Comparison of correctly and incorrectly classified patients for in-hospital mortality prediction in the intensive care unit.

BACKGROUND: The use of machine learning is becoming increasingly popular in many disciplines, but there is still an implementation gap of machine learning models in clinical settings. Lack of trust in models is one of the issues that need to be addressed in an effort to close this gap. No models are perfect, and it is crucial to know in which use cases we can trust a model and for which cases it is less reliable. METHODS: Four different algorithms are trained on the eICU Collaborative Research Database using similar features as the APACHE IV severity-of-disease scoring system to predict hospital mortality in the ICU. The training and testing procedure is repeated 100 times on the same dataset to investigate whether predictions for single patients change with small changes in the models. Features are then analysed separately to investigate potential differences between patients consistently classified correctly and incorrectly. RESULTS: A total of 34 056 patients (58.4%) are classified as true negative, 6 527 patients (11.3%) as false positive, 3 984 patients (6.8%) as true positive, and 546 patients (0.9%) as false negatives. The remaining 13 108 patients (22.5%) are inconsistently classified across models and rounds. Histograms and distributions of feature values are compared visually to investigate differences between groups. CONCLUSIONS: It is impossible to distinguish the groups using single features alone. Considering a combination of features, the difference between the groups is clearer. Incorrectly classified patients have features more similar to patients with the same prediction rather than the same outcome.

Alpha-2-adrenergic receptor agonists for the prevention of delirium and cognitive decline after open heart surgery (ALPHA2PREVENT): protocol for a multicentre randomised controlled trial.

INTRODUCTION: Postoperative delirium is common in older cardiac surgery patients and associated with negative short-term and long-term outcomes. The alpha-2-adrenergic receptor agonist dexmedetomidine shows promise as prophylaxis and treatment for delirium in intensive care units (ICU) and postoperative settings. Clonidine has similar pharmacological properties and can be administered both parenterally and orally. We aim to study whether repurposing of clonidine can represent a novel treatment option for delirium, and the possible effects of dexmedetomidine and clonidine on long-term cognitive trajectories, motor activity patterns and biomarkers of neuronal injury, and whether these effects are associated with frailty status. METHODS AND ANALYSIS: This five-centre, double-blind randomised controlled trial will include 900 cardiac surgery patients aged 70+ years. Participants will be randomised 1:1:1 to dexmedetomidine or clonidine or placebo. The study drug will be given as a continuous intravenous infusion from the start of cardiopulmonary bypass, at a rate of 0.4 µg/kg/hour. The infusion rate will be decreased to 0.2 µg/kg/hour postoperatively and be continued until discharge from the ICU or 24 hours postoperatively, whichever happens first.Primary end point is the 7-day cumulative incidence of postoperative delirium (Diagnostic and Statistical Manual of Mental Disorders, fifth edition). Secondary end points include the composite end point of coma, delirium or death, in addition to delirium severity and motor activity patterns, levels of circulating biomarkers of neuronal injury, cognitive function and frailty status 1 and 6 months after surgery. ETHICS AND DISSEMINATION: This trial is approved by the Regional Committee for Ethics in Medical Research in Norway (South-East Norway) and by the Norwegian Medicines Agency. Dissemination plans include publication in peer-reviewed medical journals and presentation at scientific meetings. TRIAL REGISTRATION NUMBER: NCT05029050.

Comparative analysis of explainable machine learning prediction models for hospital mortality.

BACKGROUND: Machine learning (ML) holds the promise of becoming an essential tool for utilising the increasing amount of clinical data available for analysis and clinical decision support. However, the lack of trust in the models has limited the acceptance of this technology in healthcare. This mistrust is often credited to the shortage of model explainability and interpretability, where the relationship between the input and output of the models is unclear. Improving trust requires the development of more transparent ML methods. METHODS: In this paper, we use the publicly available eICU database to construct a number of ML models before examining their internal behaviour with SHapley Additive exPlanations (SHAP) values. Our four models predicted hospital mortality in ICU patients using a selection of the same features used to calculate the APACHE IV score and were based on random forest, logistic regression, naive Bayes, and adaptive boosting algorithms. RESULTS: The results showed the models had similar discriminative abilities and mostly agreed on feature importance while calibration and impact of individual features differed considerably and did in multiple cases not correspond to common medical theory. CONCLUSIONS: We already know that ML models treat data differently depending on the underlying algorithm. Our comparative analysis visualises implications of these differences and their importance in a healthcare setting. SHAP value analysis is a promising method for incorporating explainability in model development and usage and might yield better and more trustworthy ML models in the future.

Characteristics of circulatory failure after out-of-hospital cardiac arrest: a prospective cohort study.

BACKGROUND: Circulatory failure after out-of-hospital cardiac arrest (OHCA) as part of the postcardiac arrest syndrome (PCAS) is believed to be caused by an initial myocardial depression that later subsides into a superimposed vasodilatation. However, the relative contribution of myocardial dysfunction and systemic inflammation has not been established. Our objective was to describe the macrocirculatory and microcirculatory failure in PCAS in more detail. METHODS: We included 42 comatose patients after OHCA where circulatory variables were invasively monitored from admission until day 5. We measured the development in cardiac power output (CPO), stroke work (SW), aortic elastance, microcirculatory metabolism, inflammatory and cardiac biomarkers and need for vasoactive medications. We used survival analysis and Cox regression to assess time to norepinephrine discontinuation and negative fluid balance, stratified by inflammatory and cardiac biomarkers. RESULTS: CPO, SW and oxygen delivery increased during the first 48 hours. Although the estimated afterload fell, the blood pressure was kept above 65 mmHg with a diminishing need for norepinephrine, indicating a gradually re-established macrocirculatory homoeostasis. Time to norepinephrine discontinuation was longer for patients with higher pro-brain natriuretic peptide concentration (HR 0.45, 95% CI 0.21 to 0.96), while inflammatory biomarkers and other cardiac biomarkers did not predict the duration of vasoactive pressure support. Markers of microcirculatory distress, such as lactate and venous-to-arterial carbon dioxide difference, were normalised within 24 hours. CONCLUSION: The circulatory failure was initially characterised by reduced CPO and SW, however, microcirculatory and macrocirculatory homoeostasis was restored within 48 hours. We found that biomarkers indicating acute heart failure, and not inflammation, predicted longer circulatory support with norepinephrine. Taken together, this indicates an early and resolving, rather than a late and emerging vasodilatation. TRIAL REGISTRATION: NCT02648061.

The inflammatory response is related to circulatory failure after out-of-hospital cardiac arrest: A prospective cohort study.

BACKGROUND: Whole body ischemia and reperfusion injury after cardiac arrest leads to the massive inflammation clinically manifested in the post-cardiac arrest syndrome. Previous studies on the inflammatory effect on circulatory failure after cardiac arrest have either investigated a selected patient group or a limited part of the inflammatory mechanisms. We examined the association between cardiac arrest characteristics and inflammatory biomarkers, and between inflammatory biomarkers and circulatory failure after cardiac arrest, in an unselected patient cohort. METHODS: This was a prospective study of 50 consecutive patients with out-of-hospital cardiac arrest. Circulation was invasively monitored from admission until day five, whereas inflammatory biomarkers, i.e. complement activation, cytokines and endothelial injury, were measured daily. We identified predictors for an increased inflammatory response, and associations between the inflammatory response and circulatory failure. RESULTS: We found a marked and broad inflammatory response in patients after cardiac arrest, which was associated with clinical outcome. Long time to return of spontaneous circulation and high lactate level at admission were associated with increased complement activation (TCC and C3bc), pro-inflammatory cytokines (IL-6, IL-8) and endothelial injury (syndecan-1) at admission. These biomarkers were in turn significantly associated with lower mean arterial blood pressure, lower cardiac output and lower systemic vascular resistance, and increased need of circulatory support in the initial phase. High levels of TCC and IL-6 at admission were significantly associated with increased 30-days mortality. CONCLUSION: Inflammatory biomarkers, including complement activation, cytokines and endothelial injury, were associated with increased circulatory failure in the initial period after cardiac arrest.

Circulatory trajectories after out-of-hospital cardiac arrest: a prospective cohort study.

BACKGROUND: Circulatory failure frequently occurs after out-of-hospital cardiac arrest (OHCA) and is part of post-cardiac arrest syndrome (PCAS). The aim of this study was to investigate circulatory disturbances in PCAS by assessing the circulatory trajectory during treatment in the intensive care unit (ICU). METHODS: This was a prospective single-center observational cohort study of patients after OHCA. Circulation was continuously and invasively monitored from the time of admission through the following five days. Every hour, patients were classified into one of three predefined circulatory states, yielding a longitudinal sequence of states for each patient. We used sequence analysis to describe the overall circulatory development and to identify clusters of patients with similar circulatory trajectories. We used ordered logistic regression to identify predictors for cluster membership. RESULTS: Among 71 patients admitted to the ICU after OHCA during the study period, 50 were included in the study. The overall circulatory development after OHCA was two-phased. Low cardiac output (CO) and high systemic vascular resistance (SVR) characterized the initial phase, whereas high CO and low SVR characterized the later phase. Most patients were stabilized with respect to circulatory state within 72 h after cardiac arrest. We identified four clusters of circulatory trajectories. Initial shockable cardiac rhythm was associated with a favorable circulatory trajectory, whereas low base excess at admission was associated with an unfavorable circulatory trajectory. CONCLUSION: Circulatory failure after OHCA exhibits time-dependent characteristics. We identified four distinct circulatory trajectories and their characteristics. These findings may guide clinical support for circulatory failure after OHCA. TRIAL REGISTRATION: ClinicalTrials.gov: NCT02648061.

Rapid declines in systolic blood pressure are associated with an increase in pulse transit time.

BACKGROUND: The correlation between pulse transit time and blood pressure has been proposed as a route to measure continuous non-invasive blood pressure. We investigated whether pulse transit time trends could model blood pressure trends during episodes of rapid declines in blood pressure. METHODS: From the Medical Information Mart for Intensive Care waveform database we identified substantial blood pressure reductions. Pulse transit time was calculated from the R-peak of the electrocardiogram to the peak of the arterial pulse waveform. The time-series were processed with a moving average filter before comparison. Averaged, continuous heart rate was also analysed as a control. The intra-individual association between variables was assessed per subject using linear regression. RESULTS: In the 511 patients included we found a median correlation coefficient between blood pressure and pulse transit time of -0.93 (IQR -0.98 to -0.76) with regression slopes of -1.23 mmHg/ms (IQR -1.73 to -0.81). The median correlation coefficient between blood pressure and heart rate was 0.46 (IQR -0.16 to 0.83). In supplementary analysis, results did not differ substantially when widening inclusion criteria, but the results were not always consistent within subjects across episodes of hypotension. CONCLUSIONS: In a large cohort of critically ill patients experiencing episodes of rapid declines in systolic blood pressure, there was a moderate-strong intra-individual correlation between averaged systolic blood pressure and averaged pulse transit time. Our findings encourage further investigation into using the pulse transit time for non-invasive real-time detection of hypotension.

Some oscillatory phenomena of blood glucose regulation: An exploratory pilot study in pigs.

It is well-known that blood glucose oscillates with a period of approximately 15 min (900 s) and exhibits an overall complex behaviour in intact organisms. This complexity is not thoroughly studied, and thus, we aimed to decipher the frequency bands entailed in blood glucose regulation. We explored high-resolution blood glucose time-series sampled using a novel continuous intravascular sensor in four pigs under general anaesthesia for almost 24 hours. In all time series, we found several interesting oscillatory components, especially in the 5000-10000 s, 500-1000 s, and 50-100 s regions (0.0002-0.0001 Hz, 0.002-0.001 Hz, and 0.02-0.01 Hz). The presence of these oscillations is not permanent, as they come and go. This is the first report of glucose oscillations in the 50-100 s range. The origin of these oscillations and their role in overall blood glucose regulation is unknown. Although the sample size is small, we believe this finding is important for our understanding of glucose regulation and perhaps for our understanding of general homeostatic regulation in intact organisms.

Transitions Between Circulatory States After Out-of-Hospital Cardiac Arrest: Protocol for an Observational, Prospective Cohort Study.

BACKGROUND: The post cardiac arrest syndrome (PCAS) is responsible for the majority of in-hospital deaths following cardiac arrest (CA). The major elements of PCAS are anoxic brain injury and circulatory failure. OBJECTIVE: This study aimed to investigate the clinical characteristics of circulatory failure and inflammatory responses after out-of-hospital cardiac arrest (OHCA) and to identify patterns of circulatory and inflammatory responses, which may predict circulatory deterioration in PCAS. METHODS: This study is a single-center cohort study of 50 patients who receive intensive care after OHCA. The patients are followed for 5 days where detailed information from circulatory variables, including measurements by pulmonary artery catheters (PACs), is obtained in high resolution. Blood samples for inflammatory and endothelial biomarkers are taken at inclusion and thereafter daily. Every 10 min, the patients will be assessed and categorized in one of three circulatory categories. These categories are based on mean arterial pressure; heart rate; serum lactate concentrations; superior vena cava oxygen saturation; and need for fluid, vasoactive medications, and other interventions. We will analyze predictors of circulatory failure and their relation to inflammatory biomarkers. RESULTS: Patient inclusion started in January 2016. CONCLUSIONS: This study will obtain advanced hemodynamic data with high resolution during the acute phase of PCAS and will analyze the details in circulatory state transitions related to circulatory failure. We aim to identify early predictors of circulatory deterioration and favorable outcome after CA. TRIAL REGISTRATION: ClinicalTrials.gov: NCT02648061; https://clinicaltrials.gov/ct2/show/NCT02648061 (Archived by WebCite at http://www.webcitation.org/6wVASuOla).

The coherence of macrocirculation, microcirculation, and tissue metabolic response during nontraumatic hemorrhagic shock in swine.

Hemorrhagic shock is clinically observed as changes in macrocirculatory indices, while its main pathological constituent is cellular asphyxia due to microcirculatory alterations. The coherence between macro- and microcirculatory changes in different shock states has been questioned. This also applies to the hemorrhagic shock. Most studies, as well as clinical situations, of hemorrhagic shock include a "second hit" by tissue trauma. It is therefore unclear to what extent the hemorrhage itself contributes to this lack of circulatory coherence. Nine pigs in general anesthesia were exposed to a controlled withdrawal of 50% of their blood volume over 30 min, and then retransfusion over 20 min after 70 min of hypovolemia. We collected macrocirculatory variables, microcirculatory blood flow measurement by the fluorescent microspheres technique, as well as global microcirculatory patency by calculation of Pv-aCO2, and tissue metabolism measurement by the use of microdialysis. The hemorrhage led to anticipated changes in macrocirculatory variables with a coherent change in microcirculatory and metabolic variables. In the late hemorrhagic phase, the animals' variables generally improved, probably through recruitment of venous blood reservoirs. After retransfusion, all variables were normalized and remained same throughout the study period. We find in our nontraumatic model consistent coherence between changes in macrocirculatory indices, microcirculatory blood flow, and tissue metabolic response during hemorrhagic shock and retransfusion. This indicates that severe, but brief, hemorrhage with minimal tissue injury is in itself not sufficient to cause lack of coherence between macro- and microcirculation.

A pig model of acute right ventricular afterload increase by hypoxic pulmonary vasoconstriction.

BACKGROUND: The aim of this study was to construct a non-invasive model for acute right ventricular afterload increase by hypoxic pulmonary vasoconstriction. Intact animal models are vital to improving our understanding of the pathophysiology of acute right ventricular failure. Acute right ventricular failure is caused by increased afterload of the right ventricle by chronic or acute pulmonary hypertension combined with regionally or globally reduced right ventricular contractile capacity. Previous models are hampered by their invasiveness; this is unfortunate as the pulmonary circulation is a low-pressure system that needs to be studied in closed chest animals. Hypoxic pulmonary vasoconstriction is a mechanism that causes vasoconstriction in alveolar vessels in response to alveolar hypoxia. In this study we explored the use of hypoxic pulmonary vasoconstriction as a means to increase the pressure load on the right ventricle. RESULTS: Pulmonary hypertension was induced by lowering the FiO2 to levels below the physiological range in eight anesthetized and mechanically ventilated pigs. The pigs were monitored with blood pressure measurements and blood gases. The mean pulmonary artery pressures (mPAP) of the animals increased from 18.3 (4.2) to 28.4 (4.6) mmHg and the pulmonary vascular resistance (PVR) from 254 (76) dyns/cm5 to 504 (191) dyns/cm5, with a lowering of FiO2 from 0.30 to 0.15 (0.024). The animals' individual baseline mPAPs varied substantially as did their response to hypoxia. The reduced FiO2 level yielded an overall lowering in oxygen offer, but the global oxygen consumption was unaltered. CONCLUSIONS: We showed in this study that the mPAP and the PVR could be raised by approximately 100% in the study animals by lowering the FiO2 from 0.30 to 0.15 (0.024). We therefore present a novel method for minimally invasive (closed chest) right ventricular afterload manipulations intended for future studies of acute right ventricular failure. The method should in theory be reversible, although this was not studied in this work.