To the Editor:
It was with great interest that we read your recent article, “Pediatric Risk Stratification Is Improved by Integrating Both Patient Comorbidities and Intrinsic Surgical Risk,”1 as this model could be useful in prognostication of negative outcomes after surgery, quality improvement, and risk adjustment. This methodologically rigorous analysis empirically derived procedural risk groupings, and added these groupings into a predictive model for 30-day postoperative mortality after common pediatric surgical procedures utilizing The American College of Surgeons National Surgical Quality Improvement Program Pediatric Surgical Risk Calculator dataset. The model also contained five patient variables (American Society of Anesthesiologists Physical Status, weight less than 5 kg, sepsis, preoperative mechanical ventilation, and preoperative vasopressors) to adjust for patient comorbidity. The inclusion of procedural risk groupings improved model discrimination significantly, and groupings were said to represent the “intrinsic surgical risk” of the procedures analyzed.
However, the procedural groupings presented in the appendix aggregate dissimilar procedures into the same risk category, and also separate similar surgical procedures with disparate indications into different risk categories. For example:
Surgeries to repair craniosynostosis and spinal fusion, which involve large fluid shifts, transfusion of blood products—and in the case of craniosynostosis surgery, a craniotomy—are grouped in the lowest risk category together with digit reconstruction, repair of syndactyly, upper endoscopy, and bilateral myringotomy tubes.
Trachesotomy is in the highest risk category, while tracheoplasty and pharyngoplasty, procedures that involve similar surgical and anesthetic risks, are in the lowest risk category.
Burr hole is grouped in the highest risk category, while craniotomy for tumor resection, which is more likely to involve blood loss and fluid shifts, is categorized as high-middle risk.
Laparoscopic ileostomy, jejunostomy, and colectomy are grouped with the lowest risk procedures, while appendectomy is categorized as low-medium risk; laparoscopic colectomy for congenital megacolon or cecostomy, proctectomy, or small intestine resection are categorized as high-medium risk.
Pancreatectomy is in the highest category risk. Ventriculoperitoneal shunt and peritoneal dialysis catheter, which typically incur much shorter operative times and do not commonly involve major blood loss or fluid shifts, are also in this highest risk category.
Is it possible that these procedural groupings reflect not only “intrinsic surgical risk” but also insufficiently adjusted-for patient risk factors and surgical circumstances? For instance, laparoscopic appendectomy may incur higher mortality risk than laparoscopic ileostomy because of unadjusted-for acute illness (i.e., the patient is acutely ill, but not septic or on vasopressors), and insufficient time for surgical optimization. Do burr hole, ventriculoperitoneal shunt, and peritoneal dialysis catheter placement confer higher risk of death than a craniotomy because they are more likely to be performed under emergent circumstances such as acutely elevated intracranial pressure, need for intrathecal chemotherapy, or urgent need for dialysis, which the model does not adjust for? Is it possible that surgery for craniosynostosis and spinal fusion, or tracheoplasty and pharyngoplasty, fall into the lowest risk category because most patients present for these elective procedures fully optimized, and are therefore at low risk of death despite the high likelihood of major blood loss and fluid shifts (in the case of craniosynostosis surgery and spinal fusion), or airway loss (in the case of tracheoplasty or pharyngoplasty)?
While sample size and event rate limitations likely limited the authors’ ability to adjust for additional patient risk factors (and indeed, in any model it is impossible to do so completely), we have concerns that the identified procedural risk groupings reflect patient risk factors and surgical circumstances in addition to—and in some cases more so than—intrinsic surgical risk, which may limit its utility for risk adjustment in other settings. Given the constraints of the data we have currently, would it be better to think of intrinsic surgical risk as a function of surgical duration, operative site, and the likelihood of major fluid shifts rather than as empirically derived groupings that are likely to reflect unadjusted-for confounding?
The authors declare no competing interests.