Prediction of cardiac arrest in critically ill patients presenting to the emergency department using a machine learning score incorporating heart rate variability compared with the modified early warning score

Practice of diagnosis and management of acute respiratory distress syndrome in mainland China: a cross-sectional study

Background

Although acute respiratory distress syndrome (ARDS) has been recognized for more than 50 years, limited information exists about the incidence and management of ARDS in mainland China. To evaluate the potential for improvement in management of patients with ARDS, this study was designed to describe the incidence and management of ARDS in mainland China.

Methods

, 2012) of all patients who fulfilled the Berlin or American European Consensus Conference (AECC) definition of ARDS in 20 intensive care units, with data collection related to the management of ARDS, patient characteristics and outcomes.

Results

was 90%. A recruitment maneuver was performed in 35.5% of the patients, and 8.7% of patients with severe ARDS received prone position. Overall hospital mortality was 34.0%. Hospital mortality was 21.8% for mild, 31.1% for moderate, and 60.0% for patients with severe ARDS (P=0.004).

Conclusions

Despite general acceptance of low Vt and limited Pplat, high driving pressure, low PEEP and low use of adjunctive measures may still be a concern in mainland China, especially in patients with severe ARDS.

Trial Registration

2012.

Increases in Heart Rate Variability Signal Improved Outcomes in Rapid Response Team Consultations: A Cohort Study

Background

Reduced heart rate variability (HRV) indicates dominance of the sympathetic system and a state of “physiologic stress.” We postulated that, in patients with critical illness, increases in HRV might signal successful resuscitation and improved prognosis.

Methods

We carried out a prospective observational study of HRV on all patients referred to the rapid response team (RRT) and correlated with serial vital signs, lactate clearance, ICU admission, and mortality.

Results

Ninety-one patients were studied. Significantly higher HRV was observed in patients who achieved physiological stability and did not need ICU admission: ASDNN 19 versus 34.5, p=0.032; rMSSD 13.5 versus 25, p=0.046; mean VLF 9.4 versus 17, p=0.021; mean LF 5.8 versus 12.4, p=0.018; and mean HF 4.7 versus 10.5, p=0.017. ROC curves confirmed the change in very low frequencies at 2 hours as a strong predictor for ICU admission with an AUC of 0.772 (95% CI 0.633, 0.911, p=0.001) and a cutoff value of −0.65 associated with a sensitivity of 78.6% and a specificity of 61%.

Conclusions

Reduced HRV, specifically VLF, appears closely related to greater severity of critical illness, identifies unsuccessful resuscitation, and can be used to identify consultations that need early ICU admission.

Association between autonomic control indexes and mortality in subjects admitted to intensive care unit

This study checks whether autonomic markers derived from spontaneous fluctuations of heart period (HP) and systolic arterial pressure (SAP) and from their interactions with spontaneous or mechanical respiration (R) are associated with mortality in patients admitted to intensive care unit (ICU). Three-hundred consecutive HP, SAP and R values were recorded during the first day in ICU in 123 patients. Population was divided into survivors (SURVs, n = 83) and non-survivors (NonSURVs, n = 40) according to the outcome. SURVs and NonSURVs were aged- and gender-matched. All subjects underwent modified head-up tilt (MHUT) by tilting the bed back rest segment to 60°. Autonomic control indexes were computed using time-domain, spectral, cross-spectral, complexity, symbolic and causality techniques via univariate, bivariate and conditional approaches. SAP indexes derived from time-domain, model-free complexity and symbolic approaches were associated with the endpoint, while none of HP variability markers was. The association was more powerful during MHUT. Linear cross-spectral and causality indexes were useless to separate SURVs from NonSURVs, while nonlinear bivariate symbolic markers were successful. When indexes were combined with clinical scores, only SAP variance provided complementary information. Cardiovascular control variability indexes, especially when derived after an autonomic challenge such as MHUT, can improve mortality risk stratification in ICU.

Discovering role interaction models in the Emergency Room using Process Mining

OBJECTIVES

A coordinated collaboration among different healthcare professionals in Emergency Room (ER) processes is critical to promptly care for patients who arrive at the hospital in a delicate health condition, claiming for an immediate attention. The aims of this study are (i) to discover role interaction models in (ER) processes using process mining techniques; (ii) to understand how healthcare professionals are currently collaborating; and (iii) to provide useful knowledge that can help to improve ER processes.

METHODS

A four step method based on process mining techniques is proposed. An ER process of a university hospital was considered as a case study, using 7160 episodes that contains specific ER episode attributes.

RESULTS

Insights about how healthcare professionals collaborate in the ER was discovered, including the identification of a prevalent role interaction model along the major triage categories and specific role interaction models for different diagnoses. Also, common and exceptional professional interaction models were discovered at the role level.

CONCLUSIONS

This study allows the discovery of role interaction models through the use of real-life clinical data and process mining techniques. Results show a useful way of providing relevant insights about how healthcare professionals collaborate, uncovering opportunities for process improvement.

Heart Rate Variability Analysis in Patients Who Have Bradycardia Presenting to the Emergency Department with Chest Pain

BACKGROUND

Heart rate variability (HRV) is a noninvasive method to measure the function of the autonomic nervous system. It has been used to risk stratify patients with undifferentiated chest pain in the emergency department (ED). However, bradycardia can have a modifying effect on HRV.

OBJECTIVE

In this study, we aimed to determine how bradycardia affected HRV analysis in patients who presented with chest pain to the ED.

METHODS

Adult patients presenting to the ED at Singapore General Hospital with chest pain were included in the study. Patients with non-sinus rhythm on electrocardiogram (ECG) were excluded. HRV parameters, including time domain, frequency domain, and nonlinear variables, were analyzed from a 5-min ECG segment. Occurrence of a major adverse cardiac event ([MACE], e.g., acute myocardial infarction, percutaneous coronary intervention, coronary artery bypass graft, or mortality) within 30 days of presentation to the ED was also recorded.

RESULTS

A total of 797 patients were included for analysis with 248 patients (31.1%) with 30-day MACE and 135 patients with bradycardia (16.9%). Compared to non-bradycardic patients, bradycardic patients had significant differences in all HRV parameters suggesting an increased parasympathetic component. Among non-bradycardic patients, comparing those who did and did not have 30-day MACE, there were significant differences predominantly in time domain variables, suggesting decreased HRV. In bradycardic patients, the same analysis revealed significant differences in predominantly frequency-domain variables suggesting decreased parasympathetic input.

CONCLUSIONS

Chest pain patients with bradycardia have increased HRV compared to those without bradycardia. This may have important implications on HRV modeling strategies for risk stratification of bradycardic and non-bradycardic chest pain patients.

The effectiveness of physiologically based early warning or track and trigger systems after triage in adult patients presenting to emergency departments: a systematic review

BACKGROUND

Changes to physiological parameters precede deterioration of ill patients. Early warning and track and trigger systems (TTS) use routine physiological measurements with pre-specified thresholds to identify deteriorating patients and trigger appropriate and timely escalation of care. Patients presenting to the emergency department (ED) are undiagnosed, undifferentiated and of varying acuity, yet the effectiveness and cost-effectiveness of using early warning systems and TTS in this setting is unclear. We aimed to systematically review the evidence on the use, development/validation, clinical effectiveness and cost-effectiveness of physiologically based early warning systems and TTS for the detection of deterioration in adult patients presenting to EDs.

METHODS

We searched for any study design in scientific databases and grey literature resources up to March 2016. Two reviewers independently screened results and conducted quality assessment. One reviewer extracted data with independent verification of 50% by a second reviewer. Only information available in English was included. Due to the heterogeneity of reporting across studies, results were synthesised narratively and in evidence tables.

RESULTS

We identified 6397 citations of which 47 studies and 1 clinical trial registration were included. Although early warning systems are increasingly used in EDs, compliance varies. One non-randomised controlled trial found that using an early warning system in the ED may lead to a change in patient management but may not reduce adverse events; however, this is uncertain, considering the very low quality of evidence. Twenty-eight different early warning systems were developed/validated in 36 studies. There is relatively good evidence on the predictive ability of certain early warning systems on mortality and ICU/hospital admission. No health economic data were identified.

CONCLUSIONS

Early warning systems seem to predict adverse outcomes in adult patients of varying acuity presenting to the ED but there is a lack of high quality comparative studies to examine the effect of using early warning systems on patient outcomes. Such studies should include health economics assessments.

Diagnostic Accuracy of Random ECG in Primary Care for Early, Asymptomatic Cardiac Autonomic Neuropathy

AIMS

Cardiac autonomic reflex tests (CARTs) are time consuming and require patient cooperation for detecting cardiac autonomic neuropathy (CAN). Heart rate variability (HRV) analysis requires less patient cooperation and is quicker to complete. However the reliability of HRV results as a clinical tool, with respect to length of recording and accuracy of diagnosis is inconclusive. The current study investigated the reproducibility associated with varying length of recording for early CAN (eCAN) assessment.

METHODS

Participants were 68 males, 72 females with average age of 55 for controls and 63 for early CAN. Inclusion criteria were that participants were medication free and presented with no comorbidities. ECGs of control and eCAN were recorded and heart rate changes analyzed with the fast Fourier transform (FFT) and Lomb-Scargle periodogram (LSP). Ten-second to 5-minute recordings were extracted from a 15-minute lead-II ECG and accuracy in assessment of eCAN determined.

RESULTS

The eCAN group was older ( P < .001) and systolic blood pressure was higher ( P < .01). HDL-cholesterol was also higher in the eCAN group ( P < .05). HRV analysis showed that both FFT and LSP results were significantly different between eCAN and control down to a 10-second ECG length for low frequency (LSP: P = .013, FFT: P = .024) and high frequency (HF-LSP: P = .002, FFT: P = .002) power. eCAN assessment was optimal down to 90-second recordings with a sensitivity of 100% and specificity of 29.49%.

CONCLUSION

HRV is suitable for clinical practice from ECG recordings of more than 90 seconds with high accuracy and repeatability within a session for each participant.

Fuzzy risk stratification and risk assessment model for clinical monitoring in the ICU

BACKGROUND

The decisions that clinicians make in intensive care units (ICUs) based on monitored parameters reflecting physiological deterioration are of major medical and biomedical engineering interest. These parameters have been investigated and assessed for their usefulness in risk assessment.

METHODS

Totally, 127 ICU adult patients were studied. They were selected from a MIMIC II Waveform Database Matched Subset and had continuous monitoring of heart rate, invasive blood pressure, and oxygen saturation. The monitored data were dimension reduced using deep learning autoencoders and then used to train a support vector machine model (SVM). A combination of methods including fuzzy c-means clustering (FCM), and a random forest (RF) was used to determine the risk levels.

RESULTS

When classifying patients into stable or deteriorating groups the main performance parameter was the receiver operating characteristics (ROC). The area under the ROC (AUROC) was 93.2 (95% CI (92.9-93.4)) with sensitivity and specificity values of 0.80 and 0.89, respectively. The suggested fuzzy risk levels using the combined method of the FCM clustering and RF achieved an accuracy of 1 (0.9999, 1), with both sensitivity and specificity values equal to 1.

CONCLUSIONS

The potential for using models in risk assessment to estimate a patient's physiological status, stable or deteriorating, within 4 h has been demonstrated. The study was based on retrospective analysis and further studies are needed to evaluate the impact on clinical outcomes using this model.

How to improve vital sign data quality for use in clinical decision support systems? A qualitative study in nine Swedish emergency departments

Background

Vital sign data are important for clinical decision making in emergency care. Clinical Decision Support Systems (CDSS) have been advocated to increase patient safety and quality of care. However, the efficiency of CDSS depends on the quality of the underlying vital sign data. Therefore, possible factors affecting vital sign data quality need to be understood.

This study aims to explore the factors affecting vital sign data quality in Swedish emergency departments and to determine in how far clinicians perceive vital sign data to be fit for use in clinical decision support systems. A further aim of the study is to provide recommendations on how to improve vital sign data quality in emergency departments.

Methods

Semi-structured interviews were conducted with sixteen physicians and nurses from nine hospitals and vital sign documentation templates were collected and analysed. Follow-up interviews and process observations were done at three of the hospitals to verify the results. Content analysis with constant comparison of the data was used to analyse and categorize the collected data.

Results

Factors related to care process and information technology were perceived to affect vital sign data quality. Despite electronic health records (EHRs) being available in all hospitals, these were not always used for vital sign documentation. Only four out of nine sites had a completely digitalized vital sign documentation flow and paper-based triage records were perceived to provide a better mobile workflow support than EHRs. Observed documentation practices resulted in low currency, completeness, and interoperability of the vital signs. To improve vital sign data quality, we propose to standardize the care process, improve the digital documentation support, provide workflow support, ensure interoperability and perform quality control.

Conclusions

Vital sign data quality in Swedish emergency departments is currently not fit for use by CDSS. To address both technical and organisational challenges, we propose five steps for vital sign data quality improvement to be implemented in emergency care settings.

Electronic supplementary material

The online version of this article (doi:10.1186/s12911-016-0305-4) contains supplementary material, which is available to authorized users.

Temporal patterns of change in vital signs and Cardiac Arrest Risk Triage scores over the 48 hours preceding fatal in-hospital cardiac arrest

AIM

To determine temporal patterns of vital sign and Cardiac Arrest Risk Triage score changes over the 48-hour period preceding cardiac arrest in an ICU setting.

BACKGROUND

Vital sign instability usually occurs prior to cardiac arrest. However, few studies have been conducted on the temporal patterns of individual vital signs preceding cardiac arrest.

DESIGN

A retrospective case-control study.

METHODS

The study subjects were 140 ICU patients (1 June 2011-31 December 2012): 46 died of cardiac arrest (case group), 45 died of other illnesses (control I group) and 49 were discharged after recovering (control II group).

RESULTS

Initial detectable changes in blood pressure appeared 18-20 hours and became dramatic at 5-10 hours before cardiac arrest. Noticeable changes in heart rates began at 4 hours and became more prominent at 2 hours pre-arrest. No apparent patterns in respiratory rate changes were observed. Body temperatures usually indicated a hypothermic state pre-arrest. Cardiac Arrest Risk Triage scores were 16-18 at 48 hours pre-arrest and then continuously increased to 20. Only mean values of systolic blood pressures were significantly different between the three study groups. Mean diastolic blood pressures, heart rates, respiratory rates and Cardiac Arrest Risk Triage scores differed between the case and control II groups and between the control I and II groups.

CONCLUSION

The study demonstrates vital sign instability preceded cardiac arrest and that the temporal patterns of changes in individual vital signs and Cardiac Arrest Risk Triage scores differed between groups. The findings of this study may aid the development of management strategies for cardiac arrest.

Integrating Real Time Data to Improve Outcomes in Acute Kidney Injury

Critically ill patients with acute kidney injury requiring renal replacement therapy have a poor prognosis. Despite well-known factors, which contribute to outcomes, including dose delivery, patients frequently miss the target dose and volume removal. One major barrier to effective care of these patients is the traditional dissociation of dialysis device data from other clinical information systems, notably the electronic health record (EHR). This lack of integration and the resulting manual documentation leads to errors and biases in documentation and missed opportunities to intervene in a timely fashion. This review summarizes the technological advancements facilitating direct connection of dialysis devices to EHRs. This connection facilitates automated data capture of many variables - including delivered dose, ultrafiltration rate and pressure measurements - which in turn can be leveraged for data mining, quality improvement and real-time targeted therapy adjustments. These interventions hold the promise to significantly improve outcomes for this patient population.

Predicting Health Care Utilization After Behavioral Health Referral Using Natural Language Processing and Machine Learning

Mental health problems are an independent predictor of increased healthcare utilization. We created random forest classifiers for predicting two outcomes following a patient's first behavioral health encounter: decreased utilization by any amount (AUROC 0.74) and ultra-high absolute utilization (AUROC 0.88). These models may be used for clinical decision support by referring providers, to automatically detect patients who may benefit from referral, for cost management, or for risk/protection factor analysis.

Manifold ranking based scoring system with its application to cardiac arrest prediction: A retrospective study in emergency department patients

BACKGROUND

The recently developed geometric distance scoring system has shown the effectiveness of scoring systems in predicting cardiac arrest within 72h and the potential to predict other clinical outcomes. However, the geometric distance scoring system predicts scores based on only local structure embedded by the data, thus leaving much room for improvement in terms of prediction accuracy.

METHODS

We developed a novel scoring system for predicting cardiac arrest within 72h. The scoring system was developed based on a semi-supervised learning algorithm, manifold ranking, which explores both the local and global consistency of the data. System evaluation was conducted on emergency department patients׳ data, including both vital signs and heart rate variability (HRV) parameters. Comparison of the proposed scoring system with previous work was given in terms of sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV).

RESULTS

Out of 1025 patients, 52 (5.1%) met the primary outcome. Experimental results show that the proposed scoring system was able to achieve higher area under the curve (AUC) on both the balanced dataset (0.907 vs. 0.824) and the imbalanced dataset (0.774 vs. 0.734) compared to the geometric distance scoring system.

CONCLUSIONS

The proposed scoring system improved the prediction accuracy by utilizing the global consistency of the training data. We foresee the potential of extending this scoring system, as well as manifold ranking algorithm, to other medical decision making problems. Furthermore, we will investigate the parameter selection process and other techniques to improve performance on the imbalanced dataset.

Novel Methods for Pulse Wave Velocity Measurement

The great incidence of cardiovascular (CV) diseases in the world spurs the search for new solutions to enable an early detection of pathological processes and provides more precise diagnosis based in multi-parameters assessment. The pulse wave velocity (PWV) is considered one of the most important clinical parameters for evaluate the CV risk, vascular adaptation, and therapeutic efficacy. Several studies were dedicated to find the relationship between PWV measurement and pathological status in different diseases, and proved the relevance of this parameter. The commercial devices dedicate to PWV estimation make a regional assessment (measured between two vessels), however a local measurement is more precise evaluation of artery condition, taking into account the differences in the structure of arteries. Moreover, the current devices present some limitations due to the contact nature. Emerging trends in CV monitoring are moving away from more invasive technologies to non-invasive and non-contact solutions. The great challenge is to explore the new instrumental solutions that allow the PWV assessment with fewer approximations for an accurately evaluation and relatively inexpensive techniques in order to be used in the clinical routine.

Factors affecting the ambulance response times of trauma incidents in Singapore

OBJECTIVES

Time to definitive care is important for trauma outcomes, thus many emergency medical services (EMS) systems in the world adopt response times of ambulances as a key performance indicator. The objective of this study is to examine the underlying risk factors that can affect ambulance response times (ART) for trauma incidents, so as to derive interventional measures that can improve the ART.

MATERIAL AND METHODS

This was a retrospective study based on two years of trauma data obtained from the national EMS operations centre of Singapore. Trauma patients served by the national EMS provider over the period from 1 January 2011 till 31 December 2012 were included. ART was categorized into "Short" (<4min), "Intermediate" (4-8min) and "Long" (>8min) response times. A modelling framework which leveraged on both multinomial logistic (MNL) regression models and Bayesian networks was proposed for the identification of main and interaction effects.

RESULTS

Amongst the process-related risk factors, weather, traffic and place of incident were found to be significant. The traffic conditions on the roads were found to have the largest effect-the odds ratio (OR) of "Long" ART in heavy traffic condition was 12.98 (95% CI: 10.66-15.79) times higher than that under light traffic conditions. In addition, the ORs of "Long ART" under "Heavy Rain" condition were significantly higher (OR 1.58, 95% CI: 1.26-1.97) than calls responded under "Fine" weather. After accounting for confounders, the ORs of "Long" ART for trauma incidents at "Home" or "Commercial" locations were also significantly higher than that for "Road" incidents.

CONCLUSION

Traffic, weather and the place of incident were found to be significant in affecting the ART. The evaluation of factors affecting the ART enables the development of effective interventions for reducing the ART.

Using what you get: dynamic physiologic signatures of critical illness

The development and resolution of cardiopulmonary instability take time to become clinically apparent, and the treatments provided take time to have an impact. The characterization of dynamic changes in hemodynamic and metabolic variables is implicit in physiologic signatures. When primary variables are collected with high enough frequency to derive new variables, this data hierarchy can be used to develop physiologic signatures. The creation of physiologic signatures requires no new information; additional knowledge is extracted from data that already exist. It is possible to create physiologic signatures for each stage in the process of clinical decompensation and recovery to improve outcomes.

Autonomic nervous system activity as risk predictor in the medical emergency department: a prospective cohort study

OBJECTIVES

To evaluate heart rate deceleration capacity, an electrocardiogram-based marker of autonomic nervous system activity, as risk predictor in a medical emergency department and to test its incremental predictive value to the modified early warning score.

DESIGN

Prospective cohort study.

SETTING

Medical emergency department of a large university hospital.

PATIENTS

Five thousand seven hundred thirty consecutive patients of either sex in sinus rhythm, who were admitted to the medical emergency department of the University of Tübingen, Germany, between November 2010 and March 2012.

INTERVENTIONS

None.

MEASUREMENTS AND MAIN RESULTS

Deceleration capacity of heart rate was calculated within the first minutes after emergency department admission. The modified early warning score was assessed from respiratory rate, heart rate, systolic blood pressure, body temperature, and level of consciousness as previously described. Primary endpoint was intrahospital mortality; secondary endpoints included transfer to the ICU as well as 30-day and 180-day mortality. One hundred forty-two patients (2.5%) reached the primary endpoint. Deceleration capacity was highly significantly lower in nonsurvivors than survivors (2.9 ± 2.1 ms vs 5.6 ± 2.9 ms; p < 0.001) and yielded an area under the receiver-operator characteristic curve of 0.780 (95% CI, 0.745-0.813). The modified early warning score model yielded an area under the receiver-operator characteristic curve of 0.706 (0.667-0.750). Implementing deceleration capacity into the modified early warning score model led to a highly significant increase of the area under the receiver-operator characteristic curve to 0.804 (0.770-0.835; p < 0.001 for difference). Deceleration capacity was also a highly significant predictor of 30-day and 180-day mortality as well as transfer to the ICU.

CONCLUSIONS

Deceleration capacity is a strong and independent predictor of short-term mortality among patients admitted to a medical emergency department.

Predictive modeling in pediatric traumatic brain injury using machine learning

BACKGROUND

Pediatric traumatic brain injury (TBI) constitutes a significant burden and diagnostic challenge in the emergency department (ED). While large North American research networks have derived clinical prediction rules for the head injured child, these may not be generalizable to practices in countries with traditionally low rates of computed tomography (CT). We aim to study predictors for moderate to severe TBI in our ED population aged < 16 years.

METHODS

This was a retrospective case-control study based on data from a prospective surveillance head injury database. Cases were included if patients presented from 2006 to 2014, with moderate to severe TBI. Controls were age-matched head injured children from the registry, obtained in a 4 control: 1 case ratio. These children remained well on diagnosis and follow up. Demographics, history, and physical examination findings were analyzed and patients followed up for the clinical course and outcome measures of death and neurosurgical intervention. To predict moderate to severe TBI, we built a machine learning (ML) model and a multivariable logistic regression model and compared their performances by means of Receiver Operating Characteristic (ROC) analysis.

RESULTS

There were 39 cases and 156 age-matched controls. The following 4 predictors remained statistically significant after multivariable analysis: Involvement in road traffic accident, a history of loss of consciousness, vomiting and signs of base of skull fracture. The logistic regression model was created with these 4 variables while the ML model was built with 3 extra variables, namely the presence of seizure, confusion and clinical signs of skull fracture. At the optimal cutoff scores, the ML method improved upon the logistic regression method with respect to the area under the ROC curve (0.98 vs 0.93), sensitivity (94.9% vs 82.1%), specificity (97.4% vs 92.3%), PPV (90.2% vs 72.7%), and NPV (98.7% vs 95.4%).

CONCLUSIONS

In this study, we demonstrated the feasibility of using machine learning as a tool to predict moderate to severe TBI. If validated on a large scale, the ML method has the potential not only to guide discretionary use of CT, but also a more careful selection of head injured children who warrant closer monitoring in the hospital.

Prediction of adverse cardiac events in emergency department patients with chest pain using machine learning for variable selection

BACKGROUND

The key aim of triage in chest pain patients is to identify those with high risk of adverse cardiac events as they require intensive monitoring and early intervention. In this study, we aim to discover the most relevant variables for risk prediction of major adverse cardiac events (MACE) using clinical signs and heart rate variability.

METHODS

A total of 702 chest pain patients at the Emergency Department (ED) of a tertiary hospital in Singapore were included in this study. The recruited patients were at least 30 years of age and who presented to the ED with a primary complaint of non-traumatic chest pain. The primary outcome was a composite of MACE such as death and cardiac arrest within 72 h of arrival at the ED. For each patient, eight clinical signs such as blood pressure and temperature were measured, and a 5-min ECG was recorded to derive heart rate variability parameters. A random forest-based novel method was developed to select the most relevant variables. A geometric distance-based machine learning scoring system was then implemented to derive a risk score from 0 to 100.

RESULTS

Out of 702 patients, 29 (4.1%) met the primary outcome. We selected the 3 most relevant variables for predicting MACE, which were systolic blood pressure, the mean RR interval and the mean instantaneous heart rate. The scoring system with these 3 variables produced an area under the curve (AUC) of 0.812, and a cutoff score of 43 gave a sensitivity of 82.8% and specificity of 63.4%, while the scoring system with all the 23 variables had an AUC of 0.736, and a cutoff score of 49 gave a sensitivity of 72.4% and specificity of 63.0%. Conventional thrombolysis in myocardial infarction score and the modified early warning score achieved AUC values of 0.637 and 0.622, respectively.

CONCLUSIONS

It is observed that a few predictors outperformed the whole set of variables in predicting MACE within 72 h. We conclude that more predictors do not necessarily guarantee better prediction results. Furthermore, machine learning-based variable selection seems promising in discovering a few relevant and significant measures as predictors.