Reply Letter: A Commentary on “A Predictive Model of Failure to Rescue After Thoracolumbar Fusion”

To the editor,

We appreciate the editorial team’s thoughtful commentary on our article [1] analyzing predictors of failure to rescue (FTR) in patients undergoing thoracolumbar fusion. They raised valuable concerns about the impact of postoperative complications like cardiac arrest in affecting FTR rates. We acknowledge that the incremental addition of this particular variable could have allowed for a better understanding of its applicability in clinical practice.

We recognize that cardiac arrest often occurs in close proximity to mortality, which may limit its value as a predictor for early intervention. However, due to the structure of the National Surgical Quality Improvement Program database, we were unable to determine whether cardiac arrest was due to an underlying complication. The lack of temporal granularity in the dataset prevents establishing causal relationships between complications, making it difficult to exclude cardiac arrest without omitting other key postoperative predictors in predicting FTR. To further address this concern, a sensitivity analysis could assess the model’s performance excluding cardiac arrest as a predictor. This approach would clarify its impact on the model and determine whether its inclusion significantly affects predictive accuracy. This analysis could also help to differentiate whether the model predicts mortality rather than actionable opportunities for intervention.

Our decision to include cardiac arrest as a predictor aligns with prior research demonstrating that postoperative complications such as myocardial infarction, unplanned intubation and stroke have been independently associated with FTR in patients undergoing brain tumor resection [2,3]. While frailty and preoperative characteristics remain crucial predictors, excluding major postoperative events underestimates the multifactorial nature of FTR.

Therefore, our analysis was not limited to preoperative characteristics and incorporated both patient-specific and postoperative risk factors. FTR is defined as death occurring after a major complication. However, the definition of “major complication” varies across specialties, with certain specialties including cardiac arrest as a postoperative predictor of FTR [4,5]. FTR was originally developed as a quality metric in measuring a hospital’s ability to prevent mortality after the occurrence of a major complication [6]. However, patient-specific characteristics and postoperative complications have also been shown to affect FTR.

To improve future iterations of this study, we propose a 2-step modeling approach:

• (1) Develop an initial model using only preoperative and early postoperative predictors available at the time of the first recorded complication.

• (2) Access the incremental performance changes from adding later-stage complications such as cardiac arrest, thereby distinguishing between early actionable risks and terminal evens that confirm patient mortality risk.

Although we agree that cardiac arrest may have limited actionability as a late-stage predictor, its inclusion stresses the importance of enhanced surveillance and targeted postoperative care. Future studies incorporating real-time monitoring of complications may enhance the predictive value of FTR models, allowing for risk assessment and timely intervention strategies. While we acknowledge the limitations of our current study, we believe that refining risk stratification and postoperative care protocols remain critical avenues for future research aimed at reducing FTR.