M R Wilkins1, F W Fitzke, P T Khaw. 1. Wound Healing Research Unit, Institute of Ophthalmology, London, UK. mail@markwilkins.co.uk
Abstract
BACKGROUND: Current pointwise linear regression (PLR) change criteria for visual field analysis are largely empirical. METHODS: Two independent sets of Humphrey Field Analyzer fields were analysed using PLR. Set i, 56 patients, and set ii, 97 patients, were followed over 16 months. Criteria were tested against set i, and then validated using set ii. Each criterion specified a fixed critical slope of 1 dB/year and with a range of significance from P<0.001 to 0.05. The criteria were varied by altering location number, cluster arrangement, and by requiring points to show change over both 12 and 16 months. True glaucomatous change was differentiated from noise by looking for exclusive progression (EP), the detection of progression without detection of improvement. RESULTS: Set i required 1 point to have a slope of 1 dB/year and P<0.05 labelled 64% progressing and 58% improving, whereas several stricter criteria were capable of detecting EP. Two points in a perimetric nerve fibre bundle (PNFB) cluster gave optimal EP detection, labelling 8.9% progressing in set i and 7.2% progressing in set ii with a cutoff P-value of 0.026 inset i and 0.013 inset ii. CONCLUSION: Lax PLR criteria detect large amounts of change. Validating criteria using two data sets allow selection of better criteria, capable of detecting EP. The criterion involving 2 points changing in a PNFB cluster offers the best option for exclusively detecting progression.
BACKGROUND: Current pointwise linear regression (PLR) change criteria for visual field analysis are largely empirical. METHODS: Two independent sets of Humphrey Field Analyzer fields were analysed using PLR. Set i, 56 patients, and set ii, 97 patients, were followed over 16 months. Criteria were tested against set i, and then validated using set ii. Each criterion specified a fixed critical slope of 1 dB/year and with a range of significance from P<0.001 to 0.05. The criteria were varied by altering location number, cluster arrangement, and by requiring points to show change over both 12 and 16 months. True glaucomatous change was differentiated from noise by looking for exclusive progression (EP), the detection of progression without detection of improvement. RESULTS: Set i required 1 point to have a slope of 1 dB/year and P<0.05 labelled 64% progressing and 58% improving, whereas several stricter criteria were capable of detecting EP. Two points in a perimetric nerve fibre bundle (PNFB) cluster gave optimal EP detection, labelling 8.9% progressing in set i and 7.2% progressing in set ii with a cutoff P-value of 0.026 inset i and 0.013 inset ii. CONCLUSION: Lax PLR criteria detect large amounts of change. Validating criteria using two data sets allow selection of better criteria, capable of detecting EP. The criterion involving 2 points changing in a PNFB cluster offers the best option for exclusively detecting progression.