Thirty-day outcomes of cerebrospinal fluid shunt surgery: The National Surgical Quality Improvement Program-Pediatrics.

Transcription

1 J Neurosurg Pediatrics 14: , 2014 AANS, 2014 Thirty-day outcomes of cerebrospinal fluid shunt surgery: data from the National Surgical Quality Improvement Program-Pediatrics Clinical article Joseph H. Piatt Jr., M.D. Division of Neurosurgery, Nemours Neuroscience Center, Alfred I. dupont Hospital for Children, Wilmington, Delaware; and Department of Neurological Surgery and Pediatrics, Thomas Jefferson University, Philadelphia, Pennsylvania Object. Cerebrospinal fluid shunts are the mainstay of the treatment of hydrocephalus. In past studies, outcomes of shunt surgery have been analyzed based on follow-up of 1 year or longer. The goal of the current study is to characterize 30-day shunt outcomes, to identify clinical risk factors for shunt infection and failure, and to develop statistical models that might be used for risk stratification. Methods. Data for 2012 were obtained from the National Surgical Quality Improvement Program-Pediatrics (NSQIP-P) of the American College of Surgeons. Files with index surgical procedures for insertion or revision of a CSF shunt composed the study set. Returns to the operating room within 30 days for shunt infection and for shunt failure without infection were the study end points. Associations with a large number of potential clinical risk factors were analyzed on a univariate basis. Logistic regression was used for multivariate analysis. Results. There were 1790 index surgical procedures analyzed. The overall rates of shunt infection and shunt failure without infection were 2.0% and 11.5%, respectively. Male sex, steroid use in the preceding 30 days, and nutritional support at the time of surgery were risk factors for shunt infection. Cardiac disease was a risk factor for shunt failure without infection, and initial shunt insertion, admission during the second quarter, and neuromuscular disease appeared to be protective. There was a weak association of increasing age with shunt failure without infection. Models based on these factors accounted for no more than 6% of observed variance. Construction of stable statistical models with internal validity for risk adjustment proved impossible. Conclusions. The precision of the NSQIP-P dataset has allowed identification of risk factors for shunt infection and for shunt failure without infection that have not been documented previously. Thirty-day shunt outcomes may be useful quality metrics, possibly even without risk adjustment. Whether important variation in 30-day outcomes exists among institutions or among neurosurgeons is yet unknown. ( Key Words CSF shunt hydrocephalus shunt failure shunt infection National Surgical Quality Improvement Program-Pediatrics The National Surgical Quality Improvement Program-Pediatrics (NSQIP-P; acsnsqip.org/about.jsp) is a prospective project organized by the American College of Surgeons in collaboration with the American Pediatric Surgical Association. 6 Participating hospitals employ dedicated medical record analysts who abstract randomly selected surgical charts for a large number of demographic and clinical variables. The NSQIP-P calculates rates of adverse events such as surgical site infection, transfusion, or readmission within 30 days for each participating institution that can be compared anonymously with other participating institutions. In the past, most analysis of CSF shunt outcomes have been based on follow-up of 1 year or longer or have used statistical methods for analysis of survival to account for variable periods of follow-up. 2 4,7 10,12,13,16 20,22 25,27 29 Study Abbreviations used in this paper: CPT = Common Procedural Terminology; ICD-9 = International Classification of Diseases, Ninth Revision; NSQIP-P = National Surgical Quality Improvement Program-Pediatrics. designs featuring 30-day follow-up have clear advantages from the standpoints of expense and methodological simplicity. The goal of the current investigation was to document 30-day outcomes and to determine how useful they can be as measures of quality in the shunt management of childhood hydrocephalus. Methods Data Extraction The current study is based on data collected by the NSQIP-P. The dataset for 2012 has been released to participating institutions. At each participating institution, surgical cases are selected on a random basis, and the associated medical records are reviewed for the following 30 days. In the current analysis the selected case is referred to as the index case. Each file in the dataset contains 129 fields, including detailed descriptors of the preoperative clinical state; diagnostic and procedural codes for the index surgical procedure, for concurrent surgical procedures under the same anesthetic, and for returns to 179

2 J. H. Piatt Jr. the operating room within 30 days; and descriptors of a large variety of postoperative events. Successful 30-day follow-up for at least 80% of all selected surgical patients is a requirement for continuing institutional participation in NSQIP, but this requirement is not stratified further by procedure. Duration of follow-up of individual patients is not documented in the dataset. Files with Common Procedural Terminology (CPT; American Medical Association) codes for ventriculoperitoneal shunt insertion, proximal revision, distal revision, and removal with replacement (62223, 62225, 62230, and 62258, respectively) for the index procedure were extracted for study. The outcomes of interest were returns to the operating room for infection and for failure without infection. was defined as a return to the operating room associated with the International Classification of Diseases, Ninth Revision (ICD-9) diagnostic code Failure without infection was defined as a return to the operating room for any procedure reflecting unsuccessful treatment of hydrocephalus, including external ventricular drainage, shunt revision or removal, endoscopic or stereotactic third ventriculostomy, open or endoscopic fenestration of cyst membranes or septum pellucidostomy, repair of cephalic or spinal CSF fistula, and bur hole drainage or shunt insertion for subdural hematoma. Index procedures with the diagnostic code for shunt infection were excluded from analysis of both end points; that is, if a shunt was recognized to be infected at the time of the index operation, it was not analyzed further. Statistical Analysis Data were organized and analyzed using SPSS (version 19.0, IBM Corp.) and Minitab (Minitab Inc.). Associations of both end points with categorical factors were analyzed with cross-tabulation and chi-square testing or Fisher s exact test. Associations of end points with quantitative factors were analyzed by ANOVA. Factors with univariate associations significant at the a < 0.05 level were used to construct multivariate models with logistic regression using the backward likelihood ratio method. The dataset was divided randomly into training subsets containing 67% of files and validation subsets containing the remaining 33% of files. The intent was to develop models in the training subsets and to assess the accuracy of the models in the validation subsets with receiver-operator characteristic curves. This project was exempted from supervision by the Nemours Delaware Valley Institutional Review Board. Results There were 1790 CSF shunt operations. In 56 cases, shunt infection was recognized at the index operation; thus, 1734 cases remained for analysis. There were 34 subsequent episodes of shunt infection (2.0%, 95% CI 1.4% 2.7%) and 200 episodes of shunt failure without infection (11.5%, 95% CI 10.1% 13.1%) within 30 days. The overall 30-day shunt failure rate was 13.5% (95% CI 11.9% 15.2%). Univariate Associations Univariate associations of these end points with various categorical factors are presented in Table 1. Male sex, structural pulmonary/airway abnormality (but not tracheostomy), steroid use within 30 days, and nutritional support had significant associations with infection. Initial shunt insertion was associated with greater risk of infection than distal shunt revision (3.1% as compared with 1.2%), but this difference did not attain significance. The risk of infection was lowest during the second quarter of the year (1.2% as compared with 2.7% in the third quarter), but this difference did not attain significance. Common Procedural Terminology code, cardiac risk factors, degree of urgency, and wound classification all had significant adverse associations with shunt failure without infection. Neuromuscular disease, age < 1 year, and admission in the second quarter had significant protective effects. With respect to CPT code, the lowest risk of failure was associated with new shunt insertion (7.4%), followed by distal shunt revision (10.9%), removal and replacement (13.9%), and proximal revision (16.2%). As a group, shunt revisions entailed greater risk of failure than new insertions. Cases of shunt infection recognized at the time of the index procedure having been excluded, cases with clean wounds had lower risk of failure than cases with contaminated wounds or dirty/infected wounds. Approximately 90% of cases were performed by surgeons identified as pediatric neurosurgeons and the remainder by neurosurgeons, as determined by the dedicated institutional NSQIP-P record abstractor. Rates of shunt infection and failure without infection were not different between pediatric neurosurgeons and other neurosurgeons for index shunt operations altogether or for individual index CPT codes. Univariate associations of quantitative factors, such as age at time of surgery, weight at surgery, duration of surgery, duration of anesthesia, and many others, were analyzed with respect to shunt infection and shunt failure without infection as well. The association of older age at surgery with failure without infection was the only significant finding: the mean age ± SD of patients experiencing shunt failure was 2733 ± 2182 days, and the mean age of patients who did not experience failure was 2327 ± 2102 days (p = 0.011). Multivariate Associations Construction of statistical models of infection and failure without infection were attempted using logistic regression applied to randomly selected training subsets of the study dataset, as described in the Methods section. Stable models proved impossible for both end points. Repeated application of the same technique to different random training subsets yielded models with different sets of retained, significant independent variables. Therefore, models were constructed for both end points using the entire dataset and all variables that were significant in the univariate analysis. Neither the model for shunt infection nor the model for shunt failure without infection accounted for more than 6% of total variance. Adjusted odds ratios for the variables with retained significance are presented in Table 2. Discussion The NSQIP-P dataset is relatively rich in clinical de- 180

3 Thirty-day outcomes of CSF shunt surgery in the NSQIP-P TABLE 1: Univariate associations of 30-day infection and failure without infection after CSF shunt surgery* Variable Failure w/o male sex p = NS CPT code NS p < transfer status NS NS surgical specialty NS NS do not resuscitate status NS NS ventilator dependence NS NS current pneumonia NS NS history of asthma NS NS history of cystic fibrosis NS NS bronchopulmonary dysplasia NS NS oxygen support NS NS tracheostomy NS NS structural pulmonary/airway abnormality p = NS esophageal/gastric/intestinal disease NS NS biliary/liver/pancreatic disease NS NS cardiac risk factors NS p = previous cardiac surgery NS NS acute renal failure NS NS dialysis NS NS coma >24 hrs NS NS acute neurological deficit NS NS CNS tumor NS NS developmental delay/impaired cognitive NS NS status seizure disorder NS NS cerebral palsy NS NS structural CNS abnormality NS NS neuromuscular disease NS p = 0.05 myelomeningocele NS NS intraventricular hemorrhage NS NS immune disease NS NS bone marrow transplant NS NS solid organ transplant NS NS open wound NS NS weight loss/failure to thrive NS NS bleeding disorders NS NS hematological disorder NS NS chemotherapy for malignancy w/in 30 days NS NS radiotherapy for malignancy w/in 90 days NS NS any sepsis w/in 24 hrs of surgery NS NS inotropic support at time of surgery NS NS previous cardiopulmonary resuscitation NS NS w/in 7 days prior operation w/in 30 days NS NS blood transfusion w/in 48 hrs NS NS childhood malignancy NS NS (continued) TABLE 1: Univariate associations of 30-day infection and failure without infection after CSF shunt surgery* (continued) Variable Failure w/o steroid use w/in 30 days p = NS nutritional support p = NS location of birth NS NS small for gestational age NS NS mode of delivery NS NS degree of urgency (elective, urgent, NS p = emergency) wound classification NS p = ASA classification NS NS quarter of admission NS p = race (white, black, other) NS NS age <1 yr NS p = term neonate NS NS preterm neonate (gestational age 36 wks) NS NS extreme preterm neonate (gestational age NS NS 28 wks) other neurosurgical procedures at time of NS NS index shunt surgery concurrent procedures by other surgeons at time of index shunt surgery NS NS * ASA = American Society of Anesthesiologists; NS = not significant. tail compared with administrative and retrospective institutional sources that have been mined in past studies of outcomes of childhood hydrocephalus. This dataset has been collected prospectively by dedicated, trained analysts, and both positive and negative observations have been captured for a very large number of clinical variables. Thus, the current study has identified a number of clinical factors associated with shunt outcomes that have received little prior attention. The variable structural pulmonary/airway abnormality was associated weakly with shunt infection. Tracheo-bronchomalacia, vocal cord palsy, and obstructive sleep apnea are common conditions among children with hydrocephalus meeting the definition of this variable. The statistical significance of this variable was not retained in multivariate analysis. The associations of steroid use within 30 days and nutritional support with early shunt infection are plausible and in some circumstances potentially actionable. Oral or parenteral but not topical or inhalational steroid administration fulfilled the criterion of the former variable. Total parenteral nutrition or tube feedings of any kind at the time of surgery fulfilled the criterion for the latter. The current study found an association between male sex and early shunt infection. A recent, large-scale study from the Pediatric Health Information System found female sex to be a risk factor for shunt infection, 26 and we have made similar observations 181

4 J. H. Piatt Jr. TABLE 2: Multivariate analysis of 30-day infection and failure without infection after CSF shunt surgery End Point Odds Ratio (95% CI) p Value infection steroid administration ( ) nutritional support ( ) 0.02 male sex ( ) failure w/o infection initial insertion ( ) cardiac risk factors ( ) second quarter admission ( ) neuromuscular disease ( ) age in days (see formula below)* * Odds ratio for the variable age in days is (age in days). in an analysis of the Kids Inpatient Database (Piatt JH and Freibott CE, unpublished data, 2013). In several large institutional patient series, sex was determined to not be a risk factor at all. 5,14,15,21 How the sex of the patient affects the risk of shunt infection must be considered an unresolved question. Failure Without The adverse association of cardiac risk factors with early shunt failure and the protective association of neuromuscular disease were unexpected observations, and they remain obscure. Likewise obscure is the association of wound classification with failure without infection. The NSQIP-P classifies wounds with respect to the index surgical procedure, so the wounds in question were opened or reopened for the purpose of shunt surgery. Notable here is the absence of any association between early shunt failure and prior operation within 30 days. That a neurosurgeon would insert or revise a shunt in a contaminated, dirty, or infected wound is difficult to credit. For this reason, the coding of wound classification was judged to be unreliable, and wound classification was not included in the multivariate analysis. The observation of the protective effect of admission in the second quarter supports the concept of the so-called July effect. This effect retained statistical significance in multivariate analysis of failure without infection, and rates of infection were lowest in the second quarter as well, although statistical significance was not confirmed. Kestle et al. 11,12 and Drake et al. 7 found evidence for seasonal variation in adverse shunt outcomes peaking in July and August in data from the Design and the Endoscopic Insertion Trials and from the Canadian Hospital Discharge Database. That younger patients experienced lower rates of early shunt failure without infection than older patients is perhaps counterintuitive, but the explanation lies in the confounding effect of CPT code: infancy accounted for the preponderance of new shunt insertions, and new insertions enjoyed lower rates of early failure than revisions. In layered cross-tabulation analysis controlled for CPT code, rates of early failure in infancy were no different from rates in later childhood. Nevertheless, consistent with the retention of age in days in the multivariate model, for each CPT code the mean age of patients who experienced early shunt failure was slightly greater than the mean age of patients who did not (data not shown). Risk Adjustment and Quality Measurement The larger goal of the current investigation was to develop risk-adjusted models of 30-day shunt outcomes that neurosurgical practices might use as benchmarks. Using a standard methodology for establishing internal validity, stable models proved unattainable. Adjusted odds ratios were calculated for factors that appeared predictive in this particular dataset, but they cannot be construed as more than hypotheses requiring confirmation in future research. The most likely explanation for this disappointing conclusion is the relatively small number of end points reached within such a short follow-up duration. Collection of additional years of data in the NSQIP-P format may permit development of stable models with internal validity, but there is no reason to expect that they will account for much larger fractions of observed variance than the models constructed in the current investigation. Failure to construct stable models for risk adjustment may have a positive interpretation: 30-day shunt outcomes may be valid metrics without risk adjustment. Further analysis of data by institution and by neurosurgeon is needed. Follow-up involves cost, so 30-day outcomes are attractive quality metrics for large-scale use. A very recent report by Al-Tamimi and colleagues 1 described 30-day shunt failure rates from a consortium of pediatric centers in the United Kingdom and Ireland and compared those rates with a secondary analysis from the prospective Design and the Endoscopic Insertion Trials. 7,12 The 30-day consortium failure rate was 12.9%, and the rates from the 2 prospective trials were 14% and 16%, respectively. These investigators noted that pediatric neurosurgeons enjoyed lower early failure rates after shunt revisions than general neurosurgeons. The NSQIP-P data do not confirm this observation, perhaps because the number of general neurosurgeons practicing in NSQIP centers is so small. The combined NSQIP-P 30-day failure rate in the current study (13.5%) is quite similar to the rates reported by Al-Tamimi et al. 1 The tight clustering of aggregate 30- day failure rates from different continents and, indeed, from different centuries is a confirmation of the generalizability of the observations and supports the external validity of 30-day outcomes as a quality metric. The data that NSQIP-P has released do not permit analysis of variation among participating centers or among individual surgeons; whether much variation exists is unknown. The rates reported by Al-Tamimi et al. and in the current study may be useful benchmarks for major pediatric centers, but their relevance to neurosurgical practices in other hospital settings cannot be assumed. Conclusions This analysis of 30-day shunt outcomes in the management of pediatric hydrocephalus identified steroid use within the preceding 30 days and nutritional support at the time of surgery as possible risk factors for early infection. 182

5 Thirty-day outcomes of CSF shunt surgery in the NSQIP-P Patients undergoing initial shunt insertion, patients admitted in the second quarter, and patients with neuromuscular disease appear to enjoy lower risk of early failure without infection, but heart disease emerged as a possible adverse factor. The tight clustering of 30-day failure rates from contemporary practice in the United States, the United Kingdom, and Ireland and from major North American pediatric centers in the 1990s is evidence for their external validity as quality metrics. Whether there is significant variation in 30-day failure rates among institutions or among surgeons is unknown, but the current study supports further investigation of 30-day shunt outcomes as quality indicators. Acknowledgment I thank Charles D. Vinocur, M.D., the NSQIP-P Surgical Champion at my institution, for facilitating access to the study data and for helpful conversations. Disclosure The author reports no conflict of interest concerning the materials or methods used in this study or the findings specified in this paper. References 1. Al-Tamimi YZ, Sinha P, Chumas PD, Crimmins D, Drake J, Kestle J, et al: Ventriculoperitoneal shunt 30-day failure rate: a retrospective international cohort study. Neurosurgery 74:29 34, Caldarelli M, Di Rocco C, La Marca F: complications in the first postoperative year in children with meningomyelocele. Childs Nerv Syst 12: , Chittiboina P, Pasieka H, Sonig A, Bollam P, Notarianni C, Willis BK, et al: Posthemorrhagic hydrocephalus and shunts: what are the predictors of multiple revision surgeries? Clinical article. J Neurosurg Pediatr 11:37 42, Cochrane DD, Kestle JR: The influence of surgical operative experience on the duration of first ventriculoperitoneal shunt function and infection. Pediatr Neurosurg 38: , Dallacasa P, Dappozzo A, Galassi E, Sandri F, Cocchi G, Masi M: Cerebrospinal fluid shunt infections in infants. Childs Nerv Syst 11: , Dillon P, Hammermeister K, Morrato E, Kempe A, Oldham K, Moss L, et al: Developing a NSQIP module to measure outcomes in children s surgical care: opportunity and challenge. Semin Pediatr Surg 17: , Drake JM, Kestle JR, Milner R, Cinalli G, Boop F, Piatt J Jr, et al: Randomized trial of cerebrospinal fluid shunt valve design in pediatric hydrocephalus. Neurosurgery 43: , Hanlo PW, Cinalli G, Vandertop WP, Faber JA, Bøgeskov L, Børgesen SE, et al: Treatment of hydrocephalus determined by the European Orbis Sigma Valve II survey: a multicenter prospective 5-year shunt survival study in children and adults in whom a flow-regulating shunt was used. J Neurosurg 99:52 57, Hatlen TJ, Shurtleff DB, Loeser JD, Ojemann JG, Avellino AM, Ellenbogen RG: Nonprogrammable and programmable cerebrospinal fluid shunt valves: a 5-year study. Clinical article. J Neurosurg Pediatr 9: , Kestle J, Drake J, Milner R, Sainte-Rose C, Cinalli G, Boop F, et al: Long-term follow-up data from the Design Trial. Pediatr Neurosurg 33: , Kestle JR, Cochrane DD, Drake JM: insertion in the summer: is it safe? J Neurosurg 105 (3 Suppl): , Kestle JR, Drake JM, Cochrane DD, Milner R, Walker ML, Abbott R III, et al: Lack of benefit of endoscopic ventriculoperitoneal shunt insertion: a multicenter randomized trial. J Neurosurg 98: , Kestle JR, Walker ML: A multicenter prospective cohort study of the Strata valve for the management of hydrocephalus in pediatric patients. J Neurosurg 102 (2 Suppl): , Kontny U, Höfling B, Gutjahr P, Voth D, Schwarz M, Schmitt HJ: CSF shunt infections in children. 21:89 92, Kulkarni AV, Drake JM, Lamberti-Pasculli M: Cerebrospinal fluid shunt infection: a prospective study of risk factors. J Neurosurg 94: , Kulkarni AV, Riva-Cambrin J, Butler J, Browd SR, Drake JM, Holubkov R, et al: Outcomes of CSF shunting in children: comparison of Hydrocephalus Clinical Research Network cohort with historical controls. Clinical article. J Neurosurg Pediatr 12: , Liptak GS, Masiulis BS, McDonald JV: Ventricular shunt survival in children with neural tube defects. Acta Neurochir (Wien) 74: , McGirt MJ, Buck DW II, Sciubba D, Woodworth GF, Carson B, Weingart J, et al: Adjustable vs set-pressure valves decrease the risk of proximal shunt obstruction in the treatment of pediatric hydrocephalus. Childs Nerv Syst 23: , McGirt MJ, Leveque JC, Wellons JC III, Villavicencio AT, Hopkins JS, Fuchs HE, et al: Cerebrospinal fluid shunt survival and etiology of failures: a seven-year institutional experience. Pediatr Neurosurg 36: , McGirt MJ, Wellons JC III, Nimjee SM, Bulsara KR, Fuchs HE, George TM: Comparison of total versus partial revision of initial ventriculoperitoneal shunt failures. Pediatr Neurosurg 38:34 40, McGirt MJ, Zaas A, Fuchs HE, George TM, Kaye K, Sexton DJ: Risk factors for pediatric ventriculoperitoneal shunt infection and predictors of infectious pathogens. Clin Infect Dis 36: , Piatt JH Jr, Carlson CV: A search for determinants of cerebrospinal fluid shunt survival: retrospective analysis of a 14-year institutional experience. Pediatr Neurosurg 19: , Pollack IF, Albright AL, Adelson PD: A randomized, controlled study of a programmable shunt valve versus a conventional valve for patients with hydrocephalus. Neurosurgery 45: , Sainte-Rose C, Piatt JH, Renier D, Pierre-Kahn A, Hirsch JF, Hoffman HJ, et al: Mechanical complications in shunts. Pediatr Neurosurg 17:2 9, Shah SS, Hall M, Slonim AD, Hornig GW, Berry JG, Sharma V: A multicenter study of factors influencing cerebrospinal fluid shunt survival in infants and children. Neurosurgery 62: , Simon TD, Hall M, Riva-Cambrin J, Albert JE, Jeffries HE, Lafleur B, et al: rates following initial cerebrospinal fluid shunt placement across pediatric hospitals in the United States. Clinical article. J Neurosurg Pediatr 4: , Simon TD, Whitlock KB, Riva-Cambrin J, Kestle JR, Rosenfeld M, Dean JM, et al: Association of intraventricular hemorrhage secondary to prematurity with cerebrospinal fluid shunt surgery in the first year following initial shunt placement. Clinical article. J Neurosurg Pediatr 9:54 63, Tuli S, Drake J, Lamberti-Pasculli M: Long-term outcome of hydrocephalus management in myelomeningoceles. Childs Nerv Syst 19: , Tuli S, Drake J, Lawless J, Wigg M, Lamberti-Pasculli M: Risk factors for repeated cerebrospinal shunt failures in pediatric patients with hydrocephalus. J Neurosurg 92:31 38, 2000 Manuscript submitted January 14, Accepted May 1, Please include this information when citing this paper: published online June 13, 2014; DOI: / PEDS1421. Address correspondence to: Joseph H. Piatt Jr., M.D., Nemours Neuroscience Center, Division of Neurosurgery, 1600 Rockland Rd., Wilmington, DE jpiatt@nemours.org. 183