Placement and revision of ventricular shunts for hydrocephalus

J Neurosurg 120:684 696, 2014 AANS, 2014 Improvement in clinical outcomes following optimal targeting of brain ventricular catheters with intraoperative imaging Clinical article Christopher G. Janson, M.D., 1,2 Liudmila G. Romanova, Ph.D., 3 Kyle D. Rudser, Ph.D., 4 and Stephen J. Haines, M.D. 1 Departments of 1 Neurosurgery, 2 Neurology, 3 Medicine, and 4 Biostatistics, University of Minnesota School of Medicine, Minneapolis, Minnesota Object. The accurate placement of cerebral ventricular shunt catheters in hydrocephalus is an important clinical problem. Malfunction of shunts remains their most common complication and greatest liability, and the influence of catheter position on shunt function remains poorly defined. The objectives of this study were as follows: 1) determine the accuracy of intraventricular catheter placement with respect to a historically favored target, defined as a 1-cm radius sphere at the anterior lip of the ipsilateral foramen of Monro; 2) confirm that this target represents a satisfactory site for frontal and occipital catheter placement by examining whether inaccuracy is associated with more shunt failures; and 3) determine whether catheter trajectory, use of image confirmation, or other factors are associated with either the accuracy or the longevity of shunts. Methods. A retrospective cohort analysis was conducted on 236 patients with 426 ventricular shunts placed or revised at the University of Minnesota over a 10-year period. Results. Accuracy of shunt placement was optimal in 43.9% of patients and suboptimal or poor in 56.1% of patients. Time to failure was significantly affected by the accuracy of catheter placement with respect to the ipsilateral foramen of Monro, with a 57% higher risk of failure with suboptimal placement (hazard ratio [HR] 1.57, 95% CI 1.26 1.96; p < 0.001) and a 66% higher risk with poor placement (HR 1.66, 95% CI 1.45 1.89; p < 0.001) relative to optimal placement. The odds of highly suboptimal or unacceptable placement were significantly increased by lack of any intraoperative imaging (OR 5.89, 95% CI 2.36 14.65; p < 0.001). Use of a nonfrontal posterior trajectory also showed a trend toward poor placement (OR 1.64, p = 0.138). Conclusions. The historical target for catheter tip placement within 1 cm of the foramen of Monro in the ipsilateral lateral ventricle was associated with significantly longer revision-free survival compared with other locations. This effect remained significant after adjusting for age and whether there was a prior history of shunting. The accuracy of catheter placement in both pediatric and adult patients was strongly associated with use of intraoperative fluoroscopic confirmation. In analyses comparing intraoperative fluoroscopy and no imaging, there was a non statistically significant difference in the 3-year time to failure, but the worst-case scenario of catastrophic short-term failure was almost completely avoided with fluoroscopy. The authors conclude that accuracy of placement is critical for shunt survival, and that use of intraoperative imaging confirmation may optimize outcomes by avoiding the majority of unacceptable placements. (http://thejns.org/doi/abs/10.3171/2013.8.jns13250) Key Words shunt brain cerebral intraventricular ventriculoperitoneal catheter accuracy hydrocephalus Placement and revision of ventricular shunts for hydrocephalus is an important clinical problem with considerable economic impact. Shunt placement is both prevalent and costly, with an estimated more than 5500 annual US hospital admissions, representing over $1.2 billion in annual direct medical costs. 15 Malfunction Abbreviation used in this paper: HR = hazard ratio. of ventricular shunts remains their most common complication and greatest liability, and can be life-threatening if not promptly treated. The required hospitalization and surgery with their associated complications are a source of significant medical risk, social cost, and expense for these patients. Any improvement in shunting outcomes has the potential for significant quality-of-life improvement and financial savings. Proper positioning of CSF shunt cath- 684 J Neurosurg / Volume 120 / March 2014

Optimal targeting of brain intraventricular catheters eters is commonly believed to be important for preventing complications including malfunction, intracerebral hemorrhage, and new lesions causing neurological dysfunction (such as hemiparesis and seizures). It is therefore surprising that there are very few studies directly comparing surgical approaches, intraventricular targets, and image guidance techniques, or that examine long-term shunt outcomes as a function of catheter position. A variety of techniques have been reported to accurately place ventricular, atrial, and peritoneal catheters, yet there is wide variation in utilization of these surgical techniques. In this study, we review our own experience with respect to preferred locations and accuracy of shunt catheter placement, to identify ways to reduce or eliminate the misplacement of shunt catheters. The main hypothesis tested was that a position exists for the cerebral ventricular catheter that leads to the lowest risk of complications and longest shunt survival. While it is generally agreed that the catheter tip should be as remote as possible from both the choroid plexus and the ependymal lining of the cerebral ventricles to prevent occlusion and shunt failure, this hypothesis has never been rigorously tested. It has been presumed on anatomical grounds that the optimal placement of a ventricular catheter tip is anterior to the choroid plexus of the lateral ventricle, which otherwise is liable to occlude the catheter holes. 3 Aside from the tip position, another variable believed to be important is the catheter trajectory. In a retrospective study of 114 patients who underwent freehand catheter placement, Albright et al. concluded that the final location of the catheter tip and the catheter trajectory are both important factors in shunt survival. 1 In that study, which radiographically verified catheter tip position and correlated it with outcomes, catheters placed in the ipsilateral frontal horn functioned significantly longer than those in other locations, and the use of the frontal approach led to a longer time to failure than the posterior approach. In other words, a frontally placed catheter with its tip in the anterior horn of the ipsilateral lateral ventricle was predictive of better long-term outcomes. However, it has been proposed that catheter placement in the ventricular atrium, the largest part of the ventricle, is theoretically more advantageous. In support of this approach, a prospective, randomized study concluded that outcomes using a parietal approach were no different from the frontal approach, but unfortunately these findings were never correlated with the actual catheter tip location. 4 Another group also concluded that the two approaches were equivalent, but again without verifying the catheter tip position. 5 Because our review of the literature did not support a single optimal target for ventricular shunt catheters, we relied on the historical definition of preferred placement generally accepted in our institution for the past 30 years. The intended target of intraventricular shunt catheters has been the frontal horn of the lateral ventricle, ipsilateral to the point of entry. We adopted this traditional location as the intended target, unless the surgeon declared a different intent preoperatively and formulated a grading scheme for the final catheter tip location. We began the study by measuring the accuracy of all catheter placements with respect to the intended target, based on objective postoperative imaging. The next question J Neurosurg / Volume 120 / March 2014 we asked was if the catheter position was associated with time to failure. Indeed, there was a very strong correlation between catheter position and outcomes, and so we examined all the factors that we believed might be associated with accuracy of catheter placement, including intraoperative imaging (fluoroscopy, endoscopy, ultrasonography, MRI, and stereotaxy), surgeon, origin of hydrocephalus, number of antecedent shunts, shunt trajectory (frontal, occipital, or parietal), and whether the catheter tip passed through the interventricular septum. Methods Literature Review Our initial literature review consisted of all prospective, randomized, controlled studies of shunt function in the National Center for Biotechnology Information Pub- Med database since the introduction of modern cerebral ventricular shunts in 1962, using standard search strategies. 4,6,11 Prior studies meeting these criteria examined the effect of catheter position (no difference among trajectories, although there were methodological flaws); 4 the effect of shunt type (no significant difference among hardware); 6 and the effect of endoscopic technique (no significant difference in outcome from nonendoscopic techniques). 11 Because these studies were all negative, the effect of catheter position and targeting methods remains controversial, and the paucity of quality trials rendered meta-analysis of existing data infeasible. Retrospective Database Analysis After determining that there were limited data to support any particular surgical approach or accepted target, we proceeded to conduct our own retrospective analysis of shunt accuracy and outcomes. The study was approved by the University of Minnesota Institutional Review Board. Patients were identified by querying a billing database at the University of Minnesota Hospitals from 2001 to 2010, the years for which reliable digital imaging data were available. A report query was generated using the search criteria of all neurosurgical patients, and further refined by including Boolean searches with ICD-9 codes corresponding to terms for hydrocephalus or shunt. Patients were included as part of this retrospective cohort study if they received a ventriculoperitoneal or ventriculoatrial shunt over the time period of interest. Subgaleal shunts and other temporary catheters were not included in the analysis, because these had a predetermined lifespan. Patients were excluded if no postoperative images were available with which to assess catheter tip placement, if an unconventional target was chosen but the intended target could not be determined, or if relevant records and images could not be obtained. Of 276 patients identified by the search terms above, 236 patients and 426 shunt events met selection criteria and were analyzed. Shunt Survival Shunt survival was defined as days from the time of placement or revision of the ventricular catheter until the first day of shunt failure (and subsequent revision) or death. Any operation requiring removal or revision of the 685

C. G. Janson et al. ventricular catheter or sacrifice of the shunt system was defined as an event. Censoring was based on the date of last known follow-up. Although the likelihood is high that any patient under our care would have returned to our hospital or would have contacted us in the event of shunt failure, we did not assume that a patient s shunt remained intact and functioning as long as the patient did not present for revision or with signs of failure. Instead, the last positive recorded date of shunt functioning was treated as the censoring date. Therefore, patients with a single shunt and final censoring date had not suffered shunt failure at the time of censoring. Imaging Modalities Methods used to assure proper location of the catheter tip included freehand placement with surface landmarks in all patients, and a subset of patients additionally had intraoperative fluoroscopic imaging, endoshunt endoscopy, ultrasonography, MRI, or frame-based stereotaxy. In some patients, multiple imaging modalities were used, such as endoscopy as well as ultrasonography, or fluoroscopy in addition to endoscopy. For freehand placement, the standard surface landmarks were generally measured 11 12 cm posterior to the nasion and 2.5 3 cm lateral to the midline, and aimed at a point half way between the tragus and lateral canthus and a point at the ipsilateral medial canthus. When intraoperative fluoroscopy or a postoperative shunt series was used for target validation, the surgeon used standard operational guidelines: on lateral view, the catheter tip should lie two-thirds of the distance from bregma to the posterior clinoid process, and on anterior view, the catheter tip should lie on the line from the insertion point to the nasion, at the same distance from the outer table of the skull as the distance determined above. 14 Techniques for endoscopy, ultrasonography, or stereotaxy were in accordance with standard surgical practices for these modalities. Stereotaxy used the frame-based Medtronic Stealth system with rigid head fixation, because the Medtronic AxiEM electromagnetic system was not adopted until after the study was completed. There were no specific indications stated for selecting any of these modalities, except that stereotaxy was typically used in the cases of patients with slit ventricles and endoscopy was used for multiloculated hydrocephalus. Postoperative imaging confirmation included plain radiographs (a shunt series) and/or CT or MRI during the same hospital stay or at a scheduled follow-up visit; 71.4% of shunts were examined using postoperative CT or MRI within 48 hours, 79.1% were examined using postoperative radiography (shunt series) within 48 hours, and 87.3% were investigated using either of these modalities, or both, within 48 hours. In addition, all patients (100%) underwent CT or MRI between the time of hospitalization and the first clinic follow-up evaluation. If a surgeon used absorbable sutures, the first postoperative visit with CT or MRI was typically within 4 weeks, but if nonabsorbable sutures were used, the first follow-up was within 2 weeks. Target Measurement For the purposes of this study, we defined the target as a sphere 1 cm in diameter, centered 1 mm anterior and lateral to the ipsilateral foramen of Monro. The OsiriX DICOM viewer was used for primary data analysis with 2D and 3D measurements of the catheter tip placement. Measurements were performed using OsiriX by a reviewer (C.G.J.) who was initially blinded to the insertion technique and outcomes. Because all patients underwent postoperative CT or MRI within 6 weeks, including those with unsuccessful shunt placement within that time frame, these imaging studies were used as the most precise measurement of actual postoperative shunt location. Where the surgeon deliberately selected a different target, the same target sphere centered on the intended target was used. Rules for classification of tip placement were as follows: 1) optimal location was defined as the catheter tip lying within the 1-cm target sphere in the ipsilateral ventricle; 2) suboptimal placement was defined as tip placement outside the target sphere, but within 1 2 cm, or if the catheter tip was present in the contralateral ventricle but with a tip location within 1 cm of the contralateral foramen of Monro; 3) highly suboptimal was defined as tip location beyond 2 cm of the target, and/or if the catheter tip was intraparenchymal; and 4) unacceptable was defined as highly suboptimal placement, and in addition the surgeon returned to the operating room within 2 weeks of the original operation, whether electively or urgently. Reoperation in this early time period was performed either because the patient was having clinical or radiographic signs of shunt failure, or because the surgeon determined that the catheter was in a location that made failure inevitable. In a few cases, the catheter was in extremely poor position and came to attention more gradually as a clinical shunt failure beyond 2 weeks, but was included in this category due to an egregiously bad positioning that was missed on initial imaging. The highly suboptimal and unacceptable categories were grouped together as poor position to simplify some analyses. Additional categories were defined for the catheter tip location with respect to the septum pellucidum: 1) tip through septum pellucidum was defined as the catheter tip passing greater than or equal to 0.5 cm through the septum, as a best estimate from imaging parameters in OsiriX with 3D reconstruction; and 2) tip in septum was defined as the distal end of the catheter lying inside the septum and protruding not more than 0.5 cm, as a best estimate using OsiriX. Descriptive data collected on each patient (and shunt event) included age, sex, number of prior shunt procedures, origin of hydrocephalus (posthemorrhagic, Chiari malformation Type I/II, arachnoid cyst, communicating or idiopathic hydrocephalus, normal pressure hydrocephalus, pseudotumor cerebri, obstructive hydrocephalus from tumor or aqueductal stenosis, or postinfectious), type of shunt (ventriculoperitoneal or ventriculoatrial), shunt valve type, reason for failure (proximal occlusion, distal occlusion, proximal and distal failure, valve failure or disconnection, infection, and idiopathic), surgical trajectory (frontal, nonfrontal occipital, and nonfrontal other), attending surgeon, presence or absence of intraoperative imaging, type of postoperative imaging studies, and tip position with respect to the midline. Statistical Analysis Aside from shunt malfunction or failure, other com- 686 J Neurosurg / Volume 120 / March 2014

Optimal targeting of brain intraventricular catheters plications were classified as infection, new hemorrhage, new neurological deficit, or death. For statistical analysis, in addition to descriptive statistics evaluated across accuracy and imaging categories, we used logistic regression to calculate odds ratios describing the effects of age, sex, imaging modalities, and origin of hydrocephalus on placement. Robust variance estimation was used for confidence intervals and p values. For survival analysis, unadjusted Kaplan-Meier survival plots were examined, including summaries at 6-month increments between placement accuracy or imaging modalities. The log-rank test is based on the robust score statistic of the Cox proportional hazards estimand. Adjusted analyses used Cox proportional hazards models with robust variance estimation and recurrent events to examine factors associated with time to failure. Baseline hazard functions were stratified for incident and recurrent shunt placements and robust variance estimates were based on patient clusters. All analyses were conducted using S-PLUS (Insightful Corp.) and R software packages (http://www.r-project.org). In a separate qualitative cost analysis, we examined the cost effectiveness of intraoperative imaging by determining how many early shunt-related complications led to a prolongation of hospitalization. We calculated the differential cost of fluoroscopy using an estimate of these potentially avoidable costs from our own billing data and from the published literature, and by multiplying these extra costs (due to early complications) by the percentage that were not associated with fluoroscopy, and then subtracting the actual costs of routine fluoroscopy. Results Accuracy of Shunt Placement and Time to Failure We found that, without regard to patient age or method of placement, the accuracy of shunts placed was poor (highly suboptimal or unacceptable) in 96 (22.5%) of 426 cases, suboptimal in 143 (33.6%) of 426 cases, and optimal in 187 (43.9%) of 466 cases. Descriptive statistics for our patient population are shown in Tables 1 and 2 with respect to the accuracy of placement or use of image confirmation. For the primary analysis, the male-female ratio was nearly equal, but only 1 of every 5 patients was in the pediatric age range, suggesting that our analysis was somewhat weighted toward adult patients. For this reason, we performed a separate analysis including only pediatric patients (Table 2), which was defined as age < 19 years at the time of shunt placement. Overall, the mean age at the time of shunt placement was 42 years (range 0 87 years). For all patients, approximately 85% had no shunt history prior to coming to our institution (range 0 35 prior shunts). The underlying reasons for shunt placement were varied, and in some cases there were multiple indications for shunting; for example, a patient may have had a tumor causing obstructive hydrocephalus and then a shunt infection. Thus, the reasons for shunt placement outnumber the number of shunt events. The vast majority of shunts (399, or 93.7% of total), were ventriculoperitoneal shunts, but there were 27 ventriculoatrial shunts (6.3% of total). Outcomes appeared homogeneous for ventriculoperitoneal J Neurosurg / Volume 120 / March 2014 or ventriculoatrial shunts. Of the ventriculoatrial shunts, 17 (63%) of the 27 shunt events and 7 (53.8%) of the 13 patients involved adults, most of whom had complicated shunt histories. The most striking result, as observed in the Kaplan- Meier survival curves (Fig. 1), is that the shunt catheter position had a large effect on time to failure in both adults and children, which was highly significant according to the log-rank (robust score) statistic (p < 0.001). In adjusted multivariate analyses, suboptimal and poor shunt placement were associated with a 1.57-fold (95% CI 1.26 1.96) and 1.66-fold (95% CI 1.45 1.89) increased risk of failure over the first 3 years, respectively (p < 0.001 for both; Table 3). The same pattern was evident in pediatric patients analyzed separately (Table 4). Once we had demonstrated that catheter location was important for time to failure, we investigated other factors that might be associated with accuracy and/or longevity of catheter placement. Intraoperative Imaging, Accuracy of Shunt Placement, and Failure Rates In terms of the accuracy of shunt placement, use of intraoperative image guidance is one factor that is expected to affect the final location of the catheter tip. Although a randomized clinical trial of endoscopy 11 showed no differences in outcomes between shunts placed with or without direct visualization, we hypothesized that other imaging modalities might be superior. In fact, we found that fluoroscopy did confer a highly statistically significant benefit to accuracy of placement and that no imaging or nonfluoroscopic imaging was associated with significantly higher odds of poor placement. With respect to use of imaging, the adjusted odds of poor placement were almost 6-fold higher without any imaging (Table 5). Given the fact that better accuracy was shown to improve shunt survival, it is interesting that the clear benefit of greater accuracy conferred by fluoroscopy did not immediately translate into a long-term benefit in shunt survival at 3 years. Although there was a trend toward improved short-term survival at 6 months in both the aggregate and pediatric cases, the effect was not statistically significant (Fig. 2). The effect of imaging on shunt survival manifested primarily over the first 2 years. For example, looking at the point estimates of the Kaplan-Meier survival curves, the difference in 1-year survival between fluoroscopic imaging and no imaging for all patients (Fig. 2 upper) is 6% and for pediatric patients is 17% (Fig. 2 lower). A short-term benefit to fluoroscopic confirmation is also suggested when noting that a mere 4.2% of poor placements (requiring short-term revision) used fluoroscopy while 47.1% of optimal placements were performed using fluoroscopy (Table 1), even though fluoroscopy accounted for only 26.3% of total shunt placements. Conversely, 81.2% of poor placements were performed without using any imaging and only 43.9% of optimal placements were performed without using imaging, with nonimaging placements representing 62.0% of total placements. Thus, a greater proportion of optimal placements and a lower proportion of unacceptable placements were obtained using image confirmation. 687

C. G. Janson et al. TABLE 1: Characteristics of shunts according to placement (all patients)* Shunt Placement Covariate Total Optimal Suboptimal Poor no. of patients 236 111 87 38 females (%) 115 (48.7) 57 (51.4) 35 (40.2) 23 (60.5) age <19 yrs (%) 49 (20.8) 18 (16.2) 22 (25.3) 9 (23.7) first shunt (%) 201 (85.2) 94 (84.7) 73 (83.9) 34 (89.5) mean age (yrs) ± SD 42.0 ± 25.8 45.0 ± 25.1 39.5 ± 26.2 39.0 ± 26.7 no. of shunt events 426 187 143 96 reason for shunting (%) posthemorrhagic 79 (18.5) 29 (15.5) 30 (21.0) 20 (20.8) Chiari malformation Type I/II 13 (3.1) 7 (3.7) 4 (2.8) 2 (2.1) communicating or idiopathic 71 (16.7) 36 (19.3) 22 (15.4) 13 (13.5) normal pressure hydrocephalus 50 (11.7) 26 (13.9) 17 (11.9) 7 (7.3) pseudotumor 62 (14.6) 24 (12.8) 18 (12.6) 20 (20.8) obstructive (tumor, aqueductal stenosis) 141 (33.1) 66 (35.3) 44 (30.8) 31 (32.3) postinfectious 23 (5.4) 9 (4.8) 9 (6.3) 5 (5.2) imaging (%) any intraop imaging 162 (38.0) 105 (56.1) 39 (27.3) 18 (18.8) no intraop imaging 264 (62.0) 82 (43.9) 104 (72.7) 78 (81.2) intraop cranial fluoroscopy 112 (26.3) 88 (47.1) 20 (14.0) 4 (4.2) intraop stereotaxy 27 (6.3) 17 (9.1) 9 (6.3) 1 (1.0) intraop MRI 4 (0.9) 3 (1.6) 0 (0.0) 1 (1.0) intraop endoscopy or ultrasonography 44 (10.3) 16 (8.6) 16 (11.2) 12 (12.5) postop CT or MRI (within 48 hrs) 304 (71.4) 117 (62.6) 104 (72.7) 83 (86.5) postop shunt series (within 48 hrs) 337 (79.1) 150 (80.2) 112 (78.3) 75 (78.1) tip position (%) through septum 83 (19.5) 14 (7.5) 33 (23.1) 36 (37.5) inside septum 34 (8.0) 15 (8.0) 10 (7.0) 9 (9.4) trajectory (%) frontal 313 (73.5) 151 (80.7) 101 (70.6) 61 (63.5) nonfrontal posterior 101 (23.7) 33 (17.6) 35 (24.5) 33 (34.4) nonfrontal, other 12 (2.8) 3 (1.6) 7 (4.9) 2 (2.1) surgeon (%) A 111 (26.1) 83 (44.4) 23 (16.1) 5 (5.2) B 72 (16.9) 32 (17.1) 27 (18.9) 13 (13.5) C 137 (32.2) 36 (19.3) 57 (39.9) 44 (45.8) D 44 (10.3) 11 (5.9) 12 (8.4) 21 (21.9) E 29 (6.8) 13 (7.0) 12 (8.4) 4 (4.2) F 33 (7.7) 12 (6.4) 12 (8.4) 9 (9.4) cause of failure (%) proximal occlusion 66 (15.5) 14 (7.5) 24 (16.8) 28 (29.2) distal occlusion 12 (2.8) 4 (2.1) 7 (4.9) 1 (1.0) proximal & distal 16 (3.8) 5 (2.7) 4 (2.8) 7 (7.3) valve failure or disconnection 41 (9.6) 12 (6.4) 14 (9.8) 15 (15.6) idiopathic 273 (64.1) 147 (78.6) 89 (62.2) 37 (38.5) infection 18 (4.2) 5 (2.7) 5 (3.5) 8 (8.3) (continued) 688 J Neurosurg / Volume 120 / March 2014

Optimal targeting of brain intraventricular catheters TABLE 1: Characteristics of shunts according to placement (all patients)* (continued) Shunt Placement Covariate Total Optimal Suboptimal Poor postoperative complications (%) new neurological deficit 4 (0.9) 2 (1.1) 1 (0.7) 1 (1.0) new hemorrhage on imaging 21 (4.9) 2 (1.1) 11 (7.7) 8 (8.3) * Values are number of patients (%) unless otherwise indicated. Percentages for shunt variables are shown with respect to number of shunt events in each column. Combined data from the 5 surgeons with the lowest operative frequency. Other Factors Affecting the Accuracy or Longevity of Shunts In terms of other factors affecting time to failure, the effect of surgeon was not statistically significant, although there was some variation with respect to accuracy or rate of complications among the surgeons sampled (Table 1), as well as in their preferences for image confirmation. Thus, a surgeon effect was incorporated into the adjusted multivariate analysis. Older patient age showed a statistically significant effect on hazard ratio (HR), with each 10-year increment conferring a 0.91 HR (p = 0.011). For all patients considered together, there was better survival for those with idiopathic/communicating and posthemorrhagic hydrocephalus origin. The data also support a possible effect of surgical approach or trajectory on both accuracy and time to failure. Relative to accuracy of placement, frontal trajectory represented 73.5% of overall shunting events, but 80.7% of optimal placements and only 63.5% of poor placements. Conversely, the nonfrontal placements were 23.7% of overall shunting events, but only 17.6% of optimal placements and 34.4% of poor placements. The logistic regression model showed higher odds of poor placements among the nonfrontal trajectory placements (adjusted OR 1.64) for all patients, although it was not statistically significant (p = 0.138; Table 5). This effect was independent of imaging use, as frontal and posterior trajectory were similarly distributed with respect to use of imaging, and imaging was adjusted for in the logistic regression model. Apart from placement accuracy, with respect to the ultimate time to failure, use of frontal placement did show a statistically significant effect on outcomes in pediatric patients (HR 0.51, 95% CI 0.29 0.91; p = 0.02; Table 4). Regarding other aspects of catheter tip position, poor accuracy and no imaging were both correlated with placements through the septum. There were 83 placements through the midline, representing 19.5% of total placements. Compared with no imaging, fluoroscopic imaging conferred a 3.1% less proportion of tips through the septum, 2.0% less proportion of tips in the septum, and a 7.1% less proportion of contralateral placements, consistent with better accuracy. These specific adverse placements affected failure rates, because placing the catheter tip inside the septum (protruding < 0.5 cm) was associated with 2.38-fold higher risk of failure (p < 0.001), although penetrating the septum more than 0.5 cm but within 2 cm J Neurosurg / Volume 120 / March 2014 of the target actually appeared to have a 44% lower risk (p = 0.002). The shunt valves used by different surgeons were quite varied, and consistent with the prior randomized clinical trial of shunt design, 6 we found that the effect of shunt valve system on time to failure was negligible. The most represented valves were the Hakim fixed pressure valve, Codman programmable valve, Medtronic Delta valve, and Medtronic Strata valve. Causes for shunt failure were listed when available, but in many cases could not be determined retrospectively; however, the majority appeared to be proximal catheter malfunctions. Aside from shunt failure, other major complications directly caused by shunt placement were recorded (such as new neurological deficit, new hemorrhage, and even death) but were so rare that formal analysis was not possible. There were no deaths attributable to shunt placement, and the incidence of new neurological deficit or new asymptomatic hemorrhage represented 0.9% or 4.9% of the total shunting events, respectively. Among shunts with hemorrhage as a complication, which represented 4.9% of total shunt events, there was a relative underrepresentation of fluoroscopic cases, in which hemorrhage was noted in only 1.8% of such cases, compared with 5.7% in no imaging cases. Estimated Cost Analysis The economic impact of reducing less than optimal catheter placements is important. After concluding that the use of imaging affected the accuracy of shunt placement and led to a decrease in the number of egregious placements, we attempted to estimate the potential financial benefit of real-time confirmation of placement. As part of the retrospective analysis over a decade, we found a total of 32 cases in which shunt complications were associated with prolongation of hospital stay. After excluding 4 highly complex cases unrelated to imaging issues and 2 cases for which intraoperative imaging was also associated with unacceptable results, we found a net 26 prolongations of hospitalization that were all associated with nonuse of fluoroscopy, or roughly 6% of total shunting events. A recent cost analysis from a limited demographic 16 found an interquartile range of reimbursement for shunt hospitalizations to be between $2428 and $18,582. Another study that analyzed the costs of shunt placement using a national database 15 found that the average cost per ad- 689

C. G. Janson et al. TABLE 2: Characteristics of shunts according to imaging (all patients/pediatric patients)* Imaging Type Covariate Total No Intraop Nonfluoroscopic Fluoroscopic no. of patients 236 147 29 60 females (%) 115 (48.7) 70 (47.6) 13 (44.8) 32 (53.3) age <19 yrs (%) 49 (20.8) 24 (16.3) 7 (24.1) 18 (30.0) first shunt (%) 201 (85.2) 129 (87.8) 20 (69.0) 52 (86.7) mean age (yrs) ± SD 42.0 ± 25.8 45.9 ± 24.9 30.9 ± 19.9 37.9 ± 28.6 no. of shunt events 426 264 50 112 placement (%) optimal 187 (43.9) 82 (31.1) 17 (34.0) 88 (78.6) suboptimal 143 (33.6) 104 (39.4) 19 (38.0) 20 (17.9) highly suboptimal 61 (14.3) 50 (18.9) 8 (16.0) 3 (2.7) unacceptable 35 (8.2) 28 (10.6) 6 (12.0) 1 (0.9) tip position (%) through septum 83 (19.5) 53 (20.1) 11 (22.0) 19 (17.0) inside septum 34 (8.0) 24 (9.1) 2 (4.0) 8 (7.1) trajectory (%) frontal 313 (73.5) 193 (73.1) 32 (64.0) 88 (78.6) nonfrontal posterior 101 (23.7) 62 (23.5) 17 (34.0) 22 (19.6) nonfrontal, other 14 (3.3) 9 (3.4) 2 (4.0) 3 (2.7) postoperative complications (%) new neurological deficit 4 (0.9) 2 (0.8) 0 (0.0) 2 (1.8) new hemorrhage on imaging 21 (4.9) 15 (5.7) 4 (8.0) 2 (1.8) pediatric patients only no. of patients 49 28 11 23 females (%) 19 (38.8) 11 (39.3) 4 (36.4) 9 (39.1) age <19 yrs (%) 49 (100) 28 (100) 11 (100) 23 (100) first shunt (%) 41 (83.7) 22 (78.6) 3 (27.3) 16 (69.6) mean age (yrs) ± SD 3.3 ± 5.3 2.8 ± 5.1 6.3 ± 7.5 3.2 ± 4.9 no. of shunt events 111 63 13 35 placement (%) optimal 39 (35.1) 7 (11.1) 5 (38.5) 27 (77.1) suboptimal 47 (42.3) 33 (52.4) 7 (53.8) 7 (20.0) highly suboptimal 17 (15.3) 15 (23.8) 1 (7.7) 1 (2.9) unacceptable 8 (7.2) 8 (12.7) 0 (0.0) 0 (0.0) tip position (%) through septum 9 (8.1) 6 (9.5) 1 (7.7) 2 (5.7) inside septum 3 (2.7) 3 (4.8) 0 (0.0) 0 (0.0) interhemispheric 8 (7.2) 6 (9.5) 1 (7.7) 1 (2.9) trajectory (%) frontal 47 (42.3) 23 (36.5) 3 (23.1) 21 (60.0) nonfrontal posterior 57 (51.4) 35 (55.6) 10 (76.9) 12 (34.3) nonfrontal, other 8 (7.2) 5 (7.9) 0 (0.0) 3 (8.6) postoperative complications (%) new neurological deficit 1 (0.9) 1 (1.6) 0 (0.0) 0 (0.0) new hemorrhage on imaging 5 (4.5) 3 (4.8) 1 (7.7) 1 (2.9) * Values are number of patients (%) unless otherwise indicated. Percentages for shunt variables are shown with respect to number of shunt events in each column. 690 J Neurosurg / Volume 120 / March 2014

Optimal targeting of brain intraventricular catheters to the hospital is $365/hr or approximately $550 per case, in addition to capital costs of $140,000 per unit. Assuming that it had been used in every case, the unit cost of fluoroscopy would be approximately $880 per case, compared with thousands of dollars per case in potential cost savings if intraoperative fluoroscopy were used routinely. In addition to the stated assumptions, this estimate does not take into account the economic losses to the patient and family of additional hospital time, discomfort, and risk in the event of early complications due to no imaging. Fig. 1. Kaplan-Meier plots of revision-free shunt survival according to placement accuracy. Upper: This plot shows shunt survival with interval survival estimates and log-rank p values for aggregate data. There was a highly statistically significant relationship between accuracy of placement and time to failure to the 3-year end point. Lower: This plot is similar to the plot above, but shows data from pediatric patients only (age < 19 yrs). As with the aggregate data, there was a highly statistically significant relationship between accuracy of placement and time to failure to the 3-year end point. mission was $35,816. Approximately half of the patients from that analysis required a hospitalization of more than 5 days, although it was unclear if the prolonged hospitalizations were due to infection, multiple shunting procedures during the same hospitalization, complications, or other factors. For the cases in our study in whom detailed costs were readily available, we found that the mean cost for additional shunting procedures or hospitalizations was $161,176 (range $38,047-$580,799). This assumes that the hospitalization or prolongation of stay was primarily related to shunt complication, which was a fair assumption in these cases based on our chart review. Using the lower estimate of $35,816, this suggests that intraoperative fluoroscopy created a $931,216 cost reduction opportunity for the hospital during the study period, or up to $4.2 million for an estimate based on our small sample of actual patient cost data. The operating cost of intraoperative fluoroscopy J Neurosurg / Volume 120 / March 2014 Discussion Our study suggests a number of clinical implications for placement of ventriculoperitoneal shunts. First, we found that the accuracy of catheter placement, irrespective of technique, was less than ideal. We believe that this underrepresentation of ideal placements is the norm in practice, and has been noted in other published reports that tabulated the accuracy of freehand catheter placements. 10,13 Moreover, we demonstrated that placement accuracy is highly correlated with long-term shunt survival to at least 3 years, and that use of intraoperative confirmatory imaging improves the accuracy of placement. While it would be logical to presume that intraoperative imaging would therefore lead to improved long-term shunt survival, we found that this was not necessarily the case. While we did observe a trend toward better longterm survival with fluoroscopy or other imaging, the effect was not statistically significant. The potential reasons for this fact are numerous, the most important of which are that numerous factors aside from imaging affect shunt longevity (such as origin of hydrocephalus, patient age, prior shunt history, tip placement with respect to septum, and frontal trajectory), and in our patient series there were enough good placements without the use of imaging that the effect was diluted. Use of fluoroscopic imaging tended to result in better placement, but there were still 21.4% suboptimal placements. Moreover, the no imaging group still received optimal placement in 31% and suboptimal placement in 39% overall. The process of shunt failure is complicated and even highly accurate placements occasionally fail at early time points for unknown reasons. Aside from shunt failure, other complications were so infrequent that no firm conclusions could be drawn. In the shunt valve design study by Drake and coworkers 6 that examined first-time pediatric shunt placements, approximately 40% of shunts failed during the 1st year and 50% at 2 years, with median time to failure of 1.8 years. Drake and Sainte-Rose 7 have discussed the hazard function for shunt failure and suggest that pediatric shunt failure is highest in the first few months after surgery, with an average 25% 40% failure rate at 1 year. After this critical period, the rate is approximately 5% per year, with a median failure rate of approximately 5 years across several large series with a combined 2651 patients. This estimate is consistent with a Markov mathematical modeling of 126 case series 17 that predicted a 36% failure rate for children at 1 year and 20% for adults at 1 year, with a median shunt survival in children of approximately 5 years. Our data follow a similar curve with an aggregate 691

C. G. Janson et al. TABLE 3: Adjusted relative hazard ratios over 3 years after placement (aggregate data)* Covariate HR (95% CI) p Value shunt failure risk according to imaging modality unadjusted no imaging imaging (not fluoroscopic) 1.01 (0.82 1.25) 0.906 fluoroscopy 0.95 (0.84 1.07) 0.394 adjusted for all covariates, including surgeon effect no imaging imaging (not fluoroscopic) 0.90 (0.70 1.16) 0.413 fluoroscopy 1.05 (0.88 1.25) 0.622 shunt failure risk according to accuracy of placement unadjusted optimal placement suboptimal placement 1.40 (1.17 1.68) <0.001 poor placement 1.53 (1.38 1.70) <0.001 adjusted for all covariates including surgeon effect optimal placement suboptimal placement 1.57 (1.26 1.96) <0.001 poor placement 1.66 (1.45 1.89) <0.001 female patient 1.25 (0.92 1.68) 0.148 age (per 10 yrs) 0.91 (0.85 0.98) 0.011 frontal trajectory 1.04 (0.75 1.45) 0.796 position through septum 0.56 (0.39 0.81) 0.002 tip inside septum 2.38 (1.48 3.82) <0.001 posthemorrhagic origin 0.64 (0.40 1.03) 0.065 idiopathic communicating origin 0.63 (0.40 0.97) 0.036 obstructive aqueductal stenosis origin 0.86 (0.58 1.28) 0.461 * The HR for shunt failure is lower when fluoroscopic imaging is used compared to when no imaging is used, adjusted for sex and age. The effect is the same magnitude but in the opposite direction after adjusting for all covariates. However, neither of these effects was statistically significant at the 3-year time point, which suggests that the magnitude of effects attributable to fluoroscopy for long-term time to failure are negated by other random factors. In a similar fashion, looking at risk stratified with respect to accuracy of placement, the HR for poor placement is 53% higher than with optimal placement, adjusted only for age and sex. When adjusting simultaneously for all covariates, the HR for poor placement is 66% higher than with optimal placement. Other statistically significant effects include age (protective), idiopathic communicating hydrocephalus origin (protective), and catheter tip interhemispheric or inside the septum at midline (harmful). median survival greater than 3 years. However, what is most striking in our data are that the 3-year survival for optimally placed shunts is greater than 70%, while the 3-year survival for poorly placed shunts is less than 20%. In other words, once the accuracy of placement is taken into account, it is clear that the widely cited median survival time of 5 years may be further risk stratified based on accuracy of placement. The ideal target for catheter tip placement remains unknown, although our previous work 1 and the current study both suggest that frontal placement with a target anterior to the foramen of Monro optimizes accuracy and shunt longevity. Only a prospective, randomized trial directly comparing surgical approaches and imaging modalities (including the use of multiple simultaneous modalities) can definitively answer this question. While frontal and posterior (occipital or parietal) approaches were statistically unable to be distinguished in this study, the odds of a poor outcome were somewhat higher with a posterior approach. Although both approaches are possible for many patients, more bad outcomes occurred using the posterior approach, which is probably due to the more limited angle of approach and less tolerance for error. 12,13 It is possible that because our center has preferred the anterior approach for many decades, we may not be as experienced with posterior approaches as surgeons in centers that generally use that approach. This study was not designed to definitively answer the question of the ideal position for the placement of the tip of the ventricular catheter in CSF shunts for the majority of patients, and it is unclear if there is a single best position for all patients. Given a historically preferred target, we merely sought to justify or refute the designated target with empirical data on shunt survival. Even in the 692 J Neurosurg / Volume 120 / March 2014

Optimal targeting of brain intraventricular catheters TABLE 4: Adjusted relative hazard over 3 years after placement (pediatric patients only)* Covariate HR (95% CI) p Value shunt failure risk according to imaging modality unadjusted no imaging imaging (not fluoroscopic) 0.78 (0.50 1.20) 0.253 fluoroscopic 0.94 (0.77 1.16) 0.581 adjusted for all covariates including surgeon effect no imaging imaging (not fluoroscopic) 0.74 (0.41 1.34) 0.324 fluoroscopic 0.99 (0.75 1.30) 0.943 shunt failure risk according to accuracy of placement unadjusted optimal placement suboptimal placement 1.41 (1.02 1.99) 0.028 poor placement 1.35 (1.10 1.65) 0.004 adjusted for all covariates including surgeon effect optimal placement suboptimal placement 1.56 (1.00 2.43) 0.048 poor placement 1.44 (1.11 1.85) 0.005 female patient 1.15 (0.65 2.05) 0.623 age (per 10 yrs) 0.56 (0.32 0.96) 0.035 frontal trajectory 0.51 (0.29 0.91) 0.023 position through septum 0.64 (0.23 1.75) 0.384 interhemispheric 1.08 (0.38 3.06) 0.881 tip inside septum 1.04 (0.23 4.69) 0.958 posthemorrhagic etiology 0.48 (0.27 0.84) 0.010 idiopathic communicating etiology 0.38 (0.16 0.93) 0.034 obstructive aqueductal stenosis etiology 0.57 (0.33 0.96) 0.036 * These data were restricted to 2 pediatric neurosurgeons over 3 years after placement. Interpretation is the same as Table 3. J Neurosurg / Volume 120 / March 2014 absence of a definitive optimal target, it is clear that the catheter should be in the ventricle and that our preferred target, within the ventricle and within 1 cm of the foramen of Monro, produces the best results in our institution. While postoperative imaging within 48 hours of surgery was useful in confirming catheter position, and serves a similar role to fluoroscopy in terms of avoiding longerterm failures because patients can be taken back during the same hospitalization if necessary, correction of position requires a second trip to the operating room. We believe that our results support a policy that one should not leave the operating room without proving that the catheter appears to be in the desired position. Intraoperative fluoroscopy has achieved this goal for most of our patients. Similar results can probably be achieved with ultrasonography in children with a patent fontanelle or older children or adults without a patent fontanelle, in whom some advocate making an additional small bur hole adjacent to the entry site to facilitate ultrasonography confirmation. 18 Intraoperative CT or MRI confirmation could also achieve or possibly improve upon these results. In a randomized, prospective multicenter trial, 11 the use of intraoperative endoscopy did not improve accuracy of first-time shunt placement and shunt survival in children when compared with freehand placement using a frontal trajectory. This study has occasionally been used as a justification against the use of any image guidance or confirmation, because it is reasoned that accuracy and outcomes with some kind of guidance or confirmation were no better than freehand placement. However, there were clearly stated limitations to that study, and it cannot be used to generalize about other guidance devices or patient populations. It is important to remember that the endoscopy study did not look at adults, nonvirgin shunt placement (patients whose brains had been penetrated multiple times by shunts in the past), or multiloculated or otherwise complex patients. Using the endoscope, the surgeon s visualized impression of the catheter tip placement often did not agree with postoperative radiographically verified placement. In fact, the proximity to choroid plexus by endoscopy was no different from freehand placement, and was accurate only about two-thirds of the time. The most plausible explanation for this phenomenon is that some degree of bending was applied to the proximal catheter through the cortical mantle so that the tip of the catheter was subject to migration toward the choroid plexus when 693

C. G. Janson et al. TABLE 5: Odds ratios of shunt placement associated with imaging categories and other selected factors* Covariate OR (95% CI) p Value odds for poor placement unadjusted, all patients fluoroscopy imaging (not fluoroscopic) 10.50 (3.08 35.85) <0.001 no imaging 11.32 (4.12 31.12) <0.001 adjusted, all patients fluoroscopy imaging (not fluoroscopic) 5.71 (1.84 17.74) 0.003 no imaging 5.89 (2.36 14.65) <0.001 lack of prior shunt placement 0.66 (0.39 1.11) 0.115 frontal trajectory 0.61 (0.32 1.17) 0.138 odds for optimal placement unadjusted, all patients no imaging imaging (not fluoroscopic) 1.14 (0.61 2.14) 0.675 fluoroscopy 8.14 (4.76 13.91) <0.001 adjusted, all patients no imaging imaging (not fluoroscopic) 1.34 (0.66 2.73) 0.424 fluoroscopy 5.81 (2.64 12.77) <0.001 * When results are adjusted for age, sex, origin of hydrocephalus, surgical trajectory, and individual surgeon (holding the other variables constant), real-time fluoroscopy shows an almost 6-fold decreased odds of poor placement versus no imaging or nonfluoroscopic imaging. Unadjusted values are shown above adjusted values for comparison. The converse was also true, that real-time imaging (and specifically fluoroscopy) improved the odds of optimal placement. It should be noted that approximately 15% of cases used multiple imaging modalities. There was also a favorable trend associated with frontal trajectory and firsttime shunt placements. the stylet or sheath was taken out, at which point the surgeon was blind to this interval movement unless another intraoperative imaging modality was used. That result merely reinforces the need for verification of the catheter tip before leaving the operating room. Our study suggests that accuracy of freehand placement is inferior when compared with placement confirmed by fluoroscopy. Because the combined use of ultrasonography or fluoroscopy together with endoscopy has never been tested, it is possible that using more than one guidance or confirmation device may be superior to just one (or none) in terms of accuracy and outcomes. An analogy is that a pilot flying through fog with instrument guidance does not depend on a single instrument, but rather a combination of GPS, radar, altimetry, and his/her own flying skill or experience. We included in our analysis cases utilizing intraoperative ultrasonography and/or endoscopy, but the number of those cases was too small to allow useful analysis. Currently a randomized, controlled trial of intraoperative ultrasonography for shunt placement is underway, and we agree that nonradiographic techniques should be evaluated for effectiveness in assuring optimal placement of catheters. Fig. 2. Kaplan-Meier plots of revision-free shunt survival according to imaging modality. Upper: This plot shows shunt survival with interval survival estimates and log-rank p values for aggregate data. Unlike the accuracy data, there was a trend for better long-term outcomes with imaging, but the result was not statistically significant. Lower: This plot is similar to the plot above, but shows data from pediatric patients only (age < 19 yrs). Unlike the accuracy data, there was a trend for better long-term outcomes with imaging, but the result was not statistically significant. The main difference from the aggregate data was that the initial interval data in the first several months showed a stronger trend toward early benefit of imaging on positive outcomes. A small prospective cohort study of patients in Europe who received shunts using an exclusively parietal (not frontal) trajectory in combination with intraoperative electromagnetic frameless neuronavigation suggested a benefit for accuracy using electromagnetic targeting, although without an overall benefit in shunt survival. 8 That study was limited by the fact that it enrolled only patients who had never received shunts before and was also a small study with only 75 patients and shunting events. Because those investigators chose not to directly compare frontal versus parietal placement, the issue of optimal surgical approach remains controversial. The main disadvantage of electromagnetic guidance 2,9 is very similar to that of endoscopy, for the following reasons: a rigid trocar is inserted inside the catheter to guide it to its final position, any microadjustments in trajectory that are made as the 694 J Neurosurg / Volume 120 / March 2014