**Purpose.**:
To develop a model for the Cirrus HD-OCT that allows for the comparison of retinal nerve fiber layer (RNFL) thickness measurements with dissimilar signal strengths (SS) and accounts for test–retest variability.

**Methods.**:
Retinal nerve fiber layers were obtained in normals using the Cirrus optic disc cube 200 × 200 protocol during a single encounter. Five RNFL scans were obtained with a SS of 9 or 10. Diffusion lens filters were used to degrade SS to obtain five scans at each SS group of 7 or 8, 5 or 6, and 3 or 4. The relationship between average RNFL thickness and SS was established, and an equation was developed to allow for adjustment of an RNFL measurement had it been a SS of 7. Intravisit interclass correlation coefficient (ICC) and coefficient of variation (CV) parameter estimates for each SS group were calculated. Repeatability and upper tolerance limit were calculated as 1.96 × √2 × within-subject standard deviation (S_{w}) and 1.645 × √2 × S_{w}, respectively.

**Results.**:
There was a linear relationship between average RNFL and SS. RNFL_{adj} = RNFL − 1.03*SS + 7.21 allows for the adjustment of RNFL readings to the same SS. Interclass correlation coefficients and CVs were good for all measurements down to SS of 3 or 4. Repeatability and upper tolerance limit were 5.24 and 4.40 μm, respectively.

**Conclusions.**:
Our model adjusts RNFL readings based on SS and includes an upper tolerance limit of 5 μm. If validated, this model could improve the detection of real RNFL changes. Further study to validate this model should be performed before widespread use is adopted.

^{1,2}The reproducibility of measurements from the Cirrus OCT has been shown to be 4 to 5 μm for average RNFL thickness measurements.

^{1,2}

^{3}Numerous studies have documented that RNFL measurements are positively correlated with SS.

^{3–9}Most of these studies have been performed using time-domain technology and have shown that: RNFL measurement variability increases with decreasing SS, when SS is greater than 7, RNFL thickness is reproducible, and, if SS is less than 4, RNFL measurements may be considered unreliable. To better characterize this relationship, researchers have artificially changed the SS in patients by defocusing the image,

^{9}by placing contact lenses during scans,

^{10}and through introduction of optical filters in between the subject and the OCT machine.

^{9,11,12}Simple formulas correcting for SS changes in average RNFL measurements have, at this time, shown limited predictive capability, in part due to not accounting for test–retest variability and being applied in unstable testing conditions.

^{11}Although not specifically looking at RNFL values, more complex formulas analyzing reflectivity peaks in the macula have been developed to correct for changes due to SS

^{12}and show potential, but as of now are too complicated and are not commercially available.

_{w}),

^{2}where S

_{w}is the square root of the mean of within-subject variance. The repeatability equation implies that a true change could occur in either direction. For glaucoma patients, the only real change that could occur is deterioration. Thus a one-tailed statistical test is more appropriate and is defined as the intravisit upper tolerance limit and is 1.645 × √2 × S

_{w}.

^{1,13}

*P*values: degree 1,

*P*< 0.001; degree 2,

*P*= 0.620; degree 3,

*P*= 0.527; logarithmic,

*P*= 0.840; and exponential,

*P*= 0.295. We concluded that only a linear relationship was significant and strong enough to be useful. We are left with an equation of the form

**Figure 1**

**Figure 1**

*a*and

*b*. We noted the sources of variability: For any reading,

*a*and

*b*vary; for initial and corrected readings (say,

*t*= 1 and 2) we have

*a*

_{1},

*b*

_{1}, and

*a*

_{2},

*b*

_{2}, each with its own variability. However, if SS changes, there is a change in RNFL, another source of variability, and the value we have set out to estimate. We found that

*b*is similar throughout different SS values. As evidence, we note the following. The distribution of SS ranged from 3 to 10 in our data. The distribution of slopes was normal (Shapiro-Wilk normality test gives

*P*= 0.144 and the Skewness-Kurtosis test's skewness component

*P*= 0.098). Using the mean and its confidence interval for descriptive statistics, we found that the mean slope, that is, the value of

*b*, was 1.03 with 95% confidence interval of 1.01 to 1.05, clinically negligible. Thus, the change in RNFL from reading 1 to reading 2 occurs primarily in

*a*. Therefore, we base our correction on a revision of equation 1 as

_{1}from SS

_{1}; and, for the second, we read RNFL

_{2}from SS

_{2}. SS

_{2}is different from SS

_{1}, so we do not know how much of the difference between RNFL

_{2}and RNFL

_{1}is due to a change in the patient's RNFL and how much to the change in SS. If we find and adjust for what RNFL

_{2}would have been if that reading had arisen from SS

_{1}, then the difference would be due solely to a change in the patient's RNFL. Substituting RNFL

_{2}and SS

_{2}in Equation 2 and solving for

*a*

_{2}, we find

*a*

_{2}= RNFL

_{2}− 1.03 SS

_{2}. Substituting that in the equation for the first reading, we find that the adjusted RNFL (RNFL

_{adj}) is

^{10}For this reason, we chose a SS of 7 as our adjustment SS value, which would allow some measurements to be adjusted positively and some negatively (in theory, providers could adjust it to whatever SS they desired, so long as they were consistent). When 7 is substituted for SS

_{1}and the subscript 2 is removed to generalize the equation, we find that the adjustment equation reduces to

**Figure 2**

**Figure 2**

**Figure 3**

**Figure 3**

_{w}being 5.23 and 5.29 μm, respectively, compared to 2.73 μm for average RNFL and 2.52 and 4.27 μm for the temporal and nasal quadrants, respectively. For average RNFL and each of the quadrants, the variability did not vary substantially as SS decreased. In fact, S

_{w}was lower for SS group D compared to SS group C for all quadrants and group B for three quadrants. The intravisit repeatability and tolerance limit for all measurements were 7.57 and 6.35 μm, respectively (Table 2). After all the RNFL values were adjusted using Equation 4, repeatability and tolerance limit dropped to 5.24 and 4.40 μm, respectively.

**Table 1**

**Table 1**

Parameter | SS Group | Mean, μm | ICC, % | ICC CIs | CV, % | S_{w}, μm |

Average RNFL | All | 93.07 | 88.6 | 84.6–92.6 | 2.93 | 2.73 |

A | 95.57 | 97.2 | 96.1–98.4 | 1.40 | 1.34 | |

B | 94.17 | 94.8 | 92.6–96.9 | 1.94 | 1.83 | |

C | 91.18 | 93.5 | 90.9–96.1 | 2.11 | 1.92 | |

D | 90.25 | 95.3 | 92.8–97.7 | 1.88 | 1.70 | |

Superior | All | 117.72 | 86.6 | 82.0–91.2 | 4.44 | 5.23 |

A | 121.31 | 90.2 | 86.2–94.1 | 3.55 | 4.31 | |

B | 118.97 | 89.9 | 86.0–93.9 | 3.64 | 4.33 | |

C | 115.21 | 89.1 | 85.0–93.3 | 5.20 | 4.63 | |

D | 114.01 | 93.4 | 90.0–96.8 | 4.33 | 4.04 | |

Temporal | All | 63.28 | 90.3 | 86.8–93.8 | 3.98 | 2.52 |

A | 63.93 | 89.7 | 85.5–93.8 | 3.75 | 2.40 | |

B | 63.80 | 92.8 | 89.8–95.7 | 3.50 | 2.23 | |

C | 62.60 | 92.2 | 89.2–95.3 | 3.61 | 2.26 | |

D | 62.44 | 93.2 | 89.7–96.7 | 3.56 | 2.22 | |

Inferior | All | 122.72 | 84.4 | 79.2–89.7 | 4.31 | 5.29 |

A | 124.84 | 90.2 | 86.2–94.1 | 3.41 | 4.26 | |

B | 124.13 | 91.8 | 88.5–95.1 | 3.13 | 3.88 | |

C | 120.55 | 82.6 | 76.4–88.9 | 4.42 | 5.33 | |

D | 120.39 | 88.3 | 82.6–94.0 | 3.53 | 4.25 | |

Nasal | All | 68.53 | 83.6 | 78.2–89.1 | 6.23 | 4.27 |

A | 72.11 | 91.9 | 88.6–95.2 | 4.41 | 3.18 | |

B | 69.73 | 88.9 | 84.6–93.3 | 4.99 | 3.48 | |

C | 66.35 | 90.3 | 86.5–94.1 | 4.37 | 2.90 | |

D | 64.26 | 95.1 | 92.5–97.6 | 3.41 | 2.19 |

**Table 2**

**Table 2**

SS Group | Mean Avg RNFL, μm | ICC, % | ICC 95% CIs, % | CV, % | S_{w}, μm | Repeatability, μm | Tolerance Limit, μm |

All-Adj | 93.2 | 94.1 | 92.0–96.3 | 2.03 | 1.89 | 5.24 | 4.40 |

All | 93.1 | 88.6 | 84.2–92.6 | 2.93 | 2.73 | 7.57 | 6.35 |

A | 95.6 | 97.2 | 96.1–98.4 | 1.40 | 1.34 | 3.71 | 3.12 |

B | 94.2 | 94.8 | 92.6–96.9 | 1.94 | 1.83 | 5.07 | 4.26 |

C | 91.2 | 93.5 | 90.9–96.1 | 2.11 | 1.92 | 5.32 | 4.47 |

D | 90.2 | 95.3 | 92.8–97.7 | 1.88 | 1.70 | 4.71 | 3.95 |

_{adj}, RNFL − 1.03 SS + 7.21 (

*n*= 1383 measurements). All, all subjects at all signal strengths (

*n*= 1383); SS group A, SS 9 or 10 (

*n*= 316); B, SS 7 or 8 (

*n*= 470); C, SS 5 or 6 (

*n*= 414); and D, SS 3 or 4 (

*n*= 183); repeatability = 1.96 × √2 × S

_{w}, tolerance limit = 1.645 × √2 × S

_{w}.

- Apply the formula to RNFL OCT scans that have a SS greater than 3. To determine the adjusted RNFL value, one can use the formula directly or make an approximation by counting how many SS units away from 7 an OCT scan is and then either add, if less than 7, or subtract, if greater than 7, those units to or from the original average RNFL value. This will adjust the average RNFL thickness to what it would have been had the SS been 7. For example, if the original average RNFL thickness was 100 and the SS was 10, the adjusted RNFL value would be 97; and if the original average RNFL thickness was 90 and the SS was 5, the adjusted RNFL thickness would be 92.
- For the model to work, it should be applied to all scans with SS from 3 to 10. This allows the practitioner to establish a more level and consistent baseline (Fig. 2).
- A decrease in RNFL thickness greater than 5 μm (tolerance limit rounded up from 4.40) below baseline would suggest (with 95% confidence) pathologic loss of RNFL.

^{9}artificially changed the SS of subjects by defocusing the image using the Stratus OCT and investigated how SS affected RNFL measurements. They showed that the superior, nasal, and average RNFL measurements were positively correlated to changes in SS. They did not, however, quantify the correlation or provide clinically useful recommendations as how to deal with interpreting differences in RNFL when SS is different. Observations by Lee et al.

^{8}began to quantify this relationship. These investigators performed two OCT scans on participants and looked not only at repeatability of the measurement but also at whether SS affected the agreement of the two measurements. They found that if SS was the same, the average RNFL measurement was 2 μm different; it was 3.2 μm different if the difference in SS between the first and second measurement was 2. They did not differentiate between pairs of scans at different SS, making it difficult to interpret the results. Since their study used the time-domain Stratus OCT, their results are not directly comparable to ours. This same research team performed a similar study using the Cirrus OCT and showed that if there was no change in SS, the average RNFL did not change; if SS increased by 1 unit, then the average RNFL increased by 0.9 μm; and if SS increased by 2 units, then average RNFL increased by 1.7 μm.

^{3}As in their previous study using the Stratus, they did not differentiate between pairs of scans at different SS. Even so, their results are comparable to ours, which showed a change between SS and average RNFL thickness of approximately 1 μm RNFL to 1 SS unit. An almost identical relationship was also observed by Kok et al.

^{11}They measured average RNFL thickness using progressively denser reflective attenuation filters to artificially decrease SS in a group of four normal patients. They found that for every 0.1 change in optical density of the filter, there was a 1.07-μm change in average RNFL thickness (where a 0.1 change in optical density equals a 1.0 unit change in SS).

^{11}applied their derived model to a group of patients before and after cataract surgery to predict average RNFL changes due to changes in SS, their prediction was off by 3.71 ± 2.97 μm. They concluded that the predictive nature of their formula was “limited.” That conclusion may be premature for a number of reasons; and the relationship between RNFL and SS that they found, and that we confirmed, may still have predictive value. First, as they suggest in their discussion, RNFL thickness is affected by cataract surgery, likely via inflammatory mediators causing increased retinal vascular permeability and leakage. It has been shown that the average RNFL thickness 4 weeks following cataract surgery is approximately 4 μm greater than preoperative measurements

^{14}and that thickening could potentially remain for up to 3 months following surgery.

^{15}This transient thickening of the RNFL could account completely for the difference in their measured and predicted RNFL. Second, they cite that interindividual slopes between SS and average RNFL vary too much to be useful. That conclusion may be inaccurate, as it was based on a small sample size as well as an unstable testing environment, as noted above. In our sample of 54 eyes, we found the variability in the mean slope of a population to be very low (i.e., a mean of 1.03 and a 95% CI of 1.01–1.05). The interindividual variability was accounted for in the upper tolerance limit of 4.40 μm, which not only describes the variability due to repeated measures but also captures the individual variability between SS and RNFL. In our model, although individual slopes may not exactly match the group mean, the upper tolerance limit provides a means of accounting for this variability. Third, Kok et al.

^{11}do not consider the confounding factor of repeated measures as a reason for the “limited” predictive value of their equation. Ours and other studies have shown that their “misprediction” could result almost entirely from test–retest variability (see more on this in the discussion below). Lastly, Kok et al.

^{11}conclude that their derived relationship seems to hold only in the artificial environment of measuring RNFL with optical filters and did not persist in a real-life situation. Contrary to this line of reasoning, the same relationship between SS and RNFL was seen in the Kim et al.

^{3}study, and their study included both glaucomatous patients and normals. It would be worthwhile to validate this relationship in a different setting, that is, a retrospective or prospective study of glaucoma and/or glaucoma suspect patients.

^{12}placed progressively denser neutral density filters between normal patients and the Stratus OCT machine to develop a mathematical correction formula to improve OCT imaging of the macula. By analyzing reflectivity peaks of the macula, they were able to develop a correction equation for diffusion filter–induced degradation in SS quality. Unfortunately, this complicated correction formula has not been validated outside of an artificially reduced SS condition, is not a part of the current OCT software, and is not commercially available, making it of little clinical value at this time.

^{1–3}Notably, research using the Cirrus OCT looking at test–retest reliability have limited their samples to patients with SS greater than 5,

^{3}greater than 6,

^{1}or greater than 7.

^{2,16}This is due, in part to the manufacturer's recommendation not to use scans with SS less than 5. The current study, however, demonstrated adequate ICCs, CVs, and S

_{w}s for SS even as low as 3 or 4 (Table 2), and it showed that the intravisit tolerance limit varied from 3.12 to 4.47 depending on the SS. A few others have suggested that RNFL measurements of lower SS may actually be acceptable. Ha et al.

^{5}found that the lower limit of acceptable SS in the Stratus OCT was 4. In another study using the Stratus OCT, Wu et al.

^{7}showed that in patients with an average SS of 4.3, there was only an average of a 5-μm difference between two separate readings. That 5-μm difference is similar to previously reported repeatability testing results for the Stratus.

^{13}

_{w}s, repeatability, and tolerance limits for the average RNFL and the four quadrants but do not report the data for individual clock hours. Previous studies have already shown that reproducibility is the best for the average RNFL measurement,

^{1,2}thus making it the most clinically relevant measure to follow. For that reason, the authors focused the current research on the average RNFL and included the quadrant data as a quality control measure to be verified for agreement with previous work on the Cirrus HD-OCT. Indeed, the current study, as well as those by both Mwanza et al.

^{1}and Leung et al.,

^{2}found the least variability in the average RNFL and temporal quadrant measurements; the nasal and inferior quadrant measurements showed the most variability. Kim et al.

^{3}and Cremasco et al.

^{16}also showed average RNFL to be the least variable, but found different patterns of variability in the quadrant RNFL measurements.

^{1,2}and not clinically significant. Third, our study included only healthy nonglaucomatous eyes. Since the main intent was to determine how varying the SS changes the RNFL thickness, we chose to include only healthy eyes so that every subject would be able to have a baseline SS of 9 or 10. The design of the study required normal subjects so that the authors could reproducibly decrease SS using diffusion filters. This may not have been possible had other media opacities or optic nerve abnormalities been present. This is an important point, as it is debatable whether the relationship between SS and RNFL thickness is the same in glaucoma and normal patients.

^{3,17}Vermeer et al.

^{17}found significant differences between the SS of glaucoma and normal patients, suggesting that glaucoma somehow directly affects the SS parameter on the OCT machine. On the other hand, Kim et al.

^{3}found in their multivariate analysis that the presence of glaucoma was not associated with interscan SS differences, suggesting that there is a similar relationship between RNFL and SS for both glaucoma and nonglaucomatous patients. If it is determined that this relationship is different between glaucoma and normal patients, the model that we propose may potentially be better suited for the monitoring of glaucoma suspect patients until they actually develop glaucoma.

**D.J. Russell**, None;

**S. Fallah**, None;

**C.J. Loer**, None;

**R.H. Riffenburgh**, None

*. 2010; 51: 5724–5730. [CrossRef] [PubMed]*

*Invest Ophthalmol Vis Sci**. 2009; 116: 1257–1263, 1263.e1–e2. [CrossRef] [PubMed]*

*Ophthalmology**. 2011; 55: 220–227. [CrossRef] [PubMed]*

*Jpn J Ophthalmol**. 2007; 114: 1505–1512. [CrossRef] [PubMed]*

*Ophthalmology**. 2012; 26: 182–188. [CrossRef] [PubMed]*

*Korean J Ophthalmol**. 2012; 26: 104–110. [CrossRef] [PubMed]*

*Korean J Ophthalmol**. 2009; 18: 213–216. [CrossRef] [PubMed]*

*J Glaucoma**. 2010; 54: 414–422. [CrossRef] [PubMed]*

*Jpn J Ophthalmol**. 2008; 115: 1347–1351, 1351.e1–e2. [CrossRef] [PubMed]*

*Ophthalmology**. 2011; 20: 371–376. [CrossRef] [PubMed]*

*J Glaucoma**. 2013; 91: 418–424. [CrossRef] [PubMed]*

*Acta Ophthalmol**. 2008; 49: 1609–1614. [CrossRef] [PubMed]*

*Invest Ophthalmol Vis Sci**. 2008; 115: 661–666. [CrossRef] [PubMed]*

*Ophthalmology**. 2014; 98: 52–58. [CrossRef] [PubMed]*

*Br J Ophthalmol**. 2009; 34: 1036–1041. [CrossRef] [PubMed]*

*Curr Eye Res**. 2011; 21: 264–270. [CrossRef] [PubMed]*

*Eur J Ophthalmol**. 2012; 53: 6102–6108. [CrossRef] [PubMed]*

*Invest Ophthalmol Vis Sci*