ZENON MARIAK, M.D., PH.D., JAROSLAW KREJZA, M.D., PH.D., MIROSLAW SWIERCZ, PH.D., KAZIMIERZ KORDECKI, M.D., PH.D., AND JANUSZ LEWKO, M.D., PH.D.

Transcription

1 J Neurosurg 96: , 2002 Accuracy of transcranial color Doppler ultrasonography in the diagnosis of middle cerebral artery spasm determined by receiver operating characteristic analysis ZENON MARIAK, M.D., PH.D., JAROSLAW KREJZA, M.D., PH.D., MIROSLAW SWIERCZ, PH.D., KAZIMIERZ KORDECKI, M.D., PH.D., AND JANUSZ LEWKO, M.D., PH.D. Departments of Neurosurgery and Radiology, Bialystok Medical Academy, Bialystok, Poland B Object. The value of transcranial Doppler ultrasonography for the detection of middle cerebral artery (MCA) spasm has been asserted. None of the published studies, however, has adequately scrutinized the overall diagnostic accuracy of this procedure. There are only sporadic reports concerning the utility of transcranial color Doppler (TCCD) ultrasonography, although this method has been proved to be more precise. In this study the authors attempted to estimate the performance of TCCD ultrasonography in detecting MCA narrowing by using receiver operating characteristic (ROC) curve analysis, based on TCCD studies obtained in a relatively large, randomly selected population of patients. Methods. Transcranial color Doppler ultrasonography studies were obtained in 100 consecutive patients (54 men and 46 women ages years, median age 50 years) routinely referred by neurosurgeons for intraarterial angiography. The M 1 segment of the MCA was insonated using a 2.5-MHz probe via a temporal acoustic window, and angle-corrected flow velocities were obtained. Angiographically depicted vasospasm was graded as none, mild ( 25% vessel caliber reduction), and moderate to severe ( 25% vessel caliber reduction). The effectiveness of TCCD ultrasonography in diagnosing MCA spasm was evaluated by calculating the areas under the ROC curves (A z ). Of the 200 MCAs examined, 173 were successfully visualized with the aid of TCCD ultrasonography. Mild vasospasm was angiographically diagnosed in 15 arteries and moderate-to-severe vasospasm in 28. The best-performing TCCD parameter for the detection of MCA narrowing was revealed to be peak systolic velocity. The A z value for moderate-to-severe vasospasm only was 0.93 and that for all vasospasms was 0.8. The best efficiency, that is, the optimal tradeoff between sensitivity and specificity in diagnosing vasospasms, was associated with a peak systolic velocity of 182 cm/second. Conclusions. The performance of TCCD ultrasonography in the diagnosis of advanced MCA narrowing is very good, and is acceptable for all vasospasms. The best-performing parameter was peak systolic velocity. KEY WORDS transcranial color Doppler ultrasonography vasospasm blood flow velocity receiver operating characteristic Abbreviations used in this paper: AVM = arteriovenous malformation; A z = area under the curve; BFV = blood flow velocity; DS = digital subtraction; GCS = Glasgow Coma Scale; MCA = middle cerebral artery; ROC = receiver operating characteristic; SAH = subarachnoid hemorrhage; TCCD = transcranial color Doppler; TCD = transcranial Doppler. LOOD flow velocity is known to increase with vessel narrowing. 53 This phenomenon has given rise to attempts to use TCD ultrasonography for the diagnosis of cerebral vasospasm. 1,2,16,21 Numerous researchers have concentrated their efforts on establishing a threshold BFV that would reliably differentiate cerebral vasospasm from normal states of the artery. 1,15,16,21,40 42,61,62 For the MCA, the most commonly cited velocity threshold for vasospasm is 120 cm/second. 1,21,43,46,57,58,62 Some investigators, however, contest this value and propose instead that the vasospasm threshold falls within the range of velocities from 120 to 200 cm/second. 63,65,66 According to other concepts, it is not possible to establish a vasospasm threshold for the general population. 22,31,52,65 A critical BFV value for confirming vasospasm would therefore be 200 cm/second, and a BFV value for excluding the condition would be 120 cm/second. 63,65 This has raised concerns about the utility of TCD ultrasonography in the reliable detection of vasospasm. 38,52 The difficulties in establishing a universal BFV threshold for diagnosing vasospasm may arise for biological reasons: vasospasm is not an all-or-none condition, and flow in the vessel in question is influenced by many factors in a variety of ways. 11,18,45,51,67 Another source of variability is strictly methodological. Conventional TCD ultrasonography is a blind method, that is, it prevents the operator from controlling the angle between the visualized vessel and the ultrasound beam, which can be a source of error in velocity measurement. 2,17,32,35 Several recent publications have documented the better accuracy and repeatability of blood flow measurements obtained using TCCD ultrasonography. 7,9,37,44,47,60 In contrast to the blind method, TCCD ultrasonography allows the parenchymal structures to be outlined and the visualized vessel to be examined. 13,35 This enables more consistent placement of the sample volume in the vessel in question, as well as J. Neurosurg. / Volume 96 / February,

2 Z. Mariak, et al. TABLE 1 Diagnostic performance of TCCD ultrasonography in MCA vasospasm as determined by comparing the A z * Moderate-to-Severe Vasospasm angle corrections for the blood velocity to be measured. Thus, the velocities obtained are closer to their true values and they have been shown to be higher compared with those obtained using TCD ultrasonography. 7,60 Consequently, the velocity threshold for vasospasm, which was established with conventional TCD ultrasonography, should not be used for TCCD studies. To our knowledge, only one paper has been published to date in which the focus was on TCCD criteria for the diagnosis of cerebral vasospasm; however, along with the bulk of research on TCD ultrasonography, it shares the assumption that there exists only one velocity value that reliably detects vasospasm. 54 It is apparent that the lower the threshold, the better the sensitivity, whereas the higher the threshold the better the specificity of the diagnostic test at the expense of sensitivity. 19,55 To assess the diagnostic performance of the test, there is no need to choose any particular decision threshold, but rather to explore the entire range of relationships between true-positive results and false-positive results over all decision thresholds. 19,68 Such overall evaluation of the sensitivity and specificity of the test for a broad range of velocities is provided by plotting the ROC curve. 19,24,68 To our knowledge this methodology has not been applied to the diagnosis of cerebral vasospasm, either with conventional or color Doppler methods, although ROC plots are widely used to assess the diagnostic performance of various tests in other areas of clinical medicine. 68 Materials and Methods All Vasospasm Velocity A z total (SEM) A z partial A z total (SEM) A z partial peak systolic (0.0280) (0.0417) mean (0.0338) (0.0431) end diastolic (0.0359) (0.0452) *A z partial = A z confined to the high sensitivity region (80%); A z total = A z in the entire range of sensitivity and specificity; SEM = standard error of the mean. Significantly different from other values in the column. Background on Patients in Whom Studies Were Obtained Transcranial color Doppler ultrasonography is routinely performed at our hospital in all patients who suffer an intracranial hemorrhage and ischemic stroke, as well as after clipping of an aneurysm. Some patients are subsequently referred for DS angiography. All patients included in this investigation underwent TCCD studies immediately before angiography. The program was approved by the Medical Academy s ethics committee. Each patient or members of his or her family gave fully informed consent. Studies were obtained in consecutive patients (54 men and 46 women ages years, median age 50 years) who were routinely referred by neurosurgeons for intraarterial DS angiography between July 1, 1999, and November 15, Ultrasonography was performed in 46 patients who had suffered nontraumatic SAH, 31 who had suffered intracerebral hemorrhage, six who had severe headaches, two who had brain tumors, two who suffered ischemic stroke, and one who had suffered traumatic SAH. Twelve patients were examined after aneurysm clipping. The distribution of the GCS-based 64 clinical grades at admission were as follows: Grade 1 (GCS score of 15), 22 patients; Grade 2 (GCS score of 12 14), 46 patients; Grade 3 (GCS score of 9 11), 16 patients; Grade 4 (GCS score of 6 8), 12 patients; and Grade 5 (GCS score of 3 5), four patients. Angiograms were obtained mostly during the acute stage of illness, except in patients examined after aneurysm clipping. Although the TCCD examinations were repeated at least every 2 days during hospitalization, only the one TCCD study performed just before angiography in each patient was included in this investigation. As a control, eight healthy volunteers (four men and four women, all 23 years old) were examined 10 times (once a day for 10 subsequent days) to obtain intraobserver reproducibility levels for the measurements. 37 Transcranial Color Doppler Ultrasonography Intracranial cerebral arteries were studied using an ultrasound scanner (model SSA 140 A; Toshiba Medical Systems, Tokyo, Japan) equipped with a 2.5-MHz 90 phased-array probe for both B-mode and Doppler imaging. The anterior cerebral artery, MCA, and posterior cerebral artery were studied via temporal acoustic windows by using methods described previously; 13,35 however, only data from the MCAs were submitted to further analysis. The M 1 segments were insonated by placing a 3-mm-wide sample volume within a straight arterial segment located 10 mm from the carotid artery bifurcation to avoid the risk of sampling a branch of the segment. 36 The mean, peak systolic, and end diastolic velocities were calculated by tracing the maximum frequency envelope of the Doppler waveform. The angles of insonation were measured visually for both MCAs. This allowed the angle-corrected BFVs to be measured. Angiographic Studies Selective intraarterial DS angiography was performed via the Seldinger approach through the femoral artery in both internal carotid arteries and at least one vertebral artery in every patient. Standard DS angiography images included anteroposterior, lateral, and one oblique view, which were obtained routinely at injection rates of 6 ml/second and filming rates of three frames per second; the field of view was 30 cm for all views. Two neuroradiologists who were unaware of the ultrasonography findings reviewed all angiograms to detect and quantify the cerebral vasospasm. The view showing the most severe MCA narrowing was used in the comparison with TCCD findings. All measurements were performed using the calipers on the digital display. The resolution of this method is 0.01 mm. The angiographic diagnosis of vasospasm was made using criteria widely adopted by other authors. 3,59,65 Briefly, to quantify the degree of vasospasm the M 1 segment of the MCA was measured at its point of maximum reduction and compared with the normal section of the artery adjacent to the narrowed segment. Angiographic vasospasm was graded as none, mild ( 25% vessel caliber reduction), and moderate-to-severe ( 25% vessel caliber reduction). Where possible, the diameter of the vessel in question was also compared with the contralateral MCA to facilitate classification. Discordant readings by the observers were resolved by consensus. Separate ROC analyses were performed for a subgroup of moderate-to-severe vasospasms and for all others, that is, mild plus moderate-to-severe vasospasms. Statistical Analysis The entire span for each velocity found empirically in all the arteries examined (peak systolic, mean, and end diastolic) was grouped for the same number of 40 bins, to reduce possible error when comparing areas under the curve for particular velocities. 30 The ROC curve was constructed by graphing the sensitivity on the ordinate as a function of the false-positive rate (or 1 specificity), for all 40 cutoff values of BFV. The sensitivity for each of the 40 velocity thresholds was calculated as a quotient of the number of arteries with true spasm in which velocity exceeded a particular threshold by the number of arteries with angiographically confirmed spasm (a ratio of true-positive vasospasms to the number of arteries with angiographically detected vasospasm). Similarly, specificity was computed as a quotient of the number of non-narrowed arteries in which 324 J. Neurosurg. / Volume 96 / February, 2002

3 Transcranial color Doppler ultrasonography in diagnosis of MCA spasm FIG. 1. Graphs showing ROC curves for peak systolic, mean, and end diastolic BFVs at different levels of vasospasm. Left: The ROC curves for 25% or greater MCA narrowing (moderate-to-severe vasospasm). The high A z values indicate very good performance of TCCD ultrasonography in diagnosing vasospasm. The area of high sensitivity of the ROC curve is above the horizontal line, at 0.8 (80%) sensitivity. Right: The ROC curves for all vasospasm. The A z value places the test in the range of moderate diagnostic performance. velocity was lower than the velocity threshold by the number of arteries with no angiographically confirmed vasospasm. The statistical significance of the difference between the particular areas under the ROC curves was calculated using the paired t-test, with necessary correction for correlated data. 19,25,68 The intraobserver reproducibility was evaluated by comparing measurements from separate days. Two statistical measures of reproducibility were calculated: the first, repeatability, was computed by averaging the coefficient of the day-to-day Pearson correlation; the second, the coefficient of variance, was calculated as the ratio between the variance explained by day-to-day differences and the total variance. A one-way analysis of variance for univariate repeated measures was also performed. J. Neurosurg. / Volume 96 / February, 2002 Results Of 200 MCAs examined, 173 were successfully visualized with the aid of TCCD ultrasonography. Images of 22 arteries in 11 patients could not be obtained because of the inadequacy of the temporal window. In five more MCAs no flow was detected, despite the adequate visualization of the parenchymal structures. These patients were excluded from the ROC study, as were another four in whom large AVMs were detected on angiographic studies. Ultimately, TCCD findings in 160 arteries in 80 patients were included in the construction of the ROC plots. Angiographic studies were of sufficient quality in all patients to determine the presence of vasospasm. In the subgroup of 80 patients included in the ROC analysis, mild vasospasm ( 25% MCA narrowing) was diagnosed from angiographic findings in 15 arteries and moderateto-severe vasospasm ( 25% of MCA narrowing) was diagnosed in 28 arteries. Forty-one narrowed arteries were found in patients with intracranial hemorrhage, one in a patient with ischemic stroke, and one in a patient with a severe headache. Vasospasm was also diagnosed in two of the 11 patients, in whom inadequate temporal windows were present. In patients with intracerebral hematoma, peak systolic, mean, and end diastolic BFVs were on average 20, 15, and 9 cm/second, respectively, which were lower than those recorded in the remaining patients, but the differences did not reach statistical significance (nonpaired t-test). The ROC curves for moderate-to-severe and all vasospasms are shown in Fig. 1. The A z s for the two groups of vasospasm are also given in Table 1. By comparing these areas one can evaluate the diagnostic performance of a particular Doppler waveform parameter for the detection of vasospasm of different severity. Although the figures show the entire area under the curve, partial areas corresponding to regions of high sensitivity of ROC plots are also given in Table 1. The sensitivity threshold for the partial area index computed was chosen to be 80%. From Fig. 1 and the partial area indices given in Table 1 it is seen that the diagnostic performance of TCCD ultrasonography is very good for moderate-to-severe vasospasm and approximately 15% lower for all vasospasms. The difference, assessed with a nonpaired t-test, is statistically significant. The peak systolic velocity performs best as a diagnostic parameter in both groups of vasospasm, although only in the moderate-to-severe vasospasm group is the peak systolic velocity shown to be significantly better than the two remaining velocities, as assessed with the paired t-test. Figure 2 shows the ROC curves for the peak systolic velocity based on Fig. 1 (moderate-to-severe and all vasospasms: left and right, respectively). Some arbitrary values of peak systolic velocity are localized on these curves, which enables both the sensitivity and specificity of these 325

4 Z. Mariak, et al. FIG. 2. Graphs showing ROC curves for peak systolic velocity at different levels of vasospasm. Left: The ROC curve for moderate-to-severe MCA narrowing with several velocity thresholds marked to diagnose vasospasm. The best efficiency of the TCCD ultrasonography was found to be associated with a peak systolic velocity threshold of 182 cm/second. Right: The ROC curve for peak systolic velocity for all vasospasm. The best efficiency of TCCD ultrasonography was found for a peak systolic velocity threshold of 182 cm/second. S = second; SE = standard error of the mean. thresholds to be delineated in the detection of vasospasm. For example, in the group with moderate-to-severe vasospasm, a low velocity threshold of 140 cm/second results in 93% sensitivity and only 63% specificity, whereas a high velocity threshold of 220 cm/second results in only 50% of the sensitivity and in a very high specificity, approaching 99% (Fig. 1 and Table 2). The efficiency of the test matched with these threshold equals 68% and 90%, respectively (Table 2). Whereas the particular thresholds can serve various clinical purposes, a specific velocity value may also be selected: one associated with the highest efficiency of the test, that is, with an optimal tradeoff between sensitivity and specificity. The software we use has a program for a calculation of this type; for the moderateto-severe vasospasm, the best efficiency of 92% was identified for a peak velocity value of 182 cm/second. For the group of all vasospasms the same peak systolic velocity threshold was shown to be optimal, but the associated efficiency was reduced to 86%. The full set of ROC analysis parameters for these velocity thresholds is listed in Table 2. Reproducibility for the measurements of the mean BFV in the MCA, which was determined using the sample correlation coefficient (r) was 0.8; the coefficient of variations was 9.3%. Discussion Transcranial Doppler ultrasonography has been established as a valuable tool for the detection of increased BFVs in major cerebral arteries, and this phenomenon is known to be associated with vasospasm. 1,4,21,40,43 There are some methodological and interpretive hindrances, however, that limit the potential of this method in the diagnosis of cerebral vasospasm. The former is related to excessive error in velocity measurement. 17,32,60 This shortcoming of the blind variant of TCD ultrasonography has been overcome by introduction of the color version of this method (TCCD studies), 13,35 which has been demonstrated to allow for more consistent and more reproducible velocity measurements, as stated earlier. Better reproducibility of repeated TCD ultrasonographic measurements in comparison with conventional TCD ultrasonography was also found in this study. A major interpretive problem, however, is determination of how great an increase in BFV is diagnostic of vasospasm and how consistent this threshold is. The most convenient answer to this question can be found in the ROC methodology. The ROC Methodology A formal definition of the ROC curve states that it is a plot of sensitivity compared with 1 specificity for a family of cut points that define positive and negative values for a test. 19,24,68 The diagnostic performance of the test can be quantified for the complete range of cut points by calculating the A z. Because this parameter is universal, it allows the diagnostic performance of different tests to be easily compared. 19,68 The performance of different Doppler waveform parameters can also be compared in our model by using this methodology. 19,25,68 An ideal diagnostic performance for the test would produce an area of 1, whereas lack of ability to discriminate between two states (for example, between spasm and nonspasm) would be equal to a value of ,25,68 Estimation of the diagnostic performance of a clinical test is indispensable for strictly practical reasons. The ROC methodology offers easy calculation of the test efficiency 326 J. Neurosurg. / Volume 96 / February, 2002

5 Transcranial color Doppler ultrasonography in diagnosis of MCA spasm for any required decision threshold. 19,68 For a given clinical situation, the clinician may want maximum sensitivity or maximum specificity, neither of which usually corresponds to maximum efficiency. For example, a lower velocity threshold will perform better in a screening investigation, whereas to exclude a healthy individual from the disease group, a higher threshold would be more suitable. Generally a threshold is chosen that corresponds to the best efficiency, and this is usually a trade-off between the maximum sensitivity and specificity. Overall Performance of TCCD Ultrasonography in the Diagnosis of Cerebral Vasospasm The value of the ROC area for TCCD ultrasonography in the diagnosis of moderate-to-severe cerebral vasospasm in the MCA was found to be greater than 0.9; greater than the value for most diagnostic imaging procedures now in use. For example, the value of A z for mammography, a universally accepted screening examination, was reported to be 0.84 for a large population sample. 10 It is also comparable to the A z value for Doppler imaging in the diagnosis of extracranial carotid artery stenosis; this method is gaining widespread acceptance as a definitive examination prior to endarterectomy. 12 The value of A z for the all vasospasms group is somewhat lower (approximately 0.8) but it still matches the performance of most diagnostic tests used in clinical practice. 10,12,68 Generally, a value of A z greater than 0.9 places a test in the high accuracy range and a value of A z between 0.75 and 0.9 indicates moderate accuracy for a test. 19,68 Selection of the Best Doppler Parameter in the Diagnosis of Vasospasm A partial area index, restricted to the high sensitivity region of the ROC, is particularly useful here because detection of vasospasm demands high sensitivity to allow the patient to benefit from timely treatment. 29 Although in most reports (Table 3) various thresholds of mean velocity are proposed to differentiate levels of vasospasm, from our ROC analysis it is apparent that the peak systolic velocity performs better than the mean and end diastolic velocity in this capacity (Table 1). This finding is in agreement with reports that the end diastolic velocity and, consequently, the mean velocity are more heavily influenced by the status of the peripheral cerebral circulation than the peak systolic velocity. 34,51 The microcirculation can be affected in several ways by such factors as normal aging, 27,37 hormonal status, 34 arteriosclerosis, 6,23,49,56 increased intracranial pressure, 26,31 disturbed autoregulation, 67 or embolic occlusion. 4,5 The aforementioned conditions, which to some extent also reduce peak systolic velocity, are likely to explain the false negatives or lower than 100% sensitivity of TCCD ultrasonography in diagnosing vasospasm. On the other hand, false positives and, consequently, the lack of perfect specificity were caused by the incidence of very high flow velocity with no arterial narrowing. This can be ascribed to hyperemia/hyperperfusion, reduced hematocrit, or anemia. 14,57 Because TCCD ultrasonography allows for direct imaging of the vessel being studied, its caliber can be estimated, and in this way differentiation of hyperemia from vasospasm may be attempted. 50 Whether this will help increase the specificity of TCCD ultrasonography further J. Neurosurg. / Volume 96 / February, 2002 TABLE 2 Measures of accuracy for TCCD ultrasonography in the diagnosis of moderate-to-severe vasospasm for different thresholds of peak systolic velocity* Velocity (cm/sec) Parameter (%) efficiency sensitivity % CI specificity % CI PPV NPV * CI = confidence interval; NPV = negative predictive value; PPV = positive predictive value. The best efficiency was associated with the velocity of 182 cm/second. in diagnosing vasospasm is a question that requires separate studies. Diagnostic Performance of TCCD Compared With Conventional TCD Ultrasonography From the data published hitherto, no conclusion can be drawn regarding the diagnostic performance of TCD ultrasonography, because virtually all these authors provided only one, or more rarely, two velocity thresholds and related them to measures of efficiency, that is, positive and negative predictive values. From Table 3 one can see that both the prevalence and velocity thresholds reported vary considerably from one study to another, as do the measures of test efficiency. It should be noted that all but one of the studies listed in Table 3 were performed with the conventional, blind Doppler method. Straightforward comparison of the efficiency of both methods at particular velocity thresholds is difficult, if not impossible, for a number of reasons. First, the BFVs in the basal cerebral arteries in healthy controls are approximately 10 to 30% higher when measured using TCCD ultrasonography, than those obtained using TCD studies. 8,60 These differences have been reported even to enhance with age and in certain intracranial diseases like intracranial vessel stenosis. 32 Therefore, the values of BFV obtained with TCD ultrasonography cannot be applied to the diagnosis of vasospasm with the newer TCCD procedure. 36 In our report, the mean and peak systolic velocity thresholds associated with optimal efficiency were indeed higher than in conventional TCD studies (136 cm/second compared with 120 cm/second and 182 cm/second compared with 160 cm/second, respectively). In addition, our group of patients included those with spontaneous intracerebral hematomas and increased intracranial pressure. The BFVs in this subgroup of patients were on average 20 cm/second lower than in patients with straightforward SAH. One may therefore reasonably assume that the BFV thresholds for diagnosis of vasospasm in patients with plain SAH would be higher by up to 20 cm/ second. 26,39,48 An ROC study of an SAH-only group of patients is currently underway at our institution. A direct comparison of our results with those obtained using the conventional Doppler procedure is also difficult 327

6 Z. Mariak, et al. TABLE 3 Literature review of sensitivity and specificity of the two types of ultrasonography for detection of vasospasm in the M 1 segment of the MCA* Mean Sensi- No. of No. of Prev of Velocity tivity Specific- PPV NPV Authors & Year Arteries Patients Spasm (%) (cm/sec) (%) ity (%) (%) (%) TCD studies Aaslid, et al., Compton, et al., Grolimund, et al., 1987 NA Lindegaard, et al., Sekhar, et al., 1988 NA Sloan, et al., Lennihan, et al., Burch, et al., Proust, et al., NA NA Sloan, et al., NA Vora, et al., TCCD studies Proust, et al., present study * The specified measures of accuracy were recalculated from data reported by the authors. Abbreviations: NA = not available; prev = prevalence. because in early evaluations of the utility of TCD ultrasonography in diagnosis of vasospasm, overly positive findings were often reported. This is mainly because populations with a high incidence of the condition were studied, that is, patients between 7 and 14 days after SAH. The prevalence of vasospasm was comparatively low in our sample because we included a broad spectrum of patients, dictated by the inflow of 100 consecutive persons referred for angiographic examination. 55 Thus, the confidence intervals, particularly those for sensitivity, were rather wide, despite the fact that our group was larger than those included in most previous studies. Furthermore, we were able to use angiography as a method to diagnose vasospasm, whereas in a number of previous studies 38,52,61 clinical criteria were used, and it is known that angiographically detected vasospasm can be asymptomatic in up to 40% of patients. 28 With the aforementioned limitations in mind, a comparison of the efficiency of both methods might be attempted, assuming that a mean velocity threshold of 136 cm/second for TCD ultrasonography corresponds to 120 cm/second for the conventional method. From among the reports cited in Table 3, only in the study of Sloan, et al., 63 was the number of patients and the prevalence of vasospasm similar to ours. In their study, the reported specificity is similar to ours, but the sensitivity was shown to be much lower for the color method. This indirectly implies that the efficiency of TCCD ultrasonography, at least for this range of velocity thresholds, is better than that of conventional TCD studies. Full evidence would only be provided, however, by a comparative study of the same population with both methods by using ROC analysis. Limitations of TCD Ultrasonography in Diagnosis of Cerebral Vasospasm A frequently overlooked fact is that the efficiency of a test changes with two variables: the prevalence of disease in the population studied and the discriminator position (in our case the velocity threshold used to diagnose vasospasm). If the sensitivity related to a selected discriminator position is higher than the specificity, then the efficiency will increase with increasing prevalence of disease. In the alternative situation (when specificity is higher than sensitivity) the efficiency will decrease with increasing prevalence of disease. 55 Consequently, a velocity threshold claimed as having the best efficiency in a population with high prevalence of disease could be associated with suboptimal efficiency when used in a population with a lower prevalence of the disease studied. 19 We performed our study on a random population of 100 consecutive patients referred for angiography and thus the prevalence of vasospasm was low in comparison with most studies shown in Table 3. Our resulting index of diagnostic performance (A z = 0.93) might be expected to be higher, not lower if a group of patients with higher prevalence of vasospasm were included in the study, for example, patients who had suffered SAH. A quite separate problem is whether the increased BFV, as measured with TCCD ultrasonography, is consistent specifically with vasospasm. It must be stressed here that all estimations of TCCD performance in diagnosing cerebral vasospasm in this study were calculated in relation to angiographically confirmed diagnoses of this condition. Nevertheless, the angiographically determined diagnosis of cerebral vasospasm also has limitations. The percentage of lumen narrowing, used in this study according to the methods of other investigators, 3,59,65 has inherent inaccuracies. These occur because vasospasm is sometimes not focal, and comparison to the adjacent segment will not necessarily provide a comparison with the unaffected artery. Repeated angiographic examination could differentiate real vasospasm from other types of arterial narrowing. 3,56 Nevertheless, this is rarely performed. 328 J. Neurosurg. / Volume 96 / February, 2002

7 Transcranial color Doppler ultrasonography in diagnosis of MCA spasm It is notable that with TCCD ultrasonography we were able to detect all five MCA occlusions and four large AVMs. These patients were excluded from the ROC analysis. Because TCCD ultrasonography enables a straightforward diagnosis of these conditions, 8,33 the incidence of MCA occlusion and AVMs does not seem to affect its performance in the detection of vasospasm. Because conventional TCD ultrasonography is much less precise in depicting these lesions, its performance in diagnosing vasospasm might be lower in comparison with TCCD studies. 8 An important limitation of TCCD ultrasonography in the diagnosis of vasospasm is related to the fact that flow velocity could not be examined in 11 of 100 patients because of inadequacy of the temporal window. The number of patients suitable for TCCD investigation can be increased, however, by using echo contrast agents, which are becoming widely available. 20 Conclusions Three points should be stressed to summarize our findings. 1) The performance of TCCD ultrasonography in the diagnosis of advanced MCA narrowing is very high, as estimated from the area under the ROC curve. For the diagnosis of mild MCA narrowing the performance is acceptable. 2) The best-performing TCCD parameter for the detection of MCA narrowing is peak systolic velocity. 3) With ROC methodology one can establish a velocity threshold associated with optimal efficiency in diagnosing vasospasm. In our investigation an optimal tradeoff between sensitivity and specificity was associated with a peak systolic velocity of 182 cm/second. References 1. Aaslid R, Huber P, Nornes H: Evaluation of cerebrovascular spasm with transcranial Doppler ultrasound. J Neurosurg 60: 37 41, Aaslid R, Markwalder TM, Nornes H: Noninvasive transcranial Doppler ultrasound recordings of flow velocity in basal cerebral arteries. J Neurosurg 57: , Adams CBT, Fearnside MR, O Laoire SA: An investigation with serial angiography into the evolution of cerebral artery spasm following aneurysm surgery. J Neurosurg 49: , Babikian VL, Feldmann E, Wechsler LR, et al: Transcranial Doppler ultrasonography: year 2000 update. J Neuroimaging 10: , Babikian VL, Wijman CAC, Hyde C, et al: Cerebral microembolism and early recurrent cerebral or retinal ischemic events. Stroke 28: , Baker AB, Iannone A: Cerebrovascular disease. III. The intracerebral arterioles. Neurology 9: , Baumgartner RW, Mathis J, Sturzenegger M, et al: A validation study on the intraobserver reproducibility of transcranial colorcoded duplex sonography velocity measurements. Ultrasound Med Biol 20: , Baumgartner RW, Mattle HP, Aaslid R: Transcranial colorcoded duplex sonography, magnetic resonance angiography, and computed tomography angiography: methods, applications, advantages, and limitations. J Clin Ultrasound 23:89 111, Baumgartner RW, Mattle HP, Schroth G: Assessment of /= 50% and 50% intracranial stenoses by transcranial color-coded duplex sonography. Stroke 30:87 92, 1999 J. Neurosurg. / Volume 96 / February, Beam CA, Layde PM, Sullivan DC: Variability in the interpretation of screening mammograms by US radiologists. Findings from a national sample. Arch Intern Med 156: , Bederson JB, Levy AL, Ding WH, et al: Acute vasoconstriction after subarachnoid hemorrhage. Neurosurgery 42: , Bluth EI, Sunshine JH, Lyons JB, et al: Power Doppler imaging: initial evaluation as a screening examination for carotid artery stenosis. Radiology 215: , Bogdahn U, Becker G, Winkler J, et al: Transcranial color-coded real-time sonography in adults. Stroke 21: , Brown MM, Marshall J: Regulation of cerebral blood flow in response to changes in blood viscosity. Lancet 1: , Burch CM, Wozniak MA, Sloan MA, et al: Detection of intracranial internal carotid artery and middle cerebral artery vasospasm following subarachnoid hemorrhage. J Neuroimag 6: 8 15, Compton JS, Redmond S, Symon L: Cerebral blood velocity in subarachnoid hemorrage: a transcranial Doppler study. J Neurol Neurosurg Psychiatry 50: , Eicke BM, Tegeler CH, Dalley G, et al: Angle correction in transcranial Doppler sonography. J Neuroimaging 4:29 33, Ekelund A, Saveland H, Romner B, et al: Transcranial Doppler ultrasound in hypertensive versus normotensive patients after aneurysmal subarachnoid hemorrhage. Stroke 26: , Gerhardt W, Keller H: Evaluation of test data from clinical studies. I. Terminology, graphic interpretation, diagnostic strategies, and selection of sample groups. II. Critical review of the concepts of efficiency, receiver operated characteristics (ROC), and the likelihood ratios. Scand J Clin Lab Invest Suppl 181:1 74, Gerriets T, Seidel G, Fiss I, et al: Contrast-enhanced transcranial color coded duplex sonography: efficiency and validity. Neurology 52: , Grolimund P, Seiler RW, Aaslid R, et al: Evaluation of cerebrovascular disease by combined extracranial and transcranial Doppler sonography. Experience in 1,039 patients. Stroke 18: , Grosset GD, Straiton J, McDonald I, et al: Use of transcranial Doppler sonography to predict development of a delayed ischemic deficit after subarachnoid hemorrhage. J Neurosurg 78: , Hacker H: Abnormal middle cerebral artery, in Newton TH, Potts DG (eds): Radiology of the Skull and Brain. Great Neck, NY: Mosby, 1979, Vol 2, pp Hanley JA, McNeil BJ: The meaning and use of the area under a receiver operating characteristic (ROC) curve. Radiology 143: 29 36, Hanley JA, McNeil BJ: A method of comparing the areas under receiver operating characteristic curves derived from the same cases. Radiology 148: , Hassler W, Steinmetz H, Gawlowski J: Transcranial Doppler ultrasonography in raised intracranial pressure and in intracranial circulatory arrest. J Neurosurg 68: , Hayashi K, Handa H, Nagasawa S, et al: Stiffness and elastic behavior of human intracranial and extracranial arteries. J Biomech 13: , Heros RC, Zervas NT, Varsos V: Cerebral vasospasm after subarachnoid hemorrhage: an update. Ann Neurol 14: , Jiang Y, Metz CE, Nishikawa RM: A receiver operating characteristic partial area index for highly sensitive diagnostic tests. Radiology 201: , Kairisto V: Optimal bin widths for frequency histograms and ROC curves. Clin Chem 41: , 1995 (Letter) 31. Klingelhofer J, Dander D, Holzgraefe M, et al: Cerebral vasospasm evaluated by transcranial Doppler ultrasonography at different intracranial pressures. J Neurosurg 75: ,

8 Z. Mariak, et al. 32. Krejza J, Mariak Z, Babikian VL: Importance of angle-correction in measurement of blood flow velocity with transcranial Doppler sonography. AJNR 22: , Krejza J, Mariak Z, Bert RJ: Transcranial colour Doppler sonography in emergency management of intracerebral haemorrhage caused by arteriovenous malformation: case report. Neuroradiology 42: , Krejza J, Mariak Z, Huba M, et al: Effect of endogenous estrogen on blood flow through carotid arteries. Stroke 32:30 36, Krejza J, Mariak Z, Melhem ER, et al: A guide to the identification of major cerebral arteries with transcranial color Doppler sonography. AJR 174: , Krejza J, Mariak Z, Walecki J: Usefulness of transcranial color-coded sonography in the diagnosis of cerebral vasospasm. Stroke 30: , 1999 (Letter) 37. Krejza J, Mariak Z, Walecki J, et al: Transcranial color doppler sonography of basal cerebral arteries in 182 healthy subjects: age and sex variability and normal reference values for blood flow parameters. AJR 172: , Laumer R, Steinmeier R, Gonner F, et al: Cerebral hemodynamics in subarachnoid hemorrhage evaluated by transcranial Doppler sonography. Part 1. Reliability of flow velocities in clinical management. Neurosurgery 33:1 9, Lee EJ, Chio CC, Lin HJ, et al: Application of transcranial Doppler sonography in surgical aspects of hypertensive putaminal haemorrhage. Acta Neurochir 138:60 67, Lennihan L, Petty GW, Fink ME, et al: Transcranial Doppler detection of anterior cerebral artery vasospasm. J Neurol Neurosurg Psychiatry 56: , Lindegaard KF, Nornes H, Bakke SJ, et al: Cerebral vasospasm after subarachnoid haemorrage investigated by means of transcranial Doppler ultrasound. Acta Neurochir Suppl 42:81 84, Lindegaard KF, Nornes H, Bakke SJ, et al: Cerebral vasospasm diagnosis by means of angiography and blood velocity measurements. Acta Neurochir 100:12 24, Lupetin AR, Davis DA, Beckman I, et al: Transcranial Doppler sonography. Part 2. Evaluation of intracranial and extracranial abnormalities and procedural monitoring. Radiographics 15: , Maeda H, Etani H, Handa N, et al: A validation study on the reproducibility of transcranial Doppler velocimetry. Ultrasound Med Biol 16:9 14, Manno EM, Gress DR, Schwamm LH, et al: Effects of induced hypertension on transcranial Doppler ultrasound velocities in patients after subarachnoid hemorrhage. Stroke 29: , Martin NA, Doberstein C, Zane C, et al: Posttraumatic cerebral arterial spasm: transcranial Doppler ultrasound, cerebral blood flow, and angiographic findings. J Neurosurg 77: , Martin PJ, Evans DH, Naylor AR: Measurement of blood flow velocity in the basal cerebral circulation: advantages of transcranial color-coded sonography over conventional transcranial Doppler. J Clin Ultrasound 23:21 26, Mayer AS, Thomas EC, Diamond BE: Asymmetry of intracranial hemodynamics as an indicator of mass effect in acute intracerebral hemorrage. A transcranial Doppler study. Stroke 27: , Michotey P, Moscow NP, Salamon G: Anatomy of the cortical branches of the middle cerebral artery, in Newton TH, Potts DG (eds): Radiology of the Skull and Brain. Great Neck, NY: Mosby, 1979, Vol 2, pp Muller M, Schwerdtfeger K, Zieroth S: Assessment of middle cerebral artery diameter after aneurysmal subarachnoid hemorrhage by transcranial color-coded duplex sonography. Eur J Ultrasound 11:15 19, Ohkuma H, Manabe H, Tanaka M, et al: Impact of cerebral microcirculatory changes on cerebral blood flow during cerebral vasospasm after aneurysmal subarachnoid hemorrhage. Stroke 31: , Okada Y, Shima T, Nishida M, et al: Comparison of transcranial Doppler investigation of aneurysmal vasospasm with digital subtraction angiographic and clinical findings. Neurosurgery 45: , Poiseuille JLM: Recerches expérimentales sur le mouvement des liquides dans les tubes de très petits diamètres. Acad Sci Comp Rend 11: , Proust F, Callonec F, Clavier E, et al: Usefulness of transcranial color-coded sonography in the diagnosis of cerebral vasospasm. Stroke 30: , Ransohoff DF, Feinstein AR: Problems of spectrum and bias in evaluating the efficacy of diagnostic tests. N Engl J Med 299: , Ring BA: Normal middle cerebral artery, in Newton TH, Potts DG (eds): Radiology of the Skull and Brain. Great Neck, NY: Mosby, 1979, Vol 2, pp Romner B, Bellner J, Kongstad P, et al: Elevated transcranial Doppler flow velocities after severe head injury: cerebral vasospasm or hyperemia? J Neurosurg 85:90 97, Romner B, Ljunggren B, Brandt L, et al: Transcranial Doppler sonography within 12 hours after subarachnoid hemorrhage. J Neurosurg 70: , Saito I, Shigeno T, Aritake K, et al: Vasospasm assessed by angiography and computerized tomography. J Neurosurg 51: , Schöning M, Buchholz R, Walter J: Comparative study of transcranial color duplex sonography and transcranial Doppler sonography in adults. J Neurosurg 78: , Sekhar LN, Wechsler LR, Yonas H, et al: Value of transcranial Doppler examination in the diagnosis of cerebral vasospasm after subarachnoid hemorrage. Neurosurgery 22: , Sloan MA, Haley EC Jr, Kassell NF, et al: Sensitivity and specificity of transcranial Doppler ultrasonography in the diagnosis of vasospasm following subarachnoid hemorrage. Neurology 39: , Sloan MA, Wozniak MA, Macko RF: Monitoring of vasospasm after subarachnoid hemorrhage, in Babikian VL, Wechsler LR (eds): Transcranial Doppler Ultrasonography, ed 2. Boston: Butterworth Heinemann, 1999, pp Teasdale G, Jennett B: Assessment of coma and impaired consciousness. A practical scale. Lancet 2:81 84, Vora YY, Suarez-Almazor M, Steinke DE, et al: Role of transcranial Doppler monitoring in the diagnosis of cerebral vasospasm after subarachnoid hemorrhage. Neurosurgery 44: , Wardlaw JM, Offin R, Teasdale GM, et al: Is routine transcranial Doppler ultrasound monitoring useful in the management of subarachnoid hemorrhage? J Neurosurg 88: , Yundt KD, Grubb RL Jr, Diringer MN, et al: Autoregulatory vasodilation of parenchymal vessels is impaired during cerebral vasospasm. J Cereb Blood Flow Metab 18: , Zweig MH, Campbell G: Receiver-operating characteristic (ROC) plots: a fundamental evaluation tool in clinical medicine. Clin Chem 39: , 1993 Manuscript received April 4, Accepted in final form October 16, This study was supported by the Polish Committee on Scientific Investigations Grant No. 6 P05C05020 to Dr. Krejza. Address reprint requests to: Jaroslaw Krejza, M.D., Ph.D., Department of Radiology, Bialystok Medical Academy, Sklodowskiej-Curie 24a, Bialystok, Poland. jkrejza@cksr.ac. bialystok.pl. 330 J. Neurosurg. / Volume 96 / February, 2002