**Purpose.**:
We compared the detection of visual field progression and its rate of change between standard automated perimetry (SAP) and Matrix frequency doubling technology perimetry (FDTP) in glaucoma.

**Methods.**:
We followed prospectively 217 eyes (179 glaucoma and 38 normal eyes) for SAP and FDTP testing at 4-month intervals for ≥36 months. Pointwise linear regression analysis was performed. A test location was considered progressing when the rate of change of visual sensitivity was ≤−1 dB/y for nonedge and ≤−2 dB/y for edge locations. Three criteria were used to define progression in an eye: ≥3 adjacent nonedge test locations (conservative), any three locations (moderate), and any two locations (liberal) progressed. The rate of change of visual sensitivity was calculated with linear mixed models.

**Results.**:
Of the 217 eyes, 6.1% and 3.9% progressed with the conservative criteria, 14.5% and 5.6% of eyes progressed with the moderate criteria, and 20.1% and 11.7% of eyes progressed with the liberal criteria by FDTP and SAP, respectively. Taking all test locations into consideration (total, 54 × 179 locations), FDTP detected more progressing locations (176) than SAP (103, *P* < 0.001). The rate of change of visual field mean deviation (MD) was significantly faster for FDTP (all with *P* < 0.001). No eyes showed progression in the normal group using the conservative and the moderate criteria.

**Conclusions.**:
With a faster rate of change of visual sensitivity, FDTP detected more progressing eyes than SAP at a comparable level of specificity. Frequency doubling technology perimetry can provide a useful alternative to monitor glaucoma progression.

^{ 1 }50 years ago. When achromatic sinusoidal gratings of low spatial frequency flicker at a high temporal frequency, the spatial frequency of the grating appears to be doubled. This phenomenon was suggested to be related to the appearance of a second-harmonic distortion that may involve rectification and response compression.

^{ 2,3 }Despite the findings that subjects may, in fact, report a higher or lower spatial frequency than its true frequency, and that the frequency-doubling effect may not be generated by the retinal ganglion cells (RGCs) of the magnocellular pathway, but by many types of RGCs,

^{ 4–7 }the first and second generations of the frequency doubling technology perimetry (FDTP) have been shown to be useful to detect early visual field damage in glaucoma.

^{ 8 }However, less is known about the performance of FDTP in following visual field progression. The rate of change of visual sensitivity has not been established in the Matrix FDTP and the role of FDTP in monitoring glaucoma progression remains unclear.

^{2}. The FDTP was performed with the Matrix FDT perimeter (Carl Zeiss Meditec, Inc.), and 0.5-cycle/degree sinusoidal gratings undergoing counter phase flickering at 18 Hz were displayed sequentially over 54 5° × 5° squares using the 24-2 zippy estimation by sequential testing (ZEST) thresholding algorithm. For SAP, the patients' appropriate near refraction was corrected with a trial lens inserted in the lens holder. For Matrix FDTP, the patients wore their own prescription glasses as recommended by the manufacturer. Only reliable visual fields (fixation losses, and false-negative and false-positive errors ≤20% for SAP and FDTP) were analyzed (Table 1 shows the reliability indices of Matrix FDTP and SAP). A visual field defect was defined as a cluster of three nonedge points, all of which were indicated at a probability level <5% with at least one <1% in the pattern deviation plot, and confirmed with at least two consecutive examinations by the same types of perimetry. The same definition of visual field defect was applied for SAP and FDTP. In the glaucoma group (179 eyes of 148 patients), only eyes with visual field defects detected in both perimetric tests at the time of enrollment were included. In the normal group (38 eyes of 28 subjects, Matrix FDTP was performed in both eyes in 10 subjects and in one randomly selected eye in each of the remaining 18 subjects), eyes had no visual field defects in SAP and FDTP throughout the follow-up period.

**Table 1**

**Table 1**

Mean, % | SD, % | Median, % | Minimum, % | Maximum, % | P Value* | |

Glaucoma group | ||||||

Matrix FDTP | ||||||

Fixation loss | 4.0 | 6.3 | 0.0 | 0.0 | 20.0 | 0.001 |

False-negative | 2.8 | 5.5 | 0.0 | 0.0 | 20.0 | 0.757 |

False-positive | 2.6 | 6.1 | 0.0 | 0.0 | 16.7 | 0.193 |

SAP | ||||||

Fixation loss | 4.8 | 5.6 | 5.3 | 0.0 | 20.0 | |

False-negative | 2.7 | 4.2 | 0.0 | 0.0 | 20.0 | |

False-positive | 2.2 | 2.8 | 1.0 | 0.0 | 17.0 | |

Normal group | ||||||

Matrix FDTP | ||||||

Fixation loss | 4.3 | 6.5 | 0.0 | 0.0 | 20.0 | 0.337 |

False-negative | 1.9 | 4.6 | 0.0 | 0.0 | 20.0 | <0.001 |

False-positive | 0.3 | 2.2 | 0.0 | 0.0 | 16.7 | <0.001 |

SAP | ||||||

Fixation loss | 4.9 | 5.5 | 6.3 | 0.0 | 18.8 | |

False-negative | 0.7 | 1.7 | 0.0 | 0.0 | 12.0 | |

False-positive | 2.4 | 3.0 | 1.0 | 0.0 | 15.0 |

^{ 9–12 }Because two of the 16 possible combinations of the four yes/no responses gave a similar mean, the threshold estimates assumed 15 discrete levels of threshold measurement. Figures 1A and 1B show the distribution of the threshold values of individual test locations obtained from Matrix FDTP and SAP in the glaucoma group. While the range of visual sensitivity was comparable (0–41 dB for SAP, 0–38 dB for Matrix FDPT), the step sizes were different between the two perimetric tests. For Matrix FDTP, the thresholds were found at 15 discrete levels at 0, 2, 3, 6, 7, 11, 12, 13, 18, 20, 23, 27, 32, 34, and 38 dB, whereas for SAP, the step size was fixed at 1 dB. To facilitate a more direct comparison between SAP and Matrix FDTP, all the SAP threshold values were transformed to the nearest threshold values fitting into the 15 thresholds as in Matrix FDTP (i.e., the SAP thresholds were rebinned into the 15 threshold values, Fig. 1C). Progression analyses were performed with and without such transformation (designated as SAP [transformed] and SAP, respectively).

**Figure 1**

**Figure 1**

^{ 13–15 }In this study, progression at a test location was identified when the rate of change of visual sensitivity was ≤−1 dB/y for nonedge locations, and ≤−2 dB/y for edge locations, with

*P*< 0.05. Three levels of criteria were used to define progression: (1) at least three adjacent nonedge (except for the two nasal edge points) test locations detected with progression, (2) any three locations detected with progression, and (3) any two locations detected with progression. The same criteria were applied for SAP and Matrix FDTP. An improvement at a particular test location was defined when the rate of change of visual sensitivity was ≥+1 dB/y for nonedge locations, and ≥+2 dB/y for edge locations, with

*P*< 0.05.

^{2}test was used to compare the proportion of progression and improvement events between the perimetric tests. The agreement of progression detection between SAP and FDTP was calculated with

*κ*statistics. A value between 0.0 and 0.2 indicates slight agreement, 0.21 and 0.40 is fair, 0.41 and 0.60 is moderate, 0.61 and 0.80 is substantial, and 0.81 and 1 is almost perfect agreement.

^{ 16 }Longitudinal MD, superior and inferior hemifield visual sensitivity measurements were fitted with linear mixed models with follow-up duration as the fixed effect. Random intercepts and coefficients were included at the subject and eye levels (each eye nested within subject) for the effect of follow-up duration (random effects). Comparisons of the rates of change of MD and visual sensitivities between FDTP and SAP were evaluated with likelihood ratio test between the goodness of fit of the linear mixed models. The sample size required for the linear mixed models was calculated based on the estimates of random intercept variance, random slope variance, residual variance, the covariance between random slope and random intercept, and the working correlation matrix derived from the linear mixed models, and assuming that all longitudinal measurements were collected at regular time intervals (every 4 months for 36 months).

^{ 17 }Sample size calculation revealed that at least 78 eyes would be required to detect a rate of change of MD ≤ −0.5 dB/y with a statistical power of 80%. A post hoc power calculation comparing the differences in proportions of progressing locations between SAP/SAP (transformed) and Matrix FDT perimetry indicated that the current sample size attained a power of ≥84.3% at an alpha of 0.05. To compute the Kaplan-Meier survival curves for Matrix FDTP and SAP, progression analysis was performed at each follow-up visit from the third follow-up visit onwards. The survival probabilities of Matrix FDTP and SAP/SAP (transformed) were compared using log-rank tests.

*P*< 0.05 was considered statistically significant in all analyses.

**Table 2**

**Table 2**

Demographics | ||

Glaucoma | Normal | |

Total n of eyes | 179 | 38 |

Age, y ± SD | 50.9 ± 14.2 | 60.2 ± 8.0 |

Duration of follow-up, mo ± SD | 46.9 ± 8.9 | 49.1 ± 4.0 |

Baseline Matrix FDTP MD, dB ± SD | −10.53 ± 6.57 | −1.34 ± 2.55 |

Baseline Matrix FDTP PSD, dB ± SD | 6.49 ± 2.14 | 2.84 ± 0.51 |

Baseline SAP MD, dB ± SD | −9.20 ± 8.03 | −0.69 ± 1.49 |

Baseline SAP PSD, dB ± SD | 7.07 ± 4.45 | 1.67 ± 0.88 |

*P*≤ 0.042). Likewise, the survival probability of Matrix FDTP for detection of progression was significantly worse compared to SAP (

*P*≤ 0.042) and SAP (transformed) (

*P*≤ 0.005) using the liberal and the moderate criteria (Fig. 3). The agreement of progression detection between Matrix FDTP and SAP, and between Matrix FDTP and SAP (transformed) was poor to fair (

*κ*ranged between 0.066 and 0.403) independent of the criteria of progression. In the normal group, Matrix FDTP detected two progressing eyes using the liberal criteria. No eyes showed progression in Matrix FDTP/SAP using the conservative and the moderate criteria.

**Figure 2**

**Figure 2**

**Figure 3**

**Figure 3**

*P*< 0.05), Matrix FDTP detected a total of 176 progressing locations in the glaucoma group, which was significantly greater than SAP (103,

*P*< 0.001), and SAP (transformed) (124,

*P*= 0.003, Fig. 4). There were no significant differences in the number of significant improvement (≥1 dB/y for nonedge locations and ≥2 dB/y for edge locations) between Matrix FDTP (

*n*= 5) and SAP (

*n*= 10,

*P*= 0.804), and between Matrix FDTP and SAP (transformed) (

*n*= 14,

*P*= 0.848, Fig. 5). This frequency distribution of the rates of change of visual sensitivity of individual locations of Matrix FDTP, SAP, and SAP (transformed) are shown in Figure 6. Figure 7 shows the spatial distribution of the progressing locations detected by Matrix FDTP and SAP for individual eyes. The spatial agreement of progression detection between the two perimetric tests was poor.

**Figure 4**

**Figure 4**

**Figure 5**

**Figure 5**

**Figure 6**

**Figure 6**

**Figure 7**

**Figure 7**

*P*< 0.001). With respect to the superior hemifield, the mean rate of change of visual sensitivity was faster for Matrix FDTP (−0.31 dB/y; 95% CI, −0.45 to −0.18 dB/y) than SAP (−0.17 dB/y; 95% CI, −0.25 to −0.09 dB/y;

*P*< 0.001; Table 4). Likewise, in the inferior hemifield, Matrix FDTP also detected visual sensitivity reduction at a significantly faster rate (−0.30 dB/y; 95% CI, −0.43 to −0.17 dB/y) than SAP (−0.24 dB/y; 95% CI, −0.31 to −0.16 dB/y;

*P*< 0.001; Table 5). While the rate of change of MD for Matrix FDTP was insignificant (

*P*= 0.846) in the normal group (Table 6), a positive trend was observed for SAP (0.104 dB/y,

*P*= 0.005).

**Table 3**

**Table 3**

Fixed Effect | P Value | Random Effect | |||||

Parameter | Coefficient | 95% CI | Level | SD | |||

Matrix FDTP MD | Duration, y | −0.260 | −0.364 | −0.156 | <0.001 | Subject | 0.458 |

Intercept | −10.278 | −11.327 | −9.229 | <0.001 | 4.080 | ||

Duration, y | Eye | 0.114 | |||||

Intercept | 4.883 | ||||||

Residual | 1.525 | ||||||

SAP MD | Duration, y | −0.202 | −0.282 | −0.122 | <0.001 | Subject | 0.331 |

Intercept | −9.340 | −10.491 | −8.189 | <0.001 | 0.000 | ||

Duration, y | Eye | 0.212 | |||||

Intercept | 7.828 | ||||||

Residual | 1.137 |

**Table 4**

**Table 4**

Fixed Effect | P Value | Random Effect | |||||

Parameter | Coefficient | 95% CI | Level | SD | |||

Matrix FDTP | Duration, y | −0.313 | −0.450 | −0.177 | <0.001 | Subject | 0.503 |

Intercept | 25.163 | 24.397 | 25.929 | <0.001 | 2.109 | ||

Duration, y | Eye | 0.077 | |||||

Intercept | 4.215 | ||||||

Residual | 2.630 | ||||||

SAP | Duration, y | −0.167 | −0.245 | −0.089 | <0.001 | Subject | 0.197 |

Intercept | 28.256 | 27.499 | 29.013 | <0.001 | 3.598 | ||

Duration, y | Eye | 0.112 | |||||

Intercept | 2.342 | ||||||

Residual | 1.804 |

**Table 5**

**Table 5**

Fixed Effect | P Value | Random Effect | |||||

Parameter | Coefficient | 95% CI | Level | SD | |||

Matrix FDTP | Duration, y | −0.295 | −0.425 | −0.165 | <0.001 | Subject | 0.399 |

Intercept | 25.916 | 25.226 | 26.605 | <0.001 | 2.526 | ||

Duration, y | Eye | 0.367 | |||||

Intercept | 3.138 | ||||||

Residual | 2.433 | ||||||

SAP | Duration, y | −0.236 | −0.311 | −0.161 | <0.001 | Subject | 0.127 |

Intercept | 29.886 | 29.475 | 30.297 | <0.001 | 1.510 | ||

Duration, y | Eye | 0.343 | |||||

Intercept | 1.899 | ||||||

Residual | 1.316 |

**Table 6**

**Table 6**

Fixed Effect | P Value | Random Effect | |||||

Parameter | Coefficient | 95% CI | Level | SD | |||

Matrix FDTP | Duration, y | −0.016 | −0.174 | 0.143 | 0.846 | Subject | 0.179 |

Intercept | −1.057 | −1.890 | −0.223 | 0.013 | 1.897 | ||

Duration, y | Eye | 0.139 | |||||

Intercept | 0.936 | ||||||

Residual | 1.855 | ||||||

SAP MD | Duration, y | 0.104 | 0.031 | 0.177 | 0.005 | Subject | 0.124 |

Intercept | −0.788 | −1.065 | −0.512 | 0.000 | 0.540 | ||

Duration, y | Eye | 0.035 | |||||

Intercept | 0.425 | ||||||

Residual | 0.747 |

*P*< 0.001) and SAP (transformed) (124,

*P*= 0.003), with a similar number of locations showing improvement (Matrix FDTP, 5; SAP, 10; SAP (transformed), 14;

*P*≥ 0.804). With a faster rate of reduction in visual sensitivity (Tables 3 1552–5), Matrix FDTP may detect visual field changes earlier than SAP and provide an effective alternative to follow visual field progression in glaucoma management.

^{ 18,19 }the role of FDTP in following glaucoma progression remains elusive. Prospective studies evaluating the performance of the FDTP for progression detection are sparse. To our knowledge, only two longitudinal studies have investigated glaucoma progression using Matrix FDTP. Xin et al.

^{ 20 }followed 33 glaucoma patients for 21.1 ± 1.8 months (range, 18–26 months) and analyzed progression by event analysis (defined as a change in MD greater than the test–retest variability). They showed that SAP and Matrix FDTP detected 8 (14.5%) and 13 (23.6%) progressing eyes, respectively (the number of eyes with improvement was not reported).

^{ 20 }In our previous study following 76 glaucoma suspects with normal visual fields at the baseline examination for at least 30 months, we demonstrated that the rate of change of PSD was faster for Matrix FDTP compared to SAP and that Matrix FDTP could detect development of visual field defects (defined as appearance of a cluster of 3 nonedge points, all of which were indicated at a probability level ≤5% with at least one ≤1% in the pattern deviation plot, and confirmed by at least two consecutive examinations) earlier than SAP.

^{ 21 }These studies, however, did not examine visual field progression with trend analysis. The criteria of detecting the development of new visual field defects in glaucoma suspects as used in our previous study would be inadequate to detect change in patients with moderate and advanced visual field loss. To our knowledge, this is the first study comparing Matrix FDTP and SAP with trend analysis in glaucoma patients.

*P*< 0.001, Tables 3 1552–5), it is reasonable to infer that Matrix FDTP may detect visual field progression earlier than SAP. However, lacking a reference standard, it is not feasible to compare directly the sensitivity between the perimetric tests. We did not take reference from structural measures as it has been shown that detectable change in the optic disc often precedes visual field progression, and that progression in structure and function may not occur concomitantly within the study period.

^{ 22,23 }

^{ 24 }the strategies for progression detection have not been optimized for Matrix FDTP. Unlike SAP, no commercially available packages are available for progression analysis in FDTP. The test–retest repeatabilities of SAP and Matrix FDTP are different

^{ 25 }and a different set of criteria (including the rate of change of visual sensitivity threshold and the

*P*value) may be necessary to define progression in Matrix FDTP. However, without a validated algorithm for detection of progression in Matrix FDTP, standardizing the criteria for progression analysis would deem appropriate to obtain a meaningful comparison between the perimetric tests. Adopting a

*P*value of 0.05 to define a significant rate of change of visual sensitivity for an individual location, we observed 5, 10, and 14 locations showing a significant positive trend for Matrix FDTP, SAP, and SAP (transformed), respectively, in the glaucoma group. With a total of 9666 (54 × 179) test locations examined, the respective proportions were 0.05%, 0.10%, and 0.14%. In the normal group, no eyes showed progression using the conservative and moderate criteria. These findings suggested that adopting a

*P*value of 0.05 already provided a high specificity for change analysis. Using a smaller

*P*value to define progression would increase the specificity at the expense of a reduced sensitivity, which may render the comparison between SAP and Matrix FDT perimetry less practical.

**S. Liu**, None;

**M. Yu**, None;

**R.N. Weinreb**, Carl Zeiss Meditec (F, C), Heidelberg Engineering (F), Optovue (F), Nidek (F), Topcon (F, C);

**G. Lai**, None;

**D.S.-C. Lam**, None;

**C.K.-S. Leung**, Carl Zeiss Meditec (F, R), Optovue (F)

*. 1981; 71: 1051–1055. [CrossRef] [PubMed]*

*J Opt Soc Am**. 1974; 64: 1251–1255. [CrossRef] [PubMed]*

*J Opt Soc Am**. 2003; 44: 1111–1116. [CrossRef] [PubMed]*

*Invest Ophthalmol Vis Sci**. 2002; 43: 3590–3599. [PubMed]*

*Invest Ophthalmol Vis Sci**. 2008; 85: 318–329. [CrossRef] [PubMed]*

*Optom Vis Sci**. 2011; 52: 764–771. [CrossRef] [PubMed]*

*Invest Ophthalmol Vis Sci**. 2011; 118: 986–1002. [CrossRef] [PubMed]*

*Ophthalmology**. 2002; 43: 322–331. [PubMed]*

*Invest Ophthalmol Vis Sci**. 1999; 76: 588–595. [CrossRef] [PubMed]*

*Optom Vis Sci**. 2002; 43: 709–715. [PubMed]*

*Invest Ophthalmol Vis Sci**. 1994; 34: 885–912. [CrossRef] [PubMed]*

*Vision Res**. 1996; 80: 40–48. [CrossRef] [PubMed]*

*Br J Ophthalmol**. 1996; 234: 411–418. [CrossRef] [PubMed]*

*Graefes Arch Clin Exp Ophthalmol**. 1997; 81: 1037–1042. [CrossRef] [PubMed]*

*Br J Ophthalmol**. 1977; 33: 159–174. [CrossRef] [PubMed]*

*Biometrics**. New York, NY: Oxford University Press; 2002: 22–32.*

*Analysis of Longitudinal Data*. 2nd ed*. 2004; 137: 863–871. [CrossRef] [PubMed]*

*Am J Ophthalmol**. 2010; 117: 1530–1537. [CrossRef] [PubMed]*

*Ophthalmology**. 2011; 52: 519–526. [CrossRef] [PubMed]*

*Invest Ophthalmol Vis Sci**. 2014; 132: 77–83. [CrossRef] [PubMed]*

*JAMA Ophthalmology**. 2009; 116: 2110–2118. [CrossRef] [PubMed]*

*Ophthalmology**. 2009; 127: 1250–1256. [CrossRef] [PubMed]*

*Arch Ophthalmol**. 2012; 3: 135–139. [CrossRef]*

*Asia Pac J Ophthalmol**. 2005; 46: 2451–2457. [CrossRef] [PubMed]*

*Invest Ophthalmol Vis Sci*