Prediction Algorithm for ICU Mortality and Length of Stay Using Machine Learning

Abstract Background: Machine learning can predict outcomes and determine variables contributing to precise prediction, and can thus classify patients with different risk factors of outcomes. This study aimed to investigate the predictive accuracy for mortality and length of stay in intensive care unit (ICU) patients using machine learning, and to identify the variables contributing to the precise prediction or classification of patients.Methods: Patients (n=12,747) admitted to the ICU at Chiba University Hospital were randomly assigned to the training and test cohorts. After learning using the variables on admission in the training cohort, the area under the curve (AUC) was analyzed in the test cohort to evaluate the predictive accuracy of the supervised machine learning classifiers, including random forest (RF) for outcomes (primary outcome, mortality; secondary outcome, and length of ICU stay). The rank of the variables that contributed to the machine learning prediction was confirmed, and cluster analysis of the patients with risk factors of mortality was performed to identify the important variables associated with patient outcomes.Results: Machine learning using RF revealed a high predictive value for mortality, with an AUC of 0.945. In addition, RF showed high predictive value for short and long ICU stays, with AUCs of 0.881 and 0.889, respectively. Lactate dehydrogenase (LDH) was identified as a variable contributing to the precise prediction in machine learning for both mortality and length of ICU stay. LDH was also identified as a contributing variable to classify patients into sub-populations based on different risk factors of mortality.Conclusion: The machine learning algorithm could predict mortality and length of stay in ICU patients with high accuracy. LDH was identified as a contributing variable in mortality and length of ICU stay prediction and could be used to classify patients based on mortality risk.

Download Full-text

Provider-Documented Anxiety in the ICU: Prevalence, Risk Factors, and Associated Patient Outcomes

Journal of Intensive Care Medicine ◽

10.1177/0885066620956564 ◽

2020 ◽

pp. 088506662095656

Author(s):

Andrew D. May ◽

Ann M. Parker ◽

Ellen S. Caldwell ◽

Catherine L. Hough ◽

Jennifer E. Jutte ◽

...

Keyword(s):

Risk Factors ◽

Substance Abuse ◽

Length Of Stay ◽

Patient Outcomes ◽

Chart Review ◽

Severity Of Illness ◽

Surgical Intensive Care Unit ◽

Icu Patients ◽

Icu Stay ◽

History Of

Purpose: To determine the prevalence of provider-documented anxiety in critically ill patients, associated risk factors, and related patient outcomes. Method: Chart review of 100 randomly sampled, adult patients, with a length of stay ≥48 hours in a medical or trauma/surgical intensive care unit (ICU). Provider-documented anxiety was identified based on a comprehensive retrospective chart review of the ICU stay, searching for any acute episode of anxiety (e.g., documented words related to anxiety, panic, and/or distress). Results: Of 100 patients, 45% (95% confidence interval: 35%-55%) had documented anxiety, with similar prevalence in medical vs. trauma/surgical ICU. Patients with documented anxiety more frequently had history of anxiety (22% vs. 4%, p = .004) and substance abuse (40% vs. 22%, p = .048). In the ICU, they had greater severity of illness (median (IQR) Acute Physiology Score 16(13,21) vs. 13(8,19), p = .018), screened positive for delirium at least once during ICU stay, (62% vs. 31%, p = .002), benzodiazepines and antipsychotics use (87% vs. 58%, p = .002; 33% vs. 13%, p = .013, respectively), and mental health consultation (31% vs. 18%, p = .132). These patients also had longer ICU and hospital lengths of stay (6(4,11) vs. 4(3,6), p<.001 and 18(10,30) vs. 10(6,16) days, p<.001, respectively) and less frequent discharge back to home (27% vs. 44%, p = .079). Conclusions: Documented anxiety, occurring in almost half of ICU patients with length of stay ≥48 hours, was associated with a history of anxiety and/or substance abuse, and greater ICU severity of illness, delirium, psychiatric medications, and length of stay. Increased awareness along with more standardized protocols for assessment of anxiety in the ICU, as well as greater evaluation of non-pharmacological treatments for anxiety symptoms in the ICU are warranted.

Download Full-text

Outcome Predictive Value of Serum Ferritin in ICU Patients with Long ICU Stay

Medicina ◽

10.3390/medicina57010001 ◽

2020 ◽

Vol 57 (1) ◽

pp. 1

Author(s):

Daniel Rusu ◽

Mihaela Blaj ◽

Irina Ristescu ◽

Emilia Patrascanu ◽

Laura Gavril ◽

...

Keyword(s):

Length Of Stay ◽

Organ Dysfunction ◽

Serum Ferritin ◽

Sofa Score ◽

Predictive Value ◽

Serum Iron ◽

Predictive Accuracy ◽

University Hospital ◽

Icu Stay ◽

Icu Length Of Stay

Background and Objectives: The simplified interpretation of serum ferritin levels, according to which low ferritin levels indicate iron deficiency and high levels indicate hemochromatosis is obsolete, as in the presence of inflammation serum ferritin levels, no longer correlate with iron stores. However, further data are needed to interpret serum ferritin levels correctly in patients with ongoing inflammation. Our study aimed to assess serum iron and ferritin dynamics in patients with long ICU stay and the possible correlations with organ dysfunction progression and outcome. Materials and Methods: We conducted a prospective study in a university hospital intensive care unit (ICU) over six months. All patients with an ICU length-of-stay of more than seven days were enrolled. Collected data included: demographics, Sequential Organ Failure Assessment (SOFA) score, admission, weekly serum iron and ferritin levels, ICU length-of-stay and outcome. Interactions between organ dysfunction progression and serum iron and ferritin levels changes were investigated. Outcome predictive value of serum ferritin was assessed. Results: Seventy-two patients with a mean ICU length-of-stay of 15 (4.4) days were enrolled in the study. The average age of patients was 62 (16.8) years. There were no significant differences between survivors (39 patients, 54%) and nonsurvivors (33 patients, 46%) regarding demographics, serum iron and ferritin levels and SOFA score on ICU admission. Over time, serum iron levels remained normal or low, while serum ferritin levels statedly increased in all patients. Serum ferritin increase was higher in nonsurvivors than survivors. There was a significant positive correlation between SOFA score and serum ferritin (r = 0.7, 95%CI for r = 0.64 to 0.76, p < 0.01). The predictive outcome accuracy of serum ferritin was similar to the SOFA score. Conclusions: In patients with prolonged ICU stay, serum ferritin dynamics reflects organ dysfunction progression and parallels SOFA score in terms of outcome predictive accuracy.

Download Full-text

Supervised machine learning and logistic regression identifies novel epistatic risk factors with PTPN22 for rheumatoid arthritis

Genes and Immunity ◽

10.1038/gene.2009.110 ◽

2010 ◽

Vol 11 (3) ◽

pp. 199-208 ◽

Cited By ~ 19

Author(s):

F B S Briggs ◽

P P Ramsay ◽

E Madden ◽

J M Norris ◽

V M Holers ◽

...

Keyword(s):

Rheumatoid Arthritis ◽

Machine Learning ◽

Risk Factors ◽

Logistic Regression ◽

Supervised Machine Learning

Download Full-text

A proof-of-concept study applying machine learning methods to putative risk factors for eating disorders: results from the multi-centre European project on healthy eating

Psychological Medicine ◽

10.1017/s003329172100489x ◽

2021 ◽

pp. 1-10

Author(s):

I. Krug ◽

J. Linardon ◽

C. Greenwood ◽

G. Youssef ◽

J. Treasure ◽

...

Keyword(s):

Machine Learning ◽

Risk Factors ◽

Logistic Regression ◽

Predictive Accuracy ◽

Area Under The Curve ◽

Prediction Rule ◽

Predictive Performance ◽

Individual Risk ◽

European Project ◽

Wide Range

Abstract Background Despite a wide range of proposed risk factors and theoretical models, prediction of eating disorder (ED) onset remains poor. This study undertook the first comparison of two machine learning (ML) approaches [penalised logistic regression (LASSO), and prediction rule ensembles (PREs)] to conventional logistic regression (LR) models to enhance prediction of ED onset and differential ED diagnoses from a range of putative risk factors. Method Data were part of a European Project and comprised 1402 participants, 642 ED patients [52% with anorexia nervosa (AN) and 40% with bulimia nervosa (BN)] and 760 controls. The Cross-Cultural Risk Factor Questionnaire, which assesses retrospectively a range of sociocultural and psychological ED risk factors occurring before the age of 12 years (46 predictors in total), was used. Results All three statistical approaches had satisfactory model accuracy, with an average area under the curve (AUC) of 86% for predicting ED onset and 70% for predicting AN v. BN. Predictive performance was greatest for the two regression methods (LR and LASSO), although the PRE technique relied on fewer predictors with comparable accuracy. The individual risk factors differed depending on the outcome classification (EDs v. non-EDs and AN v. BN). Conclusions Even though the conventional LR performed comparably to the ML approaches in terms of predictive accuracy, the ML methods produced more parsimonious predictive models. ML approaches offer a viable way to modify screening practices for ED risk that balance accuracy against participant burden.

Download Full-text

Abstract 13811: Cognitive and Psychological Outcomes of Patients 6 Months After Cardiac Arrest or Intensive Care Stay

Circulation ◽

10.1161/circ.144.suppl_2.13811 ◽

2021 ◽

Vol 144 (Suppl_2) ◽

Author(s):

Matthew R Potter ◽

Marco Mion ◽

Eleni A Nikolopoulou ◽

Neil Magee ◽

Kelly Farrell ◽

...

Keyword(s):

Cardiac Arrest ◽

Cognitive Impairment ◽

Length Of Stay ◽

Assessment Tools ◽

Hospital Length Of Stay ◽

Post Discharge ◽

Icu Patients ◽

Icu Stay ◽

Sf 36 ◽

Significant Difference

Background: The neuropsychological and cognitive consequences of an ICU stay can cause a large burden on many patients. In this study, we assessed the outcomes of a group of patients attending a newly set up, multi-disciplinary outpatient clinic focused on assessing neuropsychological and cognitive outcomes following a significant (>72 hours) ICU stay, and compared patients whose ICU was post-OHCA (out of hospital cardiac arrest), and those non-OHCA. Methods: Between 2016 and 2019, 152 patients were assessed within the Care after REsuscitation (CARE) / ICU follow up clinic, 6 months following hospital discharge, using the SF-36, HADS, PTSS-14 and MoCA. The OHCA group were compared to other non-OHCA, ICU patients (>72 hour stay). Results: No significant differences (p<0.05) were found between the groups outcomes, however we found that 6 months post-discharge, the non-OHCA group experience significantly more pain, are older, and required a longer length of stay (p<0.05). However, we found compared to normative data of the SF-36, over half of the OHCA group (on 6 out of the 8 subscales) and the non-OHCA (on 7 out of the 8) were below population norms, especially Role-Physical (66.7% OHCA and 71.6% non-OHCA) and Energy/ fatigue (66.7% OHCA and 61.4% non-OHCA). Anxiety was observed in 33.3% of the OHCA group, and 35.2% of the non-OHCA group had an abnormal total HADS score. PTSD was seen in 12.7% of the OHCA group and 10.2% of the non-OHCA group. Cognitive impairment was observed in 61.9% of OHCA and 59.1% of non-OHCA patients. Age and hospital length of stay had no significant effect on outcomes on our OHCA population, however females had significantly worse health related quality of life (HRQoL) on 6 out of the 8 subscales (p<0.05). Conclusion: Despite the OHCA and non-OHCA groups having no significant difference between their outcomes, there is a great disease burden upon many individuals following ICU stay, with many experiencing poor HRQoL, mood disorders, PTSD and cognitive impairment. The factors contributing to poor outcome following both ICU related illness and cardiac arrest should be further studied. The creation and validation of new assessment tools is imperative to ensure we fully appreciate the extent of the morbidity in this group to improve care for all ICU patients.

Download Full-text

Attributable Cost and Length of Stay for Patients With Central Venous Catheter—Associated Bloodstream Infection in Mexico City Intensive Care Units A Prospective, Matched Analysis

Infection Control and Hospital Epidemiology ◽

10.1086/510812 ◽

2007 ◽

Vol 28 (1) ◽

pp. 31-35 ◽

Cited By ~ 73

Author(s):

Francisco Higuera ◽

Manuel Sigfrido Rangel-Frausto ◽

Victor Daniel Rosenthal ◽

Jose Martinez Soto ◽

Jorge Castañon ◽

...

Keyword(s):

Intensive Care ◽

Length Of Stay ◽

Central Venous Catheter ◽

Bloodstream Infection ◽

Mexico City ◽

Venous Catheter ◽

Central Venous ◽

Icu Stay ◽

The Mean ◽

Length Of Icu Stay

Background.No information is available about the financial impact of central venous catheter (CVC)-associated bloodstream infection (BSI) in Mexico.Objective.To calculate the costs associated with BSI in intensive care units (ICUs) in Mexico City.Design.An 18-month (June 2002 through November 2003), prospective, nested case-control study of patients with and patients without BSI.Setting.Adult ICUs in 3 hospitals in Mexico City.Patients and Methods.A total of 55 patients with BSI (case patients) and 55 patients without BSI (control patients) were compared with respect to hospital, type of ICU, year of hospital admission, length of ICU stay, sex, age, and mean severity of illness score. Information about the length of ICU stay was obtained prospectively during daily rounds. The daily cost of ICU stay was provided by the finance department of each hospital. The cost of antibiotics prescribed for BSI was provided by the hospitals' pharmacy departments.Results.For case patients, the mean extra length of stay was 6.1 days, the mean extra cost of antibiotics was $598, the mean extra hospital cost was $11,591, and the attributable extra mortality was 20%.Conclusions.In this study, the duration of ICU stay for patients with central venous catheter-associated BSI was significantly longer than that for control patients, resulting in increased healthcare costs and a higher attributable mortality. These conclusions support the need to implement preventive measures for hospitalized patients with central venous catheters in Mexico.

Download Full-text

Clinical outcomes of gastroscopy in critically ill patients using high-dose proton pump inhibitor for suspected bleeding

10.21203/rs.3.rs-106503/v1 ◽

2020 ◽

Author(s):

Won Gun Kwack

Keyword(s):

Retrospective Study ◽

Proton Pump Inhibitor ◽

Proton Pump ◽

Hemoglobin Level ◽

High Dose ◽

Gi Bleeding ◽

Icu Mortality ◽

Icu Patients ◽

Icu Stay ◽

Length Of Icu Stay

Abstract Background: Gastroscopy is a useful procedure for gastrointestinal (GI) bleeding. No definite clinical guidelines recommend on the choice of gastroscopy implementation in the intensive care unit (ICU) patient with suspected GI bleeding. The objective of this retrospective study was to compare the clinical effectiveness of gastroscopy in critically ill patients using high-dose proton pump inhibitor for suspected bleeding.Methods: ICU patients using a high-does proton pump inhibitor for suspected GI bleeding from January 2015 to February 2020 were retrospectively included. Massive GI bleeding, such as hematemesis and hematochezia, were excluded. After propensity score matching (PSM) between the gastroscopy and no gastroscopy groups, the change in hemoglobin level, requirement of RBC transfusion, length of ICU stay, and ICU mortality were compared. Results: Of the 116 subjects included, 34 patients had gastroscopy during ICU stay. Among the gastroscopy group, 13 (38.2%) patients showed normal findings, and the most frequent abnormal finding was gastric ulcer (n = 9, 26.5%), and 12 patients (35.3%) had a hemostatic procedure. After PSM, the gastroscopy group needed more red blood cell transfusion than the no-gastroscopy group (P = 0.01). There was no significant difference in the change in hemoglobin level (P = 0.10), length of ICU stay (P = 0.64), and ICU mortality (P = 0.55).Conclusion: This retrospective study showed that gastroscopy had no definite clinical benefit in ICU patients using high-dose proton pump inhibitor for suspected GI bleeding.

Download Full-text

Development and Validation of Unplanned Extubation Prediction Models Using Intensive Care Unit Data: Comparative Machine Learning Study (Preprint)

10.2196/preprints.23508 ◽

2020 ◽

Author(s):

Sujeong Hur ◽

Ji Young Min ◽

Junsang Yoo ◽

Kyunga Kim ◽

Chi Ryang Chung ◽

...

Keyword(s):

Machine Learning ◽

Intensive Care Unit ◽

Patient Safety ◽

Intensive Care ◽

Predictive Value ◽

Prediction Models ◽

Support Vector ◽

Unplanned Extubation ◽

Electronic Health Record Data ◽

Icu Patients

BACKGROUND Patient safety in the intensive care unit (ICU) is one of the most critical issues, and unplanned extubation (UE) is considered as the most adverse event for patient safety. Prevention and early detection of such an event is an essential but difficult component of quality care. OBJECTIVE This study aimed to develop and validate prediction models for UE in ICU patients using machine learning. METHODS This study was conducted an academic tertiary hospital in Seoul. The hospital had approximately 2,000 inpatient beds and 120 intensive care unit (ICU) beds. The number of patients, on daily basis, was approximately 9,000 for the out-patient. The number of annual ICU admission was approximately 10,000. We conducted a retrospective study between January 1, 2010 and December 31, 2018. A total of 6,914 extubation cases were included. We developed an unplanned extubation prediction model using machine learning algorithms, which included random forest (RF), logistic regression (LR), artificial neural network (ANN), and support vector machine (SVM). For evaluating the model’s performance, we used area under the receiver operator characteristic curve (AUROC). Sensitivity, specificity, positive predictive value negative predictive value, and F1-score were also determined for each model. For performance evaluation, we also used calibration curve, the Brier score, and the Hosmer-Lemeshow goodness-of-fit statistic. RESULTS Among the 6,914 extubation cases, 248 underwent UE. In the UE group, there were more males than females, higher use of physical restraints, and fewer surgeries. The incidence of UE was more likely to occur during the night shift compared to the planned extubation group. The rate of reintubation within 24 hours and hospital mortality was higher in the UE group. The UE prediction algorithm was developed, and the AUROC for RF was 0.787, for LR was 0.762, for ANN was 0.762, and for SVM was 0.740. CONCLUSIONS We successfully developed and validated machine learning-based prediction models to predict UE in ICU patients using electronic health record data. The best AUROC was 0.787, which was obtained using RF. CLINICALTRIAL N/A

Download Full-text

A Statistical Analysis of Risk Factors and Biological Behavior in Canine Mammary Tumors: A Multicenter Study

Animals ◽

10.3390/ani10091687 ◽

2020 ◽

Vol 10 (9) ◽

pp. 1687

Author(s):

Giovanni P. Burrai ◽

Andrea Gabrieli ◽

Valentina Moccia ◽

Valentina Zappulli ◽

Ilaria Porcellato ◽

...

Keyword(s):

Machine Learning ◽

Risk Factors ◽

Statistical Analysis ◽

Multicenter Study ◽

Malignant Tumors ◽

Mammary Tumors ◽

Supervised Machine Learning ◽

Biological Behavior ◽

Clinical Staging ◽

Canine Mammary Tumors

Canine mammary tumors (CMTs) represent a serious issue in worldwide veterinary practice and several risk factors are variably implicated in the biology of CMTs. The present study examines the relationship between risk factors and histological diagnosis of a large CMT dataset from three academic institutions by classical statistical analysis and supervised machine learning methods. Epidemiological, clinical, and histopathological data of 1866 CMTs were included. Dogs with malignant tumors were significantly older than dogs with benign tumors (9.6 versus 8.7 years, p < 0.001). Malignant tumors were significantly larger than benign counterparts (2.69 versus 1.7 cm, p < 0.001). Interestingly, 18% of malignant tumors were smaller than 1 cm in diameter, providing compelling evidence that the size of the tumor should be reconsidered during the assessment of the TNM-WHO clinical staging. The application of the logistic regression and the machine learning model identified the age and the tumor’s size as the best predictors with an overall diagnostic accuracy of 0.63, suggesting that these risk factors are sufficient but not exhaustive indicators of the malignancy of CMTs. This multicenter study increases the general knowledge of the main epidemiologica-clinical risk factors involved in the onset of CMTs and paves the way for further investigations of these factors in association with CMTs and in the application of machine learning technology.

Download Full-text

Epidemiology and prognosis of anti-infective therapy in the ICU setting during acute pancreatitis: a cohort study

Critical Care ◽

10.1186/s13054-019-2681-5 ◽

2019 ◽

Vol 23 (1) ◽

Cited By ~ 3

Author(s):

Philippe Montravers ◽

Elie Kantor ◽

Jean-Michel Constantin ◽

Jean-Yves Lefrant ◽

Thomas Lescot ◽

...

Keyword(s):

Acute Pancreatitis ◽

Hospital Mortality ◽

Therapeutic Interventions ◽

Therapeutic Management ◽

Pulmonary Infections ◽

Prescribing Practices ◽

Icu Patients ◽

Icu Stay ◽

Large Heterogeneity ◽

Length Of Icu Stay

Abstract Background Recent international guidelines for acute pancreatitis (AP) recommend limiting anti-infective therapy (AIT) to cases of suspected necrotizing AP or nosocomial extrapancreatic infection. Limited data are available concerning empirical and documented AIT prescribing practices in patients admitted to the intensive care unit (ICU) for the management of AP. Methods Using a multicentre, retrospective (2009–2014), observational database of ICU patients admitted for AP, our primary objective was to assess the incidence of AIT prescribing practices during the first 30 days following admission. Secondary objectives were to assess the independent impact of centre characteristics on the incidence of AIT and to identify factors associated with crude hospital mortality in a logistic regression model. Results In this cohort of 860 patients, 359 (42%) received AIT on admission. Before day 30, 340/359 (95%) AIT patients and 226/501 (45%) AIT-free patients on admission received additional AIT, mainly for intra-abdominal and lung infections. A large heterogeneity was observed between centres in terms of the incidence of infections, therapeutic management including AIT and prognosis. Administration of AIT on admission or until day 30 was not associated with an increased mortality rate. Patients receiving AIT on admission had increased rates of complications (septic shock, intra-abdominal and pulmonary infections), therapeutic (surgical, percutaneous, endoscopic) interventions and increased length of ICU stay compared to AIT-free patients. Patients receiving delayed AIT after admission and until day 30 had increased rates of complications (respiratory distress syndrome, intra-abdominal and pulmonary infections), therapeutic interventions and increased length of ICU stay compared to those receiving AIT on admission. Risk factors for hospital mortality assessed on admission were age (adjusted odds ratio [95% confidence interval] 1.03 [1.02–1.05]; p < 0.0001), Balthazar score E (2.26 [1.43–3.56]; p < 0.0001), oliguria/anuria (2.18 [1.82–4.33]; p < 0.0001), vasoactive support (2.83 [1.73–4.62]; p < 0.0001) and mechanical ventilation (1.90 [1.15–3.14]; p = 0.011), but not AIT (0.63 [0.40–1.01]; p = 0.057). Conclusions High proportions of ICU patients admitted for AP receive AIT, both on admission and during their ICU stay. A large heterogeneity was observed between centres in terms of incidence of infections, AIT prescribing practices, therapeutic management and outcome. AIT reflects the initial severity and complications of AP, but is not a risk factor for death.

Download Full-text