The glaucomas are a heterogeneous group of progressive

Transcription

1 Multivariate Sensory Model in Glaucoma Diagnosis Peter Martus, 1 Matthias Korth, 2 FolkertHorn, 2 nselm Jiinemann 2 Martin Wisse 2 and Jost B. Jonas" PURPOSE. TO evaluate whether the combination of two psychophysical and two electrophysiological procedures improves diagnostic validity compared with single procedures. METHODS. In a clinical study, 73 patients with glaucoma from the University Eye Hospital in Erlangen and 122 healthy control subjects from the university staff, ranging in age from 19 to 62 years, underwent measurement of temporal contrast sensitivity using a full-field flicker test, spatiotemporal contrast sensitivity, blue-on-yellow visual evoked potential (EP), and a black-and-white, pattern-reversal electroretinogram. Diagnostic reference criteria included applanation tonometry, optic disc morphometry, and automated perimetry. Sensitivity was determined univariately with a fixed specificity of 80% and in a multivariate approach using logistic regression analysis. The classification rate was estimated using the leaving-one-out method. The correlation with intraocular pressure, visual field defects, and optic nerve defects was determined. RESULTS. Contrast sensitivity measurements and the blue-on-yellow pattern-onset EP showed comparable sensitivity (85%, 84%, and 85%) with 80% specificity, and a pattern-reversal electroretinogram showed lower sensitivity (64%). The first three methods contributed independent information to a diagnostic score. This score improved sensitivity to 94%, with a specificity of 89%. ll procedures moderately correlated with the neuroretinal rim area of the optic disc (r = ). The psychophysical tests showed a higher correlation with visualfielddefects (r > 0.5) than the electrophysiological tests (r < 0.3). CONCLUSIONS. The multivariate approach substantially increased the diagnostic validity compared with single procedures. This was probably because the diagnostic procedures under investigation tested different aspects of visual function. (Invest Ophthalmol is Sci. 1998;39: ) The glaucomas are a heterogeneous group of progressive eye diseases. Diagnosis is based on optic disc damage and visual field defects. In many cases, an increased level of intraocular pressure GOP) is observed before other symptoms occur, but cases of ocular hypertensive patients who never suffer glaucomatous damage and cases of lowtension glaucoma show that increased IOP is neither a sufficient nor a necessary criterion for glaucoma. Because perimetric deficits can be considered as relatively late symptoms, 1 numerous different sensory tests have been devised to find a more sensitive method of detecting glaucomatous damage before perimetric defects occur. lthough rather good sensitivities for certain procedures have been claimed, none of them has become a generally accepted part of a standard repertoire of procedures to diagnose glaucomas. Part of the problem is the heterogeneity of the damage to visual function in patients with glaucoma. The use of only a single sensory test might not, therefore, be a useful device to achieve a high sensitivity in glaucoma diagnosis for all patients. In several studies, multivariate analyses have been used to discriminate between patients with glaucoma and healthy sub- From the Departments of "Medical Statistics and Documentation and 2 Ophthalmology, University of Erlangen-Niirnberg, Germany. Supported by grants from the Deutsche Forschungsgemeinschaft (DFG Na 55/6 to 2). Submitted for publication February 11, 1997; revised February 13, 1998; accepted March 24, Proprietary interest category: N. Reprint requests: Peter Martus, Department of Medical Statistics and Documentation, University of Erlangen-Niirnberg, Waldstrafte 6, Erlangen, Germany. jects 2 6 or to predict visual field loss in the future. 7 " 9 Patients were classified according to history and systemic data,'^89 IOP, 389 morphometric data, 2 ' 4 ' 6 " 9 electrophysiological measurements, 4 ' 5 and early visualfieldloss. 6 However, multivariate analyses using intrinsically different functional diagnostic procedures (excluding visual field data) have remained exceptional. 5 In the Erlangen Glaucoma Study, among others, two psychophysical and two electrophysiological procedures have been considered. ll procedures had been developed earlier and were standardized based on results found in previous patients. The two psychophysical tests under investigation were the temporal contrast sensitivity (TCS) and the spatiotemporal contrast sensitivity (STCS). The validity of both methods has been shown previously. 10 " The two electrophysiological procedures were the recording of the pattern electroretinogram (PERG) in response to a rapid pattern-reversal stimulus and the recording of the visual evoked potential (EP) in response to the onset of a blue pattern presented on a strong yellow adaptation light (BY-EP). The validity of the PERG and BY-EP test as applied by us also has been shown previously. 121^ The four procedures have been investigated in the same patients. This opened the possibility to prove whether the different procedures contain independent diagnostic information using a multivariate analysis. In this analysis, we constructed a diagnostic score and examined how the classification rates were improved by stepwise inclusion of the different procedures. We present the results of the multivariate analysis and illustrate the construction of the diagnostic score. Investigative Ophthalmology & isual Science, ugust 1998, ol. 39, No. 9 Copyright ssociation for Research in ision and Ophthalmology 1567

2 1568 Martus et al. IOS, ugust 1998, ol. 39, No. 9 TBLE 1. Subject Characteristics ge (y) IOP (mm Hg) Neuroretinal Rim rea (mm 2 ) Mean Perimetric Defect (db) Glaucoma (rc = 73; 28 men, 45 women) POG (n = 53) LTG (n = 20) Control Subjects (n = 122; 70 men, 52 women) 50.5 ± 9.7 [53; 21-62] 51.2 ± 9.6 [54; 21-62] 48.9 ± 10.0 [51; 27-62] 41.2 ± 11.5 [42; 19-62] 28.1 ± 10.8 [24; 16-75] 31.2 ± 11.2 [27; 21-75] 19.9 ± 14 [20; 16-21] 16.5 ± 30 [17; 10-21] 1.01 ± 0.44 [1.00; ] 1.02 ± 0.46 [1.06; ] 0.97 ± 0.38 [0.89; ] 1.68 ± 0.31 [1.6; ] 7.2 ± 4.9 [6.0; ] 7.2 ± 5.2 [6.1; ] 7.0 ± 4.0 [6.0; ] 1.21 ± 1.18 [1.30; ] alues are mean ± SD, with median and minimum-maximum. Mann-Whitney test (53 with POG, 20 with LTG); P < IOP, intraocular pressure; LTG, low tension glaucoma; POG, primary open-angle glaucoma. METHODS Patients and Control Subjects Patients with Glaucoma. Eyes (n = 119) from 73 patients with glaucoma (53 with primary open-angle glaucoma [POG], IOP > 21 mm Hg; and 20 with low tension glaucoma [LTG], IOP < 21 mm Hg), 50.5 ±97 years old (mean ± SD), were included in the study. Forty-five patients were women, and 28 were men. The diagnosis of LTG was based on at least two IOP measurements before initial medical therapy. There were no significant differences in age, neuroretinal rim area, and visual defect between the patients with POG and those with LTG. The patients fulfilled the following criteria: optic disc damage (classification of Jonas, 14 glaucomatous optic disc stages I-) and pathologic cumulative perimetric defect curves (graphical display of ranked local defects compared with 95th and 99th percentiles of normal curves with identification of localized, diffuse, and broadly distributed visual field losses, Bebie 15 ). For 27 patients, only one eye was included because there was no evidence of glaucoma for the fellow eye (21 eyes with a normal visual field and normal optic disc) or the diagnosis remained unclear (6 eyes). Sixty-four (88%) of the patients with glaucoma were treated with eye drops. When pilocarpine drops had been prescribed, administration was discontinued in the morning of the day on which visual function was tested. Other eye drops (beta blocker, clonidine) were continued (Table 1). Control Subjects. Eyes (n = 223) from 122 healthy subjects aged 41.2 ± 11.5 years were included in the study; 52 were women, 70 were men. They were recruited from the staff of the hospital and the university administration. The subjects had normal IOP (<21 mm Hg), normal optic discs, and neither local nor diffuse visualfieldlosses. 15 Slit lamp examination and funduscopy revealed no pathologies (Table 1). For 21 subjects, only one eye was included because the fellow eye showed ocular hypertension (6 eyes), was suspect for optic disc damage (3 eyes), or had at least one diagnostic measurement that had not been performed (12 eyes). ll subjects had clear optic media and no elevation of IOP greater than 21 mm Hg on the day of examination. Exclusion criteria were other eye diseases (optic media opacities, vascular disease, and retinal disease) and systemic diseases (diabetes or arterial hypertension). Subjects who had false-positive and false-negative responses of more than 12% in visualfieldtesting were excluded. ll subjects had a visual acuity of 0.6 or better. Informed consent was obtained from each subject. Diagnostic Reference Criteria ll patients and control subjects were thoroughly examined using slit lamp inspection, applanation tonometry, funduscopy, gonioscopy, perimetry, and papillometry. In addition, a 24-hour IOP curve was measured in all patients (6 determinations: 7 M, 12 M, 5 PM, 9 PM, 12 PM, 7 M). Papillometric evaluations of patients and healthy subjects were based on 15 color photographs and subsequent planimetry (Morphomat 30; Carl Zeiss, Oberkochen, Germany) of the neuroretinal rim area. 14 Perimetry was performed in all subjects with a computerized static projection perimeter (Octopus 500, program Gl, 3 phases; Interzeag, Schlieren, Switzerland). Diagnostic Tests The technical details of the diagnostic procedures under investigation have been presented in several publications. We only give short descriptions here. Spatiotemporal Contrast Sensitivity Test. 10 n alternating sinusoidal stripe pattern was projected onto a temporal upper retinal area (13-8 horizontal, 4.2 vertical) using a screen (field size 53 X 4.3 ) surrounded by a white, square-shaped piece of cardboard (size 1.4 m X 1.4 m) that was illuminated by a bluish light of similar color and equal luminance (30 candela [cd]/m 2 ) as the screen. The choice of the target was motivated by a study by ulhorn and Karmeyer, 16 who investigated the frequency distribution in early glaucomatous visual field defects. The spatial frequency was 1 cyc/deg, and the pattern-reversal frequency was 2.5 Hz sinusoidal. For determining contrast sensitivity, CS = log([z max + mln )]/([Z max - I ml J), a two-alternative, forced-choice procedure combined with a staircasetracking procedure was used. No pupil width correction was performed because it has been shown that CS depends on age-related differences in pupil size only if spatial frequencies greater than those used in the present study are applied. 17 Temporal Contrast Sensitivity Test. 11 system with a full-field bowl (58-cm diameter) and a white flicker light was used. The test was performed under photopic conditions and required no fixation by the subject. The flicker threshold was determined at a constant frequency of 37.1 Hz at a timeaveraged luminance of 10 cd/m 2. Before each measurement, the mean luminance of the full-field bowl was corrected by taking into account the pupil diameter and the Stiles-Crawford effect, because a previous study 11 had shown that TCS has a

3 IOS, ugust 1998, ol. 39, No. 9 Sensory Model in Glaucoma Diagnosis 1569 strong dependence on pupil width. The CS was assessed using a staircase-tracking procedure. The mean value of at least six threshold crossings entered the evaluation. Both electrophysiological tests were performed using a two-channel Maxwellian-view system with a Xenon-arc lamp as the light source. The circular field size was 32 in diameter in all recordings. With this stimulus system, retinal illuminance is independent of pupil width, and, therefore, no correction of pupil width was necessary. Pattern-Reversal Electroretinogram. 12 Only one channel of the system was used to record the PERG. The stimulus was a vertical high-contrast (0.93), black-and-white, squarewave stripe pattern with a spatial frequency of the fundamental component of 0.88 cyc/deg. The pattern reversal was squarewave and occurred at a frequency of 7.8 Hz. The mean luminance was 4263 photopic Td. The responses were recorded with a carbon-glide electrode hooked over the patient's lower eye lid. fter amplification (EEP402, Hz, 50-Hz notch filter), the responses were averaged and were stored in a digital computer GBM-T; IBM, Danbury, CT; sampling rate of 500 Hz, 256-msec sweep, n = 30). Four pattern-reversal responses were analyzed within one sweep. subsequent discrete Fourier analysis evaluated the amplitude of the second harmonic component of a total of 120 pattern-reversal responses. Blue-on-Yellow Onset isual Evoked Potential. 13 One channel provided a high-contrast 0.88 cyc/deg stripe pattern with blue light (460 nm, 3-3 X 10 2 Td), and the other channel provided a homogeneous yellow adaptation light (570 nm, 1.3 X 10'' Td), which was superimposed on the stripe pattern. Stimulation was in the onset (200 msec)-offset (500 msec) mode. The recording was monopolar from the inion against the left earlobe, whereas the right earlobe was grounded. fter amplification ( Hz, 50-Hz notch), 150 sweeps (400-msec length) were averaged (500-Hz sampling rate). Peak-time measurements of the onset responses were made from the moment of pattern onset to the peak of the main negative wave (Nl). In both procedures two recordings were made to check for reproducibility. Statistical Methods Because there were no differences between subjects contributing one or two eyes (P > 0.4 for all diagnostic procedures, separately tested in patients and control subjects), the mean values of left and right eyes were used from persons contributing two eyes. To adjust for age dependency, residuals from a linear regression analysis in the control group were determined for patients and control subjects. These residuals were added to the mean value of the control group. ll comparisons between patients and control subjects were performed using a Mann-Whitney test. Correlation analyses (Pearson) between the diagnostic procedures and perimetric/morphometric measurements were performed in the patient group only. The mean defect among the patients with glaucoma was not normally distributed (Kolmogorov Smirnov test, P = 0.012). fter logarithmic transformation, normality was achieved (P = 0.9). Sensitivities were compared using a McNemars test in the patient group for a prefixed specificity of 80%. The level of significance was 0.05 (two-sided) in all statistical testing; in cases of multiple testing the Bonferroni correction was applied. Using multiple logistic regression with forward selection we constructed a diagnostic score S = a 0 + a^x % + a^x^ a M x, r In this formula x u..., x n are the results of the different diagnostic procedures. The coefficients a,,..., a n give special weights to the diagnostic procedures to yield the best discrimination between patients with glaucoma and control subjects. The coefficient a 0 only depends on the prevalence of glaucoma in the study population and has no meaning by itself. The probability of glaucoma disease is related to the score 5 by means of the formula/? = exp(5)/(l + exp(5)). The term "probability" is partially misleading because this probability depends on the prevalence in the study population.the probability P itself cannot be transferred to populations with different prevalence. However, with the exception of a 0, die coefficients #,,..., a n in the score do not depend on the prevalence. Therefore, the score itself may be used (namely, in clinical situations with lower prevalence compared with the study situation). If, for example, in the study situation the prevalence of patients with glaucoma is selected to be.prev x and in the clinical situation the prevalence is expected to be prev 2, the constant term a 0 has to be transformed into a 0 * = a 0 + In [(prev 2 1 prev^/cprev^ 1 prev 2 )]. is The leavingone-out method 19 was applied to estimate the sensitivity of the score. With that method, a subject is classified using a score based on the analysis of all the other subjects. This procedure is repeated for every subject. RESULTS fter age correction, no significant correlations with age were found in the glaucoma group (STCS, r = [P > 0.9]; TCS, r = [P > 0.9]; BY-EP, r = 0.19 [P > 0.1]; and PERG, r 0.02 [P > 0.9]). No differences between sexes were found (P > 0.1, Mann-Whitney test). There was a significant difference in visual acuity between patients (1.06 ± 0.16) and control subjects (1.11 ± 0.12; P < 0.05) as a result of the different ages in both groups. However, there was no significant correlation between visual acuity and the four diagnostic procedures (age-corrected values) and the diagnostic score (see below) in patients and in control subjects (r < 0.2, P> 0.1). The four diagnostic procedures significantly differed between patients and control subjects but not between LTG and POG; however, the LTG patients revealed slightly better results (not significant) than the POG patients (Table 2). The sensitivities achieved by fixing a specificity of 80% were comparable for STCS (11 false-negative responses), TCS (12 falsenegative responses), and BY-EP (11 false-negative responses) with approximately 84%. The PERG amplitude showed a significantly lower sensitivity (26 false-negative responses; sensitivity, 64%; P < 0.05; McNemar test after Bonferroni correction; Table 2). mong the patients with glaucoma, the four diagnostic procedures were correlated significantly with the mean defect of the visual field and the area of the neuroretinal rim but not with IOP (Table 3). The psychophysical procedures STCS and TCS showed higher correlations with the mean defect than the electrophysiological procedures. For the neuroretinal rim area, the four correlations had comparable sizes. Correlation analysis between the new diagnostic procedures (Table 4) was very different between patients with glaucoma and control subjects. mong the control subjects, there was no strong correlation of any two diagnostic variables (r < 0.2). mong the patients with glaucoma, a moderate correlation of

4 1570 Martus et al. IOS, ugust 1998, ol. 39, No. 9 TBLE 2. Diagnostic Procedures in Patients and Control Subjects STCS(%) TCS* (%) BY-EP Peak Time (msec) PERG mplitude Glaucoma (n = 73) POG (n = 53) LTG (n = 20) Control subjects (n = 122) Mann-Whitney testf Cutoff point Sensitivity (% 44.8 ± 26.5t [43.1; ]* 44.0 ± 26.8 [43.1; ] 47.1 ± 26.5 [41.7; ] 82.9 ± 16.7 [83.8; ] < ± ; ] 14.5 ± 7.3 [15.5; ] 17.6 ± 8.3 [15.5; ] 33.7 ± 13.7 [29.9; ] < ± 12.1 [128.8; ] ± 11.7 [128.9; ] ± 11.6 [125.9; ] ± 6.9 [115.9; ] < ± 0.97 [2.69; ] 2.79 ± 1.05 [2.66; ] 2.85 ± 0.72 [2.92; ] 3.87 ± 1.06 [3.78; ] < alues are means ± SD, with median and minimum-maximum in parentheses. STCS, spatiotemporal contrast sensitivity; TCS, temporal contrast sensitivity; BY-EP, blue-on-yellow visual evoked potential; PERG, pattern-evoked electroretinogram. * Raw values, logarithmic transformation for statistical testing, t Glaucoma versus controls, two-sided P, P < after Bonferroni correction. $ Fixed specificity of 80% could be found between the two psychophysical procedures, STCS and TCS, in contrast with lower correlations using the BY-EP and the PERG measurements. Multivariate nalysis The diagnostic score was constructed using a stepwise selection of variables. In the first step of the analysis, all four procedures were highly significant (P < ). The strongest separation provided STCS (P < ,)? = 85.5), followed by TCS (x 2 = 82.0). Therefore, in the first step STCS was included in the score. Using only STCS, 26 patients and control subjects were misclassified. In step 2 only, the additional information of the remaining three variables, TCS, BY-EP, and PERG, was considered. BY-EP then showed a higher chi-square value (46.36) than TCS (32.4). This means that the additional information provided by BY-EP, after the STCS test, was greater than that provided by TCS. This was because TCS and STCS were more strongly correlated in the glaucoma group (r = 0.43) than BY-EP and STCS (r = -0.07; Table 4). Therefore, the next variable to be included was BY-EP. In Figure 1, the abscissa is STCS and the ordinate is the "new" variable (BY- EP). Loosely speaking, the smaller the angle between the discrimination line and the abscissa, the greater is the additional information of the variable to be included in the step. vertical discrimination line would show that the new variable did not contribute any additional information to the score on the x-axis. In contrast, a horizontal discrimination line results if all diagnostic information is contained in the variable on the ordinate. It can be seen from Figure 1 that BY-EP indeed improved the diagnostic separation of patients with glaucoma and control subjects (22 versus 26 falsely classified subjects). Subsequently (Fig. IB), TCS was included in the score, and the separation was improved from 22 to 15 falsely classified subjects. From Figure 1C it can be seen that PERG did not contain additional diagnostic information (x 2 = 1.2, P =0.29). Therefore, PERG was not included in the score. The final score was S = STCS TCS BY-EP. The total false classification rate decreased from 13-3% (26/195 patients and control subjects) in the model with STCS only to 7.7% (15/195 patients and control subjects) in the final model with STCS, BY-EP, and TCS. fter correction with the leaving-one-out method, the false classification rate was 8.2% (16/195). There were no differences in misclassification between POG and LTG, however, the score values showed a tendency to be more pathologic in the POG group (P = 0.12, Mann-Whitney test) compared with the LTG group. To obtain a specificity of 100%, the cutpoint of classifying a person as having glaucoma has to be greater than the TBLE 3. Correlation nalysis of Diagnostic Procedures with IOP in 73 Patients STCS (%) TCS (%)* BY-EP Peak Time (msec) PERG mplitude (/u,) Intraocular pressure (mm Hg)f Neuroretinal rim area (mm 2 )j: Mean defect (db)* (P > 0.2) 0.46 (P < 0.001) (P < 0.001) (P> 0.1) 0.32 (P < 0.01) (P < 0.01) 0.13 (P > 0.3) (P < 0.01) 0.28 (P < 0.05) (/ > >0.1) 0.41 (P < 0.001) (P < 0.05) * Pearsons correlation coefficient, P value, logarithmic transformation. t Fifty patients with POG. i P < 0.004, respectively; P < 0.04 after Bonferroni correction. STCS, spatiotemporal contrast sensitivity; TCS, temporal contrast sensitivity; BY-EP, blue-on-yellow visual evoked potential; PERG, patternevoked electroretinogram; POG, primary open-angle glaucoma.

5 IOS, ugust 1998, ol. 39, No. 9 Sensory Model in Glaucoma Diagnosis 1571 TBLE 4. Correlation nalysis mong Diagnostic Procedures in Patients and Control Subjects BY-EP: Peak PERG: STCS Time mplitude (%) TCS (%) (msec) (>) STCS (%) P < 0.001* P > 0.5 P < TCS (%) P>0.3 P<0.1 P < 0.01* BY-EP: peak time (msec) P > 0.6 P > 0.1 P < 0.05 f PERG: amplitude P > 0.7 P > 0.1 P = 0.05 Right upper entries: correlation within glaucoma patients (n = 73); left lower entries: correlation within control subjects (n = 122); STCS, spatiotemporal contrast sensitivity; TCS, temporal contrast sensitivity; BY-EP, blue-on-yellow visual evoked potential; PERG, patternevoked electroretinogram. * P < 0.05 after Bonferroni correction. t P > 0.05 after Bonferroni correction. maximum score value in the control group. Under this condition, 65.8% of the patient group still were correctly classified. Effect of ge Table 5 and Figure 2 show the validity of the final diagnostic score depending on age using a poststratification split of the sample into three age groups in the table. The total classification rates varied between 88% and 99% for the different age groups. fter correction with the leaving-one-out method, the sensitivity in the group ranging in age from 51 to 62 years decreased from 89.1% (41/46) to 87.0% (40/46). DISCUSSION There are several reasons why it makes sense to condense the results of different diagnostic procedures into one quantitatitve variable. First, if one accepts the possibility of heterogeneous damage to visual function in glaucoma disease, sensitivity can be improved if procedures are combined that test different aspects of visual function (namely, temporal resolution, spatial resolution, and color vision). Second, any measurement not only reflects the "true" glaucoma damage but also some "noise" pertaining to patient concentrations (psychophysics) or artifacts in signal processing (electrophysiology). Therefore, it may even be useful to combine procedures measuring the same aspect of glaucoma damage if they are perturbed by noise of different sources. This strategy, especially, may be more successful than repeating only one identical measurement several times. One particular kind of damage to visual function may occur first in one patient, whereas different kinds of damage may occur first in other patients. Persisting parts of the "noise," characteristic to the patient under examination, cannot be averaged away by multiple measurements of the same kind. s a consequence, the selected procedures should show only moderate correlations within patients and within control TBLE 5. Sensitivity and Specificity of the Diagnostic Score by ge Groups Patients/false negative Control Subjects/ false positive Sensitivity (%) Specificity (%) Years Years Years Total in = 69) in = 49) (u = 77) in = 195) 12/0 57/ /3 34/ /5 31/ /8 122/ fter correction by the leaving-one-out method, the sensitivity in the group aged 50 to 62 years decreases from 89.1% (41/46) to 87.0% (40/46). In all other groups there are no differences. For a specificity of 100%, a sensitivity of 65.8% was achieved. subjects. Of course, at least the "true" glaucoma components have to be associated within the group of patients with glaucoma because they all reflect the progression of the disease. There is an identical problem in correlating new diagnostic measures with the gold standard. If these measures are intended to replace the gold standard, maximum correlation should be achieved. If they are intended to be used in combination with the gold standard, moderate correlations are to be preferred. One way of deriving a summary variable is in the construction of a multivariate diagnostic score. n alternative method uses neuronal networks 20 ' 21 that allow highly nonlinear associations between diagnostic measurements and the probability of disease. In our approach, this association is expressed by die logistic transformation, which, however, has been proved useful in a variety of biological applications. In our study, three of four "candidate" diagnostic procedures in a multivariate score were included. This score substantially improved the discrimination between patients with glaucoma and control subjects. Our results may correspond to (but of course do not prove) a widely discussed theory about the biological nature of glaucoma damage, the distinction between the so-called magnocellular (M) and parvocellular (P) pathways, which can be assigned different visual functions It has been suggested that in chronic experimental glaucoma in monkeys 2 ' 1 and in chronic human glaucoma 25 losses of retinal ganglion cells occur more rapidly among the large M-cells than among the smaller P-cells. Using only one diagnostic test, this theory suggests that probing M-cell function should be preferred to P-cell function. However, if one considers information of more than one diagnostic measurement in a multivariate diagnostic model, it might be sensible to include tests probing both cell functions. In the context of the reduced redundancy hypothesis, it has been argued by another investigator 26 that it might be useful to evaluate M and P functions. CS and the detection of luminance modulation might be related to the M-pathway. 22 ' 27 It has been suggested that the PERG, as recorded in the present study with a rapidly phasereversing pattern, reflects a reduced temporal discrimination in glaucomas, which in turn is assumed to be the result of damage to the M-cell pathway. 28 The BY-EP as evaluated here selectively probes die bluesensitive pathway. 29 This part of die visual system is regarded as belonging to the P-pathway. The corresponding retinal neu-

6 1572 Martus et al. IOS, ugust 1998, ol. 39, No. 9 1 I\J a. LU 130- > CO ^ \ ^*. v v \ r 7 ^ v J k Y > i Discrimination Line 73 Cases Controls Diagnostic Score: *STCS Diagnostic score containing STCS Discrimination Line B Diagnostic score containing STCS and BY-EP Diagnostic Score: 'STCS *BY-EP Discrimination Line 73 glaucoma 122 controls Diagnostic score containing STCS, BY-EP and TCS C Diagnostic Score: *STCS BY-EP 'TCS FIGURE 1. () Step 2 of the multivariate analysis. The inclusion of blue-on-yellow visual evoked potential (BY-EP) increases the discrimination between patients with glaucoma and control subjects significantly compared with spatiotemporal contrast sensitivity (STCS) alone. (B) Step 3 of the multivariate analysis. The inclusion of temporal contrast sensitivity (TCS) increases the discrimination between patients with glaucoma and control subjects significantly compared with the score of STCS and BY-EP. (C) Step 4 of the multivariate analysis. The inclusion of the pattern-reversal electroretinogram (PERG) does not improve discrimination between patients with glaucoma and control subjects after incorporation of additional knowledge from the TCS, the STCS, and BY-EP tests.

7 10S, ugust 1998, ol. 39, No. 9 Sensory Model in Glaucoma Diagnosis CO _Q o,6",5" - # k ^ igl^ i ^ _ v 9 ^P_. WWW 20 LTG 1,1-1, ,7-,3-,2-,1- o,o- ^ 53 POWG 122 Controls age [years] )) = *STCS+0.21*BY-EP-13.0*TCS FIGURE 2. Final multivariate diagnostic model for discrimination between patients and control subjects for different ages. Spatiotemporal contrast sensitivity (STCS), blue-on-yellow visual evoked potential (BY-EP), and temporal contrast sensitivity (TCS) are included. The best separation is for younger patients and control subjects. rons seem to have thick axons and large cell bodies because they have higher conduction velocities than red- and greensensitive cells. 3 "' 31 In our study, the univariate psychophysical procedures STCS and TCS, which are related to the M-cell pathway, showed the best discrimination between patients with glaucoma and control subjects. STCS showed the best separation (not significant compared with TCS and BY-EP), which confirms the results of ulhorn and Karmeyer' 6, because the localized target of STCS was motivated by their analysis of the spatial pattern of early visual field loss. However, in the multivariate analysis the (probable) P- pathway- associated BY-EP contributed the maximum additional information after including the most valid psychophysical procedure, STCS. lso, the noise perturbing TCS and STCS, namely, patient's compliance, should be highly correlated, whereas the noise of BY-EP, namely, artifacts in signal processing, should be a rather independent entity. In the third step of our model, the selection procedure for TCS was preferred to that for PERG, which is consistent with the univariate analysis. Only a larger number of patients and control subjects could prove the question of whether the PERG might contain useful additional information for TCS, BY-EP, and STCS. With 195 subjects, it is not sensible to prove or reject this hypothesis using a multivariate model. The result of our multivariate analysis was a weighted sum of different tests. The application of such a result in a clinical routine might be questionable. First, it is expensive to perform a battery of diagnostic tests in clinical routine. Second, our study only examined patients with early visualfieldloss. Third, clinicians usually prefer norm values for the different procedures and intuitively combine the information in a "normal versus pathologic result." However, the true clinical value of diagnostic procedures can only be assessed by the long-term follow-up of patients with an unknown disease status at the point of examination. The final efficacy of inventing a diagnostic measure may be evaluated only in a study that takes into account the effects of therapy and that is dependent on the results of diagnostic measurements. In our study, we showed the increase in information achieved by using more than one diagnostic method and we gave an interpretation of this fact combining medical and methodological arguments. One has to keep in mind not only the difficulty of evaluating the "absolute" gain of diagnostic procedures in a cross-sectional study, but also the fact that the standard visual fields may lead to misclassifications and, therefore, to an underestimation of the sensitivity and the specificity of a new diagnostic measure. References 1. Quigley H, ddicks EM, Green RW. Optic nerve damage in human glaucoma: III, quantitative correlation of nerve fiber loss and visual field defect in glaucoma, ischemic neuropathy, papilledema, and toxic neuropathy. rch Ophthalmol. 1982; 100: Drance SM. Correlation between optic disc changes and visual field defects in chronic open angle glaucoma. Trans m cad Ophthalmol Oto-laryngol. 1976;81: Drance SM, Schulzer M, Gordon RD, Sweeney P. Use of discriminant analysis: II, identification of persons with glaucomatous visual field defects. rch Ophthalmol. 1978;96: Drance SM, iraksinen PY, Price M, Schulzer M, Douglas GR, Tansley 13. The use of psychophysiological, structural, and electrodiagnostic parameters to identify glaucomatous damage. Graefe's rch Clin Exp Ophthalmol. 1987;225: Price MJ, Drance SM, Price M, Schulzer M, Douglas GR, Tansley B. The pattern electroretinogram and visual-evoked potential in glaucoma. Graefe's rch Clin Exp Ophthalmol. 1988;226: Caprioli J. Discrimination between normal and glaucomatous eyes. Invest Ophthalmol is Sci. 1992;33: Susanna R, Drance SM. Use of discriminant analysis: I, prediction of visual field defects from features of the glaucoma disc. rch Ophthalmol. 1978;96:

8 1574 Martus et al. IOS, ugust 1998, ol. 39, No Hart WM, Yablonski M, Kass M, Becker B. Multivariate analysis of the risk of glaucomatous visual field loss. rch Ophthalmol. 1979; 97: Drance SM, Schulzer M, Thomas B, Douglas GR. Multivariate analysis in glaucoma. Use of discriminant analysis in predicting glaucomatous visualfielddamage. rch Ophthalmol. 1981;99: Horn F, Martus P, Korth M. Comparison of temporal and spatiotemporal contrast-sensitivity tests in normal subjects and glaucoma patients. Ger J Ophthalmol. 1995;4: Horn FK, Korth M, Martus P. Quick full-field flicker test in glaucoma diagnosis: correlations with perimetry and papillometry. / Glaucoma. 1994;3: Korth M, Horn F, Storck B, Jonas J. The pattern-evoked electroretinogram (PERG): age-related alterations and changes in glaucoma. Graefe'srch Clin Exp Ophthalmol. 1989;227: Korth M, Nguyen NX, Jiinemann, Martus P, Jonas JB. EP test of the blue-sensitive pathway in glaucoma. Invest Ophthalmol is Sci. 1994;35: l Jonas JB, Gusek GC, Naumann GOH. Optic disc morphometry in chronic primary open-angle glaucoma: I, morphometric intrapapillary characteristics. Graefe's rch Clin Exp Ophthalmol. 1988; 226: Bebie H, Flammer J, Bebie T. The cumulative defect curve: separation of local and diffuse components of visual field damage. Graefe'srch Clin Exp Ophthalmol. 1989;227: ulhorn E, Karmeyer H. Frequency distribution in early glaucomatous visual field defects. Doc Ophthalmol Proc Series : Owsley C, Sekuler R, Siemsen D. Contrast sensitivity throughout adulthood. ision Res. 23; 1983: Hosmer DW, Lemeshow S. pplied Logistic Regression. New York: Wiley; 1989: Hand DJ. Discrimination and Classification. New York: Wiley; 1981: Goldbaum MH, Sample P, White H, et al. Interpretation of automated perimetry for glaucoma by neural network. Invest Ophthalmol is Sci. 1994;35: Brigatti L, Hoffman D, Caprioli J. Neural networks to identify glaucoma with structural and functional measurements. m J Ophthalmol. 1996;121: Kaplan E, Shapley RM. The primate retina contains two types of ganglion cells, with high and low contrast sensitivity. Proc Natl cad Sci US. 1986;83: Kaplan E, Lee BB, Shapley RM. New views of primate retinal function. In: Osborne N, Chader J, eds. Progress in Retinal Research. Oxford: Pergamon; 1990: Quigley H, Sanchez RM, Dunkelberger GR, L'Hernault NL, Baginski T. Chronic glaucoma selectively damages large optic nerve fibers. Invest Ophthalmol is Sci. 1987;28: Quigley H, Dunkelberger GR, Green WR. Chronic human glaucoma causing selectively greater loss of large optic nerve fibers. Ophthalmology. 1988;95: Johnson C. Selective versus nonselective losses in glaucoma. / Glaucoma (Suppl). 1994;3:S32-S Lee BB, Pokorny J, Smith C, Martin PR, alberg. Luminance and chromatic modulation sensitivity of macaque ganglion cells and human observers./ Opt Soc m []. 1990;7: Trick GL. Retinal potentials in patients with primary open-angle glaucoma: physiological evidence for temporal frequency tuning deficits. Invest Ophthalmol is Sci. 1985;26: Korth M, Nguyen NX, Rix R, Sembritzki O. Interactions of spectral, spatial, and temporal mechanisms in the human pattern visual evoked potential. ision Res. 1993:33: Malpeli JG, Schiller P. Lack of blue OFF-center cells in the visual system of the monkey. Brain Res. 1978;l4l: De Monasterio FM. symmetry of on- and off-pathways of bluesensitive cones of the retina of macaques. Brain Res. 1979;l66: