No full-text available
To read the full-text of this research,
you can request a copy directly from the authors.
Artificial Intelligence and Predictive Analytics
Abstract
Artificial intelligence (AI) and predictive analytics—a subset of AI techniques—are increasingly used in medicine, and the field of hemodynamic monitoring is no exception. These techniques are used to predict adverse hemodynamic events such as hypotension or tachycardia before they actually occur. In the development of predictive models, steps have to be taken from collection of monitoring data to feature extraction and feature evaluation. Finally, the best model should be selected and the predictive performance evaluated. Predictive models show promising results, yet only few are currently used in clinical practice.
... This early warning system will allow for timely resource allocation, treatment planning, and public health interventions, ultimately improving the response to emerging health challenges and enhancing patient care. [9] Education and Training AI holds great promise in advancing homeopathic education and training. It has the ability to generate engaging learning experiences through the simulation of patient cases, enabling students and professionals to practice the process of diagnosing and treating in a well-regulated setting. ...
Background: Homeopathy, a holistic system of medicine founded on the principles of "like cures like" and individualized treatment, has been in practice for over two centuries. The vast progress of the healthcare sector has brought Artificial Intelligence (AI) to the forefront as a transformative technology.
Objective: This study’s objective is to understand the integration of AI into homeopathic practice and assess its impact on patient care.
Materials and Methods: A comprehensive review of the literature was conducted in PubMed and Google searches were conducted, utilizing the keywords 'artificial intelligence.' to collect insights,
Results: The results of this study indicate that AI has the potential to revolutionize homeopathic practice in multiple ways, including personalized treatment planning, enhanced research capabilities, streamlined patient management, and improved remedy selection. Also, the AI-powered systems such as ChatGPT can assist with patient education, symptom analysis, preliminary guidance, remedy suggestions, appointment scheduling, and patient record keeping. Furthermore, AI can contribute to the quality control of homeopathic remedies, provide language translation support, offer continuing education resources, and even assist in emergency situations and public health information dissemination.
Conclusion: The integration of Artificial Intelligence into homeopathy holds significant promise for optimizing patient care and improving the overall efficiency of homeopathic practice. While AI and ChatGPT can provide valuable information and preliminary guidance, they should not replace the expertise of trained homeopathic practitioners
Arterial hypotension during the early phase of anesthesia can lead to adverse outcomes such as a prolonged postoperative stay or even death. Predicting hypotension during anesthesia induction is complicated by its diverse causes. We investigated the feasibility of developing a machine-learning model to predict postinduction hypotension. Naïve Bayes, logistic regression, random forest, and artificial neural network models were trained to predict postinduction hypotension, occurring between tracheal intubation and incision, using data for the period from between the start of anesthesia induction and immediately before tracheal intubation obtained from an anesthesia monitor, a drug administration infusion pump, an anesthesia machine, and from patients’ demographics, together with preexisting disease information from electronic health records. Among 222 patients, 126 developed postinduction hypotension. The random-forest model showed the best performance, with an area under the receiver operating characteristic curve of 0.842 (95% confidence interval [CI]: 0.736-0.948). This was higher than that for the Naïve Bayes (0.778; 95% CI: 0.65-0.898), logistic regression (0.756; 95% CI: 0.630-0.881), and artificial-neural-network (0.760; 95% CI: 0.640-0.880) models. The most important features affecting the accuracy of machine-learning prediction were a patient’s lowest systolic blood pressure, lowest mean blood pressure, and mean systolic blood pressure before tracheal intubation. We found that machine-learning models using data obtained from various anesthesia machines between the start of anesthesia induction and immediately before tracheal intubation can predict hypotension occurring during the period between tracheal intubation and incision.
Tachycardia is defined as a heart rate greater than 100 bpm for more than 1 min. Tachycardia often occurs after endotracheal intubation and can cause serious complication in patients with cardiovascular disease. The ability to predict post-intubation tachycardia would help clinicians by notifying a potential event to pre-treat. In this paper, we predict the potential post-intubation tachycardia. Given electronic medical record and vital signs collected before tracheal intubation, we predict whether post-intubation tachycardia will occur within 10 min. Of 1931 available patient datasets, 257 remained after filtering those with inappropriate data such as outliers and inappropriate annotations. Three feature sets were designed using feature selection algorithms, and two additional feature sets were defined by statistical inspection or manual examination. The five feature sets were compared with various machine learning models such as naïve Bayes classifiers, logistic regression, random forest, support vector machines, extreme gradient boosting, and artificial neural networks. Parameters of the models were optimized for each feature set. By 10-fold cross validation, we found that an logistic regression model with eight-dimensional hand-crafted features achieved an accuracy of 80.5%, recall of 85.1%, precision of 79.9%, an F1 score of 79.9%, and an area under the receiver operating characteristic curve of 0.85.
Tachycardia is a strong though non-specific marker of cardiovascular stress that proceeds hemodynamic instability. We designed a predictive model of tachycardia using multi-granular intensive care unit (ICU) data by creating a risk score and dynamic trajectory. A subset of clinical and numerical signals were extracted from the Multiparameter Intelligent Monitoring in Intensive Care II database. A tachycardia episode was defined as heart rate ≥ 130/min lasting for ≥ 5 min, with ≥ 10% density. Regularized logistic regression (LR) and random forest (RF) classifiers were trained to create a risk score for upcoming tachycardia. Three different risk score models were compared for tachycardia and control (non-tachycardia) groups. Risk trajectory was generated from time windows moving away at 1 min increments from the tachycardia episode. Trajectories were computed over 3 hours leading up to the episode for three different models. From 2809 subjects, 787 tachycardia episodes and 707 control periods were identified. Patients with tachycardia had increased vasopressor support, longer ICU stay, and increased ICU mortality than controls. In model evaluation, RF was slightly superior to LR, which accuracy ranged from 0.847 to 0.782, with area under the curve from 0.921 to 0.842. Risk trajectory analysis showed average risks for tachycardia group evolved to 0.78 prior to the tachycardia episodes, while control group risks remained < 0.3. Among the three models, the internal control model demonstrated evolving trajectory approximately 75 min before tachycardia episode. Clinically relevant tachycardia episodes can be predicted from vital sign time series using machine learning algorithms.
Traumatically brain injured (TBI) patients are at risk from secondary insults. Arterial hypotension, critically low blood pressure, is one of the most dangerous secondary insults and is related to poor outcome in patients. The overall aim of this study was to get proof of the concept that advanced statistical techniques (machine learning) are methods that are able to provide early warning of impending hypotensive events before they occur during neuro-critical care. A Bayesian artificial neural network (BANN) model predicting episodes of hypotension was developed using data from 104 patients selected from the BrainIT multi-center database. Arterial hypotension events were recorded and defined using the Edinburgh University Secondary Insult Grades (EUSIG) physiological adverse event scoring system. The BANN was trained on a random selection of 50% of the available patients (n = 52) and validated on the remaining cohort. A multi-center prospective pilot study (Phase 1, n = 30) was then conducted with the system running live in the clinical environment, followed by a second validation pilot study (Phase 2, n = 49). From these prospectively collected data, a final evaluation study was done on 69 of these patients with 10 patients excluded from the Phase 2 study because of insufficient or invalid data. Each data collection phase was a prospective non-interventional observational study conducted in a live clinical setting to test the data collection systems and the model performance. No prediction information was available to the clinical teams during a patient’s stay in the ICU. The final cohort (n = 69), using a decision threshold of 0.4, and including false positive checks, gave a sensitivity of 39.3% (95% CI 32.9–46.1) and a specificity of 91.5% (95% CI 89.0–93.7). Using a decision threshold of 0.3, and false positive correction, gave a sensitivity of 46.6% (95% CI 40.1–53.2) and specificity of 85.6% (95% CI 82.3–88.8). With a decision threshold of 0.3, > 15 min warning of patient instability can be achieved. We have shown, using advanced machine learning techniques running in a live neuro-critical care environment, that it would be possible to give neurointensive teams early warning of potential hypotensive events before they emerge, allowing closer monitoring and earlier clinical assessment in an attempt to prevent the onset of hypotension. The multi-centre clinical infrastructure developed to support the clinical studies provides a solid base for further collaborative research on data quality, false positive correction and the display of early warning data in a clinical setting.
Ventricular tachycardia (VT) is a potentially fatal tachyarrhythmia, which causes a rapid heartbeat as a result of improper electrical activity of the heart. This is a potentially life-threatening arrhythmia because it can cause low blood pressure and may lead to ventricular fibrillation, asystole, and sudden cardiac death. To prevent VT, we developed an early prediction model that can predict this event one hour before its onset using an artificial neural network (ANN) generated using 14 parameters obtained from heart rate variability (HRV) and respiratory rate variability (RRV) analysis. De-identified raw data from the monitors of patients admitted to the cardiovascular intensive care unit at Asan Medical Center between September 2013 and April 2015 were collected. The dataset consisted of 52 recordings obtained one hour prior to VT events and 52 control recordings. Two-thirds of the extracted parameters were used to train the ANN, and the remaining third was used to evaluate performance of the learned ANN. The developed VT prediction model proved its performance by achieving a sensitivity of 0.88, specificity of 0.82, and AUC of 0.93.
In the intensive care unit (ICU), clinical staff must stay vigilant to promptly detect and treat hypotensive episodes (HEs). Given the stressful context of busy ICUs, an automated hypotensive risk stratifier can help ICU clinicians focus care and resources by prospectively identifying patients at increased risk of impending HEs. The objective of this study was to investigate the possible existence of discriminatory patterns in hemodynamic data that can be indicative of future hypotensive risk.
Given the complexity and heterogeneity of ICU data, a machine learning approach was used in this study. Time series of minute-by-minute measures of mean arterial blood pressure, heart rate, pulse pressure, and relative cardiac output from 1,311 records from the MIMIC II Database were used. An HE was defined as a 30-minute period during which the mean arterial pressure was below 60 mmHg for at least 90% of the time. Features extracted from the hemodynamic data during an observation period of either 30 or 60 minutes were analyzed to predict the occurrence of HEs 1 or 2 hours into the future. Artificial neural networks (ANNs) were trained for binary classification (normotensive vs. hypotensive) and regression (estimation of future mean blood pressure).
The ANNs were successfully trained to discriminate patterns in the multidimensional hemodynamic data that were predictive of future HEs. The best overall binary classification performance resulted in a mean area under ROC curve of 0.918, a sensitivity of 0.826, and a specificity of 0.859. Predicting further into the future resulted in poorer performance, whereas observation duration minimally affected performance. The low prevalence of HEs led to poor positive predictive values. In regression, the best mean absolute error was 9.67%.
The promising pattern recognition performance demonstrates the existence of discriminatory patterns in hemodynamic data that can indicate impending hypotension. The poor PPVs discourage a direct HE predictor, but a hypotensive risk stratifier based on the pattern recognition algorithms of this study would be of significant clinical value in busy ICU environments.
Background:
Intraoperative hypotension is associated with worse perioperative outcomes for patients undergoing major noncardiac surgery. The Hypotension Prediction Index is a unitless number that is derived from an arterial pressure waveform trace, and as the number increases, the risk of hypotension occurring in the near future increases. We investigated the diagnostic ability of the Hypotension Prediction Index in predicting impending intraoperative hypotension in comparison to other commonly collected perioperative hemodynamic variables.
Methods:
This is a 2-center retrospective analysis of patients undergoing major surgery. Data were downloaded and analyzed from the Edwards Lifesciences EV1000 platform. Receiver operating characteristic curves were constructed for the Hypotension Prediction Index and other hemodynamic variables as well as event rates and time to event.
Results:
Two hundred fifty-five patients undergoing major surgery were included in the analysis yielding 292,025 data points. The Hypotension Prediction Index predicted hypotension with a sensitivity and specificity of 85.8% (95% CI, 85.8%-85.9%) and 85.8% (95% CI, 85.8%-85.9%) 5 minutes before a hypotensive event (area under the curve, 0.926 [95% CI, 0.925-0.926]); 81.7% (95% CI, 81.6%-81.8%) and 81.7% (95% CI, 81.6%-81.8%) 10 minutes before a hypotensive event (area under the curve, 0.895 [95% CI, 0.894-0.895]); and 80.6% (95% CI, 80.5%-80.7%) and 80.6% (95% CI, 80.5%-80.7%) 15 minutes before a hypotensive event (area under the curve, 0.879 [95% CI, 0.879-0.880]). The Hypotension Prediction Index performed superior to all other measured hemodynamic variables including mean arterial pressure and change in mean arterial pressure over a 3-minute window.
Conclusions:
The Hypotension Prediction Index provides an accurate real time and continuous prediction of impending intraoperative hypotension before its occurrence and has superior predictive ability than the commonly measured perioperative hemodynamic variables.
Background:
Predicting hypotension well in advance provides physicians with enough time to respond with proper therapeutic measures. However, the real-time prediction of hypotension with high positive predictive value (PPV) is a challenge. This is due to the dynamic changes in patients' physiological status following drug administration, which limits the quantity of useful data available for the algorithm.
Method:
To mimic real-time monitoring, we developed a machine-learning algorithm that uses most of the available data points from patients' records to train and test the algorithm. The algorithm predicts hypotension up to 30 min in advance based on the data from only 5 min of patient physiological history. A novel evaluation method is also proposed to assess the performance of the algorithm as a function of time at every timestamp within 30 min of hypotension onset. This evaluation approach provides statistical tools to find the best possible prediction window.
Results:
During about 181,000 min of monitoring of 400 patients, the algorithm demonstrated 94% accuracy, 85% sensitivity and 96% specificity in predicting hypotension within 30 min of the events. A high PPV of 81% was obtained, and the algorithm predicted 80% of hypotensive events 25 min prior to onset. It was shown that choosing a classification threshold that maximizes the F1 score during the training phase contributes to a high PPV and sensitivity.
Conclusions:
This study demonstrates the promising potential of machine-learning algorithms in the real-time prediction of hypotensive events in ICU settings based on short-term physiological history.
Background:
Acute hypotensive episodes (AHE), defined as a drop in the mean arterial pressure (MAP) <65 mm Hg lasting at least 5 consecutive minutes, are among the most critical events in the intensive care unit (ICU). They are known to be associated with adverse outcome in critically ill patients. AHE prediction is of prime interest because it could allow for treatment adjustment to predict or shorten AHE.
Methods:
The Super Learner (SL) algorithm is an ensemble machine-learning algorithm that we specifically trained to predict an AHE 10 minutes in advance. Potential predictors included age, sex, type of care unit, severity scores, and time-evolving characteristics such as mechanical ventilation, vasopressors, or sedation medication as well as features extracted from physiological signals: heart rate, pulse oximetry, and arterial blood pressure. The algorithm was trained on the Medical Information Mart for Intensive Care dataset (MIMIC II) database. Internal validation was based on the area under the receiver operating characteristic curve (AUROC) and the Brier score (BS). External validation was performed using an external dataset from Lariboisière hospital, Paris, France.
Results:
Among 1151 patients included, 826 (72%) patients had at least 1 AHE during their ICU stay. Using 1 single random period per patient, the SL algorithm with Haar wavelets transform preprocessing was associated with an AUROC of 0.929 (95% confidence interval [CI], 0.899-0.958) and a BS of 0.08. Using all available periods for each patient, SL with Haar wavelets transform preprocessing was associated with an AUROC of 0.890 (95% CI, 0.886-0.895) and a BS of 0.11. In the external validation cohort, the AUROC reached 0.884 (95% CI, 0.775-0.993) with 1 random period per patient and 0.889 (0.768-1) with all available periods and BSs <0.1.
Conclusions:
The SL algorithm exhibits good performance for the prediction of an AHE 10 minutes ahead of time. It allows an efficient, robust, and rapid evaluation of the risk of hypotension that opens the way to routine use.
What we already know about this topic:
WHAT THIS ARTICLE TELLS US THAT IS NEW: BACKGROUND:: Hypotension is a risk factor for adverse perioperative outcomes. Machine learning methods allow large amounts of data for development of robust predictive analytics. The authors hypothesized that machine learning methods can provide prediction for the risk of postinduction hypotension METHODS:: Data was extracted from the electronic health record of a single quaternary care center from November 2015 to May 2016 for patients over age 12 that underwent general anesthesia, without procedure exclusions. Multiple supervised machine learning classification techniques were attempted, with postinduction hypotension (mean arterial pressure less than 55 mmHg within 10 min of induction by any measurement) as primary outcome, and preoperative medications, medical comorbidities, induction medications, and intraoperative vital signs as features. Discrimination was assessed using cross-validated area under the receiver operating characteristic curve. The best performing model was tuned and final performance assessed using split-set validation.
Results:
Out of 13,323 cases, 1,185 (8.9%) experienced postinduction hypotension. Area under the receiver operating characteristic curve using logistic regression was 0.71 (95% CI, 0.70 to 0.72), support vector machines was 0.63 (95% CI, 0.58 to 0.60), naive Bayes was 0.69 (95% CI, 0.67 to 0.69), k-nearest neighbor was 0.64 (95% CI, 0.63 to 0.65), linear discriminant analysis was 0.72 (95% CI, 0.71 to 0.73), random forest was 0.74 (95% CI, 0.73 to 0.75), neural nets 0.71 (95% CI, 0.69 to 0.71), and gradient boosting machine 0.76 (95% CI, 0.75 to 0.77). Test set area for the gradient boosting machine was 0.74 (95% CI, 0.72 to 0.77).
Conclusions:
The success of this technique in predicting postinduction hypotension demonstrates feasibility of machine learning models for predictive analytics in the field of anesthesiology, with performance dependent on model selection and appropriate tuning.
Background:
With appropriate algorithms, computers can learn to detect patterns and associations in large data sets. The authors' goal was to apply machine learning to arterial pressure waveforms and create an algorithm to predict hypotension. The algorithm detects early alteration in waveforms that can herald the weakening of cardiovascular compensatory mechanisms affecting preload, afterload, and contractility.
Methods:
The algorithm was developed with two different data sources: (1) a retrospective cohort, used for training, consisting of 1,334 patients' records with 545,959 min of arterial waveform recording and 25,461 episodes of hypotension; and (2) a prospective, local hospital cohort used for external validation, consisting of 204 patients' records with 33,236 min of arterial waveform recording and 1,923 episodes of hypotension. The algorithm relates a large set of features calculated from the high-fidelity arterial pressure waveform to the prediction of an upcoming hypotensive event (mean arterial pressure < 65 mmHg). Receiver-operating characteristic curve analysis evaluated the algorithm's success in predicting hypotension, defined as mean arterial pressure less than 65 mmHg.
Results:
Using 3,022 individual features per cardiac cycle, the algorithm predicted arterial hypotension with a sensitivity and specificity of 88% (85 to 90%) and 87% (85 to 90%) 15 min before a hypotensive event (area under the curve, 0.95 [0.94 to 0.95]); 89% (87 to 91%) and 90% (87 to 92%) 10 min before (area under the curve, 0.95 [0.95 to 0.96]); 92% (90 to 94%) and 92% (90 to 94%) 5 min before (area under the curve, 0.97 [0.97 to 0.98]).
Conclusions:
The results demonstrate that a machine-learning algorithm can be trained, with large data sets of high-fidelity arterial waveforms, to predict hypotension in surgical patients' records.
At present many hospitals have to deal with the patient’s care and nursing for Acute Hypotensive Episodes (AHE) occurring in Intensive Care Units. AHE can cause fainting or shock suddenly, leading to irreversibility organ damage, and even death. Therefore, forecasting of occurrence of AHE is of practical value. However, the prediction of clinical AHE largely depends on the doctors’ experience, which cannot guarantee the high rate of success. It is thus very meaningful for the clinical care to use appropriate methods to predict the AHE with an automatic and reliable method. In this study, a Probability Distribution Patterns Analysis (PDPA) method is presented to solve the time series prediction problem of AHE. In the first phase, the features are extracted from the PDPA in the global and integral time series, and the partial local time series in the fixed time window. In the second phase, the proposed algorithm combining Genetic Algorithm (GA) and Support Vector Machine (SVM), namely GA-SVM is adopted to select the vital features for the effective classification. In order to demonstrate the generality of our method, we also conduct experiments on a classical time series problem-Control Chart Patterns (CCPs) multi-class time series, which is a benchmark problem in the process control. For CCPs problem, the experimental results demonstrate that the proposed method outperforms several traditional methods. The obtained accuracy is 98.65%, which is superior to listed previous works using the same CCPs model. For AHE classification and forecasting, the methodology is applied in two data sets, a small data set (37 records) and a big one (2892 records). The test accuracy of 89.19%, sensitivity of 91.67%, specificity of 88% in the small data set, and a test accuracy of 80.76%, sensitivity of 78.19%, specificity of 81.51% in the big data set are achieved with the classification model.
Acute hypotensive episodes (AHEs) are serious clinical events in intensive care units (ICUs), and require immediate treatment to prevent patient injury. Reducing the risks associated with an AHE requires effective and efficient mining of data generated from multiple physiological time series. We propose HeartCast, a model that extracts essential features from such data to effectively predict AHE. HeartCast combines a non-linear support vector machine with best-feature extraction via analysis of the baseline threshold, quartile parameters, and window size of the physiological signals. Our approach has the following benefits: (a) it extracts the most relevant features; (b) it provides the best results for identification of an AHE event; (c) it is fast and scales with linear complexity over the length of the window; and (d) it can manage missing values and noise/outliers by using a best-feature extraction method. We performed experiments on data continuously captured from physiological time series of ICU patients (roughly 3 GB of processed data). HeartCast was found to outperform other state-of-the-art methods found in the literature with a 13.7% improvement in classification accuracy.
This work proposes the application of neural network multi-models to the prediction of adverse acute hypotensive episodes (AHE) occurring in intensive care units (ICU). A generic methodology consisting of two phases is considered. In the first phase, a correlation analysis between the current blood pressure time signal and a collection of historical blood pressure templates is carried out. From this procedure the most similar signals are determined and the respective prediction neural models, previously trained, selected. Then, in a second phase, the multi-model structure is employed to predict the future evolution of current blood pressure signal, enabling to detect the occurrence of an AHE. The effectiveness of the methodology was validated in the context of the 10th PhysioNet/Computers in Cardiology Challenge-Predicting Acute Hypotensive Episodes, applied to a specific set of blood pressure signals, available in MIMIC-II database. A correct prediction of 10 out of 10 AHE for event 1 and of 37 out of 40 AHE for event 2 was achieved, corresponding to the best results of all entries in the two events of the challenge. The generalization capabilities of the strategy was confirmed by applying it to an extended dataset of blood pressure signals, also collected from the MIMIC-II database. A total of 2344 examples, selected from 311 blood pressure signals were tested, enabling to obtain a global sensitivity of 82.8% and a global specificity of 78.4%.
Perioperative hypotension is associated with adverse outcomes in patients undergoing surgery. A computer-based model that integrates related factors and predicts the risk of hypotension would be helpful in clinical anesthesia. The purpose of this study was to develop artificial neural network (ANN) models to identify patients at high risk for postinduction hypotension during general anesthesia.
Anesthesia records for March through November 2007 were reviewed, and 1017 records were analyzed. Eleven patient-related, 2 surgical, and 5 anesthetic variables were used to develop the ANN and logistic regression (LR) models. The quality of the models was evaluated by an external validation data set. Three clinicians were asked to make predictions of the same validation data set on a case-by-case basis.
The ANN model had an accuracy of 82.3%, sensitivity of 76.4%, and specificity of 85.6%. The accuracy of the LR model was 76.5%, the sensitivity was 74.5%, and specificity was 77.7%. The area under the receiver operating characteristic curve for the ANN and LR models was 0.893 and 0.840. The clinicians had the lowest predictive accuracy and sensitivity compared with the ANN and LR models.
The ANN model developed in this study had good discrimination and calibration and would provide decision support to clinicians and increase vigilance for patients at high risk of postinduction hypotension during general anesthesia.
Prediction of an acute hypotensive episode during an ICU hospitalization with a super learner machine-learning algorithm
- M Cherifa
- A Blet
- A Chambas