source tools when developing and disseminating new methods, similar to current standards for raw accelerometer and gyroscope data. Collectively, our comments show the strong potential of raw PPG data to enhance the use of research-grade wearable activity monitors in science over the coming years.

Identification of Time-Series Pattern Marker in Its Application to Mortality Analysis of Pneumonia Patients in Intensive Care Unit

Electronic Health Records (EHRs) are a significant source of big data used to track health variables over time. The analysis of EHR data can uncover medical markers or risk factors, aiding in the diagnosis and monitoring of diseases. We introduce a novel method for identifying markers with various temporal trend patterns, including monotonic and fluctuating trends, using machine learning models such as Long Short-Term Memory (LSTM). By applying our method to pneumonia patients in the intensive care unit using the MIMIC-III dataset, we identified markers exhibiting both monotonic and fluctuating trends. Specifically, monotonic markers such as red cell distribution width, urea nitrogen, creatinine, calcium, morphine sulfate, bicarbonate, sodium, troponin T, albumin, and prothrombin time were more frequently observed in the mortality group compared to the recovery group throughout the 10-day period before discharge. Conversely, fluctuating trend markers such as dextrose in sterile water, polystyrene sulfonate, free calcium, and glucose were more frequently observed in the mortality group as the discharge date approached. Our study presents a method for detecting time-series pattern markers in EHR data that respond differently according to disease progression. These markers can contribute to monitoring disease progression and enable stage-specific treatment, thereby advancing precision medicine.

Using machine learning to predict the risk of short-term and long-term death in acute kidney injury patients after commencing CRRT

BACKGROUND

The mortality rate and prognosis of short-term and long-term acute kidney injury (AKI) patients who undergo continuous renal replacement therapy (CRRT) are different. Setting up risk stratification tools for both short-term and long-term deaths is highly important for clinicians.

METHOD

A total of 1535 AKI patients receiving CRRT were included in this study, with 1144 from the training set (the Dryad database) and 391 from the validation set (MIMIC IV database). A model for predicting mortality within 10 and 90 days was built using nine different machine learning (ML) algorithms. AUROC, F1-score, accuracy, sensitivity, specificity, precision, and calibration curves were used to assess the predictive performance of various ML models.

RESULTS

A total of 420 (31.1%) deaths occurred within 10 days, and 1080 (68.8%) deaths occurred within 90 days. The random forest (RF) model performed best in both predicting 10-day (AUROC: 0.80, 95% CI: 0.74-0.84; accuracy: 0.72, 95% CI: 0.67-0.76; F1-score: 0.59) and 90-day mortality (AUROC: 0.78, 95% CI: 0.73-0.83; accuracy: 0.73, 95% CI: 0.69-0.78; F1-score: 0.80). The importance of the feature shows that SOFA scores are rated as the most important risk factor for both 10-day and 90-day mortality.

CONCLUSION

Our study, utilizing multiple machine learning models, estimates the risk of short-term and long-term mortality among AKI patients who commence CRRT. The results demonstrated that the prognostic factors for short-term and long-term mortality are different. The RF model has the best prediction performance and has valuable potential for clinical application.

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.

A Systematic Approach Focused on Machine Learning Models for Exploring the Landscape of Physiological Measurement and Estimation Using Photoplethysmography (PPG)

A non-invasive optical technique known as photoplethysmography (PPG) can be used to provide various physiological measurements and estimations. PPG can be used to assess cardiovascular disease (CVD). Hypertension is a primary risk factor for CVD and a major health problem worldwide. PPG is popular because of its important applications in the evaluation of cardiac activity, variations in venous blood volume, blood oxygen saturation, blood pressure and heart rate variability, etc. In this study, we provide a comprehensive analysis of the extraction of various physiological parameters using PPG waveforms. In addition, we focused on the role of machine learning (ML) models used for the estimation of blood pressure and hypertension classification based on PPG waveforms to make future research and innovation recommendations. This study will be helpful for researchers, scientists, and medical practitioners working on PPG waveforms for monitoring, screening, and diagnosis, as a comparative study or reference.

An Extensible Evaluation Framework Applied to Clinical Text Deidentification Natural Language Processing Tools: Multisystem and Multicorpus Study

BACKGROUND

Clinical natural language processing (NLP) researchers need access to directly comparable evaluation results for applications such as text deidentification across a range of corpus types and the means to easily test new systems or corpora within the same framework. Current systems, reported metrics, and the personally identifiable information (PII) categories evaluated are not easily comparable.

OBJECTIVE

This study presents an open-source and extensible end-to-end framework for comparing clinical NLP system performance across corpora even when the annotation categories do not align.

METHODS

As a use case for this framework, we use 6 off-the-shelf text deidentification systems (ie, CliniDeID, deid from PhysioNet, MITRE Identity Scrubber Toolkit [MIST], NeuroNER, National Library of Medicine [NLM] Scrubber, and Philter) across 3 standard clinical text corpora for the task (2 of which are publicly available) and 1 private corpus (all in English), with annotation categories that are not directly analogous. The framework is built on shell scripts that can be extended to include new systems, corpora, and performance metrics. We present this open tool, multiple means for aligning PII categories during evaluation, and our initial timing and performance metric findings. Code for running this framework with all settings needed to run all pairs are available via Codeberg and GitHub.

RESULTS

From this case study, we found large differences in processing speed between systems. The fastest system (ie, MIST) processed an average of 24.57 (SD 26.23) notes per second, while the slowest (ie, CliniDeID) processed an average of 1.00 notes per second. No system uniformly outperformed the others at identifying PII across corpora and categories. Instead, a rich tapestry of performance trade-offs emerged for PII categories. CliniDeID and Philter prioritize recall over precision (with an average recall 6.9 and 11.2 points higher, respectively, for partially matching spans of text matching any PII category), while the other 4 systems consistently have higher precision (with MIST's precision scoring 20.2 points higher, NLM Scrubber scoring 4.4 points higher, NeuroNER scoring 7.2 points higher, and deid scoring 17.1 points higher). The macroaverage recall across corpora for identifying names, one of the more sensitive PII categories, included deid (48.8%) and MIST (66.9%) at the low end and NeuroNER (84.1%), NLM Scrubber (88.1%), and CliniDeID (95.9%) at the high end. A variety of metrics across categories and corpora are reported with a wider variety (eg, F2-score) available via the tool.

CONCLUSIONS

NLP systems in general and deidentification systems and corpora in our use case tend to be evaluated in stand-alone research articles that only include a limited set of comparators. We hold that a single evaluation pipeline across multiple systems and corpora allows for more nuanced comparisons. Our open pipeline should reduce barriers to evaluation and system advancement.

Salzburg Intensive Care database (SICdb): a detailed exploration and comparative analysis with MIMIC-IV

The utilization of artificial intelligence (AI) in healthcare is on the rise, demanding increased accessibility to (public) medical data for benchmarking. The digitization of healthcare in recent years has facilitated medical data scientists' access to extensive hospital data, fostering AI-based research. A notable addition to this trend is the Salzburg Intensive Care database (SICdb), made publicly available in early 2023. Covering over 27 thousand intensive care admissions at the University Hospital Salzburg from 2013 to 2021, this dataset presents a valuable resource for AI-driven investigations. This article explores the SICdb and conducts a comparative analysis with the widely recognized Medical Information Mart for Intensive Care - version IV (MIMIC-IV) database. The comparison focuses on key aspects, emphasizing the availability and granularity of data provided by the SICdb, particularly vital signs and laboratory measurements. The analysis demonstrates that the SICdb offers more detailed information with higher data availability and temporal resolution for signal data, especially for vital signs, compared to the MIMIC-IV. This is advantageous for longitudinal studies of patients' health conditions in the intensive care unit. The SICdb provides a valuable resource for medical data scientists and researchers. The database offers comprehensive and diverse healthcare data in a European country, making it well suited for benchmarking and enhancing AI-based healthcare research. The importance of ongoing efforts to expand and make public datasets available for advancing AI applications in the healthcare domain is emphasized by the findings.

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.

A new cloud-based method for composition of healthcare services using deep reinforcement learning and Kalman filtering

Healthcare has significantly contributed to the well-being of individuals around the globe; nevertheless, further benefits could be derived from a more streamlined healthcare system without incurring additional costs. Recently, the main attributes of cloud computing, such as on-demand service, high scalability, and virtualization, have brought many benefits across many areas, especially in medical services. It is considered an important element in healthcare services, enhancing the performance and efficacy of the services. The current state of the healthcare industry requires the supply of healthcare products and services, increasing its viability for everyone involved. Developing new approaches for discovering and selecting healthcare services in the cloud has become more critical due to the rising popularity of these kinds of services. As a result of the diverse array of healthcare services, service composition enables the execution of intricate operations by integrating multiple services' functionalities into a single procedure. However, many methods in this field encounter several issues, such as high energy consumption, cost, and response time. This article introduces a novel layered method for selecting and evaluating healthcare services to find optimal service selection and composition solutions based on Deep Reinforcement Learning (Deep RL), Kalman filtering, and repeated training, addressing the aforementioned issues. The results revealed that the proposed method has achieved acceptable results in terms of availability, reliability, energy consumption, and response time when compared to other methods.

[Artificial intelligence in intensive care medicine]

The integration of artificial intelligence (AI) into intensive care medicine has made considerable progress in recent studies, particularly in the areas of predictive analytics, early detection of complications, and the development of decision support systems. The main challenges remain availability and quality of data, reduction of bias and the need for explainable results from algorithms and models. Methods to explain these systems are essential to increase trust, understanding, and ethical considerations among healthcare professionals and patients. Proper training of healthcare professionals in AI principles, terminology, ethical considerations, and practical application is crucial for the successful use of AI. Careful assessment of the impact of AI on patient autonomy and data protection is essential for its responsible use in intensive care medicine. A balance between ethical and practical considerations must be maintained to ensure patient-centered care while complying with data protection regulations. Synergistic collaboration between clinicians, AI engineers, and regulators is critical to realizing the full potential of AI in intensive care medicine and maximizing its positive impact on patient care. Future research and development efforts should focus on improving AI models for real-time predictions, increasing the accuracy and utility of AI-based closed-loop systems, and overcoming ethical, technical, and regulatory challenges, especially in generative AI systems.

Automatic identification of hypertension and assessment of its secondary effects using artificial intelligence: A systematic review (2013-2023)

Artificial Intelligence (AI) techniques are increasingly used in computer-aided diagnostic tools in medicine. These techniques can also help to identify Hypertension (HTN) in its early stage, as it is a global health issue. Automated HTN detection uses socio-demographic, clinical data, and physiological signals. Additionally, signs of secondary HTN can also be identified using various imaging modalities. This systematic review examines related work on automated HTN detection. We identify datasets, techniques, and classifiers used to develop AI models from clinical data, physiological signals, and fused data (a combination of both). Image-based models for assessing secondary HTN are also reviewed. The majority of the studies have primarily utilized single-modality approaches, such as biological signals (e.g., electrocardiography, photoplethysmography), and medical imaging (e.g., magnetic resonance angiography, ultrasound). Surprisingly, only a small portion of the studies (22 out of 122) utilized a multi-modal fusion approach combining data from different sources. Even fewer investigated integrating clinical data, physiological signals, and medical imaging to understand the intricate relationships between these factors. Future research directions are discussed that could build better healthcare systems for early HTN detection through more integrated modeling of multi-modal data sources.

Fair non-contact blood pressure estimation using imaging photoplethysmography

Hypertension is typically manifested as a latent symptom that requires detection through specialized equipment. This poses an inconvenience for individuals who need to undergo long-term blood pressure monitoring in their daily lives. Therefore, there is a need for a portable, non-contact method for estimating blood pressure. However, current non-contact blood pressure estimation methods often rely on relatively narrow datasets, lacking a broad range of blood pressure distributions. Additionally, their applicability is confined to controlled experimental environments. This study proposes a non-contact blood pressure estimation method suitable for various life scenarios, encompassing multiple age groups, diverse ethnicities, and individuals with different skin tones. The aim is to enhance the practicality and accuracy of existing non-contact blood pressure estimation methods. The research extracts the imaging photoplethysmogram (IPPG) signal from facial videos and processes the signal through four layers of filtering operations to obtain an IPPG signal reflecting pulse wave variations. A CNN+BiLSTM+GRU network structure is constructed to improve the accuracy of current non-contact blood pressure estimation methods. In comparison to existing approaches, the mean absolute error (MAE) for systolic blood pressure (SBP) and diastolic blood pressure (DBP) is reduced by 13.6% and 16.4%, respectively.

Artificial intelligence in the NICU to predict extubation success in prematurely born infants

OBJECTIVES

Mechanical ventilation in prematurely born infants, particularly if prolonged, can cause long term complications including bronchopulmonary dysplasia. Timely extubation then is essential, yet predicting its success remains challenging. Artificial intelligence (AI) may provide a potential solution.

CONTENT

A narrative review was undertaken to explore AI's role in predicting extubation success in prematurely born infants. Across the 11 studies analysed, the range of reported area under the receiver operator characteristic curve (AUC) for the selected prediction models was between 0.7 and 0.87. Only two studies implemented an external validation procedure. Comparison to the results of clinical predictors was made in two studies. One group reported a logistic regression model that outperformed clinical predictors on decision tree analysis, while another group reported clinical predictors outperformed their artificial neural network model (AUCs: ANN 0.68 vs. clinical predictors 0.86). Amongst the studies there was an heterogenous selection of variables for inclusion in prediction models, as well as variations in definitions of extubation failure.

SUMMARY

Although there is potential for AI to enhance extubation success, no model's performance has yet surpassed that of clinical predictors.

OUTLOOK

Future studies should incorporate external validation to increase the applicability of the models to clinical settings.

Unlocking the Secrets Behind Advanced Artificial Intelligence Language Models in Deidentifying Chinese-English Mixed Clinical Text: Development and Validation Study

BACKGROUND

The widespread use of electronic health records in the clinical and biomedical fields makes the removal of protected health information (PHI) essential to maintain privacy. However, a significant portion of information is recorded in unstructured textual forms, posing a challenge for deidentification. In multilingual countries, medical records could be written in a mixture of more than one language, referred to as code mixing. Most current clinical natural language processing techniques are designed for monolingual text, and there is a need to address the deidentification of code-mixed text.

OBJECTIVE

The aim of this study was to investigate the effectiveness and underlying mechanism of fine-tuned pretrained language models (PLMs) in identifying PHI in the code-mixed context. Additionally, we aimed to evaluate the potential of prompting large language models (LLMs) for recognizing PHI in a zero-shot manner.

METHODS

We compiled the first clinical code-mixed deidentification data set consisting of text written in Chinese and English. We explored the effectiveness of fine-tuned PLMs for recognizing PHI in code-mixed content, with a focus on whether PLMs exploit naming regularity and mention coverage to achieve superior performance, by probing the developed models' outputs to examine their decision-making process. Furthermore, we investigated the potential of prompt-based in-context learning of LLMs for recognizing PHI in code-mixed text.

RESULTS

The developed methods were evaluated on a code-mixed deidentification corpus of 1700 discharge summaries. We observed that different PHI types had preferences in their occurrences within the different types of language-mixed sentences, and PLMs could effectively recognize PHI by exploiting the learned name regularity. However, the models may exhibit suboptimal results when regularity is weak or mentions contain unknown words that the representations cannot generate well. We also found that the availability of code-mixed training instances is essential for the model's performance. Furthermore, the LLM-based deidentification method was a feasible and appealing approach that can be controlled and enhanced through natural language prompts.

CONCLUSIONS

The study contributes to understanding the underlying mechanism of PLMs in addressing the deidentification process in the code-mixed context and highlights the significance of incorporating code-mixed training instances into the model training phase. To support the advancement of research, we created a manipulated subset of the resynthesized data set available for research purposes. Based on the compiled data set, we found that the LLM-based deidentification method is a feasible approach, but carefully crafted prompts are essential to avoid unwanted output. However, the use of such methods in the hospital setting requires careful consideration of data security and privacy concerns. Further research could explore the augmentation of PLMs and LLMs with external knowledge to improve their strength in recognizing rare PHI.

Prognostic value of the systemic immuno-inflammatory index in critically ill patients with vertebral fractures

Inflammation plays a critical role in vertebral fractures. However, there is a lack of sufficient evidence regarding the prognostic significance of the systemic immuno-inflammatory index (SII), a novel marker of systemic inflammation, in patients with vertebral fractures. In this study, we aimed to assess the predictive value of SII in critically ill patients with vertebral fractures. The data were from the Medical Information Mart for Intensive Care III (MIMIC-III) version 1.4 and Wenzhou Hospital of Traditional Chinese Medicine. The cutoff values for SII were determined using the receiver operating characteristic curve, and the subjects were grouped accordingly. The clinical outcome measured was mortality within 30 days, 90 days, or 1 year. The following formula was used to calculate the SII: SII = (platelet count) × (neutrophil count)/ (lymphocyte count). Cox proportional-hazard models were employed to assess the relationship between SII and survival. Additionally, propensity score matching analysis and COX models were utilized to examine the association between SII and survival outcomes. The Pearson correlation test confirmed the correlation between SII and vertebral T-values measured by bone mineral density and pain indicator. A total of 354 patients were finally included from MIMIC-III in the univariate analysis, for the 30-day mortality, SII ≥ 3164 group, the hazard ratio (HR) (95% confidence interval) was 1.71 (1.01, 2.94). After adjusting for age, gender, race, anion gap, creatinine, systolic blood pressure (SBP), DBP MBP, SOFA, acute physiologic score III, chronic kidney disease, and SAPS II, SII ≥ 3164 was found to be an independent significant risk factor for death in patients (HR = 1.85, 95% CI: 1.06-3.24, P = .0315). A similar trend was observed for 90-day mortality and 1-year mortality. Propensity scores matching analysis further confirmed the association of SII and the prognosis of patients. Our validation results were consistent with it. Besides, the Pearson correlation test confirmed a significant correlation between SII and vertebral T-values measured by bone mineral density and pain indicator. The study findings revealed that SII is an independent predictor of mortality in patients with vertebral fractures. This indicates that SII can serve as a reliable and easily accessible prognostic indicator for newly diagnosed critically ill patients with vertebral fractures.

Hyperbolic graph convolutional neural network with contrastive learning for automated ICD coding

The International Classification of Diseases (ICD) is a widely used criterion for disease classification, health monitoring, and medical data analysis. Deep learning-based automated ICD coding has gained attention due to the time-consuming and costly nature of manual coding. The main challenges of automated ICD coding include imbalanced label distribution, code hierarchy and noisy texts. Recent works have considered using code hierarchy or description for better label representation to solve the problem of imbalanced label distribution. However, these methods are still ineffective and redundant since they only interact with a constant label representation. In this work, we introduce a novel Hyperbolic Graph Convolutional Network with Contrastive Learning (HGCN-CL) to solve the above problems and the shortcomings of the previous methods. We adopt a Hyperbolic graph convolutional network on ICD coding to capture the hierarchical structure of codes, which can solve the problem of large distortions when embedding hierarchical structure with graph convolutional network. Besides, we introduce contrastive learning for automatic ICD coding by injecting code features into text encoder to generate hierarchical-aware positive samples to solve the problem of interacting with constant code features. We conduct experiments on the public MIMIC-III and MIMIC-II datasets. The results on MIMIC III show that HGCN-CL outperforms previous state-of-art methods for automatic ICD coding, which achieves a 2.7% and 3.6% improvement respectively compared to previous best results (Hypercore). We also provide ablation experiments and hierarchy visualization to verify the effectiveness of components in our model.

A Novel Signal Restoration Method of Noisy Photoplethysmograms for Uninterrupted Health Monitoring

Health-tracking from photoplethysmography (PPG) signals is significantly hindered by motion artifacts (MAs). Although many algorithms exist to detect MAs, the corrupted signal often remains unexploited. This work introduces a novel method able to reconstruct noisy PPGs and facilitate uninterrupted health monitoring. The algorithm starts with spectral-based MA detection, followed by signal reconstruction by using the morphological and heart-rate variability information from the clean segments adjacent to noise. The algorithm was tested on (a) 30 noisy PPGs of a maximum 20 s noise duration and (b) 28 originally clean PPGs, after noise addition (2-120 s) (1) with and (2) without cancellation of the corresponding clean segment. Sampling frequency was 250 Hz after resampling. Noise detection was evaluated by means of accuracy, sensitivity, and specificity. For the evaluation of signal reconstruction, the heart-rate (HR) was compared via Pearson correlation (PC) and absolute error (a) between ECGs and reconstructed PPGs and (b) between original and reconstructed PPGs. Bland-Altman (BA) analysis for the differences in HR estimation on original and reconstructed segments of (b) was also performed. Noise detection accuracy was 90.91% for (a) and 99.38-100% for (b). For the PPG reconstruction, HR showed 99.31% correlation in (a) and >90% for all noise lengths in (b). Mean absolute error was 1.59 bpm for (a) and 1.26-1.82 bpm for (b). BA analysis indicated that, in most cases, 90% or more of the recordings fall within the confidence interval, regardless of the noise length. Optimal performance is achieved even for signals of noise up to 2 min, allowing for the utilization and further analysis of recordings that would otherwise be discarded. Thereby, the algorithm can be implemented in monitoring devices, assisting in uninterrupted health-tracking.

Analysis of pathogenic factors on the death rate of sepsis patients

BACKGROUND

The Surviving Sepsis Campaign (SSC) believed that early identification of septic shock, aggressive fluid resuscitation and maintenance of effective perfusion pressure should be carried out. However, some of the current research focused on a single death factor for sepsis patients, based on a limited sample, and the research results of the relationship between comorbidities and sepsis related death also have some controversies.

METHOD

Therefore, our study used data from a large sample of 9,544 sepsis patients aged 18-85 obtained from the MIMIC-IV database, to explore the risk factors of death in patients with sepsis. We used the general clinical information, organ dysfunction scores, and comorbidities to analyze the independent risk factors for death of these patients.

RESULTS

The death group had significantly higher organ dysfunction scores, lower BMI, lower body temperature, faster heart rate and lower urine-output. Among the comorbidities, patients suffering from congestive heart failure and liver disease had a higher mortality rate.

CONCLUSION

This study helps to identify sepsis early, based on a comprehensive evaluation of a patient's basic information, organ dysfunction scores and comorbidities, and this methodology could be used for actual clinical diagnosis in hospitals.

Medical Informatics Operating Room Vitals and Events Repository (MOVER): a public-access operating room database

Objectives

Artificial intelligence (AI) holds great promise for transforming the healthcare industry. However, despite its potential, AI is yet to see widespread deployment in clinical settings in significant part due to the lack of publicly available clinical data and the lack of transparency in the published AI algorithms. There are few clinical data repositories publicly accessible to researchers to train and test AI algorithms, and even fewer that contain specialized data from the perioperative setting. To address this gap, we present and release the Medical Informatics Operating Room Vitals and Events Repository (MOVER).

Materials and Methods

This first release of MOVER includes adult patients who underwent surgery at the University of California, Irvine Medical Center from 2015 to 2022. Data for patients who underwent surgery were captured from 2 different sources: High-fidelity physiological waveforms from all of the operating rooms were captured in real time and matched with electronic medical record data.

Results

MOVER includes data from 58 799 unique patients and 83 468 surgeries. MOVER is available for download at https://doi.org/10.24432/C5VS5G, it can be downloaded by anyone who signs a data usage agreement (DUA), to restrict traffic to legitimate researchers.

Discussion

To the best of our knowledge MOVER is the only freely available public data repository that contains electronic health record and high-fidelity physiological waveforms data for patients undergoing surgery.

Conclusion

MOVER is freely available to all researchers who sign a DUA, and we hope that it will accelerate the integration of AI into healthcare settings, ultimately leading to improved patient outcomes.

Association between platelet-lymphocyte ratio and 90-day mortality in patients with intracerebral hemorrhage: data from the MIMIC-III database

Background

Recent evidence suggested that platelet-lymphocyte ratio (PLR) may play a role in the pathophysiology of intracerebral hemorrhage (ICH), but the results are controversial. This study aimed to explore the relationship between PLR and mortality in patients with ICH.

Methods

All data were extracted from the Medical Information Mart for Intensive Care (MIMIC) III database. The study outcome was 90-day mortality. Multivariable Cox regression analyses were used to calculate the adjusted hazard ratio (HR) with a 95% confidence interval (CI), and curve-fitting (restricted cubic spline) was used to assess the non-linear relationship.

Results

Of 1,442 patients, 1,043 patients with ICH were included. The overall 90-day mortality was 29.8% (311/1,043). When PLR was assessed in quartiles, the risk of 90-day mortality for ICH was lowest for quartile 2 (120.9 to <189.8: adjusted HR, 0.67; 95% CI: 0.48-0.93; P = 0.016), compared with those in quartile 1 (<120.9). Consistently in the threshold analysis, for every 1 unit increase in PLR, there was a 0.6% decrease in the risk of 90-day mortality for ICH (adjusted HR, 0.994; 95% CI: 0.988-0.999) in those with PLR <145.54, and a 0.2% increase in 90-day mortality (adjusted HR, 1.002; 95% CI: 1.000-1.003) in participants with PLR ≥145.54.

Conclusion

There was a non-linear relationship between PLR and 90-day mortality for patients with ICH, with an inflection point at 145.54 and a minimal risk at 120.9 to <189.8 of PLR.

Automated ICD coding using extreme multi-label long text transformer-based models

Encouraged by the success of pretrained Transformer models in many natural language processing tasks, their use for International Classification of Diseases (ICD) coding tasks is now actively being explored. In this study, we investigated two existing Transformer-based models (PLM-ICD and XR-Transformer) and proposed a novel Transformer-based model (XR-LAT), aiming to address the extreme label set and long text classification challenges that are posed by automated ICD coding tasks. The Transformer-based model PLM-ICD, which currently holds the state-of-the-art (SOTA) performance on the ICD coding benchmark datasets MIMIC-III and MIMIC-II, was selected as our baseline model for further optimisation on both datasets. In addition, we extended the capabilities of the leading model in the general extreme multi-label text classification domain, XR-Transformer, to support longer sequences and trained it on both datasets. Moreover, we proposed a novel model, XR-LAT, which was also trained on both datasets. XR-LAT is a recursively trained model chain on a predefined hierarchical code tree with label-wise attention, knowledge transferring and dynamic negative sampling mechanisms. Our optimised PLM-ICD models, which were trained with longer total and chunk sequence lengths, significantly outperformed the current SOTA PLM-ICD models, and achieved the highest micro-F1 scores of 60.8 % and 50.9 % on MIMIC-III and MIMIC-II, respectively. The XR-Transformer model, although SOTA in the general domain, did not perform well across all metrics. The best XR-LAT based models obtained results that were competitive with the current SOTA PLM-ICD models, including improving the macro-AUC by 2.1 % and 5.1 % on MIMIC-III and MIMIC-II, respectively. Our optimised PLM-ICD models are the new SOTA models for automated ICD coding on both datasets, while our novel XR-LAT models perform competitively with the previous SOTA PLM-ICD models.

Cuffless Hypertension Detection using Swarm Support Vector Machine Utilizing Photoplethysmogram and Electrocardiogram

Background

Hypertension is associated with severe complications, and its detection is important to provide early information about a hypertension event, which is essential to prevent further complications.

Objective

This study aimed to investigate a strategy for hypertension detection without a cuff using parameters of bioelectric signals, i.e., Electrocardiogram (ECG), Photoplethysmogram (PPG,) and an algorithm of Swarm-based Support Vector Machine (SSVM).

Material and Methods

This experimental study was conducted to develop a hypertension detection system. ECG and PPG bioelectrical records were collected from the Medical Information Mart for Intensive Care (MIMIC) from normal and hypertension participants and processed to find the parameters, used for the inputs of SSVM and comprised Pulse Arrival Time (PAT) and the characteristics of PPG signal derivatives. The SSVM was n Support Vector Machine (SVM) algorithm optimized using particle swarm optimization with Quantum Delta-potential-well (QDPSO). The SSVMs with different inputs were investigated to find the optimal detection performance.

Results

The proposed strategy was performed at 96% in terms of F1-score, accuracy, sensitivity, and specificity with better performance than the other methods tested and methods and also could develop a cuff-free hypertension monitoring system.

Conclusion

Hypertension using SSVM, ECG, and PPG parameters is acceptably performed. The hypertension detection had lower performance utilizing only PPG than both ECG and PPG.

Explainable deep learning in plant phenotyping

The increasing human population and variable weather conditions, due to climate change, pose a threat to the world's food security. To improve global food security, we need to provide breeders with tools to develop crop cultivars that are more resilient to extreme weather conditions and provide growers with tools to more effectively manage biotic and abiotic stresses in their crops. Plant phenotyping, the measurement of a plant's structural and functional characteristics, has the potential to inform, improve and accelerate both breeders' selections and growers' management decisions. To improve the speed, reliability and scale of plant phenotyping procedures, many researchers have adopted deep learning methods to estimate phenotypic information from images of plants and crops. Despite the successful results of these image-based phenotyping studies, the representations learned by deep learning models remain difficult to interpret, understand, and explain. For this reason, deep learning models are still considered to be black boxes. Explainable AI (XAI) is a promising approach for opening the deep learning model's black box and providing plant scientists with image-based phenotypic information that is interpretable and trustworthy. Although various fields of study have adopted XAI to advance their understanding of deep learning models, it has yet to be well-studied in the context of plant phenotyping research. In this review article, we reviewed existing XAI studies in plant shoot phenotyping, as well as related domains, to help plant researchers understand the benefits of XAI and make it easier for them to integrate XAI into their future studies. An elucidation of the representations within a deep learning model can help researchers explain the model's decisions, relate the features detected by the model to the underlying plant physiology, and enhance the trustworthiness of image-based phenotypic information used in food production systems.

Automated myocardial infarction and angina detection using second derivative of photoplethysmography

Photoplethysmography (PPG) based healthcare devices have gained enormous interest in the detection of cardiac abnormalities. Limited research has been implemented for myocardial infarction (MI) detection. Moreover, PPG-based detection of angina is still a research gap. PPG signals are not always informative. Therefore, this research work presents the use of PPG signals and their second derivative to evaluate myocardial infarction and angina using a novel set of morphological features. The obtained morphological features are fed onto the feed-forward artificial neural network for the identification of the type of MI and unstable angina (UA). The initial experiments have been carried out on non-ambulatory (public) subjects for feature extraction and later evaluated on ambulatory (self-generated) databases. The intended method attains accuracy, sensitivity, and specificity of 98%, 97%, 98% on the public database and 94%, 94%, 94% on the self-generated database. The result shows that the proposed set of features can detect MI and UA with significant accuracy.

Fully convolutional neural network and PPG signal for arterial blood pressure waveform estimation

Objective. The quality of the arterial blood pressure (ABP) waveform is crucial for predicting the value of blood pressure. The ABP waveform is predicted through experiments, and then Systolic blood pressure (SBP), Diastolic blood pressure, (DBP), and Mean arterial pressure (MAP) information are estimated from the ABP waveform.Approach. To ensure the quality of the predicted ABP waveform, this paper carefully designs the network structure, input signal, loss function, and structural parameters. A fully convolutional neural network (CNN) MultiResUNet3+ is used as the core architecture of ABP-MultiNet3+. In addition to performing Kalman filtering on the original photoplethysmogram (PPG) signal, its first-order derivative and second-order derivative signals are used as ABP-MultiNet3+ enter. The model's loss function uses a combination of mean absolute error (MAE) and means square error (MSE) loss to ensure that the predicted ABP waveform matches the reference waveform.Main results. The proposed ABP-MultiNet3+ model was tested on the public MIMIC II databases, MAE of MAP, DBP, and SBP was 1.88 mmHg, 3.11 mmHg, and 4.45 mmHg, respectively, indicating a small model error. It experiment fully meets the standards of the AAMI standard and obtains level A in the DBP and MAP prediction standard test under the BHS standard. For SBP prediction, it obtains level B in the BHS standard test. Although it does not reach level A, it has a certain improvement compared with the existing methods.Significance. The results show that this algorithm can achieve sleeveless blood pressure estimation, which may enable mobile medical devices to continuously monitor blood pressure and greatly reduce the harm caused by Cardiovascular disease (CVD).

Profiling of patients with type 2 diabetes based on medication adherence data

Introduction

Type 2 diabetes mellitus (T2DM) is a complex, chronic disease affecting multiple organs with varying symptoms and comorbidities. Profiling patients helps identify those with unfavorable disease progression, allowing for tailored therapy and addressing special needs. This study aims to uncover different T2DM profiles based on medication intake records and laboratory measurements, with a focus on how individuals with diabetes move through disease phases.

Methods

We use medical records from databases of the last 20 years from the Department of Endocrinology and Diabetology of the University Medical Center in Maribor. Using the standard ATC medication classification system, we created a patient-specific drug profile, created using advanced natural language processing methods combined with data mining and hierarchical clustering.

Results

Our results show a well-structured profile distribution characterizing different age groups of individuals with diabetes. Interestingly, only two main profiles characterize the early 40-50 age group, and the same is true for the last 80+ age group. One of these profiles includes individuals with diabetes with very low use of various medications, while the other profile includes individuals with diabetes with much higher use. The number in both groups is reciprocal. Conversely, the middle-aged groups are characterized by several distinct profiles with a wide range of medications that are associated with the distinct concomitant complications of T2DM. It is intuitive that the number of profiles increases in the later age groups, but it is not obvious why it is reduced later in the 80+ age group. In this context, further studies are needed to evaluate the contributions of a range of factors, such as drug development, drug adoption, and the impact of mortality associated with all T2DM-related diseases, which characterize these middle-aged groups, particularly those aged 55-75.

Conclusion

Our approach aligns with existing studies and can be widely implemented without complex or expensive analyses. Treatment and drug use data are readily available in healthcare facilities worldwide, allowing for profiling insights into individuals with diabetes. Integrating data from other departments, such as cardiology and renal disease, may provide a more sophisticated understanding of T2DM patient profiles.

Non-invasive estimation of arterial blood pressure fluctuations using a peripheral photoplethysmograph inside the MRI scanner

The blood-oxygen-level-dependent (BOLD) signal measured by functional magnetic resonance imaging (fMRI) is modulated by neural activity through the neurovascular coupling effect, as well as non-neural factors of physiological origin such as heart rate, respiration, and arterial blood pressure (ABP). While the former two effects have been previously characterized, the modulation of the BOLD signal by ABP fluctuations is still poorly understood. This is largely due to the difficulty of obtaining reliable ABP measurements in the MRI environment. Here, we propose a combined experimental and mathematical modeling framework to estimate ABP fluctuations inside the MRI scanner using photoplethysmography (PPG). Specifically, we used concurrent PPG and ABP measurements obtained outside the scanner to train the mathematical model and applied it to PPG measurements obtained inside the MRI scanner. Our results suggest good agreement between the model-predicted and experimentally measured ABP fluctuations and region specific correlations with the BOLD fluctuations.

PPGFeat: a novel MATLAB toolbox for extracting PPG fiducial points

Photoplethysmography is a non-invasive technique used for measuring several vital signs and for the identification of individuals with an increased disease risk. Its principle of work is based on detecting changes in blood volume in the microvasculature of the skin through the absorption of light. The extraction of relevant features from the photoplethysmography signal for estimating certain physiological parameters is a challenging task, where various feature extraction methods have been proposed in the literature. In this work, we present PPGFeat, a novel MATLAB toolbox supporting the analysis of raw photoplethysmography waveform data. PPGFeat allows for the application of various preprocessing techniques, such as filtering, smoothing, and removal of baseline drift; the calculation of photoplethysmography derivatives; and the implementation of algorithms for detecting and highlighting photoplethysmography fiducial points. PPGFeat includes a graphical user interface allowing users to perform various operations on photoplethysmography signals and to identify, and if required also adjust, the fiducial points. Evaluating the PPGFeat's performance in identifying the fiducial points present in the publicly available PPG-BP dataset, resulted in an overall accuracy of 99% and 3038/3066 fiducial points were correctly identified. PPGFeat significantly reduces the risk of errors in identifying inaccurate fiducial points. Thereby, it is providing a valuable new resource for researchers for the analysis of photoplethysmography signals.

Deep-learning-based blood pressure estimation using multi channel photoplethysmogram and finger pressure with attention mechanism

Recently, several studies have proposed methods for measuring cuffless blood pressure (BP) using finger photoplethysmogram (PPG) signals. This study presents a new BP estimation system that measures PPG signals under progressive finger pressure, making the system relatively robust to errors caused by finger position when using the cuffless oscillometric method. To reduce errors caused by finger position, we developed a sensor that can simultaneously measure multi-channel PPG and force signals in a wide field of view (FOV). We propose a deep-learning-based algorithm that can learn to focus on the optimal PPG channel from multi channel PPG using an attention mechanism. The errors (ME ± STD) of the proposed multi channel system were 0.43±9.35 mmHg and 0.21 ± 7.72 mmHg for SBP and DBP, respectively. Through extensive experiments, we found a significant performance difference depending on the location of the PPG measurement in the BP estimation system using finger pressure.

β-Blockers could improve the 28-day and 3-year survival of patients with end-stage renal disease: a retrospective cohort study

BACKGROUND

Dialysis or non-dialysis end-stage renal disease (ESRD) patients are accompanied by cardiovascular disease (CVD) or hypertension. We aimed to study the effect of a common treatment for CVD, β-blockers, on the survival of ESRD patients, improving their prognosis from the perspective of drug therapy.

METHODS

It was a retrospective cohort study using the Medical Information Mart for Intensive Care dataset. ESRD patients in the intensive care unit from June 2001 to October 2012 were included. We examined the effect of using versus not using β-blockers in the overall population and subgroups with the risk of 28-day and 3-year mortality through Cox proportional hazards models and Kaplan-Meier curves.

RESULTS

A total of 1639 participants were included with 371 (22.64%) β-blockers users. There were 315 (19.22%) 28-day and 970 (59.18%) 3-year mortality events during follow-up. Using β-blockers in overall ESRD patients could reduce all-cause 28-day mortality [adjusted hazard ratio (HR) 0.450, 95% confidence interval (CI) 0.325-0.624] and 3-year mortality (adjusted HR 0.695, 95% CI 0.589-0.821). This result was consistent among subgroups (ESRD without hypertension: adjusted HR 0.412, 95% CI 0.289-0.588; with CVD: adjusted HR 0.478, 95% CI 0.321-0.711; without CVD: adjusted HR 0.448, 95% CI 0.248-0.810; with dialysis: adjusted HR 0.471, 95% CI 0.320-0.694) in 28-day mortality, and the 3-year mortality was consistent. In ESRD patients with hypertension and without dialysis subgroups, β-blockers had no effect on survival.

CONCLUSION

Using β-blockers could reduce the risk of 28-day and 3-year mortality in ESRD patients, including those with CVD. This study provided a reference for the treatment of β-blockers in patients with ESRD.

PPG2ECGps: An End-to-End Subject-Specific Deep Neural Network Model for Electrocardiogram Reconstruction from Photoplethysmography Signals without Pulse Arrival Time Adjustments

Electrocardiograms (ECGs) provide crucial information for evaluating a patient's cardiovascular health; however, they are not always easily accessible. Photoplethysmography (PPG), a technology commonly used in wearable devices such as smartwatches, has shown promise for constructing ECGs. Several methods have been proposed for ECG reconstruction using PPG signals, but some require signal alignment during the training phase, which is not feasible in real-life settings where ECG signals are not collected at the same time as PPG signals. To address this challenge, we introduce PPG2ECGps, an end-to-end, patient-specific deep-learning neural network utilizing the W-Net architecture. This novel approach enables direct ECG signal reconstruction from PPG signals, eliminating the need for signal alignment. Our experiments show that the proposed model achieves mean values of 0.977 mV for Pearson's correlation coefficient, 0.037 mV for the root mean square error, and 0.010 mV for the normalized dynamic time-warped distance when comparing reconstructed ECGs to reference ECGs from a dataset of 500 records. As PPG signals are more accessible than ECG signals, our proposed model has significant potential to improve patient monitoring and diagnosis in healthcare settings via wearable devices.

Emerging sensing and modeling technologies for wearable and cuffless blood pressure monitoring

Cardiovascular diseases (CVDs) are a leading cause of death worldwide. For early diagnosis, intervention and management of CVDs, it is highly desirable to frequently monitor blood pressure (BP), a vital sign closely related to CVDs, during people's daily life, including sleep time. Towards this end, wearable and cuffless BP extraction methods have been extensively researched in recent years as part of the mobile healthcare initiative. This review focuses on the enabling technologies for wearable and cuffless BP monitoring platforms, covering both the emerging flexible sensor designs and BP extraction algorithms. Based on the signal type, the sensing devices are classified into electrical, optical, and mechanical sensors, and the state-of-the-art material choices, fabrication methods, and performances of each type of sensor are briefly reviewed. In the model part of the review, contemporary algorithmic BP estimation methods for beat-to-beat BP measurements and continuous BP waveform extraction are introduced. Mainstream approaches, such as pulse transit time-based analytical models and machine learning methods, are compared in terms of their input modalities, features, implementation algorithms, and performances. The review sheds light on the interdisciplinary research opportunities to combine the latest innovations in the sensor and signal processing research fields to achieve a new generation of cuffless BP measurement devices with improved wearability, reliability, and accuracy.

Artificial intelligence for clinical decision support for monitoring patients in cardiovascular ICUs: A systematic review

Background

Artificial intelligence (AI) and machine learning (ML) models continue to evolve the clinical decision support systems (CDSS). However, challenges arise when it comes to the integration of AI/ML into clinical scenarios. In this systematic review, we followed the Preferred Reporting Items for Systematic reviews and Meta-Analyses (PRISMA), the population, intervention, comparator, outcome, and study design (PICOS), and the medical AI life cycle guidelines to investigate studies and tools which address AI/ML-based approaches towards clinical decision support (CDS) for monitoring cardiovascular patients in intensive care units (ICUs). We further discuss recent advances, pitfalls, and future perspectives towards effective integration of AI into routine practices as were identified and elaborated over an extensive selection process for state-of-the-art manuscripts.

Methods

Studies with available English full text from PubMed and Google Scholar in the period from January 2018 to August 2022 were considered. The manuscripts were fetched through a combination of the search keywords including AI, ML, reinforcement learning (RL), deep learning, clinical decision support, and cardiovascular critical care and patients monitoring. The manuscripts were analyzed and filtered based on qualitative and quantitative criteria such as target population, proper study design, cross-validation, and risk of bias.

Results

More than 100 queries over two medical search engines and subjective literature research were developed which identified 89 studies. After extensive assessments of the studies both technically and medically, 21 studies were selected for the final qualitative assessment.

Discussion

Clinical time series and electronic health records (EHR) data were the most common input modalities, while methods such as gradient boosting, recurrent neural networks (RNNs) and RL were mostly used for the analysis. Seventy-five percent of the selected papers lacked validation against external datasets highlighting the generalizability issue. Also, interpretability of the AI decisions was identified as a central issue towards effective integration of AI in healthcare.