The relationship between coagulation abnormality and mortality in ICU patients: a prospective, observational study

Explainable machine learning model for prediction of 28-day all-cause mortality in immunocompromised patients in the intensive care unit: a retrospective cohort study based on MIMIC-IV database

OBJECTIVES

This study aimed to develop and validate an explainable machine learning (ML) model to predict 28-day all-cause mortality in immunocompromised patients admitted to the intensive care unit (ICU). Accurate and interpretable mortality prediction is crucial for clinical decision-making and optimal allocation of critical care resources for this vulnerable patient population.

METHODS

We utilized retrospective clinical data from the MIMIC-IV (version 2.2) database, encompassing ICU admissions at Beth Israel Deaconess Medical Center from 2008 to 2019. Eligible immunocompromised patients, including those with primary immunodeficiencies and chronic acquired conditions, such as hematological malignancies, solid tumors, and organ transplantation, were selected. Data were randomly split into training (80%) and testing (20%) cohorts. Ten ML models (logistic regression, XGBoost, LightGBM, AdaBoost, Random Forest, Gradient Boosting, Gaussian Naive Bayes, Complement Naive Bayes, Multilayer Perceptron, and Support Vector Machine) were developed and evaluated using area under the receiver operating characteristic curve (AUROC), area under the precision-recall curve (AUPRC), sensitivity, specificity, accuracy, and F1 score. Model explainability was achieved through SHapley Additive exPlanations (SHAP), and decision curve analysis (DCA) assessed clinical utility. In addition, Cox proportional hazards regression was conducted to evaluate the impact of predictive factors on time-to-event outcomes.

RESULTS

Among the evaluated models, the Support Vector Machine (SVM) demonstrated the highest AUROC of 0.863 (95% CI 0.834-0.890) and a notable AUPRC of 0.678 (95% CI 0.624-0.736). Key predictive factors consistently identified across multiple ML models included 24-h urine output, blood urea nitrogen (BUN) levels, presence of metastatic solid tumors, Charlson Comorbidity Index (CCI), and international normalized ratio (INR). SHAP analyses provided detailed insights into how these features influenced model predictions.

CONCLUSIONS

The explainable ML models based on various artificial intelligence methods demonstrated promising clinical applicability in predicting 28-day mortality risk among immunocompromised ICU patients. Factors such as urine output, BUN, metastatic solid tumors, CCI, and INR significantly contributed to prediction outcomes and may serve as important predictors in clinical practice.

Interpretable machine learning model for early prediction of disseminated intravascular coagulation in critically ill children

Disseminated intravascular coagulation (DIC) is a thrombo-hemorrhagic disorder that can be life-threatening in critically ill children, and the quest for an accurate and efficient method for early DIC prediction is of paramount importance. Candidate predictors encompassed demographics, comorbidities, laboratory findings, and therapy strategies. A stepwise logistic regression model was employed to select the features included in the final model. Six machine learning algorithms-logistic regression (LR), extreme gradient boosting (XGB), random forest (RF), support vector machine (SVM), decision tree (DT), and k-nearest neighbor (KNN)-were employed to construct predictive models for DIC in critically ill children. Models were then evaluated by using area under the curve (AUC), accuracy, specificity, sensitivity, positive predictive value (PPV), negative predictive value (NPV), precision, recall and decision curve analysis (DCA). Interpretation of the optimal model was conducted using shapley additive explanations (SHAP). A total of 6093 critically ill children were encompassed in this study, of whom 681 (11.2%) developed DIC. The RF model exhibited the highest levels of accuracy (0.856), sensitivity (0.866), Kappa (0.472), NPV (0.423), and recall (0.866). However, the XGB model outperformed RF, LR, SVM, DT, and KNN in terms of AUC (0.908), specificity (0.859), PPV (0.978), and precision (0.969). Decision curve analysis (DCA) confirmed the superior clinical utility of the XGB model. Overall, the XGB model demonstrated superior clinical utility compared to RF, LR, SVM, DT, and KNN. We named the final model Alfalfa-PICU-DIC. SHAP analysis identified D-dimer, INR, PT, TT, and PLT count as the top predictors of DIC. Machine learning models can be a reliable tool for predicting DIC in critically ill children, which will facilitate timely intervention, thereby reducing the burden of DIC on patients in the pediatric intensive care unit (PICU).

Risk factors and predictive model for mortality in patients undergoing allogeneic hematopoietic stem cell transplantation admitted to the intensive care unit

Hematological malignant tumors represent a group of major diseases carrying a substantial risk to the lives of affected patients. Risk factors for mortality in critically ill patients have garnered substantial attention in recent research endeavors. The present research aimed to identify factors predicting intensive care unit (ICU) mortality in patients undergoing allogeneic hematopoietic stem cell transplantation (allo-HSCT). Furthermore, the present study analyzed and compared the mortality rate between patients undergoing haploidentical hematopoietic stem cell transplantation (Haplo-SCT) and those undergoing identical sibling donor (ISD) transplantation. A total of 108 patients were included in the present research, 83 (76.9%) of whom underwent Haplo-SCT. ICU mortality was reported in 58 (53.7%) patients, with the values of 55.4 and 48.0% associated with Haplo-SCT and ISD, respectively (P=0.514). The mortality rate of patients undergoing Haplo-SCT was comparable to that of patients undergoing ISD transplantation. The present study found that reduced hemoglobin, elevated total bilirubin, elevated brain natriuretic peptide, elevated fibrinogen degradation products, need for vasoactive drugs at ICU admission, need for invasive mechanical ventilation and elevated APACHE II scores were independent risk factors for ICU mortality. Among patients presenting with 5-7 risk factors, the ICU mortality reached 100%, significantly exceeding that of other patients. The present research revealed that ICU mortality rates remain elevated among patients who underwent allo-HSCT, especially those presenting multiple risk factors. However, the outcome of patients undergoing Haplo-SCT were comparable to those of patients undergoing ISD transplants.

A High Level of Fibrinogen Degradation Product on Arrival as the Only Clue Suggesting Deterioration in a Blunt Trauma Patient

We report the case of an 89-year-old woman who was struck by a car while walking and fell to the ground. She had hypertension, dyslipidemia, and cerebral infarction requiring medication. She was transported to a nearby acute critical care center. Upon arrival, her vital signs were stable. A physical examination showed right facial and hip contusion, right shoulder tenderness, a right elbow contusional lacerated wound, and bilateral knee abrasion wounds. She vomited when her face moved. Radiological studies showed a right proximal humerus fracture and a right minor ischial fracture. Her injury severity score (ISS) was 5 points, and her probability of surviving was 97.8%. However, a blood test revealed an extremely high fibrinogen degradation product (FDP) level (573.3 μg/mL). Because of this elevated FDP value and her inability to walk due to vomiting on motion, she remained in the emergency room (ER) for monitoring. At five hours from arrival, she became comatose, and hypotension and bradycardia (30 beats per minute) were noted followed by cardiac arrest. She underwent advanced cardiac life support and obtained spontaneous circulation. Repeated blood tests showed hyperkalemia, anemia, and hypoglycemia. She immediately underwent infusion of glucose and insulin and continuous infusion of catecholamine. Repeated whole-body CT scans revealed only increased hematomas where the fractures and contusions existed. She was admitted to the ICU. Her post-admission course was quite eventful. She required transfusion until the fourth hospital day to control circulation and anemia and underwent transfusion of 28 units of red blood cells, 30 units of platelets, and four units of fresh-frozen plasma in total. After her circulation and respiratory function had stabilized, she was extubated. However, her condition became complicated with the deterioration of her knee wounds and gall bladder inflammation in the ward. All complications were treated by non-operative management. She was transferred to another hospital for rehabilitation on day 70. This report discusses our experience with a blunt trauma patient in whom a high FDP level on arrival was the only clue indicating the deterioration of her condition. Such patients need close observation with hospitalization.

Sonoclot’s usefulness in prediction of cardiopulmonary arrest prognosis: A proof of concept study

The aim of the present study was to evaluate the usefulness of measuring whole blood coagulation with Sonoclot to predict return of spontaneous circulation (ROSC) in patients with out-of-hospital cardiopulmonary arrest (OHCA). This was a prospective, observational clinical study on patients with OHCA who were transferred to our emergency department between August 2016 and July 2018. Patients were divided into two groups: patients with return of spontaneous circulation (ROSC[+] group) and those without (ROSC[−] group). We compared the activated clotting time (ACT), clot rate (CR), and platelet function (PF) as measured with Sonoclot, and the fibrinogen degradation products (FDP) level and D-dimer level between the two groups. We analyzed 87 patients: 37 in the ROSC(+) and 50 in the ROSC(−) groups. Regarding ACT, CR, PF, FDP, and D-dimer, we used receiver operating characteristic (ROC) curves to examine how well each factor predicts ROSC. The area under the ROC curve (AUC) of CR was higher than that of the FDP and D-dimer levels. Among patients with cardiogenic cardiac arrest, the AUC of CR was higher than the AUCs of other coagulation factors. In conclusion, viscoelastic blood coagulation measurements using Sonoclot may be useful for predicting ROSC in OHCA patients.

Value of Hematological and Coagulation Parameters as Prognostic Factors in Acute Coronary Syndromes

The values of hematological and coagulation biomarkers were evaluated as predictors of in hospital mortality and complications, in patients with acute coronary syndromes (ACS). This retrospective observational study enrolled 936 ACS subjects admitted to the Clinical Emergency Hospital of Oradea, Romania, between January-December 2019. Hematological and coagulation parameters were obtained at admission. During hospitalization, the following adverse events were recorded: death, ventricular rhythm disturbances, atrial fibrillation, heart failure, re-infarction, and stroke. Accuracy of hematological and coagulation parameters as predictors of adverse outcome were also evaluated. The diagnosis was unstable angina in 442 patients (47.22%), non-ST-elevation myocardial infarction (NSTEMI) in 113 patients (12.1%) and ST-elevation myocardial infarction (STEMI) in 381 patients (40.70%); 87 patients (9.29%) died during hospitalization and 193 (20.7%) developed complications. Predictors for in hospital mortality were as follows: red cell distribution width (RDW) (AUC 0.691, p < 0.0001), white blood cells (WBC) (AUC 0.684, p < 0.0001), neutrophils (NEU) (AUC 0.684, p < 0.0001), and prothrombin time (PT) (AUC 0.765, p < 0.0001). WBC (AUC 0.659, p < 0.0001), NEU (AUC 0.664, p < 0.0001), RDW (AUC 0.669, p < 0.0001), and PT (AUC 0.669, 95% CI 0.622-0.714, p < 0.0001) also had accuracy for complications prediction. RDW had a good ability to predict heart failure in NSTEMI patients (AUC 0.832, p < 0.0001). An acceptable ability to predict ventricular rhythm disturbances occurrence had WBC (AUC 0.758, p < 0.0001) and NEU (AUC 0.772, p < 0.0001). Hematological and coagulation parameters can help in risk stratification of ACS patients. RDW, WBC, NEU, and PT were able to predict mortality and in-hospital complications in ACS patients. RDW has a good accuracy in predicting complications and heart failure in NSTEMI patients. WBC and NEU are good predictors for ventricular rhythm disturbances.

Noninvasive Real-Time Mortality Prediction in Intensive Care Units Based on Gradient Boosting Method: Model Development and Validation Study

BACKGROUND

Monitoring critically ill patients in intensive care units (ICUs) in real time is vitally important. Although scoring systems are most often used in risk prediction of mortality, they are usually not highly precise, and the clinical data are often simply weighted. This method is inefficient and time-consuming in the clinical setting.

OBJECTIVE

The objective of this study was to integrate all medical data and noninvasively predict the real-time mortality of ICU patients using a gradient boosting method. Specifically, our goal was to predict mortality using a noninvasive method to minimize the discomfort to patients.

METHODS

In this study, we established five models to predict mortality in real time based on different features. According to the monitoring, laboratory, and scoring data, we constructed the feature engineering. The five real-time mortality prediction models were RMM (based on monitoring features), RMA (based on monitoring features and the Acute Physiology and Chronic Health Evaluation [APACHE]), RMS (based on monitoring features and Sequential Organ Failure Assessment [SOFA]), RMML (based on monitoring and laboratory features), and RM (based on all monitoring, laboratory, and scoring features). All models were built using LightGBM and tested with XGBoost. We then compared the performance of all models, with particular focus on the noninvasive method, the RMM model.

RESULTS

After extensive experiments, the area under the curve of the RMM model was 0.8264, which was superior to that of the RMA and RMS models. Therefore, predicting mortality using the noninvasive method was both efficient and practical, as it eliminated the need for extra physical interventions on patients, such as the drawing of blood. In addition, we explored the top nine features relevant to real-time mortality prediction: invasive mean blood pressure, heart rate, invasive systolic blood pressure, oxygen concentration, oxygen saturation, balance of input and output, total input, invasive diastolic blood pressure, and noninvasive mean blood pressure. These nine features should be given more focus in routine clinical practice.

CONCLUSIONS

The results of this study may be helpful in real-time mortality prediction in patients in the ICU, especially the noninvasive method. It is efficient and favorable to patients, which offers a strong practical significance.

Prognostic factors for the short-term mortality of patients with rheumatoid arthritis admitted to intensive care units

Background

Patients with rheumatoid arthritis (RA) have high mortality risk and are frequently treated in intensive care units (ICUs).

Methods

This was a retrospective observational study. This study included 67 patients (20 males, 47 females) with RA who were admitted at the ICU of our institution for ≥48 h between January 2008 and December 2017. We analyzed the 30-day mortality of these patients and the investigated prognostic factors in RA patients admitted to our ICU.

Results

Upon admission, the median age was 70 (range, 33–96) years, and RA duration was 10 (range, 0–61) years. The 5-year survival after ICU admission was 47%, and 30-day, 90-day, and 1-year mortality rates were 22, 27, and 37%, respectively. The major reasons for ICU admission were cardiovascular complications (24%) and infection (40%) and the most common ICU treatments were mechanical ventilation (69%), renal replacement (25%), and vasopressor (78%). In the 30-day mortality group, infection led to a fatal outcome in most cases (67%), and nonsurvival was associated with a significantly higher glucocorticoid dose, updated Charlson’s comorbidity index (CCI), and acute physiology and chronic health evaluation (APACHE) II score. Laboratory data obtained at ICU admission showed that lower platelet number and total protein and higher creatinine and prothrombin time international normalized ratio (PT-INR) indicated significantly poorer prognosis. The multivariate Cox proportional hazard model revealed that nonuse of csDMARDs, high updated CCI, increased APACHE II score, and prolonged PT-INR were associated with a higher risk of mortality after ICU admission.

Conclusion

Our study demonstrated that the nonuse of csDMARDs, high updated CCI, elevated APACHE II score, and coagulation abnormalities predicted poorer prognosis in RA patients admitted to the ICU.

Coagulation tests on admission correlate with mortality and morbidity in general ICU patients: An observational study

BACKGROUND

It is well known that low platelet count on admission to intensive care units (ICU) is associated with increased mortality. However, it is unknown whether prothrombin time (PT-INR) and activated partial thromboplastin time (APTT) on admission correlate with mortality and organ failure. Therefore, the aim of this study was to investigate whether PT-INR and APTT at admission can predict outcome in the critically ill patient after adjusting for severity of illness measured with Simplified Acute Physiology Score 3 (SAPS 3).

MATERIALS AND METHODS

Data were retrospectively collected. APTT and PT-INR taken on admission and SAPS 3 score were independent variables in all regression analyses. Survival analysis was done with Cox regression. Organ failure was reported as days alive and free (DAF) of vasopressors and invasive ventilation, need of continuous renal replacement therapy (CRRT) and Acute Kidney Injury Network creatinine score (AKIN-crea).

RESULTS

A total of 3585 ICU patients were included. Prolonged APTT correlated with mortality with 95% confidence interval (CI) of hazard ratio 1.001-1.010. Prolonged APTT also correlated with DAF vasopressor, CRRT and AKIN-crea with 95% CI of odds ratio (OR) 1.009-1.034, 1.016-1.037 and 1.009-1.028, respectively. Increased PT-INR correlated with DAF vasopressor and DAF ventilator with 95% CI of OR 1.112-2.014 and 1.135-1.847, respectively.

CONCLUSIONS

Activated partial thromboplastin time prolongation was associated with mortality and all morbidity outcomes except the DAF ventilator. PT-INR increase at admission was associated with DAF vasopressor and DAF ventilator. APTT and PT-INR at admission correlate with morbidity, which is not accounted for in the SAPS 3 model.

Derivation of a model to predict mortality in urban patients with accidental hypothermia: a retrospective observational study

Aim

Accidental hypothermia in urban settings is associated with high mortality rates. However, the predictors of mortality remain under discussion. The purpose of this study was to evaluate prognostic factors and develop a prediction model in patients with accidental hypothermia in urban settings.

Methods

We retrospectively reviewed medical records in patients with hypothermia brought to our hospital by ambulance in a 7-year study period. Patients' records of survival discharge or in-hospital death and clinical data were collected from medical records. We analyzed factors to predict in-hospital death using multiple logistic regression analysis. Recursive partitioning analysis was used to construct a prediction model using predictors from multiple logistic regression analysis.

Results

In the study period, 192 patients were included in this study. Of them, 154 patients were discharged alive and 38 patients died. Multiple logistic regression analysis revealed that in-hospital death was related to Glasgow Coma Scale (GCS) score, prothrombin time - international normalized ratio (PT-INR) value, and fibrin degradation product (FDP). Recursive partitioning analysis revealed that patients with accidental hypothermia could be divided into four groups: very high risk (FDP ≥ 14 µg/mL, PT-INR ≥ 1.4), high risk (FDP ≥ 14 µg/mL, PT-INR < 1.4), moderate risk (FDP < 14 µg/mL, GCS < 10), and low risk (FDP < 14 µg/mL, GCS ≥ 10).

Conclusion

High FDP and PT-INR values and low GCS score on arrival at the emergency department were associated with in-hospital mortality in urban patients with hypothermia. A simple prediction model for grouping risk was developed using these predictors.

Biomarkers of disseminated intravascular coagulation in pediatric intensive care unit in Thailand

INTRODUCTION

Disseminated intravascular coagulation (DIC) is a systemic activation of hemostatic system caused by several causes. Biomarkers including antithrombin (AT), protein C (PC), and thrombomodulin (TM) were reported as the additional markers for DIC in adults. This study aimed to determine the association between biomarkers among patients with overt DIC (ODIC) and nonovert DIC (NDIC) in children in PICU.

METHODS

We enrolled 103 subjects, aged 1 month-18 years, who were admitted to PICU at Chiang Mai University (CMU) Hospital >24 hours with underlying conditions predisposing to DIC were enrolled. Biomarkers were tested after 24 hours of admission. Subject who had NDIC on the 1^st investigations would have other tests on days 3-5 of admission.

RESULTS

The incidence of ODIC by the International Society on Thrombosis and Hemostasis (ISTH) DIC score was found 24%. The bleeding, thrombosis, and death were significantly higher in ODIC group (P < 0.05). Mean levels of AT and PC in ODIC group were significantly different from NDIC one (66.9% vs 79.9%, P < 0.001 and 46.1% vs 59.2%, P = 0.004, respectively) while mean level of TM was not different between two groups. Adding AT to DIC score was better than the original score for predict mortality [area under curve (AUC) = 0.662 vs AUC = 0.65] and bleeding (AUC = 0.751 vs AUC = 0.732).

CONCLUSIONS

ODIC is prevalent among critically ill children. Adverse outcomes were more commonly found in children with ODIC. AT and PC levels after 24 hours of PICU admission seem to be the useful biomarkers for ODIC in PICU.

Preoperative elevated FDP may predict severe intraoperative hypotension after dural opening during decompressive craniectomy of traumatic brain injury

Purpose

Coagulation disorder and intraoperative hypotension are representative complications of traumatic brain injury which cause worse perioperative outcome. The aim of this study was to survey the relation of coagulation disorder and intraoperative hypotension (IH) during decompressive craniectomy.

Method

Patients who underwent emergency decompressive craniectomy due to traumatic brain injury were retrospectively surveyed. The relation between preoperative coagulation date and intraoperative hypotension (systolic blood pressure < 60 mmHg after dural opening) was analyzed.

Results

Of 41 patients screened, 12 patients (27.9%) developed IH. Fibrinogen degradation products (314 vs 64.4 μg/mL; p = 0.01) were significantly higher in the IH group. In contrast, fibrinogen (181 vs 239 mg/dL; p = 0.01) was significantly lower in the IH group. Reduction rate of sBRP before and after dural opening (%) was higher in IH group than in non-IH group (49.1 vs 27.6%: p = 0.001).

Conclusions

Preoperative elevated FDP may predict IH after dural opening during traumatic decompressive craniectomy.

The relation between international normalized ratio and mortality in acute pulmonary embolism: A retrospective study

BACKGROUND

Acute pulmonary embolism (PE) is a serious clinical disease characterized by a high mortality rate. The aim of this study was to assess the prognostic value of international normalized ratio (INR) in acute PE patients not on anticoagulant therapy.

METHODS

The study included 244 hospitalized acute PE patients who were not receiving previous anticoagulant therapy. Based on their 30-day mortality, patients were categorized as survivors or non-survivors. INR was measured during the patients' admission, on the same day as the diagnosis of PE but before anticoagulation started.

RESULTS

Thirty-day mortality occurred in 39 patients (16%). INR was higher in non-survivors than in survivors (1.3±0.4 vs 1.1±0.3, P=.003). In multivariate analysis, INR (HR: 3.303, 95% CI: 1.210-9.016, P=.020) was independently associated with 30-day mortality from PE. Inclusion of INR in a model with simplified pulmonary embolism severity index (sPESI) score improved the area under the receiver operating characteristics (ROC) curve from 0.736 (95% CI: 0.659-0.814) to 0.775 (95% CI: 0.701-0.849) (P=.028). Also, the addition of INR to sPESI score enhanced the net reclassification improvement (NRI=8.8%, P<.001) and integrated discrimination improvement (IDI=0.043, P=.027).

CONCLUSION

Elevated INR may have prognostic value for 30-day mortality in acute PE patients not on anticoagulation. Combining INR with sPESI score improved the predictive value for all-cause mortality. However, further large-scale studies are needed to confirm it's prognostic role.