Development and validation of 'Patient Optimizer' (POP) algorithms for predicting surgical risk with machine learning.

BACKGROUND: Pre-operative risk assessment can help clinicians prepare patients for surgery, reducing the risk of perioperative complications, length of hospital stay, readmission and mortality. Further, it can facilitate collaborative decision-making and operational planning. OBJECTIVE: To develop effective pre-operative risk assessment algorithms (referred to as Patient Optimizer or POP) using Machine Learning (ML) that predict the development of post-operative complications and provide pilot data to inform the design of a larger prospective study. METHODS: After institutional ethics approval, we developed a base model that encapsulates the standard manual approach of combining patient-risk and procedure-risk. In an automated process, additional variables were included and tested with 10-fold cross-validation, and the best performing features were selected. The models were evaluated and confidence intervals calculated using bootstrapping. Clinical expertise was used to restrict the cardinality of categorical variables (e.g. pathology results) by including the most clinically relevant values. The models were created with logistic regression (LR) and extreme gradient-boosted trees using XGBoost (Chen and Guestrin, 2016). We evaluated performance using the area under the receiver operating characteristic curve (AUROC) and the area under the precision-recall curve (AUPRC). Data was obtained from a metropolitan university teaching hospital from January 2015 to July 2020. Data collection was restricted to adult patients undergoing elective surgery. RESULTS: A total of 11,475 adult admissions were included. The performance of XGBoost and LR was very similar across endpoints and metrics. For predicting the risk of any post-operative complication, kidney failure and length-of-stay (LOS), POP with XGBoost achieved an AUROC (95%CI) of 0.755 (0.744, 0.767), 0.869 (0.846, 0.891) and 0.841 (0.833, 0.847) respectively and AUPRC of 0.651 (0.632, 0.669), 0.336 (0.282, 0.390) and 0.741 (0.729, 0.753) respectively. For 30-day readmission and in-patient mortality, POP with XGBoost achieved an AUROC (95%CI) of 0.610 (0.587, 0.635) and 0.866 (0.777, 0.943) respectively and AUPRC of 0.116 (0.104, 0.132) and 0.031 (0.015, 0.072) respectively. CONCLUSION: The POP algorithms effectively predicted any post-operative complication, kidney failure and LOS in the sample population. A larger study is justified to improve the algorithm to better predict complications and length of hospital stay. A larger dataset may also improve the prediction of additional specific complications, readmission and mortality.

Study protocol for PREPARE: a phase II feasibility/safety randomised controlled trial on PeRiopErative Penicillin AlleRgy TEsting.

INTRODUCTION: Patient-reported antibiotic allergy labels (AALs) are common. These labels have been demonstrated to have a negative impact on use of appropriate antibiotics and patient-related health outcomes. These patients are more likely to receive suboptimal antibiotics, have increased rates of surgical site infections and are more likely to be colonised with multidrug-resistant organisms. Increasing recognition that antibiotic allergy forms a key part of good antimicrobial stewardship has led to calls for greater access to antibiotic allergy assessment.PREPARE is a pilot randomised controlled trial of beta-lactam allergy assessment and point of care delabelling in perioperative patients utilising a validated antibiotic allergy assessment tool that has been repurposed into a smartphone application. The aim of the study is to assess the feasibility and safety of this approach in the perioperative outpatient setting. METHODS AND ANALYSIS: Adult participants requiring elective surgery and are likely to require prophylactic intravenous antibiotics will be recruited. During the intervention phase, participants will be randomised to the intervention or control arm, with control patients receiving usual standard of care. Those randomised to intervention undertake a risk assessment via the smartphone application, with those deemed low risk proceeding to direct oral provocation with either a penicillin or cephalosporin. Study outcomes will be evaluated in the postintervention phase, 30 and 90 days after surgery.Feasibility of intervention delivery and recruitment will be reported as proportions with respective 95% CIs. Participants who experience an antibiotic adverse event will be reported by group with respective 95% CIs and compared using modified Poisson regression model with robust SE estimation. ETHICS AND DISSEMINATION: This protocol has received approval from the Austin Health human research and ethics committee, Heidelberg, Victoria, Australia (HREC/17/Austin/575). Results will be disseminated via publication in peer-reviewed journals as well as presentation at international conferences. TRIAL REGISTRATION NUMBER: ACTRN12620001295932.

Comparison of Cardiac Output of Both 2-Dimensional and 3-Dimensional Transesophageal Echocardiography With Transpulmonary Thermodilution During Cardiac Surgery.

OBJECTIVES: To compare agreement and variability of cardiac output measurement of 2-dimensional (2D) and 3D transesophageal echocardiography (TEE) with thermodilution before and after bypass. DESIGN: Prospective observational study. SETTING: Two tertiary hospitals. INTERVENTIONS: Cardiac output (CO) was measured simultaneously with thermodilution and TEE by multiplying either the left ventricular outflow tract area (LVOTA) or aortic valve area (AVA), the velocity-time integral (VTI) of flow at the same site, and heart rate. The LVOTA was calculated using diameter for 2D TEE. Planimetry was used for 3D TEE. The AVA was measured using planimetry. PARTICIPANTS: The study comprised 82 adult patients undergoing coronary or valve surgery. MEASUREMENTS AND MAIN RESULTS: One hundred fifty-four complete sets of measurements were obtained (82 prebypass and 72 postbypass). All TEE methods had acceptable correlation and absence of proportional or fixed bias except for the left ventricular outflow tract (LVOT) VTI modal trace method, which had poor correlation and proportional but not fixed bias (regression coefficient [95% confidence interval], bias [percentage of mean CO]): 2D LVOT VTI modal trace 0.67 (0.54-0.80), -36.4%; 2D LVOT VTI outer edge trace 0.96 (0.80-1.12), -15.3%; 2D AVA planimetry 0.96 (0.75-1.18), +4.9%; 3D LVOT area planimetry 1.18 (0.96-1.41), +0.8%; 3D AVA planimetry 1.20 (0.93-1.46), +0.4%. All TEE methods had wide levels of agreement compared with thermodilution (-3.94 to +0.23 L/min, -2.83 to +1.28 L/min, -2.23 to +2.73 L/min, -2.35 to +2.42 L/min, and -2.57 to +2.61 L/min, respectively). Measurement variability was superior for all TEE methods compared with thermodilution before but not after bypass. CONCLUSIONS: Although limits of agreement of CO measurement with 3D TEE and thermodilution are wide, 2D planimetry of the AVA and continuous wave Doppler may be substituted for thermodilution before and after bypass.

Tension pneumothorax at anaesthetic induction in an ex-premature infant with bronchopulmonary dysplasia.

ME, a 12-week-old premature infant with a history of complicated ventilatory support at birth, was referred to as a large tertiary centre for urgent retinal laser photocoagulation for retinopathy of prematurity. Following routine induction by a consultant paediatric anaesthetist, immediate ventilation difficulties were encountered, associated with a distended abdomen. Rapid assessment and escalation of the situation occurred, including assistance from several consultants. ME was diagnosed as having a pneumothorax. Rapid intervention, including needle decompression and insertion of a chest drain stabilised the patient. The procedure was cancelled, and the patient returned to intensive care. Over the next few days the patient gradually improved, and was gradually stepped down. The patient continued to improve, successfully overcoming the pneumothorax, pneumonia and insertion of a ventriculoperitoneal shunt, and eventually was discharged back to the referring hospital.