The role of amyloid PET in patient selection for extra-ventricular shunt insertion for the treatment of idiopathic normal pressure hydrocephalus: A pooled analysis.

BACKGROUND
Idiopathic Normal Pressure Hydrocephalus (iNPH) can be effectively treated through shunt insertion. However, most shunted patients experience little or no clinical benefit, which suggests suboptimal patient selection. While contentious, multiple studies have reported poorer shunt outcomes associated with concomitant Alzheimer's disease. Prompted by this observation, multiple studies have assessed the role of amyloid PET, a specific test for Alzheimer's disease, in patient selection for shunting.


METHODS
A comprehensive literature search was performed to identify studies that assessed the association between amyloid PET result and the clinical response to shunting in patients with suspected iNPH. Pooled diagnostic statistics were calculated.


RESULTS
Across three relevant studies, a total of 38 patients with suspected iNPH underwent amyloid PET imaging and shunt insertion. Twenty-three patients had a positive clinical response to shunting. 18/28 (64.3%) of patients with a negative amyloid PET and 5/10 (50%) with a positive amyloid PET had a positive response to shunting. The pooled sensitivity, specificity and accuracy was 33.3%, 76.2% and 58.3%. None of these statistics reached statistical significance.


CONCLUSION
The results of this pooled analysis do not support the selection of patients with suspected iNPH for shunting on the basis of amyloid PET alone. However, due to small cohort sizes and weakness in study design, further high-quality studies are required to properly determine the role of amyloid PET in assessing this complex patient group.

Unlike most other causes of dementia which have no disease modifying treatment, iNPH can improve with CSF diversion [1]. Accurate patient selection is important due to the risk of serious complications. Haemorrhage, infection, over-drainage, obstruction and device failure occurs in 38% of patients [8,9].
Despite ongoing re nement of the iNPH diagnostic guidelines [11][12][13][14][15] and imaging criteria [16], patient selection is suboptimal. As few as 28% of patients improve after shunting mainly due to the overlap in the clinical and imaging features of iNPH with other conditions, such as Alzheimer's disease (AD) and progressive supranuclear palsy [17], that are not treated with shunting.
The implications of AD pathology based on cortical biopsy or CSF analysis is contentious. Evidence of comorbid AD has been associated with poorer [18][19][20][21][22], similar [23,24] and better [25] outcomes following CSF drainage. The discrepancy between these studies may be related to the biopsy of unrepresentative cerebral cortex or the reduced reliability of CSF analysis in patients with iNPH where toxic metabolic clearance is impaired [26].
Working on the hypothesis that amyloid deposition is associated with AD, which is associated with a poor response to shunting, multiple studies have examined the association between amyloid PET and clinical outcomes following shunt insertion in patients with suspected iNPH. In this article we present a pooled analysis of their results.

Inclusion and exclusion criteria
Studies were included if they reported the association between amyloid PET imaging ndings and the response to a shunt in patients with suspected iNPH.
Studies were excluded if amyloid imaging was performed after shunting, if the decision to shunt was based on the amyloid PET result or if individual patient outcomes were not reported.

Data extraction
Data extraction was performed independently by two readers with adjudication by a third, if necessary.

Statistical analysis
Results from individual studies were combined to pool estimates of sensitivity, speci city, negative predictive value, positive predictive value, and accuracy.
Diagnostic statistics are presented as a percentage alongside the 95% con dence interval calculated using the Clopper-Pearson method. Numerical data are presented as a mean with the standard deviation, unless otherwise speci ed. Statistics were calculated using the statsmodels (version 0.12) package in Python 3.7.

Results
Three studies were identi ed that compared the clinical response to shunting between those with and without a positive amyloid PET scan [46][47][48].
Two further studies of amyloid PET performed before shunting were found but were excluded as the response to shunting was not reported [49,50].
Additionally, both studies included patients from the same group and registry as one of the three studies that was already included. No detail was provided on any overlap in patient cohorts between these studies.
In a sixth study by Jang et al [51], the primary outcome was the association between CSF tap test and amyloid PET results. As only patients with a negative amyloid PET were offered a shunt, the study was excluded.
Details of the three studies that were included in the pooled analysis are provided in Table 1. The association between amyloid PET imaging and shunt response was the primary study outcome in a study by Hiraoka et al, whereas it was a secondary outcome in the other studies. Patients were recruited between approximately 2007 and 2016. In the study by Rinne et al, the recruitment period was not provided. Across all studies, 38 patients were shunted.
The technical detail of PET imaging including the choice of radiotracer is provided in Table 2. 18 F-utemetamol was used in two studies and 11 C-BF227 in one study.
Of the 38 patients shunted and followed-up, 23 had a positive clinical response. 18/28 (64.3%) and 5/10 (50%) of patients with a negative amyloid PET and positive amyloid PET had a positive response to shunting, respectively. The Fisher's exact test P value for the association between shunt response and amyloid PET result was 0.473.
The diagnostic statistics for the prediction of a shunt response in those with a negative amyloid PET is provided in Table 3. Sensitivity, speci city and accuracy ranged from 0-60.0%, 60.0-87.5% and 52.9-72.7%, respectively. The respective pooled sensitivity, speci city and accuracy was 33.3%, 78.3% and 60.5%.
All three studies reported the SUVR (Standardised Uptake Value Ratio) for each patient, which are provided in Supplementary Table 1. The box and whisker plot in Figure 2 shows that SUVRs are generally higher in those who did not respond to shunting than those who did respond to shunting, although, neither individually nor when pooled, was the difference statistically signi cant.
Amyloid PET was well tolerated with only two cases of minor adverse reactions (nausea) reported by Rinne  Discussion iNPH can be effectively treated through shunt insertion [52]. However, as many patients who undergo this procedure experience short-lived or no bene t from the procedure, patient selection is suboptimal. Prompted by reports of better shunt outcomes in those without evidence of AD [20,53,54], multiple studies have sought to determine how amyloid PET could improve patient selection. We present a pooled analysis of three non-randomised studies that have examined the association between amyloid PET and the clinical response to shunt insertion.

Association between amyloid PET and shunt response
The pooled results of the three studies shows low sensitivity (33.3%) and a moderate speci city (78.3%) in identifying patients with a poor response to shunting. Proportionally more patients who responded to a shunt had a negative amyloid PET than a positive amyloid PET (61.5% versus 50%), although the difference was not statistically signi cant. Therefore, the main nding in this pooled analysis is that amyloid PET, when considered in isolation, does not accurately identify patients with suspected iNPH who are likely to respond to shunting.

Weaknesses of included studies
There are multiple inherent weaknesses within the studies included in this pooled analysis.
All of the studies were non-randomised and therefore inherit all of the limitations of such a study design. Furthermore, assessors of the clinical response to shunting were not blinded to the result of the amyloid PET, representing a potential source of bias.
All studies were small, yielding a total of only 38 patients. Therefore, each study was only powered to identify large effects. As such, differential effects on each component of the iNPH clinical triad could not be assessed. Similarly, in such small cohorts, it was not possible to control for other factors associated with shunt responsiveness, such as age [11,55], duration of symptoms [11,[56][57][58], gait disturbance predominance over cognitive impairment [10,11,57,58], the presence of co-morbidities [55,59], and the structural imaging features of iNPH and AD.
In the study by Leinonen et al, amyloid imaging was performed between 9 and 38 months after the biopsy and shunt insertion. This is problematic because the effects of shunting on amyloid PET is unknown. Secondly, since amyloid deposition is progressive, the amyloid PET is likely to be an overestimate of the amyloid burden at the time the decision was taken to insert a shunt. Lastly, a signi cant interval between shunting and imaging may cause selection bias based on disease severity or, more signi cantly, the response to shunting.
In addition to these inherent weaknesses of the studies included in this pooled analysis, there was also signi cant variation between the studies in the clinical, radiological and non-radiological investigations performed, which has implications for pooling results.
Recruitment by Leinonen et al required only "enlarged ventricles" and any one of the iNPH clinical triad, whereas the other two studies also required sulcal effacement and two of the iNPH clinical triad. Leinonen et al performed 24-hour intracranial pressure (ICP) monitoring in all patients before consideration of shunt insertion; in this study, all but one patient responded to shunting. Rinne et al used ICP monitoring "if required", although speci c indications were not discussed. Hiraoka et al performed CSF tap testing. As inclusion criteria become less stringent, the pre-test probability for iNPH decreases and so does the expected response rate to shunting.
The method of assessing the clinical response to shunting was variable. Methods included the iNPHGS [60], the Black score [61] or a non-disclosed "clinical assessment". Different methods are likely to in uence the reported rate of shunt response, which has obvious implications when pooling results. Furthermore, the timing of the assessment ranged from 2 months to over 20 months. Clinical improvement may occur over many months, which is re ected in the ongoing changes in the parenchyma and CSF spares through the rst postoperative year [62]. On the other hand, clinical improvements can be eeting [63][64][65] and therefore follow-up of at least one year is often used in clinical trials [66]. In two of the studies, assessment was performed at three months or earlier. The value of a clinical improvement lasting three months is debatable, and, in many centres, this would be considered unsatisfactory.
Two different radiotracers were used in the three studies. Multiple radiotracers have been developed with differing pharmacokinetics although all are highly speci c for Aβ [67] and correlation between radiotracer uptake is high [68][69][70]. Therefore, the diagnostic accuracy of different amyloid-speci c radiotracers, and hence their role in patient selection for shunt insertion, is likely to be comparable.

Reasons for lack of association between amyloid PET and shunt response
Firstly, amyloid PET may accumulate in the brain for reasons that have no effect on shunt response. While Aβ deposition, alongside hyperphosphorylated tau protein, is a histopathological hallmark of AD [71], it is also observed in other neurodegenerative conditions such as Lewy Body Dementia [72,73] and cerebral amyloid angiopathy [74]. Amyloid deposition also occurs in healthy aging [75][76][77], albeit with an increased risk of subsequent cognitive decline [78,79]. Secondly, the potential for improvement post-shunting will depend on the degree of permanent neurological damage caused by NPH or any other comorbid condition [80]. Thirdly, a recent study showed CSF biomarkers of AD were associated with a positive response to tap-test [25]. This observation raises the possibility of a so-called neurodegenerative NPH (that would be amyloid PET positive) that is separate from an idiopathic NPH. Despite the result of this pooled analysis, amyloid PET may still have a role in the wider workup of patients for shunt insertion, particularly in identifying dual pathology. Even if amyloid PET does not predict immediate shunt outcomes, cerebral amyloid is likely to in uence long-term prognosis, which would be relevant when counselling patients on the expected risks and bene ts of shunt insertion.

Requirements for future study
There remains a need for a blinded and randomized study to de nitively determine the role of qualitative and quantitative interpretation of amyloid PET in predicting shunt response in patients with suspected iNPH. The study should be adequately powered to assess for differential effects of each domain of iNPH and to control for disease severity and other clinical features associated with long-term shunt outcomes. The predictive power of amyloid PET should be compared with the features apparent on structural imaging and biomarkers from CSF analysis. Finally, adequate follow-up of at least a year is required due to the progressive nature of AD and the often-transient improvements experienced following shunting.

Conclusion
This pooled analysis does not support the use of amyloid PET in the selection of patients with suspected iNPH for shunting. Amyloid PET may however remain a useful adjunct in the workup of these patients, speci cally for informing longer term prognosis. There remains the need for higher quality prospective studies to more conclusively evaluate the role of amyloid PET in this complex patient group.    Figure 1 Imaging features of normal pressure hydrocephalus and Alzheimer's disease. In a 70-year-old patient who presenting with cognitive decline, the coronal T1weighted imaging (A) and axial T2-weighted MR imaging (B), suggested normal pressure hydrocephalus due to ventriculomegaly and crowding near the vertex. However, the 18F-orbetapir PET imaging (C) showed generalised increased grey matter uptake with loss of grey-white matter differentiation, particularly within the bilateral posterior parietal cortices. This signi ed extensive amyloid deposition prompting a diagnosis of Alzheimer's disease.

Figures
Conversely, in a 67-year-old patient also with ventriculomegaly on MR imaging (D, E), the 18F-orbetapir PET imaging showed a normal pattern of uptake and preserved grey-white matter differentiation indicating no signi cant amyloid deposition. This patient had a signi cant and sustained response to shunting.