Data Source

The ACS-NSQIP-P is a nationwide, prospectively collected patient database with over 50 participating institutions and more than 300 patient variables.¹ It includes the following de-identified and Health Insurance Portability and Accountability Act (HIPAA)–compliant variable categories: patient demographics, comorbidities, operative variables, preoperative lab values, primary procedure current procedural terminology (CPT) codes, International Classification of Diseases, Ninth Revision (ICD-9) codes, and 30-day postoperative events such as readmission, reoperation, mortality, and complications.

Rates of discrepancies between data abstractors are less than 2% as data abstractors are trained by the ACS to ensure the collection of high-quality data.¹ Patient followup is performed by telephone or letter for a maximum of 30 days after discharge.¹ Previous studies have shown that the NSQIP-P achieved 91.4% confirmed 30-day followup.⁴ Institutional participation in the NSQIP is associated with reductions in postoperative adverse events¹¹ and allows for a more thorough risk-adjusted analysis than an institutional administrative database.²⁵ The NSQIP has been shown to capture all-cause and unplanned readmission occurrences more accurately than information from medical records regarding the cause of readmission.²² Our institution does not require institutional review board approval for NSQIP studies given that NSQIP data are non-identifiable and HIPAA compliant.²⁴

Data Acquisition

We queried the NSQIP-P database for patients younger than 18 years (at the time of the procedure) who had undergone a procedure performed by a neurosurgeon or pediatric neurosurgeon during the period from 2012 to 2013. Procedures were grouped by CPT codes into the following procedural categories: spine; craniotomy for craniosynostosis; craniotomy for neoplasm; craniotomy for Chiari decompression; shunt or ventricular catheter placement; shunt or ventricular catheter revision, removal, or irrigation; myelomeningocele (MMC) repair; skin lesion; and other. Procedure categories and specific CPT codes within each category are listed in Fig. 1.

Cohort selection by surgical subspecialty and procedural classification by CPT code. N = number of cases.

Patient demographic information included age, sex, and race. Patient comorbidities of interest included obesity, pulmonary comorbidity, gastrointestinal comorbidity, renal comorbidity, CNS comorbidity, cardiac comorbidity, steroid use (within 30 days before the principal procedure or at the time the patient was being considered as a candidate for surgery; it did not include short-term use, such as a 1-time pulse, limited short course, or taper of ≤ 10 days), chemotherapy within 30 days prior to surgery, radiotherapy within 90 days prior to surgery, open wound (with or without infection), tracheostomy at the time of surgery, immune disease or immunosuppressant use, nutritional support (intravenous or nasogastric tube), bleeding disorder, hematological disorder, current or previous history of malignancy, history of prematurity, intraventricular hemorrhage, preoperative systemic inflammatory response syndrome (SIRS) or sepsis within 48 hours prior to surgery, and congenital malformation of any organ system (a detailed list of conditions included within each NSQIP-P categorical comorbidity variable is shown in Supplement S1).

Operative and hospital variables of interest included hospital length of stay; inpatient or outpatient status; concurrent procedure status; prior operation within 30 days; transfer status; discharge destination; operative time; American Society of Anesthesiologists (ASA) score; blood transfusion; elective, urgent, or emergent triage; and wound classification.

Preoperative lab values of interest included hypoalbuminemia, hyponatremia, hypernatremia, elevated white blood cell (WBC) count, thrombocytopenia, elevated aspartate transaminase (AST), elevated blood urea nitrogen (BUN), abnormal prothrombin time (PT), and abnormal partial thromboplastin time (PTT).

Noted postoperative complications included infections (surgical site infection [SSI], sepsis, urinary tract infection [UTI], pneumonia, central line bloodstream infection), wound disruption, unplanned intubation, renal failure or insufficiency, coma lasting more than 24 hours, cerebrovascular accident (CVA; stroke or intracerebral hemorrhage) or intracranial hemorrhage, seizure, peripheral nerve injury, cardiac arrest, graft or prosthesis failure, deep venous thrombosis (DVT), and pulmonary embolism (PE). The NSQIP-P data on time to postoperative complications were analyzed.

Unplanned readmission is defined in the NSQIP database as “any unplanned readmission for any reason within 30 days of the principal surgical procedure. The readmission has to be classified as an ‘inpatient’ stay by the readmitting hospital, or reported by the patient/family as such.”¹ Unplanned readmission is coded in the NSQIP “if the readmission was unplanned.” Previous studies have demonstrated that an unplanned readmission designation in the NSQIP has greater than 95% agreement with hospital chart records regarding planned versus unplanned readmission designation.²²

We defined readmission risk factors as any patient characteristic or event (pre- or postoperatively) that increased the likelihood of postoperative readmission. This definition included postoperative complications (for example, SSI) that could also be reasons for readmission.

Data Analysis

Univariate analysis of unplanned readmission outcome association with procedure type, patient demographics, patient comorbidities, preoperative lab values, operative variables, and postoperative complications was performed using the chi-square test, Fisher's exact test, or univariate logistic regression where appropriate. Variables with significance of p ≤ 0.20 in the univariate analyses were then entered into a multivariate analysis via binary logistic regression and were considered independently significant when p ≤ 0.05. As a secondary analysis, we applied a Bonferroni correction to the univariate analysis to correct for the higher risk of Type I error due to the use of over 100 independent variables. In this case, only variables with p ≤ 0.002 in the univariate analysis were entered into the multivariate analysis (the α_corrected value for reaching significance in the Bonferroni correction is calculated by dividing the original α value—in our case, 0.2—by the number of independent variables in the analysis). Receiver operating characteristic (ROC) curve analysis was performed for multivariate regression model validation. Statistical analyses were performed using SPSS version 22.0 (IBM Corp.).

Complications

Postoperative infection, particularly SSI, was the strongest predictor of readmission. This observation is not surprising given that SSI is not uncommon and is associated with significant morbidity and mortality, particularly if not caught and treated early. Surgical site infection has been shown to cause up to 70% of reoperations after pediatric spine surgery¹⁷ and is one of the strongest predictors of readmission after pediatric plastic surgery.²³ However, the relatively high rate of SSI and readmissions related to SSI could indicate a need to continue to study and improve preventative measures before, during, and after surgery (prophylactic antibiotic use, close monitoring of sterile technique, postoperative wound monitoring or cleaning) to prevent occurrences of SSI and SSI-related return to system. Additionally, wound disruption was a common reason for readmission and was independently associated with readmission.

Shunt failure was the most common ICD-9 diagnosis on readmission. In pediatric neurosurgery, shunt failure is a common and oftentimes unpreventable phenomenon, occurring in approximately 12% of pediatric shunt procedures within 30 days or less,¹¹ and up to 46% of procedures had reported failure rates when the follow-up period lasted up to 1 year.^6,10 The 30-day readmission rate for shunt operations (15.7%) in this study is substantially similar to that published in previous work²⁰ and may be useful as a benchmark for future studies aimed at lowering adverse shunt surgery events. As expected, the majority of readmissions after a CSF shunting procedure were related to recurrent shunt difficulties. Aside from shunt failure, however, infection-related readmissions (organ or space SSI and wound disruption) were the most common reasons for readmission among the shunt procedures with the highest readmission rates. These results indicate that measures to prevent SSI and wound disruption could help to reduce unplanned readmission after pediatric neurosurgery.

Readmission occurred, on average, approximately 2 weeks postoperatively. Given that postoperative infections (systemic or SSI) were the strongest predictors of readmission, it is expected that the time course to readmission would follow a course similar to the time required for common infection complications to develop. Again, this finding highlights the need for improved postoperative wound monitoring and better operative wound management to prevent unplanned readmission. The present study supports the notion that ongoing efforts to minimize SSIs in pediatric neurosurgery have the greatest potential to reduce readmission rates.

Patient-Related Factors

The percentage of patients with any comorbidity was 91.3%, and 45.3% of patients had any non-CNS comorbidity. The high comorbidity rate in pediatric neurosurgery relative to that in other specialties could be one possible cause of the higher readmission, morbidity, and mortality rates observed in this specialty relative to the rates in others.⁴ Comorbidities independently associated with readmission included the presence of any comorbidity, steroid use, oxygen support, nutritional support, and seizure disorder. Chronic steroid use has not been well studied in pediatric populations, although some institutions have initiated steroid avoidance protocols before renal transplantation surgery.¹⁵ Nutritional support has been indicated as a risk factor for readmission after pediatric cardiac surgery¹⁶ and cleft palate repair.¹⁹ The comorbid conditions that have a strong association with unplanned readmission could be used as preoperative risk stratification variables. Patients who are considered high risk based on these findings may benefit from more careful discharge planning or increased outpatient attention in an effort to prevent readmission.

Native American race emerged as an independently significant risk factor for readmission. Native American race has been reported as a risk factor for readmission after orthopedic surgery¹² and following sepsis,⁸ although its association with readmission after pediatric neurosurgical procedures has not been observed to our knowledge. While a plausible explanation for this is not apparent from these data, it is important to note disparities such as this as an area for future study. An important caveat to this observation, however, is that Native American patients make up a very small proportion of the NSQIP sample. Therefore, observations about this population are potentially subject to error given the magnified effect of small numbers of readmissions in a population with a small denominator. Furthermore, we were unable to determine socioeconomic or geographical data from the NSQIP-P data set, both of which may contribute to demographic differences in readmission rates.

Procedure- and Hospital-Related Factors

We observed a higher rate of readmission among procedures related to CSF shunts, MMC repair, and craniotomy for infratentorial tumor excision. These findings are perhaps unsurprising as these procedures are typically performed in patients with multiple comorbidities and often require close follow-up for revisions. Previous studies have shown high rates of return to system in pediatric shunt surgery, with readmission rates similar to those reported here.²⁰ Interestingly, 4 of the 5 procedures with the highest readmission rates accounted for over 40% of all procedures in the NSQIP-P neurosurgery cohort. Of the procedures with the highest readmission rates, operation for MMC repair was the least frequently performed (1.3% of all procedures), and its high readmission rate is expected given the myriad comorbidities that often occur in patients with MMC.⁵

Emergent or urgent triage status and admission through the emergency department were significant predictors of unplanned readmission on univariate analysis, although emergent or urgent triage was not a predictor on multivariate analysis. These findings are not surprising, yet they may aid in patient risk stratification when ascertaining patient readmission risk preoperatively.

Interestingly, a longer hospital stay was a significant protective factor in multivariate regression. This observation lends credence to the idea that for health care providers the goals of decreased hospital stay and decreased readmission may be competing.¹⁴ Efforts to decrease hospital length of stay may lead to increased readmission and vice versa. However, our data capture readmission within 30 days of the primary procedure, not from discharge; therefore, it follows that a longer hospitalization decreases the chance of readmission within 30 days because of a smaller window of time in which readmission (as defined within 30 days of the primary procedure) can occur.

Study Limitations

This study has several limitations. First, the NSQIP-P database is limited by the type of data provided by the participating institutions and by the NSQIP categorical variables. Although the NSQIP-P is a national database, the case sample is not necessarily nationally representative. Trauma cases are not included in the database. Certain procedures (for example, CPT code 62201: endoscopic third ventriculostomy with choroid plexus cauterization) are not included either. The severity of categorically coded conditions cannot be ascertained, limiting our ability to associate certain patient conditions with readmission outcomes. Some comorbidities (for example, cardiac risk factors) are insufficiently granular and do not allow for differentiation between different comorbidity subtypes (for example, atrial septal defect vs hypertrophic cardiomyopathy). Facility identifiers are not included in the NSQIP, which prevents analysis of facility outliers with substantially higher or lower rates of readmission. As is the case with many patient databases, preventative preoperative measures are not tracked, probably resulting in overestimation of risk factors. Furthermore, not all patients had their preoperative laboratory values measured; albumin and AST had missing value rates of greater than 80%, limiting our ability to interpret statistical analysis of the differences in preoperative laboratory values between groups. Data on the exact reason (that is, the ICD-9 diagnosis) for readmission in the NSQIP are not always reported or entirely clear. Of the 1098 unplanned readmissions, 746 (67.9%) were directly related to the primary procedure. Of the remaining 32.1% of readmissions unrelated to the primary procedure, the NSQIP data abstractors may not have captured the reason for readmission, which could limit the thoroughness of our readmission reason results. Using readmission as an outcome measure has particular caveats as higher or lower rates of readmission alone may not necessarily indicate poorer or better surgical care quality.

Finally, the large number of variables analyzed increases the risk of Type I error, or false-positive findings. The Bonferroni method of multiple measures correction is very conservative, greatly decreasing the risk of incorrectly rejecting the null hypothesis (Type I error), but at the cost of increasing the risk of incorrectly accepting the null hypothesis (Type II error). In most multivariate logistic regression analyses, a higher α is accepted when selecting variables to include in the model in an effort to avoid inappropriately excluding important variables. Thus, we present the corrected model for reference but base our discussion on the uncorrected model.

Despite these limitations, this study may aid surgeons in identifying procedures and patient risk factors that predispose patients to a higher risk of readmission after pediatric neurosurgery. Patients undergoing longer procedures or procedures related to CSF shunting, MMC repair, or craniectomy for infratentorial brain tumor excision are at greater risk for readmission, especially if they are transferred from the emergency department. Patients undergoing spine procedures or craniotomy for craniosynostosis have a lower risk of readmission compared with those undergoing other procedures. Patients who are Native American, have any preexisting comorbidity, have undergone an operation in the previous 30 days, or have a seizure disorder should be considered at greater risk for unplanned readmission. In addition, patients who require oxygen supplementation, nutritional support, or long-term steroids should also be considered to have a greater risk for readmission. Finally, patients who experience postoperative infection (SSI or systemic infection) should be considered at the greatest risk for readmission. These data may also prove useful for family and patient counseling prior to neurosurgical operations.

Online-Only Content

Supplemental material is available with the online version of the article.

• Supplement S1. http://thejns.org/doi/suppl/10.3171/2016.2.PEDS15604.