A method for estimating neck circumference using ubiquitous health variables for sleep apnea screening

Obstructive sleep apnea (OSA) is a prevalent condition that frequently goes undiagnosed. NoSAS Score provides a straightforward approach for rapid and accurate OSA screening, although it requires measuring neck circumference (NC). We aimed to develop a linear regression model to estimate NC using NoSAS variables. 3,109 subjects from the Sleep Health Heart Study database were included (56% man; 62 ± 10 years; body mass index [BMI] = 28.6 ± 4.8 kg/m2; NC = 38.5 ± 4.8 cm; apnea-hypopnea index = 14.5). Four multiple linear regression models were developed: 1) Age, BMI, sex, snoring; 2) Age, height, weight, sex, snoring; 3) Sex, BMI; 4) Sex, weight. Models 3 and 4 included the most significant variables of models 1 and 2, identified through permutation feature importance. For fitting and evaluating the models, participants were allocated into training/validation (80%) and testing (20%) sets. Model 3 (mean absolute error = 1.72 cm, mean absolute percentage error = 4.49%, and R2 = 0.72) was selected for further evaluation due to its simplicity, resulting in the following formula for estimated NC: eNC = 22.9 + 0.438×BMI + 5.435×Male. Using eNC instead of true NC did not reduce the performance of NoSAS in identifying individuals at high risk of OSA. The presented model for NC estimation can be utilized to assess OSA risk when NC measurement is unavailable.