Predicting Medical Emergency Team Calls, Cardiac Arrest Calls and Re-Admission after Intensive Care Discharge: Creation of a Tool to Identify At-Risk Patients

Abstract

We aimed to develop a predictive model for intensive care unit (ICU)-discharged patients at risk of post-ICU deterioration. We performed a retrospective, single-centre cohort observational study by linking the hospital admission, patient pathology, ICU, and medical emergency team (MET) databases. All patients discharged from the Alfred Hospital ICU to wards between July 2012 and June 2014 were included. The primary outcome was a composite endpoint of any MET call, cardiac arrest call or ICU re-admission. Multivariable logistic regression analysis was used to identify predictors of outcome and develop a risk-stratification model. Four thousand, six hundred and thirty-two patients were included in the study. Of these, 878 (19%) patients had a MET call, 51 (1.1%) patients had cardiac arrest calls, 304 (6.5%) were re-admitted to ICU during the same hospital stay, and 964 (21%) had MET calls, cardiac arrest calls or ICU re-admission. A discriminatory predictive model was developed (area under the receiver operating characteristic curve 0.72 [95% confidence intervals {CI} 0.70 to 0.73]) which identified the following factors: increasing age (odds ratio [OR] 1.012 [95% CI 1.007 to 1.017] P <0.001), ICU admission with subarachnoid haemorrhage (OR 2.26 [95% CI 1.22 to 4.16] P=0.009), admission to ICU from a ward (OR 1.67 [95% CI 1.31 to 2.13] P <0.001), Acute Physiology and Chronic Health Evaluation (APACHE) III score without the age component (OR 1.005 [95% CI 1.001 to 1.010] P=0.025), tracheostomy on ICU discharge (OR 4.32 [95% CI 2.9 to 6.42] P <0.001) and discharge to cardiothoracic (OR 2.43 [95%CI 1.49 to 3.96] P <0.001) or oncology wards (OR 2.27 [95% CI 1.05 to 4.89] P=0.036). Over the two-year period, 361 patients were identified as having a greater than 50% chance of having post-ICU deterioration. Factors are identifiable to predict patients at risk of post-ICU deterioration. This knowledge could be used to guide patient follow-up after ICU discharge, optimise healthcare resources, and improve patient outcomes and service delivery.

Value of laboratory results in addition to vital signs in a machine learning algorithm to predict in-hospital cardiac arrest: A single-center retrospective cohort study

“…Third, our model might have failed to learn from the laboratory results. However, laboratory results, which have been clinically and theoretically proven to be associated with IHCA [13,15,30], have a high feature importance in the model (Fig 4), which indicates they were successfully learned. In addition, the newer variables (i.e.…”

“…To support the necessity of preventive monitoring, prior studies found that clinical deterioration is common prior to cardiac arrest [4,5]. Based on these findings, various early warning scores have been developed, ranging from an analog model based on vital signs to a digital scoring system using both vital signs and laboratory results [6][7][8][9][10][11][12][13][14][15]. Recently, a variety of early warning scores have utilized machine learning to account for the non-linear relationships in various input variables [16][17][18].…”

“…Up to 80% of patients with IHCA have signs of deterioration in the eight hours before cardiac arrest [4,5], and various early warning scores based on vital signs have been developed [6][7][8][9][10][11]. The widespread implementation of electronic health records enables large datasets of laboratory results to be used in the development of early warning scores [12][13][14][15]. Recently, automated scores using machine learning models with and without laboratory results have been widely investigated, and both have achieved promising results [16][17][18].…”

Transition of care at discharge from the Intensive Care Unit: a scoping review

“…A recent study found that 21% of the discharged patients had post-ICU deterioration, including cardiac arrest, RRT calls, and readmission ( 63 ) . Patients undergoing lung transplantation and other thoracic surgery, as well as advanced age, increased severity of the disease estimated by the Acute Physiology and Chronic Health Evaluation III (APACHE III) score, bradycardia, abnormal levels of albumin in the admission to the ICU, hyperkalemia and high level of activated partial thromboplastin time (APTT) at discharge from the ICU, presented a higher risk regardless of deterioration.…”

Identifying predictive factors of rapid response/code blue: A retrospective study using logistic regression analysis

“…Predictors of patient deterioration included bradycardia, tachypnea, pyrexia, hypotension, low mean arterial pressure, and hyperkalemia 4,5 . Recognition of early deterioration has the potential to enhance patient safety, decrease mortality rates, identify patients that need to be closely assessed by a critical care nurse or medical doctor, and potentially save lives 6 .…”