Respiratory distress syndrome prediction at birth by optical skin maturity assessment and machine learning models for limited-resource settings: a development and validation study

Abstract

Background: A handheld optical device was developed to evaluate a newborn's skin maturity by assessing the photobiological properties of the tissue and processing it with other variables to predict early neonatal prognosis related to prematurity. This study assessed the device's ability to predict respiratory distress syndrome (RDS).

Methods: To assess the device's utility we enrolled newborns at childbirth in six urban perinatal centers from two multicenter single-blinded clinical trials. All newborns had inpatient follow-up until 72 h of life. We trained supervised machine learning models with data from 780 newborns in a Brazilian trial and provided external validation with data from 305 low-birth-weight newborns from another trial that assessed Brazilian and Mozambican newborns. The index test measured skin optical reflection with an optical sensor and adjusted acquired values with clinical variables such as birth weight and prenatal corticoid exposition for lung maturity, maternal diabetes, and hypertensive disturbances. The performance of the models was evaluated using intrasample k-parts cross-validation and external validation in an independent sample.

Results: Models adjusting three predictors (skin reflection, birth weight, and antenatal corticoid exposure) or five predictors had a similar performance, including or not maternal diabetes and hypertensive diseases. The best global accuracy was 89.7 (95% CI: 87.4 to 91.8, with a high sensitivity of 85.6% (80.2 to 90.0) and specificity of 91.3% (95% CI: 88.7 to 93.5). The test correctly discriminated RDS newborns in external validation, with 82.3% (95% CI: 77.5 to 86.4) accuracy. Our findings demonstrate a new way to assess a newborn's lung maturity, providing potential opportunities for earlier and more effective care.

Trial registration: RBR-3f5bm5 (online access: http://www.ensaiosclinicos.gov.br/rg/RBR-3f5bm5/), and RBR-33mjf (online access: https://ensaiosclinicos.gov.br/rg/RBR-33rnjf/).

Keywords: childbirth; equipment and supplies; machine learning; medical device; newborn; prematurity; respiratory distress syndrome; skin physiological phenomena.

Figure 1

Database, birth scenarios, and index test (outcomes). LBW, low birth weight; RDS, respiratory distress syndrome.

Figure 2

Steps of the testing process. (1) The device touches the skin over the sole of a newborn. (2) The sensor acquires skin maturity by assessing the photobiological properties of the tissue when measuring the reflection portions of the light beam incident on the skin. (3) The user inputs clinical data. (4) The data processor uses machine learning algorithms to predict respiratory distress syndrome.

Figure 3

Flowchart of participants using STARD diagram, according to development and model validation birth scenarios.

Figure 4

Attribute importance given by XGBoost when considering information gain that a variable brings when inserted into the model. (A) Model 1: trained with skin reflection + birth weight + Antenatal corticosteroid therapy for lung maturation, for the binary outcome RDS vs. non-RDS. (B) Model 2: trained with Skin reflection + birth weight + Antenatal corticosteroid therapy for lung maturation + diabetes + hypertensive diseases for the binary outcome RDS vs. non-RDS.

Figure 5

Confusion matrix for Respiratory Distress Syndrome prediction until 72 hours of life, according to gestational age at birth, using a three-variable-mode. (A) Incorrect prediction in birth scenario 1 - Cross-validation (n = 780). (B) Incorrect prediction in birth scenario 2, LBW - External validation (n = 305). (C) Correct prediction in birth scenario 1 - Cross-validation (n = 780). (D) Correct prediction in birth scenario 2, LBW - External validation (n = 305).