To investigate the relationship between the rate of retinal nerve fiber layer (RNFL) loss during initial follow-up and the magnitude of associated visual field loss during an extended follow-up period.
Retrospective cohort study.
A total of 1,150 eyes of 839 glaucoma patients extracted from the Duke Glaucoma Registry. Rates of RNFL loss were obtained from global RNFL thickness values of the first 5 optical coherence tomography (OCT) scans. Rates of visual field loss were assessed using standard automated perimetry mean deviation (SAP MD) during the entire follow-up period. Joint longitudinal mixed effects models were used to estimate rates of change. Eyes were categorized as fast, moderate or slow progressors based on rates of RNFL loss, with cutoffs of ≤-2 µm/year, -2 to -1 µm/year and ≥-1 µm/year, respectively. Univariable and multivariable regressions were completed to identify significant predictors of SAP MD loss.
The rate of RNFL change was -0.76±0.85 µm/y during initial follow-up, which occurred over 3.7±1.5 years. 765 (66%) eyes were slow, 328 (29%) moderate, and 57 (5%) fast progressors, with rates of RNFL thinning of -0.36±0.54 µm/year, -1.34±0.25 µm/year, and -2.87±1.39 µm/year respectively. The rates of SAP MD loss among slow, moderate, and fast OCT progressors were -0.16±0.35 dB/y, -0.32±0.43 dB/y, and -0.71±0.65 dB/y respectively over the extended follow-up period of 6.1±1.9 years (P<0.001). Age, OCT progressor group, and concurrent SAP rate were all significantly associated with the overall rate of SAP MD loss in a multivariable model (all P<0.001).
Rapid RNFL thinning during an initial follow-up period was predictive of concurrent and subsequent rates of visual field decline over an extended period.
O ptical coherence tomography (OCT) has been widely used to evaluate neural loss in glaucoma, a progressive optic neuropathy that is the leading cause of irreversible blindness in the world. The attractiveness of OCT resides in its ability to quickly obtain quantitative and reproducible measurements of different anatomic regions of the eye affected by glaucoma, such as the retinal nerve fiber layer (RNFL).
Since OCT does not directly represent a patient’s visual function, the use of its measurements in clinical practice relies on the belief that these measurements are predictive of clinically relevant outcomes (i.e., loss of vision). For example, in individuals who are suspected to have glaucoma, the rationale for using OCT as a diagnostic test is to detect structural damage before the appearance of visual field defects on standard automated perimetry (SAP). Similarly, in individuals already diagnosed with glaucoma, progressive loss of neural tissue detected by OCT is believed to be associated with an increased risk of disease progression and worse visual outcomes. As such, abnormalities and change detected on OCT would provide an opportunity to initiate or escalate treatment to prevent irreversible loss of vision.
Several studies have investigated whether OCT measurements are predictive of future functional deterioration. A study by Miki et al. demonstrated that RNFL thinning predicts future visual field defects in glaucoma suspects. Each 1-µm/year faster rate of RNFL loss corresponded to a hazard ratio of 2.05 for the risk of visual field loss in their cohort. Similarly, Yu et al. showed that progressive RNFL loss had a hazard ratio of 3.81 to predict progressive visual field loss in patients already diagnosed with glaucoma after adjustment for covariates. Although these studies have provided important assessments of the predictive value of OCT, they do not provide easily interpretable quantitative descriptions of how change in OCT is related to change in SAP; hazard ratios do not provide information that can be easily translated to clinical practice. For a given RNFL thinning, the risk of SAP deterioration may be 2 or 3 times higher than a certain baseline but remain relatively small in magnitude. Hazard ratios are relative measures which do not help a clinician decide whether a specific rate of change on OCT actually carries a risk of fast visual field progression for an individual patient. More specifically, it is important to demonstrate that rapid loss observed on OCT during a given follow-up period is associated with concurrent or future risk of fast deterioration on SAP.
The purpose of this study was to investigate the relationship between the magnitude of RNFL thinning seen during an initial period of follow-up of glaucoma patients and the magnitude of concurrent and subsequent visual field loss seen during an extended follow-up period of the same cohort. We hypothesized that fast initial RNFL progression would be associated with fast visual field progression, justifying the rationale for and providing guidance to the use of longitudinal OCT results in assisting clinical decision-making in glaucoma.
The dataset utilized in this study was derived from the Duke Glaucoma Registry developed by the Vision, Imaging and Performance (VIP) Laboratory of Duke University. Institutional Review Board (IRB) approval was obtained for this analysis, and a waiver of informed consent was provided due to the retrospective nature of this work. All methods adhered to the tenets of Declaration of Helsinki for research involving human participants.
The database contained clinical information from baseline and follow-up visits, including patient diagnostic and procedure codes, medical history and stereoscopic optic disc photographs. The study included patients previously diagnosed with primary open-angle glaucoma (POAG) based on International Classification of Diseases (ICD) codes. Patients were excluded if they presented with any other ocular or systemic disease that could affect the optic nerve or visual field (e.g. retinal detachment, retinal or malignant choroidal tumors, non-glaucomatous disorders of the optical nerve and visual pathways, atrophic and late-stage dry age-related macular degeneration, amblyopia, uveitis and/or venous or arterial retinal occlusion) according to ICD codes. In addition, tests performed after panretinal photocoagulation, according to Current Procedural Terminology (CPT) codes, were excluded. ICD and CPT codes used to construct this database have been detailed in a previous work. Glaucomatous eyes were required to have an abnormal visual field at baseline (i.e., GHT “outside normal limits” or PSD probability <5%). Intraocular pressure (IOP) measurements were completed using Goldmann applanation or Tonopen tonometry (Reichert Technologies, Depew, NY).
All eligible subjects had imaging with the Spectralis spectral-domain OCT (SD-OCT) system (Heidelberg Engineering, GmbH, Dossenheim, Germany) and SAP tests using the Humphrey Field Analyzer II or III (Carl Zeiss Meditec, Inc., Dublin, CA). The mean global RNFL thickness was calculated by averaging the measurements acquired from a 12-degree (for single circle scans) or a 3.45-mm peripapillary circle scan (for scans from the Glaucoma Module Premium Edition), as described in detail previously. OCT scans were excluded if the scan quality score was less than 15. In addition, eyes with a baseline RNFL value ≤38 µm were excluded due to the “floor effect,” which would preclude subsequent trend analysis of RNFL thickness during follow-up. Since manual review of all tests was impractical, we excluded scans that had a global RNFL thickness greater than 130 µm, representing implausible measurements above the higher range of reported RNFL thickness for normal controls. Values outside this range likely indicate the presence of acquisition or segmentation errors in the presence of otherwise adequate quality scores. SAP tests included 24-2 and 30-2 Swedish Interactive Threshold Algorithm (SITA) tests with size III white stimulus. Visual fields were excluded from this analysis if they had more than 33% fixation losses or more than 15% false-positive errors, or if the result of the glaucoma hemifield test (GHT) was “abnormally high sensitivity”. All reliable visual fields from each enrolled eye were utilized, starting with the first visual field within 6 months of the first RNFL OCT visit. Subjects were required to have 5 good-quality OCT scans and at least 5 reliable SAP tests.
In the present study, we were interested in assessing whether fast progression detected on OCT during the initial follow-up period would be associated with concurrent or subsequent visual field loss. The rationale for this design was based on the concept that if such an association is demonstrated, it would then be justified for clinicians to make clinical decisions based on the initial assessment provided by the OCT, rather than having to wait until irreversible loss of vision occurs as measured by SAP (see Discussion). In order to assess rates of change on OCT during the initial follow-up period, we used global RNFL thickness values of the first 5 valid SD-OCT visits. Rates of visual field loss and magnitude of visual field decline were then measured by SAP mean deviation (MD) during the entire follow-up period, starting with the baseline visit corresponding to the first SD-OCT date through the last available visit. A joint longitudinal linear mixed effects model was used to estimate rates of change for both SD-OCT and SAP, with random effects applied at the eye level. Details about this model have been presented elsewhere. In short, linear mixed models estimate the average rate of change in an outcome variable using a linear function of time, and subject- and eye-specific deviations from this average rate are introduced by random slopes. In joint longitudinal modeling, both OCT and SAP data are modeled concurrently, allows a better determination of the true underlying relationship between the two outcomes by taking into account measurement error.
Eyes were categorized as fast, moderate or slow progressors based on the rates of change in global RNFL as follows: fast OCT progressors were those with a change in global RNFL thickness faster than -2 µm/year, moderate progressors were those with a rate between -2 and -1 µm/year, and slow progressors were those with a rate slower than -1 µm/year.
We then performed univariable and multivariable regression to evaluate the impact of age, gender, race, mean IOP during the initial follow-up period, OCT progressor group, and baseline MD on the rate of SAP loss. Adjusted rates of SAP loss from the multivariable model were then used in conjunction with the mean total follow-up period in this study to estimate the total magnitude of SAP loss during the study period among OCT progressor groups.
Analyses were performed using R 3.6.1 (R Core Team, Vienna, Austria) within the Protected Analytics Computing Environment (PACE), a highly protected virtual network space developed by Duke University for analysis of identifiable protected health information.
The study included 1,150 eyes of 839 patients with POAG extracted from the Duke Glaucoma Registry. Baseline characteristics are provided in Table 1 . Mean age was 66.0 ± 9.9 years, with African-Americans comprising 32.2% of the cohort. Average RNFL thickness at baseline was 73.9 ± 16.7 µm, while baseline MD was -5.44 ± 5.56 dB. The initial follow-up period corresponding to the 5 OCT scans used in this analysis was 3.7 ± 1.5 years. Mean IOP during this initial follow-up period was 14.5±3.2 mmHg. The average total follow-up period was 6.1 ± 1.9 years.
|Characteristic||n = 1,150 eyes of 839 patients|
|Age (years), Mean ± SD||66.0 ± 9.9|
|Sex, female (%)||452 (53.9)|
|Race, (%) Black or African American||270 (32.2)|
|Mean initial IOP (mmHg)||14.5 ± 3.2|
|SD – OCT|
|Number of tests, n||5,750|
|Follow-up time, for the first 5 SD-OCT tests (years), Mean ± SD Median (IQR)||3.7 ± 1.5 3.6 (2.3; 4.6)|
|Baseline mean RNFL thickness (µm), Mean ± SD Median (IQR)||73.9 ± 16.7 73.0 (61.0; 85.0)|
|Baseline mean SD-OCT quality, Mean ± SD Median (IQR)||25.1 ± 4.5 25.0 (22.0; 28.0)|
|Number of tests, n||10,259|
|Total follow-up time (years), Mean ± SD Median (IQR)||6.1 ± 1.9 5.9 (4.8; 7.5)|
|Number of tests per eye, n (%) Mean ± SD Median (IQR)||8.9 ± 4.4 7.0 (6.0; 11.0)|
|Baseline SAP MD (dB), Mean ± SD Median (IQR)||-5.44 ± 5.56 -3.61 (-7.37; -1.84)|
|Baseline SAP PSD (dB), Mean ± SD Median (IQR)||5.50 ± 3.81 3.81 (2.39; 8.16)|