purpose. To determine the probability of future glaucomatous visual field (VF) progression with clinical and perimetric data.

methods. One hundred sixty-one eyes of patients (161) enrolled in the Advanced Glaucoma Intervention Study (AGIS) with ≥8 years of follow-up and a baseline VF score ≤16 were selected. VF progression at 8 years was determined with point-wise linear regression (PLR) analysis, using a two-omitting algorithm. The course of VF series over the first 4 years of follow-up was quantified by an index, the sum of slopes, which is the sum of all slopes of VF thresholds with *P* < 0.05 when PLR was performed on the 4-year data. The following parameters were included in a logistic regression model to predict 8-year outcomes from the first 4 years of follow-up: intervention sequence, age, AGIS VF score, mean IOP, IOP fluctuation, and sum of slopes.

results. Sixty-four (40%) eyes progressed after 8 years as determined by PLR analysis. Two parameters were predictive of subsequent VF progression, as identified at 8 years (predictive power: 76%): more negative sum of slopes (i.e., faster or more extensive deterioration; *P* < 0.001) and older age at 4 years (*P* = 0.049). When sum of slopes alone was used to predict outcomes at 8 years, the predictive power was the same.

conclusions. The VF sum of slopes can be used to estimate the probability of subsequent VF worsening with reasonable, clinically useful accuracy. This probability may be combined with other clinical information for more effective clinical predictions and treatment decisions.

^{ 1 }and previous AGIS reports

^{ 2 }

^{ 3 }have shown the most important predictors for visual field (VF) progression to be older age at the time of first glaucoma intervention, greater intraocular pressure (IOP) fluctuation, higher mean IOP, and lower baseline AGIS VF score. Length of follow-up and number of glaucoma interventions have also been found to be less important, but significant, risk factors for glaucomatous VF progression. Clinicians often have available VF series obtained during earlier follow-up and depend on both kinds of data to make an informed decision about glaucoma treatment in a given patient. No study has been undertaken to investigate whether and how combining these different types of clinical information would be helpful for predicting the subsequent course of glaucoma in an individual patient.

^{ 4 }Phakic patients, 35 to 80 years of age, with open-angle glaucoma no longer controlled by maximally tolerated medical treatment were recruited. The eligible eyes had to have a best corrected visual acuity score of at least 56 letters (Early Treatment Diabetic Retinopathy Study charts) and meet specified criteria for combinations of consistently elevated IOP, despite maximum-tolerated and effective medical therapy, glaucomatous VF defect, and/or optic disc rim deterioration.

^{ 4 }Between 1988 and 1992, investigators at 12 participating AGIS clinical centers enrolled 789 eyes of 591 patients. Eyes were randomly assigned to one of two surgical-intervention sequences: argon laser trabeculoplasty–trabeculectomy–trabeculectomy (ATT) or trabeculectomy–argon laser trabeculoplasty–trabeculectomy (TAT). Follow-up study visits were scheduled 3 and 6 months after enrollment and every 6 months thereafter. The institutional review boards at each of the participating centers approved the AGIS protocol, and all patients provided informed consent. The research was conducted in accordance with the tenets of the Declaration of Helsinki. Data in this report are based on database closure of March 31, 2001.

^{ 5 }Study measurements were made at baseline, 3 months after initial intervention, and at each 6-month follow-up examination. Baseline, or reference, measurements were performed after the eligibility measurements but before the first surgical intervention, to avoid the effect of regression to the mean.

^{ 6 }We used the two-omitting regression algorithm recently described by Gardiner and Crabb

^{ 7 }for definition of change versus stability at each point at 8 years. In summary, a test location is considered progressing or improving during the follow-up period if the regression slope is statistically and clinically significant (as defined later) in both of the following regression analyses: (1) after omitting the last threshold in a series and (2) after deleting the threshold before last for the same series. This approach has been shown, in simulation experiments, to be more specific than using all the data points for a single regression analysis, and it maintains a sensitivity comparable to other stringent algorithms used for the same purpose, such as two of two

^{ 8 }or three of four.

^{ 9 }

^{ 10 }Regression slopes were considered significant if ≥1.00 dB/year or ≤−1.0 dB/year in presence of

*P*≤0.01.

^{ 6 }: the two point Glaucoma Hemifield Test (GHT) change criterion. According to this criterion, a VF series is designated as changing if two test locations belonging to the same GHT cluster demonstrate change in the same direction. This set of criteria was found to be the most conservative among various PLR approaches. VF outcomes from PLR at 8 years were classified as progressing or nonprogressing. Improving and stable eyes were categorized together as nonprogressing.

*P*< 0.05.

^{2}test, unpaired

*t*-test, or Mann-Whitney test, depending on the type of data) at

*P*≤ 0.20 were included in the final model. In addition, we included all clinically relevant variables that might predict or confound detection of VF progression. The following risk factors or potentially confounding factors were entered into the final logistic model for prediction of VF worsening at 8 years: intervention sequence, age at 4-year follow-up, mean IOP and IOP fluctuation during the first 4 years of follow-up, AGIS VF score, and sum of slopes at 4 years. Standard deviation of the IOP at all visits after the initial surgery was used as a measure of IOP fluctuation.

*P*

_{ x }is the probability of occurrence of an outcome (VF progression here),

*e*is the root of the natural logarithm,

*b*

_{0}is the intercept,

*b*

_{1… n }is the regression coefficient(s) for independent variable(s), and

*X*

_{1… n }is independent variable(s) included in the logistic regression equation.

*P*≤ 0.05 or less were considered statistically significant.

*P*< 0.001; Mann-Whitney test, Fig. 1 ), age at 4-year follow-up (

*P*< 0.001; unpaired

*t*-test), and AGIS scores at 4 years (

*P*= 0.001, unpaired

*t*-test). Neither average IOP nor IOP fluctuation during the first 4 years of follow-up was significantly different between the progressing and nonprogressing eyes. On multivariate analysis, the following two variables were associated with VF progression (Table 2) : a more negative sum of slopes during the first 4 years (

*P*< 0.001) and older age at 4 years of follow-up (

*P*= 0.049).

*P*≤ 0.01) had any predictive power for the fate of VF series at 8 years. We found that PLR at 4 years missed 67% (43/64 eyes) of progressing eyes, whereas it rarely detected progression (1/ 97 eyes, 1%) in the absence of worsening at 8 years.

*P*< 0.001).

^{ 11 }compared different curve-fitting models for predicting individual VF thresholds over time. They found that PLR was the best model for prediction of threshold sensitivity at individual test locations when the first five VFs were used to predict the threshold sensitivity at selected test locations on the 15th VF examination performed during follow-up of patients with normal-tension glaucoma. Our study is different in that we addressed the problem of predicting the course of the entire VF series at 8 years using information available during the first 4 years of follow-up, including clinical and perimetric data. The AGIS database was considered an appropriate database for this investigation, because a large proportion of patients enrolled in AGIS were followed for >8 years. The AGIS patients also had regular VF examinations every 6 months during the course of the study, making this database an excellent medium for application of PLR. However, it must be emphasized that AGIS patients represent a subgroup of patients with advanced glaucoma who are no longer controlled on medical treatment. Hence, the rate and pattern of glaucoma progression is not necessarily generalizable to patients at large with primary open-angle glaucoma. This, though, does not diminish the value of the model presented in this investigation.

*P*

_{ x }: 20% and 66%) provide quantitative probabilities that are more meaningful and can be taken into account along with other clinical findings for better clinical decision making.

^{ 1 }

^{ 2 }In the present study, neither mean IOP nor IOP fluctuation during the first 4 years of follow-up had predictive value for the VF status at 8 years. We speculate that there may be an inadequate number of IOP measurements during the first 4 years for a robust estimate of mean IOP or IOP fluctuation. These results do not necessarily contradict the findings of the aforementioned studies.

^{ 12 }

^{ 13 }

^{ 14 }Therefore, assuming that VFs are performed every 6 months after the baseline examination, PLR can be applied only after three or more years of follow-up.

^{ 6 }Of note, when AGIS criteria were used for definition of VF outcomes at 8 years, the sum of slopes was still the strongest and only significant predictor of subsequent VF change. This confirms the value of sum of slopes for forecasting the subsequent course of VFs.

Progressing | Nonprogressing | ||||||
---|---|---|---|---|---|---|---|

No. | % | No. | % | P | |||

Total | 64 | 39.8 | 97 | 60.2 | |||

Eye | |||||||

Right | 36 | 56.3 | 40 | 41.2 | 0.060 | ||

Left | 28 | 43.8 | 57 | 58.8 | |||

Gender | |||||||

Male | 29 | 44.4 | 48 | 49.5 | 0.518 | ||

Female | 35 | 55.6 | 49 | 50.5 | |||

Race | |||||||

Black | 35 | 54.7 | 47 | 48.5 | 0.687 | ||

White | 28 | 43.8 | 49 | 50.5 | |||

Hispanic | 1 | 1.6 | 1 | 1.0 | |||

Age after 4 years of follow-up (y) | |||||||

Mean | 71 | 66 | < 0.001^{*} | ||||

SD | 6.7 | 10.8 | |||||

Range | 49–82 | 45–85 | |||||

Intervention sequence | |||||||

ATT | 32 | 50 | 51 | 52.6 | 0.749 | ||

TAT | 32 | 50 | 46 | 47.4 | |||

Cataract surgery during follow-up | |||||||

No | 34 | 53.1 | 58 | 59.8 | 0.403 | ||

Yes | 30 | 46.9 | 39 | 40.2 | |||

No. of visual field exams per eye during first 4 years | |||||||

Mean | 9.6 | 9.6 | 0.370 | ||||

SD | 0.9 | 1 | |||||

Range | 6–11 | 6–11 | |||||

Mean IOP during first 4 years (mm Hg) | |||||||

Mean | 16.6 | 15.7 | 0.080 | ||||

SD | 3 | 3.4 | |||||

Range | 7.2–22.7 | 6.9–22.3 | |||||

IOP fluctuation (mm Hg) | |||||||

Mean | 3.2 | 2.8 | 0.190 | ||||

SD | 1.9 | 2 | |||||

Range | 1.0–12.0 | 0.7–15.9 | |||||

AGIS visual field score at 4 years | |||||||

Mean | 9.5 | 7 | 0.001^{*} | ||||

SD | 4.1 | 4.5 | |||||

Range | 0–16 | 0–16 | |||||

Sum of slopes at 4 years (dB/y) | |||||||

Mean | −28.7 | 1.9 | < 0.001^{, †} | ||||

SD | 38.9 | 14.5 | |||||

Range | −236.9 to +10.4 | −58.6 to +42.2 |

**Figure 1.**

**Figure 1.**

P | Odds Ratio | 95% CI for OR | |||
---|---|---|---|---|---|

Lower | Upper | ||||

Intervention sequence (TAT sequence) | 0.709 | 1.172 | 0.509 | 2.699 | |

Age at 4 years of follow-up (y) | 0.049 | 1.048 | 1.000 | 1.097 | |

AGIS VF score at 4 years | 0.590 | 1.028 | 0.931 | 1.135 | |

Mean IOP during the first 4 years (mm Hg) | 0.246 | 1.083 | 0.947 | 1.238 | |

IOP fluctuation during the first 4 years (mm Hg) | 0.966 | 0.995 | 0.803 | 1.233 | |

Sum of slopes at 4 years (dB/y) | < 0.001 | 0.918 | 0.886 | 0.951 |

P | Odds Ratio | 95% CI for OR | |||
---|---|---|---|---|---|

Lower | Upper | ||||

No cataract extraction | |||||

Intervention sequence (TAT sequence) | 0.827 | 0.891 | 0.316 | 2.510 | |

Age at 4 years of follow-up (years) | 0.040 | 1.068 | 1.003 | 1.138 | |

AGIS VF score at 4 years | 0.061 | 1.128 | 0.995 | 1.279 | |

Mean IOP during the first 4 years (mm Hg) | 0.740 | 0.971 | 0.819 | 1.152 | |

IOP fluctuation during the first 4 years (mm Hg) | 0.540 | 1.115 | 0.786 | 1.582 | |

Sum of slopes at 4 years (dB/year) | 0.027 | 0.952 | 0.912 | 0.994 | |

Cataract extraction during follow-up | |||||

Intervention sequence (TAT sequence) | 0.693 | 1.439 | 0.237 | 8.759 | |

Age at 4 years of follow-up (years) | 0.371 | 1.046 | 0.947 | 1.156 | |

AGIS VF score at 4 years | 0.076 | 0.799 | 0.623 | 1.024 | |

Mean IOP during the first 4 years (mm Hg) | 0.362 | 1.170 | 0.834 | 1.641 | |

IOP fluctuation during the first 4 years (mm Hg) | 0.542 | 1.108 | 0.797 | 1.540 | |

Sum of slopes at 4 years (dB/year) | 0.001 | 0.826 | 0.737 | 0.927 |

**Figure 2.**

**Figure 2.**

**Figure 3.**

**Figure 3.**

*Ophthalmology*

*.*2004;111:1627–1635. [CrossRef] [PubMed]

*Am J Ophthalmol*

*.*2000;130:429–440. [CrossRef] [PubMed]

*Am J Ophthalmol*

*.*2002;134:499–512. [CrossRef] [PubMed]

*Control Clin Trials*

*.*1994;15:299–325. [CrossRef] [PubMed]

*Ophthalmology*

*.*1994;101:1445–1455. [CrossRef] [PubMed]

*Arch Ophthalmol*

*.*.In press.

*Invest Ophthalmol Vis Sci*

*.*2002;43:1400–1407. [PubMed]

*Br J Ophthalmol*

*.*1991;75:493–495. [CrossRef] [PubMed]

*Br J Ophthalmol*

*.*2000;84:1154–1158. [CrossRef] [PubMed]

*Br J Ophthalmol*

*.*2001;85:696–701. [CrossRef] [PubMed]

*Graefes Arch Clin Exp Ophthalmol*

*.*1995;233:750–755. [CrossRef] [PubMed]

*Acta Ophthalmol (Copenh)*

*.*1982;60:267–274. [PubMed]

*Acta Ophthalmologica Suppl*

*.*1985;173:19–21.

*Invest Ophthalmol Vis Sci*

*.*2000;41:2192–2200. [PubMed]