The role of magnetic resonance imaging in the identification of suspected acoustic neuroma: a systematic review of clinical and cost effectiveness and natural history

Audiological tests

Patients typically present with one or more audiovestibular symptoms that result in referral to an otolaryngology clinic. If a history and physical examination suggest the possibility of an acoustic neuroma, an audiological examination will be requested. This will generally consist of pure tone and speech recognition audiometry, although different criteria are used by different departments in defining what constitutes an abnormal result warranting further investigation. Historically these tests proved to be of limited value, as they often gave a mixed picture11 and did not in fact diagnose acoustic neuroma; rather the likelihood of this pathology was increased in the clinician’s mind. More recent work has indicated that the mixed picture was the result of the acoustic neuroma potentially causing both cochlear hearing impairment because of ischaemia or biochemical degradation21 and retrocochlear hearing impairment by nerve compression. Recent work has indicated that measures of audiological handicap may have utility in the management of patients with acoustic neuroma. 22 A brief summary of the role of audiological investigation in the identification of acoustic neuroma is presented in Chapter 2 (see Comparison of the use of auditory brainstem response with the use of magnetic resonance imaging).

Electrophysiological tests

Electrodiagnostic recordings of the ABR enable the investigation of the functioning of the peripheral auditory nerve and the auditory brainstem pathways. 23,24 For ABR (or other physiological tests such as caloric testing, speech audiometry, pure tone audiometry) to indicate a retrocochlear lesion that might be tumour, the tumour has to exert some effect (most commonly thought to be physical pressure) on the neurological structures involved. An acoustic neuroma usually has the effect of slowing the speed of propagation of the action potentials as they progress through the ascending auditory pathway or causing dys-synchrony of the potentials. These two mechanisms lead to a delayed latency response or an abolished response respectively. However, it is important to remember that an abolished response can be the result of a severe/profound cochlear impairment as well as the result of an acoustic neuroma. The ABR has been studied extensively over the last 20 years with respect to its diagnostic capability, it’s relationship to other diagnostic tests, its effect on the management of patients,25,26 including comparative roles of ABR and other diagnostic investigations,12,27,28 it’s limitations in the identification of small tumours,29,30 and potential techniques to enhance it’s diagnostic power. 31,32 ABR can also provide additional information such as the degree of brainstem compression by measurement of the ABR on the non-tumour side. 11,33 This may be monitored postoperatively as an index of recovery.

Chapter 2 (see Comparison of the use of ABR with the use of MR imaging) summarises the published data to evaluate the role of ABR with respect to other diagnostic tests in the screening and diagnosis of patients with suspected acoustic neuroma. When evaluating the role of ABR compared with other technologies such as MR imaging, it should not be overlooked that there are some patients for whom MR imaging is not appropriate and for whom some alternative strategy is needed to cater for them. ABR or CT may represent an option for the majority of such patients. 34–36

Imaging tests

MR imaging detects tumours on the basis of abnormal anatomy and is a structural investigation, whereas ABR relies on altered physiology and is a functional investigation.

Imaging has evolved from diagnostic tests that involved clinical risk (CT with cisternography), through non-invasive, low spatial resolution axial examinations (CT plus contrast), to high-resolution MR imaging using three-dimensional T2W or gradient echo sequences and possibly T1W sequences post contrast. Current 1.5- or 3.0-Tesla (T) clinical MR imaging scanners offer high-resolution three-dimensional volume sequences that enable effective screening of the majority of patients without a requirement for T1W sequences post gadolinium (reducing cost). Most units will reserve postcontrast sequences for those cases in which the diagnosis of acoustic neuroma cannot be confidently excluded by a three-dimensional T2W or gradient echo sequence. 37,38

MR imaging with gadolinium (GdT1W) is considered by many as the gold standard diagnostic test for acoustic neuroma,39–44 and is used to evaluate positive findings following the MR screening examination. There are patients for whom MR imaging is inappropriate, such as claustrophobic patients and those with implanted metal. In these patients CT scanning is the alternative imaging test.

One important issue is that of supply and demand. The prevalence of ‘incidental acoustic neuroma’ has been estimated at 2/10,000. 45,46 The incidence of patients presenting to a general ENT clinic with symptoms that might indicate a CPA lesion has been reported as nearly 20%. 47 Even among those exhibiting symptoms of asymmetrical hearing impairment or unilateral tinnitus who have been referred for investigation, as few as 1% might have acoustic neuromas. 13,48,49 Thus, the large number of patients ‘eligible’ for investigation generates waiting list issues in many NHS MR imaging units, and the low incidence coupled with a high relative cost of GdT1W MR imaging (see Chapter 3) have contributed to reservations regarding its use as a first-line (albeit definitive) basis for diagnosis.

Debate therefore surrounds the most efficient MR imaging protocol to use in screening. Most NHS units are selecting sequence protocols that balance scanner/radiology time with waiting list pressures. In some centres excessive waiting list times for MR imaging are influencing clinical decisions and referral patterns, and selected patients are undergoing CT examinations to exclude CPA space-occupying lesions as the first (and possibly only) investigation, particularly in elderly subjects in whom diagnosis of intracanalicular lesions is not a clinical priority. CT cannot exclude small tumours within the IAC but, given the low prevalence of the condition and the reduced cost, this approach may be cost-effective, providing the majority of patients do not go on to MR imaging or to develop tumours.

Other resource savings may be made if medical staff do not directly supervise the screening process. Suitable patients can be ‘batched’, allowing large numbers to be scanned in a given MR imaging session. 50 However, this approach may not be suitable for older generation or lower field (< 1 T) scanners in which spatial resolution is insufficient to clearly define the individual nerves. It may prove necessary to obtain additional contrast-enhanced T1W images for a small percentage of patients with equivocal findings on T2W images or when patient movement leads to an inability to resolve the individual components of the nerve complex. For scans that are medically unsupervised, this requires patients to be recalled. It should be recognised that radiologists reporting these scans will have varying degrees of training and familiarity with imaging this region, and some may have a preference for and greater confidence with reporting contrast-enhanced T1W images.

Chapter 2 (see Comparison of the role of different protocols for MR imaging) summarises the recent evidence on the effectiveness of different MR imaging protocols in the investigation of patients presenting with symptoms that might indicate the presence of an acoustic neuroma.

Diagnostic criteria of the ABR waveform

The various peaks and troughs of the ABR waveform originate from the auditory nerve and progressively higher nuclei and tracts within the brainstem as reviewed by Moller. 23 Using a click stimulus results in a sequence of components labelled as waves I–VII, where waves I, III and V are the most robust and are employed as the main diagnostic components of the response. The origins of the components of the click-evoked ABR are as follows (Figure 4):

• wave I – the distal auditory nerve close to the cochlea

• wave II – the proximal auditory nerve as it approaches the cochlear nucleus

• wave III – the cochlear nucleus

• wave IV – the superior olivary complex

• wave V – the lateral lemniscus

• waves VI and VII – the inferior colliculus.

FIGURE 4.

Generators of the auditory brainstem response (ABR) and the ABR waves. (a) A schematic drawing of the presumed generators of the ABR. BIC, brachium of the inferior colliculus; CIC, Commisure of the inferior colliculus; CN, cochlear nucleus; CP, Commisure of the Probst; GC, Gudden’s commisure; IC, inferior colliculus; LL, lateral lemniscus; MGB, medial geniculate body; NLL, nucleus of the lateral lemniscus; RE, in reference to; SG, spiral ganglion; SOC, superior olivary complex; TB, trapezoid body. (b) ABR showing presumed anatomic correlations of major peaks. Note that one anatomic structure may give rise to more than one ABR wave and, conversely, more than one anatomic structure may contribute to a single ABR wave. From James W. Hall. New Handbook For Auditory Evoked Responses, 1/e. Published by Allyn and Bacon/Merrill Education, Boston, MA. Copyright © 2007 by Pearson Education. Reprinted by permission of the publisher.

The location of acoustic neuroma usually involves regions of the auditory nerve beyond the generator site for wave I. The normality of components originating more proximally than wave I is therefore compromised in cases of acoustic neuroma. For this reason the waves I–III and waves I–V interpeak intervals are the main diagnostic indicators of normality. The wave V component is the single most robust component of the ABR. This has resulted in the waves I–V component being the gold standard measurement and the one referred to most extensively in the literature. In some recordings the identification of wave I can be problematic because of the level of hearing impairment, and in these cases some reports refer to the absolute wave V latency. Correction of wave V latency with respect to hearing impairment can be applied. 54 As acoustic neuroma are predominantly unilateral, the interaural differences in the waves I–V latency interval and absolute wave V latency are important, as the ‘normal’ side can be used as a control for the suspect side. A summary of the important diagnostic criteria is as follows:

• monaural waves I–III and waves I–V interpeak latency intervals

• interaural difference of the waves I–V interpeak interval

• absolute latency of wave V with correction for hearing impairment

• interaural difference of wave V latencies (ILDV or IT⁵), corrected for hearing impairment.

In addition to the above criteria the total absence of a recognisable ABR is sometimes seen in acoustic neuroma. However, the ABR may also be absent in cases of severe/profound cochlear hearing impairment (especially at high frequencies) and so the audiometric status of the patient must be known to interpret the clinical significance of an absent ABR. If the patient’s hearing threshold is greater than 75 dBHL (decibel hearing level) at 4 kHz, for example, an absent ABR is non-diagnostic; as milder hearing thresholds are considered, an absent ABR becomes increasingly suggestive of neuropathy.

It is important to note that the characteristics of the ABR are also sometimes affected by a range of different pathologies (e.g. multiple sclerosis, other space-occupying tumours of the brainstem) in addition to acoustic neuroma. Interpretation of the waveform must take this into account in more complex cases.

Interpretive strategy

It is common practice to define a test’s performance in terms of sensitivity, the rate of true positives identified, and specificity, the rate of true negatives identified. However, these two measures do not carry equal weight, as the consequences of a false-negative result (a tumour that is present but that is missed) are more serious than those of a false-positive result (identification of a tumour leading to further testing, with no tumour being present).

Although the performance of individual ABR measurements may be known, patients are often judged using a combination of ABR measurements that are imperfectly correlated. There have been very few attempts to construct and evaluate efficient ABR protocols using a number of measurements, and the true specificity of most ABR protocols is likely to be somewhat worse than clinicians expect. 55

Study groups

Reports on the performance of the ABR in identification of acoustic neuroma cases are almost exclusively retrospective studies. Patients within a study group will have had ABR investigation previously and will have subsequently received confirmation of the presence of an acoustic neuroma or not using imaging techniques and/or during surgery. In earlier ABR reports the imaging confirmation will be based on CT scanning or early developments of MR imaging technology and procedures. This should be considered when interpreting the results on the diagnostic performance of the ABR presented by the papers included in this review.

Pooled synthesis

The 15 studies that assessed the sensitivity of ABR against MR imaging were highly heterogeneous (p < 0.0001, I² = 88%), ranging from 64% (95% CI 54–72%)66,68 to 100% (95% CI 40–100%)67 (Figure 5a). Only three studies reported the specificity of ABR versus MR imaging and these were also highly heterogeneous (p < 0.001, I² = 86%), ranging from 62% (95% CI 49–73%)67 to 88% (95% CI 78–94%)65 (Figure 5b).

FIGURE 5.

(a) Pooled sensitivity for 15 studies assessing the use of auditory brainstem response (ABR) versus the use of MR imaging. The size of the symbol is in proportion to the size and weight of each individual study. (b) Pooled specificity for 15 studies assessing the use of ABR versus the use of MR imaging. The size of the symbol is in proportion to the size and weight of each individual study. df, degrees of freedom.

Exploration of heterogeneity

A major potential driver of the heterogeneity in the diagnostic accuracy of ABR is likely to be the size of the acoustic neuroma. A number of studies reported their true-positive and false-negative results separately, according to the size of the acoustic neuroma. When we stratified the sensitivity results according to the categories of acoustic neuroma size, heterogeneity was substantially reduced and the following trend in pooled sensitivity results was obtained: acoustic neuroma size ≤ 1.0 cm: 79% (95% CI 72–85%), heterogeneity p-value = 0.98, I² = 32%; acoustic neuroma size 1.0–2.0 cm: 95% (95% CI 91–97%), heterogeneity p-value = 0.46, I² = 0%; acoustic neuroma size > 2.0 cm: 98% (95–99%), heterogeneity p-value = 0.347, I² = 10%. Sensitivity results of the individual studies are plotted by mean tumour size categories in Figure 6.

FIGURE 6.

Auditory brainstem response (ABR) sensitivity versus tumour size derived from four studies.

The effects of hearing impairment on the ABR

The recording of an ABR waveform whose clarity is sufficient to allow measurements of peak latencies requires that a well-synchronised action potential volley be induced in the auditory nerve. This is lost at low-stimulus intensities or in cases of elevated high-frequency hearing thresholds (regardless of site of lesion). In our review of the ABR literature we noted that some studies excluded those cases in which a clear (quantifiable) ABR was not seen, whereas other studies included them. Still other studies excluded cases in which the subjects’ hearing impairment exceeded a defined value, in the expectation that a measurable ABR would be unlikely. Most studies that evaluated the effect of hearing impairment on ABR included a statement which suggested that the authors considered that a quantifiable waveform was not expected if subjects’ hearing thresholds exceeded 70 or 75 dBHL at 4 kHz (or an average of two or three frequencies in the mid- to high-frequency range).

The effects of tumour size on the ABR

A review of the literature shows that the sensitivity of the ABR is dependent on the size of the tumour. High levels of ABR sensitivity are reported for tumours in excess of 1 cm, and several studies57,60,65 report values of 100%. However, in some cases this is achieved by classing no ABR as a positive finding, which results in relatively low levels of specificity. There is a significant reduction in ABR sensitivity for smaller tumours (< 1 cm). Values range from as low as 58%30 up to 92%. 29 In the work by El-Kashlan and colleagues29 this fairly high level of sensitivity again results from classing no ABR as a positive finding.

Because several studies provided ABR sensitivity data for tumours of less than 1 cm, between 1 and 2 cm and more than 2 cm, we were able to establish indicative sensitivities of 79%, 95% and 98% respectively. Thus, ABR provides a clinically acceptable sensitivity for medium to larger tumours but is not able to reliably identify small (often intracanalicular) tumours. One observation made58,61 was that the presence of a normal ABR waveform (indicating good functional integrity of the nerve) was a good prognostic indicator for hearing preservation, given an appropriate surgical approach.

Role of the ABR in screening for acoustic neuroma

In the 1980s and 1990s the ABR became popular as a cost-effective initial screen for patients considered at risk of an acoustic neuroma, its performance being judged in the light of the size of tumours then being detected radiologically (typically over 1 cm). However, the advent of gadolinium-enhanced MR imaging in the late 1980s and the subsequent development of non-contrast MR imaging has allowed much smaller tumours to be detected, leading to improved surgical outcome and hearing preservation. 70

The current aim of screening for acoustic neuroma is to detect tumours at an early stage when they are relatively small. This is a challenge for the ABR in view of the relatively low sensitivities reported for small tumours. The performance of the ABR for larger tumours is clinically more acceptable, but it is expected that identification of acoustic neuroma will be achieved before this stage. Application of ABR as a screening tool is therefore limited for the general population of patients when compared with the MR scan, particularly as the mean size at diagnosis is now 10 mm (see Chapter 4, Growth).

There are some situations in which MR imaging might not be possible or might be contraindicated. Alternative strategies, including ABR, may have a role to play in the following scenarios:

• in patients who are unable to routinely undertake an MR scan because of claustrophobia (2.8–4.0%);34,35 in such cases the use of sedation or a general anaesthesia, or an open MR imaging system or CT, could be considered as an alternative imaging strategy

• when physical limitations limit access to the scanner, such as severe obesity (1 in 1139 and 2 in 913);34,36 imaging on an open MR scanner or CT may be possible in these patients

• in patients with metallic implants, pacemakers, etc. (2 in 1139)34

• when access to an MR scan is limited because of time constraints and waiting times.

Modifications of the ABR

One research group has developed a modified ABR method that seeks to make the ABR sensitive to small (< 1 cm) acoustic neuroma. 71 The so-called ‘stacked ABR’ claims to allow the involvement of nerve fibres of all frequencies (rather than just the high-frequency fibres upon which the standard ABR relies), thus making it more sensitive to selective effects of even small acoustic neuroma. However, this method requires specialist equipment, demands a high level of operator skill and takes substantially longer than the standard ABR. Initial reports of the sensitivity of the stacked ABR for small tumours have been encouraging but further and independent studies are needed before it becomes clear whether it is a viable clinical tool worthy of routine clinical implementation.

With a similar aim, Bush and colleagues72 have explored the use of another index to improve the detection rates in smaller acoustic neuroma. Their method evaluates the difference between the behavioural and electrophysiological (ABR) thresholds, which they show to be abnormally large in acoustic neuroma. In a small series (seven acoustic neuroma cases, four of which had tumour diameters of 5 mm or less) the technique gave a positive result. Further studies will determine whether this method delivers the promise suggested by this study.

Role of other techniques in screening for acoustic neuroma

In the past, audiovestibular investigations have been employed to assist with the identification of acoustic neuroma. These tests have included stapedius reflex threshold, threshold tone decay, speech audiometry, alternate binaural loudness balance, otoacoustic emissions (OAE) and caloric testing. In early papers the value of these tests has been investigated with respect to ABR, CT and early MR imaging technology. A prospective study by Ferguson and colleagues12 showed that audiovestibular tests had sensitivities in the range of 45–85%, with caloric testing highest at 85% and speech discrimination lowest at 45%. Specificities were in the range of 66–90%, with alternate binaural loudness balance at 90% and caloric testing at 66%. The conclusion from this study was that the performance of audiovestibular tests as screening tools was limited when compared with the ABR. However, Haapaniemi and colleagues64 confirmed earlier reports that there is a positive relationship between the size of an acoustic neuroma and a decrement of caloric function.

Evoked OAE can be employed to assess outer hair cell activity. In some acoustic neuroma cases cochlear hair cell function can be preserved such that an OAE will be present, in contrast to conductive and cochlear losses in which a hearing impairment of greater than 30 dB will result in an absent emission. However, Quaranta and colleagues73 showed that only 18 of 47 patients (38%) with acoustic neuroma had an OAE present, which demonstrates the limitation of the technique as a screening tool. In cases with absent OAE the hair cells will have been compromised by the tumour.

Prasher and colleagues21 reported absent transient-evoked otoacoustic emissions (TEOAE) in 19 of 26 patients with acoustic neuroma (73%); in all patients in whom TEOAE was absent, a hearing impairment of 40 dBHL or greater was present, and this was assumed to be cochlear in origin. Telischi and colleagues74 undertook distortion product otoacoustic emission (DPOAE) measurements in 44 patients with unilateral acoustic neuroma. On the basis of the presence or absence of the DPOAE, 26 (59%) tumour ears were classified as having a cochlear loss, 13 (30%) as retrocochlear (DPOAE recorded in the presence of a hearing impairment > 40 dB) and five (11%) as mixed. Ferber-Viart and colleagues75 attempted TEOAE recordings in 168 ears with acoustic neuroma, and in 79% were not able to demonstrate good cochlear function, thus indicating a cochlear dysfunction in addition to the tumour. Ferguson and colleagues76 were unable to evoke TEOAE in 78 patients of a series of 100 with unilateral acoustic neuroma. The various mechanisms by which cochlear dysfunction may be involved in tinnitus generation are described in Chapter 4.

Pooled synthesis

Although the sensitivities were found to be relatively homogeneous (T2W: p = 0.34, I² = 11%; T2*W: p = 0.19, I² = 40%), specificities were highly heterogeneous (T2W: p < 0.0001, I² = 89%). The pooled test sensitivity for the T2W reference was 98% (95% CI 94–99%) and for the T2*W reference it was 96% (95% CI 86–99%). The specificity of the T2W studies ranged from 90% (85–94%)38 to 100% (91–100%). 88 For the T2*W studies specificity ranged from 86% (75–93%)85 to 99% (98–100%)86 (Figures 8 and 9).

FIGURE 8.

Pooled sensitivity and specificity for T2-weighted reference studies. The sizes of the symbols are in proportion to the size and weight of each individual study. df, degrees of freedom.

FIGURE 9.

Pooled sensitivity and specificity for T2-star-weighted reference studies. The sizes of the symbols are in proportion to the size and weight of each individual study. df, degrees of freedom.

Compared with the gold standard of GdT1W MR imaging, the high-resolution non-enhanced T2W and T2*W sequences appear to have good test accuracy as assessed by both sensitivity and specificity. The level of test specificity was found to be heterogeneous across the included studies in this review.

In such analyses the pooled effect may change if poorer quality studies are excluded. The number of studies included in these analyses was not large enough, nor did the studies differ sufficiently in quality (see Appendix 3) to make such a direction of analyses useful.

Evolution of MR technology subsequent to the evidence reviewed

During the 1990s progressive improvement of two-dimensional and then three-dimensional T2W or T2*W sequences acquired without Gd allowed researchers to question the requirement for GdT1W imaging in all cases, on the basis of improved image quality, diagnostic accuracy and time/cost savings. 91,92

Reports subsequent to 1998 have compared T2W with T2*W sequences for specific attributes that improve acquisition time, acoustic neuroma detection or characterisation. These papers were not included in the analysis on the basis that the sequences were not specifically evaluated for performance of acoustic neuroma detection by comparison with a GdT1W sequence. Selected references have been included as part of a short narrative review to document advances in imaging technology that have been applied to clinical practice subsequent to the data evaluated in the systematic review.

The time advantage of using an ultra-long echo train length (ETL) and half-Fourier three-dimensional FSE to acquire images with high spatial resolution in less than 3 minutes was reported by Naganawa in 1998. 86 The advantage of high-resolution T2W imaging in identifying the path of the VIIth nerve in relation to acoustic neuroma compared with the limited spatial and contrast resolution of GdT1W imaging was illustrated by Sartorelli-Schefer et al. 93 The improvement of image quality, due to reduced CSF flow artefact and shorter imaging time, by the use of a driven equilibrium radio frequency reset pulse (DRIVE) in adjunct with a T2W three-dimensional TSE sequence was reported by Ciftci et al. in 2004. 94

T2*W sequences, including balanced fast field echo (bFFE), fast imaging employing steady-state acquisition (FIESTA), or true free induction with steady-state precession (true FISP), have also become standard internal auditory meatus (IAM) sequences in clinical practice. These sequences use very short repetition times and balanced gradients to generate steady-state free precession. They use signal from free induction decays, spin echoes and stimulated echoes to generate images with a high signal-to-noise ratio.

The contrast in T2*W images is dependent on the T2/T1 ratio and CSF, which has long T2 and T1 relaxation times, returns a high signal intensity and outlines normal structures, which return a lower signal. Balanced gradients also result in flow compensation and reduce image degradation by CSF flow. 95

Some groups prefer TSE (T2W) to GE (T2*W) sequences because of their reduced degradation by magnetic susceptibility artefact. However, magnetic susceptibility is an inherent tissue parameter that cannot be completely avoided and occurs with both sequence types. Three-dimensional TSE sequences with a long ETL have very short acquisition times but are susceptible to blurring, depending on ETL and k-space trajectory. 96

The sequences used in current practice allow reliable acquisition of three-dimensional T2W or T2*W data sets that can be interrogated from any orientation using a submillimetre slice thickness on a workstation immediately after the study. Sequence selection is based on the equipment available and operator preference.

Allen et al., 199682

This study selected pathologically proven acoustic neuroma cases thereby facilitating validation of the gold standard reference test (GdT1W) but introducing a selection bias compared with a screening population.

Four neuroradiologists were blind to the clinical data and GdT1W results and used an observer confidence rating (1, ‘definitely not present’, to 5, ‘definitely present’), with scores 1–3 representing a negative test and 4 and 5 a positive test.

There was good agreement (no statistical difference using k-coefficients) between FSE and GdT1W MR imaging, but case selection could have facilitated the excellent performance of the index test. The authors’ conclusion that two-dimensional FSE was a cost-effective alternative to GdT1W MR imaging should be treated with caution.

The findings of two false-negative index test results and one false-positive result were not regarded as significant compared with the large number of observations in the study. The authors did not develop a discussion of the implications of indeterminate results, or the influence of the diagnostic threshold on sensitivity/specificity, but instead commented that, at the time of publication, the slice thickness of acquisitions had reduced from 3.0 to 0.8 mm, making a significant miss even less likely.

The authors commented that clinicians at their institution had abandoned impedance testing and brainstem-evoked responses in favour of limited-cost high-resolution MR imaging if an audiogram suggested the possibility of acoustic neuroma. They also commented that a negative MR test result would result in transfer back to the primary care provider rather than clinic follow-up. This documents a significant practice shift away from the routine use of the ‘gold standard’ test, based on case experience at the centre and the existing literature quoted in the paper.

Stuckey et al., 199683

This study emphasised the potential benefit of a high-resolution CISS sequence in which supplementary GdT1W imaging would be required only if the CISS imaging was suboptimal, equivocal or positive, or if an abnormality on whole brain T2 imaging required further characterisation.

The study included a recognised selection bias due to variable application of CT and non-imaging tests to the same cohort, with patients often referred for MR imaging to clarify existing findings. The authors acknowledged that this bias resulted in the high incidence (14%) of acoustic neuroma in this screening population.

A normal appearance was defined by the ability to identify normal-sized VIIth and VIIIth nerves clearly throughout their course, with no evidence of a mass lesion. The two independent, blinded observers graded the images as optimal, suboptimal but adequate, or inadequate and interpreted the images as normal, indeterminate or abnormal, with the latter two categories constituting a positive result. Each interpretation was rated with a subjective confidence score of low, moderate or high.

False-positive observations were caused by nerve clustering and/or a small IAC. Differences in observer performance (not confirmed statistically) were possibly related to experience and non-availability of MPR facilities. The authors commented that a high false-positive rate would have increased the necessity for GdT1W imaging and costs.

The single false-negative interpretation was the result of ‘satisfaction of search’ (identifying an obvious lesion in the opposite CPA) and the missed lesion having a homogeneous, relatively high signal intensity on the CISS sequences, resulting in reduced conspicuity compared with CSF in the IAC.

Although the study did not evaluate the role of GdT1W screening, the authors commented that omitting these sequences could result in missing lesions with no mass effect such as nerve enhancement associated with inflammatory lesions of the labyrinth or cranial nerves, or leptomeningeal sarcoidosis, metastasis or lymphoma. They commented that the incidence of such significant pathologies was very low and that the implications of missing more benign pathologies such as labyrinthitis were minimal.

The authors concluded that three-dimensional CISS was a sensitive test that could reasonably be used alone if gadolinium was contraindicated. They advised further evaluation before advocating the sequence as a stand-alone screening test.

Soulié et al., 199784

Normality was defined as a high CSF signal in the IAC and nerves visible; uncertain cases, in which the nerves could not be clearly identified, were classified as positives.

False-positive results were the result of recognised limitations of the technique (narrow IACs and CSF flow artefacts) and the authors proposed that these would be addressed by the future introduction of three-dimensional FSE scanning with submillimetre slice thicknesses.

The authors emphasised that the criteria set for normality of high-resolution T2 images were strongly predictive of normality on the GdT1W images and could rule out acoustic neuroma in 73% of their patients. The data were supportive of this and provided further support for non-contrast high-resolution screening although the methodology was flawed.

The authors recognised selection bias in their population, excluding patients with pathologies that would obligate GdT1W sequences (history of malignancy, human immunodeficiency virus, cranial neuropathy or middle ear disease); however, the prevalence of these conditions is low in the screening population and this bias was likely to be less significant than the methodological biases also acknowledged by the authors. For these reasons they did not advocate abandoning the GdT1W sequence for screening patients but proposed that this should be considered given the time/cost savings that might be involved.

Hermans et al., 199785

The independent, blinded observers looked for abnormality, defined as nodular low signal relative to CSF, and scored the degree of confidence on a five-point scale. They also documented any cause of uncertainty related to artefact or a technical limitation of the examination.

Test specificity values quoted were calculated on the basis of the number of ears examined. If recalculated on the basis of the number of patients, test specificity falls from 97% to 86% (see 2 × 2 tables in Appendix 5)

The authors commented that identification of the nerves in the IAC was not a prerequisite in the study and that making it so would have reduced the specificities obtained. This interpretive bias partly compensated for the fact that volume averaging, which was recorded as the most frequent cause of uncertainty, could have been reduced if the observers had been able to review MPR on a workstation. This facility, which might have improved specificity, was not available.

The authors explored the consequences of varying the diagnostic threshold for the test. When the operating point was set at a level felt to be equivalent to the operating point for screening, the quoted sensitivities and specificities were achieved. At this point the screening service would miss tumours in 2/18 cases and recall 9/65 negative cases for reimaging unnecessarily. At the more lenient operating point (including ‘probably negative’ cases), 0.5/18 cases would be missed (the intralabyrinthine tumour was scored ‘definitely negative’ on one occasion) and 34.5% of the patients would require GdT1W imaging.

The authors recognised the potential measurement bias in the study and commented that the significance of missing the smallest IAC tumours would depend on the natural history of that tumour and the clinical consequences of the miss. They did not discuss the cost implications of reduced test specificity (increased GdT1W usage), but they did discuss potential causes and the implications of a false-positive reference test (none were noted in the study though).

Naganawa, 199886

The high-resolution technique resulted in very high-quality imaging in the hands of these investigators, with a false-negative rate of zero. The improved spatial resolution also permitted identification of labyrinthine pathologies that could explain the presentation in 13 ears (eight patients, 4%).

Although the quality of imaging was high in this study, the acquisition time was long (11 minutes) and the technique was not available/relevant to the screening population at the time. The potential of high spatial resolution imaging to minimise false-negative test results was an important observation and the authors did recognise that false negatives may arise from a larger study population.

The false-positive rate was low but the significance of false positives was not considered, the authors commenting that they would be verified by GdT1W imaging.

Held et al., 199987

The authors’ conclusions that CISS provided better nerve attribution and improved evaluation of labyrinthine involvement in a screening population may be subject to selection and measurement biases.

The authors recognised this limitation by commenting that all cases in their institution would continue to undergo both three-dimensional CISS and three-dimensional GdT1W imaging routinely. They also acknowledged the time-penalty implications of workstation review of three-dimensional T2*W and three-dimensional GdT1W datasets, which was not a time-effective approach compared with conventional two-dimensional GdT1W imaging at the time.

Marx et al., 199988

The single independent but unblinded reviewer defined the results as positive, indeterminate or negative, but evaluation of the image quality of the index and reference test was subjective. The sensitivity and specificity of the index test were each 100%, there were no indeterminate readings and the two-dimensional FSE images were felt to be as good as the reference images in all cases. These results and the authors’ conclusion that FSE MR imaging was an excellent choice as a screening test for acoustic neuroma should be viewed with caution, given the lack of detail on subject selection and lesion measurement.

The authors intended to demonstrate the accuracy of two-dimensional FSE imaging using ‘readily available hardware’, and, although the results do not justify the conclusions, the intention recognises that a screening test has to be deliverable at a community rather than at a tertiary level. The authors emphasised that the sequence could be acquired at low cost without purchase of specialised hardware.

The authors also commented that FSE provided better evaluation of acoustic neuroma in the lateral IAC (although they did not evaluate this point) and that, ultimately, dropping GdT1W sequences would eliminate the confusion generated by a false-positive reference test.

Schmalbrock et al., 199989

The patients were prospectively selected from a cohort undergoing surveillance imaging or postoperative follow-up. Selection of pathologically proven acoustic neuroma cases facilitated validation of the gold standard reference test (GdT1W) but introduced a selection bias compared with a screening population, as did inclusion of eight patients with NF2.

The researchers were interested in clear demonstration of the outline of nerves in CSF, in attribution of a lesion to a nerve of origin and in accurately defining a lesion outline so that growth could be characterised. These objectives recognised the requirements for ‘watch and wait policies’ (for over 65-year-olds in their practice) and for accurate follow-up post surgery or radiosurgery.

The images were evaluated by a single unblinded radiologist. Grading scales were used to evaluate tumour conspicuity and contrast. Tumour volume measurements were calculated ‘offline’ on a workstation by a neuroradiologist using manual tracing and computed summation. The volumes on the three-dimensional GE sequence were assessed first to avoid biasing by the reference test. This was time-consuming and would not have been applicable to the screening population as a cost-effective technique.

Very small lesions were identified (0.05 cm³) and the cochlear, vestibule and semicircular canal were all clearly demonstrated on the three-dimensional GE sequence but not on the three-dimensional GdT1W sequence. The number of small lesions in the study provided supportive evidence of the applicability of the index test to the screening population.

The index test was shown to be highly sensitive and specific, even in small (< 5 mm) acoustic neuroma. The three-dimensional GE sequence was better at defining tumour outline with CSF (but not brain) and better at identifying the related nerves. Cystic tumour components were harder to distinguish from CSF on the index test, but cystic, necrotic or haemorrhagic components were better defined by the index than the reference test.

The authors emphasised the importance of the very high spatial resolution required for a screening test if subtle nerve pathologies were to be detected without the use of GdT1W imaging, and commented that achieving a slice section thickness comparable to, or less than, the width of a cranial nerve would allow this to be achieved in practice.

Evidence from the study did support the use of both the index and the reference test sequences to provide optimal evaluation of acoustic neuroma post treatment.

Zealley et al., 200044

A criterion for normality in this study was clear visualisation of the VIIth and VIIIth nerves, but further specific criteria for abnormality were not defined.

The authors discussed different screening strategies including FSE alone, GdT1W alone, FSE with GdT1W in selected patients and FSE plus GdT1W imaging in all patients. Their preferred strategy was the last, on the basis that the strengths/weaknesses of both sequences were complimentary and the approach would not require supervised image review or delayed recall of patients.

The authors’ conclusions were strongly influenced by the relatively high percentage of tumours (44% overall) that were not allocated to the ‘definitely tumour’ group in their study. They did not explore the implications of varying the diagnostic threshold of the test on test performance or of individual operator performance on test results.

The results presented in Appendix 5 were extracted from the data presented in the paper. If the diagnostic threshold is set at a level appropriate to a population screening test (including ‘uncertain’, ‘probable acoustic neuroma’ and ‘definite acoustic neuroma’ categories as a positive result), test sensitivity is 97%, specificity is 95% and there would be 32 true-positive, one false-negative, 57 false-positive and 1143 true-negative results in the population of 1233 patients screened.

If an ‘FSE alone’ policy had been followed using the suggested ‘screening threshold’, 1144 patients would have not required GdT1W imaging with one false-negative result. In addition to the 32 proven cases, only 57 would have required GdT1W imaging to confirm their false-positive status. The published results would appear to support an ‘FSE alone’ policy with this selected diagnostic threshold.

If the threshold is moved to include ‘acoustic neuroma probably not present’ as well, test sensitivity becomes 100% but specificity falls to 61% and there are 473 false-positive results with 506 patients requiring GdT1W imaging to characterise the tumour or confirm the false-positive status. This represents a significant cost to exclude one false-negative result in the population.

The range in the percentage of FSE scans correctly allocated to ‘definitely normal’ (44–77%), and the moderate correlation for interobserver score allocation, was an illustration in practice of the potential implications of variation in image quality, operator experience and confidence. Variable image quality and operator experience could increase or reduce the number of patients in the uncertain categories and therefore the performance of the screening test.

Two cases of labyrinthine enhancement (2/1233) were detected in the screening population. These were not characterised or discussed.

Annesley-Williams, 200138

This study excluded patients from the results analysis when the reference test was graded inadequate. This resulted in 52 (15%) of the two-dimensional cases and 16 (9.25%) of the three-dimensional cases being excluded. The author stated that they would have recalled these cases for GdT1W imaging and so, in the context of evaluating the FSE sequence as a screening test, these should have been regarded as false-positive results. As a consequence, the sensitivity and specificity calculable for the two-dimensional and three-dimensional FSE sequences in Appendix 5 differ from those quoted in the original paper.

The author rightly comments that the two-dimensional FSE sequence was accurate in detecting acoustic neuroma in the IAC (20/20 detected), providing image quality was adequate. However, the recalculated false-negative and false-positive rates for the two-dimensional FSE sequence illustrate the limitations of this sequence in practice at the time, with the inability to detect lesions as filling defects in the labyrinth and a relatively higher number of inadequate scans because of the inability to define nerve roots in the IAC on 3-mm sections.

Despite the biases described, the results suggest that the performance of the index test in practice improved after the introduction of the three-dimensional FSE sequence, with no false-negative results reported for tumour, and a reduced indeterminate scan rate. The author commented that they had expected a more significant reduction in inadequate rates using the three-dimensional TSE sequence with 1-mm slices; however, they suspected that the scan time of 8 minutes was associated with increased movement artefact.

The author advocated confirmation of all pathological cases with GdT1W imaging to avoid surgery on false-positive cases and discussed the role of GdT1W imaging in detecting subtle intralabyrinthine tumours and other inflammatory lesions. They speculated that, with improved acquisition times, the image quality of three-dimensional FSE might improve to a level at which it was possible to detect intralabyrinthine schwannoma. They also highlighted the potential value of detecting such lesions reliably with GdT1W imaging, facilitating appropriate management of patients with possible NF2.

Ben Salem, 200190

Two radiologists were blinded but reviewed the images jointly, first the index and then the reference test. Only five of the 19 acoustic neuroma were 5 mm or smaller, but the smallest (2.5 mm) was detected. Sensitivity and specificity were excellent and there were no false-negative and only 15 false-positive readings.

The author acknowledged a selection bias in their population resulting in the high incidence (10%) of acoustic neuroma.

Quality of the included papers

Appendix 3 details the results of the application of the QUADAS tool to the selected studies. The following discussion gives a narrative summary of these conclusions.

All of the selected studies set a clear research question and were either prospective or retrospective case–control studies; 6/12 studies included a spectrum of patients that was comparable with a screening population. 38,44,84–86,90

None of the selected papers specified inclusion and exclusion criteria for patients and only four described the subjects’ symptoms explicitly. 84–87 The papers in which the study population was most comparable with the screening population included consecutive series of patients with appropriate symptoms. 38,44,84–86,90

All of the papers compared an index sequence with the reference test, GdT1W imaging, and applied both the index and the reference tests to all cases, without a delay between acquisitions. Although all patients underwent GdT1W imaging, three series employed three-dimensional T1W sequences with MPR with submillimetre slice thickness86,87,89 rather than two-dimensional T1W sequences with 3-mm axial slices. This is unlikely to have influenced the sensitivity of the index test with regard to lesion detection (but does raise the theoretical possibility of a false-negative result for the lower resolution reference test).

Acquisition of the index test was independent of the reference test in all cases, and all studies described the execution of both index and reference tests in sufficient detail for other investigators to replicate them. The index test was interpreted without knowledge of the results of the reference standard in eight out of 12 studies. 38,44,81–83,85,86,88 The reference test was interpreted without knowledge of the results of the index test in five out of 12 studies. 81,82,85,86,88

Three out of 12 of the selected papers did not define the parameters that determined a positive or negative index test result. 81,82,86 This is unlikely to have biased the results of the studies given that the investigators knew what they were looking for.

None of the selected papers provided sufficient data to evaluate test performance for detection and characterisation of acoustic neuroma by patient age, sex or specific subsite location (such as IAC, porus acousticus or CPA). Although this does not undermine the performance of the index test as a screening tool, it impairs the extraction of natural history data from these cohorts.

The method used to evaluate tumour size or volume was documented in only two out of 12 of the selected papers. 38,89 Again, this did not influence lesion detection but illustrates the potential for measurement bias within the population. Significant intra- or interobserver variation at follow-up has profound implications for the evaluation of natural history and clinical effectiveness, as the proportion of patients under serial observation is substantial and increasing.

Indeterminate test results or observer confidence ratings were reported in five out of 12 of the selected series. 38,44,82,83,85 There were no withdrawals from any of the studies, other than omission of the indeterminate results from the analysis of test sensitivity and specificity in one paper;38 these results have been incorporated with the original data for analysis in this review.

Although the evidence evaluated in this review was acquired using ‘old technology’ and might be regarded as out of date, re-evaluation has provided additional supportive evidence for screening using non-contrast sequences, on the basis of data derived from the clinical use of a relatively low-resolution two-dimensional T2W sequence, which was initially reported with a cautious recommendation to use both non-contrast and GdT1W sequences routinely. 44

Factors such as case selection, equipment and image quality, variability in the number and experience of observers, variation in the diagnostic threshold and counting ears evaluated rather than cases, all have the potential to affect the calculation of sensitivity and specificity. It was not possible to control for these factors.

The evidence in this systematic review was therefore extracted from a heterogeneous population with variation in demographic features and investigator methodology over a time frame in which technology evolved and clinical practice changed. Allowing for this, the following specific themes may be drawn from the evidence evaluated in this chapter and from related papers that were reviewed informally as part of a narrative process during the project.

Detection of small acoustic neuromas

None of the papers evaluated in the evidence review had a significant population of patients with small acoustic neuroma. As such, the evidence extracted from the papers that compared a new index sequence with the reference standard (GdT1W) can only robustly support a statement that two-dimensional or three-dimensional T2W or T2*W sequences can reliably detect acoustic neuroma larger than 5 mm without the requirement for GdT1W imaging. However, technological advances in MR have improved the spatial resolution of current three-dimensional T2W or T2*W sequences to a degree that illustrates that this statement is not applicable to current practice, in which high-resolution sequences are used routinely to screen for acoustic neuroma of all sizes.

The evidence evaluated in this review was published between 1995 and 2001. To our knowledge, no studies comparing a T2W or T2*W sequence with a reference GdT1W sequence have been published since. We believe that this reflects the wide acceptance within the diagnostic community that good-quality, high-resolution T2W or T2*W sequences permit exclusion of acoustic neuroma of any size with sufficiently high diagnostic confidence to abandon routine GdT1W imaging. Practice change reflecting this belief was documented in 199682 and discussed in editorials shortly afterwards. 91,92

The refinement of MR technology over time has been a continual process that has resulted in either improved image quality or preserved image quality with shorter imaging times (reducing patient motion). Evidence drawn from a brief narrative review of publications since 2001 is supportive of the assertion that current three-dimensional T2W or T2*W sequences allow confident, reliable identification of the VIIIth and VIIth nerves within the CPA and IAM. In the context of population screening, the criteria that define normality would exclude acoustic neuroma.

This issue of size of tumour impacts on the quality assessment using the QUADAS tool. Only two points are lost in this assessment if there is significant selection bias in the study population that could be responsible for the apparently good performance of a test evaluated using excellent methodology in all other regards (e.g. selecting patients with moderate-sized tumours unlikely to be missed by any radiologist using any MR sequence).

Definition of normality

The majority of papers in the review defined criteria for a normal or abnormal study. The definition of a normal, high-resolution, non-contrast study should include clear visualisation of normal VIIIth and VIIth nerves within the CPA and IAC, along with clear identification of normal cochlear and labyrinthine structures. If a high-resolution study is confidently normal, GdT1W imaging is unlikely to improve diagnostic confidence. 38,44,83–85

Reported advantages of high-resolution T2W or T2*W imaging without contrast

• Exquisite anatomy: high-resolution T2W or T2*W images provide an accurate anatomical representation of the CPA and temporal bone and enable detection of the normal nerves, or acoustic neuroma, without requiring GdT1W sequences. 81–86

• Identification of VIIIth and VIIth nerve relationships: high-resolution T2W or T2*W images allow lesion attribution to specific nerve branches. 81,87,89 This can inform the surgical approach and may have prognostic implications, as tumours arising from the superior division of the vestibulocochlear nerve have a higher rate of hearing preservation. Spatial relationships between acoustic neuroma and the VIIth nerve can be determined preoperatively. 93

• Improved evaluation of inner ear structures: high-resolution T2W or T2*W images permit detection of inner ear extension87 and evaluation of inner ear pathology and dysplasias. 86,97

Reported disadvantages of high-resolution T2W and T2*W imaging without contrast

• Multiplanar reformats (postprocessing time) are required to confidently exclude subtle pathology at the porus acousticus or in the IAC. 86 Advances in computer workstation technology now permit rapid evaluation of thin-section MPRs, in any orientation, immediately following image acquisition.

• The incidence of small tumours in the study populations reported is relatively low, with approximately 30% of the reported acoustic neuroma being smaller than 5 mm. Very small tumours may be missed if there is nerve root clumping,44,82 and if there is nerve root clumping or image degradation by artefact, the examination should be reported as indeterminate (requiring recall for GdT1W imaging).

• High-resolution three-dimensional T2W or T2*W sequences will identify structures that are outlined by, or contain, fluid, for example, CSF or perilymphatic fluid. They will not clearly identify inflammatory or other processes within the temporal bone or brain that may also be responsible for the patient’s symptoms. Other sequences are required for this. 38,81,86

• High-resolution three-dimensional T2W or T2*W sequences will not necessarily characterise a filling defect that may also be a normal variant, vessel loop, exostosis, lipoma, cavernoma, meningioma or metastasis, although more complex presentations are more likely with some of these pathologies. 38,87

• High-resolution three-dimensional T2W or T2*W sequences will not detect microscopic acoustic neuroma that do not alter the contour of the VIIIth or VIIth nerves, or the anatomy of the cochlea or labyrinth. GdT1W imaging might detect such subtle pathology but cannot distinguish acoustic neuroma from other inflammatory pathologies (the specificity of GdT1W imaging would be low, the false-negative rate for acoustic neuroma would be high).

Image quality and radiologist experience

Image quality and diagnostic ability are paramount for the clinical effectiveness and cost-effectiveness of a non-contrast screening test.

False positives or negatives may occur as a result of poor image quality and limited radiologist experience or confidence. 38,83 Optimal image quality can be ensured by using modern equipment with good quality assurance and experienced radiographers. Normality should be defined (as above) and indeterminate scans should be either reviewed (second opinion) or recalled for GdT1W imaging.

Radiologists with a detailed knowledge of temporal bone anatomy and neuroanatomy, including anatomical variations and pathology, will reduce the false-negative rate and the requirement for GdT1W imaging for ‘uncertain’ cases. 44,83 The QUADAS assessment tool does not account for the experience of radiologists interpreting images.

Variation in observer confidence and sensitivity/specificity was illustrated by two studies,83,85 and there were moderate interobserver agreement scores in another. 44

The potential impact of radiologist experience can be illustrated by the effect of varying the diagnostic threshold for a positive test result on the false-positive result rate in the study by Zealley and colleagues;44 the false-positive rate increases from 57 to 473 patients when the threshold includes ‘acoustic neuroma probably not present’ cases. In this scenario an additional 416 patients would require GdT1W imaging to prevent one false-negative examination result. 44

Timing of radiology review

If a review is undertaken during the first attendance, an immediate decision regarding the requirement for GdT1W may be made, resulting in no need for recall or a second appointment and a reduced time to diagnosis. However, the radiology cost of supervision of all IAM scans is prohibitive given the low prevalence of acoustic neuroma in the screening population.

Thus, delaying the radiology review and undertaking unsupervised examinations may allow efficient high patient volume and out of hours scanning. The following factors will enhance the efficacy of this strategy: experienced radiographers, optimal scan quality, an experienced radiologist (low recall rate for GdT1W) and a relatively low prevalence of acoustic neuroma in the population.

GdT1W imaging as the gold standard test

GdT1W imaging has evolved from being the gold standard diagnostic test to the gold standard reference test for further evaluation of indeterminate images or characterisation of defined pathology. As very few false-negative cases have been reported for GdT1W imaging,98,99 its role as the gold standard reference test for lesion confirmation has remained unchallenged.

However, as lesion size reduces, the specificity and ultimately the sensitivity of the gold standard declines. 98 Although the sensitivity of GdT1W imaging approaches 100%, the specificity for acoustic neuroma is lower. The reported differential diagnosis for enhancing lesions in the IAC or CPA includes meningioma, metastasis, lymphoma, labyrinthitis, sarcoid and haemangioma. 100 False-positive results for GdT1W MR imaging have been widely reported. 81,82,101–103

Three-dimensional GdT1W sequences have improved the signal-to-noise ratio compared with a two-dimensional T1W sequence, with the equivalent slice thickness, and the MPR facility allows evaluation in any slice orientation, resulting in improved comparability with the three-dimensional T2W or T2*W sequence. 87

The relatively large (3 mm) section thickness of conventional two-dimensional GdT1W images and the low signal of non-enhancing structures (nerves/CSF) limit its accuracy for the measurement of tumour volume in small tumours, the identification of cystic components (particularly within the IAC) and the attribution of a lesion to a specific nerve. 89 Areas of non-enhancement of acoustic neuroma within the IAC may impact on outcome by misleading the choice of surgical approach, for example selecting a retrosigmoid approach to acoustic neuroma involving the lateral IAC. 104

In practice, GdT1W images are rarely interpreted in isolation as they are usually acquired to clarify findings on the screening three-dimensional T2W or T2*W sequence. The described limitations are of limited relevance in the context of screening but are more significant for follow-up of known acoustic neuroma.

Philosophy of cost-effective limited screening versus a more complete examination to detect all possible causes of hearing impairment

The quoted rate of normal scans in acoustic neuroma screening populations is 86–88%. 34,97 The abnormal rate of 12–14% includes a heterogeneous definition of abnormalities and inclusion/exclusion of pathologies. In addition to detection of acoustic neuroma or other soft tissue lesions involving the VIIIth or VIIth nerves, high-resolution three-dimensional T2W or T2*W imaging enables detection of arachnoid or epidermoid cysts and vessel loops within the CPA or IAC and evaluation of the inner ear for malformations and tumours.

Pathologies that may be missed by omitting the GdT1W sequence include small acoustic neuroma, intralabyrinthine acoustic neuroma, labyrinthitis, metastasis, and inflammatory dural and leptomeningeal lesions such as sarcoidosis. Pathologies that may be misinterpreted on three-dimensional T2W or T2*W imaging alone also include vessel loop, arachnoid cyst and lipoma. 38,82,83,85

Arguments for omitting GdT1W imaging from a screening examination include the fact that the incidence of the more aggressive pathologies in the screening population is low and that their presentation with isolated symptoms would be unusual. The benefit of identifying benign processes such as labyrinthitis is limited. 38,82,83,85

A number of abnormalities potentially related to hearing impairment or clinically significant but incidental to the patient’s presentation may be missed if a T2W evaluation of the brain is omitted from the screening evaluation. The ‘incidental abnormality rate’ detected in larger screening cohorts depends on what observers regard as pathological and also on the philosophy of the screening process (cost-minimal focused on acoustic neuroma detection versus inclusive aimed at detecting CNS problems as well). The incidence of unexpected pathology in the healthy population has been documented in a study from Rotterdam105 in which the authors reported the identification of 212 asymptomatic brain lesions in a population of 2000 (10.6%).

The average age of acoustic neuroma screening populations is around 50 years. Small vessel white matter lesions have been detected in up to 20% of such patients. 106 White matter lesions are a marker for cardiovascular risk factors, increased stroke risk and risk for intracerebral haemorrhage in patients placed on anticoagulants. 107 Such patients may benefit from investigation and treatment of previously undetected cardiovascular risk factors (e.g. a potential beneficial ‘side effect’ of the scan).

The prevalence of serious incidental pathologies requiring referral to neurology or neurosurgery was (10/1139) 0.9%34 to (11/644) 1.7%. 106 An approximation to a 1% prevalence would seem reasonable and would correlate with the findings in the Rotterdam study. 105 In our view, this is sufficiently high to justify inclusion of whole brain imaging with the screening IAM scan (which is current practice in the UK).

Contraindications to MR imaging

Absolute contraindications to MR imaging include the presence of a cardiac pacemaker, an intraocular metallic foreign body and some neurosurgical aneurysm clips. In such cases, an alternative imaging strategy is required.

Other factors that may limit the success of an MR examination include claustrophobia and severe obesity (increasing in the UK population). In this case, options include sedation, access to open MR units, CT or alternative non-imaging strategies.

Staged ABR and GdT1W MR imaging protocols

The study by Welling and colleagues114 compared the cost of a ‘traditional’ diagnostic protocol [including electronystagmography (ENG), CT and unenhanced MR imaging as well as ABR and GdT1W MR imaging in the algorithm] with the cost of one that deployed GdT1W MR imaging or ABR after audiometry was used to assess the likelihood that a neuroma may be present. Strength of suspicion was determined using an assessment of the patient’s history, a physical examination and audiometry results (pure tone air and bone conduction levels and speech discrimination). Only those patients considered to have a low risk of neuroma (less than 5%) were referred to ABR. Those at intermediate or high risk (5% or above as assessed by reviewer) were referred for GdT1W MR imaging immediately after audiometry. (If an individual’s ABR results were abnormal among the low-risk group, a subsequent referral for GdT1W MR imaging was advocated.)

Unit costs were estimated from a survey of four independent institutions conducted in 1989. (The average cost of ABR was reported as US$220 and of GdT1W MR imaging as US$1235.) The average cost of the traditional diagnostic work-up (in which ENG and CT, etc. might feature) was based on a retrospective review of 70 patient records and application to these of unit costs. The average cost of the traditional protocol was estimated at US$2568 (this figure excluded office visits and treatment). By comparison, the estimated average cost of the alternative diagnostic strategy (ABR and/or GdT1W MR imaging immediately after audiometry based on an assessment of risk) was US$1150. (Assuming a dollar to sterling conversion of £1 = US$2, the cost of the traditional protocol in 1989 prices would be £1284 and that of the alternative strategy would be £575.)

The paper only considers other costs related to treatment of the neuroma, long-term sequelae and litigation, and offers some estimates in relation to the first two from a range of sources. The average cost of hospitalisation for excision of an acoustic neuroma was reported as US$9400 when there are no complications and US$39,950 when major complications arise. (Surgery was only attempted on a subgroup of patients, the team opting to manage seven through observation rather than intervention. Six of those operated on, however, experienced major complications.) When a delay in diagnosis results in profound unilateral SNHI, an impairment equivalent to 6% of annual salary – US$18,000 over the average remaining working life – was hypothesised as an estimate of the cost of lost production (35% lost income being suggested for bilateral deafness).

A number of problems arise with this study. First, the 70 patients in the study were individuals known to have tumours. Given this, although the detection rates across protocols could be treated as being equal and thus the study amounts effectively to a cost-minimisation analysis, whether this group would be representative of those to whom the protocols would be applied in practice is debatable. Individuals exhibiting symptoms who are not imaged may well experience a different diagnostic process (and cost) to that of those encountered in this group. From this study, for example, we do not know how many false negatives might have arisen in the ABR examinations of those judged ‘low risk’ among an unselected group. In consequence, we do not know the cost associated with false negatives in this protocol. Similarly, we do not know what the compliance might be for those assigned to recall monitoring within such a protocol, or again what costs might arise from this in terms of missed tumours. Although a sensitivity analysis could potentially have gone some way to addressing these issues, none was undertaken in the study.

A second, although relatively minor, issue concerns the discussion of the treatment and costs of long-term sequelae. This study is almost unique in making reference to such costs, but is somewhat naïve in its treatment of them. Considerable caution must therefore be exercised in interpretation of this discussion. For example, no discounting of future lifetime earnings is undertaken in the estimate of future lost earnings. Again, although this is an issue that could perhaps have been dealt with in a sensitivity analysis, that none was undertaken means that the result must be viewed cautiously. Although the study makes no claim to have definitively established the cost-effectiveness of a risk-assessed protocol over a traditional approach, the inference that this is the case is made.

The study by Robinette and colleagues65 essentially follows the same approach as that of Welling and colleagues114 and is again US based. The authors take data relating to 75 patients (all of whom had been surgically confirmed as having tumours) and assign the patients to the Welling and colleagues’114 groupings of high, intermediate and low risk. (As the symptoms presented in case notes did not always fit the Welling and colleagues’114 criteria, some interpretation was used on occasion.) Based on the prevalence of tumours among individuals exhibiting the symptoms described by Welling and colleagues114 (prevalence rates as reported in Buach and colleagues118), the number of individuals needed to generate the number of tumours observed was calculated. The costs of screening this number with an GdT1W MR imaging protocol as the first test after audiometry and of using a protocol that incorporates a risk-assessed screen in which ABR is included (in line with Welling and colleagues114) are calculated. The sensitivity for ABR and GdT1W MR imaging from Buach and colleagues118 is then used to determine the expected number of false negatives in each risk grouping, and, finally, the additional cost required to detect these by adopting an GdT1W MR imaging first test protocol is calculated. The study design can be described as a cost-effectiveness analysis in as much as costs are related to the number of expected missed tumours under the two protocols.

Unit costs are presented as US$300 for ABR and US$1500 for GdT1W MR imaging. The source for these is not identified, although from the context in which they are presented they are likely to reflect billable charges levied at the study site. The cost-effectiveness of the risk-assessed diagnostic algorithm over the GdT1W MR imaging first test algorithm is demonstrated by reference to the cost that would be incurred were one required to detect the tumours that would otherwise be missed. This cost is shown to rise across the risk groups, the low-risk group being higher than the intermediate-risk group and the intermediate-risk group being higher than the high-risk group. Thus, to find the one tumour missed in the high-risk group, it is estimated that a protocol deploying GdT1W MR imaging immediately after audiometry would cost an additional US$30,900. To find the four missed tumours in the intermediate risk group would cost an estimated additional US$858,000 (or US$214,500 per tumour), and to find the one missed tumour in the low-risk group would cost an estimated US$1.6 million.

The study is not compromised by selection bias to the same extent as that of Welling and colleagues. 114 Although a selected sample is used – all individuals were known to have tumours – the number of persons one would expect to screen in order to generate the number of tumours observed was derived from a study by Bauch and colleagues118 in which both tumour and non-tumour patients were present. In this respect, the data are more likely to be representative of patients to whom tests would be applied in practice. However, the use of prevalence data from another study as well as one set of unit costs does build assumptions into the analysis upon which the study’s results and conclusions are predicated. Varying these assumptions, as the authors state (and provide some guidance on), would alter the study results. A formal sensitivity analysis is, however, not undertaken nor are dollar estimates of costs per missed tumour across groups given under different assumptions.

These caveats noted (and broadly acknowledged by the authors), the study nevertheless extends the conclusion of Welling and colleagues114 that a diagnostic strategy based on risk assessment in which ABR has a role is not only likely to be more cost-effective than one based on a more protracted ad hoc battery of tests but also more cost-effective than one that deploys GdT1W MR imaging as a first-line test after audiometry.

Robson and colleagues,43 Saeed and colleagues42 and Ravi and Wells41 report the results of studies undertaken in the UK. In each, comparisons are made between protocols involving specific GdT1W MR imaging tests immediately after audiometry and traditional diagnostic protocols involving a range of intervening tests. Each is undertaken broadly within what might be considered as being cost-minimisation analyses.

Robson and colleagues43 report what is effectively a prospective study of 99 patients. The authors state that all 99 patients were referred for GdT1W MR imaging before the results of their audiovestibular examinations were known. The results of these tests should not therefore have influenced the decision to refer patients for GdT1W MR imaging and thus the study can be thought of as being prospective. Although it is conceivable that an elevated suspicion of neuroma must have been present for GdT1W MR imaging to have been requested, that other tests were requested first suggests that any elevated risk cannot have been very high. Unit costs are based on charges levied on private patients at a single UK hospital (Radcliffe Infirmary, Oxford) in 1992. The cost of directing 99 patients to imaging immediately after audiometry is estimated at £12,900, the unit cost of GdT1W MR imaging being reported at £130. The cost of the existing protocol that included CT and ABR (as indicated by need) before imaging is estimated at £12,545 (these costs exclude GdT1W MR imaging and any repeat CT scans that may in reality have been undertaken). The imaging protocol involves patients being processed in batches (thus obviating the need to adjust equipment, and saving on scan time) as well as the use of a reduced amount of gadolinium. The elimination of the need for repeat CT and/or ABR or GdT1W MR imaging when equivocal results from CT and ABR remain equivocal, or of missed false negatives when GdT1W MR imaging is not used to exclude a tumour, would likely erase the small cost difference between the two protocols. As with the other papers reviewed thus far, however, no sensitivity analysis is undertaken to illustrate this.

The prospective study design in which all patients presenting with symptoms giving rise to a suspicion of acoustic neuroma are included in the study is one of the paper’s strengths, although the absence of a sensitivity analysis means that the conclusions must be treated with some caution. In as much as only costs are formally compared, it can be thought of as a cost-minimisation analysis. GdT1W MR charges reflect in part the reduced scan time (20 minutes compared with the standard of approximately 60 minutes) as well as the reduced dose of gadolinium. Using an exchange rate of £1 = US$2 and tripling the scan time involved, the dollar equivalent would be US$780. As can be seen, this is somewhat less than the costs reported by Welling and colleagues114 and Robinette and colleagues. 65 The reduced dose of gadolinium may in part explain the difference as may the different geographical settings for the studies; however, the possibility that unit costs are not calculated on the same basis remains.

With these caveats in mind the study does seem to provide prima facie evidence that GdT1W MR imaging immediately after audiometry for patients in whom an acoustic neuroma is suspected is more cost-effective than a strategy in which GdT1W MR imaging is deployed only after further intervening tests. Although this would appear to contradict the findings of the study by Robinette and colleagues,65 the different unit costs and the different protocol of tests used between the two studies must be borne in mind. To what extent a direct comparison is possible is debatable.

In the study by Saeed and colleagues,42 a traditional protocol (involving CT as well as ENG, ABR and GdT1W MR imaging) is again compared with a protocol in which GdT1W MR imaging is conducted immediately after audiometry has failed to remove suspicion of a neuroma. The cost of processing patients who would undergo GdT1W MR imaging was estimated using details of tests involved in the traditional protocol. Unit costs relate to the authors’ own department (Manchester, UK), although how exactly these have been arrived at is unclear. The unit costs are similar to those of Robson and colleagues43 for imaging (£165) and, as with Robson and colleagues,43 were based on a protocol that involved processing patients in batches (requiring a 30-minute scan time) as well as use of a reduced dose of gadolinium. (Adjusting these for a 60-minute scan time and converting to dollars at an exchange rate of £1 = US$2, the GdT1W MR unit cost would equate to US$660.) The cost per patient of the traditional diagnostic process is estimated at £188.22, and the cost of an GdT1W MR imaging first protocol is estimated at £180.05. Cases in which tumours may have been missed because of equivocal CT or ABR results are presented suggesting that the two protocols would not be equally effective. As with Robinette and colleagues,65 costs per tumour missed are not calculated. The costs of operating the two protocols are very similar. The similarity of the costs forms the basis for the study’s conclusion and thus it can be thought of as a cost-minimisation analysis. Given the similarity of costs under the two protocols and the likely incomplete nature of costs under the traditional protocol, the dominance of the GdT1W MR imaging first protocol over the traditional diagnostic strategy is implicit.

The relative cost-effectiveness of the protocol in which imaging is undertaken more widely and earlier in part reflects the effect of eliminating equivocal intervening tests. If these results were from an unselected group of patients (as was the case in Robson and colleagues43) there is also a strong likelihood that the traditional protocol would miss some tumours because of the lower sensitivity of the tests deployed (i.e. like is not compared with like in terms of effectiveness). The corollary of this is that if this was an unselected group of patients there would be good reason to believe that the cost advantage reported for a GdT1W MR imaging first test protocol after audiometry would be greater than is indicated here. However, it is not entirely clear that this is an unselected group of patients. It is not stated what prompted the GdT1W MR imaging request for this group, merely that ‘139 requests were made as part of the investigative protocol for patients with unilateral or asymmetrical sensorineural hearing impairment’. In other words, it is possible, as with Welling and colleagues,114 that the request for imaging was informed by the results of other tests besides audiometry.

If we assume that this is not the case, then the study is similar to that of Robson and colleagues43 and provides further evidence that a strategy in which GdT1W MR imaging is deployed immediately after audiometry is more cost-effective than one in which it is deployed only after other intervening tests. If we do not make this assumption then the result must be treated with somewhat more care. A straight to GdT1W MR strategy may see many referred for GdT1W MR imaging who could be (and implicitly were) eliminated from further investigation by cheaper tests. That a GdT1W MR image was requested for all in this group may indicate that it was in fact needed to positively diagnose or eliminate acoustic neuroma as a cause of the symptoms exhibited. In this situation, savings attributed to the exclusion of intermediate diagnostic steps would then merely reflect that, for a group that could be defined ex post (i.e. once its participants have been the tested we know its value), by definition such tests were of no value. As with the other studies, no formal sensitivity analysis is undertaken, again indicating the need for some care to be exercised in drawing conclusions from the study.

In Ravi and Wells,41 similar issues arise. These authors compare the cost of a traditional protocol, in which CT as well as ENG and GdT1W MR imaging are used to investigate suspected acoustic neuroma, with that of a GdT1W MR image immediately after audiometry protocol. Case notes on all tests conducted on 100 patients are examined to identify those in whom a GdT1W MR image was requested to eliminate an acoustic neuroma. The unit costs for GdT1W MR imaging are taken from one of the study’s centres, although these are demonstrated to be similar to those at the other study sites (£110, £137 and £158, £137 being that used in the study) and are similar to those of Robson and colleagues. 43 When, following a pure tone audiogram, patients were referred to GdT1W MR scanning as their next investigation, the cost per patient is estimated at £220.72 (if results were communicated by the GP) or £258.72 (if results were communicated via an outpatient visit). The £220.72 comprises £137 for a GdT1W MR image, £75 for an ENT consultation and £8.72 for pure tone audiometry. By comparison, when other tests besides a pure tone audiogram were undertaken before GdT1W MR imaging was conducted, the cost per patient is markedly higher – £417.31 for those who underwent a pure tone audiogram, ‘special’ audiological tests and ENG as well as GdT1W MR imaging. (The precise GdT1W MR test is not described, although the costs suggest it is similar to that in the studies by Robson and colleagues43 and Saeed and colleagues. 42) As with Saeed and colleagues,42 the analysis of cost-effectiveness in terms of cost per tumour missed or detected is not pursued, and the study may be thought of in terms of a cost-minimisation analysis. That a protocol in which GdT1W MR imaging is deployed as a first test costs less than a protocol involving intervening steps is sufficient to demonstrate its dominance.

As with Saeed and colleagues,42 the interpretation of results here depends on the interpretation of recruitment to the study. If the 100 participants whose notes were examined reflect a selected sample – individuals for whom other tests after pure tone audiometry were equivocal – they cannot be viewed as being equivalent to the general population of individuals for whom audiometry has failed to rule out a tumour (among a group comprising the latter, other tests may not have been as equivocal and imaging may have been pursued). Were this the case, any savings attributed to the exclusion of intermediate diagnostic steps would merely reflect that, for a group defined ex post (i.e. for when having conducted the tests we know they were of no value), savings could be derived by the elimination of intermediate steps. If, on the other hand, the sample represents an unselected group (everyone for whom pure tone audiometry has failed to exclude a tumour) then, as with Robson and colleagues,43 the study lends weight to the argument that GdT1W MR imaging immediately after audiometry is a more cost-effective diagnostic strategy for acoustic neuroma than one that involves additional steps. From the text, one cannot be certain of which of these is the case. The fact that it is not stated explicitly that this is an unselected group adds to the suspicion that this was not the case. As with the other studies, no formal sensitivity analysis is undertaken and again caution is warranted in regard to the study’s results.

In Cheng and colleagues,115 the cost of a protocol that involves GdT1W MR investigation after audiometry is compared with that associated with a traditional protocol. Included in the battery of tests deployed in the traditional protocol are ABR, CT and ENG as well as GdT1W MR imaging. Data on the tests performed on 270 patients who presented consecutively at a Canadian hospital and in whom an acoustic neuroma was suspected were extracted from their notes. Unit costs for the tests used were derived from the Ontario Health Insurance Plan for 1998 and combined with data on test use to produce an estimate of total costs for all patients. The unit cost of an uncontrasted MR image is reported as C$260 and that of an ABR as C$44.60 (broadly equivalent to US$244 and US$42 respectively). This was then averaged across the 270 patients to produce an estimate for the traditional protocol of C$191.19 (US$180). The average cost for the 270 patients to receive audiometry (pure tone, speech discrimination and acoustic reflex tests) followed by GdT1W MR imaging without other intervening tests was estimated at C$310.30 (US$292).

In as much as the study deals with an unselected group of patients, comparing the actual costs of administering one protocol with the estimated costs of another (effectively a cost-minimisation analysis), it does not suffer from selection bias as was the case in some of the other studies. However, against this, under the traditional protocol the number of false negatives from ABR and other tests are unknown. (‘Most patients with normal ABRs were assumed to have no retrocochlear pathology. No further testing was performed on such patients’.) As the number of tumours missed is unknown, like is not being compared with like in terms of effectiveness across the two protocols. It follows that the costs associated with an unknown number of potentially missed tumours are implicitly ignored in the study. Although the study could be classified as a cost-minimisation analysis, the validity of the assumption that the two diagnostic strategies have the same outcomes is difficult to sustain (an unknown amount of costs associated with missed tumours has been omitted from the traditional protocol). As no formal sensitivity analysis is undertaken, the impact on the findings of relaxation of the implicit assumptions regarding missed tumours or of varying the costs associated with the conduct of any of the diagnostic tests is unknown. [The figures quoted in the paper are somewhat confusing. The authors quote a unit cost of GdT1W MR imaging of C$310 but use a cost of C$260 in their calculations, the unit cost they quote for non-contrast enhanced MR imaging. The total cost of processing the 270 patients would be C$83,700; adding to this audiometry costs (C$13,581) the total for a GdT1W MR imaging protocol would be C$97,281 and not the C$83,781 quoted in the paper.]

In Carrier and Arriaga,13 a protocol that involves imaging immediately after audiometry is compared with a protocol that permits further investigation before imaging based on ABR – similar to the study by Welling and colleagues114 but without formal risk assessment. As with Robson and colleagues,43 a specific imaging strategy in which a reduced dose of gadolinium is administered is followed, as well as a reduction in the number of images printed. Although the authors make no reference to batching patients, the total imaging time including set-up and administration of gadolinium is reported at between 20 and 25 minutes, similar to the times reported by Robson and colleagues43 and Saeed and colleagues42 when batching does take place.

The images of 485 patients who presented with symptoms of ASHI or unilateral non-pulsate tinnitus were reviewed and a diagnosis arrived at. The cost of the ‘straight to imaging’ protocol is estimated based on direct progression to imaging after audiometry. This is estimated at US$300–500 per scan, the lower cost reflecting the lower scan time as well as use of a reduced amount of gadolinium. The cost of an ABR is quoted at US$200–300 per test. In what amounts to a cost-minimisation analysis the authors conclude that, given the lower sensitivity of ABR compared with GdT1W MR imaging and the associated costs of additional confirmatory tests that would likely be involved if ABR was deployed, as well as the costs of potentially missed tumours, the direct to imaging protocol is more cost-effective than traditional algorithms.

As with the other papers reviewed, however, a number of issues arise in relation to this study. First, it is not clear if the sample used here has been selected – it is one for which images exist, and others may have been filtered out before reaching this stage. It seems unlikely that imaging would have been requested unnecessarily and thus it is likely that it is a selected sample. This need not effect the conclusion of the study as the costs of the diagnostic tests are not actually calculated based on the use of tests in the sample. Rather, the relative cost-effectiveness of the direct to GdT1W MR imaging strategy hinges on the relative unit costs and assumptions regarding repeat examinations and missed tumours. Although the unit cost reported for an ABR (US$300) is in line with the costs reported by Robinette and colleagues65 and Welling and colleagues,114 it is at variance with those reported in the UK studies as well as in other US studies. Thus, the unit cost reported for an ABR by Cheng and colleagues,115 for example, is US$44.60 (£22.30); in Robson and colleagues43 a unit cost of £55 (approximately US$110) is reported and in Saeed and colleagues42 there is a cost of £23 (approximately US$46). As with the other studies, a sensitivity analysis could have assessed the impact of uncertainties regarding these costs (as well as those regarding the sensitivity of ABR and the cost of missed tumours) but none was undertaken. If the unit costs are considered credible the study does provide evidence supporting the contention that a direct to GdT1W MR imaging protocol is likely to be more cost-effective than one that entails intervening tests.

Finally, the prospective study by Rupa and colleagues67 examined 90 patients, comparing a protocol using ABR and GdT1W MR imaging in the investigation of suspected acoustic neuroma with a protocol using GdT1W MR imaging only after audiometry. The study was conducted in India with unit costs for GdT1W MR imaging and ABR based on those operating at the hospital in which the study was conducted (although it is not stated, one assumes that these reflect charges levied on users). In total, 18 patients were unsuitable for ABR in the study. Among the remainder, all received ABR and GdT1W MR imaging. ABR had 100% sensitivity among those tested (as confirmed by GdT1W MR imaging results) and cost one-fifteenth of GdT1W MR imaging (US$13.33 versus US$200). If all 90 patients were to receive GdT1W MR imaging after audiometry, diagnostic costs would have been US$18,000. If only the 18 unsuited to ABR and the 30 whose ABR results suggested retrocochlear pathology were to receive GdT1W MR imaging, all others receiving only ABR, total costs would have been US$10,800. This indicates that a diagnostic algorithm incorporating ABR as an initial screen followed by GdT1W MR imaging is more cost-effective than one deploying GdT1W MR imaging immediately. This study thus apparently contradicts that of Carrier and Arriaga13 as well as the UK studies but is in line with those of Welling and colleagues114 and Robinette and colleagues. 65

At face value, this study offers perhaps the most convincing evidence of the various studies examined thus far that a diagnostic algorithm incorporating GdT1W MR imaging after initial screening with ABR is more cost-effective than one with GdT1W MR imaging after audiometry. The patient group is unselected and information on false negatives under the different search strategies is known (there were none). On closer examination, however, the patient group studied is found to be atypical of that likely to be encountered in most Western hospitals. Of the six tumours detected among this group, four were in excess of 1.5 cm in size and two were 3 cm in size. These are somewhat larger than is typically the case in other studies (see Chapter 4, Growth). This clearly explains the high sensitivity of ABR in this study (see Chapter 2, Comparison of the use of auditory brainstem response with the use of magnetic resonance imaging). Moreover, although the unit cost of GdT1W MR imaging is broadly comparable to that reported in other studies, the unit cost of ABR does seem somewhat cheap. Although the study does appear valid – even in the absence of a sensitivity analysis – whether the results are applicable to a UK setting is highly questionable.

GdT1W MR imaging versus non-contrast-enhanced MR imaging

The study by Daniels and colleagues116 compares a diagnostic algorithm that deploys non-contrast-enhanced MR imaging as a first test after audiometry with a diagnostic strategy in which ABR is used as a screen followed by GdT1W MR imaging. Unit costs are based on the charges levied at the study institution and are US$245 for ABR, US$1200 for GdT1W MR imaging and US$415 for non-contrast-enhanced MR imaging. The patient group studied is selected, comprising only those who have experienced sudden SNHI. The sample includes 58 individuals, and the use of various tests is based on a retrospective review of patient notes regarding ABR testability and results. Those who were not ABR testable were assigned the cost of GdT1W MR imaging (this being the diagnostic test they would have been referred to) and those who were ABR testable were assigned the cost of an ABR and the cost of GdT1W MR imaging if their ABR results were abnormal (for those with normal ABR results five cycles of audiometry follow-up were assigned).

Total costs for the ABR/GdT1W MR imaging protocol for the 58 patients were estimated as US$52,175. Using non-contrast-enhanced MR imaging as the sole test after audiometry, total costs were estimated as $24,070, a difference of $485 per patient evaluated. If we assume that the two protocols are equally effective in the diagnosis of tumours this would indicate that the non-contrast-enhanced MR diagnostic protocol is more cost-effective than the traditional protocol. If, more realistically, we assume that the traditional protocol is less sensitive than non-contrast-enhanced MR imaging (because of the role of accorded ABR), more false negatives would be associated with the traditional protocol and costs associated with complications and long-term sequelae would also be higher. Interpreting this as a cost-minimisation study it seems reasonable to assume that, for this patient group, the extent to which non-contrast-enhanced MR imaging is more cost-effective than the traditional protocol is underestimated. The extent of any overestimation though is impossible to assess.

An attempt at a sensitivity analysis is undertaken in this study. In the sensitivity analysis the percentages of patients that one would expect to be ABR non-testable and, of those who are ABR testable, that one would expect to exhibit abnormal ABR results are estimated for a hypothetical cohort of 100 patients. Rates are based on a meta-analysis of the literature for patients with sudden SNHI. Under these revised assumptions the non-contrast-enhanced MR imaging protocol was estimated to have a slightly lower saving of US$431 per patient associated with it relative to the ABR/GdT1W MR imaging protocol.

A key issue with this study, however, (echoing the criticisms levelled at a number of the other studies already examined) is that the findings relate to a selected (and in this case very specific) patient group – namely those who have experienced sudden SNHI. This group represents only a small proportion of those who experience acoustic neuroma – reported as 5–15% in this study based on the literature (see Chapter 4, Symptoms). How representative the detection rates and costs associated with the different diagnostic strategies applied to this group are for the other 85–95% is unclear (the authors make no claims in regard to this broader group).

A second issue with the paper (as with various others) is the absence of a formal sensitivity analysis. Although the unit costs deployed, for example, are in line with those contained in several of the studies already examined, as has already been noted, considerable variation in costs exists in the literature and the costs reported here are somewhat more expensive than those reported by other studies. A sensitivity analysis would have let the authors identify the range over which costs could vary without effecting their conclusions. Similarly, based on the literature, the paper reports the sensitivity of non-contrast-enhanced MR imaging as 100%. This is by no means universally agreed (as discussed below). The authors could have used sensitivity analyses to explore the range of sensitivities for non-contrast-enhanced MR imaging and ABR for which one protocol was more cost-effective than another in this patient group; however, no such analysis was undertaken.

In the study by Allen and colleagues,82 a patient group with more general symptoms was examined with non-contrast-enhanced MR imaging and conventional GdT1W MR imaging to determine the relative cost-effectiveness of the two imaging approaches. A group of 25 patients with GdT1W MR imaging-confirmed diagnoses of acoustic neuroma and 25 control subjects were evaluated in blinded readings of non-contrast-enhanced MR images and GdT1W MR images by four neuroradiologists. GdT1W MR imaging was assumed to have 100% sensitivity and non-contrast-enhanced MR imaging was demonstrated in the study to have 98% sensitivity. Unit costs were based on billable charges; for GdT1W MR imaging the unit cost was US$1200 and for the non-contrast-enhanced MR examination the unit cost was US$400. The authors assert that there is no statistical difference between the two tests in terms of the results obtained but that non-contrast-enhanced MR imaging is US$800 cheaper per test than GdT1W MR imaging and therefore more cost-effective. The non-contrast-enhanced MR imaging protocol indeed can be made even cheaper by removing unnecessary ABR and impedance testing from the GdT1W MR imaging protocol and made more effective by deploying a more refined search approach (taking thinner sliced images), which is now possible. In effect, the study amounts to a cost-minimisation analysis in which non-contrast-enhanced MR imaging is clearly less costly.

However, the false-negative rates reported in the study indicate that GdT1W MR imaging and non-contrast-enhanced MR imaging are not equally effective despite their treatment as such by the authors in this study. It follows that a cost-minimisation analysis is not appropriate here. The cost of missed tumours should have been factored into the analysis. Although the low false-negative rates may make it unlikely that this would have effected the result (non-contrast-enhanced MR imaging would have remained the more cost-effective strategy), no sensitivity analyses were undertaken that could perhaps have addressed this issue. A further issue commented on by Jackler91 is whether the high sensitivity achieved at this centre could reasonably be expected to be replicated elsewhere. Other centres are unlikely to have the experience and expertise in the use of the non-contrast-enhanced MR technology that exists at this site. Other centres may not achieve the high sensitivity achieved here and may experience higher false-negative rates (and costs associated with these). Thus, the study’s findings must be treated with some caution, especially given the small sample size involved.

The study by Marx and colleagues,88 although not claiming to examine cost-effectiveness, nevertheless contains information from which an assessment of cost-effectiveness can be made. The study compared the sensitivity of non-contrast-enhanced MR imaging with GdT1W MR imaging prospectively in 25 patients whose history, physical examination and audiometry results gave rise to a suspicion of acoustic neuroma. Patients were investigated with both protocols; all tumours detected with GdT1W MR imaging were detected with non-contrast-enhanced MR imaging giving 100% sensitivity. The cost of non-contrast-enhanced MR imaging is reported as being one-third of that for GdT1W MR imaging at this institution, although no actual figures are given. The lower cost is attributed to a shorter scan time, absence of contrasting and a reduced number of printed images.

The study’s strength is its prospective design of an unselected group of patients. That the study group exhibited many smaller tumours – as small as 4 mm – adds weight to the contention that non-contrast-enhanced MR imaging is as sensitive as GdT1W MR imaging (the average size of tumours in the group was 1 cm), although the fact that this is a small sample studied by an experienced team and that these are results obtained in a trial, in which greater care is deployed than might be under normal circumstances, may all have affected the outcome. The advantages of GdT1W MR imaging in detecting other pathology are, as the authors point out, also perhaps worth considering. The authors are cautious in recommending non-contrast-enhanced MR imaging over GdT1W MR imaging commenting on the need to ensure that the skills of the team using the technology are adequate. In as much as this was not a cost-effectiveness analysis the study cannot really be criticised for failing to conduct a sensitivity analysis, although one would certainly have been useful. With these caveats in mind the study does provide evidence that a diagnostic algorithm deploying non-contrast-enhanced MR imaging after audiometry may be more cost-effective than one relying on GdT1W MR imaging.

Finally, as with Marx and colleagues,88 the study by Tan117 does not purport to be a cost-effectiveness analysis but it contains information from which inferences regarding cost-effectiveness may be drawn. The study relates to 123 individuals who presented with audiovestibular symptoms indicative of acoustic neuroma. As non-contrast-enhanced MR tests that were negative for acoustic neuroma were not confirmed by GdT1W MR imaging, the sensitivity of non-contrast-enhanced MR imaging cannot be assessed in this study. The authors, however, refer to the literature, including the study by Allen and colleagues,82 to assert that non-contrast-enhanced MR imaging may miss between 6% and 8% of smaller tumours. The cost of GdT1W MR imaging (S$950 or US$627) is based on billable charges at this institution; this compares with a cost of S$400 (US$264) for non-contrast-enhanced MR imaging. The lower charge is attributed to a shorter scan time as a result of not using contrast. The authors are conservative in their conclusions, suggesting that non-contrast-enhanced MR imaging be used for those who cannot afford GdT1W MR imaging, for those whose presenting symptoms indicate a low probability of acoustic neuroma (e.g. vertigo and/or tinnitus without SNHI or when the history suggests that any tumour present is likely to be large) and in the case of elderly and/or medically infirm patients in whom no real harm may result from small tumours missed. GdT1W MR imaging by inference is still seen as the gold standard and the test that should be used when a strict budget or other information does not contraindicate it.

Risk factors

Public concern that the use of mobile telephones might increase the risk of brain tumours because of microwave exposure led to a range of epidemiological studies designed to estimate any possible injurious effects, especially of prolonged use. The proximity of the acoustic nerve to the handset and hence to the radiation field might be expected to place the nerve at particular risk. There is no agreement between studies but it would appear that, at least in the short term (i.e. less than 10 years of mobile phone use), the risk is likely to be small; beyond that period the risk seems to be elevated but it is not sufficiently quantified at present. 123

Occupational noise exposure has also been evaluated as a possible risk factor for acoustic neuroma. Preston-Martin and colleagues124 compared 86 acoustic neuroma patients with 86 control subjects and reported an odds ratio of 2.2 for noise exposure, which increased to 13.2 for noise exposure of more than 20 years. The most recent Swedish study125 examined occupational noise exposure data for 599 cases of acoustic neuroma and 101,756 control subjects, and reported no increased risk of acoustic neuroma even after allowing for a long latency period.

Using material from a population-based tumour registry (National Cancer Registry 1961–79), Haas and colleagues126 evaluated the influence of pregnancy at the time of diagnosis. Observed cases for malignancy and meningioma were the same as for a control population, except for acoustic tumours, the latter presenting more frequently during pregnancy. Although interesting, this finding has not been independently replicated.

Incidence and prevalence

Various attempts have been made to evaluate the incidence and prevalence of acoustic neuroma in the general population. Original attempts were based on hospital autopsy studies. These studies represented a valuable source of tumour material, but their usefulness is restricted by limited availability. Hardy and Crowe127 examined the available autopsy material at the Johns Hopkins Hospital in Baltimore, representing material from 1928 to 1941, and found six tumours in the 250 temporal bone pairs (2.4%). Leonard and Talbot,128 in a later series of similar size from the same institution, revealed acoustic neuromas in 0.8% of subjects. In the Hamburg series based on the Wittmaack Collection,129 1720 temporal bones were collected between 1906 and 1945; 30 tumours were identified, many of them (n = 22) being large tumours. In the Harvard temporal bone collection,130 in 893 temporal bones from 517 individuals, five tumours were diagnosed. The authors considered that ‘the finding of acoustic neuromas in 0.9% of individuals in this series indicates the high incidence of this tumour in the general population’. However, to suggest that findings in a temporal bone collection reflect the incidence in the general population is not tenable given the highly selected provenance and small sample size of the material in such collections. Interestingly, they also commented that ‘the location and size of these tumours indicate that clinical diagnosis would have been difficult or impossible by any method of study’, reflecting the rather crude diagnostic techniques available at that time.

In 1984, a Finnish study131 reported an analysis of 298 temporal bones from 168 cases (129 paired specimens). No occult neuromas were discovered in the material, although one or two cases were expected based on previous autopsy studies. The authors found only one large CPA acoustic neuroma in a patient who died after a neurosurgical operation. A further autopsy study from Copenhagen132 based on a collection of 150 temporal bones revealed eight (2.4%) acoustic tumours.

Tos and colleagues132 have used the above data to infer tumour incidence in the general population (i.e. the number of new cases of acoustic neuroma in 1 year), which resulted in conservative estimates of acoustic neuroma of 8000 tumours per million population, which they recognised as totally unrealistic. Although autopsy studies might at best give an idea of prevalence (i.e. the total number of cases of acoustic neuroma in the population at a given time), they are unlikely to provide data on incidence. Some authors have also confused the terms ‘incidence’ (i.e. the number of new cases of a disease during a given time interval, usually 1 year) and ‘prevalence’ (defined as the total number of cases of a disease in the population at a given time), which has caused confusion in the literature.

The advent of non-invasive diagnostic intracranial imaging by means of CT and MR imaging has afforded another means of assessing the prevalence in the general population. Thus, more recently, epidemiological studies (e.g. from tumour registries and population-based studies), clinical studies based on referral to specialist units and databases of confirmed cases, and exploration of MR imaging findings in large numbers of individuals have been used to estimate tumour incidence and prevalence. We present data from such studies that met the inclusion criteria for this review.

Data from registries

Table 11 details three sets of studies on registry data. Propp and colleagues133 analysed the epidemiology of acoustic neuroma using combined data from central brain tumour registries in the USA (1975–99). The overall incidence of primary nerve sheath tumours increased significantly over the study period, as did the incidence of acoustic neuroma, whereas the incidence of benign schwannomas at other sites decreased or remained static. The annual incidence of acoustic neuroma was found to be between six and eight per million person-years. In their data, acoustic neuroma accounted for 57% of benign intracranial schwannoma but further scrutiny of data from Los Angeles county confirmed that acoustic neuroma accounted for 90% of intracranial nerve sheath tumours. Acoustic neuroma were found to occur with equal incidence in men and women, but the non-white population had a significantly lower incidence than the white population. The authors felt that better diagnostic techniques, especially for the symptomatic elderly, contributed to the apparent increased incidence. They also felt that more accurate coding, avoidance of misclassification and more complete active case ascertainment would provide a much more accurate picture. Adding to the difficulty is that many cases are now ‘diagnosed’ on imaging without having histological confirmation (as biopsy alone would require major surgery), thus the accuracy of the diagnosis in these cases is open to question; in addition, these histologically unconfirmed cases may not be reported to a tumour registry.

In 2005, Evans and colleagues134 reported incidence data from the North West of England, accrued from the major neurosurgical centres and cross-referenced with the regional tumour registry (data from 1990 to 1999). From a sample of 419 sporadic acoustic neuroma and 64 NF2-related acoustic neuroma, they estimated the incidence of acoustic neuroma as 1.04 per 100,000 population per year in the first 5 years of the study period, increasing to 1.4 per 100,000 population in the last 5 years. They also pointed out that more acoustic neuroma than previously considered were the result of NF2, due to a greater awareness of the mosaic forms of this disease. An epidemiological study of the incidence of brain tumours in Devon and Cornwall confirmed that the incidence of cranial nerve tumours (unspecified) was approximately 2.3 per 100,000 person-years, with the peak in incidence between the fifth and sixth decades, the precise incidence of acoustic neuromas not being stated. 149

The leading studies on epidemiology have come from Denmark (population 5.2 million), in large part because of the establishment of a comprehensive database for these tumours in 1975. In this central register, data are entered from all neurological and otological clinics in the country on all confirmed tumours. This has resulted in sequential publications9,122,132,150, covering the time span from 1975 to 2001, offering the most comprehensive data yet available on the epidemiology of these tumours in a stable population of meaningful size.

The incidence of these tumours has changed over the years since data collection commenced. 8,9,122 There has been a dramatic decline in the incidence of giant tumours (> 40 mm extracanalicular extension), from 28% initially down to 1%, with a commensurate increase in tumours with an extracanalicular extension of 11–40 mm; the incidence of small tumours (between 1 and 10 mm extracanalicular extension) has remained steady at 20%. The mean size of the extrameatal portion of the tumour has dropped from 28 to 16 mm. These improvements are largely attributable to the advent of MR imaging and increased accessibility to the technology over time. In particular, the non-invasive nature of MR imaging has resulted in the evaluation of elderly patients who might not otherwise have been fit to undergo more invasive assessments. Also, there has been a reduction in the time taken for patients to seek advice (‘patient delay’) and physician time to respond (‘doctor delay’) because of greater awareness and public education. Patient organisations (acoustic neuroma support groups, tinnitus groups, etc.) have also played a very positive role in creating a much more informed and demanding public.

Clinical studies based on referral to specialist units and databases of confirmed cases

Table 12 summarises the studies that have reported incidence data for populations of patients confirmed to have acoustic neuromas, and Table 13 summarises the prevalence data from papers that examined populations of patients presenting with various audiovestibular symptoms suggesting the possibility of acoustic neuroma.

In Manitoba, Canada, Frohlich and colleagues137 reported an overall incidence of acoustic neuroma of 1.27 cases per 100,000 per year, with differences reported in the incidence and peak age between men and women; however, numbers in this study (n = 69) were small.

Moffat and colleagues,138 evaluating the population in the greater Cambridge area in the UK, estimated an overall annual incidence for 1981–91 of 2.02 per 100,000 per year, which was an exceptionally high incidence for this time period (before routine MR scanning). In 2002,139 an incidence of 0.83 was reported and the authors estimate the overall incidence to be 1.36 per 100,000 per year accounting for changes in the catchment population. Mean tumour size at presentation in the period 1982–2002 (1.8–3.0 cm) showed a slight trend towards an overall decrease, which was paradoxically accompanied by an increase in the numbers of large tumours (e.g. those > 45 mm).

The incidence of acoustic neuroma in the South African population has been reported by Seedat and colleagues140 as being approximately 0.3 per 100,000 population per year; interestingly, racial differences were apparent, with the incidence among the white population being significantly higher than that among the black population. Whether this represents a true racial difference in the susceptibility to develop these tumours or whether it reflects social and economic factors is uncertain; for instance, of the 65 MR imaging units in South Africa, 63 are in the private sector, resulting in restricted access for the poorer (black) population.

Szyfter and colleagues151 undertook a study of the epidemiology of acoustic neuroma in the Polish population based on questionnaires sent to seven neurosurgical and three ENT centres to determine the numbers undergoing surgery in an indicative year (1997–8). In total, 72 patients underwent surgery, equating to an annual surgical incidence of 1.9 acoustic neuroma per million population; however, the true incidence (as distinct from the surgical incidence) of these tumours in the general Polish population could not be assessed.

The prevalence of acoustic neuroma reported in Table 13 for symptomatic populations varies from 0% to 6.1%. The design of each study was different, and the populations studied were heterogeneous in terms of age, sex and presenting symptoms, and firm conclusions cannot be drawn.

MR findings in studies of large numbers of individuals

Table 14 summarises the data from MR studies of large numbers of patients who were being scanned for reasons other than to exclude an acoustic tumour. These have been used to estimate the prevalence of acoustic tumours in the general population. Anderson and colleagues148 retrospectively evaluated 24,246 brain MR studies (19,405 with contrast and 4841 without contrast) and found seven unsuspected cases of acoustic neuroma per 10,000 brain MR imaging studies. They concluded that the true prevalence of acoustic neuroma is likely to be greater ‘than the 1.0 per 100,000 population per year previously reported’, but their data (on prevalence) cannot be used to evaluate the incidence of these tumours. Besides, it is uncertain how representative MR imaging data of this kind is of the general population. Lin and colleagues45 studied 46,414 brain scans on the MR database at the University of California in San Francisco, and eight tumours were discovered incidentally (the scans being undertaken for reasons other than to investigate the presence of an acoustic neuroma). The figures suggested that acoustic tumours may be present in at least 0.02% of the population; however, it is difficult to extrapolate prevalence data in a hospital-based population to the general population.

Hearing impairment

Table 16 summarises the data from the included papers on the hearing impairment at presentation and the audiological findings on investigation.

Unilateral sensorineural hearing impairment

There is a consensus in the literature that the majority of patients who come to be diagnosed with an acoustic neuroma had a principal presenting symptom of a progressive unilateral SNHI. The incidence of SNHI in acoustic neuroma has been reported as over 90%;160,163,164,173 the incidence has been stable over time, specifically comparing presenting symptoms for diagnoses in 1981–93 and 1994–2002. 139 Authors have considered whether tumour size is related to the extent of hearing impairment, but this has not been robustly demonstrated. 164,179 Tumour morphology has been categorised and some associations with the extent of SNHI have been reported152,172 in that, when laterally and medially arising lesions are compared, the extent of SNHI is greater in the laterally arising tumours. 172

Sudden hearing impairment

There are a number of other symptoms that can be associated with the diagnosis of acoustic neuroma. Sudden SNHI is a surprisingly prevalent condition, affecting perhaps 7500 individuals in the UK each year. 183 The proportion of acoustic neuroma patients who experience a sudden hearing impairment and who have this as their principal presenting symptom is variable, ranging from 1.7%171 to 27.0%. 159 This may represent variance in the extent to which sudden hearing impairment is seen as requiring urgent otological consutation. 184 Interestingly, of those patients with sudden SNHI, it has been reported that approximately 2% are diagnosed with an acoustic neuroma when investigated radiologically. 144,161

Normal hearing

There are reports in the literature of patients being diagnosed with acoustic neuroma but having normal hearing. Analysis is confounded by variation in the definition of normal hearing, some choosing 20 dBHL174 and others 25 dBHL as the cut-off. 165 Some authors consider normal hearing as being symmetrical, hence allowing for some age-appropriate hearing impairment. 182 Unsurprisingly, there is marked variability in reports of the incidence of normal hearing in acoustic neuroma patients, ranging from 5.0%177 to 12.5%. 176 In papers in which normal hearing is strictly defined, the incidence is lower. 181

Table 17 summarises the data on tinnitus and vestibular symptoms at presentation and/or as determined on further investigation.

Tinnitus

Several studies152,164,175 have reported that the incidence of tinnitus as a principal presenting symptom of acoustic neuroma ranges from 8% to 13%. As stated above, this is at odds with the prevalence of tinnitus in this patient group, which has been reported to be between 60% and 83%,59,158,160,163 and this has led to a proposal that tinnitus associated with a acoustic neuroma may evoke less distress than a clinician would predict. 22,183

Further exploration of the data of Moffat and colleagues138 by Baguley et al. 180 reviewed the characteristics of 941 patients with a radiological diagnosis of unilateral sporadic acoustic neuroma. Of these, 717 experienced tinnitus (76%) and in 114 (12%) tinnitus was the principle presenting symptom. Statistically significant associations were found between tinnitus presence/absence and tumor size (p = 0.012) (although these were non-linear) and type of hearing impairment (progressive, sudden, fluctuant, nil), with a tendency for patients without hearing impairment to be less likely to experience tinnitus. Statistically significant associations were identified between classification of tinnitus severity and age at diagnosis (p < 0.001) (greater age being associated with greater tinnitus severity), abnormal findings on caloric testing (p = 0.01) (abnormal calorics being associated with greater tinnitus severity) and tinnitus as a principal presenting symptom (p < 0.001) (this being associated with greater tinnitus severity). The common sense expectation that those patients who indicate tinnitus to be their principle presenting symptom experience more severe tinnitus than those who do not was supported.

Imbalance

A further proportion of patients describe symptoms of imbalance at presentation, which have been reported to include rotary vertigo, unsteadiness and imbalance. There is a lack of consistency in how these terms are applied in the literature regarding patients with acoustic neuroma, and this is a significant impediment to interpretation. There is evidence that vestibular symptoms in acoustic neuroma tend to be mild22,160 and involve non-specific dysequilibrium. 170,174,182 A small proportion of patients, between 10%176 and 19%170 of the patient population with acoustic neuroma, are described as experiencing rotary vertigo. Relatively few patients with acoustic neuroma present with a primary complaint of symptoms of imbalance, reported as between 7% and 26% when the principle presenting symptom is robustly defined. 139,164,168,172,175

Table 18 summarises the data on all other symptoms at presentation or as determined by further investigation. These data illustrate the relative frequency of trigeminal (Vth) nerve involvement, a nerve that supplies sensation to the face and eyes. As an acoustic neuroma expands superiorly, the sensory fibres of this nerve are particularly vulnerable to compression by the tumour. Hence, impairment of facial sensation (loss of sensation, paraesthesia, etc.) or loss of sensation on the cornea are typically seen in larger tumours. A small number of patients may even present with excruciating facial pain, evoking a misdiagnosis of trigeminal neuralgia. The facial nerve, although more intimately related to the tumour, is less frequently involved clinically, presumably because of the greater resilience of motor nerve fibres to the effects of compression. Indeed, the presence of a facial paralysis in a patient presenting with what appears to be an acoustic neuroma should lead to the consideration of other petrous apex pathology (e.g. non-acoustic tumours such as cholesteatoma). The evidence also demonstrates that the facial nerve may appear to have normal function to an examining clinician but may, on investigation, exhibit signs of weakness.

Symptoms and tumour characteristics

The association between tumour characteristics and symptom profile has been considered. The influence of tumour size has been analysed and the observation made that small tumours (< 1 cm diameter) may be associated with fewer symptoms;185 however, the most rigorous analysis failed to demonstrate any association between size and symptoms. 164

Tumour morphology has also been considered. Medially arising tumours have been demonstrated to be associated with a higher incidence of symptoms deriving from cerebellar, trigeminal nerve and brainstem involvement,152 such as ataxia, disdiadokinesis and decreased facial and corneal sensation. Conversely, an association between medial acoustic neuroma and normal hearing has been reported. 170,172

Studies have failed to demonstrate a consistent correlation between the pattern or degree of hearing impairment and any aspect of tumour morphology. Many studies demonstrate a worsening of hearing threshold over time159,186–195 but relatively few are able to report a statistically significant correlation with growth. 179,186,190–192

Diagnosis of asymptomatic tumours

Table 14 summarises the data from two studies reporting on large series of MR scans. Anderson and colleagues148 retrospectively evaluated 24,246 brain MR imaging studies (19,405 with contrast and 4841 without contrast) and found 17 incidental acoustic neuromas or approximately seven per 10,000 routine brain MR studies being undertaken for reasons other than the exclusion of an acoustic neuroma. A further study at the University of California in San Francisco45 analysed the MR imaging database of 46,414 patients who had brain scans with no documented audiovestibular symptoms, and revealed eight patients with incidental acoustic neuromas. Three of these patients were found to have audiovestibular symptoms on enquiry after diagnosis; three patients had asymmetry on audiometry at 4 kHz, with otherwise normal audiometry for age in the remaining patients. Tumour size ranged from 3 to 28 mm.

Duration of symptoms and delay to diagnosis

An interesting consideration is that of the duration of symptoms. This is a complex area as some of the symptoms of an acoustic neuroma are insidious and may not be given due attention by the patient or GP. Van Leeuwen and colleagues184 studied a series of 164 patients who underwent surgical removal of a unilateral acoustic neuroma in Nijmegen, the Netherlands, between 1980 and 1992. The mean delay from initial symptom onset to seeing a specialist was 35.7 months [standard deviation (SD) 62.2, range 0–468 months]; such delay could be due to patient reluctance to seek advice or to the GP not referring on for specialist consultation. A further mean delay of 15.2 months (SD 36.2, range 0–242 months) was evident from specialist input to diagnosis. Although there is an obvious caveat in that the period studied predates the modern era of straightforward access to MR imaging, this work does demonstrate the complexity of the situation regarding the duration of symptoms and the delay before definitive diagnosis. Moffat and colleagues139 report the length of history in a cohort of acoustic neuroma patients undergoing surgery in the years 1981–3 as 42 months (SD 51) and from 1994 to 2004 as 45 months (SD 108.64). These figures are not statistically significantly different (p = 0.4726, Mann–Whitney U-test) and therefore access to MR imaging may not have influenced this issue.

Within the literature, the complexity of the issue of length of symptoms and diagnostic delay has not been adequately addressed and, although some papers report a duration of symptoms,59,152,163,164 the lack of rigour hampers systematic analysis. Specifically, a possible association between size of tumour and duration of symptoms has either been excluded164 or been suggested and not statistically verified. 59,163

Measurement

An important issue in measuring acoustic neuromas is how their size is measured or calculated even within the same imaging technique. The most common procedure is to measure the maximum diameter of the tumour; however, this entails gross simplification, because tumours are three-dimensional structures and often irregular in shape. Moreover, a change in the diameter implies an exponential change in the volume of the lesion. For example, when the diameter has doubled, the volume has increased eight times, assuming that the shape of the tumour has not changed. 197 Rosenberg237 compared rigorous computer analysis of size and growth rate with the radiologists’ usual measurements of maximal diameters. 237 The computer analysis used a new method of measurement: equivalent diameter as the diameter of a perfect circle that corresponds to the measured tumour area (largest area in both axial and coronal plane). Rosenberg237 concluded that the measurement of the maximal tumour diameter on MR imaging is still a reliable method for following acoustic neuroma growth and that there is no need to perform a rigorous analysis of tumour size to determine whether the tumour is growing significantly.

The confusion and possible bias in measuring acoustic neuroma size is confounded when we take into account that, in addition to the maximal diameter method, numerous other descriptions of measurement have been reported by different centres. These are summarised in Table 20.

To limit the variance, the American Academy of Otolaryngology – Head and Neck Surgery – recommended that size should be measured by taking the square root of the product of the two standardised diameters in the axial plane. Although this method was introduced in 1995, very few centres have been using it. In addition, the procedure has its own drawbacks. One is that the superior and inferior extents of the tumour as well as the volume are not taken into account. Another is that this procedure is intended to classify tumours in such a manner as to allow comparison of the results of surgery, not to establish changes in the size over time. 197

Apart from the numerous different formulae used to measure size and/or volume, there is also evidence that the reliability of measuring acoustic neuromas is far from satisfactory. Marshall and colleagues241 assessed the inter- and intraobserver reliability of measuring acoustic neuromas, including in their study the method proposed by the American Academy of Otolaryngology, and concluded that, in routine clinical practice, differences in tumour size of the order of 2 mm cannot be reliably measured, even by the same radiologist. On the other hand, in another study,197 when the maximal surface of the tumour in the axial plane and SD were calculated – SD was taken as the indicator of growth or shrinkage – interobserver agreement was found to be very high and clinical judgements agreed with surface axial computer measurements. However, this study had several weaknesses including a small number of patients (n = 44) and the fact that interobserver agreement was dependent on the threshold set (e.g. if growth is defined as a large difference of 3 mm, interobserver agreement becomes very high, and vice versa). Reported growth of acoustic tumours should therefore be interpreted with caution, especially if this is the criterion for recommending treatment.

Definition of growth

It might be thought that any growth, even that which is minimal, should be considered as such; however, the reality is quite different as terms such as ‘important growth’, ‘definite growth’, ‘probable growth’, ‘minimal growth’, ‘slow or rapid growth’ and various other definitions have been used in the literature. Table 21 summarises the various definitions that have been used to report growth. For example, an increase of 1 mm in size may be considered as growth in one study198 and ‘stable’ in another. 186 In addition, regression has been separately reported in some papers,186,198,210 whereas in others221 stability and regression count as one. Finally, mathematical formulae, both simple and complex, have been used to assess growth and growth rates.

Radiosurgical tumour control

In addition to the problems related to measuring tumour size and growth, comparisons between conservative management and radiosurgery have introduced new challenges. Radiosurgeons also have various definitions of tumour control, i.e. non-growth. A widely accepted radiosurgical definition of tumour control is that of Flickinger and colleagues242 who described tumour control as a less than 1-mm increase in tumour diameter in any two directions or 2 mm in one direction. Another widely accepted definition of tumour control is freedom from surgical resection or lack of surgical intervention. In a study that compared conservative management in patients from a single centre with radiosurgery patients from the literature,229 using Flickinger’s242 definition of control, the control rate in their population of 71 patients after a mean of 3.1 years was 87%, which was only 6% worse than Flickinger’s242 results after a 24-month median follow-up period with 313 patients, and 1% better than that reported by Iwai and colleagues243, the radiosurgical study with one of the longest median follow-up periods (60 months).

In another report,223 data from a study of conservative management were compared with data from three studies of radiosurgery. The results revealed that the risk of growth was statistically higher in the conservative management group, whereas stability was comparable. Regarding other factors, the risks of losing useful hearing, developing trigeminal nerve dysfunction and/or developing hydrocephalus requiring shunt placement were statistically higher after radiosurgery. Ultimately, however, without standardised ways of measuring tumour growth and reporting the results, a fair comparison of radiosurgical results with the natural history of acoustic neuromas is not possible. 229

Growth pattern during conservative management

Studies that have compared the various methods of assessing growth have revealed that reports of growth depend on the measurement method used and the respective differences may be huge. In one study that compared different measurement methods, tumour volume doubling time assessed by a Bayesian partial volume tissue segmentation method demonstrated tumour growth in 80% of cases that appeared to show no growth when measured by manual segmentation techniques. 207,209 In another study, volume calculation was carried out manually and compared with a computerised method; the difference was statistically significant and in some cases as much as double or triple the volume. 215

Although the literature reports different methods of measuring and comparing growth, any review may only be a rough estimate of what actually happens in ‘real life’. Table 22 summarises the results of the papers included in this review that reported the growth pattern of acoustic neuroma during conservative management, that is, ‘wait and rescan’. Besides the absolute growth, an estimate of the growth rate of these tumours will contribute to the decision of when, if at all, to intervene. A very slowly growing tumour may have a completely different management in comparison to a rapidly growing tumour, which may result in neurological complications in the short term. Of course, the task of assessing growth rate has the same weaknesses mentioned previously. In addition, the various studies on growth rate have used different criteria or methods of reporting. For example, some studies report the growth rate of tumours that were increasing in size, whereas other studies report the growth rate of all tumours. Finally, some studies introduce the term ‘tumour doubling time’, which is another type of growth rate. Table 22 summarises the outcomes of papers reporting on acoustic neuroma growth rate.

To facilitate comparisons within the table we have ordered the studies as follows: (1) growth rate applies to all tumours, (2) growth rate applies only to tumours that grow, (3) application of growth rate to tumour state is not reported or is not clear or (4) growth rate has another definition.

In a meta-analysis comprising 1244 patients in 19 studies, 51% of tumours showed no growth, 43% of tumours grew and 6% showed regression. 194 In another meta-analysis of 555 patients in 13 studies, 54% showed growth (range 14–74%) and 26% had rapid growth of more than 2 mm/year (range 0–47%). 240 In a systematic review of 1340 patients, 46% showed growth (95% CI 43–48%, range 15–85%) and 8% regression (95% CI 6–10%). 16 In a small group of four prospective studies, 29% of tumours showed growth (95% CI 21–37%). 16 In another systematic review of 879 patients, 51% of tumours grew and 4% showed regression. 239 An additional problem is that there are various patterns of growth, and a tumour that shows growth may stop doing so and vice versa. In a study215 that assessed these patterns, tumours were classified into five categories: continuous growth (15%), negative growth (5%), growth followed by shrinkage (40%), negative growth followed by growth (20%) and no variation in tumour size (20%). In another study,228 five patterns of growth were also found: continuous growth (25%), stability/growth (16%), stability (50%), negative growth (7%) and growth/stability (3%). Finally, a third study237 reported six patterns: no growth (35%), growth only (21%), no growth to growth (13%), growth to stability (13%), growth to regression (10%) and regression (8%).

Growth rate

In one meta-analysis180 of 793 patients in 13 studies, the mean growth rate was 1.9 mm/year (range 0–10 mm/year) and in another meta-analysis240 of 508 patients in 13 studies, the mean growth rate was 1.8 mm/year (range 0.5–3.2 mm/year). In a systematic review of 964 patients, the mean growth rate was 1.2 mm/year (range 0.4–2.9 mm/year). 16 In another systematic review of 879 patients, the mean growth rate was 1.8 mm/year (range 0.3–30 mm/year) (including those with regression the mean growth rate was –3.9 to 30.0 mm/year). 239 It seems that the mean growth rate for all tumours varies between 1 and 2 mm/year. Including only tumours that grow, the growth rate varies between 2 and 4 mm/year. However, there are cases with significant regression or exceptional growth that may exceed 18 mm/year.

Time frame of growth

Irrespective of growth rate, the time scale in which this growth may occur is very important. For example, if growing in a specific time period (e.g. within 1 or 2 years from diagnosis) predicts later tumour behaviour, this would influence management and follow-up intervals. Therefore, some studies have analysed the time frame of growth (Table 23).

Many of these studies suggest that the first year after diagnosis is crucial for determining the pattern of tumour growth, but it is clear from several other reports that this may not be the case for some patients. For example, a patient with a 4-mm extension in the CPA did not have any growth for 6 years and then the tumour started to grow requiring radiosurgery. 219 Another case showed an exceptional growth in just 7 months from diagnosis (17 mm),221 whereas another small tumour was stable for 6 years and then showed a continuous growth for the next 5 years. 237

Predictors of growth

Determinants or predictors of growth would be very helpful in management planning and patient counselling. Therefore, many studies have attempted to identify such factors and apply them to clinical practice (Table 24).

It seems that most studies fail to identify predictors of growth. Some studies have found large initial size to be a determinant of later growth although the opposite has also been reported. In a systematic review of 1340 patients16 no correlation was found between tumour growth frequency and patient age at diagnosis or follow-up duration; prospective design and serial MR imaging were associated with lower tumour growth frequency and larger tumours were associated with a lower risk of enlargement.

A meta-analysis of 10 studies and 620 patients did not find any predictive factors for tumour growth, whereas in four studies including 255 patients positive growth at 1 year was found to be predictive of future growth. 194 In another meta-analysis of 555 patients (13 studies) no statistically significant differences were found regarding mean age and mean initial size between growth and non-growth groups. 240

Failure of conservative management mainly due to tumour growth

Conservative management or the ‘wait and see’ de facto policy fails when treatment is decided. However, this does not necessarily mean that the tumour has grown because patients may change their minds and have treatment for other reasons such as personal preference, symptoms, etc. Table 25 summarises the ‘failure’ rates of conservative management taken from the literature.

In a meta-analysis of 15 studies and 1000 patients,194 20% of patients required treatment. In another meta-analysis of 13 studies and 1000 patients,240 26% of patients required treatment (range in these studies 0–50%). Finally, in a systematic review of 1340 patients,16 18% (95% CI 16–21%) required treatment. It seems that approximately one in every four or five cases fails conservative management; however, the range is very wide (0–50%), suggesting that there are many factors influencing the percentage of patients who finally have treatment, such as length of follow-up, selection of patients, counselling, etc.

AMED

1. acoustic neuroma$.mp.

2. acoustic lesion$.mp.

3. acoustic neurinoma$.mp.

4. vestibular nerve tumor$.mp.

5. vestibular nerve tumour$.mp.

6. vestibular nerve lesion$.mp.

7. vestibular schwannoma$.mp.

8. intracranial tumor$.mp.

9. intracranial tumour$.mp.

10. intracranial lesion$.mp.

11. cerebellopontine angle lesion$.mp.

12. cerebellopontine angle tumor$.mp.

13. cerebellopontine angle tumour$.mp.

14. or/1–13

Filters were not used on the AMED database.

BIOSIS

British Nursing Index

1. ACOUSTIC ADJ NEUROMA$

2. ACOUSTIC ADJ NEURINOMA$

3. ACOUSTIC ADJ TUMOR$

4. ACOUSTIC ADJ TUMOUR$

5. ACOUSTIC ADJ LESION$

6. VESTIBULAR ADJ NERVE ADJ TUMOR$

7. VESTIBULAR ADJ NERVE ADJ TUMOUR$

8. VESTIBULAR ADJ NERVE ADJ LESION$

9. VESTIBULAR ADJ SCHWANNOMA$

10. INTRACRANIAL ADJ TUMOUR$

11. INTRACRANIAL ADJ TUMOR$

12. INTRACRANIAL ADJ LESION$

13. CEREBELLOPONTINE ADJ ANGLE ADJ LESION$

14. CEREBELLOPONTINE ADJ ANGLE ADJ TUMOR$

15. CEREBELLOPONTINE ADJ ANGLE ADJ TUMOUR$

16. CPA

17. 1 OR 2 OR 3 OR 4 OR 5 OR 6 OR 7 OR 8 OR 9 OR 10 OR 11 OR 12 OR 13 OR 14 OR 15 OR 16

Filters were not used on the BNI database.

CRD

acoustic AND neuroma* OR acoustic AND lesion* OR acoustic AND neurinoma* OR vestibular AND nerve AND tumor* OR vestibular AND nerve AND tumour* OR vestibular AND nerve AND lesion* OR vestibular AND schwannoma* OR intracranial AND tumor* OR intracranial AND tumour* OR intracranial AND lesion* OR cerebellopontine AND angle AND lesion* OR cerebellopontine AND angle AND tumor* OR cerebellopontine AND angle AND tumour*

Filters were not used on the CRD databases.

CINAHL

The Cochrane Library including Cochrane Reviews, CENTRAL, DARE, HTA and NHS EED

DH-Data

1. ACOUSTIC ADJ NEUROMA$

2. ACOUSTIC ADJ NEURINOMA$

3. ACOUSTIC ADJ TUMOR$

4. ACOUSTIC ADJ TUMOUR$

5. ACOUSTIC ADJ LESION$

6. VESTIBULAR ADJ NERVE ADJ TUMOR$

7. VESTIBULAR ADJ NERVE ADJ TUMOUR$

8. VESTIBULAR ADJ NERVE ADJ LESION$

9. VESTIBULAR ADJ SCHWANNOMA$

10. INTRACRANIAL ADJ TUMOUR$

11. INTRACRANIAL ADJ TUMOR$

12. INTRACRANIAL ADJ LESION$

13. CEREBELLOPONTINE ADJ ANGLE ADJ LESION$

14. CEREBELLOPONTINE ADJ ANGLE ADJ TUMOR$

15. CEREBELLOPONTINE ADJ ANGLE ADJ TUMOUR$

16. CPA

17. 1 OR 2 OR 3 OR 4 OR 5 OR 6 OR 7 OR 8 OR 9 OR 10 OR 11 OR 12 OR 13 OR 14 OR 15 OR 16

Filters were not used on the DH-Data database.

EMBASE