The Missing Waveform Information in the Orthogonal Electrocardiogram (Frank Leads)

Transcription

1 The Missing Waveform Information in the Orthogonal Electrocardiogram (Frank Leads) IV. Computer Diagnosis of Biventricular Hypertrophy From 'Maximal" Surface Waveform Information By F. KORNREICH, M.D., P. BLOCK, M.D., AND D. BRISMEE, PH.D. SUMMARY A new lead system recording nine surface ECGs was found capable of resynthesizing in each patient all the QRS waveforms recorded on the thoracic surface. Observations were made in 416 patients: 150 normal (N) individuals, 95 patients with left ventricular hypertrophy (LVH), 97 patients with right ventricular hypertrophy (RVH), and 74 patients with biventricular hypertrophy (BVH). After time-normalization and division of the nine waveforms into eight equal parts, 72 variables were determined for each patient. Recordings from Frank XYZ leads were also available for each subject. After time-normalization and division in eight equal parts, 24 variables were obtained from Frank lead recordings in each individual. The 72 variables from the new system and 24 variables from the were then submitted to multivariate statistical procedures in order to differentiate BVH from N, LVH, and RVH. Four parameters were finally retained for both the and the as the best discriminators for the separation of BVH from normal, BVH from LVH, and BVH from RVH. The diagnostic performances, defined as 1/2 (specificity + sensitivity), were 93.5%,, and 81%, respectively, for the and 88.5%, 69%, and 75% for the. The repeatability of the results was tested on independent control samples and found reasonable. The limited clinical use of independent pairwise comparisons is discussed and the alternative of stepwise pairwise comparisons following a decision tree is proposed. Some deterioration in the final classification is noticed; the recognition rates are 90% for normal, 79% for LVH, 75% for RVH and 73% for BVH (average 79%) with the, and 90% for N, 55% for LVH, 46% for RVH and 59% for BVH (average 62%) with the. Additional Indexing Words: Nine-lead system Maximal waveform information Linear discriminant function Biomedical computer program BMD 07M (stepwise discriminant analysis) Biomedical computer program BMD 04M (multiple discriminant analysis) Multivariate analysis Pairwise comparisons THE ECG DIAGNOSIS of biventricular hypertrophy (BVH) still represents a challenge. The percentages of correct classifications of BVH reported in the literature range between 4% and 46%.'`7 More recently, Gamboa, Klingeman, and Pipberger, applying multivariate statistical procedures to the Frank leads, were able to recognize 69% of a BVH sample8; their series included 87 patients with necropsy evidence of biventricular hypertrophy. The subsequent analysis of a sample with clinically diagnosed BVH resulted in a correct classification of 44% of the From the Unit for Cardiovascular Research, School of Medicine, Free University, Brussels, Belgium. Address for reprints: F. Kornreich, M.D., Director, Unit for Cardiovascular Research, School of Medicine, Free University, Brussels (V.U.B.), Eversstraat 2, 1000 Brussels, Belgium. Received January 15, 1974; revision accepted February 20, Circulationl, Volume XLIX, Juine new cases. The factors contributing to the low recognition rate of BVH in the literature are manifold; one of them is probably related to the selection of the lead system. In a previous paper9 a lead system, consisting of nine surface electrocardiogram recordings taken at fixed, well-defined anatomical locations, was designed. In each patient these nine leads, linearly combined, were found capable of resynthesizing, within the noise level, all the waveforms recorded on the thoracic surface. The object of the present paper was to apply this 9- lead system to patients with biventricular hypertrophy and to compare its diagnostic performance to that of the. Both systems were submitted to identical statistical procedures. Differentiation between the diagnoses of biventricular hypertrophy and normal persons, patients with left ventricular hyper-

2 1224 Classification of Patients Studied Table 1 KORNREICH, BLOCK, BRISMEE Number of patients Number of patients Diagnosis (learning groups) (control groups) Normal Left ventricular hypertrophy (LVH) Right ventricular hypertrophy (RVH) Biventricular hypertrophy (BVH) Total (416) trophy, and patients with right ventricular hypertrophy is also discussed. Material and Methods The material consisted of 416 well-documented patients (table 1); for each patient nine surface leads were retained for further analysis.*le The were also available for each patient. All patients underwent a clinical evaluation, including chest X-ray and ECG (the ECG data was not used as a criterion for the building up of the series). The patients with a negative history, a normal physical examination, and a normal chest X-ray constituted the normal group. The ages of this group ranged from 20 to 50 years. The diagnosis of left ventricular hypertrophy (LVH) was based on cardiac catheterization during which hemodynamic data and biplane angiograms were obtained, permitting the evaluation of LV volumes and muscle weights. Left ventricular hypertrophy was defined as left ventricular muscle weight (LVMW) greater than two standard deviations above the mean LVMW for adult male subjects, which is 188 g + 33 g.", 12 The upper limit for normal is thus 254 g. All the patients in the LVH learning group (52 patients) had isolated aortic valve disease. The 43 patients in the LVH control group had aortic valve disease which could be combined with other heart conditions such as hypertension (six cases), myocardial infarction (four cases) or angina pectoris (seven cases). The right ventricular systolic pressure never exceeded 30 mm Hg. The patients with a right ventricular systolic pressure (RVSP) exceeding 35 mm Hg'3 were considered as having right ventricular hypertrophy. The learning RVH group (57 patients) was composed of patients with mitral stenosis (46 cases), cor pulmonale (four cases), ventricular septal defect with pulmonic stenosis (three cases), or tetralogy of Fallot (four cases). None of these patients presented a LVMW above 254 g. Most of the 40 patients of the control group had mitral stenosis (31) or cor pulmonale (nine cases). The biventricular hypertrophy group was classified on the basis of presence of both a LVMW > 254 g and a RVSP > 35 mm Hg. Various heart diseases were encountered, but most of them were related to multivalvular lesions with or without coronary artery disease. The ECG was used to exclude from either group records with a QRS duration sec. Only the QRS complex *ln each patient, these nine leads were found capable of resynthesizing all 126 waveforms recorded on the thoracic surface. This material was most kindly provided by Drs. J. Holt, A. Barnard, and J. Kramer, Jr. (Birmingham, Alabama). was considered in the present study. Time normalization was performed by dividing the QRS wave in eight equal parts,'4 a process which produced 72 variables for each individual in the new system. Identical time normalization was performed with the, so that each subject was represented by 24 variables. The statistical procedures involved in this work were also described in more detail in a previous paper.'0 Briefly stated, the best discriminators between two groups were selected by stepwise discriminant analysis (Biomedical Computer Program 07M)"; these selected variables were then entered into another program (Biomedical Computer Program 04M)" for the computation of the linear function L: L = x1x1 + x2x XnXn in which n is the number of arbitrarily chosen variables and X the corresponding coefficients. First, the discriminant functions were determined on the learning groups and the patients classified by means of a likelihood ratio test; then the classification method was tested on new independent samples (control groups) in order to assess the repeatability of the statistical procedures. Results Table 2 gives the differentiation between records from normal patients and those from patients with biventricular hypertrophy; four variables were retained for the and for the. The best discriminators for the were located in the upper sternal and upper middorsal regions and depicted in figure 1. The performance defined as 1/2 (sensitivity + specificity)'1 was 93.5% for the and 88.5% for the. The performance on the control groups of normals and BVH for both systems gave values of 88% for the nine leads and 84.5% for the. In table 3 differentiation between BVH and LVH is shown: the performance for the reached and for the, 69%. The best discriminators between LVH and BVH were found on the left precordial leads in the (fig. 2). Most discriminating measurements on the were situated on the X lead. The performance levels obtained with the control groups of BVH and LVH for the nine leads and the were 82% and 70%, respectively. Circulationl. Volumnie XLIX, June 1974

3 MISSING INFORMATION IN FRANK LEADS 1225 Table 2 Differentiation between Normal Records and Those From Patients with Biventricular Hypertrophy 6/8 lead 3 5/8 lead 1 2/8 lead 8 2/8 lead 2 Diagnostic accuracy classified) (N) % 92% 93.5% % 88% 4/8 Z 3/8 X 3/8 Y 7/8 X classified) (N) *Product of the coefficients and averaged means of the normal and biventricular hypertrophy group. These products indicate the true contribution to differentiation.19 The differentiation between BVH and RVH is described in table 4. The four variables chosen for the as best discriminators between these two pathological entities were equally distributed over the left sternal and left precordial and the upper mid and left dorsal regions (fig. 3). The yielded a performance of 81% and the, 75%. The controls were 78% and 76%, respectively. The pairwise differentiations thus far presented - normal vs BVH, BVH vs LVH, and BVH vs RVHare useful for the determination of the best discriminators if one particular pathological entity is considered, but, in a normal clinical setting we deal with patients in whom the diagnosis is not known, or at best, presumed. All ECG diagnoses have to be considered simultaneously.`6 Dealing in this paper with the particular problem of ventricular hypertrophy and wishing to enhance the clinical usefulness of the discriminant functions, a stepwise procedure (fig. 4) is proposed which could be applied to patients in which some kind (not a priori known) of cardiac hypertrophy, whether pure right, pure left, or combined is present. In a first step, we differentiated the normal individuals from a pool of patients with pure left, pure right, or combined ventricular hypertrophy. Table 5 Table 3 Differentiation Between Records from Patients with Biventricular Hypertrophy and Those from Patients with Left Ventrictular Hypertrophy 95% 82% 88.5% % 1 78% 84.a-% 1/8 lead 7 1/8 lead 5 6/8 lead 6 3/8 lead 5 Diagnostic accuracy /8 X 1/8 X 3/8 Y 6/8 X classified) (LVH) 81% 83% 82% classified) (LVH) 53% 69% 78% 62% 70% *See note to table 2 for derivation of this term. Circulation, Volutme XLIX, June 1974

4 1226 KORNREICH, BLOCK, BRISMEE BVH vs N S lis 7 ir22 30&t~ 37450s_ 53 9.d, e J6 2? 3! *.s S.'67n!? J s _1~,,, Le Figure 1 The locations of the electrodes for the are indicated with large dots. The corresponding waveforms are depicted and named L,,L2 L, and referred to in the tables as lead 1, lead 2.. lead 9. The waveforms represent averaged, time-normalized QRS complexes, divided in eight equal parts: on the abscissa, 1, 3, 5 and 7 mean '8 QRS, 3/s 9QRS, 5/ QRS, and -/ pqrs. The voltages are given in microvolts and 0 represents the baseline. The best discriminators are depicted on the waveforms with vertical bars. This figure represents the discrimination between normal patients (N) and patients with biventricular hypertrophy (BVH). gives the differentiation achieved between these two groups, using eight discriminators: a performance of 89.5% was obtained with the and of 82.5% with the. The repetition of these measurements on control groups yielded 86.5% for the nine leads and 79% for the. In the second step, differentiation between left and right ventricular hypertrophy was looked for in the group of patients classified as "hypertrophy" after the first step. The discriminant function computed for this differentiation can be seen in table 6. s of 87.5% and 76% were achieved for the nine leads and the respectively; the controls yielded 82% and 69%, respectively. Finally, in a last step differentiation between BVH and LVH, and BVH and RVH was sought using the discriminant functions shown in tables 3 and 4. The final results of this stepwise procedure and the results in separate control groups are shown in table 7 for both lead systems. The total percentage of correctly classified patients was 79% (control 73%) with the nine lead system and 62% (control 55%) with the. Constitution of the Series Discussion The diagnostic crite'ria on which the selection of patients was based can always be questioned if the selection is made with determining left ventricular mass and the pressure in the right ventricle. In this (,'irccilation. Volumnte XLIX, June 1974

5 MISSING INFORMATION IN FRANK LEADS V/- vs LVH 14 Y , 6 S.J h* 52 *' 00t et, 7, e.s'!5 ss >' Figure 2 For explanation of figure format see legend to figure 1. This figure represents the discrimination between patients with left ventricular hypertrophy (LVH) and patients with biventricular hypertrophy (BVH). Table 4 Differentiation Between Records from Patients with Biventricular Hypertrophy and Those from Patients Right Ventricular Hypertrophy wuih 3/8 lead 4 2/8 lead 8 5/8 lead 9 7/8 lead 6 Diagnostic accuracy /8 Z 8/8 X 2/8 Y 8/8 Y classified) (RVH) 77% 81% 82% 74% 78% classified) (RVH) 65% 75% 78% 75% 76.5% *See note to table 2 for derivation of this term. Circulation, Volunme XLIX, June 1974

6 1228 KORNREICH, BLOCK, BRISMEE =L1 o1 A~~~~~ A\ S. / Ll t s5 22 e 3? 4551 S5 es.e S8VH vss R H M 23 * 4 * n 7 52 lb5 t > no 1r. T X1 tcp 7s l2 T ". l. 01 ton4 m if, If ^ ~~~ L L j 1.1.~~~~~~~~~~~~~. 3 0 Figure 3 See legend to figure 1 for explanation of the format. This figure represents the discrimination between patients with right ventricular hypertrophy (RVH) and patients with biventricular hypertrophy (BVH). study, left ventricular hypertrophy was estimated from biplane angiography and right ventricular hypertrophy from the systolic pressure in the right ventricle; biventricular hypertrophy was based on the presence of both criteria. Whether other parameters N-VH L VH-BVH VH AV1I SBVH BVH Figure 4 RVH-BVH Flow diagram of the different steps followed in the classification of patients with ventricular hypertrophy. N = normal; VH = ventricular hypertrophy, (not localized); LVH = left ventricular hypertrophy; RVH = right ventricular hypertrophy; BVH = biventricular hypertrophy. ought to have been used remains an open question. The main point of this study is that the diagnostic performance of the was comparatively better than that of the Frank lead system when applied to identical normal and pathological groups. Differentiation between Normal and Biventricular Hypertrophy The recognition rate of BVH with both systems was rather high, indicating that the patients of the BVH group had clear-cut abnormal electrocardiographic patterns, which is not in favor of a postulated pseudonormalization7 of the ECG when simultaneous hypertrophy of both ventricles is present. It appears that the characteristic ECG patterns of BVH as compared to the normals was not merely the sum of the LVH patterns plus the RVH patterns. First, when records of patients with pure LVH and others with pure RVH were pooled into one single group and discriminated with respect to the normals, the discriminators were C'irtctlaitiotn, Volutnme XLIX, Jttne 1974

7 MISSING INFORMATION IN FRANK LEADS 1229 Table 5 Differentiation Between Normal Records and Records from Patients with Either Pure Right, Pure Left, or Cambined Ventricular Hypertrophy 6/8 lead /8 lead /8 lead /8 Iead /8 lead /8 lead /8 lead /8 lead Diagnostic accuracy /8 Z 8/8 Z 3/8 X 4/8 Z 8/8 X 1/8 X 6/8 Z 8/8 Y classified) (N) classified) (VH) 95% 84% 89.5% 93% 80% 86.a-% classified N) classified VH) 95% 70% 82.5% 90% 68% 79% *See note to table 2 for derivation of this term. different than those distinguishing BVH from the same normals. Second, a fairly good differentiation was achieved between BVH and the above mentioned pool of LVH and RVH, resulting in a performance of 82% for the and only 66% for the Frank leads, stressing further the significance of the selection of a proper lead system. Differentiation of BVH and LVH and BVH and RVH The performance level of the was markedly better than the for differentiation of BVH and LVH. The discrepancy between the systems was less important in discrimination of BVH from RVH. Only three leads were involved in the separation of BVH from LVH and four leads for the comparison BVH with RVH. The finding of a quantitative relationship between the degree of enlargement of either ventricle and the accuracy of the diagnosis would have been of interest. One would expect that BVH might be misclassified as RVH or LVH when there was a predominance of enlargement in one ventricle. Unfortunately the BVH able 6 Classification of Patients with Ventricular Hypertrophy into Those twvith Left Ventricular Hypertrophy and Those with Right Ventricular Hypertrophy 4/8 lead /8 X /8 lead /8 Z /8 lead /8 X /8 lead /8 X Diagnostic accuracy (Correctly 90% 84% Correctly 90% 82% classified) (LVH) classified LVH (Correctly 80%, Correctly 62% 56% classified) (RVH) classified RVH 87.5% 82% 76% 69% *See note to table 2 for derivation of this term. Circuilatiokn. volume XLIX, June 1974

8 1230 KORNREICH, BLOCK, BRISMEE Table 7 Percent Correct Differentiation Between Records of Normal Cardiac Patients and Records from Patients with LVH, RVH, and BVH by the Steptvise Procedure in the Learning Groups and Control Groups N 90%. 84% N 90% 82% LVH 79% 72%, LVH 515% 51% ' RVH 75% 66% RVH 46% 37%. BVH 73% 69% BVH 59% 39Cr Abbreviations: N = normal; LVH = left ventricular hvpertrophy; RVH = right ventricular hypertrophy; BVH = biventricular hypertrophy. cases misclassified as LVH were not the patients with highest LVMW and lowest RVSP; neither were the BVH cases misclassified as RVH, the patients with the lowest LVMW and highest RVSP. Conversely, there was no relationship between the patients with pure LVH or pure RVH misclassified as BVH and either low LVMW or low RVSP. No satisfactory explanation for these discrepancies can be proposed by the authors; further studies on larger populations have to be conducted before definite conclusions can be drawn. Relation to Previous Work The authors wish to limit their comparison to an extensive well-documented study8 on biventricular hypertrophy using the and involving similar statistical procedures, namely, multivariate analysis and linear discriminant functions. In a paper published in 1969, Gamboa, Klingeman, and Pipberger were able to separate BVH from normals, LVH, and RVH with a performance of 84%, 78% and 78.5%, respectively. In our study the figures with the were 88.5%, 69%, and 75%, respectively. Except for the separation of BVH from LVH, where our figures were markedly lower, our other results by are very similar to the previous study. With the an over-all better performance is found: 93.5%,, and 81%, respectively, are found for the above diagnoses. Discriminant Functions in Clinical Practice As stated earlier (see Results) the practical utilization of separated, pairwise discrimination might lead the clinician to contradictory findings. For instance, a patient with BVH could be recognized as normal if the question normal or BVH is asked and yet be classified as LVH when the question BVH or LVH is raised and RVH when confronted with a BVH-RVH choice; it is indeed confusing for the clinician to find a patient with both LVH and RVH and still labeled normal when compared to BVH. Two solutions to this problem are possible: First, use of a decision tree leading ultimately to one unequivocal response, or second, use of multigroup diagnosis. The second possibility, although probably the most accurate one because it analyzes all possible ECG diagnoses simultaneously, is beyond the scope of this work and will be dealt with in a later paper. The first method leads to the results summarized in table 7. The advantage of this approach is an unequivocal classification, but this is offset by diminishing recognition rates as misclassifications occur at each step and are finally cumulative. Repeatability of the Results The reproducibility of the results on new independent samples, although reasonable, was not entirely satisfactory (tables 3-7): the main reason for this must be found in the marginal size of the samples involved in this study The amount of deterioration of results varies with the groups considered and indicates that one should adapt the sample size to the pathological condition under study.18 Larger populations have to be investigated before practical clinical use can be made of the computed discriminant functions. The main goal of this paper was the illustration of the diagnostic potential of a new, as compared to the. References 1. ROSEMAN RH, KRAUSE S, HIVANG W, KATZ LN: Electrocardiographic diagnosis of combined left and right ventricular strain. Am Heart J 40: 453, LIPSETT M, ZINN W: Anatomic and electrocardiographic correlation in combined ventricular hypertrophy. Am Heart J 45: 86, PAGNONI A, GOODWIN JF: Cardiographic diagnosis of combined ventricular hypertrophy. Am Heart J 14: 451, LANGENDORF R, HURWITZ M, KATZ LN: Electrocardiographic patterns of combined ventricular strain. Br Heart J 5: 27, HATTORI L: Study of correlation between ECG findings and Circildation. Voltume XLIX, June 1974

9 MISSING INFORMATION IN FRANK LEADS ventricular wall thickness based on cardio-synchronous angiocardiography: II. Study of ECG criteria for ventricular hypertrophy. Jap Circ J 30: 655, KOVATs-HOPFF L, WYss OAM: Vectorcardiographic signs of biventricular hypertrophy. Cardiologia (Basel) 48: 269, WOLFF R, MORSE RL, M AZZOLENI A: Combined ventricular hypertrophy. In Proc Long Island Jewish Hospital Symposium, Vectorcardiography. Amsterdam, North-Holland Publishing Co, 1966, p GAMBOA R, KLINGEMANJD, PIPBERGERHV: Computer diagnosis of biventricular hypertrophy from the orthogonal electrocardiogram. Circulation 39: 72, KORNREICH F: The missing waveform information in the orthogonal electrocardiogram (): I. Where and how can this missing waveform information be retrieved. Circulation 48: 984, KORNREICH F: The missing waveform information in the orthogonal electrocardiogram (): II. Diagnosis of left ventricular hypertrophy and myocardial infarction from "total" surface waveform information. Circulation 48: 996, HOLT JH JR, BARNARD ACL, LYNN MS: Study of the human heart as a multiple dipole electrical source: IL. Diagnosis and quantitation of left ventricular hypertrophy. Circulation 40: 697, KENNEDYJW, BAXLEYWA, FIGLEYMM, DODGEHT, BLACKMON 1231 JR: Quantitative angiocardiography: Normal left ventricle in man. Circulation 34: 272, HOLT JH JR, BARNARD ACL, LYNN MS, KRAMERJO J R A study of the human heart as a multiple dipole electrical source: III. Diagnosis and quantitation of right ventricular hypertrophy. Circulation 40: 711, PIPBERGER HV: Computer analysis of the electrocardiogram. In Computers in Biomedical Research, vol 1. New York, Academic Press, Inc., 1965, p DIXON WJ: Biomedical Computer Programs (B.M.D.). University of California Publications in Automatic Computation, No 2. Berkeley, University of California Press, CORNFIELD J, DUNN RA, BATCHLOR CD, PIPBERGER HV: Multigroup diagnosis of electrocardiograms. Comput Biomed Res 6: 97, CORNFIELD J: Statistical classification methods. In Computer Diagnosis and Diagnostic Methods, ed by JACQUEZ JA. Springfield, Illinois, Charles C Thomas, 1972, p PIPBERGER HV, SCHNEIDERMAN MA, KLINGEMANJD: The loveat-first-sight effect in research (Editorial). Circulation 38: 822, PIPBERGER HV, BERSON AS, KLINGEMAN JD, BATCHLOR CD: Diagnostic classifications of orthogonal electrocardiograms and vectorcardiograms. In Vectorcardiography 2, New York, North-Holland Publishing Company, 1970, p 157 Circulation, V'olhme XLIX, June 1974