Introduction

Cerebrospinal fluid (CSF) shunts allow children with hydrocephalus, a common cause of neurological disability [1, 2], to survive and avoid ongoing brain injury. A recent meta-analysis demonstrated a pooled incidence of congenital hydrocephalus in the United States at 68 per 100,000 live births [3], with nearly 400,000 new cases of hydrocephalus globally per year [3, 4] CSF shunt placement has been the mainstay of hydrocephalus treatment for over 60 years [5]. However, CSF shunts are associated with repeated revision surgeries and risk of infection [5]. Mechanical malfunction is frequent, and 60% of shunts require surgical revision within 4 years [6,7,8]. In the United States there are approximately 20,000 pediatric CSF shunt surgeries annually [9]. With each CSF shunt surgery, the risk of shunt infection increases [10, 11], and in the United States there are approximately 2,000 pediatric CSF shunt infections per year [9]. The burden to children, families, and the healthcare system of CSF shunt infections in terms of costs [9], morbidity over the life span [12], and quality of life [13] are substantial and preventable [14].

Controversies have emerged in the field of hydrocephalus about optimal peri-operative techniques to prevent CSF shunt infections in addition to the now-standard use of prophylactic IV antibiotics [11]. The BASICS trial demonstrated that antibiotic impregnated catheters (AIC) had lower infection rates as compared to standard shunt catheters, while silver-impregnated catheters did not, in a mixed population of children and adults in the United Kingdom [15]. The Hydrocephalus Clinical Research Network (HCRN) has instituted multiple quality improvement protocols that have been shown to decrease infection rates compared to pre-operative prophylactic antibiotic administration: initially with intrathecal (IT) antibiotics between 2007–2009 [16], followed by its replacement with AIC in 2012–2013 in North America [17]. Despite the benefit of AICs suggested in the BASICS trial, there have been few large scale studies directly comparing different infection prevention techniques that have shown to be superior to standard shunt catheters in children and none in a low-risk population [18, 19]. The aim of this study was to compare the odds of infection following the use of IT antibiotics, AIC, and standard care during low-risk CSF shunt surgery in children using large scale, multi-center administrative data augmented with clinical data.

Design/methods

Study design and setting

This was a retrospective cohort study conducted at 6 large pediatric neurosurgical practices at tertiary care children’s hospitals (Boston Children’s Hospital, Children’s Hospital of Philadelphia, Children’s Hospital of Pittsburgh, Cincinnati Children’s Hospital Medical Center, Primary Children’s Hospital, and Seattle Children’s Hospital) between 2007 and 2015 [20]. These hospitals were selected due to their inclusion in the Pediatric Health Information System + (PHIS + ; Children’s Hospital Association, Lenexa, KS) database that includes detailed administrative, laboratory, microbiology, and radiology data for all children receiving care at participating centers [20]. The hospital names were blinded for the presentation of the results; Hospitals A, D, E and F were all HCRN sites [20].

Study population

The study population included children ≤ 18 years of age who underwent initial shunt placement between January 1, 2007 and December 31, 2012 at one of the six study sites [20]. During the screening phase, medical records for 5,903 children and 11,121 shunt procedures were abstracted from PHIS based on evidence of initial shunt placement or shunt revision between January 1, 2007 and December 31, 2012. Initial CSF shunt placements were defined as admissions with any International Classification of Diseases, Ninth Revision, Clinical Modification procedure code for extracranial ventricular shunt placement (02.3–02.35 except 02.39 alone), excluding those with any concurrent procedure code for replacement (02.42) or removal of ventricular shunt (02.43), and/or any diagnosis code for shunt malfunction (996.2), and/or shunt infection (996.63) [20]. CSF shunt revisions were defined as admissions with a primary diagnosis code for shunt malfunction (996.2) excluding those with concurrent CSF shunt infection (996.63) [20]. Dates of initial shunt placement and any subsequent revision surgeries were also abstracted [20]. Medical records were screened by trained study staff at each participating site to confirm that each initial shunt placement identified through PHIS screening represented that child’s true initial placement and that we were able to capture details of any preceding neurosurgical procedures [20]. Surgical procedure data including all initial CSF shunt placements, CSF shunt revisions, and first CSF shunt infections was collected for each eligible child through December 31, 2015, allowing each child at least three years of follow up time since the initial shunt placement [20].

Data sources

The PHIS + database was augmented with detailed clinical data obtained from chart review to create a database with over 3,000 CSF shunt surgeries for the investigation of CSF shunt infection prevention. This approach permitted us both to confirm critical variables (e.g. use of IT antibiotics and AIC) and to obtain additional variables unavailable in PHIS + (e.g. surgical decisions in the operating room).

All site investigators participated in a group consensus process to determine which additional variables were feasible and accurate to collect in chart review. We used Research Electronic Data Capture (REDCap), a secure web-based application for electronic data capture, to ensure consistent chart review data collection across sites [21, 22]. Data obtained through chart review were matched to PHIS + data using hospital, medical record number, and date of surgery.

A comprehensive data quality assurance plan, explained in detail in Podkovik et al., was implemented to ensure that data collected from PHIS and PHIS + adhered to internally consistent definitions and accurately reflected clinical course and outcomes [20].

Outcome variables

The outcome of CSF shunt infection was defined adopting the widely-used HCRN consensus definition of CSF shunt infection, which [11, 16, 17] is either 1) microbiological determination of presence of bacteria in culture of CSF, wound swab, and/or pseudocyst fluid; or 2) shunt erosion (visible hardware); or 3) abdominal pseudocyst (even without positive culture) [11, 16, 17]. The primary outcome was infection within 6 months from the most recent surgery. Subjects were censored at the time of their first infection or at the conclusion of the observation period, whichever came first.

Secondary outcomes included length of hospital stay (days) and rates of post-operative complications: bacteremia, CSF leak, pseudomeningocele, meningitis, need for antibiotic treatment for wound site, bowel perforation and other complications.

Predictor variables

We took advantage of the natural experiment that occurred in PHIS + hospitals from 2007 to 2012 with the use of IT antibiotics and AIC. During the study period, most patients received standard care, defined as receiving prophylactic IV antibiotics (either cefazolin or vancomycin) without IT antibiotics and having conventional shunt tubing.

IT antibiotics were defined by an appropriate antibiotic (e.g., vancomycin, gentamicin) with an appropriate intrathecal dose (i.e., 0–10 mg). Corroborating information from the operative report and/or surgeon survey were likewise evaluated.

AIC use was determined by documentation from the operative report. Corroborating information from the operative report and/or surgeon survey were likewise evaluated.

All outcomes were associated with the technique used in the preceding CSF shunt surgery. Because a given patient may undergo multiple CSF shunt surgeries for which different infection prevention techniques might be used, the predictor variables are time-varying in the analysis.

Statistical analysis

For descriptive statistics we reported means and standard deviations for the continuous variables. For categorical variables we reported counts, proportions and 95% confidence intervals.

Due to the observational nature of our study design, we performed propensity score analyses with inverse probability treatment weighting to estimate the relationship between prevention techniques (standard technique, intrathecal antibiotics, and antibiotic impregnated catheter) and shunt infection within 6 months of shunt placement. The propensity score, defined as the conditional probability of receiving treatment given covariates, plays a central role in causal inference. Under certain assumptions, an unbiased estimate of the average treatment effect can be obtained by adjusting for the propensity score alone rather than a vector of covariates, which is often of high dimension [23].

For our primary analysis, we first applied the covariate balancing propensity score (CBPS) methodology [24] to model the probability of shunt infection prevention techniques while optimizing the covariates balance [20]. The CBPS takes advantage of the dual characteristics of the propensity score as a covariate balancing score and the conditional probability of treatment assignment. This method does two things simultaneously: it 1) balances covariates and 2) optimizes predicted probability of treatment given covariates. CBPS has been extended to more than two treatment options. We estimated CBPS for initial shunt placement and revision placement separately. A list of predetermined covariates based on previous research [10, 11, 25,26,27,28] were included in the CBPS models: patient age, patient biological sex, patient race, primary insurance, patient weight, weekday or weekend of the procedure, complex chronic conditions, admission priority, etiology of patient hydrocephalus, concurrent non-neurosurgical procedure, concurrent neurosurgical procedure, prior CSF leak, prior gastrostomy, prior inpatient antibiotics, prior CNS surgeries, prior non-CNS surgeries, and prior tracheostomy. The CBPS and weights were calculated using [29] and [30] packages in R (R Core Team, 2022) [31]. Covariate balance between infection prevention techniques in the propensity score weighted sample was assessed by balance tables and density plots, using the {cobalt} [32] package in R. After we derived CBPS weights, inverse probability treatment weighting was applied to logistic regression models, with infection within 6 months as outcomes, and infection prevention techniques as the sole predictor. We reported adjusted odds ratios (aORs) and 95% confidence intervals. For our secondary analyses, continuous outcome variables were compared using a Kruskal–Wallis rank sum test, and binary outcomes were compared using Fisher’s Exact Test. All analyses were conducted in R statistical software (R Core Team, 2022) version 4.2.2.

Role of the funding source

The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.

Results

From a total of 5,903 unique patients within the PHIS + data set, 1,723 patients had an initial shunt placed amongst six PHIS + pediatric hospitals between January 1, 2007 and December 31, 2012. These children experienced 3,094 initial shunt placements and shunt revisions prior to development of first CSF shunt infection or censoring at the end of the observation period, December 31, 2015. Patient-level demographics at the time of the initial shunt placement are provided in Table 1.

Table 1 Patient-level characteristics for the overall cohort and in association with CSF shunt infection within 6 months
Full size table

There were 138 shunt infections identified within 6 months of the antecedent surgery. Table 1. provides a bivariate analysis between patient level characteristics and CSF shunt infections. The only patient-level factor that differed between children who developed CSF shunt infection and those who did not was etiology of hydrocephalus. Table 2 provides a bivariate analysis between procedure-level characteristics and CSF shunt infections. There were significant differences between procedures with and without infection in procedure type, age at surgery, weight at surgery, year of shunt surgery, and use of antibiotic impregnated sutures. There was a larger proportion of infections following initial placements compared to following shunt revisions. Patients with shunt infections tended to be younger (2.25 ± 4.61 years vs 3.03 ± 4.61 years) and lower weight at surgery (13.41 ± 18.24 kg vs 15.45 ± 17.29 kg). Antibiotic impregnated sutures were associated with infections.

Table 2 Procedure-level characteristics for the overall cohort and in association with CSF shunt infection within 6 months
Full size table

We compared infection rates between the three shunt prevention techniques (Table 3). The overall 6-month infection rate of shunt placements (both initial and revision) was 4.5% [95% CI: 3.8,5.3], with no significant differences observed between infection prevention techniques (Fig. 1). Adjusted odds ratios generated from CBPS are also presented in Table 3. Among all procedures, compared to standard care, IT antibiotics had an aOR of 1.4, [95% CI: 0.7, 2.7], p = 0.4 and AICs had an aOR of 0.7, [95% CI: 0.5, 1.2], p = 0.2. None of the shunt infection prevention techniques showed a significant independent association with infection at 6 month when separated by initial versus revision placements.

Table 3 6 month risk of infection overall and by infection prevention technique and propensity score adjusted odds ratios (aOR) by infection prevention technique
Full size table
Fig. 1
figure 1

Observed rate of 6 month risk of infection (%, 95% confidence intervals) by infection prevention technique

Full size image

Table 4 reports secondary outcomes and post-operative complications within seven days of surgery. There was no significant difference in hospital length of stay between the infection prevention techniques. There were no significant differences in any other complication rates amongst the procedures except for the presence of a post-operative pseudomeningocele (1.5% in standard care group compared to 0.1% for both IT antibiotics and AICs) and other complications (11% for both IT antibiotics and AICs compared to other groups).

Table 4 Secondary outcomes by infection prevention technique
Full size table

Discussion

We took advantage of the natural experiment that occurred in PHIS + hospitals from 2007 to 2012 with the use of IT antibiotics and AIC to compare these techniques to standard care in the cohort of children undergoing initial CSF shunt placement and CSF shunt revisions. In this retrospective analysis of over 3,000 low-risk surgeries at six institutions between 2007 through 2015, there were no differences in 6-month infection rates between standard care, IT antibiotics and AICs. AICs tended to have a favorable odds ratio compared to standard care and IT antibiotics tended to have an unfavorable odds ratio compared to standard care; however, no significant differences were observed between the techniques. This was observed both when evaluating all procedures combined and then both initial and revision placements separately.

The HCRN has implemented multiple peri-operative infection prevention protocols over the last 15 years. In 2007, the HCRN protocol recommended that surgeons utilize a one-time instillation of IT antibiotics, consisting of vancomycin and gentamicin, for all shunt surgeries in addition to pre-operative intravenous antibiotics [16]. A 2011 study demonstrated a reduction in infection rates from 8.8% to 5.7% (p = 0.003) following the implementation of the IT antibiotic protocol [16]. Subsequently there was increasing adoption and research into the efficacy and utility of AICs [33,34,35,36,37,38,39,40,41,42] coated with rifampin and clindamycin [43]. The subsequent HCRN protocol replaced the use of IT antibiotics with AICs [17]. A subsequent 2016 study showed a similar infection rate of 6.0% (p = 0.002) following the protocol replacing IT antibiotics with AIC [17]. Our recent study reviewed utilization trends of the three infection prevention techniques in six PHIS + hospitals and demonstrated that AIC use increased and IT antibiotic use decreased during the study period, except for Hospital B which consistently used AICs [20].

Our unadjusted 6 month infection rates across all techniques were 4.5% for standard care, 5.2% for IT antibiotics, and 3.7% for AICs. These rates are lower than the HCRN cohorts. Most previous studies evaluating IT antibiotics and AICs incorporated all children who received a shunt surgery prior to enrollment, which included children presenting with a previous shunt infection. Our cohort design allowed us to have complete shunt history and thus minimize variation in infection risk. Hence lower infection rates were observed due to the inherently lower risk patient population within our study.

A 2012 study by Simon et al. evaluated 1000 children undergoing shunt placements, and after controlling for baseline factors, it was noted that infection risk was most significantly associated with the need for revision [11]. In this and multiple other studies, it was concluded that relatively few patient, medical, or surgical risk factors – other than revision surgery itself—were associated with first infection [10, 11, 26, 44] Paradoxically, our cohort demonstrates a higher percentage of infections in the initial placement compared to the revision placements. Of note, we observed a decrease in the number of overall infections following the year 2012. This is explained by the fact that no new children were enrolled in the subject pool following this year, but infection events were continued to be monitored for the existing study population.

Since this is a retrospective cohort study, we measured the association between techniques and infection risk, rather than causality. It might be argued that a clinical trial is optimal, however, the use of a large database permits us to efficiently capitalize upon the existence of detailed data on large numbers of CSF shunt surgeries (far larger cohorts than previously assembled) and allowed us to examine a wider spectrum of children. We were also able to use sophisticated analytic approaches to optimize predicted probability of treatment given practice variation we observed in earlier work [20]. This study provides relevant information about the newest CSF shunt infection prevention technique in use today, AIC, and suggests limited benefit in a low-risk population. There is a relatively small number of surgeons and hospitals that limit our ability to study surgeon and hospital effects on patient outcome systematically; however, the multi-institutional nature of this study gives it greater generalizability than previous studies. Similarly, while minority children are under-represented in these data, its multi-institutional nature provides greater generalizability than previous studies. Because we observed much lower infections across all three techniques than previously reported, using standard care as the reference group, our sample sizes provided only 12% power to detect the difference between intrathecal vs. standard, and 13% power to detect the difference between AIC versus standard care. Therefore, because of the relatively rare occurrence of infection events, even with our large multi-year sample we were under-powered for most comparisons. Despite this limitation, this real-world evidence provides little support for routine use of IT antibiotics or AICs compared to standard care in low-risk CSF shunt surgeries.

Conclusion

We did not observe a difference in 6 month infection rates or adjusted odds of infection between AIC or IT compared to standard care for children undergoing initial CSF shunt placement and CSF shunt revisions. Compared to previous studies, the benefit provided by AICs and IT compared to standard care may not be as large as previously believed amongst low-risk patients once cohorts are appropriately balanced. The real-world benefit of AIC among low-risk patients should be evaluated carefully using current data given interim changes in surgical practice and their widespread adoption.