Purpose
To report on visual field variability, outliers, and the ability to detect change over time in a large dataset of patients followed at the QEII Eye Care Centre in Halifax, Nova Scotia.
Methods
Visual field series (SITA Standard 24-2 test, Humphrey Field Analyzer) were analyzed from all patients who had undergone 5+ examinations over 2+ years. Variability over time was estimated by the median absolute deviation (MAD) of differences in Mean Deviation (MD) between consecutive tests, and related to the median MD of the series using quantile regression. Outliers were determined with MM-estimation, a regression technique that iteratively determines "robustness weights" to reduce the impact of suspect observations on the estimated rate of change and its statistical significance. We classified observations with weights <0.5, <0.05, and <0.001 as suspect, probable, and definite outliers.
Results
Visual field series were analyzed from 2,200 patients (4,160 eyes; n=36,600 exams; median age [interquartile range, IQR], 66 [57,74] years; MD, -2.4 [-5.3, -0.9] dB; follow-up, 7 [5, 10] years; 8 [6, 11] exams). Suspect, probable, and definite outliers were present in 13%, 7%, and 5% of visual field series. On average, the MD difference between consecutive tests was 0.6 dB in eyes with near-normal visual fields (median MD of series, 0.0 dB) and 1.3 dB in eyes with moderately advanced damage (median MD, -10.0 dB, Fig. 1). However, the within-series variability varied more than 2-fold between patients, from 0.8 dB (25th percentile) to 2.0 dB (75th percentile).
Conclusions
Visual field series from patients followed in clinical practice often contain outliers which violate the assumptions of ordinary-least-squares (OLS) regression. Robust regression techniques should be used routinely to estimate rates of change over time and to alert clinicians to suspect data. Variability over time varies more than 2-fold between patients with similar visual field damage, underscoring the importance of determining the significance of change based on individualized rather than population-based criteria.