This study developed and validated a novel prediction model for PMV following TTVR using the Fine-Gray competing risks regression approach. Our model demonstrated good discriminative ability and calibration in both the training and validation sets, with AUCs of 0.747 and 0.833, respectively. The identified predictors of PMV included CPB time, ejection fraction, NYHA grade, serum albumin, atelectasis, pulmonary infection, pulmonary edema, age, need for postoperative dialysis, hemoglobin levels, and PaO2/FiO2 ratio.
Our findings align with previous studies that have identified similar risk factors for PMV after cardiac surgery. Cislaghi et al. reported that advanced age, low EF, low hemoglobin level, and prolonged CPB time were associated with increased PMV risk,7 while Oura et al. and Rajakaruna et al. identified NYHA grading as a predictor of PMV.89 The consistent association of CPB time with increased PMV risk underscores the importance of minimizing CPB duration when possible.10 The impacts of EF and NYHA grading highlight the significance of preoperative cardiac function in determining postoperative respiratory outcomes.11 Including serum albumin as a predictor underscores the role of nutritional status in postoperative recovery, suggesting that preoperative nutritional optimization could reduce PMV risk.12 Additionally, advanced age emerged as a significant predictor, reflecting the impact of patient frailty on recovery. The influence of hemoglobin levels suggests that managing preoperative anemia may be crucial in reducing PMV risk.13
Postoperative pulmonary complications, such as atelectasis, infection, and edema, are common complications following cardiac surgery.14 Due to the necessity of accessing the right chest during thoracoscopic surgery, there is inevitably an impact on the lungs, increasing the incidence of complications like postoperative atelectasis, which can affect ventilation duration. Our prediction model incorporates these early postoperative factors, along with the need for postoperative dialysis and the PaO2/FiO2 ratio, as important predictors of PMV. The PaO2/FiO2 ratio further highlights the critical role of immediate postoperative pulmonary function in determining ventilation duration.15 The need for postoperative dialysis is another crucial factor influencing PMV. Renal dysfunction following cardiac surgery can exacerbate fluid imbalance and impair pulmonary function, contributing to longer ventilation times.16 Postoperative dialysis, therefore, serves as an indicator of severe complications and systemic stress, which correlates with extended PMV. Defined as occurring within the first 24 hours after surgery, these factors provide timely information for clinical decision-making while capturing most relevant early complications. This dynamic approach enhances the model's predictive capability, allowing for both preoperative risk assessment and postoperative updates. It guides management strategies such as preventing atelectasis, promptly treating pulmonary infections, managing fluid balance, and optimizing ventilator settings to improve the PaO2/FiO2 ratio, thereby potentially reducing PMV risk. The inclusion of postoperative factors, supported by previous research demonstrating improved accuracy in PMV prediction,1718 represents a significant strength of our model. Its robust performance in both training and validation sets suggests potential generalizability, emphasizing the critical role of immediate postoperative pulmonary function in determining ventilation duration and the need for strategies to prevent and promptly manage these complications.
One important aspect to consider in the context of PMV prediction is the precise definition of prolonged mechanical ventilation, as substantial variation exists in the terminology and definitional criteria for cohorts of subjects receiving PMV.19 In our study, PMV was defined as the need for continuous mechanical ventilation for more than 72 hours following surgery. However, if mechanical ventilation was discontinued within 72 hours but needed to be resumed within 48 hours, it was still considered as PMV. Successful weaning from the ventilator was defined as the patient's ability to tolerate 48 hours without ventilator support. This definition is based on clinical practice and previous studies, but there is a recognized need for standardization in this area.
Additionally, the methodology used in our study to account for competing risks further strengthens the predictive power of our model. The use of the Fine-Gray competing risks regression model is a notable aspect of our study. This approach accounts for the competing risk of death, which is crucial in the context of cardiac surgery where early mortality can preclude the occurrence of PMV. Previous studies, such as that by Staffa et al., have highlighted the importance of considering competing risks in prognostic models for cardiac surgery outcomes.20
The clinical implications of our predictive model are significant, as it enables accurate identification of patients at high risk for PMV, thereby allowing clinicians to implement targeted interventions to prevent or mitigate this complication. These interventions might include preoperative optimization of comorbid conditions, meticulous intraoperative management to reduce CPB time, and aggressive postoperative care to prevent complications like atelectasis and pulmonary infection. The model also aids in patient counseling and resource allocation in the intensive care unit. The use of a nomogram in clinical practice provides an intuitive and user-friendly tool for risk assessment,21 allowing clinicians to input individual patient variables and generate a personalized risk score for PMV, thus supporting informed decision-making and efficient resource allocation.
Our study has several limitations. First, its retrospective nature introduces the potential for bias and unmeasured confounding. Second, while our model showed good performance in internal validation, external validation in different patient populations and healthcare settings is necessary to confirm its generalizability. Third, our study was conducted at a single center with extensive experience in TTVR, which may limit its applicability to centers with less experience in this technique. Future research should prospectively validate our model in diverse clinical settings to confirm its generalizability and assess its impact on patient outcomes. While our model performed well, the inclusion of early postoperative factors limits its use for pure preoperative risk prediction. Therefore, future studies should develop two complementary models: one for preoperative and intraoperative risk assessment, and another for postoperative risk reassessment and management. Additionally, exploring targeted interventions based on our model's risk stratification would be valuable. Incorporating dynamic variables, such as intraoperative hemodynamic parameters and real-time postoperative data, could enhance the model's predictive accuracy. As minimally invasive cardiac surgery techniques evolve, integrating machine learning may refine the model,22 identify new predictors of PMV, and optimize patient care.