Abstract

Objectives

This study aims to design a predictive model for extension of length of stay (LOS) in very preterm infants (VPIs), for risk management and assisted decision making in the early postnatal period.

Methods

VPIs in the development cohort were randomly divided into training and testing sets in a 7:3 ratio. A total of 5 machine learning algorithms were used to construct and evaluate the model. LOS extension was defined as exceeding the 75th percentile of total hospitalization days for different gestational age groups.

Results

This study included a total of 1044 VPIs in the development cohort, and 23.9% (n = 250) were classified as having LOS extension. Seven variables were screened and selected to construct the prediction model based on the best algorithm, logistic regression (LR). In the internal validation, compared with other algorithms, the LR algorithm achieved the highest area under the curve (AUC) of 0.773 (95% CI 0.717-0.830). The accuracy was 0.729, specificity was 0.782, recall was 0.566, and F1 score was 0.503. External validation of the LR model yielded an AUC value of 0.727 (95% CI 0.674-0.780). In terms of calibration curves, apart from the internal validation set showing a slight overestimation, both the training set and the external validation set demonstrated good consistency. Moreover, the decision curve analysis showed that the model has appropriate clinical applicability.

Conclusions

The predictive model could help healthcare professionals predict and address potential risks of LOS extension in VPIs.