Abstract
Abstract: :
Purpose: Current methods for determining glaucomatous visual field deterioration with trend analysis lack specificity (Spry, IOVS 2000) and assume time-independence of sensitivities. This study addresses these concerns by examining a novel application of time series analysis. Methods: Humphrey 24-2 field series from 104 patients (mean follow up 8.3 years) participating in the Advanced Glaucoma Intervention Study were analyzed. Absolute sensitivities were fitted against time with a time series model (Auto-Regressive Integrated Moving Average or ARIMA). The ability of the ARIMA model to correctly identify progression was evaluated with ROC analysis, with field series classified as "definitely stable" (n = 53) and "definitely progressing" (n = 51) according to standard pointwise linear regression (PLR) criteria (≷= 1 field locations with significant slope of deterioration ≷ 1dB /year) plus clinical assessment based upon the independent, masked evaluation of 4 clinicians. Results: The best ARIMA parameter was overall mean deterioration in sensitivity (µ) + 2 S.D. of that change, which achieved good sensitivity (94%) and 92% specificity, as against clinician plus PLR. Results show that ARIMA is conservative, with fewer locations per field identified as deteriorating compared with PLR. Conclusion: ARIMA identified glaucomatous visual field deterioration with good sensitivity and specificity compared with current methods, and can effectively take into account the high variability of deteriorating field series. Further research will investigate whether the more conservative nature of the model leads to improved specificity in determining field change, compared with PLR.
Keywords: 624 visual fields • 359 clinical research methodology