Construction and validation of a risk prediction model for early ventilator-induced diaphragm dysfunction in...

Mechanical ventilation is a core life-support technique in intensive care units (ICUs), but it can easily induce ventilator-induced diaphragm dysfunction (VIDD) in the early stage of mechanical ventilation (within 24 h). VIDD is defined as new-onset diaphragm dysfunction caused by mechanical ventilation, and diaphragm ultrasound is the gold standard for its diagnosis. This study aimed to explore the risk factors of early VIDD in mechanically ventilated patients and construct a temporally validated risk prediction model using clinical baseline indicators.

A prospective study was conducted on 470 mechanically ventilated patients who were admitted to the ICU of Shenzhen Nanshan People's Hospital from May 2024 to May 2025. Patients were divided into a modeling group (n = 328) and a temporal validation group (n = 142) at a 7:3 ratio. VIDD was diagnosed by the simultaneous fulfillment of three ultrasound indicators (AND logic): diaphragm thickening fraction (TFdi) < 20%, diaphragm excursion(DE) < 1 cm, and diaphragm thickness(Tdi) < 0.2 cm. These ultrasound indicators served exclusively as the outcome diagnostic criteria, not as predictive variables. Based on the incidence of VIDD, patients were categorized into VIDD and non-VIDD groups. In total, 34 potential predictive variables were collected, including general data, mechanical ventilation-related parameters, medication use, treatment methods, laboratory test indicators, and relevant rating scales. Independent risk factors were identified via Lasso regression directly without univariate pre-screening, followed by multivariate logistic regression. R software was applied to construct a nomogram model. The discriminative ability, calibration, and clinical validity of the prediction model were evaluated using the area under the receiver operating characteristic (ROC) curve (AUC), calibration curve, and decision curve analysis (DCA).

Among 328 patients in the modeling group, 160 developed VIDD, with an incidence rate of 48.78%. Independent risk factors for VIDD included complicated sepsis, controlled mechanical ventilation mode, Richmond agitation-sedation scale (RASS) score ≤ - 4, acute physiology and chronic health evaluation II (APACHE II) score >20, and complicated intra-abdominal hypertension (IAP ≥ 16 mmHg) (all p < 0.05). No diaphragm ultrasound-derived indicators were included in the final model, as they were used for outcome diagnosis rather than predictive variables. The nomogram model constructed based on the above factors indicated an AUC of 0.820 (95% CI: 0.775 ~ 0.865) in the modeling group, with a Youden index of 0.514, a sensitivity of 0.806 (0.745, 0.867), and a specificity of 0.708 (0.640, 0.777). The AUC of the temporal validation group was 0.816 (95% CI: 0.748 ~ 0.885), with a Youden index of 0.564, a sensitivity of 0.783 (0.685, 0.880), and a specificity of 0.781 (0.686, 0.876), suggesting excellent discriminative ability and favorable external stability of the model. In the Hosmer-Lemeshow test (modeling group: χ ² = 4.957, p = 0.549; validation group: χ ² = 8.431, p = 0.208), the calibration curve indicated a high degree of agreement between the predicted values and the actual values. The DCA threshold probabilities of the model were 0.11-0.88 and 0.13-0.77 in the modeling and validation groups, respectively, suggesting a wide range of beneficial thresholds and good clinical validity.

The nomogram model constructed based on the five risk factors exhibited favorable efficacy for predicting early VIDD. Therefore, the model can be applied as a convenient quantitative tool for the early identification of high-risk patients in clinical practice.