A prognostic tool to predict outcomes in children undergoing the Norwood operation

Using machine learning to predict five-year transplant-free survival among infants with hypoplastic left heart syndrome

Hypoplastic left heart syndrome (HLHS) is a congenital malformation commonly treated with palliative surgery and is associated with significant morbidity and mortality. Risk stratification models have often relied upon traditional survival analyses or outcomes data failing to extend beyond infancy. Individualized prediction of transplant-free survival (TFS) employing machine learning (ML) based analyses of outcomes beyond infancy may provide further valuable insight for families and healthcare providers along the course of a staged palliation. Data from both the Pediatric Heart Network (PHN) Single Ventricle Reconstruction (SVR) trial and Extension study (SVR II), which extended cohort follow up for five years was used to develop ML-driven models predicting TFS. Models incrementally incorporated features corresponding to successive phases of care, from pre-Stage 1 palliation (S1P) through the stage 2 palliation (S2P) hospitalization. Models trained with features from Pre-S1P, S1P operation, and S1P hospitalization all demonstrated time-dependent area under the curves (td-AUC) beyond 0.70 through 5 years following S1P, with a model incorporating features through S1P hospitalization demonstrating particularly robust performance (td-AUC 0.838 (95% CI 0.836-0.840)). Machine learning may offer a clinically useful alternative means of providing individualized survival probability predictions, years following the staged surgical palliation of hypoplastic left heart syndrome.

Textbook outcome for the Norwood operation-an informative quality metric in congenital heart surgery

Objectives

To develop a more holistic measure of center performance than operative mortality, we created a composite "textbook outcome" for the Norwood operation using several postoperative end points. We hypothesized that achieving the textbook outcome would have a positive prognostic and financial impact.

Methods

This was a single-center retrospective study of primary Norwood operations from 2005 to 2021. Through interdisciplinary clinician consensus, textbook outcome was defined as freedom from operative mortality, open or catheter-based reintervention, 30-day readmission, extracorporeal membrane oxygenation, cardiac arrest, reintubation, length of stay >75%ile from Society of Thoracic Surgeons data report (66 days), and mechanical ventilation duration >75%ile (10 days). Multivariable logistic regression and Cox proportional hazards modeling were used to determine predictive factors for textbook outcome achievement and association of the outcome with long-term survival, respectively.

Results

Overall, 30% (58/196) of patients met the textbook outcome. Common reasons for failure to attain textbook outcome were prolonged ventilation (68/138, 49%) and reintubation (63/138, 46%). In multivariable analysis, greater weight (odds ratio [OR], 2.11; 95% confidence interval [CI], 1.17-3.95; P = .02) was associated with achieving the textbook outcome whereas preoperative shock (OR, 0.36; 95% CI, 0.13-0.87; P = .03) and longer bypass time (OR, 0.99; 95% CI, 0.98-1.00; P = .002) were negatively associated. Patients who met the outcome incurred fewer hospital costs ($152,430 [141,798-177,983] vs $269,070 [212,451-372,693], P < .001), and after adjusting for patient factors, achieving textbook outcome was independently associated with decreased risk of all-cause mortality (hazard ratio, 0.45; 95% CI, 0.22-0.89; P = .02).

Conclusions

Outcomes continue to improve within congenital heart surgery, making operative mortality a less-sensitive metric. The Norwood textbook outcome may represent a balanced measure of a successful episode of care.

Deep Learning for Improved Risk Prediction in Surgical Outcomes

The Norwood surgical procedure restores functional systemic circulation in neonatal patients with single ventricle congenital heart defects, but this complex procedure carries a high mortality rate. In this study we address the need to provide an accurate patient specific risk prediction for one-year postoperative mortality or cardiac transplantation and prolonged length of hospital stay with the purpose of assisting clinicians and patients' families in the preoperative decision making process. Currently available risk prediction models either do not provide patient specific risk factors or only predict in-hospital mortality rates. We apply machine learning models to predict and calculate individual patient risk for mortality and prolonged length of stay using the Pediatric Heart Network Single Ventricle Reconstruction trial dataset. We applied a Markov Chain Monte-Carlo simulation method to impute missing data and then fed the selected variables to multiple machine learning models. The individual risk of mortality or cardiac transplantation calculation produced by our deep neural network model demonstrated 89 ± 4% accuracy and 0.95 ± 0.02 area under the receiver operating characteristic curve (AUROC). The C-statistics results for prediction of prolonged length of stay were 85 ± 3% accuracy and AUROC 0.94 ± 0.04. These predictive models and calculator may help to inform clinical and organizational decision making.

Early prediction of critical events for infants with single-ventricle physiology in critical care using routinely collected data

OBJECTIVE

Critical events are common and difficult to predict among infants with congenital heart disease and are associated with mortality and long-term sequelae. We aimed to achieve early prediction of critical events, that is, cardiopulmonary resuscitation, emergency endotracheal intubation, and extracorporeal membrane oxygenation in infants with single-ventricle physiology before second-stage surgery. We hypothesized that naïve Bayesian models learned from expert knowledge and clinical data can predict critical events early and accurately.

METHODS

We collected 93 patients with single-ventricle physiology admitted to intensive care units in a single tertiary pediatric hospital between 2014 and 2017. Using knowledge elicited from experienced cardiac-intensive-care-unit providers and machine-learning techniques, we developed and evaluated the Cardiac-intensive-care Warning INdex (C-WIN) system, consisting of a set of naïve Bayesian models that leverage routinely collected data. We evaluated predictive performance using the area under the receiver operating characteristic curve, sensitivity, and specificity. We performed the evaluation at 5 different prediction horizons: 1, 2, 4, 6, and 8 hours before the onset of critical events.

RESULTS

The area under the receiver operating characteristic curves of the C-WIN models ranged between 0.73 and 0.88 at different prediction horizons. At 1 hour before critical events, C-WIN was able to detect events with an area under the receiver operating characteristic curve of 0.88 (95% confidence interval, 0.84-0.92) and a sensitivity of 84% at the 81% specificity level.

CONCLUSIONS

Predictive models may enhance clinicians' ability to identify infants with single-ventricle physiology at high risk of critical events. Early prediction of critical events may indicate the need to perform timely interventions, potentially reducing morbidity, mortality, and health care costs.

Prediction of One-Year Transplant-Free Survival after Norwood Procedure Based on the Pre-Operative Data

This paper discusses computational modeling of predictive risk factors for neonates undergoing a Norwood surgical procedure, a multi-stage cardiac procedure that restores functional systemic circulation in patients such as neonates with Hypoplastic Left Heart Syndrome (HLHS). In this model, we apply machine learning based binary classication to 549 cases reported by the Pediatric Heart Networks Single Ventricle Reconstruction Trial. We use neural networks classier to predict risk factors for individual patients undergoing a Norwood procedure for the repair of HLHS. Results indicate that independent risk can be calculated with 85% accuracy and 0.94 area under the receiver operating characteristics curve. This model may help physicians provide counseling for families and medically optimize patients prior to surgery by modifying individual risk factors.

Creation of an empiric tool to predict ECMO deployment in pediatric respiratory or cardiac failure

PURPOSE

To create a real-time prediction tool to predict probability of ECMO deployment in children with cardiac or pulmonary failure.

MATERIALS AND METHODS

Patients ≤18 years old admitted to an ICU that participated in the Virtual Pediatric Systems database (2009-2015) were included. Logistic regression models using adaptive lasso methodology were used to identify independent factors associated with ECMO use.

RESULTS

A total of 538,202 ICU patients from 140 ICUs qualified for inclusion. ECMO was deployed in 3484 patients (0.6%) with a mortality of 1450 patients (41.6%). The factors associated with increased probability of ECMO use included: younger age, pulmonary hypertension, congenital heart disease, high-complexity cardiac surgery, cardiomyopathy, acute lung injury, shock, renal failure, cardiac arrest, use of nitric oxide, use of either conventional mechanical ventilation or high frequency oscillatory ventilation, and higher annual ECMO center volume. The area under the receiver operating curve for this model was 0.90 (95% CI: 0.85-0.93). This tool can be accessed at https://soipredictiontool.shinyapps.io/ECMORisk/.

CONCLUSIONS

Here, we present a tool to predict ECMO deployment among critically ill children; this tool will help create real-time risk stratification among critically ill children, and it will help with benchmarking, family counseling, and research.

Intensive Care Mortality Prognostic Model for Pediatric Pulmonary Hypertension

OBJECTIVES

The disease burden and mortality of children with pulmonary hypertension are significantly higher than for the general PICU population. We aimed to develop a risk-adjustment tool predicting PICU mortality for pediatric pulmonary hypertension patients: the Pediatric Index of Pulmonary Hypertension Intensive Care Mortality score.

DESIGN

Retrospective analysis of prospectively collected multicenter pediatric critical care data.

SETTING

One-hundred forty-three centers submitting data to Virtual Pediatric Systems database between January 1, 2009, and December 31, 2015.

PATIENTS

Patients 21 years old or younger with a diagnosis of pulmonary hypertension.

INTERVENTIONS

Twenty-one demographic, diagnostic, and physiologic variables obtained within 12 hours of PICU admission were assessed for inclusion. Multivariable logistic regression with stepwise selection was performed to develop the final model. Receiver operating characteristic curves were used to compare the Pediatric Index of Pulmonary Hypertension Intensive Care Mortality score with Pediatric Risk of Mortality 3 and Pediatric Index of Mortality 2 scores.

MEASUREMENTS AND MAIN RESULTS

Fourteen-thousand two-hundred sixty-eight admissions with a diagnosis of pulmonary hypertension were included. Primary outcome was PICU mortality. Fourteen variables were selected for the final model: age, bradycardia, systolic hypotension, tachypnea, pH, FIO2, hemoglobin, blood urea nitrogen, creatinine, mechanical ventilation, nonelective admission, previous PICU admission, PICU admission due to nonsurgical cardiovascular disease, and cardiac arrest immediately prior to admission. The receiver operating characteristic curve for the Pediatric Index of Pulmonary Hypertension Intensive Care Mortality model (area under the curve = 0.77) performed significantly better than the receiver operating characteristic curves for Pediatric Risk of Mortality 3 (area under the curve = 0.71; p < 0.001) and Pediatric Index of Mortality 2 (area under the curve = 0.69; p < 0.001), respectively.

CONCLUSIONS

The Pediatric Index of Pulmonary Hypertension Intensive Care Mortality score is a parsimonious model that performs better than Pediatric Risk of Mortality 3 and Pediatric Index of Mortality 2 for mortality in a multicenter cohort of pediatric pulmonary hypertension patients admitted to PICUs. Application of the Pediatric Index of Pulmonary Hypertension Intensive Care Mortality model to pulmonary hypertension patients in the PICU might facilitate earlier identification of patients at high risk for mortality and improve the ability to prognosticate for patients and families.