Predicting birth weight with conditionally linear transformation models.

Area of Science:

• Perinatal Medicine
• Biostatistics
• Medical Imaging

Background:

• Accurate birth weight (BW) prediction is vital for managing neonatal morbidity and mortality risks.
• Current prediction formulas, often based on prenatal ultrasound, primarily provide point estimates and lack systematic uncertainty quantification.
• This limitation hinders the identification of fetuses with high prediction uncertainty.

Purpose of the Study:

• To introduce and evaluate conditionally linear transformation models (CLTMs) for predicting fetal birth weight.
• To model the entire conditional distribution of BW, enabling both point predictions and prediction intervals.
• To compare CLTMs against traditional linear regression and quantile regression methods.

Main Methods:

• Application of various conditionally linear transformation models (CLTMs) to a dataset of 8712 deliveries.
• Comparison of CLTM performance against standard linear regression and quantile regression techniques.
• Evaluation based on prediction accuracy, conditional coverage, and average length of prediction intervals.

Main Results:

• The best-performing CLTM variant demonstrated competitive results compared to quantile regression and linear regression.
• CLTMs successfully provide fetus-specific prediction intervals, offering a measure of prediction accuracy.
• CLTMs effectively account for heteroscedasticity, kurtosis, and skewness in birth weight distributions.

Conclusions:

• Conditionally linear transformation models (CLTMs) offer a robust approach for birth weight prediction.
• CLTMs enhance clinical practice by providing both accurate point predictions and interpretable prediction intervals.
• The ability of CLTMs to model complex distributional characteristics makes them a valuable tool in perinatal analytics.