Construction and validation of a predictive model for sleep disorders among pregnant women

Sleep disorders are among the most common major problems during pregnancy. Most studies on sleep disorders of pregnant women are closely related to adverse birth outcomes. In this study, our aim was to develop and validate a predictive model for the risk of sleep disorders in pregnant women.

A total of 2,467 healthy pregnant women were enrolled and randomly partitioned into a training set and a validation set at a ratio of 7:3. During the variable selection stage, the Pearson's chi-square test was employed to identify variables with a p-value below 0.05, which were then designated as candidate variables for subsequent logistic regression analysis. Concurrently, the LASSO regression technique was utilized to sift through and isolate the most valuable variables. Ultimately, we developed binary Logistic regression models predicated on the Pearson's chi-square test (Model 1) and the LASSO regression (Model 2). The performance of the nomograms was evaluated using the Bootstrap resampling procedure, the sensitivity and specificity of the receiver-operating characteristic (ROC), the area under the ROC curve (AUC), and decision curve analysis (DCA).

A total of 439 (25.4%) pregnant women in the training set and 208 (28.1%) in the validation set exhibited sleep disorder, respectively. The prediction models shared 6 risk factors (age, anxiety, depression, family functions, degree of pregnancy reaction, pre-pregnancy physical condition). In the Model 1, the sensitivity was 69.4%, and specificity was 59.6%. When pregnancy weeks, residence, only child were included in Model 2, the sensitivity was 82.4% and specificity was 54.8%. In the validation set, the areas under the curve of the Model 1 and Model 2 were 0.678 (0.635, 0.720), and 0.719 (0.678, 0.761), respectively. The risk prediction model of sleep disorders in pregnant women showed that the calibration curve is approximately distributed along the reference line. Decision curve analyses demonstrated a favorable net benefit within the range of the threshold probability in the nomograms.

Model 2 exhibited superior performance, can serve as a convenient and reliable tool for predicting the risk probability of sleep disorders in pregnant women.