Congenital Heart Disease
Conditions
Keywords
Machine Learning Algorithm, Pulse Oximetry
Brief summary
The purpose of this study is to implement and externally validate an inpatient ML algorithm that combines pulse oximetry features for critical congenital heart disease (CCHD) screening.
Detailed description
The study will externally validate an algorithm that combines non-invasive oxygenation and perfusion measurements as a screening tool for CCHD. In a previous study, the investigators created an algorithm that combines non-invasive measurements of oxygenation and perfusion over at least two measurements using machine learning (ML) techniques. The prior model was created and tested using internal validation (k-fold validation). Thus, the investigators will test the model on an external sample of patients to test generalizability of the model. Additionally, the team will trial a repeated measurement for any "failure" of the screen to assess impact on the false positive rate. Study team will also use repeated pulse oximetry measurements (up to 4 total and including measurements after 48 hours of age, which may be done outpatient) to create a new algorithm that incorporates new data over time. The central hypothesis is that the addition of non-invasive perfusion measurements will be superior to SpO2-alone screening for CCHD detection and a model that incorporates repeated measurements will enhance detection of CCHD while preserving the specificity.
Interventions
DIAGNOSTIC_TESTSpO2/PIx Measurement and ML Algorithm
Right upper and any lower extremity oxygen saturation (SpO2) and perfusion index (PIx) will be measured and an online ML inference model will be used to classify a newborn as healthy versus CCHD as new pulse oximetry data is collected.
Sponsors
University of California, Davis
National Institutes of Health (NIH)
Study design
Allocation
NA
Intervention model
SINGLE_GROUP
Primary purpose
DIAGNOSTIC
Masking
NONE
Intervention model description
Non-invasive measurements of oxygenation and perfusion will be measured with pulse oximeters and a machine learning algorithm to improve sensitivity of CCHD screening.
Eligibility
Sex/Gender
ALL
Age
0 Years to 21 Days
Healthy volunteers
Yes
Inclusion criteria
* Age \< 22 days * Fetuses suspected to have congenital heart disease * Newborns with suspected/confirmed critical congenital heart disease * Asymptomatic newborn undergoing SpO2 screening for CCHD
Exclusion criteria
* Echocardiogram completed prior to enrollment as the newborn would then no longer be considered "asymptomatic undergoing SpO2 screening for CCHD" * For Newborns with confirmed/suspected congenital heart disease (CHD): a) Patent ductus arteriosus and/or atrial septal defect/patent foramen ovale without other defects, b) Corrective cardiac surgical or catheter intervention performed before enrollment or c) Current infusions of vasoactive medications other than prostaglandin therapy.
Design outcomes
Primary
| Measure | Time frame | Description |
|---|---|---|
| Area under the curve for receiver operating characteristics for critical congenital heart disease using ML inpatient algorithm. | Through study completion, an average of 4 years | Receiver operating characteristics reflect a combination of sensitivity and specificity of a test. The investigators will identify the true positive and true negative rates for CCHD by confirming health status to a minimum of 2 months of age. The investigators will also utilize birth defect and death registries for missing infants. |
Secondary
| Measure | Time frame | Description |
|---|---|---|
| Sensitivity for critical congenital heart disease using ML inpatient algorithm (0-24 hours and 24-48 hours) | Through study completion, an average of 4 years | The investigators will identify the true positive rate for CCHD by confirming health status to a minimum of 2 months of age. CCHD will be defined based on echocardiogram or parent report if echocardiogram not present. The investigators will also utilize birth defect and death registries for missing infants. |
| Specificity for critical congenital heart disease using ML inpatient algorithm (0-24 hours and 24-48 hours) | Through study completion, an average of 4 years | The investigators will identify the true negative rate by confirming health status to a minimum of 2 months of age. CCHD will be defined based on echocardiogram or parent report if echocardiogram not present. The investigators will also utilize birth defect and death registries for missing infants. |
| Area under the curve for receiver operating characteristics for critical congenital heart disease using dynamic ML algorithm | Through study completion, an average of 4 years | Receiver operating characteristics reflect a combination of sensitivity and specificity of a test. The investigators will identify the true positive and true negative rates for CCHD by confirming health status to a minimum of 2 months of age. The investigators will also utilize birth defect and death registries for missing infants. |
| Sensitivity for critical congenital heart disease using dynamic ML algorithm | Through study completion, an average of 4 years | The investigators will identify the true positive rate for CCHD by confirming health status to a minimum of 2 months of age. CCHD will be defined based on echocardiogram or parent report if echocardiogram not present. The investigators will also utilize birth defect and death registries for missing infants. |
| Specificity for critical congenital heart disease using dynamic ML model | Through study completion, an average of 4 years | The investigators will identify the true negative rate by confirming health status to a minimum of 2 months of age. CCHD will be defined based on echocardiogram or parent report if echocardiogram not present. The investigators will also utilize birth defect and death registries for missing infants. |
| Sensitivity for critical coarctation of the aorta using dynamic ML algorithm | Through study completion, an average of 4 years | Critical coarctation of the aorta is the most commonly missed CCHD. The investigators will identify the true positive rate by confirming health status to a minimum of 2 months of age. The investigators will also utilize birth defect and death registries for missing infants. |
Countries
United States
Contacts
CONTACTHeather Siefkes, MD, MSCI
CONTACTElva Horath, IMG
Outcome results
None listed