- Email: [email protected]
Formal analysis of the optimal duration of tilt testing for the diagnosis of neurally mediated syncope
Electrophysiology
Formal analysis of the optimal duration of tilt testing for the diagnosis of neurally mediated syncope Kenneth M. Stein, MD, David J. Slotwiner, MD, Suneet Mittal, MD, Marc Scheiner, MD, Steven M. Markowitz, MD, and Bruce B. Lerman, MD New York, NY
Background Although tilt testing has emerged as the test of choice for assessing patients with suspected neurally mediated syncope, the optimum duration of tilt testing is poorly defined. This in part relates to the absence of a gold standard to assess test performance.
Objective Our purpose was to formally estimate the effects of varying duration of drug-free tilt testing on test performance in diagnosing neurally mediated syncope.
Design If a test’s specificity is known, then in the absence of a gold standard an imputed (estimated) sensitivity may be calculated on the basis of the observed diagnostic yield in a given population as a function of assumed population prevalence. We determined the relationship of specificity to drug-free tilt test duration by use of data from 11 previous studies reporting the results of drug-free tilt testing in a total of 435 control subjects (60 to 80 degrees of tilt, footboard support, 15to 60-minute duration). Data (weighted for study size) were fit to an exponential function relating specificity to tilt duration. Test yield was evaluated as a function of tilt duration in 213 consecutive patients referred to our laboratory for the evaluation of suspected neurally mediated syncope who underwent passive tilt testing for up to 30 to 60 minutes.
Results The estimated specificity of tilt testing was 94% at 30 minutes, 92% at 40 minutes, and 88% after 60 minutes of passive tilt. The cumulative yield of tilt testing was only 17% at 30 minutes, 22% at 40 minutes, and 28% after 60 minutes. On the basis of an estimated population prevalence of 25% to 50% in this referral population, imputed sensitivity is 27% to 48% at 30 minutes, 36% to 64% at 40 minutes, and 43% to 74% after 60 minutes of passive tilt. The overall diagnostic accuracy was not strongly influenced by tilt duration beyond 30 minutes and ranged from 60% to 84%.
Conclusions Passive tilt testing (ie, tilt testing without pharmacologic provocation) for durations of up to 60 minutes has limited sensitivity for diagnosing neurally mediated syncope. For populations with a pretest likelihood of 25% to 50%, test results are inaccurate in one to two fifths of patients. (Am Heart J 2001;141:282-8.)
Tilt table testing has emerged as the test of choice for assessing patients with suspected neurally mediated syncope. However, many different protocols are used and the optimum duration of tilt testing remains controversial, as does the use of pharmacologic maneuvers to enhance sensitivity.1 This state of affairs relates in large part to the absence of a “gold standard,” a way of perfectly identifying true positives and false negatives, for From the Division of Cardiology, Department of Medicine, New York Hospital–Cornell Medical Center, New York, NY. Supported in part by National Institutes of Health grant No. RO1 HL5139. Submitted January 31, 2000; accepted October 13, 2000. Reprint requests: Kenneth M. Stein, MD, Division of Cardiology, Starr-4, New York Hospital, 525 E 68th St, New York, NY 10021. E-mail: [email protected] Copyright © 2001 by Mosby, Inc. 0002-8703/2001/$35.00 + 0 4/1/112236 doi:10.1067/mhj.2001.112236
the diagnosis. Such a gold standard is required to accurately evaluate test sensitivity and diagnostic performance. Thus the American College of Cardiology expert consensus on tilt testing concluded that “On the basis of studies in control subjects discussed earlier, it is clear that tilt table testing exhibits a high level of diagnostic specificity. Sensitivity, in contrast, is a more difficult issue. Apart from tilt table testing itself, there is no clearcut, accepted ‘gold standard’ for establishing a diagnosis of neurally mediated (and especially vasovagal) syncope against which diagnostic procedures can be measured.”2 Owing to the absence of a gold standard, clinicians have been forced to use the yield of the test in patients believed to have a high pretest likelihood of neurally mediated syncope as an (admittedly imperfect) estimate of sensitivity.3 However, it can be shown mathematically that, even
American Heart Journal Volume 141, Number 2
Stein et al 283
in the absence of a gold standard, if specificity is known then an “imputed” (estimated) sensitivity can be calculated as a function of assumed prevalence by use of the observed diagnostic yield. The purpose of the current study was to develop a technique for computing imputed sensitivity that would potentially be applicable to all situations in which a diagnostic test of known specificity is applied in the absence of a gold standard. We then sought to use this technique to evaluate the effect of the duration of passive tilt testing on the utility of tilt testing for diagnosing neurally mediated syncope.
Methods Calculation of imputed sensitivity If n total patients undergo a diagnostic test and we represent the true positives by a, the true negatives by b, the false positives by c, and the false negatives by d (n = a + b + c + d), then we may define the true-positive rate as a/n, the falsepositive rate as c/n, the true-negative rate as b/n, and the false-negative rate as d/n. The yield (Y) (or positivity rate) of a diagnostic test as a function of duration (time, t) may then be defined as the sum of the true-positive (TP) and the falsepositive (FP) rates of the test as functions of duration: Y(t) = TP(t) + FP(t) For a given population the true-positive rate depends on the sensitivity (Se) and disease prevalence (Pr) in that population, whereas the false-positive rate depends on specificity (Sp) and prevalence. Thus yield can be expressed by reference to sensitivity, specificity, and prevalence: Y(t) = Se(t)Pr + (1 – Sp[t])(1 – Pr) Finally, if the test specificity is known and if test yield can be determined for a given population, then an imputed sensitivity (Sei) may be computed as a function of assumed disease prevalence in that population: Sei(t,Pr) =
July 1998. All patients underwent tilt-table testing after an overnight fast. After written informed consent was obtained, each patient was placed in the supine position on a motorized tilt table with footboard support and an intravenous catheter was placed in an arm vein. Continuous electrocardiographic monitoring was implemented. Noninvasive automated blood pressure monitoring was performed (Dinamap, Critikon, Tampa, Fla) and the blood pressure and heart rate were recorded at 1-minute intervals. A manual blood pressure cuff was used to verify any decrease in blood pressure recorded by the automatic cuff or when warranted by clinical symptoms. After a 15-minute supine control phase, patients were tilted upright to an angle of 60 degrees for 30 or 60 minutes or until a positive response was achieved. A positive (abnormal) response was defined as (1) sudden loss of consciousness or the development of presyncope in association with an abrupt decrease in systolic blood pressure to 80 mm Hg and (2) the reproduction of symptoms that had been associated with the patient’s clinical event(s). A positive response had to meet both criteria. With use of these data, the relationship between tilt duration and positivity rate (and standard error of the relationship) was then computed by the life-table (product limit) method.
Results Tilt specificity In the 11 published studies reporting the results of passive tilt testing in a total of 435 normal subjects, specificity varied from 83% to 100% and declined modestly with increasing duration of the test (Figure 1). The data were reasonably approximated by an exponential distribution according to the equation Sp = e–0.0022t (P < .001, R2 = 59%). The estimated specificity of tilt testing was 94% at 30 minutes, 92% at 40 minutes, and 88% after 60 minutes of passive tilt.
Y(t) – (1 – Sp[t])(1 – Pr) Pr
Tilt specificity as a function of tilt duration By computerized searching, as well as by review of the relevant literature, we identified 11 previous studies reporting the results of drug-free tilt testing in healthy control subjects subject to the following restrictions: (1) 60 to 80 degrees upright tilt, (2) footboard support, and (3) 15- to 60-minute duration.414 The data were then weighted for study size and were fit to an exponential function relating specificity to tilt duration, assuming 100% specificity at 0 minutes (Curve Estimation procedure, SPSS for Windows 8.0.1, SPSS, Chicago, Ill).
Tilt yield as a function of tilt duration Tilt yield (positivity rate) was evaluated as a function of passive tilt duration in 213 consecutive patients referred to our laboratory for the evaluation of suspected neurally mediated syncope. The first 193 consecutive patients underwent passive tilt testing for 30 minutes between May 1995 and May 1997. The remaining 20 consecutive patients underwent passive tilt testing for 60 minutes between June and
Tilt yield Demographic characteristics of the patients referred for tilt testing are summarized in Table I. There were no differences between patients with positive and negative tilt tests with respect to age, sex, or the presence of underlying heart disease. Consistent with previous studies,15 there was a relationship between the historic frequency of syncope and the likelihood of a positive tilt test. Although many patients had recurrent syncope, nearly half (99/213, 46%) were referred for the evaluation of recurrent presyncope or after a single syncopal episode. Of particular note is the extremely low yield observed in patients with a history of recurrent presyncope but without true syncope (1/35, 3% positive rate). The overall yield of passive tilt testing was low in this referral population. The cumulative yield of tilt testing was 3% ± 1% (estimate ± SE) at 10 minutes, 8% ± 2% at 20 minutes, 17% ± 3% at 30 minutes, 22% ± 6% at 40 minutes, and 28% ± 8% after 60 minutes of passive tilt.
American Heart Journal February 2001
284 Stein et al
Figure 1
Specificity as a function of tilt duration in 11 previously published studies reporting the results of drug-free tilt testing (60 to 80 degrees upright tilt with footboard support) in healthy control subjects. The size of each circle reflects the size of the control population in the study.
Table I. Patient characteristics
Age (y) Sex (male/female) Heart disease(%) Episodes of syncope Indeterminate Presyncope only 1 2-3 4+
Positive tilt test (n = 36)
Negative tilt test (n = 177)
54 ± 22 14 (39%) 22 (61%) 9 (25%)
57 ± 20 73 (41%) 104 (59%) 43 (24%)
1 (3%) 1 (3%) 12 (33%) 9 (25%) 13 (36%)
6 (3%) 34 (19%) 52 (29%) 50 (28%) 35 (20%)
Statistical significance NS NS NS P = .07
Results are expressed as mean ± SD where appropriate. Heart disease is defined as a known prior history of coronary artery disease, significant valvular heart disease, or cardiomyopathy. NS, Not significant.
Tilt performance On the basis of an assumed disease prevalence of 25% to 50% in our referral population, imputed sensitivity ranged from 27% to 48% after 30 minutes, from 36% to 64% after 40 minutes, and from 43% to 74% after 60 minutes of passive tilt (Figure 2). The overall diagnostic accuracy (true-positive + true-negative rate) was greater for 30-minute tilts than for shorter durations but only
improved modestly thereafter (Figure 3). Thus, on the basis of an assumed prevalence of 25% to 50%, the estimated diagnostic accuracy was 60% to 82% after 30 minutes, 64% to 85% after 40 minutes, and 65% to 84% after 60 minutes of passive tilt. It is possible to formally assess the information derived from a diagnostic test by computing the difference between pretest and posttest likelihood of disease on the basis of positive versus negative tests (“diagnostic information,” see Figure 4). The posttest likelihood of disease for a positive test is its positive predictive value. The posttest likelihood of disease for a negative test is equal to 1 minus the negative predictive value. Curves demonstrating the posttest likelihood of disease after negative versus positive tilt tests are shown in Figure 4, based on “best-case” estimated sensitivities of 48%, 64%, and 74%, respectively, after 30, 40, and 60 minutes and estimated specificities of 94%, 92%, and 88% after 30, 40, and 60 minutes of tilt. As with all diagnostic tests, the information derived from testing is greatest for patients with an intermediate pretest likelihood of disease. Within this range, there was a small increase in information for the 40minute test compared with the 30-minute test but no clinically important further increase for the 60-minute
American Heart Journal Volume 141, Number 2
Stein et al 285
Figure 2
Imputed sensitivity of tilt testing for the diagnosis of neurally mediated syncope as a function of tilt duration and assumed true disease prevalence in our study population.
Figure 3
Estimated diagnostic accuracy (true-positive + true-negative rate) of tilt testing for the diagnosis of neurally mediated syncope as a function of tilt duration and assumed true disease prevalence in our study population.
test. For patients with a pretest likelihood below 33%, there was no important difference among 30-, 40-, and 60-minute tests. With use of these “best case” assumptions, the positive predictive value of a 40-minute pas-
sive tilt test in a patient with a 50% pretest likelihood of disease is 88%. However, the posttest probability despite a negative test (1 minus the negative predictive value) is still 28%.
American Heart Journal February 2001
286 Stein et al
Figure 4
Posttest likelihood of disease in patients with positive versus negative tilt tests and diagnostic information (the difference between the posttest likelihoods of disease after positive vs negative tests) as a function of pretest likelihood of disease and passive tilt duration. The calculations are based on “best-case” estimated sensitivities of 48%, 64%, and 74%, respectively, after 30, 40, and 60 minutes and estimated specificities of 94%, 92%, and 88% after 30, 40, and 60 minutes of tilt. Note that the positive predictive value (PPV) curves at 30 and 40 minutes are virtually superimposed. NPV, Negative predictive value.
Discussion The current study represents the development of a formal technique for estimating the sensitivity of a diagnostic test for which no “gold standard” exists. Application of the imputed sensitivity analysis to tilt testing for the diagnosis of neurally mediated syncope leads to several conclusions: (1) the specificity of passive (drug-free) tilt testing is high for tilt durations of up to 60 minutes, (2) the yield of passive tilt testing is
relatively low in a current referral population, and (3) as a result, the imputed sensitivity of passive tilt testing is low for tilt durations of up to 60 minutes. The diagnostic information derived by tilt testing is greatest for patients with an intermediate pretest likelihood of disease. For these patients, a 40-minute tilt protocol is more informative than a 30-minute protocol, but little additional information is gained by prolonging the test to 60 minutes. Although the positive
American Heart Journal Volume 141, Number 2
predictive value of a 40-minute passive tilt test is high, there is a clinically significant false-negative rate for patients with an intermediate pretest likelihood of disease.
Specificity Although many in the cardiology community continue to express concern regarding false-positive results of tilt tests, a review of 11 studies in which passive tilt tests were performed in 453 healthy control subjects demonstrates that the test has high specificity.4-14 There was a tendency for specificity to diminish with prolongation of upright tilt. Even so, specificity is estimated to be 88% at the end of a 60-minute drug-free tilt test by use of current methods.
Yield The yield of drug-free tilt testing observed in the current study was disappointingly low (17% after 30 minutes and 28% after 60 minutes). This is in contrast with some (mostly early) studies reporting yields of 50% to 74% for passive tilt tests of 45 to 60 minutes in patients with syncope of unknown origin.5,16-18 However, the data are consistent with the substantially lower yields reported in other studies.4,11,13-15,19-22 Some of these differences in results may be attributable to subtle differences in method (eg, tilt angle, invasive vs noninvasive hemodynamic monitoring). It is also highly likely that the population referred for tilt testing is different from the population evaluated in the initial reports of the method, in that it is now more heterogeneous, and the true prevalence of neurally mediated syncope may be lower in patients referred for the test today than in the past. This reflects the natural evolution of a diagnostic modality away from being an experimental procedure applied first to a homogeneous high-risk population and toward being a test used to gain information about patients with an intermediate probability of disease.
Sensitivity and optimum duration of testing As a result of the low yield, the imputed sensitivity of the passive tilt test is low (although still within a range considered acceptable for other diagnostic modalities). Although sensitivity increases with duration of tilt testing, specificity declines. As a result, overall test accuracy (which was estimated to range from 60%-85%) is not strongly influenced by tilt durations beyond 30 minutes. The diagnostic information derived from testing was greatest at intermediate probabilities of disease and was greater for a 40-minute test than for shorter-duration tilts. Little additional information was gained by extending the tilt to 60-minute duration. The positive predictive value of passive tilt testing in the 40- to 60-minute range was estimated to be high. However, the passive tilt test probably has a clinically important false-negative rate for patients with an inter-
Stein et al 287
mediate pretest likelihood of disease. These results emphasize the potential importance of tilt protocols with pharmacologic provocation to reduce the falsenegative rate of tilt testing.14,23,24 It should be strongly emphasized that the technique for calculating imputed sensitivity yields estimates and not true values for sensitivity. These estimates are critically dependent on the validity of the assumptions regarding disease prevalence in the population studied. Although the assumption that the prevalence was in the range of 25% to 50% in the current population seems reasonable, it is possible that the actual prevalence may be higher or lower. It should be noted that if the actual prevalence were any higher than 50% then the imputed sensitivity would be correspondingly lower (eg, if the actual prevalence were 75% then the imputed sensitivity for 30 minutes of passive tilt would have only been 19%). On the other hand, on the basis of the estimated specificity of 88% at 60 minutes and the observed yield of 28% at 60 minutes, it would not have been mathematically possible for the population prevalence to have been substantially less than 20%.
Limitations It should be noted that a positive tilt test in the asymptomatic control populations consisted of hemodynamic collapse, whereas a positive tilt test in the patients with syncope consisted of hemodynamic collapse with reproduction of the clinical symptoms. Clearly, if this stricter definition were applied to the control subjects, there could have been no “positive” tests and the specificity would be 100%. It is theoretically possible that in practice some of the “false-positive” tests predicted on the basis of the specificity estimates were correctly classified as “negative” as a result of failure to reproduce clinical symptoms. Although this problem is far from trivial, in this particular population there were no tests that were reclassified from positive to negative on the basis of failure to duplicate clinical symptoms. It should also be noted that the choice of an exponential form to fit the specificity data was based on the theoretic need that the specificity could never be less than 0, should reach 0 by an infinite duration of tilt, and should be 100% at time 0. It should be observed that the actual computed curve (Sp = e–0022t) is almost precisely linear over the times that we analyzed (0 to 60 minutes) and that substitution of a linear curve fit does not affect the results. It should also be noted that relatively few patients underwent the full 60minute tilt protocol (although the SEs for the estimates of tilt yield at 40-60 minutes are low). Finally, it should be noted that the estimates of specificity were derived from pooled analysis of multiple, potentially heterogeneous studies and are subject to the limitations inherent in this approach. It should also be pointed out that, although a similar
288 Stein et al
analysis applied to isoproterenol tilt testing would be of interest, there is not sufficient standardization of isoproterenol tilt protocols in the literature, in terms of dose, duration, and the duration of any preceding passive tilt phase, to permit a comparable analysis.
Conclusions If the specificity of a diagnostic test is known, then even in the absence of a gold standard, test sensitivity can be estimated on the basis of the assumed disease prevalence in a given referral population. The results of the analysis suggest that a 40-minute tilt protocol may provide more diagnostic information than a 30-minute period but that little additional information is gained by extending the protocol to 60 minutes. Although the predictive value of a positive passive tilt test is high, there is a significant false-negative rate, emphasizing the potential importance of tilt testing with pharmacologic provocation for the diagnosis of neurally mediated syncope.
References 1. Benditt DG, Remole S, Bailin S, et al. Tilt table testing for evaluation of neurally-mediated (cardioneurogenic) syncope: rationale and proposed protocols. PACE 1991;14:1528-37. 2. Benditt DG, Ferguson DW, Grubb BP, et al. ACC expert consensus document: tilt table testing for assessing syncope. J Am Coll Cardiol 1996;28:263-75. 3. Benditt DG. Neurally mediated syncopal syndromes: pathophysiological concepts and clinical evaluation. PACE 1997;20:572-84. 4. Abi-Sainra F, Maloney JD, Fouad-Tarazi FM, et al. The usefulness of head-up tilt testing and hemodynamic investigations in the workup of syncope of unknown origin. PACE 1988;11:1202-14. 5. Fitzpatrick A, Sutton R. Tilting towards a diagnosis in recurrent unexplained syncope. Lancet 1989;1:658-60. 6. Grubb BP, Temesy-Armos P, Hahn H, et al. Utility of upright tilt-table testing in the evaluation and management of syncope of unknown origin. Am J Med 1991;90:6-10. 7. Fitzpatrick AP, Theodorakis G, Vardas P, et al. Methodology of head-up tilt testing in patients with unexplained syncope. J Am Coll Cardiol 1991;17:125-30. 8. Fitzpatrick AP, Theodorakis G, Vardas P, et al. The incidence of malignant vasovagal syndrome in patients with recurrent syncope. Eur Heart J 1991;12:389-94.
American Heart Journal February 2001
9. Grubb BP, Wolfe D, Samoil D, et al. Recurrent unexplained syncope in the elderly: the use of head-upright tilt table testing in evaluation and management. J Am Geriatr Soc 1992;40:1123-8. 10. Kapoor WN, Brant N. Evaluation of syncope by upright tilt testing with isoproterenol: a nonspecific test. Ann Intern Med 1992;116:358-63. 11. Lurie KG, Dutton J, Mangat R, et al. Evaluation of edrophonium as a provocative agent for vasovagal syncope during head up tilttable testing. Am J Cardiol 1993;72:1286-90. 12. Natale A, Akhtar M, Jazayeri M, et al. Provocation of hypotension during head-up tilt testing in subjects with no history of syncope or presyncope. Circulation 1995;92:54-8. 13. Raviele A, Menozzi C, Brignole M, et al. Value of head-up tilt testing potentiated with sublingual nitroglycerin to assess the origin of unexplained syncope. Am J Cardiol 1995;76:267-72. 14. Aerts A, Dendale P, Strobel G, et al. Sublingual nitrates during head-up tilt testing for the diagnosis of vasovagal syncope. Am Heart J 1997;133:504-7. 15. Fitzpatrick AP, Lee RJ, Epstein LM, et al. Effect of patient characteristics on the yield of prolonged baseline head-up tilt testing and the additional yield of drag provocation. Heart 1996;76:406-11. 16. Kenny RA, Ingram A, Bayliss J, et al. Head-up tilt: a useful test for investigating unexplained syncope. Lancet 1986;1:1352-5. 17. Raviele A, Gasparini G, DiPede F, et al. Usefulness of head-up tilt test in evaluating patients with syncope of unknown origin and negative electrophysiologic study. Am J Cardiol 1990;65:1322-7. 18. Ruiz GA, Peralta A, Gonzalez-Zuelgaray J, et al. Evolution of patients with clinical neurally-mediated (vasovagal) syncope not subjected to specific treatment. Am Heart J 1995;130:345-50. 19. Morillo CA, Klein GJ, Zandri S, et al. Diagnostic accuracy of a lowdose isoproterenol head-up tilt protocol. Am Heart J 1995;129: 901-6. 20. Strasberg B, Rechavia E, Sagie A, et al. The head-up tilt table test in patients with syncope of unknown origin. Am Heart J 1989;118: 923-7. 21. Lippman N, Stein KM, Lerman BB. Differential therapeutic response of patients with isoproterenol-dependent and isoproterenol-independent vasodepressor syncope. Am Heart J 1994;128:1110-6. 22. Pavri BP, Ruskin JN, Brooks R. The yield of head-up tilt testing is not significantly increased by repeating the baseline test. Clin Cardiol 1996;19:494-6. 23. Almquist A, Goldenberg IF, Milstein S, et al. Provocation of bradycardia and hypotension by isoproterenol and upright posture in patients with unexplained syncope. N Engl J Med 1989;320: 346-51. 24. Mittal S, Stein KM, Markowitz SM, et al. Induction of neurallymediated syncope with adenosine. Circulation 1999;99:1318-24.
Formal analysis of the optimal duration of tilt testing for the diagnosis of neurally mediated syncope Kenneth M. Stein, MD, David J. Slotwiner, MD, Suneet Mittal, MD, Marc Scheiner, MD, Steven M. Markowitz, MD, and Bruce B. Lerman, MD New York, NY
Background Although tilt testing has emerged as the test of choice for assessing patients with suspected neurally mediated syncope, the optimum duration of tilt testing is poorly defined. This in part relates to the absence of a gold standard to assess test performance.
Objective Our purpose was to formally estimate the effects of varying duration of drug-free tilt testing on test performance in diagnosing neurally mediated syncope.
Design If a test’s specificity is known, then in the absence of a gold standard an imputed (estimated) sensitivity may be calculated on the basis of the observed diagnostic yield in a given population as a function of assumed population prevalence. We determined the relationship of specificity to drug-free tilt test duration by use of data from 11 previous studies reporting the results of drug-free tilt testing in a total of 435 control subjects (60 to 80 degrees of tilt, footboard support, 15to 60-minute duration). Data (weighted for study size) were fit to an exponential function relating specificity to tilt duration. Test yield was evaluated as a function of tilt duration in 213 consecutive patients referred to our laboratory for the evaluation of suspected neurally mediated syncope who underwent passive tilt testing for up to 30 to 60 minutes.
Results The estimated specificity of tilt testing was 94% at 30 minutes, 92% at 40 minutes, and 88% after 60 minutes of passive tilt. The cumulative yield of tilt testing was only 17% at 30 minutes, 22% at 40 minutes, and 28% after 60 minutes. On the basis of an estimated population prevalence of 25% to 50% in this referral population, imputed sensitivity is 27% to 48% at 30 minutes, 36% to 64% at 40 minutes, and 43% to 74% after 60 minutes of passive tilt. The overall diagnostic accuracy was not strongly influenced by tilt duration beyond 30 minutes and ranged from 60% to 84%.
Conclusions Passive tilt testing (ie, tilt testing without pharmacologic provocation) for durations of up to 60 minutes has limited sensitivity for diagnosing neurally mediated syncope. For populations with a pretest likelihood of 25% to 50%, test results are inaccurate in one to two fifths of patients. (Am Heart J 2001;141:282-8.)
Tilt table testing has emerged as the test of choice for assessing patients with suspected neurally mediated syncope. However, many different protocols are used and the optimum duration of tilt testing remains controversial, as does the use of pharmacologic maneuvers to enhance sensitivity.1 This state of affairs relates in large part to the absence of a “gold standard,” a way of perfectly identifying true positives and false negatives, for From the Division of Cardiology, Department of Medicine, New York Hospital–Cornell Medical Center, New York, NY. Supported in part by National Institutes of Health grant No. RO1 HL5139. Submitted January 31, 2000; accepted October 13, 2000. Reprint requests: Kenneth M. Stein, MD, Division of Cardiology, Starr-4, New York Hospital, 525 E 68th St, New York, NY 10021. E-mail: [email protected] Copyright © 2001 by Mosby, Inc. 0002-8703/2001/$35.00 + 0 4/1/112236 doi:10.1067/mhj.2001.112236
the diagnosis. Such a gold standard is required to accurately evaluate test sensitivity and diagnostic performance. Thus the American College of Cardiology expert consensus on tilt testing concluded that “On the basis of studies in control subjects discussed earlier, it is clear that tilt table testing exhibits a high level of diagnostic specificity. Sensitivity, in contrast, is a more difficult issue. Apart from tilt table testing itself, there is no clearcut, accepted ‘gold standard’ for establishing a diagnosis of neurally mediated (and especially vasovagal) syncope against which diagnostic procedures can be measured.”2 Owing to the absence of a gold standard, clinicians have been forced to use the yield of the test in patients believed to have a high pretest likelihood of neurally mediated syncope as an (admittedly imperfect) estimate of sensitivity.3 However, it can be shown mathematically that, even
American Heart Journal Volume 141, Number 2
Stein et al 283
in the absence of a gold standard, if specificity is known then an “imputed” (estimated) sensitivity can be calculated as a function of assumed prevalence by use of the observed diagnostic yield. The purpose of the current study was to develop a technique for computing imputed sensitivity that would potentially be applicable to all situations in which a diagnostic test of known specificity is applied in the absence of a gold standard. We then sought to use this technique to evaluate the effect of the duration of passive tilt testing on the utility of tilt testing for diagnosing neurally mediated syncope.
Methods Calculation of imputed sensitivity If n total patients undergo a diagnostic test and we represent the true positives by a, the true negatives by b, the false positives by c, and the false negatives by d (n = a + b + c + d), then we may define the true-positive rate as a/n, the falsepositive rate as c/n, the true-negative rate as b/n, and the false-negative rate as d/n. The yield (Y) (or positivity rate) of a diagnostic test as a function of duration (time, t) may then be defined as the sum of the true-positive (TP) and the falsepositive (FP) rates of the test as functions of duration: Y(t) = TP(t) + FP(t) For a given population the true-positive rate depends on the sensitivity (Se) and disease prevalence (Pr) in that population, whereas the false-positive rate depends on specificity (Sp) and prevalence. Thus yield can be expressed by reference to sensitivity, specificity, and prevalence: Y(t) = Se(t)Pr + (1 – Sp[t])(1 – Pr) Finally, if the test specificity is known and if test yield can be determined for a given population, then an imputed sensitivity (Sei) may be computed as a function of assumed disease prevalence in that population: Sei(t,Pr) =
July 1998. All patients underwent tilt-table testing after an overnight fast. After written informed consent was obtained, each patient was placed in the supine position on a motorized tilt table with footboard support and an intravenous catheter was placed in an arm vein. Continuous electrocardiographic monitoring was implemented. Noninvasive automated blood pressure monitoring was performed (Dinamap, Critikon, Tampa, Fla) and the blood pressure and heart rate were recorded at 1-minute intervals. A manual blood pressure cuff was used to verify any decrease in blood pressure recorded by the automatic cuff or when warranted by clinical symptoms. After a 15-minute supine control phase, patients were tilted upright to an angle of 60 degrees for 30 or 60 minutes or until a positive response was achieved. A positive (abnormal) response was defined as (1) sudden loss of consciousness or the development of presyncope in association with an abrupt decrease in systolic blood pressure to 80 mm Hg and (2) the reproduction of symptoms that had been associated with the patient’s clinical event(s). A positive response had to meet both criteria. With use of these data, the relationship between tilt duration and positivity rate (and standard error of the relationship) was then computed by the life-table (product limit) method.
Results Tilt specificity In the 11 published studies reporting the results of passive tilt testing in a total of 435 normal subjects, specificity varied from 83% to 100% and declined modestly with increasing duration of the test (Figure 1). The data were reasonably approximated by an exponential distribution according to the equation Sp = e–0.0022t (P < .001, R2 = 59%). The estimated specificity of tilt testing was 94% at 30 minutes, 92% at 40 minutes, and 88% after 60 minutes of passive tilt.
Y(t) – (1 – Sp[t])(1 – Pr) Pr
Tilt specificity as a function of tilt duration By computerized searching, as well as by review of the relevant literature, we identified 11 previous studies reporting the results of drug-free tilt testing in healthy control subjects subject to the following restrictions: (1) 60 to 80 degrees upright tilt, (2) footboard support, and (3) 15- to 60-minute duration.414 The data were then weighted for study size and were fit to an exponential function relating specificity to tilt duration, assuming 100% specificity at 0 minutes (Curve Estimation procedure, SPSS for Windows 8.0.1, SPSS, Chicago, Ill).
Tilt yield as a function of tilt duration Tilt yield (positivity rate) was evaluated as a function of passive tilt duration in 213 consecutive patients referred to our laboratory for the evaluation of suspected neurally mediated syncope. The first 193 consecutive patients underwent passive tilt testing for 30 minutes between May 1995 and May 1997. The remaining 20 consecutive patients underwent passive tilt testing for 60 minutes between June and
Tilt yield Demographic characteristics of the patients referred for tilt testing are summarized in Table I. There were no differences between patients with positive and negative tilt tests with respect to age, sex, or the presence of underlying heart disease. Consistent with previous studies,15 there was a relationship between the historic frequency of syncope and the likelihood of a positive tilt test. Although many patients had recurrent syncope, nearly half (99/213, 46%) were referred for the evaluation of recurrent presyncope or after a single syncopal episode. Of particular note is the extremely low yield observed in patients with a history of recurrent presyncope but without true syncope (1/35, 3% positive rate). The overall yield of passive tilt testing was low in this referral population. The cumulative yield of tilt testing was 3% ± 1% (estimate ± SE) at 10 minutes, 8% ± 2% at 20 minutes, 17% ± 3% at 30 minutes, 22% ± 6% at 40 minutes, and 28% ± 8% after 60 minutes of passive tilt.
American Heart Journal February 2001
284 Stein et al
Figure 1
Specificity as a function of tilt duration in 11 previously published studies reporting the results of drug-free tilt testing (60 to 80 degrees upright tilt with footboard support) in healthy control subjects. The size of each circle reflects the size of the control population in the study.
Table I. Patient characteristics
Age (y) Sex (male/female) Heart disease(%) Episodes of syncope Indeterminate Presyncope only 1 2-3 4+
Positive tilt test (n = 36)
Negative tilt test (n = 177)
54 ± 22 14 (39%) 22 (61%) 9 (25%)
57 ± 20 73 (41%) 104 (59%) 43 (24%)
1 (3%) 1 (3%) 12 (33%) 9 (25%) 13 (36%)
6 (3%) 34 (19%) 52 (29%) 50 (28%) 35 (20%)
Statistical significance NS NS NS P = .07
Results are expressed as mean ± SD where appropriate. Heart disease is defined as a known prior history of coronary artery disease, significant valvular heart disease, or cardiomyopathy. NS, Not significant.
Tilt performance On the basis of an assumed disease prevalence of 25% to 50% in our referral population, imputed sensitivity ranged from 27% to 48% after 30 minutes, from 36% to 64% after 40 minutes, and from 43% to 74% after 60 minutes of passive tilt (Figure 2). The overall diagnostic accuracy (true-positive + true-negative rate) was greater for 30-minute tilts than for shorter durations but only
improved modestly thereafter (Figure 3). Thus, on the basis of an assumed prevalence of 25% to 50%, the estimated diagnostic accuracy was 60% to 82% after 30 minutes, 64% to 85% after 40 minutes, and 65% to 84% after 60 minutes of passive tilt. It is possible to formally assess the information derived from a diagnostic test by computing the difference between pretest and posttest likelihood of disease on the basis of positive versus negative tests (“diagnostic information,” see Figure 4). The posttest likelihood of disease for a positive test is its positive predictive value. The posttest likelihood of disease for a negative test is equal to 1 minus the negative predictive value. Curves demonstrating the posttest likelihood of disease after negative versus positive tilt tests are shown in Figure 4, based on “best-case” estimated sensitivities of 48%, 64%, and 74%, respectively, after 30, 40, and 60 minutes and estimated specificities of 94%, 92%, and 88% after 30, 40, and 60 minutes of tilt. As with all diagnostic tests, the information derived from testing is greatest for patients with an intermediate pretest likelihood of disease. Within this range, there was a small increase in information for the 40minute test compared with the 30-minute test but no clinically important further increase for the 60-minute
American Heart Journal Volume 141, Number 2
Stein et al 285
Figure 2
Imputed sensitivity of tilt testing for the diagnosis of neurally mediated syncope as a function of tilt duration and assumed true disease prevalence in our study population.
Figure 3
Estimated diagnostic accuracy (true-positive + true-negative rate) of tilt testing for the diagnosis of neurally mediated syncope as a function of tilt duration and assumed true disease prevalence in our study population.
test. For patients with a pretest likelihood below 33%, there was no important difference among 30-, 40-, and 60-minute tests. With use of these “best case” assumptions, the positive predictive value of a 40-minute pas-
sive tilt test in a patient with a 50% pretest likelihood of disease is 88%. However, the posttest probability despite a negative test (1 minus the negative predictive value) is still 28%.
American Heart Journal February 2001
286 Stein et al
Figure 4
Posttest likelihood of disease in patients with positive versus negative tilt tests and diagnostic information (the difference between the posttest likelihoods of disease after positive vs negative tests) as a function of pretest likelihood of disease and passive tilt duration. The calculations are based on “best-case” estimated sensitivities of 48%, 64%, and 74%, respectively, after 30, 40, and 60 minutes and estimated specificities of 94%, 92%, and 88% after 30, 40, and 60 minutes of tilt. Note that the positive predictive value (PPV) curves at 30 and 40 minutes are virtually superimposed. NPV, Negative predictive value.
Discussion The current study represents the development of a formal technique for estimating the sensitivity of a diagnostic test for which no “gold standard” exists. Application of the imputed sensitivity analysis to tilt testing for the diagnosis of neurally mediated syncope leads to several conclusions: (1) the specificity of passive (drug-free) tilt testing is high for tilt durations of up to 60 minutes, (2) the yield of passive tilt testing is
relatively low in a current referral population, and (3) as a result, the imputed sensitivity of passive tilt testing is low for tilt durations of up to 60 minutes. The diagnostic information derived by tilt testing is greatest for patients with an intermediate pretest likelihood of disease. For these patients, a 40-minute tilt protocol is more informative than a 30-minute protocol, but little additional information is gained by prolonging the test to 60 minutes. Although the positive
American Heart Journal Volume 141, Number 2
predictive value of a 40-minute passive tilt test is high, there is a clinically significant false-negative rate for patients with an intermediate pretest likelihood of disease.
Specificity Although many in the cardiology community continue to express concern regarding false-positive results of tilt tests, a review of 11 studies in which passive tilt tests were performed in 453 healthy control subjects demonstrates that the test has high specificity.4-14 There was a tendency for specificity to diminish with prolongation of upright tilt. Even so, specificity is estimated to be 88% at the end of a 60-minute drug-free tilt test by use of current methods.
Yield The yield of drug-free tilt testing observed in the current study was disappointingly low (17% after 30 minutes and 28% after 60 minutes). This is in contrast with some (mostly early) studies reporting yields of 50% to 74% for passive tilt tests of 45 to 60 minutes in patients with syncope of unknown origin.5,16-18 However, the data are consistent with the substantially lower yields reported in other studies.4,11,13-15,19-22 Some of these differences in results may be attributable to subtle differences in method (eg, tilt angle, invasive vs noninvasive hemodynamic monitoring). It is also highly likely that the population referred for tilt testing is different from the population evaluated in the initial reports of the method, in that it is now more heterogeneous, and the true prevalence of neurally mediated syncope may be lower in patients referred for the test today than in the past. This reflects the natural evolution of a diagnostic modality away from being an experimental procedure applied first to a homogeneous high-risk population and toward being a test used to gain information about patients with an intermediate probability of disease.
Sensitivity and optimum duration of testing As a result of the low yield, the imputed sensitivity of the passive tilt test is low (although still within a range considered acceptable for other diagnostic modalities). Although sensitivity increases with duration of tilt testing, specificity declines. As a result, overall test accuracy (which was estimated to range from 60%-85%) is not strongly influenced by tilt durations beyond 30 minutes. The diagnostic information derived from testing was greatest at intermediate probabilities of disease and was greater for a 40-minute test than for shorter-duration tilts. Little additional information was gained by extending the tilt to 60-minute duration. The positive predictive value of passive tilt testing in the 40- to 60-minute range was estimated to be high. However, the passive tilt test probably has a clinically important false-negative rate for patients with an inter-
Stein et al 287
mediate pretest likelihood of disease. These results emphasize the potential importance of tilt protocols with pharmacologic provocation to reduce the falsenegative rate of tilt testing.14,23,24 It should be strongly emphasized that the technique for calculating imputed sensitivity yields estimates and not true values for sensitivity. These estimates are critically dependent on the validity of the assumptions regarding disease prevalence in the population studied. Although the assumption that the prevalence was in the range of 25% to 50% in the current population seems reasonable, it is possible that the actual prevalence may be higher or lower. It should be noted that if the actual prevalence were any higher than 50% then the imputed sensitivity would be correspondingly lower (eg, if the actual prevalence were 75% then the imputed sensitivity for 30 minutes of passive tilt would have only been 19%). On the other hand, on the basis of the estimated specificity of 88% at 60 minutes and the observed yield of 28% at 60 minutes, it would not have been mathematically possible for the population prevalence to have been substantially less than 20%.
Limitations It should be noted that a positive tilt test in the asymptomatic control populations consisted of hemodynamic collapse, whereas a positive tilt test in the patients with syncope consisted of hemodynamic collapse with reproduction of the clinical symptoms. Clearly, if this stricter definition were applied to the control subjects, there could have been no “positive” tests and the specificity would be 100%. It is theoretically possible that in practice some of the “false-positive” tests predicted on the basis of the specificity estimates were correctly classified as “negative” as a result of failure to reproduce clinical symptoms. Although this problem is far from trivial, in this particular population there were no tests that were reclassified from positive to negative on the basis of failure to duplicate clinical symptoms. It should also be noted that the choice of an exponential form to fit the specificity data was based on the theoretic need that the specificity could never be less than 0, should reach 0 by an infinite duration of tilt, and should be 100% at time 0. It should be observed that the actual computed curve (Sp = e–0022t) is almost precisely linear over the times that we analyzed (0 to 60 minutes) and that substitution of a linear curve fit does not affect the results. It should also be noted that relatively few patients underwent the full 60minute tilt protocol (although the SEs for the estimates of tilt yield at 40-60 minutes are low). Finally, it should be noted that the estimates of specificity were derived from pooled analysis of multiple, potentially heterogeneous studies and are subject to the limitations inherent in this approach. It should also be pointed out that, although a similar
288 Stein et al
analysis applied to isoproterenol tilt testing would be of interest, there is not sufficient standardization of isoproterenol tilt protocols in the literature, in terms of dose, duration, and the duration of any preceding passive tilt phase, to permit a comparable analysis.
Conclusions If the specificity of a diagnostic test is known, then even in the absence of a gold standard, test sensitivity can be estimated on the basis of the assumed disease prevalence in a given referral population. The results of the analysis suggest that a 40-minute tilt protocol may provide more diagnostic information than a 30-minute period but that little additional information is gained by extending the protocol to 60 minutes. Although the predictive value of a positive passive tilt test is high, there is a significant false-negative rate, emphasizing the potential importance of tilt testing with pharmacologic provocation for the diagnosis of neurally mediated syncope.
References 1. Benditt DG, Remole S, Bailin S, et al. Tilt table testing for evaluation of neurally-mediated (cardioneurogenic) syncope: rationale and proposed protocols. PACE 1991;14:1528-37. 2. Benditt DG, Ferguson DW, Grubb BP, et al. ACC expert consensus document: tilt table testing for assessing syncope. J Am Coll Cardiol 1996;28:263-75. 3. Benditt DG. Neurally mediated syncopal syndromes: pathophysiological concepts and clinical evaluation. PACE 1997;20:572-84. 4. Abi-Sainra F, Maloney JD, Fouad-Tarazi FM, et al. The usefulness of head-up tilt testing and hemodynamic investigations in the workup of syncope of unknown origin. PACE 1988;11:1202-14. 5. Fitzpatrick A, Sutton R. Tilting towards a diagnosis in recurrent unexplained syncope. Lancet 1989;1:658-60. 6. Grubb BP, Temesy-Armos P, Hahn H, et al. Utility of upright tilt-table testing in the evaluation and management of syncope of unknown origin. Am J Med 1991;90:6-10. 7. Fitzpatrick AP, Theodorakis G, Vardas P, et al. Methodology of head-up tilt testing in patients with unexplained syncope. J Am Coll Cardiol 1991;17:125-30. 8. Fitzpatrick AP, Theodorakis G, Vardas P, et al. The incidence of malignant vasovagal syndrome in patients with recurrent syncope. Eur Heart J 1991;12:389-94.
American Heart Journal February 2001
9. Grubb BP, Wolfe D, Samoil D, et al. Recurrent unexplained syncope in the elderly: the use of head-upright tilt table testing in evaluation and management. J Am Geriatr Soc 1992;40:1123-8. 10. Kapoor WN, Brant N. Evaluation of syncope by upright tilt testing with isoproterenol: a nonspecific test. Ann Intern Med 1992;116:358-63. 11. Lurie KG, Dutton J, Mangat R, et al. Evaluation of edrophonium as a provocative agent for vasovagal syncope during head up tilttable testing. Am J Cardiol 1993;72:1286-90. 12. Natale A, Akhtar M, Jazayeri M, et al. Provocation of hypotension during head-up tilt testing in subjects with no history of syncope or presyncope. Circulation 1995;92:54-8. 13. Raviele A, Menozzi C, Brignole M, et al. Value of head-up tilt testing potentiated with sublingual nitroglycerin to assess the origin of unexplained syncope. Am J Cardiol 1995;76:267-72. 14. Aerts A, Dendale P, Strobel G, et al. Sublingual nitrates during head-up tilt testing for the diagnosis of vasovagal syncope. Am Heart J 1997;133:504-7. 15. Fitzpatrick AP, Lee RJ, Epstein LM, et al. Effect of patient characteristics on the yield of prolonged baseline head-up tilt testing and the additional yield of drag provocation. Heart 1996;76:406-11. 16. Kenny RA, Ingram A, Bayliss J, et al. Head-up tilt: a useful test for investigating unexplained syncope. Lancet 1986;1:1352-5. 17. Raviele A, Gasparini G, DiPede F, et al. Usefulness of head-up tilt test in evaluating patients with syncope of unknown origin and negative electrophysiologic study. Am J Cardiol 1990;65:1322-7. 18. Ruiz GA, Peralta A, Gonzalez-Zuelgaray J, et al. Evolution of patients with clinical neurally-mediated (vasovagal) syncope not subjected to specific treatment. Am Heart J 1995;130:345-50. 19. Morillo CA, Klein GJ, Zandri S, et al. Diagnostic accuracy of a lowdose isoproterenol head-up tilt protocol. Am Heart J 1995;129: 901-6. 20. Strasberg B, Rechavia E, Sagie A, et al. The head-up tilt table test in patients with syncope of unknown origin. Am Heart J 1989;118: 923-7. 21. Lippman N, Stein KM, Lerman BB. Differential therapeutic response of patients with isoproterenol-dependent and isoproterenol-independent vasodepressor syncope. Am Heart J 1994;128:1110-6. 22. Pavri BP, Ruskin JN, Brooks R. The yield of head-up tilt testing is not significantly increased by repeating the baseline test. Clin Cardiol 1996;19:494-6. 23. Almquist A, Goldenberg IF, Milstein S, et al. Provocation of bradycardia and hypotension by isoproterenol and upright posture in patients with unexplained syncope. N Engl J Med 1989;320: 346-51. 24. Mittal S, Stein KM, Markowitz SM, et al. Induction of neurallymediated syncope with adenosine. Circulation 1999;99:1318-24.