Secondary Logo

Journal Logo

Original Studies

Baseline Age and Mean Deviation Affect the Rate of Glaucomatous Vision Loss

Bommakanti, Nikhil MD*,†; De Moraes, Carlos G. MD, MPH, PhD; Boland, Michael V. MD, PhD; Myers, Jonathan S. MD§; Wellik, Sarah R. MD; Elze, Tobias PhD; Pasquale, Louis R. MD#; Shen, Lucy Q. MD**; Ritch, Robert MD#; Liebmann, Jeffrey M. MD

Author Information
doi: 10.1097/IJG.0000000000001401
  • Free

Abstract

The number of people with primary open-angle glaucoma (POAG) is projected to be over 80 million by 2040.1 Residual life expectancy is also increasing. Taken together, this will result in both an increased prevalence of the disease and longer disease exposure, which may ultimately lead to more patients experiencing significant vision loss.

Early initiation of treatment can decrease the rate of the visual field (VF) worsening and prevent the development of visual impairment.2 However, not all patients with POAG will have VF deterioration at a sufficiently rapid rate to experience a significantly worsened quality of vision within their respective lifetimes.3 Furthermore, glaucoma treatments are associated with side effects and adverse events, which also increase cost to the patient and the healthcare system. It is therefore desirable to determine characteristics at initial presentation which could help identify individuals who more likely to progress to visual impairment, so patients in need of closer monitoring and/or more aggressive treatment may be identified earlier in the course of their illness, while patients at decreased risk may be spared unnecessary intervention.

Longitudinal measurement of VF parameters allows for the determination of the rate of worsening.2,4 For a similar baseline level of damage, eyes with a more negative rate (slope) will reach the visual impairment endpoint more rapidly, therefore the rate of VF worsening can be used as a means of risk assessment to determine which eyes are more likely to experience significant impairment.

Several studies have investigated the influence of baseline risk factors, including age and mean deviation (MD), on the rate of progression, with conflicting results. In the Early Manifest Glaucoma Trial (EMGT),5 older age and worse MD were associated with faster progression, however, in Advanced Glaucoma Intervention Study6 eyes with better VFs (as measured by the Advanced Glaucoma Intervention Study score) were more likely to progress more rapidly. The Ocular Hypertension Treatment Study7 found an association with older age, whereas MD only reached borderline significance. The differences in results among these studies may be related in part to differences in disease severity of the populations investigated as well the methods for endpoint determination; of note, these clinical trials employed event-based methods to determine progression.

Data from clinical trials necessarily result from highly idealized populations which may have different characteristics than patients seen in clinical practice. Large sets of real-world data drawn from more representative patient populations may complement the findings from randomized controlled trials. To that end, our group created the Glaucoma Research Network (GRN), a research consortium comprised of 6 member institutions with VF data contributed by each institution.8–11

The purpose of this study is to investigate baseline risk factors associated with more rapid rates of VF worsening. Given the conflicting reports to date and the importance of age and disease severity on risk stratification, we hypothesized that older age and worse VFs at baseline would be associated with a more rapid rate of global and pointwise VF worsening.

METHODS

The institutional review boards at all sites approved the creation of the de-identified database of VF tests. The study adheres to the tenets of the Declaration of Helsinki. Written informed consent was waived given the retrospective nature of the study.

VF tests were selected from the GRN Database, which comprises all VF tests (standard automated perimetry) performed over varying time periods at its member institutions: Wilmer Eye Institute (Johns Hopkins University), Wills Eye Hospital (Thomas Jefferson University), Bascom Palmer Eye Institute (University of Miami), Massachusetts Eye and Ear (Harvard University), New York Eye and Ear Infirmary (Mount Sinai), and Edward S. Harkness Eye Institute (Columbia University). The complete, unfiltered dataset contains 963,913 VFs from 357,602 eyes from 190,144 patients. Patients in the dataset were treated at the discretion of attending physicians.

Inclusion and Exclusion Criteria

The subset of 24-2 VFs performed using the Swedish Interactive Thresholding Algorithm (Standard and Fast) with white-on-white stimuli of size III from patients older than 18 years of age at baseline (to exclude congenital and infantile glaucoma) and with baseline MD ≥−10 dB (to reduce floor effects due to limitations with perimetric measurement) was selected. VFs that did not meet the established reliability criteria, defined as having <15% false-positive errors and <30% fixation losses, were excluded.12 Eyes with ≥5 years of follow-up as well as those with ≥6 VFs (after removing the first test to minimize effects from learning), and a Glaucoma Hemifield Test outside normal limits or a pattern standard deviation probability <5% on at least 2 VF tests were included.

Global Analysis

Linear Regression

A least-squares linear regression model was fit using VF MD as a function of time for each patient eye; the slope was extracted, yielding a rate of VF worsening (mean deviation rate: MDR in dB/y) for each eye. Linear regression was used for the determination of MDR for consistency with measurements used in clinical practice. A linear mixed-effects model with baseline age and baseline MD as predictor variables and MDR as the response variable was fit to determine the association of MDR with baseline parameters. The model was specified with random intercepts at the patient level to account for crossed effects from some patients contributing both eyes to the dataset. A full linear mixed-effects model was subsequently fit using MDR as a function of both predictors as well as the interaction between the two, also with random intercepts for each patient.

Logistic Regression

To determine the predictive ability of baseline age and baseline MD to separate patients into a rapid versus slow progression group, eyes with MDR≤−1.0 dB/y were defined as belonging to the rapid progression group.13 The probability of belonging to the rapid progression group based on these baseline variables was determined using logistic regression.

Pointwise Analysis

MD is a summary statistic that results from the weighted average of the total deviation (age-corrected threshold sensitivity; TD) at all fifty-two (fifty-four minus two blind spot points in the 24-2 VF) test points. A pointwise total deviation rate (TDR) was determined using the slope of the linear regression model of TD as a function of time for points with baseline TD ≥−10 dB in a similar manner to MDR. Linear mixed-effects models were fit with pointwise TDR as a function of baseline age and baseline MD as above.

Statistical and Data Analysis

All analyses were performed using the R statistical programming language.14 A number of additional packages written by the R community were used in the process of data cleaning,15,16 analysis,15,17–22 and visualization23–25 as well as manuscript preparation.26–28 Statistical significance was defined as P-value <5%.

RESULTS

This study used 84,711 VFs from 8167 eyes from 5644 patients from a dataset of 963,913 VFs from 357,602 eyes from 190,144 patients. Baseline patient ages ranged from 18 to 96 years and follow-up ranged from 5 to 18 years. The mean baseline age was 66 years and the mean baseline MD was −4 dB (Fig. 1).

FIGURE 1
FIGURE 1:
Baseline characteristics. A, Number of visual field tests per eye. B, Length of follow-up per eye in years. C, Baseline age per eye in years. D, Baseline mean deviation (MD) per eye in dB.

Global Analysis

The linear mixed-effects model with baseline age and baseline MD as predictor variables yielded the following coefficients β (P): age: β=−0.0079 dB/y2 (P<0.001); MD: β=0.012/y (P<0.001), meaning MDR worsened by −0.0079 dB/y for each year increase in baseline age (older age) and by 0.012 dB/y for each dB decrease in baseline MD (worse MD) (Fig. 2).

FIGURE 2
FIGURE 2:
Linear mixed-effects models. A, Mean deviation rate (MDR) as a function of baseline age and as a function of baseline mean deviation (MD). B, Coefficients of the models with 95% confidence intervals and a dashed vertical line indicating zero.

In the full model with both parameters and their interaction, all terms were significant (baseline age: P<0.001; baseline MD: P=0.025; baseline age×baseline MD: β=0.00065, P=0.0017) (Fig. 3, Table 1). The statistically significant interaction means that MDR worsens more rapidly with older baseline age when baseline MD is more negative.

FIGURE 3
FIGURE 3:
Qualitative representation of the interaction between baseline age and baseline mean deviation (MD). The rate of change of mean deviation rate (MDR) as a function of one parameter changes based on the value of the other parameter.
TABLE 1
TABLE 1:
Coefficients With 95% Confidence Intervals and P-Values for Multivariate Linear Mixed-effects Model (With Random Intercepts for Patients) of MDR as a Function of Baseline Age, Baseline MD, and Their Interaction

In the logistic regression, 754 (10%) of the eyes were in the rapid progression group (Fig. 4), which was defined as having an MDR≤−1.0 dB/y. The odds of belonging to the rapid progression group were predicted to change by a factor of 1.033 for each unit increase in baseline age and by a factor of 0.8821 for each unit increase in baseline MD (less severe visual loss; Table 2). The probability of belonging to the rapid progression group increased with older baseline age and worse baseline MD (Fig. 5), a finding broadly consistent with the findings from the linear model.

FIGURE 4
FIGURE 4:
Histogram of mean deviation rate by eye. Rapid versus slow progression groups were defined by a cutoff of mean deviation=−1.0 dB.
TABLE 2
TABLE 2:
Odds With 95% Confidence Intervals and P-Values for Logistic Regression With Group Membership Defined as MDR≤−1.0 dB/y
FIGURE 5
FIGURE 5:
Probability of belonging to the rapid progression group, defined as having a mean deviation (MD) rate≤−1.0 dB/y, based on the logistic regression model.

Pointwise

The mean pointwise TDR ranged from −0.21 to −0.55 dB/y, with the most rapid pointwise progression observed in the nasal and paracentral fields (Fig. 6). The coefficients of the linear model of TDR as a function of baseline age were clustered around −0.01 dB/y (ranging from −0.0066 to −0.012), demonstrating a diffuse effect of baseline age. The coefficients for TDR as a function of baseline MD ranged from 0.0059 to 0.032 dB/y, suggesting that with worse baseline VFs, the nasal and paracentral test locations progress more rapidly than others (Fig. 7).

FIGURE 6
FIGURE 6:
Total deviation rate for each eye and mean value, by visual field test point. Background color corresponds to the magnitude of the mean total deviation rate.
FIGURE 7
FIGURE 7:
Coefficients of the linear mixed-effects models of total deviation rate (TDR) as a function of baseline age and as a function of baseline mean deviation.

DISCUSSION

We examined the influence of baseline age and baseline MD on the rate of VF worsening on a global and pointwise basis using a large, real-world dataset and found: (1) increased baseline age and worse baseline MD leads to an increased rate of VF deterioration and a greater probability of having a significantly rapid rate of worsening; (2) among older patients, worse baseline MD leads to an even faster deterioration in VFs; (3) the nasal and paracentral regions of the field experience the fastest rate of worsening; and (4) the pointwise VF loss is separated into a diffuse, age-related component and a localized, MD-related component. The heterogenous pointwise progression is consistent with the localized nature of glaucoma loss and demonstrates that the observed findings are not due to cataract alone.

These findings augment our understanding of the nature of VF worsening. They confirm the findings of previous studies (including prospective and retrospective cohorts as well as a large data study29–32) with regard to the effect of worse baseline MD on future progression. They do not support the findings from a randomized clinical trial (EMGT5) in which baseline MD reached borderline significance only. We confirmed previous studies (including the EMGT) on the role of age as a predictor of the rate of VF worsening.

These findings also provide new understanding about pointwise deterioration and offer actionable information to clinicians. In particular, physicians may be increasingly vigilant about monitoring older patients with existing VF loss and may pay particular attention to the nasal and paracentral regions of the field.

Global measures of progression are widely used and, being summary statistics, have lower variability. Conversely, a pointwise approach provides insight into the topographic nature of the disease, with the drawback that it has greater variability and is more difficult to track and interpret.2 For consistency with clinical practice and to deliver novel insights, this study employed the use of both approaches to measure VF worsening.

The effect sizes in this study were small (for comparison, every 10 y older baseline age and every 10 dB worse baseline MD led to an increased rate of VF deterioration by −0.1 dB/y, analogous to ∼1 mm Hg higher intraocular pressure in Ocular Hypertension Treatment Study7). Incidentally, this is likely one of the reasons why conflicting findings were previously observed: small effect sizes necessitate large sample sizes. The small effect sizes may be explained in part by the fact that this dataset employed real-world data. In practice, patients with more significant loss are treated more aggressively, which slows their rate of worsening. Therefore the true effect sizes (as would be seen in the natural history of the disease) are likely to be greater, meaning the values obtained in this study are likely to be conservative estimates of the true effect of these baseline parameters on the rate of VF worsening.

The patients in this study were followed at academic centers. As such centers have a large referral base, the initial observation in the GRN dataset may not be the first VF test for that patient. In this context, the term “baseline” used in this study cannot and does not refer to the very first examination this patient has received; instead, it refers to the first test which a physician observes in his or her first encounter with a patient.

Limitations

This was a retrospective study, which may have influenced our results due to selection and survivorship bias. All VFs performed over a fixed period of time, without selection based on disease, were contributed to the GRN dataset. Some data from patients with diseases other than POAG were therefore included in the dataset. We minimized the possibility of including these VFs in our analysis by selecting for series with multiple fields performed over a long period of follow-up, and we further enriched our sample by introducing strict criteria for the Glaucoma Hemifield Test and pattern standard deviation for tests in each patient eye series.

Using a small proportion of the full database in this way may raise questions regarding the generalizability of these results. Nonetheless, these inclusion and exclusion criteria were necessary to meet requirements for the analysis of VF progression using linear regression, such as having a minimum number of reliable tests of the same strategy and pattern.

As this dataset lacked complementary clinical information, variables including intraocular pressure, central corneal thickness, and race were not able to be assessed, and these unavailable parameters may have acted as confounders.

Furthermore, patients may have undergone cataract or glaucoma surgery during the course of their follow-up, and there maybe a concern that this would influence their measured rate of VF worsening. Such surgical intervention would increase MD on future tests, which would result in a less negative slope of MD as a function of time (ie, yield a slower rate of VF worsening than what would be observed in the absence of intervention). In fact, this means that the true effect is likely to be more dramatic than what was determined by the conservative estimates in this study.

Finally, there are certain unmet assumptions required for linear regression, including homogeneity of variance (homoscedasticity). There have been several attempts to use other mathematical models (including censored linear, exponential, quadratic, and logistic models),33,34 and some investigators have found that exponential models more accurately fit the observed data35 However, (1) the difference in fit between linear and exponential models is not clinically significant; (2) linear models are used clinically and more readily interpretable (that is, they are commercially available and are parsimonious); and (3) exponential models require non-negative inputs, which precludes the use of age-corrected threshold deviation (as was used in this study). Because this study was focused on the effect of age, not correcting for it could introduce bias. Furthermore, other investigators have found in their datasets that linear regression models demonstrated better fit and predictive capability than exponential and other models, despite the violation of statistical assumptions.34 For these reasons, this study used linear regression to measure rate of VF worsening.

Generalizability

The large sample size and real-world nature of this study improve its generalizability. However, these findings can only be applied to glaucoma patients older than 18 years of age and with baseline MD better than −10 dB on 24-2 VFs with reliability indices consistent with those used as inclusion criteria in this study. These results should not be extrapolated to patients who do not meet the inclusion criteria for this study.

Future Directions

These data can inform the development of new VF testing algorithms which may, for example, increase sampling of the 24-2 test points with the greatest likelihood of progression based upon baseline characteristics (age and MD). Figure 8 demonstrates how the test points most likely to worsen differ based on baseline MD.

FIGURE 8
FIGURE 8:
Coefficients of the linear mixed-effects model of total deviation rate (TDR) as a function of baseline mean deviation (MD) for 3 subsets grouped by baseline MD.

REFERENCES

1. Tham Y-C, Li X, Wong TY, et al. Global prevalence of glaucoma and projections of glaucoma burden through 2040. Ophthalmology. 2014;121:2081–2090.
2. De Moraes CG, Liebmann JM, Levin LA. Detection and measurement of clinically meaningful visual field progression in clinical trials for glaucoma. Prog Retin Eye Res. 2017;56:107–147.
3. Saunders LJ, Russell RA, Kirwan JF, et al. Examining visual field loss in patients in glaucoma clinics during their predicted remaining lifetime. Invest Ophthalmol Vis Sci. 2014;55:102–109.
4. Caprioli J. The importance of rates in glaucoma. Am J Ophthalmol. 2008;145:191–192.
5. Leske MC, Heijl A, Hussein M, et al. Factors for glaucoma progression and the effect of treatment: The Early Manifest Glaucoma Trial. Arch Ophthalmol. 2003;121:48–56.
6. AGIS Investigators. The Advanced Glaucoma Intervention Study (AGIS): 12. Baseline risk factors for sustained loss of visual field and visual acuity in patients with advanced glaucoma. Am J Ophthalmol. 2002;134:499–512.
7. Gordon MO, Beiser JA, Brandt JD, et al. The Ocular Hypertension Treatment Study: baseline factors that predict the onset of primary open-angle glaucoma. Arch Ophthalmol. 2002;120:714–720; discussion 829–830.
8. Wang M, Shen LQ, Boland MV, et al. Impact of natural blind spot location on perimetry. Sci Rep. 2017;7:6143.
9. Wang M, Pasquale LR, Shen LQ, et al. Reversal of glaucoma hemifield test results and visual field features in glaucoma. Ophthalmology. 2018;125:352–360.
10. Wang M, Shen LQ, Pasquale LR, et al. An artificial intelligence approach to detect visual field progression in glaucoma based on spatial pattern analysis. Invest Ophthalmol Vis Sci. 2019;60:365–375.
11. Saeedi OJ, Elze T, D’Acunto L, et al. Agreement and predictors of discordance of 6 visual field progression algorithms. Ophthalmology. 2019;126:822–828.
12. Yohannan J, Wang J, Brown J, et al. Evidence-based criteria for assessment of visual field reliability. Ophthalmology. 2017;124:1612–1620.
13. Chauhan BC, Garway-Heath DF, Goñi FJ, et al. Practical recommendations for measuring rates of visual field change in glaucoma. Br J Ophthalmol. 2008;92:569–573.
14. R Core Team. R: A Language and Environment for Statistical Computing. Vienna, Austria: R Foundation for Statistical Computing; 2019.
15. Wickham H, François R, Henry L, et al. Dplyr: a grammar of data manipulation; 2019. Available at: https://CRAN.R-project.org/package=dplyr. Accessed August 5, 2019.
16. Wickham H, Henry L. Tidyr: easily tidy data with ’spread()’ and ’gather()’ functions; 2019. Available at: https://CRAN.R-project.org/package=tidyr. Accessed August 5, 2019.
17. Henry L, Wickham H. Purrr: functional programming tools; 2019. Available at: https://CRAN.R-project.org/package=purrr. Accessed August 5, 2019.
18. Dowle M, Srinivasan A. Data.table: extension of ‘data.frame’; 2019. Available at: https://CRAN.R-project.org/package=data.table. Accessed August 5, 2019.
19. Robinson D, Hayes A. Broom: convert statistical analysis objects into tidy tibbles; 2019. Available at: https://CRAN.R-project.org/package=broom. Accessed August 5, 2019.
20. Bolker B, Robinson D. Broom.mixed: tidying methods for mixed models; 2019. Available at: https://CRAN.R-project.org/package=broom.mixed. Accessed August 5, 2019.
21. Bates D, Maechler M, Bolker B, et al. Lme4: linear mixed-effects models using ‘eigen’ and s4; 2019. Available at: https://CRAN.R-project.org/package=lme4. Accessed August 5, 2019.
22. Kuznetsova A, Bruun Brockhoff P, Haubo Bojesen Christensen R. LmerTest: tests in linear mixed effects models; 2019. Available at: https://CRAN.R-project.org/package=lmerTest. Accessed August 5, 2019.
23. Wickham H, Chang W, Henry L, et al. Ggplot2: create elegant data visualisations using the grammar of graphics; 2019. Available at: https://CRAN.R-project.org/package=ggplot2. Accessed August 5, 2019.
24. Pedersen TL. Patchwork: the composer of ggplots; 2017. Available at: https://github.com/thomasp85/patchwork. Accessed August 5, 2019.
25. Wilke CO. Cowplot: streamlined plot theme and plot annotations for ‘ggplot2’; 2019. Available at: https://CRAN.R-project.org/package=cowplot. Accessed August 5, 2019.
26. Müller K. Here: a simpler way to find your files; 2017. Available at: https://CRAN.R-project.org/package=here. Accessed August 5, 2017.
27. Xie Y. Knitr: a general-purpose package for dynamic report generation in R; 2019. Available at: https://CRAN.R-project.org/package=knitr. Accessed August 5, 2019.
28. Xie Y. Bookdown: authoring books and technical documents with R markdown; 2018. Available at: https://CRAN.R-project.org/package=bookdown. Accessed August 5, 2018.
29. Forchheimer I, de Moraes CG, Teng CC, et al. Baseline mean deviation and rates of visual field change in treated glaucoma patients. Eye. 2011;25:626–632.
30. Garg A, De Moraes CG, Cioffi GA, et al. Baseline 24-2 central visual field damage is predictive of global progressive field loss. Am J Ophthalmol. 2018;187:92–98.
31. Park H-YL, Hong KE, Park CK. Impact of age and myopia on the rate of visual field progression in glaucoma patients. Medicine (Baltimore). 2016;95:e3500.
32. Fujino Y, Asaoka R, Murata H, et al. Evaluation of glaucoma progression in large-scale clinical data: the Japanese Archive of Multicentral Databases in Glaucoma (JAMDIG). Invest Ophthalmol Vis Sci. 2016;57:2012–2020.
33. Caprioli J, Mock D, Bitrian E, et al. A method to measure and predict rates of regional visual field decay in glaucoma. Invest Ophthalmol Vis Sci. 2011;52:4765–4773.
34. Bryan SR, Vermeer KA, Eilers PHC, et al. Robust and censored modeling and prediction of progression in glaucomatous visual fields. Invest Ophthalmol Vis Sci. 2013;54:6694–6700.
35. Chen A, Nouri-Mahdavi K, Otarola FJ, et al. Models of glaucomatous visual field loss. Invest Ophthalmol Vis Sci. 2014;55:7881–7887.
Keywords:

glaucoma; progression; visual fields; age; mean deviation

Copyright © 2020 Wolters Kluwer Health, Inc. All rights reserved.