Clinical predictors of shunt response in the diagnosis and treatment of idiopathic normal pressure hydrocephalus: a systematic review and meta-analysis

Background Positive shunt response (SR) remains the gold standard for diagnosing idiopathic normal pressure hydrocephalus (iNPH). However, multiple pathologies mimic iNPH symptoms, making it difficult to select patients who will respond to shunt surgery. Although presenting features, extended lumbar drainage (ELD), infusion test (IT), intracranial pressure monitoring (ICPM), and tap test (TT) have been used to predict SR, uncertainty remains over which diagnostic test to choose. Objective To conduct a systematic review and meta-analysis to identify clinical predictors of shunt responsiveness, evaluate their diagnostic effectiveness, and recommend the most effective diagnostic tests. Methods Embase, MEDLINE, Scopus, PubMed, Google Scholar, and JSTOR were searched for original studies investigating clinical predictors of SR in iNPH patients. Included studies were assessed using the QUADAS-2 tool, and eligible studies were evaluated using univariate and bivariate meta-analyses. Results Thirty-five studies were included. Nine studies discussed the diagnostic use of presenting clinical features, 8 studies ELD, 8 studies IT, 11 studies ICPM, and 6 studies TT. A meta-analysis of 21 eligible studies was conducted for TT, ELD, IT, and ICPM. ICPM yielded the highest diagnostic effectiveness, with diagnostic odds ratio (DOR) = 50.9 and area under curve (AUC) = 0.836. ELD yielded DOR = 27.70 and AUC = 0.753, IT had DOR = 5.70 and AUC = 0.729, and TT scored DOR = 3.86 and AUC = 0.711. Conclusion Intraparenchymal ICPM is statistically the most effective diagnostic test, followed by ELD, IT, and lastly TT. Due to the higher accessibility of TT and IT, they are recommended to be used first line, using a timed-up-and-go improvement ≥ 5.6 s or a Rout cut-off range between 13 and 16 mmHg, respectively. Patients who test negative should ideally be followed up with ICPM, using mean ICP wave amplitude \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\ge$$\end{document}≥ 4 mmHg, or 1- to 4-day ELD with an MMSE cut-off improvement \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\ge$$\end{document}≥ 3. Future research must use standardized methodologies for each diagnostic test and uniform criteria for SR to allow better comparison.


Introduction
Normal pressure hydrocephalus (NPH), a syndrome discovered by Hakim and Adams [1], classically presents with dementia, gait disturbance, and urinary incontinence [23]. Idiopathic NPH (iNPH) is the most common form of adult-onset hydrocephalus, and the current gold standard for diagnosing iNPH is shunt response (SR), which is also the treatment [21,23,37,45,51,65]. Although clinical improvement has been reported in up to 90% of patients following shunt surgery [34], this value has also been as low as 46.7% [5]. For example, Hebb and Cusimano [23] found This article is part of the Topical Collection on CSF Circulation * Santhosh G. Thavarajasingam sgt16@ic.ac.uk that 59% of iNPH patients improved post-shunt, with longterm improvement reported in only 29%. This disparity in reported outcomes reflects the difficulty in selecting suitable patients for shunt surgery as many pathologies mimic iNPH symptoms [41,43,57].
Existing iNPH guidelines, such as the widely cited Japanese iNPH guidelines (2012) [44], a narrative review, outline various clinical tests that can be used in aiding iNPH diagnosis. Alongside Hakim's triad, radiological and biochemical markers, these include the tap test (TT), infusion test (IT), extended lumbar drainage (ELD), and intracranial pressure monitoring (ICPM) [44]. However, the current guidelines [44] do not include explicit diagnostic parameters and cut-off values for each test. Hence, in practice, there is a lack of consistency in the method and evaluation of each test.
The current literature is outdated with regard to the analysis of the diagnostic tests and the presenting factors which may predict SR. In the most recent literature review, Nunn et al. [48] was published last year and investigated the accuracy of ELD. This review [48] included 4 papers of which only one was published after 2003. Furthermore, the last systematic reviews to investigate the diagnosis of iNPH were Relkin et al. (2005) [56] and Hebb and Cusimano (2001) [23]. While both investigated multiple iNPH diagnostic features, there was little comparison within and between diagnostic tests, and no meta-analysis was performed [23,55]. Therefore, to fill this gap in the literature, this review aims to evaluate the diagnostic effectiveness of presenting features, TT, IT, ELD, and ICPM, by incorporating the latest primary research.

Statistical analysis
An Egger's regression and asymmetry test [9] was used to assess publication bias (p < 0.05% = significant). A univariate analysis yielded the log diagnostic odds ratio (logDOR) for each diagnostic test and the DerSimonian-Laird (DSL) summary point, visualized in forest plots. The diagnostic odds ratio (DOR) was calculated by taking the natural logarithm of the logDOR. Cochrane's Q test [9] (p < 0.05% = significant) and Higgins I 2 test [68] (heterogeneity: < 25% = low; 25-50% = moderate; > 50% = high) were used to assess heterogeneity. Finally, a bivariate analysis, which is known to give a slightly more accurate estimation of diagnostic performance than univariate analysis in small sample size analyses, was utilized to plot a summary receiver operating characteristic (SROC) curve. An area under curve (AUC) of 1 indicates perfect diagnostic effectiveness; an AUC ≤ 0.5 indicates an ineffective test. Statistical significance was assumed for p < 0.05. Statistical analysis was carried out by utilizing the mada [11] and meta [60] packages with the R software (version 4.0.4) [55].

Results
The literature search retrieved a total of 7179 papers, of which 359 papers underwent full-text review and 35 studies were included (Fig. 1). The QUADAS-2 tool [69] scored all included studies at low to moderate risk of bias overall, none scoring high risk (Fig. 2a). No significant publication bias was detected by Egger's test (p = 0.0847) (Fig. 2b) [9].

Symptom dominance
McGirt et al. [40] found that patients with gait disturbance as a primary symptom were twice as likely to respond than others, although complete triad presentation compared to the absence of one symptom did not affect SR. This was reinforced by Marmarou et al. [38] and Murakami et al. [46]. However, Meier and Miethke [41] found that the presence of dementia was significantly associated with a worse outcome (p = 0.01).

Symptom duration
Meier and Miethke [41] found that a symptom duration onset of under 12 months before shunt surgery was associated with a significantly better outcome. McGirt et al. [40] noted a 13% lower chance of positive shunt outcome for each additional symptom year before surgery. However, Bådagård et al. [3] and Mahr et al. [37] observed no effect of symptom duration on outcome.

Fig. 1 Preferred Reporting
Items for Systematic Reviews and Meta-Analyses (PRISMA) [42] flowchart outlining the study selection process for qualitative synthesis (systematic review) and quantitative synthesis (meta-analysis)

Symptom severity
Bådagård et al. [3] found a higher modified iNPH scale score, which reflects less severe symptoms, to significantly predict better prognosis. Mahr et al. [37] saw an association between lower baseline Kiefer scores and SR; a cut-off of > 5 had a sensitivity = 58% and specificity = 71%. Meier and Miethke [41] found worsening and severe dementia to be indicators of poor SR. Thomas et al. [63] reported that those with one SD below the age-corrected population average for the following tests to fair 4-6 × worse when shunted: Wechsler memory scale (WMS) immediate recall and WMS immediate recall with Rey-Osterrieth complex figure direct copy or Stroop color word. Mahr et al. [37] found modified Rankin scale (mRS), Stein and Langfitt hydrocephalus score (SLHS), and mini mental state examination (MMSE) scores to be on average higher in responders than non-responders. Using an MMSE cut-off of ≥ 21 was reported to have a specificity = 93% and sensitivity = 67% [37]. Similarly, Kazui et al. [30] found those with a high MMSE total score to be 1.1 × more likely to respond. Overall, the worse the symptomology, the worse the prognosis.

Age
Mahr et al. [37] found the mean age in years of non-shunt responders to be 73.9 compared to 68.8 for excellent responders, while Murakami et al. [46] observed shunt responders mean age to be 75.8 and non-shunt responders 79.9. Bådagård et al. [3] split patients into 3 age groups; in those aged < 70, the response rate was 62% and then 52% and 39% for 70-80 and > 80, respectively. However, Marmarou et al. [38] and McGirt et al. [40] reported no significant relationship between age and SR.

Co-morbidities
Bådagård et al. [3] reported ischemic stroke and transient ischemic attack to be significant negative predictors of SR, while Kazui et al. [30] found patients with hypertension to be twice as likely not to respond to shunting. However, Bådagård et al. [3] reported that hypertension, cardiovascular disease, diabetes, hyperlipidemia, and anti-platelet and anticoagulant use had no significant impact on SR.

Gait tests
Three studies used gait function to assess TT response, namely the timed up and go (TUG) [26,27,71], 10-m walk in time (10Ti) [27], and in step (10St) tests [27]. Using the TUG test, Yamada et al. [73] showed that the absolute difference in time between pre-and post-TT was more accurate in predicting SR than percentage change. Using the absolute time difference, a ≥ 5.6-s threshold within 24 h after TT was reported to have a sensitivity = 83.3% and specificity = 81.0% for predicting an improvement of ≥ 10 s 12-month post-shunt [73]. Ishikawa et al. [26] set  Fig. 2a, risk of bias summary plot with bar chart of the distribution of risk-of-bias judgments for all included studies (n = 35) across the domains of the QUADAS-2 tool [69], shown in percentages (%), is shown. In the bottom, an overall risk of bias, which represents the collated risk-of-bias judgments for all domains, is depicted. The reference standard refers to shunt responsiveness. In Fig. 2b, an Egger's asymmetry test funnel plot [9] of all data points included in the meta-analysis (n = 24; two studies used twice due to discussing two index tests) indicating presence and degree of publication bias is shown. P-value < 0.05 is deemed significant and implicates publication bias. Egger's asymmetry test yielded 0.0847%, calculated running an Egger's regression [9] (see Egger's regression line) on the collated logDOR and standard errors of all data used in the meta-analysis (n = 24)  • In the "improved" group, the pre-STT baseline perfusion values were significantly lower than the "no improvement" group. Significant increases from pre-to post-STT were only found in the "improved" group • STT: sensitivity = 50%, specificity = 50% • TP = 6, FN = 6, FP = 1, the threshold to a 10% improvement in the TUG, yielding sensitivity = 34.2% and specificity = 73.6%. Ishikawa et al. [27] again evaluated the TUG as well as 10Ti and 10St and found the percentage change in TUG on day 1 post-TT at the cut-off value ≥ 11.3% to have the highest sensitivity (78.3%) and specificity (80.0%).

Perfusion studies
In addition to clinical examinations, Hertel et al. [24] and Walter et al. [66] used magnetic resonance imaging (MRI) pre-and post-TT investigating cerebral perfusion and showed that TT diagnostic effectiveness can be improved by MRI use. Notably, Walter et al. [66] found that baseline perfusion values in the shunt responder group were significantly lower than in those who did not improve. Furthermore, they found that only the shunt responder group demonstrated a significant increase in cerebral perfusion from pre-to post-TT [66]. Both studies reported clinical examination to not be significantly associated with SR, unlike cerebral perfusion improvement [24,64].

Infusion test
Eight studies discussed the use of IT in predicting SR, analyzing different aspects of CSF dynamics (

Complementary variables
Besides Rout, the significance of complementary variables was reported by 4 studies [2,15,29,56]. Kahlon et al. [29] concluded that steady-state plateau pressure (PpL) was equal to Rout in terms of predicting SR, with 78% of patients with PpL > 22 mmHg improving. Eide and Brean [15] detected a significant correlation between 12-month shunt outcome and elevated CSF pulse pressure (CSFPP) amplitude of ≥ 4 mmHg measured during lumbar infusion (p < 0.002). Anile et al. [2] found that a cut-off > 0.25 for intracranial elastance (IE) slope gradient clearly differentiated shunt responders and non-responders and was superior to Rout in predicting SR in ventricular infusion test (VIT). Ryding et al. [58] concluded that the volume at plateau level of intracranial venous blood volume (PLIV) decreasing after CSF infusion was positively predictive of shunt outcomes (p < 0.01).

ICPM as assessment of ELD response
Eide and Stanisic [17] reported a correlation of ICPM with ELD, showing that all patients with a raised pulsatile ICP of ≥ 5 mmHg in 10% of recording time were ELD responders (PPV = 100% and NPV = 77%). A reduction in ICP wave amplitude during ELD was a powerful indicator of SR (Spearman correlation − 0.6; p < 0.001) [17].

Memory tests as criteria for ELD response
Gallina et al. [19] showed that a 1-day ELD had a sensitivity = 100% and specificity = 75% for SR when using a ≥ 3-point increase in MMSE for ELD response. This was supported by Chotai et al. [8] who used a 4-day ELD with ≥ 2-point MMSE increase, yielding sensitivity = 100% and specificity = 60%. Similarly, Mahr et al. [37] found that an ELD response, defined as a 10% improvement in gait or

Global symptom improvement as criterion for ELD response
Panagiotopoulos et al. [47] found that a ≥ 1-point increase in NPH score, an objective assessment of symptom severity, after ELD was significantly associated with SR. This finding was supported by Woodworth et al. [71] who reported that ≥ 1 objective symptom improvement predicted SR with sensitivity = 91%, and specificity = 70%. Marmarou et al. [38] showed that when improvement was reported by the patient or family, ELD response was significantly correlated to SR (p < 0.001) and had sensitivity = 95% and specificity 64%.

Static ICPM
Six studies evaluated the use of the static ICP parameter (mean ICP and B-wave activity) as a predictor of SR [13,16,35,49,59,60]. Overall, 4 studies reported no significant correlation between mean ICP or B-wave activity and SR [13,35,59,60]. However, Pfisterer et al. [51] found a significant correlation (p < 0.004) between patients with a preoperative basal ICP level > 10 mmHg and SR. Eide and Sorteberg [16] found the mean ICP to be significantly higher in shunt responders (8 mmHg vs. 7 mmHg), although static ICP overall was a poor predictor of SR.

RAP index
The relationship between static and dynamic ICP parameters was measured in 2 studies using the RAP index (Pearson correlation between mean ICP and mean ICP wave amplitude) [16,35]. Eide and Sorteberg [16] reported that although the RAP average was similar between responders and non-responders, the percentage of RAP ≥ 0.6 or ≥ 0.8 was higher in shunt responders. The use of a RAP threshold of 0.8 yielded sensitivity = 66%, specificity = 48%, PPV = 82%, and NPV = 27% for predicting SR [16]. In line with this, Mahr et al. [37] reported sensitivity = 74% and specificity = 70% for RAP ≥ 0.8.

Meta-analysis
In the meta-analysis, sensitivity and specificity of the bestperforming diagnostic parameters were selected for each included study to facilitate consistent comparisons. Studies were excluded from the meta-analysis in the case of insufficient reporting of statistical findings. A meta-analysis was not conducted for the diagnostic ability of presenting features due to significant methodological heterogeneity. The meta-analysis for TT included 5 studies, and the DSL summary point is at 1.35 (95% CI: − 0.18-2.89) (Fig. 3a,  Fig. 3e) [24,26,64,68,71]. The analysis of IT included 5 studies, and the DSL summary point is at 1.74 (95% CI: 0.75-2.73) (Fig. 3b, Fig. 3f) [4,15,29,35,56]. For ELD, 6 studies were included, and the DSL summary point is at 3.31 (95% CI: 2.19-4.43) (Fig. 3c, Fig. 3g) [8,17,19,36,47,69]. Finally, for ICPM, 7 studies were included, and the DSL summary point is at 3.93 (95% CI: 2.79-5.07) (Fig. 3d,  Fig. 3h) [12, 13, 16-18, 20, 35]. Cochrane's Q test [9] and  Fig. 3d, as well 95% confidence intervals in brackets. At the bottom of each graph, the DerSimonian-Laird (DSL) summary point is shown, which is the summary diagnostic odd ratio (DOR), drawn as a diamond with width inversely proportional to its standard error. A higher DOR implies higher diagnostic effectiveness. Cochrane Q test [9] is shown with p < 0.05% indicating significance, and Higgin's I 2 test [68] with < 25%, 25-50%, > 50% indicating low, moderate, and high heterogeneity, respectively. Study-specific estimates of sensitivity and false-positive rate values are shown in summary receiver operating characteristics (SROC) curve for TT in Fig. 3e, IT in Fig. 3f, ELD in Fig. 3 g, and ICPM in Fig. 3 h. The SROC curve is drawn in black, and the confidence region (95%) is drawn in grey. The data points are symbolized by triangles, and the summary estimate as a circle. The area under the curve (AUC) is shown in the bottom right corner with an AUC of 1 indicating highest diagnostic effectiveness and an AUC of 0.5 or below indicating an ineffective test Higgins' I 2 test [68] indicate low heterogeneity for all subanalyses (Fig. 3).

Discussion
Correctly identifying suitable patients for shunt surgery remains challenging but crucial, as unlike secondary NPH, a large proportion of iNPH patients have poor shunt outcomes. Given that both up to 41% of iNPH patients do not have a positive SR [23], as well as shunting carrying a significant procedural risk of 22-28% [31,70], the need for an accurate diagnostic test to predict SR is apparent. Hence, this review offers guidance as to which clinical tests and patient factors are most effective in predicting SR.
Our meta-analysis of 21 studies concluded that ICPM is the most effective clinical predictor of SR, followed by ELD, IT, and TT. The findings suggest that a patient with shunt-responsive iNPH is 50.1 × more likely to have positive ICPM than a shunt-unresponsive iNPH patient, compared to 27.7 × more likely with ELD. In comparison, a patient with shunt-responsive iNPH is only 5.7 × more likely to have a positive IT and only 3.86 × more likely to have positive TT.
The most predictive indicator of SR is a mean pulse amplitude cut-off of ≥ 4 mmHg, in 70% of recording times in dynamic ICPM [17,19,29]. The meta-analysis is a reliable indicator of ICPM effectiveness as the methodology of the studies included was similar, all using intraparenchymal ICP monitors and assessing similar index parameters, reflected by the relatively small spread of data (Fig. 3h). However, we found several limitations within the included studies. In papers with Eide as principal author, which comprised 5 of the 7 included ICPM studies [12,13,[16][17][18], there were notable differences in patient selection, with Sorteberg et al. [61] selecting patients with triad symptoms and ventriculomegaly for ICPM, while Eide and Stanisic [17] and others used the Relkin guidelines (Table 5) [56]. Moreover, 4 of these studies selected patients retrospectively from the same unit between 2002 and 2008, raising concerns about patient selection bias; additionally some patients may have been excluded from retrospective analysis in one study but included in another [13,[16][17][18]. Ultimately, further research is needed in multicenter settings, with more attention on complications, which Eide neglected. Overall, given the limitations of the included ICPM studies, the findings of the meta-analysis for ICPM must be interpreted with caution. Although TT is thought to simulate a shunt procedure making it an accurate assessment of the effect of a physiological reduction of CSF circulating volume, its high false-negative rates may deprive many patients of the potential benefits of shunt surgery. ICPM could help identify these patients when used as a second-line test given its higher specificity. The mechanism of ICPM's greater accuracy may be due to its ability to uncover CSF pathology, which affects the compliance of brain parenchyma, by direct analysis of its reflection in ICP values. However, as the pathophysiology of iNPH remains unknown [44], the biological mechanism behind the statistical diagnostic superiority of ICPM remains elusive. Regardless of this, a significant advantage of ICPM is that it utilizes objective cut-off values based on monitor readings, and, therefore, is not subject to assessor bias like in some subjective assessments of ELD or TT improvement. ICPM is negatively regarded due to a high perceived risk of complications. However, it must be noted that intraparenchymal ICPM has been found to be much safer than external ventricular drain (EVD)-ICPM: Raboel et al. [54] reported a 5% infection risk for intraparenchymal ICPM versus 27% for EVD-ICPM [22]. In line with this, Vonhoff et al. [64] reported no patients with major complications and only 7% with minor complications, such as accidental removal of the probe, in 152 adults undergoing hydrocephalus ICPM. Although currently less accurate than invasive ICPM, noninvasive ICPM, using transcranial Doppler ultrasonography or MRI, might provide an entirely complication-free assessment of SR in the future [22,54].
ELD is the second most effective test. Despite ELD techniques and response criteria varying significantly between studies, there was relatively low spread in the data (Fig. 3c, Fig. 3g), indicating that ELD is highly accurate regardless of methodological heterogeneity. That said, 3-point MMSE improvement after 1-day ELD had the highest sensitivity and specificity and hence should be given preference [19]. Furthermore, as Chotai et al. [8] found that 1-day ELD is as effective as 4-day ELD, a 1-day ELD should be used to avoid procedural complications and financial costs of a 4-day ELD. However, it must be noted that the study by Chotai et al. [8] had several weaknesses, particularly supplementary tests, such as radionucleotide cisternography and MRI, being used, which hinders drawing valid conclusions regarding ELD as a sole diagnostic test for SR. Interestingly, the Japanese NPH guidelines [44] recommend TT over ELD, citing fewer complications in the TT, while still reporting higher sensitivity and specificity of ELD. However, we only found 3 cases of infections and no other long-term significant complications in 8 ELD studies (n = 425), consistent with Walchenbach et al. [65] who reported only 2 major complications (n = 49) without long-term sequelae. We conclude that the much higher diagnostic effectiveness of ELD relative to TT and IT outweighs the potential complication risk, which has been decreasing in the last decade most likely due to evolving ELD techniques [8].
The financial burden of ICPM and ELD is regarded by many as significant obstacle to clinical implementation. In 2006, Burnett et al. [6] found that an undiagnosed NPH patient would incur mean costs of $108,842. In comparison, it costs $7000 to insert a shunt and $2870 for an ELD, both values signifying the cost of the respective devices only [6]. However, given the low NPV and specificity of ELD, it may incur additional costs by not selecting shunt-responsive patients (Table 4). This could be avoided by using a highly specific diagnostic test, namely ICPM; however, the same authors reported that a hypothetical diagnostic test with specificity = 80% would incur $83,000 per quality-adjusted life year, due to the complexity of prolonged clinical monitoring which may exceed cost-effectiveness thresholds [6]. That said, since 2006, many studies reported ICPM and ELD specificity to be > 80% [13,15,17,20,47]; hence, we recommend an update on the current cost-effectiveness of both tests.
Given the high accessibility of IT and TT, they can be used as first-line diagnostic procedures. Patients who test negatively, but have a high index of clinical suspicion, should be followed up with ICPM, or alternatively with ELD, as these tests have significantly higher specificity and sensitivity. For IT, we recommend using Rout with a cut-off range between 13 and 16 mmHg [4,35,39], as well as CSFPP with a cut-off ≥ 4 mmHg [15] as predictive parameters of SR. These should be used in conjunction with PPPA [29], PpL [29], PLIV [58], and IE [2], for which future research must identify optimal cut-off values. For TT, TUG is reported as the most accurate, specifically an absolute time (≥ 5.6-s improvement) is advised [73], although there was no clear consensus on which parameter is best. The authors also recommend that TT should be followed up with an MRI assessing cerebral perfusion post-TT, to improve its sensitivity [24,64]. The patient's age, symptom severity, and co-morbidities all influence SR outcomes and should be considered in patient selection [3,35,39,44,61]. The longer symptoms are present before surgery, the worse the prognosis is [38,39]; hence, early identification and treatment are critical.
In 2001, Hebb and Cusimano [23] stated that more prospective studies, which shunt iNPH patients regardless of their diagnostic test results, were needed to yield precise overall sensitivities and specificities of each test. Twenty years later, we found that this remains the most significant limitation of the literature. Given that in many studies the results of the diagnostic test influence the decision to shunt, accurate false-negative values are difficult to ascertain. We emphasize the importance that future research must aim to shunt all patients to allow for highly valid conclusions on diagnostic efficiency to be derived.
Radiology is also pivotal in the diagnosis of iNPH; however, until recently, studies have not been able to show an association between radiological features and SR [33]. The iNPH Radscale may potentially play an important role in SR prediction and is a promising new tool in the iNPH patient workup [32]. However, due to its relatively recent introduction in 2017, there is currently insufficient literature evaluating its use. Although Kockum et al. [32] have reported 98.5% predictive accuracy of NPH Radscale, another study found it to be unable to predict motor outcomes post-shunting [35]. Nonetheless, the literature on the role of radiology in iNPH is extensive; hence, the authors believe that a separate meta-analysis on the role of radiological tools in the prediction of SR is required. Additionally, biochemical markers have been hypothesized to be valuable in the diagnosis of iNPH. As reported by Leinonen et al. [36], the most important biomarker indicating iNPH may be abnormally low CSF TNF-α concentrations. Nevertheless, this study, as well as a systematic review by Pfanner et al. [50] recently in 2017, reported that no biomarker was able to predict SR. However, Zhang et al. [74] in 2020 reported that studies had found that high concentrations of vascular endothelial growth factor in the CSF may be associated with worse SR. A meta-analysis of both radiology and biochemical markers, together with this meta-analysis, can give more extensive guidance regarding the clinical prediction of SR.

Limitations
The key limitations of this study were the differences in SR criteria and diagnostic test methodologies between studies. Employing a meta-regression on these independent variables might have increased the validity of our findings. The lack of methodological consistency, as well as not all patients receiving the reference standard (shunt), is reflected in the risk analysis (Fig. 2a) which showed a moderate overall risk of bias for almost half of the studies. The number of studies used in each sub-analysis was moderate (n = 5-7), including more studies may have decreased the risk of sample size bias. Finally, the true diagnostic effectiveness of all diagnostic tests may be higher than reported by the meta-analysis, as many studies either used a medium-pressure valve or did not report using a programmable valve, despite pressure adjustments often necessary postoperatively to avoid over-and underdrainage [7,28,36,46,47,69].

Conclusion
Intraparenchymal ICPM is statistically the most effective diagnostic test, followed by ELD, IT, and lastly TT. The best clinical predictors of SR were found to be a mean ICP wave amplitude ≥ 4 mmHg and secondly using a 1-to 4-day ELD with an MMSE cut-off improvement ≥ 3. If IT is used, Rout with a cut-off range between 13 and 16 mmHg, in conjunction with a CSFPP cut-off of ≥ 4 mmHg, should be used. Additionally, for TT, a TUG improvement of ≥ 5.6 s can be used. Despite the statistical superiority of ICPM and ELD, the financial cost associated with their use as well as the potential for complications renders it most useful in identifying SR patients with negative IT or TT results. The latter are clinically easier to perform with low complication rates and hence may be used as firstline diagnostic tests to predict SR. When used in addition to diagnostic tests, the severity of symptoms, patient age, and co-morbidities may aid in predicting SR. In the future, standardized methodologies for each diagnostic test and uniform criteria for SR must become the norm to draw better comparisons.